Advances in X-Ray Analysis - Volume 33 - 1989
Advances in X-Ray Analysis - Volume 33 - 1989
Advances in X-Ray Analysis - Volume 33 - 1989
X-RAY ANALYSIS
Volume 33
A Continuation Order Plan is available for this series. A continuation order will bring delivery of
each new volume immediately upon publication. Volumes are billed only upon actual shipment.
For further information please contact the publisher.
ADVANCES IN
X-RAY ANALYSIS
Volume 33
Edited by
Charles S. Barrett
University of Denver
Denver, Colorado
John V. Gilfrich
Sachs/Freeman Associates
Washington, D.C.
Ting C. Huang
IBM Almaden Research Center
San Jose, California
Ron Jenkins
JCPDS-International Centre
for Diffraction Data
Swarthmore, Pennsylvania
and
Paul K. Predecki
University of Denver
Denver, Colorado
Sponsored by
University of Denver Department of Engineering
and
JCPDS - International Centre for Diffraction Data
The Plenary Session on x-ray analysis of thin films did not just
happen this year but really began four years ago with Paul Predecki
suggesting a special session devoted to thin film techniques. The session
generated a great deal of interest, so Paul suggested that a workshop on
thin films should be slated for the 1987 conference. A full day was devoted
to the workshop, which was split into a half day on epitaxial thin films and
the other half day on polycrystalline thin films. The workshop attendance
indicated a great deal of interest in this topic, leading to this year's
Plenary Session.
The first two speakers of the Plenary Session (B. Tanner and K. Bowen)
have been key throughout the thin film activities. They were invited
speakers for the 1985 special session on thin films and instructors for the
1987 workshop on epitaxial thin films.
v
vi FOREWORD
parameters technique' which has extended the range of analysis to include the
determination of single and mUltiple layer unknown films without the need
for similar types of standards.
Charles Goldsmith
East Fishkill, NY
December 1989
PREFACE
vii
viii PREFACE
Tutorial workshops on various XRD and XRF topics were held during the
first two days of the conference. These are listed below with the names of
the workshop organizers and instructors.
The total number registered for the conference was 415, over 225 of whom
registered for one or more workshops. We are particularly indebted to the
workshop organizers and instructors who gave unselfishly of their time and
experience to make the workshops an outstanding part of the conference.
Paul K. Predecki
for the Organizing Committee
January 1990
UNPUBLISHED PAPERS
The following papers were presented at the conference but are not
published here for various reasons.
xiii
CONTENTS
xv
xvi CONTENTS
SPC: What is It and Why should You Use It in Your X-Ray Analytical
Laboratory? 537
P. B. DeGroot
A Compact On-Line XRF Analyzer for Chemical and Petrochemical Processes. 585
D. J. Kalnicky, A. I. Ahonen, M. J. Hietala and K. J. Kauhanen
xx CONTENTS
B K Tanner
Abstract
Use of a reference crystal to condition the beam in the double-axis diffractometer
permits the Bragg peak width to be reduced to the correlation of the two crystal
reflecting ranges. Some recent applications of double axis diffractometry to the study
of heteroepitaxial layers are discussed. The advantages of multiple reflections for
beam conditioning and the four reflection DuMond monochromator are examined.
Glancing incidence and exit diffractometry permits the study of very thin layers, down
to a few tens of nanometres in thickness and both synchrotron radiation and skew
reflections can be used to tune the glancing angle close to the critical angle. Recent
applications of triple-axis diffraction, where an analyzer crystal is used after the
specimen, to the study of very thin single epitaxial layers and multi quantum well
structures are reviewed.
Introduction
The width of the Bragg peaks, and hence the angular resolution, in single-axis
diffractometry is determined by the collimator geometry, the source size and the
spectral width of the characteristic lines in the X-ray spectrum. In practice, this
means that the Bragg peaks from a substrate and a nearly lattice-matched epitaxial
layer cannot be resolved. The requirement to determine non-destructively the
composition of heteroepitaxial layers grown for electro-optic applications has
stimulated the widespread use of high resolution diffraction over the past decade. By
simulation of rocking curves of model structures and seeking a best match with
experiment, a very great deal of structural information can be acquired. Indeed,
although not discussed here, simulation is essential for the interpretation of very thin
and multiple layer structures. While standard geometries are well suited to quality
control of wafers for commercial device fabrication and the characterization of many
research systems, there is an unrelenting drive toward the exploitation of very thin
layers where quantum confinement effects give rise to totally new physics from
which novel electronic devices can be devised. This challenge to characterize
nanometre-thick layers at the Angstrom level has been met by precision diffraction
techniques. The methodology of determining structural parameters of epitaxial layers
from the X-ray data has recently been surveyed 1,2. Extensive bibliographies are also
contained in these papers 1-3. Here, an attempt is made to compare the information
available from different settings and clarify the criteria to be used when designing
Double-Axis Diffractometry
Reference E~~~
Source ~ (a)
~ Collimator Specimen
camt•• t.
1720 1725 1730 1735 17040 1745 1750 17&5 1750 17&5 1770 1775
Fig 1. (a) The (+n,-n) double-crystal diffraction geometry. (b) Silicon 004, 004
reflection rocking curve with CuK radiation.
B. K. TANNER 3
for strained-layer epitaxial systems such as GaAs on Si, CdHgTe on CdTe, or GejSi
on Si, the mismatched layer results in very significant wafer bowing. Unless the area
of the specimen sampled by the beam is quoted, results of different investigations
cannot hope to be compared in a meaningful manner. An example of good practice is
the correlation of dislocation density and rocking-curve full width at half-height
maximum (FWHM) in CdxHgl_xTe on GaAs by Brown et aZ 6 in a volume containing
a number of high-resolution diffraction papers. While it is standard practice to use an
open detector, it should also be noted that use of a slit before the detector to
improve signal to noise can lead to a change in the FWHM of the rocking curves in
imperfect crystals.
(1)
where 8 B is the Bragg angle, ).. is the X-ray wavelength and 'lJ h is the the cosine of
the angle between the diffracted beam and the inward surface normal. Although
direct measurement of fringe period in multiple-layer structures cannot be used to
determine t 7 due to beating of the fringe systems, Fourier analysis will yield
thicknesses directly 8. Note that it is important to Fourier-analyze the rocking curve
plotted on a logarithmic scale as otherwise the Bragg peak totally dominates the
spectrum. A major importance of this technique is that it fixes the layer thicknesses.
In mu1tiple-Iay~r structures, very different structures can give rise to very similar
rocking curves and there is danger that in least-squares minimization methods with
many variables, one may end up at a local minimum which does not correspond to
the experimental structure.
(2)
where 8 1 and 8 2 _are the Bragg angles of reference and specimen and ~" is the
bandwidtb of the -X-ray spectrum passed by the collimator. In most experimental
situatio,!ls this will approximate to the splitting of the Koc. 1 and K(){2 lines. Thus, the
004, 004 rocking curve of InP and GaAs is broadened by about 25 secs which may be
quite acceptable for the study of strained-layer systems such as InGaAs or GaAsSb
on GaAs. Similarly, the mismatch between Si and GaAs results in a peak broadening
which is small compared with the rocking-curve FWHM of GaAs grown epitaxially
on Si 10. As the lattice parameters of Ge and GaAs are very nearly equal, it follows
that the double crystal setting is also satisfactory for the study of SixGe1_x on Si 11.
However, with InP and Si, the mismatch is so large that, in the 004, 004 reflection,
the Kot 1 and Koc 2 lines are resolved (Fig 2). The reason can be readily seen from the
4 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
25000
(a) 8000 (b)
7000
20000
6000
ISOOO 5000
4000
10000
3000
2000
SOOO
1000
6600 66SO 6700 67SO 6800 -140 -120 -100 -80 -60 -40 -20 -j) 20 40
Fig 2 Rocking curves in the (+n,-m) setting where the lattice parameters are not
equal; 004 reflection using CuKc:t radiation with (a) InP and GaAs. The GaAs peak is
additionally broadened to 44 secs due to an epitaxial InGaAs layer which curves the
wafer. (b) Si and InP. The two peaks are the resolved Kc:t1 and lines. K«z
DuMond diagram 1 or by ray tracing. In the (+n,-m) setting the Kl)(l and Kit. rays in
Fig 1 are no longer diffracted when the second crystal is parallel to the firs!. When
the difference in angular setting of the specimen to satisfy these two rays significantly
exceeds the correlation of the two reflecting ranges, then a split peak results.
Fortunately, lattice mismatches in current strained-layer systems is such as to
broaden, not split the peaks.
The ( + n, + n,-mJ setting is a little inconvenient experimentally as, for Bragg angles of
the order of 30 , the diffractometer body has to be at an angle of 1200 with respect
to the X-ray tube housing. A way of avoiding this is to use an extra reflection on the
reference crystal to give a (-n,+n,+n,-m) setting (Fig 4(a)). The first pair of
reflections can be conveniently made for a single-grooved, or channel-cut, crystal and
the diffractometer is now aligned at the same angle as for a single reference
reflection in the ( + n,-m) setting.
(a)
(b) (c)
Fig 3 (a) The (+n, +n,-m) setting showing true beam monochromatization (b) The
DuMond diagram for the first two reflections (c) The DuMond diagram for all three
reflections. The dotted line 3' corresponds to the specimen crystal angularly
displaced. No overlap, and hence diffracted intensity, results from a very small
rotation as in the ( + n,-n) setting 1
(a)
Specimen
(b)
Fig 4 (a) The (-n, +n, +n,-m) setting giving a convenient beam geometry and high,
dispersion-free, sensitivity to lattice strain (b) The four reflection (-n, + n, + n,-n,-m)
setting giving collinear incident and conditioned beams.
6 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
Asymmetric reflections
Symmetric reflections only provide data on the Bragg plane spacing normal to the
surface and thus give no information on the degree of relaxation in a strained layer.
Even when relaxation does not occur, serious errors can be incurred if it is not
appreciated that epilayers can be, and often are, tilted with respect to the substrate.
Provided that the layer is not sufficiently thin that interference effects lead to peak
shifts (typically < 0.5 pm) 19, the measured peak splitting A8 between layer and
substrate is
Fig 5 Multiple-reflection beam conditioner to reduce the tails of the Bragg peak for
study of diffuse scatter.
wings of the Bragg peak can then be readily distinguished. For example, the diffuse
scatter arising from silicon wafer polishing processes can be distinguished 18.
where ¢ is the angle between Bragg planes and the crystal surface and u is the linear
photoelectric absorption coefficient. This formula may be used to determine the
depth sensitivity of X-ray topographs. However, at the Bragg reflection, the
extinction distance determines the depth to which the X-ray wave penetrates. Even in
the case of zero photoelectric absorption, the depth reached by the X-ray wave will,
for strong reflections, usually be only of the order of microns due to the dynamical
diffraction effects occurring. The extinction distance h is given by
[6]
Thus, in either glancing-incidence or glancing-exit geometry, the layer thickness
ds
,,
,,
.
,,
----- ... _-_ ...... "
---.:......._..
,
(a)
IDIID
(b)
are sees
.....
..Il00
(c) (d)
'''''''''
'''''''''
Fig 7. Simulated and experimental rocking curves taken with CuK radiation and 333
reflection of a 100 nm thick GalnAsP layer on InP [28]. An&le between incident
beam and surface of (a) 1.15 0 , (b) 20 (c)11.4° (d)46.9 (glancing exit)
Glancing incidence can be achieved by varying wavelength 25,26 and this is most
appwpriate to synchrotron radiation work. In the laboratory, in order to maintain the
intensIty afforded by the characteristic X-ray lines, skew reflections are used. SARCA
(Skew Asymmetric Rocking Curve Analysis) 27 involves the use of a reflection in
which the Bragg cone is embedded in the crystal surface (Fig 6).
Rotation of the specimen about the diffraction vector thus tunes the angle between
the incident beam and the surface, from a minimum at glancing incidence to a
maximum at glancing exit. As evident from Fig 7, the layer peak intensity varies
systematically with azimuthal angle and the ~reement between simulated rocking
curve and experiment is at all points excellent . The ratio of layer to substrate peak
intensity reaches 53%, in comparison with 14%. which is the highest value in a non-
B. K. TANNER 9
-- & (a)
Speci men
(b)
AllnAs InP(substrate)
(0,39973,4074)
Fig 8 (a) Basic triple-axis diffraction geometry (b) Map of the scatter in reciprocal
space around the 440 reciprocal lattice point from a 20nm thick InAlAs layer on InP
capped by 3 nm of GaAs 34. Flat Ge (111) crystals were used as beam conditioner
and analyzer; 1.54 A radiation.
skew geometry (using the 044 reflection). It should be noted however, that the peak
intensity for a given reflection goes down as the glancing angle is reduced, and there
is thus a balance to be achieved between peak shape and intensity. Also, the peak
widths narrow in the glancing-exit case as the FWHM w is given by
[7]
However, the substrate and layer peaks move together (Fig 7), thus offsetting the
gain in angular resolution.
The method has been exploited by Pietsch and Borchard 29 for the study of lattice
matched C~Sr1_xF2 grown on GaAs. Although for high angle reflections, the Bragg
peaks of substrate and layer overlap, at extreme grazing incidence the effect of
refraction is sufficiently large to separate the two peaks. The magnitude of the shift
depends not only on the incidence angle but also the atomic numbers of substrate
and layer. For 1.35 micron band gap InGaAs on InP the shift for the 333 reflection at
glancing angle of 10 is 16 secs 28 which is rather small in comparison with the rocking
curve peak widths.
Triple-Axis Diffractometry
Triple-axis diffractometry is thus applicable not simply to the most highly perfect
crystals. Indeed it is most appropriate for the study of such systems as GaAs on Si
where it is extremely difficult in the double-axis mode to distinguish between tilts and
dilations. In the triple-axis mode a rocking curve recorded by scanning only the
specimen will provide a measure of the tilts only. A so-called e ~ scan, in which the
analyzer is stepped around the specimen for each angular position of the specimen,
permits a map of the scattering to be reconstructed in reciprocal space. (Note that,
once set up, the angular setting of the analyzer about its own axis remains
e
unchanged.) The ideal method of data collection is one in which the and1 setting
are chosen to give a grid in reciprocal space and it is straightforward to program an
instrument to perform this operation. Integration along a line in reciprocal space
recovers the tilt-free, quasi-double-axis rocking curve 34. Apart from increased
complexity of setting up the apparatus, the only disadvantage of the triple-axis mode
is the increased data collection time but this is of course offset by the increased data
collected! Careful thought to experimental design, such as use of a strong scattering
analyzer, and the size of the area element in reciprocal space from which scattering is
measured can minimize data collection times. A channel-cut InP analyzer crystal
works very well for example and has an acceptance range well matched with
optimum step on the analyzer. Because of the additional Bragg-reflecting elements,
fluorescence from the specimen is eliminated and the signal-to-noise is much better
than in the do~rle-axis diffractometer. A value of 106:1 has been reported by
Zaumseil et al in their study of ion implantation in silicon. They were able to
determine the depth profiles of strain and static lattice disorder. Their measurement
technique was to scan the analyzer at a fixed setting of the specimen and determine
the value of the maximum scattered intensity. This was then plotted as a function of
specimen angular setting.
Fornard Projection
High resolution diffraction has been performed since the 1920s but has only become
widely exploited within the past decade. Double-axis diffraction is now regarded by
many crystal growers as an llltegral part of their analytical armoury. However, most
workers have not exploited fully the variety of geometries or beam conditioners
available. It is clear that, with the recent commercial availability of a triple-axis
diffractometer, this mode of high-resolution diffraction is likely to develop rapidly
and become of comparable importance to double-axis diffraction in the 1990s.
References
D. Keith Bowen
Department of Engineering
University of Warwick
Coventry CV4 7AL, UK
ABSTRACT
INTRODUCTION
micrographs, but the scales are quite different. X-ray images are,
roughly, 1000 times more sensi ti ve to strain, cover 1000 times larger
diameter specimens but have 1000 times poorer spatial resolution. The
methods are complementary, and, for example, one should use electron
microscopy for details of interface structure but X-ray topography (XRT)
for strain mapping in whole wafers (or large portions of them) before or
after device processing.
Singlecryst.almetnQ.ds
In these methods there are no beam-conditioning optics between the X-ray
source and the specimen. The intrinsic reflectivity of the crystal must be
correlated with the properties of the source (e.g. its divergence), so
these methods do not have the highest strain sensitivity. They can,
however, reveal individual dislocations and other short-range strains and
are relatively undemanding on alignment and other experimental techniques.
dp = hb/a (2)
In the section topography and Lang methods (figure lb and lc), a point
source is used and the divergence controlled with slits, so that only the
Kal line is used. It is difficult to make b<10 mm in these cases, but
parameters may now be chosen so that d p - d i - I micron. The specimen
must be scanned through the beam to build up the image for the Lang method
(in reflection or transmission) but if left stationary in transmission the
D. K. BOWEN 15
method is that of section topography. Note that the latter provides depth
information, since it gives diffraction effects in a slice through the
crystal, and can thus yield near-surface information.
I Plate
r-~~-~-rc-e--"I-~
(a) (b)
Traverse
point _ _ _ _ _ _ _ _ _ _-:-_ _ ~
source
(c)
synchrotron
:-=--urce
I
_-Iaf--~~
Plate
It I~
(d)
Figure 1. (a) Berg-Barrett method, (b) Section topography (if the specimen
and film are translated this becomes the Lang transmission method) (c)
Lang method in reflection geometry, (d) white radiation method.
16 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
The reflection methods are much more suitable for thin layer analysis. The
X-ray beam is confined to typically 10 microns of the surface by both
extinction and absorption effects, and thus only thin layers are probed.
Contrast arises primarily through kinematic scattering of X-rays from the
distorted regions around defects in comparison with dynamic scattering in
perfect regions, provided that the defects have a strong enough strain
field (which is the case for precipitates and dislocations). Single
crystal images are 'integrated' images, that is, they may utilise all
angular and spectral components of the incident beam that they can
diffract. Under these conditions, dynamical scattering is proportional to
the structure factor but kinematic scattering to its square, hence
contrast is strong. There must still be sufficient crystal thickness as in
the transmission method for this difference to be appreciable.
How thin a layer can be imaged? This depends upon the absorption, the
extinction and the defect structure of the specimen. If the specimen is
perfect, hence scattering according to dynamical theory, the incident wave
intensity is reduced to lie of its value at the surface in a depth equal
to the extinction distance 9 • In the symmetric Bragg case this is
proportional to sinBIAF, where A is the wavelength and F the structure
factor of the reflection. For an asymmetrical reflection, sinB is replaced
by {sin(B-tlplsin(8-tpl}l12 where 4> is the angle between the diffracting plane
and the surface. For example, taking a GaAs (001) wafer, with CuKa.
radiation, the extinction distance for a symmetrical 004 reflection is 7.1
microns, whilst that for a 224 reflection using a slice cut 4° off the
exact (001) orientation (so that the glancing angle is reduced to 2.6') is
4.2 microns. The effect does not become large until glancing incidence
angles near the critical angle for total reflection are achieved. This may
be achieved in two ways: by use of synchrotron radiation, which allows
arbitrary wavelength and hence Bragg angle selection, and by the use of
skew asymmetric reflections l , in which the crystal is rotated about the
diffraction vector until the glancing angle is sufficiently low.
Extinction distances below 0.5 microns may be achieved in this way,
permitting topography of layers down to about 0.1 microns.
Where the specimen is cut slightly off the exact 001 orientation, one may
use diffraction off planes nearly - but not quite - perpendicular to the
surface, and obtain grazing incidence and grazing exit beams, reducing the
extinction distance dramatically. This was first exploited by Afanasiev 1o ,
to obtain images of features in layers some tens of nanometres in
D. K. BOWEN 17
What is the boundary between perfect and imperfect for the calculation of
penetration depth? Dudley et al. demonstrated that dynamic and kinematic
si tuations can coexist in the salle experiment. Whi 1st scattering re.ote
from a dislocation is dynamic, the local scattering and penetration near a
dislocation is always kinematic. However, the GIE method allows the
dynaaical extinction distance to be reduced so that contrast can be
achieved in submicron layers.
detector
plate or film
X-ray beam
line
source
Jr--
~
(a)
detector
plate or film
X-ray beam
(b)
these are not integrated images. The rate of change of intensity with
strain is therefore given simply by the (local) slope of the rocking
curve. Both short- and long- range strain fields are imaged.
It must once more be emphasized that topographic methods image the strain
fields of defects, not their central positions. This is significant in the
imaging of dislocations in coherent or semi-coherent epilayer materials.
Whether the dislocations are exactly in the interface, or just above or
just below it, the strain field of a dislocation will extend in both the
substrate and the epilayer. An image taken of the layer alone - that is,
with the diffractometer set on the flank of the layer peak - will still
give contrast from dislocations in the substrate, though with decreasing
contrast as their distance from the interface increases. Image simulation
is a powerful tool in this connection 16 but has not yet been sufficiently
widely applied.
(a) (b)
EXAMPLES OF APPLICATIONS
EpilaYf;)r mismatch
In strained-layer heteroepitaxial systems it is important to confirm the
absence of an interfacial dislocation network, and in relaxed or
partially-relaxed structures it is necessary to measure the degree of
relaxation and its detailed structure. Whilst the average relaxation may
be measured accurately with X-ray diffractomet ry 17 and in some cases
considerable detail may be revealed by etching 18 , only XRT will provide
large-area defect mapping of both layers and bulk, with the potential for
quantitative measurements such as Burgers vector analysis. Figure 3 shows
interface dislocation generation in an Si-Ge alloy. In the thin layer,
just above the critical thickness, nucleation has begun but the
dislocations are still confined near their initiation points and have not
spread far. In the thicker layer, the dislocation density is near the
limit of resolution yet the inhomogeneity can still be seen. Dislocation
nucleation and kinetics may thus be studied by this means (in this
material it is also known that the dislocations extend throughout the
crystal to form a relatively uniform net at device processing
temperatures).
D. K. BOWEN 21
PrQcess:::-relateddef.ects
Defects are inevitably introduced into the material during device
processing, and XRT is a powerful tool for process development and
troubleshooting. X-rays can also penetrate surface layers that are opaque
to other methods such as photoluminescence. Figure 4 is an image of a GaAs
wafer containing an array of FET devices. The long-range strain field,
substrate dislocations near the surface, the devices themselves and even
scratches on the metallisation are clearly seen.
Various doping and thermal treatments are now in use in silicon-device
fabrication with CZ silicon in order to produce a surface zone, around 20
microns thick, which is denuded of oxygen precipitates. Characterization
of this feature during process development and control is a difficult
problem by other methods, but figure 5 shows how the penetration depth may
be tuned in a topograph to examine this surface layer.
111
- 9
. "I
u
I
Figure 6. Spherical precipitate strain fields revealed by section
topographs. 422 reflection, MoKa, field width 520 microns. The simulations
are shown opposite the appropriate part of the images. Courtesy Cui, Green
and Tannerl9.
As device layers become thinner, so the quality of the surface upon which
they are grown becomes more important; moreover, the application of modern
'machining' methods such as chemi-mechanical polishing, ultraprecision
machining, laser and ion milling and reactive ion etching depends upon the
residual damage that they introduce. Surface-sensitive characterization
methods are important and the sensitivity of the double-crystal X-ray
methods in diffractometry have been shown by Abdul Gani et al. 20. A
section topograph of laser-drilled holes in silicon is shown in figure 7.
The Pendellosung fringe system is highly sensitive to strain, and the
distortion of the fringes around the hole clearly shows the near-surface
strains, even in the section taken over 2 mm away from the edge of the 200
micron hole.
The use of grazing incidence beams with short wavelengths can reveal
astonishingly small strains. Topographic contrast has recently been
observed on high quality silicon wafers polished with silica chemi-sol -
the standard silicon device material - using 880 or 886 reflections with
Mo Ka in which the intrinsic rocking curve width is 0.5 arc seconds. The
strains involved have been found by precision diffractometry to be a few
parts in 10' and are largely confined to the top 20 nm 2l .
D. K. BOWEN 23
REFERENCES
Jerome B. Cohen
INTRODUCTION
While any subject associated with films evokes immediate thoughts about
solid-state devices, there are other much more mature areas where it is impor-
tant, such as coatings, and yet work on this entire subject is only in its in-
fancy. Residual stresses are not only important because of their well-known
effect on corrosion and mechanical properties, they can also lead to other
severe problems. For example, it is now believed that stress gradients can
lead to void formation and subsequent failure in metallization in electronic
devices 1 • Thus, it is not only the level of stress that can be important, but
gradients across a surface of a thin stripe, or through the thickness.
There are really three classes of problems associated with this topic:
In this very brief paper we can only review techniques, as well as some
recent results. There have been two excellent re~ent publications that can
serve to bring the reader up-to-date in this area ,3.
Techniques
Returning to x-ray peak shifts, more often than not, we are dealing with
either single crystal films, or highly textured polycrystalline films. This
poses special problems, because the normal "d" vs sin2~ technique for stress
measurement can be difficult if you can't find the peak!
However, even for highly textured films, adequate intensity can often be
obtained, either with a synchrotron source, or with a rotating anode genera-
tor. For example, in our own laboratory, we've found that a 12 kW rotating
1.1HOC)
1.18l16O
1.18860
"Uz 1.18840
~
Vl
6 1.111820
Sample 4
1.18l1OO
Stress - 71 +/- 1 MPa
1.18780
1,16760 c....L.J...L.J...L.J...L.J....u...u........................I...L..............................w...L.L..L.L..L.L..&..L.w...&..L..&..L..&..L..&..L.J
0,0 0.2 0.4 0,11 0,11 1.0
SIN SQUARED PSI
anode generator is adequate for work on thin AI-alloy films. And it is par-
ticularly simple to pick the radiation you wish to have by plating a narrow
strip around the demountable target, masking the rest of the target with plat-
ing tape. We·ve had good experience (long life) for example, with hard
chromium plating 2~m thick on a copper target 12 . The chromium can be removed
by light polishing, with the target in a lathe.
When the peaks are weak, better precision for the peak position can be ob-
tained by fitting with pseudo-Voight functions, than with parabolae. This
also permits using a functional form for the background, especially important
if the weak peak is in the tail of a much stronger one. One example of our
results, on a highly textured Al-2 wt pct. Cu film on Si, is given in Fig. 1.
S. S~ SJ
(L 1) xf cos~cosX sin¢Cosx cos(900+X)
(2)
Therefore:
2,;..... 2 2 2
X+ X+E))cos X.
Co • Co .,;.....
+<'22 s l. n 'l"'l.n <'2)sl.n'l"'l.n (3)
28 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
With six or more independent reflections for which X,¢ can be calculated
(or experimentally measured since stress may cause them to change), the six un-
known E., can be obtained. These can then be referred to the crystal axes
using Eq~. 2 and the direction cosines between the axes Si and the crystal
basis. The stresses in the crystal axial system can be obtained from:
These may be referred back to the axes Si by the inverse of Eqn. 2 (with
a ij replacing Eij ).
(5)
and to obtain the S' from S" in the crystal axial system:
S'ijk1 (6)
Also:
(7)
C33' = Aa u + Ba 22 , (8)
where
Some Examples
Where subscripts f and s represent film and substrate and the (X are coeffi-
cients of expansion. For a stripe, with a passivation layer:
For Al on Si, Eqn. 10 gives - 2 MPa/oC, whereas Eqn. 11 gives 4.4 MPa/oC.
The films used in this study were continuous and sputtered at - 150°C and the
passivation layer was produced at 350°C and then thermally cycled at 125°C.
o
~ 150
(fJ
(fJ
~ 100
V;
50
O+nTITrrnTITrrnTITrrnTITrrnTITTIT~TITrrnTn
0.000 0.500 1.000 1.500 2.000 2.500
Film Thickness (microns)
FIG 4: Residual stresses in AI-Cu films on Si single crystals,
after long term storage (solid lines) or after cycling to
125°C for 1015 cycles and 1 month storage.
30 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
20000
16000
o
CL
::;; , '
120.00
Ifl
Ifl
Q)
U)
8000
Time (days)
250 00
20000
I
~:::::l
(f)
Stress versus Time for l micron AI-Cu film on
S, during heat treatment at 350°C and after
quench to room temperature
~ 50,00
U)
000
-50.00
J T=350'C
-10000 ~Tn-rTTn"Tn-rTTn"Tnn-rTnn-r~n-r~n-r"n-r"n
0.00 5.00 1 0.00 15.00 20.00 25.00
Time (Hr.)
alloys (which is - 60MPa), and close to the ultimate tensile strength. In ef-
fect, the film takes on some of the mechanical character of the substrate,
which is somewhat surprising at these film thicknesses. In fact, in a bend
test of one of these films on its substrate, the Si fractured when the applied
stress in the outermost layer (the film) was - 70 MPa. During this test we
followed the "d" spacing, looking for a change in slope vs strain, indicating
plastic rather than elastic deformation; the film was still behaving elasti-
cally at fracture. We could perhaps better define a film, as that layer that
has different properties than the bulk, rather than by its thickness, and/or
perhaps these layers in which a biaxial stress state prevails.
The Future
Work in this area is really just in its infancy and much more is needed:
1) We need to know more about gradients, with depth, and, with stripes,
across the face.
J. B. COHEN 31
4) Eqn. 10 (or 11) can be used to test how strongly a film is bonded to
its substrate; measurements of stress (or just "d") vs temperature should fol-
low this equation, if the bonding is strong ?1er the temperature range, as has
been demonstrated by Vreeland and co-workers .
Unfortunately, while there are now adequately intense sources for work on
actual devices, the grain size in the stripes is probably too coarse to work
at this level. We will have to be content with model systems, such as a col-
lection of parallel strips.
ACKNOWLEDGEMENTS
REFERENCES
ABSTRACT
INTRODUCTION
alloy? Can one deduce and predict the macrodeformation response and ther-
mal stability of an alloy from the deformation characteristics of its
microstructure? The dilemmas that one usually faces in answering these
questions are inextricably linked to the investigative methods applied.
Although one can well characterize the induced dislocation structure of the
grains by optical metallography and electron and scanning microscopy, the
cooperative behavior of large dislocation populations, conditioned by the
inhomogeneous distribution of deformation encountered in macrodeformation,
cannot be revealed by these methods. X-ray powder diffractometry, sampling
a large strain population, is capable of giving information of group ef-
fects on macroplasticity and on residual elastic strains. The former can
be deduced from the broadening and the latter from the peak shifts of the
X-ray lines. These investigative methods, however, lose sight of the
deformation characteristics of the individual grains and of local in-
homogeneity and severity of deformation which may disrupt the integrity of
the alloy and cause it to fail. This paper represents one aspect of a
series of studies which made it its goal to establish a link between micro-
and macro-mechanics [1-8). It will focus on the workhardening and recovery
characteristics as well as on aspects of stress corrosion cracking of the
nickel-base Alloy 600. This alloy is a solid solution alloy of Ni-Cr con-
taining chromium carbide precipitates and because of its low stacking fault
energy deforms by planar slip mode into a Taylor type of lattice. The al-
loy has considerable technological importance and has been the subject of
many studies for steam generator tubing in pressurized water reactors and
ensuing problems of stress corrosion cracking [19). In the present study
the link between the micro and macro aspects of deformation, recovery and
stress corrosion cracking is sought by characterizing quantitatively the
dislocation structures of the individual grains of a numerous grain popula-
tion via X-ray rocking curve measurements using the CARCA method as
principal research tool [13), supported by TEM measurements [8). The un-
derlying reason for linking recovery studies to the preceding deformation
process is to obtain information on the thermal stability of immobile-
dislocation obstacles, such as sessile and forest dislocations and Lomer-
Cottrell locks. These are invariably generated during the deformation of a
Taylor type lattice and can be removed by controlled isothermal anneals.
By determining the activation energies of recovery a clue can be obtained
as to what type of deformation microstructure is associated with the ther-
mal response of the microstructure.
EXPERIMENTAL METHODS
Alloy 600 tubing was investigated with the composition 0.03 wt. % C,
76.2 wt. % Ni, 15.4 wt. % Cr, 7.6 wt. % Fe, 0.25 wt. % Mn, 0.2 wt. % Si,
0.37 wt. % Al, 0.26 wt. % Ti, 0.26 wt. % Cu, 0.001 wt. % S, 0.009 wt. % P
and 0.003 wt. % B. Tensile samples were prepared by cutting longitudinal
strips from the tubes and annealing at 704°C for 15 hr. Prior to deforma-
tion, the samples were electropolished to remove a thin ground surface
layer and any residual oxide layer. The grain size of the material ranged
between 10 and 100 ~m. Deformation was carried out on samples to which
precision resistance (120 0) strain gages were attached and after which the
samples were strained to various amounts between 0.3% and 2.4% plastic
strain.
The samples for the TEM analysis were prepared by jet polishing with
an electrolyte of 7% perchloric acid and 93% methanol at -27°C. The ex-
amination was then carrieo out on a JEM-lOO CX microscope.
The dislocation density was measured by two sets of grid lines using
the expression derived by Smith and Guttman [14]. The average dislocation
density p can be expressed by the equation
p (1)
RESULTS
Figure 3 exhibits the stress-strain plot of Alloy 600 with small and
large grain size and that of pure copper having a grain size of 20~m.
Polycrystalline copper was selected as a model material to contrast the
work-hardening behavior of Alloy 600 because it is well known that copper
develops upon deformation 3 pronounced cell structure which in Alloy 600,
due to its low stacking fault energy, is totally absent.
500
ALLOY 600
400 ( 1O\Jm)
'@
a. 300
6
(jJ ALLOY 600
(jJ
w (110\Jm)
a: 200
I-
(jJ
100
o o~--------~--------~--------~--------~-
0.01 0.02 0.03 0.04
STRAIN, C
where a is the stress, a the yield stress, € the strain, A the material
constant, € the strain Xt elastic limit and n the strain-hardening ex-
ponent. y
10
6
i3 /
/
i3 0 ./ \
ALLOY 600
( 11Ollm)
4
....
.-' . " '\
,
,,
ALLOY 600
2 ( 1Ollm)
o
o 0.01 0.02 0.03
STRAIN. t
Fig. 4. Normalized plot for ~/~ vs. € for Alloy 600 of grain size
10 ~m (-----), for Al18y 600 of grain size 110 ~m (---), for
pure copper ( ).
C. F. LO ET AL. 39
For Alloy 600 with grain size 10 ~m, the microdeformation plot of Fig.
5 exhibits shape changes that reflect those of the macroscopic workharden-
ing curve of Fig. 3. There is an inflection point at £ = 0.007, designated
by I in Fig. 5, above which the increase of ~ diminished. The segment
below € = 0.007 corresponds to that of the macrodeformation plot having the n
value of 0.109. The increase of workhardening above £ = 0.007 to n = 0.343
(Fig. 3) is reflected in the microdeformation plot of Fig. 5 in much
greater detail. The diminished increase of ~ reaches a maximum at € =
0.015and at larger strains it reverses its course, and goes downward. The
C'
>
I-
4
(3
i=
(j)
«
...J
a. "-
0
a:
u 3
,
,I, ALLOY 600(110Ilm)
~
...J
«
I-
,,
"
Z
,--,
UJ
~
I ,' ... ...
+,'
2 ... ...
UJ
a:
u ,, ......
....
~ ,, i . ··
-- ALLOY 600(1Ollm)
0
0 0.01 0.02 0.03
STRAIN, (;
Fig. 5. Plot of incremental microplasticity (~=o(~/~ )/0£) vs. strain
for Alloy 600 of grain size 10 ~m (-----), f8r Alloy 600 of
grain size 110 ~:1 (- - -) and for pure copper ( - - ) .
40 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
For Alloy 600 with grain size 110 ~m the break in the stress-strain
curve can be seen at strain level € = 0.0125 when the initial parabolic
shape of the curve transforms gradually into a linearly-ascending segment
(Fig. 3). This hardening effect is manifested in the microdeformation plot
by a pronounced downward turn of~. It is of interest to note that the
strength increase of the fine-grained Alloy 600, manifested on a macro-
scopic scale by the higher a vs. € curve, is reflected on a microscopic
level in Fig. 5 by the trend of ~ values which consistently lies below that
of the large-grained alloy.
25
20 NO STRAIN
~
rn 0.3% PLASTIC STRAIN
~
a:
(!)
15
....
0
UJ
z~
10
UJ
u
a:
UJ
~ 5
0
0 10 20 30 40 50 60 70
HALFWIDTH. B (MIN OF ARC)
30
0.3% PLASTIC STRAIN (APEX POSITION)
593°C ANNEALED
ANNEALING TIME
25 (MINUTES)
o
~ 20 5
a: 10
(!)
15
~ 15 NO STRAIN
w
(!)
<{
I- 10
Z
W
()
a:
w
Il. 5
o 10 20 30 40 50 60 70
HALFWIDTH (MIN OF ARC)
Fig. 7. Microstrain distribution as a function of annealing time
at an annealing temperature of 593°C for 0.3% plastic
deformation: - - - , 0 min; - - -, 5 min;
10 min; - . -, 15 min; ...... , no strain.
With the aid of these micros train distribution curves, a much greater
precision in determining ,he recovery of the residual strain was possible
than is obtainable by con'entional X-ray methods. As shown in Fig. 8, the
residual strain was obtained by calculating the ratio of the area AB be-
tween the micros train distribution curves of a certain annealing time
(upper limit) and the unstrained state (lower limit) to the area AA between
the distribution curves of deformed samples prior to recovery annealing
(upper limit) and the unstrained state (lower limit). Thus
X (4)
The resulting curves of fraction In{l/(l-x)) of recovery vs. time are shown
in Fig. 9. A good straight-line fit was obtained except at the highest
temperature where the recovery process reached a saturation level very
quickly. By replotting these data for constant values of X, it was pos-
sible to calculate the activation energy of the recovery process. Thus,
Fig. 10 shows three curves representing time and temperature combinations
42 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
1.0 ff//
0/
: " ~,,,oc
/ ~ --0-- 649°C
- 0 - 704°C
10 100
TIME (MINUTES)
Fig. 9. Isothermal annealing response of Alloy 600 deformed to
0.3% plastic strain: '" , 593°C: 0 , 649°C; 0, 704°C.
C. F. LO ET AL. 43
TEMPERATURE ('C)
I I
STRAIN RELEASE
- t : . - 30% ~
O~
0UJ
1000 ::=~~~ ~D~
O~ ~~
~o~:~O
~
UJ
::E
i=
............ D~ ~ 0 30 %25.52 kcal/mole
~ 0 50 %:25.64 kcal/mole
100 ............ t:. 0 70 %:25.74 kcal/mole
I I
1.00 1.05 1.10 1.15 1.20
l/Tx 10 3 (oK')
to achieve 30%, 50% and 70% recovery. The slope of each line represents
the activation energy for the recovery process to that particular point.
The following activation energies were obtained:
-1
25.52 kcal mol_ l
25.64 kcal mol_ l
25.74 kcal mol
The good agreement between the three measured values indicates that only a
single atomistic process is operating during the l~y-strain recovery of
Alloy 600. The value obtained here, 25.6 kcal mol ,is in fairly good -1
agreement with the value obtained by Smidoda et al. [19) of 26.9 kcal mol
for nickel grain-boundary self-diffusion. This process as a controlling
recovery mechanism would be consistent with the results of the high voltage
TEM studies by Bruemmer and coworkers and this study, and also with the
fact that grain boundary carbides were the primary source of dislocations
generated during low strain deformation, and the fact that the dislocations
were arranged in a planar low-energy dislocation Taylor-type of lattice
adjacent to the grain bounjaries.
ALLOY 600
20 3% PLASTIC STRAIN
/", 500°C ANNEALING
(/l
Z
/ \.
<
a:
(!)
15
! \ / 2 8 0 0 MIN
u.
o .I
. ,.,
.......
''1''\
'..,.......670 MIN
w
(!)
10 j~-... ~ .~ 200 MIN
<
I- /-;~ ..... '~~40 MIN
Z
W ;' / / i .('\:\'"
t)
a: 5 ;": ,~,
w
Q. / . ..I
./
"",~ ,
...... .
.......~
o .-/ '.. ' .....
o 10 20 30 40 50 60 70 80
ROCKING CURVE HALFWIDTH (MIN OF ARC)
ALLOY 600
40 2% PLASTIC STRAIN
950°C ANNEALING
35
30
(/l
~ ~
<!
25 .:
a: c
t'J •
~
u.
0 20
W
t'J
<!
I-
z 15
w
u
a:
w
0-
10
20 40 60 80 100 120
ROCKING CURVE HALFWIDTH (MIN OF ARC)
TEMPERATURE (oG)
/o~O~
w
~
I-
100
~'"' 073.00 ""Imo"
0 40 % :67.58 kcal/mole
0 50 % :64.08 kcal/mole
unstrained state very rapidly. For an anneal at 950°C, the deformed state
was virtually fully recovered without any recrystallization taking place.
The results shown above indicated that the activation energy decreases as
the recovery proceeds. This implies that more energy was required for
strain release at the earliest stages of the recovery process.
~~
0
i 30
"
C!l
:i
b 20
f--f--'~~.
~
u.
«
-J
J: 10
--.0
o 10 20 30 40 50
DEGREES FROM APEX
Fig. 14. Microstrain relaxation on the surface of an Alloy 600
C-ring following SCC exposure. The sample was heat
treated to produce no GB carbides.
that the strains decreased along the periphery as the angular position from
the apex increased. Prior to straining,the sample was heat treated at
1070°C for 60 minutes and water quenched to maximize the potential for
stress-corrosion cracking (sensitized condition). When the strained C-ring
was exposed to a 10% mixture of NaOH and KOH for 1000 hrs, the CARCA
measurements disclosed a total strain relaxation over the range of 0 to 0
Fig. 15. Micrograph of apex region of Alloy 600 with extensive SCC
cracking. Sample was heat treated to produce no GB
carbides and,exposed to 10% caustic for 1000 hours.
C. F. LO ET AL. 47
1MM
o
~,--------------------------------------------------------,
HEAT# 2609
o
N
o
ci~----------r----------.-----------r----------r---------~
0.0 5.0 10.0 15.0 20.0 25.0
DIFFERENCE OF HALFWIDTH (MIN OF ARC)
' .. ' )
..
__ ... I
,
~
., :~ -~'
.
~ -. _ V ' ._
- " .... \
r I'-,-".
oJ
...- ; • yr '
.~ ,
,.
.... , • i
-- ·O.2mm
Fig . 18. Micrograph of apex region of Alloy 600 with no observable
SCC cracking. Sample was heat treated (704°C/1S hr) to
contain a significant number of GB carbides .
halfwidth of 21.5 min of arc. Further away from the apex there was much
less cracking with a concomitant decrease in the ~~ . In the region of 20°-
30°, the total crack length was only 2.4 mm with a ~ decrease of 2.5 min of
arc. Thus, there appears to be a mechanism operative whereby crack growth
promotes strain relaxation . A second sample was prepared to explore more
fully the above effect. l~is sample, however, was heat treated at 704°C
for 15 hr to produce a substantial number of grain boundary carbides. In
this heat treated condition the alloy is known to be much more resistant to
SCC than the sample described above . In fact no microcrack could be
detected in the metallograph of Fig. 18 when the strained sample was ex-
posed to the caustic solution. The X-ray results of Fig. 19, however, gave
clear evidence that significant microstructural changes have occurred even
though no visible cracks have been initiated . There was a substantial drop
I-EAT#2609
1070·C/2OMN + 704°C/1SHR
50 '" C-RNG CLOSURE
• C·RNG CLOSlH: +
1000 ~ IN 10% CAUSTIC
O'--~--'L.---'---L-""""-"""'-
o 10 20 30 40 50
DEGREES FROM APEX
Fig. 19. Microstrain relaxation on the surface of an Alloy 600
Coring following SCC exposure. Sample was heat treated
(704°C/1S hr) to contain a significant number of GB
carbides .
C. F. LO ET AL. 49
DISCUSSION
Correlation of Workhardening Characteristics with Micro and Macrostructure
will be seen that the slOF~S of the stress-strain curve for the fine-
grained Alloy 600 changes slightly after € = 0.7% and subsequently leads
gradually over from n = 0.109 to n = 0.343. The '1-plot of Fig. 5 however
is more explicit on the course of hardening with strain increase. By its
precipitous plunge after € = 1.5%, it shows that a stage of increased hard-
ening has occurred and that the hardening rate decreased at higher strains.
In the stress-strain curve of the macrostructure, on the other hand, the
successive group effects of microstructural grain hardening are averaged
out. If indeed one were able to resolve these effects the stress-strain
curve would have serrated features.
ACKNOWLEDGMENTS
The authors are grateful to Elsevier for their kind permission to use
several of the figures in this paper. The authors are also grateful to
EPRI for financial support of this work.
REFERENCES
[5] H.Y. Liu, W.E. Mayo and S. Weissmann, Mapping and Analysis of
Microplasticity in Tensile-Deformed Double-Notched Silicon Crystals
by Computer-Aided X-ray Double-Crystal Diffractometry, Mat. Sci. and
Eng., 63:81 (1984).
[8] C.F. Lo, H. Kamide, G.F. Feng, W.E. Mayo and S. Weissmann, X-ray
Determination of Stress-Strain Distribution in Unnotched C-rings: A
Link from Micro- to Macromechanics, Acta. Metall. 36:3069 (1988).
[12] C.L. Briant and S. Banerji, Intergranular Failure in Steel: The Role
of Grain Boundary Composition, Int. Metall. Rev. 23:164 (1978).
[15] S.M. Bruemmer and C.H. Henager, High Voltage Electron Microscopy
Observations of Microdeformation in Alloy 600 Tubing, Scr. Metall.,
20:909 (1986).
[16] S.M. Bruemmer, L.A. Charlot and C.H. Henager, Microstructure and
Microdeformation Effects on IGSCC of Alloy 600 Steam Generator
[17] W.A. Johnson and R.F. Mehl, Reaction Kinetics in Processes of
Nucleation and Growth, Metals Technology 6:5 (1939).
Department of Engineering
University of Warwick, UK
ABSTRACT
INTRODUCTION
Double-axis x-ray diffraction methods may be used to detect strains in the range
to- 5 to to- 6 , with depth resolution down to "- 50 nm, near the surface of single-
crystal semiconductor materials.
The effect of using different silica colloids and changing the polishing conditions is
also being investigated and silicon wafers from various manufacturers are being
compared.
EXPERIMENTS
The experiments were performed using a Bede Scientific Instruments model 450
diffractometer mounted to give a vertical dispersion plane. The smallest obtainable
step size of the diffractometer fine axes was 0.05 arc seconds. This was confirmed
experimentally by measurement of silicon 880 rocking curves of intrinsic width 0.5 arc
seconds. The x-ray generator was a Hiltonbrooks model DG2 run at 45 kV, 35mA
with a copper target.
The 400 reflection was used with CuKQ! 1 radiation, gIVIng a surface symmetric
reflection with Bragg angle 8 = 34.56·. The four-reflection beam conditioner was
channel cut from a single block of silicon, with a strain-relief cut for mounting, and
etched to remove sawing damage. The experiment was set up in the (+, -, +, -, +)
parallel geometry (figure 1) which is non-dispersive in wavelength. A 0.5 mm diameter
collimator was used, which gave a beam size of 0.5 x 4 mm 2 at the sample, due to
horizontal beam divergence. Care was taken to reduce the background as much as
possible without intercepting the beam, by placing narrow slits and lead perspex
shielding between the two axes.
collimator
x-ray
source
Fig. 1. Double axis diffractometry. Experimental set-up.
L. HART ET AL. 57
0.8
............ 0.6
°5
~... 0.4
0.2
0
-40 -20 0 20 40
arc seconds
Fig. 2. Instrument function measured using two
four-reflection beam conditioners.
Theory - Experiment ..
Figure 3 shows rocking curves from a rough - polished wafer which was then
etched to depths of 1 pm and 2 pm to remove strain and damage. There is no
significant difference in the rocking curve shapes produced by the 1 pm and 2 pm
etching treatments, which implies that most of the residual strain is in the top micron
below the surface. The residual strain level is greatly reduced in fully-polished wafers.
A rocking curve from such a wafer is shown in figure 4 which compares rough
polished, fully -polished and etched -wafer rocking curves. The plots are shown on a
semi-logarithmic scale.
101~--~----~----~----~ 101~--~~--~----~----~
-100 -50 o 50 100 -100 -50 o 50 100
arc sec arc sec
Rocking curves from polished silicon wafers - 400 reflection.
Fig. 3. Rough polish - Fig. 4. Rough polish -
1 11m etch Full polish
2 11m etch .. 1 11m etch ..
58 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
Most of the scattering in the tails appears between 5 and 60 arcs secs, i.e. 1 to
15 rocking-curve widths from the Bragg peak maxima, with more intensity on the low
angle side of the peak, implying a lattice dilation at the surface.
Experimental data from an etched silicon wafer was compared with a theoretical
simulation of a perfect silicon crystal rocking curve, shown in figures 5 and 6. A
background of 0.001, chosen by matching the background at 100 arc seconds away
from the peak, has been added to the theoretical simulation. This is ascribed to
Compton and thermal diffuse scattering.
A number of different strain profiles was simulated in order to find any possible
agreements with the experimental data. A positive mismatch was assumed, due to the
asymmetrically increased intensity on the low- angle side of the rocking curve.
Mismatch may be defined as follows:
'experimental' mismatch m* = a-
lia
=
lid
~ = - lie cot e
where al and as are the lattice parameters of the layer and substrate, e is the Bragg
angle, and v is the Poisson ratio. In the simulations, lattice mismatch was varied
between 3 and 20 parts per million, with total layer thicknesses from 0.5 to 20
microns, assuming maximum strain at or near the surface.
In the first instance, comparisons were made with data from rough-polished wafers.
The best fit was given by a non-linear graded mismatch of 8 ppm near the surface,
decreasing to zero at 1 pm below the surface (figure 7). This was divided into 20
layers of 50 nm each, assuming a constant strain across each layer. The model shows
10°
0.8 10-1
;;., ;;.,
.1:: .1::
.=>
0
0.6
.=>
~ 0.4 ~
~ ~
0.2
0 10-4
-100 -50 0 50 100 -100 -50 0 50 100
arc seconds arc seconds
Comparison of experiment with theory for a perfect silicon crystal.
Theory - Experiment ..
Fig. 5. Linear scale Fig. 6. Logarithmic scale
L. HART ET AL. 59
w
U
0-
~ Z
-'
O~--~ __ ~ __ ~~~~
~o ~z ~. ~. ~. ~o
a very low strain profile at the surface, nsmg to a maximum about 50 nm below.
Models with maximum strain at the surface itself consistently showed poorer agreement
with experiment. The resulting simulation is shown in figure 8, compared with a
perfect-crystal simulation. The low-angle tail of the curve shows increased intensity,
while the high-angle intensity is reduced. This asymmetrical shape was typical of all
the simulations tried, assuming a positive mismatch at the surface. Other effects
which change the peak shape include sample curvature, surface orientation and surface
roughness. These effects were eliminated by comparing rocking curves from the same
point on the sample before and after etching. Initial comparisons for fully- polished
wafers place an upper bound for the strain level at 4 ppm, with similar depth profiles.
A good fit to the experimental data may be obtained by adding intensity to both
tails of the theoretical strained -layer curve. This could be explained by diffuse
scattering from atom disorder or defects near the surface, which would increase the
intensity at the rocking curve tails on both sides of the peak (Thomas, Baldwin,
Dederichs, 1971; Dederichs, 1971).
100 ~----.---~.----,----~
10-1
10-3
104L---~~--~----~------'
-100 -50 o 50 100
arc seconds
Fig. 8. Theoretical simulations comparing perfect crystal
with model strained crystal.
Perfect crystal - Strained crystal --
60 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
investigating theoretical models to explain the mechanism by which these effects are
produced.
It has been shown (Andrews and Cowley, 1985; Robinson, 1986) that diffuse
scattering may arise from an abrupt truncation of charge density at a crystal surface
and this has been measured from polished (111) silicon wafers (Harada et ai, 1983,
1987). An asymmetry, implying lattice dilation, was also measured, in agreement with
the results presented here.
CONCLUSIONS
Double-axis x-ray diffractometry may be used to detect very small strains (mis-
matches of a few parts per million) down to - 50 nm below the surface.
This technique may be used to evaluate various stages in the silicon wafer
polishing process.
REFERENCES
Mary A G Halliwell
INTRODUCTION
layer
a unit cell
~
U substrate
unit cell
which depicts a layer with bulk lattice parameter ar grown on a substrate with lattice
parameter a . In the ideal case illustrated it is assumed that the layer unit cell remains
tetragonal w?th cell edges parallel to those of the substrate unit cell. Then the fractional
relaxation is defined by:
(1)
where a is the lattice parameter of the partially relaxed layer unit cell parallel to the
interface.
R equal to zero indicates that the layer unit cell is strained to exactly match the
substrate unit cell parallel to the interface. R equal to unity implies total relaxation with
the layer unit cell acquiring cubic symmetry. This is rarely achieved in practice. When
the lattice parameter difference between the layer and substrate is large, complete
relaxation can occur during growth at an elevated temperature. Measurements made at
room temperature frequently show a small residual strain which results from the
difference in thermal expansion of layer and substrate as the sample is cooled. This will
lead to a measured relaxation greater than unity for a layer which has a higher
expansion coefficient and higher lattice parameter than its substrate.
R can be determined if the dimensions of the partially relaxed unit cell are
measured. This can be done by recording rocking curves from reflecting planes which
are inclined to the (001) substrate surface. Measurement of a number of partially
relaxed samples has indicated that relaxation is seldom ideal. As will be described
below, all three axes of the layer unit cell may be tilted with respect to the axes of the
substrate unit cell.
The [001] vectors in the layer and substrate may also be non-parallel. A possible
mechanism for both group IV and group III-V semiconductor materials is the case of a
substrate surface slightly tilted with respect to the (001) plane. The substrate surface will
then have atomic steps which could preferentially favour the nucleation of dislocations
along one of a pair of slip planes rather than the other. An assumption which can still
reasonably be made in circumstances when there are possible differences in dislocation
densities on the four {Ill} slip planes is that the [110] and [110] directions in the
layer and substrate remain parallel. This assumption is used in the next section to
develop a general method of deriving the layer unit cell dimensions from diffraction
data.
(001)
substrate
TABLE 1
NB all values in seconds of arc for Copper K.a radiation
of the 001 point can be calculated. From these points it is possible to calculate dx and dz
(=c), the d spacings of the (110) and (001) planes for the layer. If the procedure is
repeated for the hlil and 11hl reflections a second value for dz can be obtained
together with a value for dV' the d spacing for the (110) planes for the layer. Thus the
complete symmetry of the unit cell can be determined from a minimum of eight rocking
curves. The measurement can be repeated for different hhl reflections to reduce the
experimental error. The next section considers the best choice of reflections for
determining relaxation.
Before the relaxation can be calculated, the bulk lattice parameter of the layer,
ar , is required. This will be known if the layer is silicon, germanium or a binary III-V
semiconductor. Otherwise the lattice parameter of the layer in bulk form will have to be
calculated from the distorted unit cell dimensions using elastic theory.
CHOICE OF REFLECTION
Table 1 shows the peak separations (S) which would be observed in rocking
curves recorded from unrelaxed layers with lattice parameters differing by 1% from
their substrate~ (normally referred to as a 1% mismatch), together with the change in
peak separation (dS) which occurs for a relaxation of 1% as given by equation (1). Data
is given for four of the most common substrate materials - silicon, germanium, gallium
arsenide and indium phosphide. The S value for a particular reflection is proportional to
the value of (c-a) for the tetragonally distorted layer cell while dS is P!opprtional to
R(c-a). For silicon and germanium the data are for silicon_germanium alloy layers with
the appropriate mismatch. For the III-V substrates the data are for gallium indium
arsenide heteroepitaxial layers. Because Poisson's ratio is smaller for Si and Ge than
for the III-Vs (Brantley, 1973) there is a significant difference in the S values for
germanium and gallium arsenide even though the lattice parameters of the two
materials are very close.
The 002 and 004 reflections have been included in the table for completeness.
However since they only involve the c parameter of the unit cell they cannot be used to
determine relaxation unless the layer is of known lattice parameter, as for example in
the case of a layer of GaAs on Si. The change of separation with relaxation is small but
the reflections are much simpler experimentally and may be appropriate to use in some
circumstances. In this case four rocking curves should be recorded with the x-ray beam
M. A. G. HALLIWELL 65
parallel to [011] directions, with 90° rotation about the surface normal between each. In
this way the presence of a tilt between the layer and substrate can be revealed. It will not
be possible to determine if there is a difference between dx and dy for the layer.
The reflections which are most useful for determining relaxation are those for
which the ratio dS/S is greatest. The 113 reflection appears a good choice for all
materials. However for InP the value of i for the L beam path is less than 10 and its use
is not straightforward. In this case, and for other substrates with larger lattice
parameters the 224 reflection is the best choice. The minimum relaxation that can be
measured will depend on the minimum peak movement that can be determined. This
will be instrument dependent. A well-aligned double-a.xis diffractometer can give typical
peak widths of 10 to 20 seconds of arc. It should be straightforward to measure peak
shifts of one half the peak width and hence a 1% relaxation in a layer with 1% mismatch
should be measurable on a routine basis. Because the peak movement is proportional to
(c-a)R, or the product of mismatch and relaxation, this means that 0.5% relaxation
should be measurable in a layer with 2% mismatch. The relaxation is equivalent to a
lattice parameter difference of 100 ppm (parts per million) parallel to the interface.
This method can also be used for measuring the relaxation occurring in
superlattice stacks consisting of alternating layers of two semiconductor materials. In
this case the relaxation is defined in terms of the mean lattice parameter of the stack. By
averaging over several reflections it was possible to measure relaxations equivalent to a
difference in lattice parameter of 20 ppm parallel to the interface (Halliwell et aI, 1989).
DEFINITIONS OF RELAXATION
The only established definition of relaxation is given in equation (1). This only
applies to the ideal case. Two possible definitions for the non·ideal case should be
considered:
Strain relief parallel to the < 110> directions in the inteiface. This would be the preferred
method when the asymmetry is large, particularly in circumstances where special efforts
have been made to achieve uniaxial distortion (Tuppen et aI, 1989). In this case if the
bulk lattice parameter of the layer is '!T and the d spacing parallel to the two < 110>
directions are dx and dy then the two values of the relaxation could be expressed as:
ar = a + (c-a)(l-v)/(l+v) (3)
The approximation involved in assuming tetragonal distortion should be small. The a
parameter obtained can then be substituted into equation (1) to give a value for the
extent of relaxation. In this case an R value of unity does not necessarily imply a strain-
free layer. It does imply that the c parameter of the layer is equal to the bulk lattice
66 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
parameter but the layer could still be in compression parallel to one < 110> direction
and in tension parallel to the other < 110> direction in the plane of the layer.
For the case of binary group IV, or ternary or quaternary group III-V
heteroepitaxiallayers where the composition, and hence ar' is not precisely known, then
asymmetric reflections will have to be used so that all three layer unit cell parameters
can be determined. In this case a=(d}+d})O.5 can be substituted into equation (1)
together with ar to obtain a mean value lor die relaxation.
SUMMARY
ACKNOWLEDGEMENTS
Thanks to my colleagues at BTRL, Chris Tuppen and Michael Lyons for many
discussions, and in particular to Mark Hockly who also supplied the electron microscopy
results. Acknowledgement is made to the Research and Technology Board of British
Telecom for permission to publish this paper.
REFERENCES
Brantley, W.A., 1973, Calculated elastic constants for stress problems associated
with semiconductor devices, J. App!. Phys. 44:534
Halliwell, M.A.G. 1981, X-ray measurements of lattice mismatch in
heteroepitaxiallayers, Inst. Phys. Conf. Ser. 60:271
Halliwell, M.A.G., Lyons, M.H., Davey, S.T., Hocldy, M., Tuppen, e.G., and
Gibbings, C.J., 1989, Estimation of percentage relaxation in Si/Si1_ Ge x
strained-layer superlattices, Semicond. Sci. Technol. 4: 10 x
Nagai, R., 1981, Dislocation velocities in indium phosphide, lpn. J. App!. Phys.
20:793
Tuppen, e.G., Gibbings, C.J., Rockly, M. and Halliwell, M.A.G., 1989,
Asymmetric strain distribution produced by the preferential nucleation of
misfit dislocations, submitted to Appl. Phys. Lett.
CHARACTERIZATION OF STRUCTURAL INHOMOGENEITIES IN
GaAs/AlGaAs SUPERLATTICES
ABSTRACT
INTRODUCTION
DIFFRACTION MODEL
Kinematical diffraction theory is an approximation of
dynamical theory and is an adequate model when the maximum
reflecting power is low (i.e. for relatively thin layers). The
treatment of Speriosu and Vreeland,4 for example, is quite
successful for SL analysis, and it provides the basis for the
model used here. Each period of the SL is assumed to diffract
independently, and the transition between the sublayers is
assumed to be sharp. The intensity calculated for any given
orientation is, therefore, the sum of the intensities
diffracted from each layer. The modified kinematical
diffraction model used in this paper extends the earlier
approach by correcting for normal absorption and extinction
(e.g. Batterman and cole 12 ) . Note that diffracted intensity
from both the substrate and buffer layers are explicitly
included. The accuracy of the modified kinematical diffraction
model should be comparable to that of dynamical diffraction
treatments for total layer thicknesses considerably greater
than 2 J.4m. It should be noted, however, that dynamical
diffraction must be used for more complex layered structures
where large lattice-parameter gradients are present or where
layers reach 5-10 J.4m total thickness. Another experimental
feature which should not be expected to be reproduced by the
kinematical model is the set of subsidiary maximum associated
with each satellite reflection.
EXPERIMENTAL PROCEDURES
The undoped superlattices of GaAs/AIGaAs were grown in a
VARIAN GEN II system on two degrees-off (001) GaAs substrates.
The nominal Al concentration in the AIGaAs layers was x=O.35,
P. C. HUANG ET AL. 69
Periodicity of SL
simulations of SL
10'
100
10-'
1;-
.iii 10- 2
C
2
..£ 10-3
Q)
>
:p
0
Q)
0:::
10-5
10--6
b
10- 7
-1500.0 -500.0 500.0 1500.0 -2000.0 -666.7 666.7 2000.0
W, (are seconds) W, (are seconds)
10'
10--6 E-_-~
C
10-7~------~--------~------~
-1500.0 -166.7 1166.7 2500.0
W, (are seconds)
Period Variation of SL
o (+1)
t:.. (+2)
-.-(+4)
o (+6)
a b
10~~------------~----------------~--------~~------------~----------------~------------~
0.00 333.33 666.67 1000.00 0.00000 0.33333 0.66667 1.00000
Thickness of AIGaAs Layer, W AI Cone. of AIGaAs Layer
40.0 r----------.....,
• expo • expo
A cal.&O"=O.% A cal.&0"=O.2%
10° ------------- -------------
c cal.&0"=1.%
o cal.&O"=O.5% r0-
~ cal.&0"=2.% t> 30.0
.....>.
Q)
'00 10-
1 C/l
c ~
.....Q) S
.£ 10-2 2 20.0
Q)
I
>
:,::; 3:
0 LL
V 10-
3
~
a: o
Q) 10.0
0...
10- 4
CONCLUSIONS
A modified kinematical model, which corrects for normal
absorption and extinction, is used to interpret the rocking
curves of GaAI/AlGaAs SL's of moderate thickness. The
experimental results could be matched closely using a two-
sublayer structure with sharp interfaces and slightly varying
periodicity. The difference between expected structure and
that from the rocking curves and simulations illustrates the
importance of x-ray diffraction during commissioning of MBE and
other thin film growth apparatus.
ACKNOWLEDGEMENTS
The authors wish to thank Polaroid Corporation for their
support of the MBE growth runs and Mr. Kenneth Maloney for his
support and encouragement of this work.
REFERENCES
1. L. Esak, IEEE J Electronics, 22: 1611 (1986).
2. N. Ho1onyak, Jr., R.M. Kolbas, W.O. Laidig and B.H.
Vojak, J ~ Phys, 51: 1378 (1980).
3. W.T. Tsang, ~ Phys Lett, 38: 204 (1981).
4. V.S. Speriosu and T. Vreeland Jr., J ~ Phys, 56: 1591
(1984) .
5. P. Auvray, M. Baudet and A. Regreny, J ~ Phys, 62: 456
(1987) .
6. J.M. Vandenberg, M.B. Panish, H. Temkin and R.A. Hamm,
~ Phys Lettr, 53: 1920 (1988).
7. J.M. Baribeau, Song Kechang and K. Munro, ~ Phys
Lettr, 54: 323 (1989).
8. P.F. Fewster, J ~ Cryst, 21: 524 (1988).
9. A.T. Macrander, G.P. Schwartz, and G.J. Gualtieri, J
~ Phys , 64: 6733 (1988).
10. S.J. Barnett, G.T. Brown, D.C. Houghton and J.M.
Baribeau, ~ Phys Lettr, 54: 1781 (1989).
11. B.K. Tanner, this volume
12. B.W. Batterman and H. Cole, Rev Mod Phys, 36: 681 (1964).
13. W.J. Bartels and W. Nijman, J Cryst Growth, 44: 518
(1978) .
14. J. Hornstra and W.J. Bartels, J Cryst Growth,44: 513
(1978) .
15. P.F. Fewster and C.J. Curling, J ~ Phys, 62: 4154
(1987) .
16. C.W. Wie, J ~ Phys, 66: 985 (1989).
17. M. Ettenberg and R.J. Paff, J ~ Phys, 41: 3926 (1970).
18. E. Estop, A. Izrael and M. Sauvage, Acta Cryst, A32: 627
(1976) •
19. M.C. Joncour, M.N. Charasse, and J. Burgeat, J ~ Phys,
56: 1591 (1984).
20. P.C. Huang and S.R. Stock, simulation similar to C.W. Wie,
T.A. Tombrello and T. Vreeland, Jr., J ~ Phys, 59: 3743
(1986) •
21. D. Chrzan, J ~ Phys, 59: 1504 (1986).
X-RAY DIFFRACTION ANALYSIS OF SiGe/Si SUPERLATIICES
IN1RODUCfION
It is well known that x-ray diffraction techniques provide one of the most
direct means of analyzing superlattice periodicity, layer thickness, strain, and
defect structure. This suitability of x-rays is a consequence of the fact that most
superlattice periods are in the range of 10 to 50 times the typical wavelength
(CuKa 1 =1.54 A) and are hence measurable by x-ray diffraction, and also that x-
rays are non-destructive within the energy and dosage ranges needed for these
studies. Applications of x-ray techniques to superlattices have been discussed in
many publications in the open literature. 1- 8
In general, the validity of experimental data rests upon two broad groups of
factors, namely (a) the superlattice and sample preparation procedure, and (b)
the instrumental and physical constraints of the diffraction technique employed.
These two sets of factors will be considered in the following sections, and
comments and observations will be provided on obtaining interpretable data, free
from ambiguities.
Si(400)
(Si-Ge):(Si) on Si
_ 4 SG35V2H
CI SLiOOL)
.9
1++++++++++++~~~~r-~~~~~~~~++++~
60 62 64 66 68 70
ANGLE 28, (degrees)
Fig. 2 Superiattice peaks about the Si(004) Bragg reflection in the SiGe-
Si/Si system.
on GaAs, and would hence require direct low-angle Bragg reflection measurement
appropriate to their periodicity starting, typically, with order n=1.
The angular positions of the satellites enable the periodicity of the super-
lattice to be calculated and compared with the epitaxial growth parameters. Com-
putational methods can also be used to fit the experimental data in terms of rela-
tive chemical composition and interface roughness (see below).
It is generally assumed that the average SL peak coincides with the main
substrate peak, since the two peaks cannot be resolved on a single crystal
diffraction pattern. Using high resolution x-ray double-crystal diffractometry,
however, the small but finite separation between the two peaks becomes more
easily detectable. This angular separation can be used for an estimate of the aver-
age composition of Al in AIAs/GaAs or Ge in SiGe/Si superlauices. 4
PROBLEM AREAS
Sample Preparation
In general, there are variations across the wafer in the superlattice pa-
rameters including the periodicity, composition and layer thickness caused by
flux variations during the epitaxial deposition. These variations introduce addi-
tional broadening and spurious detail in the x-ray diffraction profile, which in
turn may complicate the analysis. For this reason alone the choice of smaller
samples over the larger ones is preferred. However, as described below there are
also other considerations for choosing smaller samples.
I
I,
,
I
,
(B)
PLATE
double diffractometer simply because the former uses a much larger beam size
and sample area. Associated with this problem is the manner in which the sample
is affixed to the mount several times during the course of the investigation. It has
been found that practically any adhesive causes some degree of strain and may
unpredictably change the relative alignment of the beam with respect to the
sample. Fig. 3 shows a sketch of such an occurrence for a sample which for
simplicity has been assumed to have been bent into two flat regions so that either
one or the other may be oriented at the proper diffraction angle with the detector
set at the angle 29 B .
"'I:::=' 6E5
0 SG35/ l2xl2mm
i:l Original paraffin mount
...
~=
<Il
5E5 Paraffin remelted
&
~ 4E5
U
6
i-
.;;:: 3E5
J 2E5
l
4)
>
.~
<l 1E5
Il:::
0
J
1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0
Detector Angle, 29, degrees -+
Fig. 4 Powder diffractometer scan of the first superlattice peak about the
(000) reflection. The measurements were taken on a 12mm x 12mm
sample, as a function of wax mounting.
M. FATEMI ET AL. 79
6E5~--------------------------------,
0.5 0.6
j
0.7 0.8
~"
0.9 1.0 1.1 1.2 1.4
1.3
Specimen Angle 8, degrees
Fig. 5 Powder diffractometer 8 -scan of the first superlattice peak about
(000). The sample and mounting are identical to those of Fig. 4, but
the detector angle 28 is fixed.
.~ 1.2E5
c
.s
..s
d)
8.0E4
>
~
';:l 4.0E4 SG35
~ t-- '.
( ...
0.0
1.0 2.0 3.0 4.0 5.0
Detector Angle, 29, degrees ~
Fig. 6 Low-angle 9-29 scans (about (000» of the first three harmonics of
the SiGe/Si superlattice, showing the effect of a 100A Ni cap on the
scattering in tensi ties.
80 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
8E3 I,
II
"
OJ
I
-g 6E3 I
0 I I
~ I I
...'" ,
I I
I
8. I I
j I
II
I
I
I
20 arc-sec
1------1
U I
.~ 4E3 Before Anneal I
I
I
j
I
I I
I
I
0
I After Anneal
.~
os 63toC/30 min
'il
~
2E3
I
\
I
\
\
\
" 20
60 40 20 0 40 60
.6.e - arc-sec
Fig. 7 (004) rocking curve from the substrate in the SiGe/Si superlattice
system (004) before and after an anneal at 631°C.
Numerical Calculations
CONCLUSIONS
IE7 SiGe:SiJSi
#SG3S simulation
IE6
.~
c:
£ IES
oS.,
>
·w
lFA
U"
II::
IE3
IE2
S9 61 63 6S 67 69 71
Detector Angle, 2&, degrees
REFERENCES
1. L.L. Chang, L.Esaki, W.E. Howard, and R. Ludeke, J. Vac. Sci. lQ., (1973) 11.
3. R.M. Fleming, D.B. McWhan, A.C. Gossard, W. Weigman, and R.A. Logan, J.
Appl. Phys. n, (1980) 357.
6. M.C. Joncour, M.N. Charasse, and J. Burgeat, 1. Appl. Phys. 58, (1985) 3373.
7. Y. Kashihara, T. Kase, and J. Harada, Jap. 1. Appl. Phys . .6., (1986) 1834.
11. M. Born and E. Wolf "Principles of Optics," 6th Ed., Pergamon Press. New
York (1980).
mGH RESOLUTION MEASUREMENT OF SURFACE MISORIENTATION IN
SINGLE CRYSTAL WAFERS
M. Fatemi
ABSTRACT
INTRODUCTION
EXPERIMENTAL CONCEPr
Consider a Si wafer whose (100) planes are misoriented with respect to the
surface at an angle cp, oriented in such a way that the incident beam and the
normal vectors to the surface and to the diffracting planes are co-planar. The
(004) rocking curve from the wafer can thus be obtained and the peak angle
recorded. If the crystal is rotated in its own plane by 180° and the process is
repeated, the position of new rocking curve peak will be displaced from the old by
twice the misorientation, and the angle </l is, therefore, immediately obtained,
even though the exact diffraction angle itself is not known. The basic reversal
concept has been used in several applications by various workers. 2 - 4 For
example, Fewster 3 has applied a related technique to the determination of lattice
parameters of epitaxial layers on substrates of similar composition. In that
application, however, the surface misorientation was subtracted out, and only the
difference between the substrate and epilayer peaks was needed. In the present
application laser alignment is used to define the direction of the crystal surface,
while x-rays are used to determine the direction of the diffracting planes.
In order for the concept to be correctly applied to the present problem, two
important conditions must hold. First, it is essential that the azimuthal rotation of
the crystal in its own plane (the 0)- motion) be reproducible within a few arc-
mins, even after several rotations about the azimuth axis. Second, the O)-axis must
be centered on the region of the crystal illuminated by the x-rays so that the same
area of the crystal remains in the beam at all azimuths. The most detailed part of
instrumental preparation, a one-time process, deals with the initial permanent
adjustment of the omega axis and the tilt mechanisms as required for all double-
crystal instruments. This alignment also removes ambiguities that often arise
when the diffractometer is used for spectrometric studies.
INSTRUMENTAL MODIFICATIONS
The design of the second stage (Fig. 1) was based on the premise of opera-
tional simplicity without introducing misalignment errors due to mechanical
instability. It is for this reason that the crystal translations in both lateral and
vertical directions are controlled by the first and the last mechanisms, i.e., the
main lateral translation (part #1) and the vertical sample translation (part #8),
respectively. The lateral translation stage supports the entire weight of the
second stage and can be used to select various test points of the sample along a
horizontal strip in line with the x-ray beam. On the other hand, translation along
the vertical line is accomplished by the manually-operated micrometer (part #7).
M. FATEMI 85
Fig. 1 Exploded diagram of the components controlling the motion and align-
ment of the second crystal. See text for a description of each part.
Attached to the main lateral stage is the main vertical slide (part #2). This mech-
anism is used in the preliminary alignment of the instrument to compensate for
changes in the height of the crystal during the adjustments in the tilt of the 00-
axis. Next, the foreaft translation stage, part #3, controls the position of the
crystal surface in the x-ray beam. The main 00- axis control, part #4, is carried by
the fore-aft slide and supports, in turn, the motor-driven 00 rotation mechanism
(part #5). The sample-centering translation slides (part #6) and sample tilt
controls (part #7) are part of a commercially-available eucentric goniometer.
Since the axes of the sample tilt controls must lie in the sample surface,
appropriate spacers are needed to compensate for the extra distance between the
outer face of the goniometer and the tilt axes. Some of this space is taken up by
the mounting plate and micrometer used to vary the sample height (part #8). It is
absolutely essential that prior to any of the alignment steps described below,
spacers of correct thickness are used with each wafer. We also assume that the
vertical a-axis of the second crystal stage has already been determined, that the
incident x-rays (from the first crystal) intersect this axis horizontally, and that
the tilt center, i.e. the intersection of the two tilt axes (part #7) is in line with the
co-axis. This last operation can usually be performed "on the bench" and requires
an iterative adjustment of sample translations (part #6) and rotation about the co-
axis.
Component Calibration
Prior to any alignment, the interaction among the three critical compo-
nents, i.e. the co - axis tilt control, the main height control, and the sample
goniometer tilt mechanisms must be determined and calibrated. The calibration
process, taking advantage of both x-rays and laser optics, is as follows.
Any change in the tilt of the crystal using the co - axis tilt mechanism
affects the height of the test area. The resultant shift must then be corrected
using the main height control. The change in the sample height as a function of
variations in the co -axis tilt may be calibrated by passing the incident x-rays
directly through a narrow slit at the sample position. The beam intensity is kept
constant by adjusting the height, while the co -axis tilt screw is turned in regular
increments. The height-vs-main tilt calibration can, therefore, be easily made.
However, since adjusting the main height does not change the crystal tilt, this tilt
may be reset to the original value using the sample tilt controls (on the eucentric
goniometer). In this fashion the readings of the sample tilt help calibrate the
angles associated with the co-axis tilt mechanism.
( a) (b)
• w
c=:I7--. ~
.~
LASER SOURCE
CROSS HAIRS CB-.
FRONT-SURFACE
MIRROR
2ND CRYSTAL
CB--,--~
Fig. 3 Initial laser optical alignment of the co-axis with respect to the second
crystal. (a) co =0° orientation. The laser spot is focused at point A on
"distant" cross-hairs. (b) With a rotation of the crystal to co=180°, the
laser spot moves to point B. (c) The beam is brought to the half-way
point C by readjusting the sample tilt goniometer.
M. FATEMI 87
the measurements, It IS important to limit the test area of the crystal to within 2-3
mm, using a ring of opaque material surrounding the region, centered on both
the co and the tilt axes. A fine laser beam is centered in the ring and is reflected
toward a front-surface mirror at a distance several meters away. For the optical
alignment described later, the beam is focused on "cross-hairs" mounted on the
wall at a convenient location as close to the crystal as possible. The double
reflection increases the sensitivity of optical alignment to changes in the crystal
position. It is also necessary to ensure that the laser beam remains within the
specified area on the crystal for all rotations about the co- axis. Thus, if the angle
co is changed by 180°, and a significant shift in the position of the laser spot with
respect to the ring is noted, the alignment of the region with respect to the co-axis
must be rechecked, as well as the position of the laser beam on the crystal.
Optical Alignment
4
(a 1 (b1
DIFFRACTING PLANES
¢-o:
n1
----j,::::,CX'
w=o I w=180· I
2ND CRYSTAL TILT: 2ND CRYSTAL TILT:
¢+o: DOWN ¢. a. UP
FOR ROCKING CURVE FOR ROCKING CURVE
~ (c 1
Fig. 4 X-ray alignment of the co-axis with respect to the spectrometer plane.
The horizontal dashed line is parallel to the direction of incoming x-
rays arriving from the right. The solid line below it denotes the co-axis.
(a) Rocking curve for co=O° is obtained with the crystal tilted downward
by cp+ a. (b) For the rocking curve at co = 180° the crystal tilt is corrected
upward by cp-a. (c) The main tilt is lowered by a = 1/2 the total
difference between (a) and (b). The entire assembly is then raised by
the amount of sample drop (h).
88 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
tP"'
<l>Vj
w=oo
.. n1
U
jf="' ..
U
"1
r<l>v
\=::,Q,
A
.", U
Fig. 5 When the x-ray alignment is complete, the diffracting planes remain
normal to the 00- axis at both azimuths of 0° and 180° with only one tilt
adjustment.
tionary on the cross-hairs, except for possibly a deviation on the order of 10-20
arc-sec which may be difficult to remove by this procedure. However, as seen
below, as long as the shift in the laser beam is visible, one may correct for its
effect with a simple analysis.
X-Ray Alignment
The oo-axis should also be made "parallel" to the horizontal plane of the
diffractometer. To perform the x-ray alignment the motorized crystal tilt controls
are adjusted to obtain a rocking curve from the test area. The azimuth is then
changed to 180 0 and the process repeated. The vertical tilt angle for 00=0° is
compared with that of 00 = 180°. One-half the total change is compensated by an
appropriate adjustment of the oo-axis tilt control. The sample tilt control is next
set at the average tilt reading for the two azimuths, and the change in the crystal
height is corrected as in Fig. 4. As before, the procedure is repeated to remove
possible residual errors. When this stage of alignment is complete, the crystal tilt
will be properly set for rocking curves at both 00=0° and 00=180° (Fig. 5).
Fig. 6 shows a simplified picture in which the laser source is first directed
at the crystal, oriented at "00=0°." For this configuration (A), the surface is drawn
perpendicular to the beam so that the reflected beam returns directly to the
source. In the actual process the reflected beam for 00 =0° is adjusted for the
center of cross-hairs; that is, the incident beam will not be perpendicular to the
crystal surface. Assuming that the 00- axis is shifted with respect to the crystal by
a small angle 0, a 180° azimuthal rotation causes a shift of 40 on the cross-hairs.
Were it possible to remove the misalignment, the crystal would be rotated
clockwise by an angle 0 into the oo-axis. The reflected beam in turn would shift by
M. FATEMI 89
CRYSTAL SURFACE
w=180°
\
(B)
A
B
B
+ LASER
SOURCE
20, to the midway position (point C in the Figure), and remain there for both co=O°
and co=180° positions. In the absence of that capability, however, one may still
accurately measure the angle of shift by observing the distance the laser spot
moves using the calibrated a -drive. Clearly, the angular shift of the x-ray
diffraction peaks follows a similar pattern. Hence, in configuration B the angular
reading of the a-scale would be too low by 1/2 the observed shift in the laser spot,
namely, by 20. The factor 2 should not be misleading, since the diffraction angles
at the two azimuths are averaged, thus only adding /) to the correction.
THE MEASUREMENT
With the co-axis aligned and the tilt axes contained in the sample surface,
the area to be examined is outlined using an opaque material such as photoresist.
The mounted sample generally is in the form of wafers whose front and back sur-
faces are nearly parallel. Hence, only small corrections in the surface orienta-
tion need be made. The laser beam will again be used to define the direction of the
surface. When first mounted on the sample plate, the wafer will shift the laser
beam both laterally and vertically with changing azimuth. The object is to adjust
the wafer so as to keep the spot laterally stationary on the cross-hairs. Vertical
deviations above or below the cross-hairs are not critical. In this mode, two
rocking curves at co =0° and co = 180° are obtained from the wafer. The peak sepa-
ration gives twice the component of the misorientation angle (4)h) in the horizon-
tal direction. The wafer is next rotated to the azimuth co=90°, and the procedure is
repeated, again obtaining two rocking curves after centering the laser beam lat-
erally. The new difference in the peak positions is twice the misorientation
component (4)u) in the vertical (90-270) direction. It is known that for small
(commuting) rotation angles the total misorientation angle is given by
90 I. CHARACTERIZATION OF EPITAXIAL THIN FILMS BY X-RAY DIFFRACTION
In a typical case, a 75-mm Si wafer specified as "on axis to within ±0.5°" was
measured by this technique. The components -Ph and CPu were found to be +1260
and +330 arc-sec, respectively, with the primary flat of the wafer placed verti-
cally on the right. Both numbers were within ± 10 arc-sec. Hence, the total angle
cP was calculated to be 1300 arc-sec in the direction of 14.7° from the x-axis. Note
that an accuracy of 1-2 arc-sec could have been obtained with only a 1/2-hour
additional effort. To check the reproducibility, the wafer was rotated by 15°
clockwise in order to bring the misorientation vector in the horizontal plane.
The measurement was repeated, and a total misorientation of l300 arc-sec was
obtained along the new horizontal line.
CONCLUSIONS
ACKNOWlEDGEMENTS
The author thanks Mr. M.C. Geiger for assistance in the x-ray measure-
ments, Mr. D.F. Peters for his skillful fabrication of parts for the modified second
stage, and Ms. Mary Daley for typing the manuscript.
REFERENCES
1. K.C. Hsieh, M.S. Feng, and G.E. Stillman, Proc. Materials Research Soc., 116,
261 (1988).
T. C. Huang
ABSTRACT
I:\,TRODUCTIO~
The study of surfaces and ultra-thin films is a growing and important application of the
X-ray diffraction technique (XRD). The conventional 0-20 scanning or symmetric
Bragg-diffraction geometry is not suitable for characterizing this type of materials because
the X-ray path lengths through a surface layer or ultra-thin film are too short to produce
adequate diffraction intensities relative to the bulk or the substrate. The path length in
a film can be increased or the penetration depth of the X-rays-can be decreased by an
order of magnitude using a fixed incident angle of a few degrees such as that used in the
Seemann-Bohlin geometry.1 To further limit the incident X-rays to the top surface layer,
a grazing-incidence geometry with an incident angle of less than one degree is required.
Several techniques which utilize grazing incident X-rays for studying surfaces and thin
films have been reported. For example, the reflectivity/interference technique has been
used successfully for determining density, smoothness, and thickness of thin films.2-4 The
grazing-incidence diffraction method has been used to study the in-plane crystallography
of epitaxial films.s The grazing-incidence scattering technique has been used extensively
for the surface structure determination of two-dimensional solids. 6 ,7 The grazing-incidence
geometry has also been used for surface and near-surface X-ray spectroscopic studies of
trace elements (in the ng!g range),8-IO elemental depth-profiling,II.12 short range order /
local atomic structure,13 surface oxidation and corrosion,14 etc.
Recently, a GIABD technique has been developed for the determination of crystalline
phases on surfaces and structural depth-profiles in thin films. IS -18 In this paper, the basic
principles, the experimental techniques, and the results of characterizing iron-oxide thin
films are discussed.
THEORETICAL BACKGROUND
For X-rays, the refractive index of most materials IS slightly smaller than one, the
refractive index of air, and can be calculated from
n = 1 - t5 - iP = 1- (1 ),
where No is the Avogadro's number, e the electronic charge, me the electronic mass, c the
velocity of light, Z the atomic number, A the atomic mass, p the density, ,.! the X-ray
wavelength, and J-t the X-ray linear absorption coefficient.
When an X-ray beam incident from air onto a material with an index of refraction less
than one, there is a critical angle (a c = j2i) below which X-ray total reflection occurs.
Typical values for the critical angle are in the 0.2° - 0.6° range for"! = 1.54A. At incidence
angles less than a c' the incident X-ray beam is evanescent within the material and pene-
trates only the top 100A or thinner layer. As the incident angle a increases and ap-
proaches a c' the penetration depth increases rapidly and reaches that expected from
normal absorption at a»a c (i.e., D(a)= sina/Il)' The lie penetration d<!pth is given by
The penetration depth in Fe2 0 3 as a function of IX calculated from Eq. (2) is plotted in
Fig. 1.
6000
~ 4000
o
A = 1.54 A
J1 = 1175/cm
3 4 5
a (0)
The diffracted beam, which originates from this region of variable depth, can be used
for crystalline phase and structural depth-profile determination of surfaces and ultra-thin
films. The diffracted intensity of the reflection i from material j can be calculated by
ex p [ 0(:) ] dt (3)
where
EXPERIMENTAL TECHNIQUE
The geometry of GIABD is shown in Fig. 2. The thin-film specimen is fixed at a grazing
angle with respect to the incident X-ray beam. The detector is scanned along the 28 circle
in the vertical plane of a diffractometer to record the asymmetrically Bragg-diffracted X-
rays from crystal planes that are inclined to the specimen surface.
Incident X-Rays \
~ ____ ~~~~~-=~====~~~
a(fixed)
Monochromatic ,
f
Thin Film
Substrate
APPLI CA TI 01\S
The G IABD technique has been used successfully for characterizing single- and multi-
layer thin filmsY-18 An example of using GIABD for surface and depth-profile analysis
of iron-oxide thin films are given below.
Thin y-Fe203 films with good magnetic and mechanical properties suitable for high
density magnetic recording application can be prepared in a number of ways. One of the
most effective methods involves an initial formation of an intermediate iron-oxide,
Fe304, in an as-deposited film by reactive sputtering of an Fe target in a carefully con-
trolled Ar-02 atmosphere, and then followed by a high-temperature post-oxidation in air
to complete the conversion to y-Fe203.19 In 1986, a polarization neutron-reflection
depth-profile analysis of sputtered iron-oxide thin films revealed a previously unreported
magnetic behavior of a post-oxidized film.2o A thin magnetically-dead layer was detected,
the top 150A layer of the post-oxidized film, whereas the as-deposited film had uniform
magnetic properties throughout its 2600A depth. A structure analysis of the surface and
the bulk of the films is thcreforc important for the understanding of the observed but
unexplained magnetically-dead layer.
(311)
33-22% Fe 2+
Fe 3 0 4 -fcc
1311)
:0
OJ 22-8.5% Fe 2+ Fig. 3. Computer simulated Cu
.~ Simple Cubic Ka XRD patterns: top, face-
co 1220)
E centered cubic spinel Fe3 0 4 ;
0c 5;::
o~ middle, simple cubic iron-
~£'l
oxide; and bottom, tetragonal
1313) y- Fe20 3.
8.5-0% Fe 2+
'Y-Fe203-Tetragonal
1220)
20 30 40 50 60
'2e
T. C. HUANG 95
Apart from small shifts and some intensity changes, the transformation from a face-
centered cubic spinel to a simple cubic or a tetragonal structure has generated several
weak "superlattice" reflections, such as the (300) and (310) peaks of the simple cubic and
the (214) peak of the tetragonal phases. The presence or absence of these "superlattice"
peaks can be used to distinguish these phases.
The application of G IABD to the characterization of the iron-oxide films was done
using both Cu-target X-ray and synchrotron radiation sources; the results are discussed
below.
The basic structure of the post-oxidized film was first identified using the conventional
8-28 scanning technique. The XRD pattern, plotted in Fig. 4, shows a set of relatively
broad thin-film peaks superimposed on a sharp and intense Si(lll) substrate peak at
28.4°. The presence of the strong (214) peak at 28° and seven other weak peaks allow for
a positive identification of the y-Fe203 phase in the film. Because of preferred orientation,
the relative intensities of these peaks are different from those of a randomly oriented
powder specimen. ;\10 second phase was detected in the pattern, probably because the
X-ray path lengths through the surface layer are too short to produce surface diffraction
with adequate intensities relative to the bulk of the film. Therefore, it was necessary to
use the G IABD technique for a surtace-structure determination of the film.
The GIABD pattern of the post-oxidized film was obtained using a modified Rigaku
thin-film goniometer. To increase intensities, a 12-kW rotating Cu anode X-ray source
was used. A bent graphite incident-beam monochromator was placed btfore the specimen
to remove the Cu KP and unwanted radiations. The specimen was fixed at an incident
angle ofO.3±O.2SO. A data collection time of over 15 hours was used to obtain a GIABD
pattern with intensities sufficient for surface structure identification. The GIABD pattern
is predominately by an usually high backgrounJ superimposed on top of a relatively weak
surface-diffraction pattern .. The background intensity decreases monotonically with in-
creasing 28 and was caused by the air scattering and/or the Fe fluorescence from the film.
y(214) Si(111)
y(113)
IX1QQ
-0
<1l
.~
ro
E
a
c
20 30 40 50 60
"28
Fig. 4. 8-28 scanning Cu Ka XRD pattern of the post-oxidized iron-oxide thin film.
96 II. XRD CHARACTERIZATION OF POL YCRYST ALLINE THIN FILMS
The net surface diffraction pattern after background is subtracted numerically is plotted
at the top of Fig. 5. It shows a new polycrystalline diffraction pattern; the y-Fe203 peaks
previously observed in Fig. 4 are absent. The new diffraction peaks match those of a
hexagonal IX-Fe2 0 3 shown at the bottom of Fig. 5. It is worth noting that the relative
intensities of the GIABD peaks also match those of the JCPDS standard pattern (PDF
#33-664) indicating a randomly oriented material (see Fig. 5).
In summary, the GIABD technique has successfully detected IX-Fe203 on the surface,
while the conventional 8-28 scanning technique has identified y-Fe203 in the bulk of the
film. The formation of an anti-ferromagnetic IX-Fe203 layer on the surface caused the
magnetically-dead layer, which had previously been observed by polarized neutron re-
flection. 2o
The synchrotron diffraction analysis was performed at the eight-pole Wiggler station
VIJ-2 of the Stanford Synchrotron Radiation Laboratory.22 A Si{lll) double-crystal
monochromator was used for the selection of a proper wavelength from the continuous
synchrotron radiation source; 1.38A radiation was used initially, and a long X-ray wave-
length (i.e., 1 = 1.83A) was subsequently selected for the analysis. The GIABD patterns
were obtained using a Huber six-circle goniometer.
(104)
a-Fe 2 03
(110)
(024) (116)
(012)
(113) iii
I I
(0061 (202)
I I
~I I
20 30 40 50 60
°2{)
Fig. 5. GIABD pattern of the post-oxidized iron-oxide thin film with background
subtracted (top), and computer simulated XRD pattern of an IX-Fe203 standard (bottom).
T. C. HUANG 97
a(104)
a=O.12°
D=30'&'
"0
OJ
.~
co
E
....
o y(313)
c
a=8.0°
D=3pm
y(226)
y(213+220)
y(400)
20
The synchrotron diffraction patterns obtained initially with ). = l.38A and at ex = 0.12°
and 8.0° are plotted in Fig. 6. The synchrotron data positively confirm the results previ-
::msly obtained from conventional Cu-Kex X-rays by detecting ex-Fe203 0'1 the surface and
y-Fe203 in the bulk of the film .. The peak-to-background ratios, especially the pattern
obtained at ex=O.12°, where the entire incident beam was limited to the top surface of the
film, were also low because of the Fe fluorescence caused by the 1.38A radiation.
The transition and the structural depth-profiles of ex-Fe203 and y-Fe203 were deter-
mined quantitatively from the intensities of the ex(104) and the y (313) peaks obtained at
98 II. XRD CHARACTERIZATION OF POL YCRYSTALLINE THIN FILMS
a(104) a==O.312°
D==90A
,),(213)
~ a(113)
1J
III a==O.319°
.~ D==120A
C1l
E
.... ,),(220)
o
c ~
a==O.325°
D==20oA
30 40 50
°28
Fig. 7. Synchrotron diffraction patterns of the post-oxidized film (tl == uuA and ex ~ ex c )'
ex:S: 1.0°. I t should be noted that the ex(llO) peak is buried under the y(313) peak, and only
the net intensities of the y(313) peak were used in the least-squares refinement analysis.
The two models of the structural depth-profiles used in the least-squares refinement are:
(I) a step distribution function with a constant depth d~ at the surface for ex-Fe203, and
(2) a linear distribution function, with an average depth da and a linear width d w ' (Details
of the refinement has been given elsewhere and will not be repeated here. IS)
The best fit was obtained by a step profile and da == 90A. The observed and calculated
diffraction intensities with a da == 90A step surface of ex-Fe203 are plotted in Fig. 8. The
refinement with the linear distribution profile, however, showed that the fit is insensitive
to a small variation of d w' The maximum value of dw was found to be l20A. Combining
the results obtained from these two models, the ex-Fe203 surface was found to have an
average depth of90±ISA and a transition width of no more than ±60A.
._--1----...:-::1 ')'-Fe 20 3
10,000 25,000 Fig. 8. Plot of the experimental
intensities of the (104) re-
.~ 8,000 20,000
c: flection of ex-Fe203 (open cir-
'" cles) and the (313) reflection
:§ 6,000 15,000
-c of y-Fe203 (solid dots) as a
~
~ 4,000 10,000
function of the incident angle
c: ex. The solid curves are the
5,000 least-square fitted results ob-
tained using a step profile and
o d~ == 90A.
0.2 0.3 0.4 0.5 0.6
a (deg.)
T. C. HUANG 99
a=O.2S0
D=40~
-a
OJ
.~
co
E
o
c a=S.O°
D=2Jlm
(220)
30 40 50
°29
Fig. 9. Synchrotron diffraction patterns of the as-deposited film.
GIABD analysis of the as-deposited film has also been conducted. Synchrotron
diffraction data were recorded with (X from 0.25 to 5.0
0
Selected diffraction patterns
0
•
obtained at (X = 0.25 and 5.0 are plotted in Fig. 9. Unlike those of the post-oxidized film,
0 0
practically the same synchrotron diffraction patterns have been obtained for all (X angles
which indicate no structural change in the as-deposited film. The presence of a weak and
broad "superlattice" (300) peak at 37" of the (X = 5.0 pattern shown in Fig. 9 indicates
0
that the film is not Fe304, but has converted into a simple cubic and/or a tetragonal
iron-oxide. In terms of the degree of oxidation and the iron-cation content, the results
indicates that the film has been oxidized beyond Fe304, and has less than 22% Fe2+ and
more than 78% Fe3+.
CONCLUSION
Is has been shown that the GIABD technique can be used for the structural identification
and depth-profiling analysis of surfaces and ultra-thin films. The successful application
of GIABD using conventional X-ray and sync!lrotron radiation sources for the charac-
terization of the as-deposited and post-oxidized iron-oxide films demonstrates the effec-
tiveness of the technique.
ACKNOWLEDGMENTS
The author would like to thank S. S. P. Parkin for providing the iron-oxide films; B. R.
York for collecting the GIABD data using a rotating anode X-ray source; S. Brennan, Z.
Rek, and M. Toney for the GIABD experiments preformed at SSRL; and M. Toney for
the least-squares refinement analysis.
REFERENCES
7 R. Feidenhans'l, Surface Science Report 10, 105 (1989); and references therein.
8 Y. Yoneda and T. Horiuchi, Sci. Instr. 42,1069 (1971).
9 P. Wobrauschek and H. Aiginger, Spectrochim. Acta, B35, 607 (1980).
10 C. T. Yap, R. E. Ayala, and P. Wobrauschek, X-Ray Spectrom. 17, 171 (1988).
11 1. M. Bloch et aI., Phys. Rev. Lett. 54, 1039 (1985).
12 A. Iida, K. Sakurai, and Y. Gohshi, Adv. X-Ray Anal. 31,487 (1987).
13 S. M. Heald, E. Keller, and E. A. Stern, Phys. Lett. J03A, 155 (1984).
14 G. N. Greaves et aI., Daresbury Laboratory Technical Report DLjSCIjP586E, SERC,
Daresbury Laboratory, Daresbury, Warrington WA4 4AD, U.K.
15 B. R. York and A. B. Austin, Proc. APIE, X-Rays in Materials Analysis: Novel
Applications and Recent Developments, edited by T. W. Rush, Vol 690, P. 135 (1986).
16 T. C. Huang and B. R. York, Appi. Phys. Lett. 50,391 (1987).
17 T. C. Huang, M. F. Toney, S. Brennan, and Z. Rek, Thin Solid Films 154,439 (1987).
18 M. F. Toney, T. C. Huang, S. Brennan, and Z. Rek, 1. Mat. Res. 3, 351 (1988).
19 Y. Ishii, A. Terada, 1. Ishii, S. Ohta, S. Hattori, and K. Makino, IEEE Trans. Magn.
MAG-16, 1114 (1980).
20 S. S. P. Parkin, R. Sigsbee, R. Felici, and G. P. Felcher, Appi. Phys. Lett. 48, 604
( 1986).
21 K.l. Gallagher, W. Feitknedht, and U. Mannweiler, Nature 217, IllS (1968).
22 D. E. Moncton and G. S. Brown, Nuci. Instrum. Methods, 208, 579:1983).
STUDY OF lHIN FILMS AND MULTILAYERS USING ENERGY-DISPERSIVE
DIFFRACTION OF SYNCHROTRON RADIATION
Daresbury Laboratory
Warrington WA4 4AD
England
ABSTRACT
We present the results of a preliminary study of the use of
polychromatic synchrotron radiation in energy-dispersive mode for the rapid
characterization of thin film and multilayer materials. This technique takes
advantage of the high intensity and excellent collimation properties of white
beam synchrotron radiation. The glancing angle diffraction geometry allows
structural depth probing as well as enhanced signal-to-noise by suppressing
substrate contributions. The technique was successfully applied to metallic
films a few hundred Angstroms thick, and to W-C multilayers.
INTRODUCTION
The technique of using a glancing incidence angle to enhance surface
SenSltlVlty in X-ray diffraction experiments is well established both for
studying single crystal surfaces (Marra et al 1979 and 1982) and thin
polycrystalline films (reviewed recently by Wagner et al 1988). When using a
divergent beam from an X-ray tube, a focussing geometry imposes severe
constraints on the range of glancing angles which can be accessed without
introducing defocussing errors and consequent line broadening. Although the
Seemann-Bohlin parafocussing geometry has been used with some success in
characterizing thin films, the poor angular resolution is a serious
limitation (York and Austin, 1986). It is evident that a parallel beam
geometry offers the possibility of improved angular resolution, and the
advantages of this arrangement for studying thin films have been demonstrated
by Lim et al (1987). The ease and versatility of the parallel beam geometry
for powder diffraction applications, particularly when using synchrotron
radiation, has been extensively documented by Parrish et al (1986 and 1987)
and Hart and Parrish (1986 and 1989). In this arrangement, the angle of the
beam diffracted by the specimen is defined by a set of long, parallel foils
positioned between specimen and detector. In this way, the specimen and
detector may be coupled or uncoupled during a scan without distorting or
broadening the reflections. For enhanced surface sensitivity, the decoupled
mode is used, and a fixed incident or glancing angle (typically less than
1°) is selected. For a given wavelength in the incident monochromatic beam,
the choice of glancing angle will determine the penetration of the radiation
into the sample (Lim et al 1987).
Advances in X-Ray Ana/ysis, Vol. 33 101
Edited by C.S. Barrett el 0/.
Plenum Press, New York, 1990
102 II. XRD CHARACTERIZATION OF POL YCRYSTALLINE THIN FILMS
INSTRUMENTATION
solid state
detector
28
------
t""
~--------------~---
~ .
sample ,<fJ ~ inCident white beam
(-<1 0,1 mrad)
glanCing a~gle
3
Inllde.nt white beam spectrum
m
(throuyh 1mm2 aperture I
0~ x108
0
u
5
-
QJ
VI
.c::
-x
Cl.
:J
LL
o
o 20 40 60 80 100
Energy (keY)
where E is the incident beam energy in keY and dhkl is the interplanar
lattice spacing in Angstroms. The range of energies available in the
incident white beam is shown in Fig.2. Although useful flux is present up to
about 80 keY, experiments are usually designed so that sufficient powder
peaks fall between about 15 keY and 45 keY. A typical powder pattern from an
NBS Si sample is shown in Fig.3, which was collected at a Bragg angle of 15°.
Although this is an appropriate choice of Bragg angle for a typical
polycrystalline specimen, we will see that much smaller angles are needed
when studying multilayers. It is also clear that the broad widths of the
Bragg peaks shown in Fig.3 limit this technique to crystallographic systems
of fairly high symmetry, otherwise excessive peak overlap will complicate the
interpretation of the pattern. Fortunately, most thin film polycrystalline
materials possess a suitable high symmetry (often being simple metals or
oxides).
12
VI
§ B
o
LJ
o-~
-'I--~----~I---'----'---~
10 20 30 40
Energy (keY)
TABLE 1
Critical angle cl>c (mrad)
Energy (ke V) Cu Au
10 5.94 8.70
20 2.97 4.35
30 1.98 2.90
40 1.48 2.17
50 1.19 1.74
60 0.99 1.45
70 0.85 1.24
80 0.74 1.09
90 0.66 0.97
100 0.59 0.87
RESULTS
lOOO 2u _ 8~o
f = 1.7. 3,4 & 5 mrad
;><,00
7000
:J
0
u
10;00 -
1000
~oo
I~ 20 7~ 10 lS
Energy I keVI
29 =20·
( u 1350},) I Au (l000'\) on glass
f = 1.2.3. 4&5
2500
mrad
o ......
...,
o
...
2000 « i
,
.. U
,
:::I
o c
...... .
0
..,'"
U ~
o ~
......
o ...
,UU, , ,
N ~
' 000
o "''''
...,
o
" "
U
soo
'5 2S JO lS 40
Energy (keV)
keV and 50 ke V are of the order of a few rnrad. A glancing angle of 1 rnrad,
say, should limit the penetration of the X-ray beam to approximately the
uppermost 50A of the Cu layer. FigA shows spectra collected at a scattering
angle of 8.5 0 at various glancing angles. At a glancing angle of 1 rnrad only
the Cu(111) peak can be observed in the energy range between 15 keY and 45
keY. As the glancing angle is increased, the Au peaks rapidly become
dominant. Above 3 rnrad, the beam penetrates through both the Cu and Au
layers into the amorphous glass substrate, producing a broad hump of
scattered radiation centred around 30 keY. Increasing the scattering angle
to 20° brings many more powder peaks into the region of interest, as shown in
Fig.5. Once again, the contribution from the Au remains very small until the
glancing angle is large enough for the beam to penetrate through the Cu
layer. It is interesting to note the strong Cu(222) and, at larger glancing
angles, the Au(222) peaks in this pattern. This confirms the high degree of
texture in these sputtered films. Each of the spectra shown in Figs. 4 and 5
was collected in a measuring time of 600 seconds.
The next two examples are W-C multilayers deposited on a Si single
crystal substrate. Both multilayers consist of 30 bilayers with ad-spacing
of 28A in one case and 56A in the other. At the chosen scattering angle of
2°, the first order Bragg peaks should appear at 12.7 keY and 6.35 keY for
the 28A and 56A multilayers, respectively. The results are shown on a log-
scale in Figs. 6 and 7, where the various orders of Bragg reflections are
labelled. Again, the measuring time per spectrum was 600 seconds during
single-bunch operation of the SRS. It should be remembered that the
intensity of the incident beam beyond about 50 keY has fallen dramatically
(see Fig.2), and no correction for the incident spectral distribution has
been made. Nevertheless, diffraction peaks up to about 6th or 7th order can
be seen in the raw data shown in Fig.7. It should also be pointed out that
these spectra were collected in asymmetrical grazing incidence geometry. If
measurements are made in symmetrical geometry (ie the glancing angle is set
equal to the Bragg angle, in this case 1°), very high reflectivities can be
106 II. XRD CHARACTERIZATION OF POL YCRYSTALLINE THIN FILMS
o ~ ro ~ ~ ~ ~ ~ ~ ~ ~
Ener gy (keY I
DISCUSSION
11
w- ( multilayer
10
IN: 30 . d: 56 AI
9
29 : 2°. 4J : 2 mrad
8
.
~ 6
::>
o
!! 5
c
- 4
o~ __________________~~~~~~
o 10 ro ~ ~ W ~ ~ ~ ~
Energy I keY I
ACKNOWLEDGEMENTS
The authors wish to express their gratitude to the SERC for use of the
Daresbury Laboratory facilities, and special thanks to Mr A Neild for his
technical support.
REFERENCES
ABSTRACT
The grazing-incidence X-ray diffraction (GIXD) method
was employed to analyze two-layer thin films, which were the
samples of 100A Au/500A Cu/Si0 2 (substrate) and 250A Cu/500A
Au/Si0 2 (substrate), which were prepared by the evaporation
technique under the condition that the Si0 2 substrate was
at room temperature. Diffraction profiles were obtained at
various glancing angles (a) and the data were analyzed as a
function of a. The results were as follows: 1) Diffraction
peaks were shifted to larger diffraction angles, because of
the refraction of the incident X-ray beam. The a2gular2s~}~t
has been approximated by the equation, a-( a - ac ) ,
where ac is the total reflection critical angle of the
material. 2) As a result of the correction of angular shift,
the stress of the evaporated films was estimated to be nUll.
3) The broadening of the Cu diffraction peak and the
enhancement of the Cu diffraction intensity occurred at
angles near a c of CU,due to the reflection of the X-ray beam
at the Cu/Au interface. The angular ~hift2 ~y~ to the
reflection has been approximated by a+( il' - a c ) , where
il'c is the critical angle of the upper-layer, Cu. 4) Below
the critical angle, small angular shifts were also observed.
The shift was thought to be caused by the roughness of the
sample surface. 5) In order to calculate the crystallite size
from FWHM (full width at half maximum) obtained by GIXD, a
somewhat higher glancing angle than the critical angles of
both the upper- and the lower-layers should be adopted. 6)
The X-ray penetration depth for the multilayer sample has
been calculated as a function of the glancing angle and the
calculated results agreed with the experimental results. 7)
The intenSities of diffracted X-ray beams were also
calculated as a function of glancing angle. In consideration
of the interface-reflection, the calculated results for the
Cu in the Cu/Au sample were consistent with the experimental
values.
I. INTRODUCTION
Recently several surface analyses have been performed in
several fields by many investigators. In analyses by
employing X-ray beams, the methods with the technique of
grazing-incidence are particularly noteworthy for surface-
and near-surface-analysis ; for example, the total-reflection
X-ray flyorescence method for elemental analysis of trace
amounts, the grazing-incidence X-ray ~Jattering for
structural analysis of near-surface layers , etc. The
grazing-incidence X-ray diffraction CGIXD) method is a
similar surface-analytical technique. There are two kinds
of GIXD methods. One of them is to analyze a three-
dimensional structure Cpolycrystalline)3) The other is to
analyze a two-dimensional structure by employing both grazing
incidence and grazing observation, for determining the atomic
arrangeme~} on the surface; for example, the InSb(lll) 2 X 2
structure . The former GIXD is well suited for the near-
surface measurement, i.e., a thin film near gyrface, and
has gyen applied to the study of a nitride layer ,an oxide
film '7)etc. This GIXD is also called the Seemann-Bohlin
method .
In the present study, the former GIXD was employed for
three-dimensional structure analysis. Au/Cu and Cu/Au two-
layer thin films were analyzed by the GIXD and the phenomena
of refraction and reflection at the surface and the interface
were observed. Effects of these phenomena on the angular
shifts, the penetration depth, and the diffraction intensity
due to the very small angle of the incident X-ray were
discussed.
II. EXPERIMENTAL
The GIXD measurements were carried out using a
Geigerflex Rad-B spectrometerCRigaku Corporation) attached to
the thin-film measuring systemCModel 2655Al). The geometry of
instrument is shown in Figure 1. A rotating cobalt anode was
operated at 30kV and
100mA, and a flat
graphite monochromator
(0002) was used a
Measurement conditions
were an incident
slit width of 0.2mm
and a ~ranning speed
of 2°min . Values of
the symbols in Fig.l
are shown in Table I. X-ray
The flat-surface Source Slit
samples of two-layer SI
thin films were
prepared by the
evaporation technique,
i.e., Au and Cu were
evaporated onto an
optical flat surface
an amorphous quartz Fig.! Geometry of GIXD instrument
T. TAKAYAMAANDY. MATSUMOTO 111
-
intensities.
In Fig.2, the >. X1.7
experimental critical angles
-
I I)
t:
are almost consistent with C1>
/1
GIXD profiles were
investigated in detail Au(lll) Au 100;"
at the glancing-angle I Cu(111) ~cu 500A
range 0 f O. 1° to 2. 0° . .......___ \' , Si0 2 100~~1s
The GIXD profiles
obtained at some of Au(200) I
the glancing angles I Cu(200)
are displayed in I
Figure 3, where each
peak pOSition
indicates the value
c=
I
from JCPDS. The ~I
diffraction peaks tend X
to shift to higher
diffraction angles at
small glancing angles.
The integral peak
positions were
determined after data
treatment with Co K a2 40 50 60
stripping and
smoothing. Figure 4
Au(lll) Cu 250A
shows the diffraction- ( b)
angle differences ~Au500A J
between the Cu(ll1) ~ 1000cfsl
experimental peak 8-28 I x 112
positions and the
JCPDS positions I
plotted as a function Cu(200)
of glancing angle.
At a< a c ' total
I
reflection occurs at
the sample surface;
the diffraction peaks ~I
appear in Fig.3. One X
of the causes was
thought to be
divergence of the
incident beam the
divergence of the
incident beam 40 50 60
was geometrically 28 I degrees
calculated to be below
0.02°). Another of Fig.3 GIXD profiles and 9-29
causes was thought diffraction profiles
T. TAKAYAMAANDY. MATSUMOTO 113
~~~~
corrected -11- - - - - - - - - -
4.05
_______ J--------.JJ! (l- - - - - - -
'--a:=0.7°
uncorrected
(1) are shown in Table II, where 6 is the real Bragg's angle
obtained by Eq.(I), .8i is the integral FWHM (integral
breadth), and f is the average crystallite size. Although
the formula does not contain the instrumental correction
coefficient term, the calculated crystallite sizes of the
(a)
reflected
X-ray
interface +------"--:-,---=::::,...,:::::.....---1---_\_
Au 500A
I
(b)
I250c/s
a
Z' It
'iii ~--""""' 0.7 0
-
c
<I>
"""""""--0.65 0
.....C
F = (7 )
: : :h
refraction phenomenon is
ignored in Eq.(4).
The attenuation model
of the incident X-ray
considered for the multilayer
is
(j-l)th layer ~
""~'-'-t' t
*tj-l
sample as indicated in Figure
7. Analogous to Eq.(4), the (j)th layer ----~_~tl
wave intensity diffracted at
depth t' in the (j)th layer Fig.7 Attenuation model of
of the multilayer sample is X-ray beam in multilayer
given by sample
T. TAKAYAMAANDY. MATSUMOTO 117
j-l j-l
= 10 IT Ti(a) exp[- r ti{ai(a)+,ui cosec/d Tj(a)
i=l i=l
x exp[-t' {a j (a)+ ,uj cosec (J }], (8)
. 1
where i denotes the (i)th layer and t'= tp- rt:lt i . Assuming
that the decaying rate of X-rays reaching the surface after
being diffracted in the (j)th layer is y %, which is the
mInImum rate-value of the detected intensity, the X-ray
penetration depth tp is given by
j j-l
tp = [-In{(1-0.01y)/ IT Ti (a )}- r ti {ai (a )-a j ( a)
i=1 i=1
111-5. Intensity of
Diffracted X-ray Au
Dependent on o 0.5 1.0 1.5
Glancing Angle
For powder X-ray
diffraction analysis, Peiser
et al. has given an
equation of diffraction
intensity14). However, this
equation can not be applied
to practical measurements,
because a sample with enough
or infinitely wide surface
is needed, and refraction
and reflection are not o
considered in this equatio~5 a/ degrees
BruneI and de Bergevin
have given the equation of Fig.8 Calculated X-ray
X-ray intensity for mirror- penetration depth
surface samples. In this vs. glancing angle( a)
118 II. XRD CHARACTERIZATION OF POL YCRYSTALLINE THIN FILMS
j j-1
Id = 10 Z(a) IT T i ( a ) exp[- r t.{a.(a)+,ui coseCjl}]
i=l i=l I I
.5 5.0
~.;..
:::----
----
layer in the 100A Au/500A Cu "C
Q)
IV. CONCLUSION
ACKNOWLEDGMENT
The authors wish to express their gratitude to Mr. Y.
Matano and Mr. H. Tamura for assistance in part of the
experiment.
REFERENCE
1) P. Wobrauschek, H. Aiginger : Spectrochim. Acta, 35B, 607
(1980).
2) E. Matsubara et al. : Z. Naturforsch. 43A, 181 (1988).
3) M. BruneI, F. de Bergevin: Acta Cryst. A42, 299(1986).
4) J. Bohr et al. : Phys. Rev. Lett. H, 1275(1985).
5) N. Moncoffre et al. : Surf. Interface Anal. ~, 139(1986).
6) G. Lim et al. : J. Mater. Res. £.471(1987),
7) R. Feder, B. S. Berry: J. Appl. Cryst. ~, 327(1970).
8) L. G. Parratt : Phys. Rev. ~. 359(1954).
9) H. F. Quinn, P. Cherin : Adv. X-ray Anal. Q, 94(1961).
10) W. H. Hall: J. Inst. Met. 75, 1127(1950).
11) R. Khanna et al. : Nucl. Instrum. Methods Phys. Hes. B15,
752(1986).
12) M. Katayama, M. Shimizu: Adv. X-ray Chern. Anal. Jpn. ~,
281(1988),
13) M. F. Toney et al. : J. Mater. Res. ~, 351(1988).
14) H. S. Peiser et al. : " X-ray Diffraction by Polycrystal-
line Materials" Inst. Phys., London. p.159 (1955).
ENHANCED HANGE OF MEASURABLE THICKNESS OF THIN CRYSTALLINE
INTRODUCTION
i(t)
Y [1]
i(co) 1-exp(-2·U·t·csc~)
MODEL
Taking the expression for the diffracted intensity from a
unit volume of material and introducing the correct. ion for a
sample tilt of a degrees, we obtain the following equation:
If we define
then
dl (4J
t
K-A(9) [ ]
let) 1 - exp(-~-t-S) (5J
~-S
o
In a similar manner, the factor which modifies the intensity
ratio between two diffraction lines 291 and 292 is:
1 - exp(-~-t-S(91,a»
y (6]
1 - exp(-~-t-S(92,a»
lI": TA I- IO rtf'TnZ-OO
1.2 r------ - - - - --
IIlPIl Il - r J -
1-
IIlPI1II - S - ---
~ 1.r) -~;.':":1.;"~~;-:: ~ ... _ "7"' __
IIU'I III - 7 - -
1\1 r'~ ~9 -.
I- "
'.
II.A
IU
1-
'r n.G
•..1
UJ
(1
n n.~
II I
r I
..J
':1 n.;::
tt:
---.--_._._ ...
4
f€ ".
11.1I--- - - -y- o
UJ' ,( 'J t ) 10'
EXPERIMENTAL METHOD
2000
o Si Free I 11]0 U. IITm i!!I MgO O .4rrm
c'n
:~ I51JO
c
:~ leno
:no
Ill-L-l-"-----'--
4000
3ffiO
o 'VJ THI~~ • 'VJ O.an. tt;Poerm
3200
a H.W
<I>
\.
0>
:!! 1200
oS
1.01
<100
a
HI<.L r l l)n8
150
1<10 o r-tJl THIN • tt;P O.2nm tt;P o.an ..
130
J>
~ 120
~ 110
c IOU
0 00
~ 80
0 ( b)
\.
01 7U
:!! GO
oS
50
0
01
N 40
-
:JO
~\.
0 20
z: In
n
HI<L p lone
///
:1 . 0 r - - - - - - - - -- - - -- -- - - -/---...
~~::-
.3 2.4
o 2.2
¥ 2.0
e
E I.B
1.6 -
•. / / /
_
//
. ~
1.4
ill
L
1. 2 '
1.0 ./
ft ,,'
J
(ij
a
+
0.8-
III
:L 0.6'
..... 04 ~
~O"I/ O.O~~~~-~~-~~--~~~~-~~~~
0.0 0 .2 0.4 0.6 0.8 1.0 1. 2 1.4 1.6 1. 8 2.0 2.2 2.4 2.6 2.8 3.0
ACKNOWLEDGEMENTS
REFERENCES
ABSTRACT
1. INTRODUCTION
this goal is not simple and the formation of TiN on the surface depends
strongly on the nitriding conditions.
2. EXPERIMENTAL
Since the thickness of the layers was beyond the penetration depth of
the Cu-Ka beam (8 J.1m at 29=40°), diffractograms of several samples were run
with Mo-Ka radiation and the (110)R> line appeared. For this line, the
penetration depth is 30 J.1m (Mo-Ka) and this diffractogram was received
from both the nitrided zone and the untreated bulk. From these results, we
came to a conclusion that the nitrided zone is composed of titanium at ,
E and 6 phases.
A. RAVEH ET AL. 131
The thickness of the TiN layer measured by SEM. and the ni trogen
concentration determined by AES are correlated with the XRD results.
Figure ~ shows the I(200}!I(111} ratio as calculated by the XRIC computer
program vs. film thickness for randomly oriented 6-TiN. while Fig. 4
shows values as measured in this work. As shown in Fig.3. the
a-(Ti.N) N
0
~
E-Ti2N
0-
~
• I)-TiN
0-
§
~ (a)
N
'"
0
~
N2+H2 (80:20)
(b)
N2+H2+Ar (80:10:10)
(e)
N2+Ar (80:20)
(d)
80 70 60 50 40 30
Diffraction angle (2e)
Figure 1 X-ray diffractograms of plasma-nitrided samples at
0.5 MHz. rf power=1.1 kVAj gas pressure=7 mbarj total
feed=150 sccmj nitriding time=5 h. (a) untreatedj
(b) N2 :H2 =80:20 (c) N2 :H2 :Ar=80:10:10 (d) N2 :Ar=80:20.
132 II. XRD CHARACTERIZATION OF POL YCRYSTALLINE THIN FILMS
2.95 4.685
0$ 2.94
."
~ 2.93
4.680
..
'"
-.:
OJ
OJ
.."
S
4.675
..
~
"CI.
OJ
2.92 "CI.
OJ
~ ~
"
...l 2.91
4.670 "
...l
2.90
0 10 20 30 40 50 60
92 Concentration (vol. %)
4
B
1.8
52at.%N
1.6 ;;- 3
....
;;-
.... 1.4
:j
'-----::oN g
8eo
.----
t:: --------::oN
8<: 1.2 --. ~
~
• .... 2 II
0.6 1 "
0 2 4 6 8 0 2 4 6 8
Film Thickness (J.!m) Film Thickness (J.!m)
4.246
z ~
...1= 4.244 +
'eu"
... + ~
.!l 4.242 ~
"e
...eueu + +
Q. 4.240
~
+
~"
-=
....l
4.238
~ +
4.236
30 35 40 45 50 55
Nitrogen Content (at. %)
w (,()
~ ~
50
I I 40
40
E E
'"' 30
C "
C
.'"
0 0
U U 211
20
'"'"'"'
.&: III '"
.c
c.. 0-
CJ ()
0 10 20 50
G:lS Composition H2 Concentration (vol. %)
lOO
~ 80
Q
.... 60
Q
"
.......,
tl 40
N
e
Q
~
20
the amount of a-phase was only one third of that of E-phase. The amount
of E-phase decreased with decreasing N2 concentration in the plasma
(Fig. 6a), or when the hydrogen was replaced by argon (Fig. 6b). The
E-phase was found to have a strong texture with the c axis normal to the
surface. The preferred orientation is a function of the overall film
thickness as seen from Fig. 7. The variation of the texture for several
conditions is also shown in Fig. 1.
(c) The reaction between active nitrogen species and the titanium
depleted of Al and V forms a compound layer of E and a phases. In this
layer, the E-phase has a strong (001) orientation parallel to the
surface, but the a-phase is randomly oriented. Similar texture was
found by Rie and Lampe 11 •
REFERENCES
1. R.R. Manory and G. Kimmel, Thin Solid Films 150, 277 (1987) .
2. V. Valvoda, R. Cerny, R. Kuzel,Jr., L. Dobiasova, J. Musil, V. Poulex
and J. Vyskocil, Thin Solid Films 170, 201 (1989).
3. A. Raveh, G. Kimmel, U. Carmi, A. Inspektor, A. Grill and R. Avni,
Surface and Coatings Technology, 36, 183 (1988).
4. A. Raveh, R. Avni and A. Grill, to be publ. in Thin Solid Films (1989).
5. W.B. Pearson, Handbook of Lattice Spacing and Structure of Metals and
Alloys (Vol.4), Pergamon Press, 1965.
6. G. Kimmel, "Program XRIC (X-Ray Intensities Computations)", Rep #878,
Philips Lab. Briarcliff Manor, NY, 1986.
7. L.E. Toth, Transition Metal Carbides and Nitrides, Academic Press,
New York 1971.
8. J.E. Sundgren, B.O. Johansson, H.T.G. Hentzell and S.E. Karlsson,
Thin Solid Films 105, 385 (1983).
9. J.P. Noel, D.C. Houghton, G. Este, F.R. Shepherd and H. Plattner,
J. Vac. Sci. Technol. A2, 284 (1984).
10. Metals Handbook, 8th ed. (American Society of Metals, 1973), Vol.8,
pp 322 and 369.
11. K.T. Rie and Th. Lampe, Metallberflaeche 31, 1003 (1983).
A COMPARATIVE STUDY OF STRESS DETERMINATION
Abstract
A review of the literature has shown that two major techniques are used to determine
stress/strain in thin film structures: curvature measurement technique and internal stress measure-
ment by the traditional sin 2t/J technique. However, to date, there is no published comparison of the
results from both techniques when they are applied to a given specimen.
In this paper we compare hoth methods theoretically and experimentally. We show that,
while in some cases similar information is obtained from both techniques, there are cases where the
two measurements give different results. We will show that, in such cases, the two methods can be
complementary. Thus, they can be used to yield a more complete description of the total
stress/strain tensor in the material.
Stresses and strains are almost always present in thin films deposited on substrates. These
stresses may be intrinsic to the film, caused by defect distributions in the film volume, such as im-
purity, void and/or dislocation distributions. I - 3 They may also arise from morphology changes
within the film, such as partial grain growth, recrysta11ization or phase change. These intrinsic
strcsses may be affected by the presence of the substrate; however, they would be present even if
the film was "free-standing". The intrinsic stress field, if integrated over the entire film volume,
will average to zero. Another type of residual stress field may form solely due to the presence of the
substrate, in response to the differences of material parameters in the film/substrate interface. These
residual stresses can be called exlriflSi<' residual stresses. For example, a film deposited at elevated
temperature will have residual stresses at room temperature if its coefficient of thermal expansion
is different from that of its substrate. In epitaxial films, differences in plane spacing between the film
and the substrate may also cause an appreciable residual stress field at any temperature. "Extrinsic
Figure 1: The sample coordinate system Xi and the direction of the measured x-ray stress ".p.
stresses" will have finite values when integrated over only the film or the substrate, but will yield a
zero average if integrated over the total film + substrate volume. 4
The total stress in the film is the sum of the stresses from all the different causes. Interactions
between materials, deposition parameters, etc., can result in complex stress/strain distributions
within the film-substrate composite. Thus, even though theoretically one can calculate the stress
distribution for some cases, experimental measurement of strain/stress distributions is the preferred
approach.
There are two methods that enjoy wide spread usage in stress measurements on
polycrystalline thin films. If the thin film has sufficient thickness to yield a good diffracted beam,
one can use the traditional d vs. sin 2if; technique4 : In this method *, the plane spacing of a family
of planes d hk, are used as an internal strain gage. If dhk, are measured along various directions I1¢\I
in the film (fig. I), where the angles ,p, if; are set on a powder diffractometer, the stress in the surface
direction S+ is given by the formula:
d</>, >/1 - do (1 + v) . 2 v
d = {--
•.- } "A. sm 1/1 - (-,. )"kk (I)
o "hkl 'I' \
Thus, from the slope of the d¢~ vs. sin 2if; plot, one can obtain the stress "¢, if the elastic constant
term {I + v/Ehkl and the unstressed lattice spacing do are known. For routine analysis, one may
use the intercept of the d¢~ vs. sin 21/1 plot at if; = 0 for do. For most cases, the error due to this term
will be negligible compared to the instrumental errors4 • This method of stress analysis requires
largely untextured or weakly textured films and assumes linear variation of d¢jI with sin2if;. When
these conditions are fulfilled, one can obtain stress values to =+= IOMpa with reasonable attention to
experimental detail. The errors associated with the measurement have been studied extensively and
formulas describing them are available in the literature. 4 ,S
The stress a¢ obtained from equation (1) is the total average stress in the grains diffraeting
during the measurement. Because of the selective nature of the diffraction process, only a partieular
set of grains will diffract at any if; tilt and, thus, the strain data is from incomplete, local subsets of
*This method was originally proposed for use on bulk samples and has been in use for over 70
years. In its original form, it was applicable to bi-axial stress distributions. In the last decade or
so,this method has been modified and extended to triaxial stress distributions. For most routine
thin film deposition techniques, the bi-axial analysis is adequate. We will review this method here.
For more detailed analysis techniqllcs, the reader is referred to reference 4.
I. C. NOYAN AND C. C. GOLDSMITH 139
the total irradiated volume. Such incomplete averaging can result in fmite values for certain strain
fields that would integrate to zero over large, continuous volumes of the film. Consequently, 11~
determined from the sin 2t/1 technique may contain extrinsic and intrinsic components as well as
interaction stresses, depending on the actual stress distribution in the irradiated volume. Further-
more, 11~ is an average over the depth of penetration. Any stress gradient within the irradiated vol-
ume will be averaged over.
The second method widely used in stress detennination from thin films is the measurement
of bending. It is well known that, when a substrate is coated with a film on one side, the composite
bends. 6 This is due to the unbalanced moments caused by the asymmetric residual stress distrib-
ution. The radius of curvature R uniquely represents the internal stress distribution, and a large
body of literature exists for the formulas linking R to l1R. I • 2• 6 - 12 Por a substrate with a single film
layer on it, the relationship between Rand O"R in the film is given by+:
(2)
Here, to, tJ, Eo, EJ, VO, VI, are the thicknesses, Young's moduli and Poisson's ratio of the substrate
and the film, respectively. The term,) is the Davidenkov deltaR and describes the shift of the neutral
axis from (to + t l){2:
(3)
If olle assumes that the film and the substrate have the same elastic moduli, and that the film
thickness is much smaller that the substrate thickness; (tl < < to), equation (2) becomes:
Eo t~
(l1¢)R = (1- vo) 6Rt l ' (4)
A large number of the papers also treat multiple films on substrates. I - 3,9·- 11 None of these
approaches, on the other hand, can rigorously analyze the case where ollly the total radius of cur-
vature R, and thickness of each layer t i , i ~ IN are known. The best approach in such cases is to se-
quentially deposit the films and measure R after each deposition step. One can, then, use equation
(4) for stress analysis as long as the total thickness of the films is much smaller than that of the
substrate. One can see from equations (2-4) that thickness information for all the layers on the
substrate must be known accurately for bending analysis. When such information is not available,
measurement of R by itself yields little, if any, information on the state of stress.
X -ray diffraction can be used to measure the radius of curvature R if the substrate is a single
crystal. The technique is straightforward and has been described in detail in the literature. I ,2,13,14
Because diffraction is used to determine peak positions from which the radius of curvature R is later
calculated, there will be a statistical error associated with R. Thus, the errors associated with the
bending method depend on the accuracy of the radius of curvature R and on the errors associated
with the ti. The standard deviation in (TR calculated from equation (4) can be determined from
standard error propagation I.,: ..!..
Eo to 2 t62 t~22
SO' = (I _ vo) 6Rt l {4S Io + 7 St l + R2 SR} (5)
I
KR2
N
0.80 Jim -
KR3
N
0. 93 ,lm 1.56 Jim
KR5 - 2 Jim -
KR7 Polyimide (lO,lm) \.9 Jim (electroplated) -
KR8
N
- -
KR9
N
2.72 11m (electroplated) -
--
NI Cr (~400i\) I Jim 1.48 Jim
-----
N2 2.8 Jim
N N
N3
N N
-
~-
N4
N N
-
c--------- - -- -
N5 Cr (1.7 11m) on both 3 Jim -
sides!
'-------------- - - - - - - - - ~-
--
Where Sj denotes the standard deviation of the ith variable. It can be seen from this equation that
the film and substrate thicknesses must be known very accurately. SR will contribute less to the
overall uncertainty because of the multiplier t~/R2. For most practical cases, S. will be around 4-15
MPa.
The stress (l'R obtained from equations (2, 4) is, by defInition, an extrinsic stress, since it -rc-
suIts from the asymmetric moments in the film-substrate couple. Thus, if there are intrinsic stresses
in the film as wen, the stress determined from equations (I), (2, 4) may not agree. In such a case,
the difference, if any, may be due to the presence of such intrinsic stresses. Thus, the two methods
are complementary and may be used to fully determine the stress state in such materials. In the
following, the results of such an analysis, carried on Cu films deposited on Si substrates is reported.
Experimental Details
Thin Cu films were deposited on (100) Silicon wafers by electroplating, electro less plating,
vapor deposition or by a combination of the three methods. Three other specimens (KR7-KR9)
were prepared by spin coating 10 micrometers of polyimide on Si wafers. i\ thin film of Cu was
subsquently deposited on the polyimide by a combination of electro less and electroplating tech-
niques. Film thicknesses were measured by surface profilometry, or weight gain. In table I, perti-
nent data about the specimens used in this study are summarized.
After the specimens were prepared, residual stress measurements were carried out using both
sin 21j, and radius of curvature methods. Por the sin21j1 measurements, the 420, 220, 222 and 311 re-
flections were used, with eu radiation for the 420 reflection, Cr radiation for the 220 reflection and
Fe radiation for the 222, 311 reflections. Data collection for 420 (eu X.) and 222 , 311 (Pe K. )
was carried out using P filtered radiation, with the x-ray tube operated at 40kV-45mA. Por these
measurements the goniometer was a Rigaku powder diffractometer using normal parafocussing
optics and a scintillation detector. Por data collection from the 220 planes (CrK.) a vacuum path
I. C. NOYAN AND C. C. GOLDSMITH 141
II>
~
o
~~'--~--~~--~--~~--~
o 0 0.2 sinzt/l 0.4 0.6
figure 2: ;\ typical d"" vs. sin 21/' plot. The plot shows no psi-splitting
and is qllasi-linear (420 reflection from specimen KR2).
goniometer was used, in which the entire diffractometer (source, goniometer, and detector) was
enclosed in a vacuum chamber pumped down to rough pump vacuum. For this measurement,
parallel beam optics were utilized with unfiltered radiation and a flow proportional detector. For
all cases, the specimens exhibited partial texture, with quasi-linear d4>!1 vs. sin2'" plots. A typical plot
is shown in figure 2. For stress analysis, x-ray elastic constants, calculated from single crystal
compliances in tpe Kroner limit, were employed. A summary of the applicable elastic constants is
given in table 2.
The wafer bending measurements were carried out on a modified Lang topography camera,
with Mo radiation diffracted in transmission from the Si (220) planes. The center 60 mm of the
wafer was scanned, using step sizes ranging from 2 to 20 mm depending on the amount of curva-
ture. A typical plot of rotation vs. translation is shown in figure 3. In this figure, since the rotation
vs. translation is quite linear, the wafer is bent uniformly, with a constant radius of curvature across
the distance scanned. The intensity vs. translation, on the other hand, shows some non-linearity,
with one fairly large peak at the begining. This may indicate some thickness non-uniformity near
the edge. For the stress calculations, the elastic constant E/(l - vo) for Si (100) wafers was used
from Brantley16 and UR was calculated using both equations (2) and (4).
0.20 19000
0.15
...,...,
<f)
~ Z
-<
17000
z 0.10 '~z""
-< 16000
tS -<
~ 0.05
15000
10 20 30 40 50 6d 4OOO
POSITION IN IIILLlIIETERS
Figure 3: A typical rotation (omega) vs. position plot for measurement of curvature.
Here w is the rotation to keep the diffraction peak at its maximum intensity in
a fixed detector. R is obtained from the slope of this plot 1.
Stress results from the sin 21/! and radius of curvature methods are summarized in table 3. As
indicated in the table, all results are + 15MPa or better, at three standard deviations, as calculated
from scatter around the linear model. At ftrst glance, one can make the following observations:
I. Stresses measured from the 311, 222 and 420 reflections seem to agree within experimental er-
ror.
2. In general, stresses measured using the 220 reflection differ signiftcantly from stresses obtained
using the above reflections.
3. The stress calculated from equations (2, 4) using experimentally measured R values exceeds the
stress from sin 2'" measurement in most cases.
4. Even for the thin films used in this study, there can be significant differences between the
modified Stoney formula, eqt. (4), and the more general fonnula, eqt (2).
One may obtain further infOlmation from table III, if it is analyzed in conjunction with table
II. Consider specimens KR7, KR8, KR9: These specimens have 10 /-lm of 5878 PIon them. The
PI was spun on and oven cured (80-400 °C, with 40 min at 400 0C). Then, a thin layer (2000A
or so) of electroless Cu was plated on the PIon specimens KR7, KR9. A thin film of Cu was,then,
electroplated on this "strike layer". Diffraction data was obtained from this final Cu layer. Specimen
KR8 was kept as a reference sample for the detennination of stresses due to PI deposition. It can
be seen that the bulk of the stresses are due to the Cu plating operation. For these specimens, the
differences between the stresses from bending and from the sin2'" measurements were significant,
but not very large ( ~ 30MPa.)
Specimens KRI- KR5 and NI-N4 demonstrate the effect of different Cu plating methods on
the residual stress distribution. The data indicates that electroless Cu plating (KR I) has signif-
icantly less stress than Cu evaporation(KR2, N3, N4). On the other hand, KR5, which is evapo-
rated Cu on Si without an adhesion layer shows low stresses from all the measurements, indicating,
possibly, a weak interface. Specimens KR3, NI, N2 are composite Cu structures. Here Cu was
evaporated on an adhesion layer and e1ectroless Cu plated on the evaporated layer. Thus, there are
two layers of Cu with different stress distributions. In such a case, one would expect the total stress
to be a weighted average of these distributions, since the entire Cu volume will contribute to the
diffraction peaks in the sin 2YI method. Since we measure the composite R in the radius of curvature
measurement, "R from this method will also be an average. Specimen KR3 meets these expecta-
tions, even though the average stresses are higher than expected. On the other hand, NI, N2 have
very low stresses, indicating that the average was dominated by the electroless plating. The cause
of this effect is unknown at this time.
Specimens N3, N4 present another puzzling case. N3, N4 have the same Cu that was evap-
orated on KR2 and all three specimens yield similar stresses from the radius of curvature method.
On the other hand, both N3 and N4 have low stresses if analyzed by the sin2'" method. One can
not simply ascribe such differences to the intrinsic stress distribution, because KR2 does not behave
the same way as N3, N4.
Specimen N5 is different from all the above specimens in that it has symmetric evaporated
Cr films on both sides of the Si wafer and an evaporated Cu film on the top of the Cr film on one
side. Thus, we would expect bendIng to be representative of the Cu film only. However, this spec-
imen has the smallest radius of curvature and the highest stress among the specimens presented
here. The stress from the sin 2'" method is significantly smaller than the bending stress. Again, fur-
ther experimentation is needed to determine the cause of the disagreement.
Conclusions
It may be seen from the results presented so far that the stress results from the two most-common
techniques of stress detennination do not always agree. The reasons they do not agree are not clear
at this time. One may make the argument, made here in the theoretical analysis section , that the
disagreement is due to intrinsic and/or interaction stress components. It may also be due to error
in the f!lm thickness. In any case, however, the results of this study indicate a need for combined
usage of the sin2'" and radius of curvature methods·. Even for well-characterized specimens, the
It must be noted here that the conclusions discussed above are valid irrespective of the method used
to measure R.
144 II. XRD CHARACTERIZATION OF POLYCRYSTALLINE THIN FILMS
scatter in the data is significant enough to preclude anyone method from claiming the "correct"
stress value. If only the sin2'" method is available, data should be obtained from more than one
reflection as a check.When multi-layer ftlms arc present, stress from as many layers as possible must
be obtained before one can claim to "know" the stress in the structure.
References
Abstract
Superconducting thin films (0.3-0.7Ilm) in the TI-Ca-Ba-Cu-O system have been
prepared on various single crystal substrates by sequential electron beam
evaporation followed by appropriate sintering and annealing. Oxygen-annealed films
show Tc as high as 11 OK and critical current densities to SOO,OOO A/cm 2. X-ray
diffraction analyses of these films show predominantly the TI2Ca2Ba2Cu301 0 phase
(c-parameter near 3SA), but some also contain up to 50 at% of the TI2CaBa2Cu208
phase (c-parameter near 30A). The complete absence of hkl reflections other than
001 demonstrates the highly oriented nature of the films as well as the absence of
other TI phases. The diffraction peaks are noticeably broader for the 3sA phase than
for the 30A phase. For a 0.71lm film such broadening is consistent with coherent sizes
along the c-axis of 1200 - 1400A and sooA, respectively, for the 30A and 3sA phases,
and of strain values near 1.4 - 1.8 x1 0-3 for both phases.
Introduction
The discovery of superconducting ceramics with transition temperatures (Td
above liquid nitrogen temperatures has stimulated development of superconducting
thin films. The materials which have generated the most interest recently are those in
the TI-Ca-Ba-Cu-O system. These are tetragonal perovskite-like structures with a-axis
dimensions near 3.8SA and c-axes of different lengths characteristic of each
phase 1-4. Tc for these phases are as high as 12SK1-4.
The quaternary nature of this system as well as the volatility of TI make thin-film
production more difficult 5. For this study samples prepared using a sequential
electron beam evaporation technique were analyzed S. These samples were
prepared using a starting stoichiometry of TI2Ca2Ba2Cu301 0, hereafter refered to as
2223, which was deposited on single crystal SrTi03 and Y-Zr02. The 2223 phase
has been shown to have Tc up to 125K1.
Our experience has shown that in bulk ceramics 2223 is difficult to form as a phase
-pure superconductor, suggesting that a phase transition mechanism exists between
2223 and other TI-superconductor phases. X-ray diffraction (XRD) was used to check
the phase purity of these films, as well as to determine orientation of the crystallites.
Line broadening analysis was used to determine strain vs crystallite size
relationships. In this paper we detail this XRD study.
Sample preparation
Films were depOSited on single crystal substrates of SrTi03 and Y-Zr02 using
sequential electron beam evaporation as described by Ginley et al 5. Cleaned
substrates were mounted into a Leybold-Hereaus ESV-6 e-beam evaporator loaded
with a four-pocket crucible containing metallic TI, Ca, Ba, and Cu. Sequential layers
of Cu-Ba-Ca-TI were deposited so as to start and end with Cu. Film thickness was
predetermined by varying the number of layers deposited. The net stoichiometry of
the deposit was near that of 2223. Chamber pressure was maintained at 1.5 X1 0- 5
mbar using a controlled oxygen leak. Film thickness for this study was 0.7I1m.
The materials were calcined at 850°C for 15 minutes in air followed by a 3O-minute
oxygen anneal at 750°C. A TI vapor overpressure was maintained by bulk 2223
ceramic during calcination and annealing in order to prevent loss of TI. The films with
I
a=2223
b=2122
I .
;;
0.0 '"
0
-'"0 ~
o~
0 '" en
E
en
e.
.!t
'"
0
e. .
co
0
e. .!t
'"
'"
e.
0
.0
.0
N . ;;
0
e.
8J
:E.~
0
.0
[ AlA
0
0-
~~ ~ T 0
J: V\
000
I. -d"-
4 18 32 46 E o
Two Theta (degrees)
I a=2223
b=2122
.0
c;;-
.i?:' o
e.
'iii
c .llt '"
~
~
.E '"e.
0 0
e.
Q)
.0
> c;;-
"iU o
'iii e.
a: III
<0'.0
.0 (; ;¢
_o 0~
<0'
0
e.", Q '" e.
N
<0
1l
0
4
1.11 118
1JJ
32 46 E o
Two Theta (degrees)
the best superconducting properties had Tc to 111 K and critical current densities
above 600,000 Ncm 2 as detailed by Ginley et a1 5.
X-ray Diffraction
XRD analyses were done using a Siemens D500 automated powder diffractometer
in a standard Bragg configuration using Cu radiation. The detector was a scintillation
counter preceded by a diffracted beam monochromator. The samples were mounted
onto a custom sample holder to insure alignment with the focusing circle. Data were
collected and analyzed using Siemens Diffrac 500 software.
In figure 1, 2223 is the dominate phase, but in many other films 2122 appears to be
in equal abundance. An example is shown in figure 2 for a film depOSited on Zr02,
Finally, in films containing both phases, the diffraction peaks for the 2223 phase
are broader than those for 2122. Figure 3 more clearly shows this point. This
observation was made regardless of the substrate material or dominant phase. This
precipitated a study to determine the cause of line broadening.
148 II. XRD CHARACTERIZATION OF POLYCRYST ALLINE THIN FILMS
4 5 6 7
Degrees (two-theta)
Figure 3. This enlargement of the low angle region from figure 2 clearly
shows that the diffraction peaks for 2223 are broader than those for 2122
(left to right respectively).
where B is the broadening of the peak less the instrumental breadth, e is the
diffraction angle and A is the wavelength of the x-rays used (Cu Ka1 = 1.540SA). DG
M. O. EATOUGH ET AL. 149
10 .,----_ _ _ _ _ _ _ _- - ,
-<
~
8
(0
b
~ 6 q, = 490A
X
C\J
2'
<D
4
'"o() 2
~
O+-~-.~_.,----~-.~~
0.00 0.02 0.04 0.06 0.08
(sin 8/')..)2
is the size value, and eG is the strain value. Thus the slQpe of the least-squares line
through the breadth function in each phase determines the strain (16eG 2) while the
intercept is the crystallite size functiQn ((1/DG)2).
Because of the peak overlaps a pseudo Voight function 9 was used to profile all
sets of peaks in each pattern. Instrumental brQadening was removed by subtracting
the peak breadths calculated from a well characterized, recently identified general
reference of LaB61 0 measured just priQr to. sample data cQllectiQn.
Figure 4 gives the plQtted data frQm a film deposited Qn SrTi03. It clearly shQWS
that broadening is due to both crystallite size and strain. Strain values for 2122 and
2223 are very similar ( eG = 1.4-1.5 x 10-3 ). This is cQnsidered a significant strain
value 7. Crystallite size, however, is very different fQr each with the size Qf 2122 abQut
3x that Qf 2223 (-500A vs -1400A respectively). The result for films Qn Y-Zr02
(Figure 5) are nearly identical to those fQr films Qn SrTi03. Strain values are slightly
higher for these films.
10,-__________________,
<
<0
8
b
~ 6
X
C\J
2"
(j)
4
8'" 2
23:
Figure 5. Results for film deposited on Zr02 are similar to those in figure
4 for films deposited on SrTi03'
Conclusions
The present studies show that even with a starting stOichiometry of 2223, 2122
often forms in abundance. Synthesis protocol appears to be very important. Ginleyet
al. 5 have shown that the configuration of the experiment contributes toward the ability
to form polycrystalline C-axis oriented films or randomly oriented films. The substrate
may also contribute by influencing epitaxial-like growth. Both substrate materials
(SrTi03 d1 00=3.905A, Y-Zr02 d11 0=3.60A) have dimensions near those of the
a-parameter (3.85A) for the TI-superconductor materials. Whether this mismatch of
lattice parameters causes the formation of dislocations that lead to nonuniform strain
is not yet proven.
The fact that 2122 grows to a much larger grain size than 2223 might indicate that
the films are first oxidized and calcined to the 2223 form, appearing as many relatively
small crystallites separated by dislocation boundaries, but still c-axis oriented. The
2122 phase is then nucleated at a few points and grows to larger crystallites through
a more gradual, "phase transition" driven mechanism.
Acknowledgements
The authors gratefully acknowledge the technical advice and review given by
Martin J. Carr. They also acknowledge the help with XRD work by R. G. Tissot and R.
D. Lujan. This work supported by the U. S. Department of Energy under contract
number DE-AC04-76DP00789
References
1. S. S. Parkin, V. Y. Lee, E. M. Engler, A. I. Nazzal, T. C. Huang, G. Gorman, R.
Savoy, and R. Beyers, Bulk Superconductivity at 125K in TI2Ca2Ba2Cu30x,
Phys. Rev. Lett., 60:2539 (1988)
M. O. EATOUGH ET AL. 151
Fergus E. Moore
ABSTRACT
INTRODUCTION
SmS
JCPDS 26-1479
\ 11
1'1
IJ"
RELATIVE
INTENSITY
Fig. 1. Diffraction pattern for 2000 A SmS thin film on fused quartz with
JCPDS powder standard 26-1479.
EXPERIMENTAL
1\
L
+
DETECTOR
-;_; ::0~THIN
INTERFACE
FILM
/
/
///
/ /
~
L
l ."'"ACnOH 'UN,.
-L THIN-FILM PLANE
/; 'i~
/ ~
Fig. 2. Diffraction geometry for low incident-angle Seemann-Bohlin
thin-film diffractometer.
F. E. MOORE 155
illustrates the geometry where ~ =So and the Bragg condition selects those
crystallites whose d·iffraction planes have an angle qI with the thin-film
plane. Following the methodology of Segmuller and Murakami,2 expressions
for the change in d-spacing caused by lattice strain may be developed with
the strain tensor, E, related to the stress tensor, d, by £ij = Sijkl okl
or £'ij = S'ijkl O'kl. While the compliance tensor, S, is most simply
expressed in the crystalline coordinates (unprimed), the stress tensor is
most simply expressed in the thin-film coordinates (primed). The stress
model, in which a constant stress field is applied in the film plane, is
given by 0'11 = 0'22 = s, 0'33 = 0, and o'kl = 0 for k + 1. The stress
tensor is expressed in the crystalline coordinate system by applying
rotational transformations to the primed tensor. Briefly, the expressions
in dyadic form are d' = s (I - L' L') and d = s (I - L L) where in the
unit tensor I = Okl and the unit vector L is perpendicular to the
thin-film plane as shown in Figure 2. In order to determine L in the
crystal coordinates one may relate it to the unit vector M perpendicular
to the diffraction planes. In effect, L is rotated about M in the
crystal coordinate system with LoM = cosql. The following expression
for (8d/dlhkl per unit stress, s, is derived.
(M/d)/s cos2q1 1 1 2
+ sin 2q1 So fl(hkl) - cos 2q1 So f2(hkl)
+ cosql sin1ll So [ sinS f3(hkl) + cosS f4(hkl)
+ sin 2q1 So [ sin2S fS(hkl) + cos2S f6(hkl) 1
where So 544/2 - ( 511 - 512 )
fl(hkl) m1 2 m2 4 + m1 2 m3 4 + m2 2 m3 2 + m1 2 m2 2 m3 2 ) 1 m2
f2(hkl) 2 m1 2 m2 2 + m2 2 m3 2 + m3 2 m12)
f3(hkl) 2 ml m2 4 + ml m3 4 m1 3 m2 2 - m1 3 m3 2 ) / m
f4(hkl) 2 m2 3 m3 m2 m33 ) / m
fS(hkl) ml m2 3 m3 - ml m2 m33) / m2
f6(hkl) m2 2 m3 2 - m1 2 m2 4 - m1 2 m3 4 - m1 2 m2 2 m3 2 ) / m2
and m m2 2 + m3 2
h dhkl/a, m2 = k dhkl/a, m3 = 1 dhkl/a
6.00 r - - - - - - - - - - - - - - - - - - - - - - - , 6.00
ao = 5.9527 A
II = -0.27
5.98 5.98
LATTICE LATTICE
o PARAMETER
PARAMETER 5 96 5.96
0_-
(A) . ~_-cr (A)
• _- 0
CaF2 O_~o1f- o. 0 •
5.94 -- --
SUBSTRATE _ _
2.3 kbar
COMPRESSION
.90-
5.94
5.92 5.92
-1.0 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1.0
all COS (2~) a .1.
Fig. 3. Stress analysis of SmS thin films on fused quartz and CaF2 sub-
strates. Parameters shown are results of least-square fits.
5.74 5.74
LATTICE LATTICE
PARA~ETER 5.72
(AI
• 5.72
PARAMETER
(AI
5.70 5.70
5.68 5.68
Fig. 4. Stress analysis of metal surface layer of polished SmS thin film.
Parameters shown are result of least-square fit with fixed aO'
5.90,---,,---,I,---,----rI----r----rI- . - .I-.-.-I-,---.--
1-'--'---1'-'-1'---'1,---,---,5.90
Si02 SUBSTRATE
I
+
I- 0.67 kbar TENSION -
•
-.=-0-.04----___~·:!~~+L~T~-I~r__J
[
~ ~ ~
J.-9--~
5.88r 5.88
LATTICE
ao = 5.8797 A ~ ~_ (y- LATTICE
PARAMETER
PARAMETER
(AI
I-
-rr-- - - -
(AI
1----- _ - - - - - I
5.861- 5.86
CaF 2 SUBSTRATE
I- 1.36 kbar COMPRESSION -
p = -0.49
5.84 1 1 1 1 1 1 1 1 I 5.84
-1.0 -0.8 -0.6 -0.4 -0.2 o 0.2 0.4 0.6 0.8 _ 1.0
COS (2;(-1 a.l
Fig. 5. Stress analysis of Sm.86Gd.14S thin films on fused quartz and CaF2
substrates. Parameters shown are results of least-square fits.
158 II. XRD CHARACTERIZATION OF POLYCRYSTALLINE THIN FILMS
ao = 5.7363 A
5.84 p = -0.49 5.84
10.4 kbar
5.82 TENSION 5.82
LATTICE • LATTICE
PARA~ETER 5.80 PARAMETER
5.80
(A) (A)
5.78
---
CaF;------_
SUBSTRATE
4.9 kbBr
5.76 TENSION
BJ.
Fig. 6. Stress analysis of Sm.82Gd.18S thin films on fused quartz and CaF2
substrates. Parameters shown are results of least-square fits.
28
24
Sm.78 Gd .22 S
... 220,440
20
o .111,222
FWHM 16 o .200,400
x COS (e)
x 103 12 A,s
(2e RADIANS) 12
DEPOSITION RATE
345 A SmS
SIN (e)
SUMMARY
OF ELECTROPLATED STRUCTURES
ABSTRACT
We report the results of a recent study where nickel substrates electroplated with chromium
were loaded in-situ on an x-ray diffractometer. This technique allows determination of lattice
spacings in the vicinity of the interface for both the film and the suhstrate as a function of the ap-
plied load. We used such lattice parameter data, SEM observations of the surface and x-ray peak
breadth data to study the partitioning of deformation hetween the film and the suhstrate. The data
indicates progressive loss of adhesion between the film and the substrate with increasing deforma-
tion. We observe significant effect of electroplating residual stresses on the mechanical behavior of
the system. The loss of adhesion between the film and the substrate coupled with the initial residual
strcss profile causes an apparent 'negative Poisson's ratio' for the film during initial stages of the
loading. This effect disappears with cyelie loading and unloading.
INTRODUCTION
Electroplated systems are used in many areas of engineering. They can be found in decorative
areas such as gold-plated jewelry, or in technical applications such as device interconnections in
micro-electronic circuits. As such, the mechanical response of plated systems to applied loads and
the integrity of the interface hetween the film and the suhstrate is a subject of significant techno-
logical importance. To characterize the interface and the mechanical response various tests are
used. These tests fall into two broad groups; (i) tests applied to the plated system (film plus the
suhstrate) and (ii) tests applied to free standing filmsl.
Signifieant among the tests applied to the plated system is the standard tensile tesP. In this
test, the applied load and total elongation of the plated system arc measured. Since the film is
usually very thin, the measured parameters arc dominated hy the response of the substrate. Little
is learned ahout the stress/strain distribution in the film or ahout the deformation concentration at
the interface. Interface integrity and strength can be determined hy the"peel" test 3 , whcre the load
required to peel thc film ofT the substratc is recorded. This test, however, is gcometry sensitivc and
Advances in X-Ray Analysis, Vol. 33 161
Edited by C.S. Barrett et al.
Plenum Press, New York, 1990
162 II. XRD CHARACTERIZATION OF POL YCRYST ALLINE THIN FILMS
is hcst used as a comparative technique. It can not he applied with ease to all specimens, especially
to hrittle films. Furthermore, stress/strain partitioning between the film and the substrate can not
he studied by this method. As a result of the complications associated with the tests described
ahove, significant research effort has gone into the determination of the mechanical properties of
free standing films4. These films are tested in uniaxial tension on thin-film testers, or in bi-axial
tension on bulge testers2,5. The properties measured by such methods will be only those of the film.
The interface between the film and the substrate, and its effects on the mechanical properties can
not be studied by these methods. Reconstructing the properties of the plated system from the
stand-alone properties of the film and substrate is not a trivial task. It requires, in addition to these
properties, information about the interface and its effects.
The fundamental parameters which influence the mechanical response of a plated structure
to an applied load are the elastic constants (Young's moduli and Poisson's ratio) of the substrate
and the film. In addition, configurational parameters, which depend on the past treatment and ge-
ometry of the structure, influence the mechanical response. These include the yield and fracture
stre~ses and the respective thickness's of the substrate and the deposit. Other configurational pa-
rameters that can strongly influence mechanical response arc the residual stress fields within the
system, defect types and distributions near the interface and interface geometry. Testing free-
standing films alters the residual stress field of the e1ectrodeposit6. It also bypasses the clastic
compatibility requirements imposed upon the system due to the presence of such an interface.
Hence the results obtained from the mechanical testing of free-standing deposits, uniaxial tensile
testing or biaxial bulge testing, may yield misleading information.
Thus it is necessary to be able to distinguish the elastic response of the individual components
of the plated structure to an applied load without interfering with the structure. The standard
method of x-ray elastic constant determination utilized routinely in x-ray residual stress analysis 7 is
an ideal testing method that can he applied to this problem. This application is demonstrated in the
current study.
THEORY
Assume that a homogeneous load F is applied along the XI direction to a plated system. The
strain at any point within the system (film or substrate) is given by:7
(I)
I lere e~ is the strain due to the applied load, eij is the residual strain in the system due to plating
residual stresses, and Eif is the incompatibility strain that arises because of the presence of the film-
sllh~trate interface. This term will be zero if the elastic constants of the film and substrate are equal
and if the interface is "perfect".
The strains given hy equation (I) will affect the atomic plane spacings along a given axis
by the relation:
(2)
where Ei,res is the resolved net strain along the axis i. The changes in d j can be detected by x-ray
diffraction. This forms the basis for the majority of x-ray stress/strain determination techniques.
In an x-ray diffraction experiment using the traditional sin 2 t/' technique, the plane spacing along a
given axis II (figure I), is given by:7
d<{lift - do 2 . 2 . 2
d = {Ell cos cP + E22 sm cP - En} sm tit + E.l3 (3)
o
In this equation, the shear terms Eij , iii arc assumed to be zero. In the cases of electroplating, this is
required by the equations of equilibrium and by symmetry considerations.
G. SHEIKH AND I. C. NOYAN 163
Figure I: The sample coordinate system Xij and the direction of the measured x-ray stress (T4>'
d"'I/t - do I- v . 2 v
do =~ (T", Sill '" - E (Tkk
(4)
Here the stress (T 4> is the stress in the S~ direction (figure I) and (Tkk = L(Tii' Again, as in equation
(I), (T ~ is the total stress and contains contributions from the applied, residual and interaction stress
fields;
(5)
~
Here the terms are analogous to those in equation (I). For a uniaxial load F applied to a plated
specimen along the XI direction, all (T~ are zero except (T~I' and all (Tij are zero except (TIl = (Th. The
interaction stresses (Tir depend on geometric factors as well as material parameters and all of these
tensor components may be locally finite'·8.
Equations (4) and (5) form the basis of x-ray tensile testing+. In this procedure, the load is
applied through a jig that is mounted on the diffractometer and the strain is read through the x-ray
measurement. This strain is an x-ray average of the microstrains in the diffracting volume of the
particular V' tilt. Since, because of diffraction, it is confined to either the film or the substrate, all
the data is obtained from a particular volume. Purthermore, because of absorption, the data can
be confined to the region around the interface in the substrate by choosing the fIlm thickness or the
wavelength of the incident radiation. Thus, this method can be used to study the strain effects in
the first couple of 11m of the substrate near the interface. Of course, with such radiation, one ob-
tains data from the entire thickness of the film.
EXPERIMENTAL PROCEDURE
In this study the chromium-nickel system was used. The nickel substrates were machined
into dog-bone type tensile specimens from commercially available rolled sheets (~99.2% pure). The
dimensions of the gage section were nominally (27.00 x 6.36 x 0.77) mm. The specimens were me-
chanically polished (240, 400, 600 grit paper, 6, 3 micron diamond paste and 0.05 micron alumina)
to a mirror finish. They were subsequently annealed at 600C for four hours in a reducing atmos-
phere (forming gas) amI furnace-cooled. The specimens at this point exhibited mild texture.
tOne can also plot the "slope" of the d~1l vs. sin 2V' data vs. the applied load (eqn. 4). This is the
traditional method of determining x-ray clastic constants. However, it can also be used to detect the
onset of micro - yielding'.
164 II. XRD CHARACTERIZATION OF POL YCRYST ALLINE THIN FILMS
Chromium was, then, electroplated on the nickel substrates from a hexavalent chromium acid bath
(250 gil of er0 3 and 2.7 gil H2S04 ), Two separate plating conditions were used yielding bright
deposits·,ID. Plating at sse and 0.4 amps/em2 resulted in smooth deposits with no visible surface
defects even at high magnification (> 5k). Plating at room temperature and 0.08 amps/cm 2 resulted
in crazed deposits, where the surface had fme lines, or crazes separating large, smooth regions. The
deposit thickness's were in the range of 1-3 Ilm.
The plated specimens were, then, placed in a tensile device which, in turn, was mounted on a Picker
horizontal diffractometer. The load was applied step-wise and it was monitored by a load cell.
i
0u-__ ~ ____ ~ __ ~ ____ ~ __ ~ ____ ~ __ ~ ____ ~ __ ~ ____ ~~
Figure 2: Variation of the x-ray stress,O'~ (2-a) and the interplanar spacing at tit = 0; dhk',~~o
(2-b) with the applied load in the Ni substrate.
G. SHEIKH AND I. C. NOYAN 165
Filtered copper radiation, with the bcam centered in the gage length, was used. The 331 and 420
reflections of nickel at 20 Bragg = 1450 and 1560 respectively, and the 222 reflection of chromium
at 20 Bragg = 136° were examined during each step of the loading.
EXPERIMENTAl, RESUlTS
for each load, the complete sin 2 '" stress analysis was carried out for the Ni substrates. Represen-
tative data is shown in figure 2-a. It can be seen that Ni has compressive residual stresses at zero
applied load. This is quickly overcome by the applied tensile stress and the relationship between the
applied load and the x-ray stress is linear during load application. In addition, the d¢ll at '" = 0,
which is the (311) plane spacing in the direction of the surface normal x] was monitored (fig. 2-b).
This plane spacing dccreased monotonically with increasing applied (tensile) load along the XI di-
rection. This is expected, since Poisson's contraction must occur in all directions normal to the
tensile axis. Similar results were observed for the (420) planes. Thus, one can conclude that Ni is
behaving "normally" in its response to the applied load.
The x-ray analysis of the film was more complicated. The chromium deposits were highly
textured and the perfection of the (III) orientation increased with increasing thickness of the de-
posit. The chromium peak examined, 222, virtually disappeared at if/-tilts > 10°. llecause of the
limited if/ range, only the variation of the (222) plane spacing in the X3 direction ('" = 0) was mon-
itored as a function of the applied load. This data is shown in figure 3-a. It is seen that there are two
distinct regions: dm.~~o first decreases with applied load, showing the expected Poisson's con-
traction. Then, after a critical transition point (Ter it increases rapidly, showing effectively a negative
Poisson's ratio. All specimens examined exhibited similar behavior with the transition occurring
at applied stresses of 125-175 MPa. This behavior persisted regardless of the status of the deposits
(crazed or craze-free).
If these specimens are cycled in load; i.e, loaded, unloaded and loaded again, a ratcheting
mechanism is observed in this anomalous expansion (figure 3-b). The plane spacing does not relax
to its initial value after the first unloading step but relaxes to a higher value. If the film is loaded
again, we see some Poisson contraction until we reach a new transition load (Ter' However, in this
case, (Ter increases to approximately the highest load encountered by the specimen in the previous
loading cycle. After the transition point, we again observe some anomalous expansion of dm,~~o
with increasing tensile load. Upon unloading, the plane spacing dm.,H relaxes to value higher than
the prrvious cycle. At this point, further loading or unloading has no effect on the plane spacing
dm .,.-" from thc elcctrodeposited film.
The shapes and breadths of the x-ray diffraction peaks were also monitored during the x-ray
analysis. The peak breadth at half maximum intensity (1'.Il.lI.M.l.) of the chromium deposit re-
mains constant (within the margin of error) over the three cycles of loading. The Ni peak, however,
did show effects of plastic flow: Prior to deformation, the nickel substrate exhibited relatively 'sharp
and resolved peaks. The K." K.2 from the 311 and 420 reflections became broad and unresolved
when the load exceeded 150 MPa during the first loading cycle. This indicates some plastic flow
within the diffracting volume. However, towards the end of the second loading cycle, the peak was
resolved again, indicating some recovery.
The surfaces of the specimens were examined under scanning electron and optical micro-
scopes before and after each load cycle. After the transition point aen severe cracking was observed
in all Cr films. These cracks ran across the width of the specimen in the plane of the film, and were
normal to the tensile axis (Mode I cracks). In the crazed specimens, these cracks cut across the
craze-free regions. In addition, some crazes that were normal to the load axis enlarged considerahly
and hecame comparable in width to the cracks caused by the loading. Thc mode I cracks extended
over the edge anc! through the layer of chromium platcd on the sides. To check the interface,
transverse cross sections were prepared using an electron discharge machine. Cracks extending
parallel to the interface and occasional cracks parallel to the tensile axis were observed (Mode II
cracks). All cracks were limited to the gage length and ended at the grips of the tensile device.
166 II. XRD CHARACTERIZATION OF POLYCRYSTALLINE THIN FILMS
The peak-breadth data indicates that the Ni suffers plastic flow in the range of 125-175 MPa,
whereas the Cr film docs not. The incompatibility in strain across the boundary causes cracking in
the Cr film. Microscopic examination of the surface reveals both Mode I and Mode II cracks.
Thus, we have delamination and cracking of the film when the Ni goes into plastic flow. The
Ocr a
I
I
I
I
100 150
Applied stre•• (MPa)
Chromium Deposit Interplanar Spacing
2
+3
o 200 250
Applied StreB. (MPo)
I'igure 3: Variation of dm,~-o with the applied load "~I in the Cr film. When the critical stress
"cr is exceeded, the lattice spacing normal to the applied tensile load starts expanding (3-a).With
cyclic loading, a ratcheting mechanism is observed in the anomalous expansion (3-b).
G. SHEIKH AND I. C. NOYAN 167
anomalous expansion in the (222) plane spacing in the direction of the surface normal d 222 • \H' can
he explained hy the relaxation of plating residual stresses. If we re-arrange equation (4):
I -u r
(hkl,o/I=O = d0 + do( E ){ 11110 + "kk) (6)
(7)
For applied stresses larger than "eI' the observed variation in d hk, • ~ ~ o is possible if and only if
orrh/O(Tapp. < -I . Since the change in (Tapp. is positive, the change in the residual stress term must
be negative. Given the houndary conditions of this experiment, this is possible only by relaxing
ten~ilc rc~idual stresses in thc chromillm film.
It is possible to quantify the rate of this (residual stress) relaxation with the x-ray data. The
plot of d hk" ~=O vs. cr. pp. approximates a straight line for this specimen. If the experimentally measured
slope of this line is "M.", equation (7) can be written as:
-u aUkk
M<1=d o(T)(1 +-a-} (for "app. > "cr) (8)
, U app.
Thus, in the case of this specimen, the change in residual stresses is linearly related to the applied
load:
Here, (Ukk)'cr. is the total residual stress within the film hefore relaxation starts, C is a constant of
168 II. XRD CHARACTERIZATION OF POL YCRYSTALLINE THIN FILMS
Numerical analysis of the data using the above equations will be presented in a later article.
The analysis also permits qualitative evaluation of interface damage: The relaxation of the
residual stresses in the Cr film can be due to two possible causes;
I. Plastic flow in Cr.
2. Delamination of the interface.
Since x-ray data indicates negligible plastic flow in the Cr, most of the relaxation is due to the de-
lamination of the interface. In a recent study it was found that the cracking of brittle films may be
followed by delamination if the substrate possesed high yield strength (stainless steel) and if the
interface was not "clean". Low yield strength substrates (aluminum) enhanced film cracking tend-
ency but also inhibited delamination by blunting film cracks at the interface". Nickel lies on the
higher end of this spectrum and the lower fracture toughness of the interlayer may be due to the
presence of contaminating residues from the annealing process.
Assuming a homogeneous residual stress distribution in the film, as well as a homogeneous
distribution of deformation/damage due to the applied load, one can also conclude that delami-
nation is linear with the applied load. After the second loading cyc-le, delamination has reached
saturation and very little force is transmitted from the substrate to the film. Thus, no change at all
is seen in the d hk1 , H from the film in response to the applied load. The film is still attached to the
suhstrate in spots, since we did not observe any large scale spalling of the film. In such anchoring
spots, dhk1 , ,H mllst change. The fact that we do not see any change in the x-ray data indicates that
the total volume of such anchoring spots is much smaller than the delaminated regions·.
SUMMARY
Nickel substrates electroplated with Cr were loaded into the plastic deformation range on an
x-ray diffractometer. Plane spacing data as a function of load from the substrate and the film were
used to study the partitioning of deformation in the system. Progressive loss of adhesion was ob-
served above a critical load (Tn This loss of adhesion, coupled with plating residual stresses causes
an apparent "negative Poisson's ratio" in the film.
X-ray tensile testing is a novel and unique method which can be used to obtain qualitative and
quantitative information about the mechanical properties of plated interfaces. This technique can
be casily used as a non-destructive quality control tool in production.
ACK NOWI.EDGEMENTS
The authors would like to thank Dr. Dave Rath of IBM Yorktown for significant assistance
with plating theory and practice and many helpful discussions. Thanks are also due to Professor
Juan Sanchez of Columbia University for helping and encouraging us at all stages of this study.
REFERENCES
I. D. i\. Hardwick, The Mechanical Properties of Thin Films, Thin Solid Pilms, 154:109 (1987).
INTRODUCTION
In this paper, the change in the sin 2 1jJ diagrams obtained from y(552)
planes in Zn-Ni-alloy electroplating layers under different conditions are
shown first. Next, some fundamental problems in the x-ray stress
measurement in thin layers are discussed.
EXPERIMENTAL METHODS
Materials
The effect of heat treatment was examined on sample #4. Samples were
exposed to temperatures of 100, 200 and 300 oC, respectively, in vacuum for
two hours and a slow furnace cooling was applied.
X-Ray Instruments
~=f\ ~ 1:.
!: /
0
"'. ;' 0 B .' -
; .r." B ./ 0 "'~" \ 0
i \y
C \0 \.'. .... e,'
~O-A
c C aDa+--A
(a) (b) ( ) e...-- C
. 121 6'-:-_ _
a
---,,-=-__
.35
---'
.7 a
'"::-__--:~--~'-::-c---':::---~
.35 .7 a .35 .7
s i n2 1/J S i n2 I/J s i n2 I/J
Fig. 1. Sin2~ plot for the 552 reflection from Zn-Ni-alloy electroplating
layer obtained by (a) instruments A and B with Co-Ka, (b) instru-
ments A and B with Fe-Ka, and (c) instruments Band C with Cr-Ka.
T. SASAKI ET AL. 173
:i o
,
.35 .7
The effect of the thickness of the plating layer on the sin2~ diagram
was also examined with sample numbers 2, 3 and 4; we obtained the result
that the differences in the sin 2 ~ diagrams were very slight.
Figs. 3 and 4 show the diffraction profiles and the sin 2 ~ diagrams of
Zn-Ni plating after heat treatment. We can see that the sin2~ diagrams show
a complete change in shape, and a linear distribution is seen at 300 0 C; new
peaks emerge near 26=130 0 with an increase of temperature.
.122
r (552)
'-,..
7000
counts
e
,S
'c"
U e
...'"
co
100' e .1216
co
c
e !!
.,.'"
200'
300' e
300' e ~
e
500' e
.............. 600' e
======::::::=::::======~.~
120 130 140 150 700' e .7
28 (deg.) s i n2 1/J
Fig. 3. Diffraction profile for Zn-Ni Fig. 4. Sin2~ plot for the 552
alloy electroplating layer reflecting after heat
after heat treatment. treatment.
174 II. XRD CHARACTERIZATION OF POL YCRYSTALLINE THIN FILMS
5°1
'"' 22°1 r (552) (a)
I a re(211)
oS
~ r~---\
;., 200
I;
. .
\\ ':"
\.:
-50
, , '\
: : A\
(I)
(I)
.......'"
Vl
'''~~/''''''\''
180 -1 5 0 !:-=---'------'="----'---,--,J
- 6 ~o0::----'----';;-0---'---'6:-::'0'0 - 600 0 600
Strain (X10- 6 ) Strain (X10- 6 )
Fig. 5. Stresses during bending process in (a): the plating layer
and (b): the substrate.
surface and the boundary to the substrate were observed before the heat
treatments; the distribution became uniform after the heat treatment at
300 0 C for two hours. Such a distribution of Zn may cause the sin2~ diagrams
to be parabolic.
Fig. 5 shows the x-ray stresses in both the plating layer and the
substrate of sample #4 which was kept at 300 0 C for 2 hours while subjected
to a four-points bending. X-ray stresses in the substrate, aFe(2ll), are in
proportion to the applied strain. But x-ray stresses in the plating layer, y
(552), calculated by means of E=78 GPa, v=0.25, 8=70.5 0 are decreasing
during the tensile loading process and then change another way during the
stress releasing process. This result is thought to be caused by the
initiation or the growth of cracks in the plating layer. The residual
stress was released due to the removal of restriction.
CONCLUSIONS
(1) It was found that the sin2~ diagram for y (552) planes of the Zn-Ni-
alloy electroplated layer showed a parabolic shape.
(2) The change in the sin 2 ~ diagram due to changing the x-ray radiations
is very slight. This can be explained theoretically as being typical
of thin layers.
(3) High concentrations of Zn near the surface and the boundary to the
substrate were observed. These concentrations may be related to the
shape of the sin2~ diagram.
ACKNOWLEDGEMENTS
The authors are grateful to Dr. M. Katayama, Dr. K. Kyono and Dr. C.
Kato of Kawasaki Steel Co. for kindly supplying the samples and much helpful
T. SASAKI ET AL. 175
REFERENCES
M. J. Sullivan, C. Goldsmith
P. Ficalora
INTRODUCTION
2.EXPERIMENTAL PROCEDURE
An ultra high vacuum system was used for these experiments. The
main pump was a Varian "Vac-Ion" pump capable of pumping 200 l/sec.
This unit was backed by two oil-free pumps used for rough pumping. A
venturi jet pump was used for light roughing and a Var~3n "Vac-Sorb"
sorption pump was used to attain pressures of 1.0 x 10 torr. Using
M. J. SULLIVAN ET AL. 179
The RBS data was obtained using the instrument located at the IBM
Thomas J. Watson Research Center. Samples were first mounted on a
go~~ometer in a vacuum chamber and pumped down to approximately 1.0 x
10 torr. The samples were then bombarded with monoenergetic, singly
ionized helium ions at 2.3 MeV at current levels of 20-50 nA over a 2
mm x 2 mm area. The backscattered ions (170° take-off angle) were
analyzed using a silicon surface barrier detector which produces a
voltage pulse that is proportional to the energy of the individual
backscattered ions. This system has a sensitivity of 0.1 atomic % for
heavy elements.
- ..... ' (2) " ' ' ' ' ' '
o
o
o
<D
V1
o
1-0
o
z"ot
:::J
o
U
o
o
o
N
1.2 1.6
Backscatiering Energy (MeV)
The sheet resistance of the thin film couple is 0.208 ~/o and the
resistivity of the copper layer can be calculated assuming that the
titanium and copper layers are parallel resistors. The copper resistivity
is calculated to be 3.07 ~~-cm which is a factor of almost two greater
than the bulk value [16]. This is not unexpected since the resistivity of
copper is very sensitive to the presence of small amounts of impurities.
TABLE I
X-Ray diffraction results for
titanium/copper - time zero
d-sEacing (X) 1/10 (%) Ehase hkl
2.5553 12 Ti 110
2.3440 43 Ti 002
2.2445 52 Ti 011
1. 8032 44 Cu 200
1.7262 22 Ti 012
1.4762 4 Ti 110
1. 3325 10 Ti 103
1.2775 70 Cu 220
1. 2472 45 Ti 112
1. 2307 22 Ti 201
M. J. SULLIVAN ET AL. 181
1.0
0.8
00.6
"-
c
Q)
u
<=
III
~ 0.4
·M
Ul
Q)
p:;
....,
Q)
Q)
ti O. 2 ....:::<::::::::=:~->.~
o
o
Time (hrs)
Figure 2 Variation of sheet resistance with
heat treatment in vacuum
o
- - III$.lO
--l<>
tOO
(1 t) 511/1117
(1)2/1l1li
200
-
o
o
'"
(fl0
0
f- 0
z ~
::>
0
u
0
0
0
'"
TABLE II
X-Ray diffraction results for
titanium/copper - 368°C/lhr/Vac
d-spacing (X) 1/10 (%) Phase hkl
2.5475 7 Ti 010
2.3405 23 Ti 002
2.2402 55 Ti 011
2.0827 85 Cu 111
1. 8018 34 Cu 200
1. 7258 30 Ti 012
1.4737 5 Ti 110
1.3348 6 Ti 103
1. 2783 65 Cu 220
1.2339 26 Ti 201
Also, it has been shown that the impurities introduced during thin film
deposition can increase resistivity by factors of several hundred times
bulk values. Copper also forms a number of oxides and would be expected
to react with the residual gas during deposition. The amount of oxygen or
other impurities in the copper is, however, less than the AES sensitivity
of approximately 1.0%.
TABLE III
X-ray diffraction results for
titanium/copper - 450°C/0.75hrs/Vac
d-spacing (X) 1/10 (%) Ehase hkl
2.2369 78 Ti 011
2.0841 27 Cu III
1. 9882 90 TiCu 3 021
1. 8022 15 Cu 200
1.7301 8 Ti 012
1.4791 16 Ti 110
1.2783 16 Cu 220
1.2313 23 Ti 201
I ~ 1\\\)1
I I~I
4S0 o C
SO
42S o C
E
.5
til
til
60 0
OJ
c:
".
u 36S oC
°rl
.c:
.j.l
:l
U 40
°rl
8
3S0 o C
20
0
0"=1
0 1 2 3 4
Time (hrs)
TABLE IV
X-ray diffraction results for
titanium/copper - 360°C/1hr/H 2
2.0826 82 eu III
1.8003 38 eu 200
1.2779 93 eu 220
186 II. XRD CHARACTERIZATION OF POL YCRYSTALLINE THIN FILMS
4. CONCLUSIONS
5. REFERENCES
10. B. S. Barry and I. Ames, LB.M. J. Res. and Dev., 13 (1969), pg. 286.
16. T. Lyman, editor, Metals Handbook Volume I, ASM, Metals Park, Ohio
(1961), pg. 48.
James E. Willis
INTRODUCTION
Thin film samples represent ideal samples for analysis using X-ray fluorescence in that they
tend to be flat, smooth and homogeneous. However, before a sample can be characterized
using XRF, several questions must be answered. Some of these questions are:
IJ (E )
[ 1 - exp ( _ ( 1 pr1 + (Eg. 1)
51n("'1)
"'lV"2
1
Figure 1-Schematic of the measurement of a single layer thin sample. The sample
geometry is defined in terms of the incident angle '1'1> takeoff angle '1'2 and the thickness t1.
where
The relationship between the thickness and relative intensity of a single layer thin film
containing a single element is characterized by two fundamental parameters: (1) the mass
absorption coefficient for the incident X-rays and (2) the mass absorption coefficient for the
analyte X-rays. The sum is these two terms is the reciprocal of the lie thickness and has
units of [mg/cm2].
The reciprocal of the mass absorption coefficient for the incident X-ray beam is an
indication of the thickness of the sample for the incident beam. The reciprocal of the mass
absorption coefficient for the analyte X-rays is an indication of the thickness of the sample
to the analyte X-rays. The reciprocal of the sum of these two terms is an indication of the
analytical range of the technique.
Figure 2 shows the curve which relates the measured intensity versus thickness for a single
layer thin film. This curve is characterized by two parameters: (1) the intensity for a bulk
sample and (2) the thickness where the intensity reaches I-lie or 63% of the bulk.
Historically several terms have been used to describe the measurement of thin films in each
of three regions.
The thin region extends from zero thickness to a thickness where the film yields an
intensity of about 10% of the bulk. In this thickness region, the measured intensity is
proportional to the sample thickness and to the concentration of the analyte. This is another
way of saying there are no matrix effects. Samples in this thickness range are often
analyzed using empirical working curves established by measuring one or more similar
type standards.
...>
in
z
...w~
THICKNESS
Figure 2 - The curve of intensity vs. thickness is characterized by two parameters; the
intensity for a bulk sample of the same material and the thickness where the intensity
reaches 63% of the bulk (the lie thickness).
J. E. WILLIS 191
6
N
E
~ 5
§.
(f) 4
(f)
w
z
Q3
I
f--
10 20 30 40 50 60 70 80
ATOMIC NUMBER
Figure 3 - The lie thickness versus atomic number for elements from sodium to uranium.
The curve on the left is for Ka lines, the curve on the right is for La lines.
The critical thickness is described as that thickness where a further increase in sample
thickness causes no corresponding increase in analyte intensity. Such a sample is also said
to be infinitely thick. Most matrix correction techniques and analysis programs were
developed for use in this thickness range.
Between the thin region and the critical thickness is the intermediate region, where
matrix effects are present in varying amounts and the analyte intensity is not linearly
proportional to thickness.
For a given acquisition condition, the lie thickness can be calculated as a function of atomic
number. Figure 3 shows the curve of lie thickness versus atomic number for elements
from Na to U. The curves are suprisingly flat, ranging from I [mg/cm2] for Na Ka lines to
10 [mg/cm2] for Mo Ka lines and from 0.5 [mg/cm2] for Sn La lines to 3 [mg/cm2] for U
La lines. The analytical range of the technique for single layer thin films is about twice the
range shown in the graph.
Figure 4 - Schematic of a two layer thin film system characterized by the incident angle '1'1>
take-off angle '1'2' thickness of the top layer tl and thickness of the bottom layer t2.
192 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
o
-"
u"
Au THICKNESS
Figure 5 - Measured intensity for eu Ka X-rays from a set of NBS Au on eu thin film
samples. In these samples the eu layer is infinitely thick.
The relationship between the relative intensity for the analyte present in layer 2 and the
thickness of the layers in a two layer thin film sample is given in Equation 2.
(Sq. 2)
The measured intensities for eu Ka X-rays from a set of NBS Au on eu thin film samples
are shown in Figure 5. In this example the Au in layer one ranges from 1.5 to 13
[mg/cm2], while layer two is an infinitely thick eu substrate. The measured intensity from
the eu layer as a function of the Au thickness is characterized by absorption of the primary
X-ray beam and the eu Ka (analyte) X-rays in the Au layer.
For this example, the top term in Equation 2, which represents the intensity from the eu
layer as a function of the eu layer thickness, is one, since the eu is infinitely thick.
MULTIPLE LAYER THIN FILMS
The relationship between the measured intensities and the thickness and composition of the
layers in a multiple layer thin film sample can be understood in terms of a simple working
model. In this model the measured intensities from the first layer depend only on the
thickness and composition of the first layer. The intensities from the second layer depend
on the thickness and composition of the second layer as well as the thickness and
composition of the first layer, where the first layer acts as a transmission filter. The
intensities from the Nth layer depend on the thickness and composition of the Nth layer as
well as the thickness and composition of the intervening layers.
The determination of the thickness and composition of multiple layer thin film samples is
not easily handled using empirical techniques. In this simplified model, which corrects
only for X-ray absorption, the calculation of an unknown sample is very complicated and
requires many linear and non-linear corrections. Empirical techniques can be used for
selected cases of only a few elements in a few layers, however, as the number of elements
J. E. WILLIS 193
and layers increases so does the number of similar type standards required for calibration.
Because of this requirement for many similar type standards, the calculation of the
thickness and composition of multiple layer unknown samples is best handled by a
fundamental parameters analysis program.
Fundamental parameter equations calculate theoretical intensities for hypothetical
samples when given the spectrometer geometry, X-ray tube operating conditions,
composition and thickness of any primary and secondary filters and thickness and
composition of the hypothetical sample.
Figure 6 shows the percentage of secondary intensity from the enhancement of Fe by the
Ni plotted as a function of thickness, when the Ni is below the Fe. In this hypothetical
case, the Fe layer and the Ni layer are the same thickness. The intensity due to the
enhancement of Fe from the Ni reaches a maximum of about 6% at a thickness of a few
[mglcm2J. If the secondary fluorescence contribution due to the Ni is neglected, the
measured thickness of the Fe would be about 6% high.
6
>-
f--
Cil
Z
W
f--
Z
o o~----------------------------------------~
20
THICKNESS (mglcm 2 )
40 r---------------------------------------~
>-
t-
u;
Z
UJ
t-
Z
o ~______________________________________~
o THICKNESS (mglcm 2) 20
Figure 7 shows the percentage of secondary intensity for Fe when the Fe is underneath the
Ni. In this case the fraction of Fe intensity due to enhancement by the Ni increases as a
function of thickness until it reaches about 30% at a thickness of about 16 [mglcm2]. If
secondary effects are neglected in the measurement of the Fe thickness for this sample, the
measurement could be more than 30% high.
// :: /////////////////////// +/:/
///////////////////////////////////
///////////////////////////////////
Figure 8 - Schematic of a Ni thin film on top of a bulk sample of Sn. This sample is
unusual in that the intensity from the Ni thin film layer exceeds the intensity from a bulk
sample of Ni.
J. E. WILLIS 195
140~--------------------------------------~
OL-______________________________________~
Because of the high energy of the primary beam, it is transmitted by the Ni to the Sn and
produces characteristic Sn X-rays. The Sn X-rays are then absorbed by the Ni and
produce secondary fluorescence. The graph of Ni intensity vs. thickness, shown in Figure
9, has two unusual characteristics: (1) it exceeds 100% relative (relative to a bulk sample
of Ni excited using the same conditions) and (2) the intensity decreases with increasing
thickness between about 20 and 100 [mg/cm2J.
CONCLUSION
A relatively simple model for understanding the X-ray production and absorption in single
and multiple layer thin film samples is useful for predicting the measured intensities from
thin film samples but may not provide adequate accuracy for characterization of samples if
secondary fluorescence effects are ignored. Because of the complexity of the problem and
number of standards that would be required for calibration of an empirical analysis
program, the calculation of unknown samples is best handled by a fundamental parameters
analysis program.
Acknowledgement
This work was supported in part by Tracor Xray, 345 E Middlefield Rd, Mountain View
CA 94043.
REFERENCES
1. G. Pollai, M. Mantler, H. Abel., Spectrochemica Acta, 26B : 747 (1971)
3. T.e. Huang and W. Parrish, Adv. X-ray Anal., 29: 395 (1986)
Rigaku/USA, Inc.
Danvers, MA
INTRODUCTION
The purpose of this paper is to show how XRD and XRF can be
used as complimentary tools to determine multi-layer thin film
composition, both elemental and crystalline, as well as film
thickness.
FUNDAMENTAL PARAMETERS IN FLUORESCENCE
Using the Fundamental Parameters approach in XRF, the
thickness and composition of a multi-layer thin film can be
simultaneously determined for up to four layers. Conventional
XRF using an empirical method requires many standard samples
which must be of similar composition to the unknown. In the
case of multi-layered thin films, producing such standards can
be difficult. A large number of standards would be required
for a three or four layer film, making analysis impractical.
Using the Fundamental Parameter approach, one needs only to
produce a single-layer thin film, or have a pure metal of
"inf ini te thickness" as a standard. From these standards,
instrument sensitivity can be calculated. This in turn is used
to characterize multi-layered thin films for which no "matrix
matched" standards exist.
In order to obtain XRF data from lower layers in multi-
layered film, a high intensity source and a large incident/take
off angle are required. This geometry is achieved through the
close coupling of the end-window Rh tube, sample and detectors.
THIN FILM DIFFRACTOMETRY
Conventionally,e/2e XRD geometry is not very useful for the
study of thin multi-layered films because of poor sensitivity
and interference from the substrate.
Table 1. Equipment
X-RAY DIFFRACTION
12 KW Rotating Anode
Thin Film Attachment
Horizontal Goniometer
21theta scans with theta fixed at: 0.5 degrees
2.0 degrees
4.0 degrees
8.0 degrees
IBM PS/2 Model 80 12.0 degrees
X-RAY FLUORESCENCE
3371 Sequential X-Ray Spectrometer
Data-flex Software with on-line Fundamental Parameters Program
IBM Model 70
2800
~ 2100
~
Q
W
S
0
1400
::Ii
f/)
700
10 20 30 40 50 60 70 80
2 THETA
Figure 1. Standard powder diffraction geometry wherein
9/28 scan shows large broad hump from glass substrate as
well as not revealing the 100% relative peak for ZnO in
the ZnO/Ag sample.
R. A. BROWN ET AL. 199
350
280
~
a.. 210
It)
~
C
W
::&:
I-
0
0 140
::;
rn
70
10 20 30 40 50 60 70 80
2 THETA
AU=GOLD 3000
2500
2000
iii
Ii:
It)
1500
C
III
:r
I-
0
0
:;; 1000
rn
499
AU
AU
AU
20 30 4D 50 60 70 80
2 THETA
Figure 3. Theta set to 0.5 0 for Sample 1. Reveals small
Au peaks.
AU=GOLD M=MOUNTING MEDIUM 3000
2500
AU
2000
AU
2 THETA
Figure 4. Theta set to 2 0 for Sample 1. Large increase
in Au peaks as well as appearance of a peak "M" due to
the mounting medium.
R. A. BROWN ET AL. 201
2000
AU
iii
I- 1500
11.
~
0 AU
UJ
:r
I-
0
0
::;: 1000
Ul
AU
500
W
20 30 40 50 60 70 80
2 THETA
Figure 5. Tungsten nwn peak when theta is set to 4° for
Sample 1.
2000
.,~
1500
Q
W
i!:0
0 1Il00
::I!
In
500
M AU
20 30 40 50 60 70 80
2 THETA
Figure 6. Sample 2 with theta set to 0.5°. Au (gold)
and M (mounting medium).
2Il00
ui
I- 1500
Il.
on
~
Q
W
J:
I-
0
0 1Il00
::I!
In
CU AU
500 M AU AU
20 30 40 50 60 70 80
2 THETA
Figure 7. eu (copper) layer of Sample 2 becomes evident
when theta is set to 2°.
R. A. BROWN ET AL. 203
2000
iii
Ii: 1500
~
Q
W
j!: AU
0 U CR
0
::E
I/)
1000 R
M CU
CU AU
500
20 30 40 50 60 70 80
2 THETA
Figure 8. Increase in Cu peak intensity of Sample 2 when
theta is set to 4 0 • Cr (chromium) peaks overlap Au (gold)
peaks making identification difficult.
Au-Cu/Cr
was used to analyze for three layers, Au, Cu and Cr. The
thickness of each layer was calculated and results are shown
in Table 3 as Au-Cu-Cr.
If no XRD data had been available, a second assumption could
have been made that there are two layers where the first layer
is pure Au and the second layer is a mix of Cu and Cr metal.
This assumption was tested by a method where thickness of
the first layer and both thickness and elemental composition
of the second layer can be found.
Results are shown in Table 3 as Au-Cu/Cr.
CONCLUSION
If we assume that there are two layers, with Cr as 14% of
the second layer, we should have seen an anomaly in peak
intensities for the XRD data on those peaks where Au and Cr
overlap. Since the XRF data indicates layer one and two in
excess of 6000A, one would not expect to see the third layer
through that thickness, especially through absorbers such as
Cu and Au.
NEAR-SURFACE CHEMICAL CHARACTERIZATION
INTRODUCTION
EXPERIMENTAL
The experime~t was carried out using SR at the Photon Factory on beam
line 4A. The apparatus for chemical state analysis by SIXES is
essentially the same as usual energy-dispersive X-ray fluorescence system
using SR as reported in our previous paper. 11 This technique measures
the X-ray absorption spectrum in terms of fluorescent X-ray intensity as
a function of the incident energy scanned around the absorption edge of
interest. In the present experiment, the GIF setup (Fig.1) was used for
the surface sensitive measurement.
The sample used was Cr[200Al/Cr 203 [2000Al thin film. It was prepared
by vacuum evaporation with an electron beam gun and deposited on a flat,
~ lj:
Monochromatized
'-..... SR beams
'" ""---
Fluores~nt /
~----:JX_Rays
-~ ,,:
//,/'
Si(Li)detector
2 3 4 5 B
~'-'I---,'-..-~...,..L...-L.,-~~---.i-., ~~ I IP
,,
,
\
\
\
...>- \
'>
\
:;:
u 0
CI)
;:
CI)
rx:
,,
, Fig.2
,,
,
\ Angular dependence of the
reflectivity at lower (solid
line) and higher (dashed
line) energy side of the
absorption edge. Experimen-
5 10 15 tal (a) and calculated (b)
Glancing Angle (mrad) results are shown.
smooth synthetic quartz plate. Pure Cr and Cr 203 thin films of 2000l\.
were also prepared as reference samples.
,.;
.5
GI
U
c::
GI
U
UI
I!!
o::::I
ii:
20
X·Ray Energy (eV)
The curves A and B are clearly distinguished from each other, though
the sample is the same. For the 45 deg. incidence angle, the penetration
depth is in the order of ]Jm above the absorption edge. Therefore B
corresponds to information for the whole film, i.e., Cr metal 200 ~ layer
plus Cr 203 2000 ~ layer. From the quantitative analysis, it was found
that curve B was in good agreement with the synthesis spectrum of Cr and
Cr 203 in the layer thickness ratio of about 1:10. On the other hand, in
the grazing incidence condition, the penetration depth becomes extremely
shallow and is estimated to be less than 100 ~ for A. That is, curve A
indicates the chemical state of the near-surface layer of the sample.
Actually, it is close to the spectrum of Cr metal. The difference
between A and the Cr metal spectrum might be explained by considering the
formation of the surface Cr 203 layer by natural oxidation. Assuming that
curve A consists of two components,Cr and CrZ03 spectra, the thickness of
the surface Cr 203 layer can be calculated and was estimated to be about
40 A. From these results, it is concluded that we can analyze the near-
K. SAKURAI AND A. IIDA 209
o 10 20
Fig.4 X-Ray Energy (eV)
Fig.5
o ~------------------------~
The calculated angular depend- (c) Penetration Depth
ence of the reflecti vi ty (a),
surface X-ray intensity; (b),
and penetration depth; (c) for 10 3
pure chromium metal near the
absorption edge. Calculation
was made for +41 eV from the 10 2
absorption edge (broken line),
-22 eV (solid line), and +1 eV
(dot-dash line). Anomalous 10 ~------------------------~
scattering fa 1\ors were taken
5 10 15
from the table for the calcu-
lation of the refractive index. Glancing Angle (mrad)
210 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
surface layer separated from the bulk material information by using the
grazing incidence condition.
SUMMARY
ACKNOWLEDGF..MENTS
REFERENCES
1. INTRODUCTION
X-ray fluorescence analysis is the most suitable method. for
the characterization of the thickness and the chemical
composition of thin film samples. It is non-destructive, rapid,
precise, and accurate for both metal and oxide samples.
X-ray spectrometry by the fundamental parameter method allows
determination of the thickness and composition of multi-Iarer
thin film simultaneously using a small number of standards. ,2,3
However, the process of generating fluorescent x-rays in a multi-
layer thin film is rather complex. The intensity is affected
by both the thickness and the composition of the film. it is
also affected by the upper and/or lower layer in the film.
This paper describes the basic studies of the fundamental
parameter method for multi-layer thin film analysis using an
IBM PC-based fundamental parameter software, Dataflex 270. The
software includes a theoretical intensity calculation for primary
and secondary excitations. The data reduction routines which
take the raw intensity and calculates the thickness and the
composition are described. The influence of the primary spectral
distributions on the analyzed results are also studied.
2. EXPERIMENTAL
A wavelength-dispersive sequential x-ray spectrometer (Rigaku
Model 3270) equipped with a Rh-target end-window tube was
employed for the measurements with a tube excitation voltage
of 50kV. The fundamental parameter software package was the
Dataflex 270 (DF 270) which runs on an IBM PC Personal System/2
connected directly to the spectrometer.
Primary excitation
IPO p) =
I
r f.~epQiP(~)lo(71)eXp \f_(~(_i\)fx+~(_iP!!X)}dX_d"=
t
to I-m
I
X=~+~ X=...H.i.!£l._J:!..!..igl
4 sinli', cose ' 5 sin \'2 cose
where X5- 0
I ) = 1- exp (-Xz/'t) I-exp (-X3P!) 1- ( 1+ Xoft) exp (- X3ft)
l X, X2 X3X, + X~
Fluorescence yield : wi
Jump ratio : OJ
Transition probability : Pi
Fe -Koc Intensities
Primary Secondary
44.7192 Ip 1.8
/
NiFe(Fe20% 2000A
Fe m77t'7'7777/77;77?~-:+-;>77:m7'771 2000A
Primary Secondary
831 18.3 336 7.6
I I
/ I
.. "/<'. :/~ .
NiFe(Fe20%) -:'~ /1 : IO~m
Fe ~~~~~~~~~~2QCX)A
·1:'
Primary Secondary
44.7 1.0 16.8
MF,(::ro%JI A6 \~,(,/.
Figure 1 Fluorescent X-ray
intensities from double-layers
216 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
• Thin film
o Bulk
1.0r-------------j/
Ni-KO( P - i<o(
~ 0.1
Thin film samples 'iii Thin film samples
c
0.1 C/CoNi/Cr/NiP Q)
C/CoNi/Cr/NiP
NiP C Si02P2051Si
'C
~ Ni a1 0.01 NiP
11 0 .01 ~1
>
Q)
Bulk samples
Ni pure ~ 0.001
Sui k samples
Phosphate ro
Ni base alloy ~ Fe base alloy
~O.OOI Fe base alloy ;
~O.OOOI
~~0~.OO~I~~0~.071--~0~.I--~1.0 ~~~~--~~~~~
Relative t~etical intensity
Excitation Absorption
1.0,---------------,
0:
H .....0:
.Q o
.~
t~~
~
C
... ...
~ .~0.5
.~ 1
III
H
~1
H
O~--~----~----~--~
5 20 o t05 10 20
Intensity Intensity ~ Thickness t
IR ~ I-e- pt IR~ e -pt
Thickness of IR Th ickness of I R
In{l-IR) InIR
tIR = In 0.96 ·to.04 tIR = InO.5 ·to.5
Intens ity of t Intensity of t
~ t
IR = I - 0.96 to.04 IR = 0.5 to.S
Figure 3 Relationship between film
thickness and intensity
S - 1<0< 0.32 0.19 0.047 0.11 0.066 0.028 0.019 0.QI8 0.027 0.010 0.012
Ti - Ko( 1.9 1.2 0.29 0.23 0.20 0.16 0.1 I 0.035 0.030 0055 0.023
Fe - Koe 3.8 2.9 0.71 0.56 0.12 0.18 0.26 0.082 0.067 0.090 0.053
Cu - Koe 6.0 5.1 1.3 1.0 0.23 0.093 0.23 0.15 0.12 0.077 0.097
Mo- Koe 15 19 7.1 6.6 1.5 0.61 0.43 0.41 0.81 0.34 0.14
Sn- Ka( 30 40 17 17 4.3 1.8 1.2 0.47 1.1 0.97 0.41
Tb- La( 2.2 2.1 0.60 0.51 0.12 0.17 0.21 0.078 0.064 0.081 0.051
Pb - Lo< 5.2 6.2 2.1 1.9 0.45 0.18 0.13 0.32 0.24 0.1 I 0.12
Density 2.25 1.50 2.71 2.0 4.60 7.874 8.94 10.22 7.29 8.272 11.34
218 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
F - K« 0.30 - OAO 0.28 0.065 0.20 0.12 0.034 0.043 0.083 0.035
AI- Kae 2.5 1.3 - 1.3 0.40 0.14 0.096 0.13 0.16 0.047 0.081
S - Kae 7.6 3.6 0.77 - 1.0 0.37 0.24 0.22 0.35 0.13 0.14
I=ali + bIM +c
Ir =Ip + Is
initial values for thickness and composition are set for the
following iteration process. During the iteration the program
calculates the theoretical intensity of the thickness and
composition based upon the last iteration. It then calculates
the deviations for the thickness and concentrations, comparing
the theoretical intensities with measured intensities. A
difference equation is introduced in this part of the
calculation, after which new thickness and concentrations are
SI S2 S3 TI T2
NigoFelo Cr 2cxx)A
Cu Cu 20001\.
Cr NigoFelO 2cxx)A
Si02 Si02
Rh torget 50kV,90°,O.127mmBe
1.0 x Pella et 01.
o XRFII
Continuum o Ochietal.
• DF270
,.,
50.5
-CI)
c:
Characteristic lines .......... DF270
-XRFII
~
.~ RhL
.~ ~ RhK - Pella et 01.
.~
~
~
Q)
E 05
1 -Ochi et 01.
O~ ____~~____~~____~__~~~____~~~~~
o 1.0 2.0 30
--~ Wavelength (A)
(%)
Rh target
10 50KV
....
....
0
....
Q)
0
u 0
+=
>.
"0
c
0
-
0
.~
0 Ag
~ -5 x Pella et 01.
o XRFII
c Ochi et 01.
1 • DF270
-10
U5 2 3 4 5 678910
(AI
- - - Wavelength
Figure 7 Analytical errors in thin
film analysis using various kinds of
primary spectral distributions
(standards: bulk standard materials)
8. CONCLUSIONS
The basic performance of the fundamental-parameter software
of the DF 270 has been evaluated and good linear relationships
between the measured and the theoretical intensities were
obtained for both bulk and thin film samples. We introduced
a difference equation in the iteration processing of thickness
and composition calculations, permitting the analysis of
complicated multi-layer thin films. The accuracies obtained
for single-component film analyses are approximately 1% over
most of the analytical elements by using pure element and pure
compound bulk standards. The analyzed results by four different
primary distributions were compared, with large differences
found among them. Our distribution gave the smallest average
analytical errors.
REFERENCES
1. D. Laguitton and M. Mantler, Adv. X-ray Anal., 20 (1977),
p.515
2. D. Laguitton and W. Parrish, Anal. Chem., 49 (1977), p.1152
3. T.C. Huang and W. Parrish, Adv. X-ray Anal., 29 (1986), p.395
4. T. Shiraiwa and N. Fujino, Japan J. Appl. Phys., 2 (1966),
p.886
5. K. Ohno J. Fujiwara and 1. Morimoto, X-ray Spectrum, .2.
(1980) p.138
6. P.A. Pella, L. Feng and J.A. Small, X-ray Spectrum, 14
(1985) p.125
7. J.W. Criss, Adv. X-ray Anal., 23 (1980), p.93
8. H. Ochi and H. Okashita, Shimazu Rev., 45 (1988), p.51
9. R. Tertian and N. BroIl, X-ray Spectrum, 13 (1984), p. 134
APPLICATION OF MULTI-LAYER THIN FILM ANALYSIS BY X-RAY
SPECTROMETRY USING THE FUNDAMENTAL PARAMETER METHOD
1. Introduction
The fundamental parameter method for x-ray spectrometry
has been used most commonly for bulk samples, because it
permits an analysis utilizing a minimum number of standards,
even for samples with complicated matrices. The need for the
analysis of thin film materials, which includes multi-layer
films, has been increasing in recent years along with the rapid
progress of high technologies. However, there have been few
reports that deal with the application of fundamental parameter
methods to multi-layer thin films. 1,2,3,4 There may be two
situations in the analysis of thin films. In routine analysis
of quality control applications, they usually require precise
analysis. Fortunately, it is possible to prepare well
characterized standards similar to the unknowns. In the case
of research and development and service laboratories, there is
not limit to the type of samples they can see, nor can they
prepare adequate standards for each type of sample. We have
found that the fundamental parameter method to be described
fulfills the needs of both of these situations.
This paper describes some of the applications of the Rigaku
multi-layer thin film fundamental parameter program of the
DATAFLEX 270, including a standard less analysis method
utilizing a sensitivity library. 5,6
2. Experimental
A brief configuration of the spectrometer system including
the computer system used in this study is shown in Table 1.
The spectrometer system is a wavelength-dispersive type (Rigaku
sequential spectrometer Model 3270). The spectrometer can
analyze down to boron (Z=5) utilizing a Rh-target, end-window
x-ray tube and new analyzing devices, such as synthetic multi-
layer and total reflection devices. The analysis of ultra
light elements from boron to oxygen in thin films have broad
Off-line
quantitative analysis
Theoretical intensity
co leulation
Unit of thickness
moss thickness or thickness
Measuring conditions
Film st ructure
Spectral lines
Similar or'dissimilar
standards
IntensiTy conversion
DI",hrogm
Diaphragm: 30mm dia.
Attenuator : III
ISensitivity I ibraryl
for all elements
Allenuator
>-/
51 yDB42
TI V Cr Mn Fe Co NI Cu
15 I I I I
Cu
,,
00933
Calculation af thickness and
.! tiD
.~
c concentrations with F. P.
method
r5~
22 23 24 25 26 27 26 29
--Atomic number
(Note) Arrow: Library sample
Spectrum : K«
Samples : Pure materials and compounds
X-ray tube: Rh-target
15
}I.O
\I)
~IO
H
8'
1 lO.5
5
Ti 0.0
Ti
o 20 30 40 0.5 1.0 1.5
- - Atomic number ----+ log E (Energy: keV)
Note: Intensities of compounds converted to the intensities of pure materials
Figure 4. Instrument sensitivity curve
230 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
Calibration formula
Ie • a Ii+ bIM + c
Ie: calibrated intensity
1M: measured intensity
a. b,c : sensitivity constants
°0~------~5~0~0--------~IOO~0--
Thickness of carbon (~)
Figure 6. Analysis of carbon on magnetic disks
232 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
~20
'iii .£
c ~ 20
~ ~
.~ .~
"0
~IO
::> ~Vl 10
Vl
g 0
Q)
::?! ::?!
1 0 1 0
450 0 5 10 15 0 10 20 30
~ Theoreticol intensity ~ Theoretical intensity ~ Theoretical intensity
WDXRF EDXRF
LiF(220)
S.C. Si (Li) detector
Fine slit
4,---------------------,
;!?
:2
1
.~ 15 c If
c <1 ::> 3 '
::> ~
,
>.
.... u" c:-
g
QJ
E 1
:€IO .Q
- ....
0
_OJ 4 .3
...I
~7<1 1 >. N
.~ 5
§LF!T
N ..oN
OJ aJ
II)
c ,
.J
,u"'(l
f-<1
1...1
II)
c ,
...I
~ .0 .0' , 1 ~
c f- .0 C ~
H ~ ,
f- H
0 75 80 85 90 7.5
2 - Theta (deg.)
8I
~
PC Substrate 1 I. 2 mm I
2.0 u" m
...J
2. Analytical technique
· Selection of analytical line
· Sensitivity calibration
3. Standards
· Pure materials or dissimilar standards
· Simi lor standards
4. Analytical problem
· Uniformity of thickness and composition
· Flatness of substrate
· Impurity in substrate
· Standard value errars
· Errors in physical parameter' and instrument factors
· Errors caused by density in thickness analysis
· Microscopic uniformity of film surface
of the FeTbCo layer. The lines measured were the Ka's of Si,
N, Fe and Co, and the Tb-La line. The spectrum shown in the
figure was obtained from one of the actual samples. The Co-Ka
line is clearly separated from the Tb-LB line. The thicknesses
of the two silicon nitride layers are determined by measuring
both silicon and nitrogen. Very accurate results were obtained
for the thickness and composition of the FeTbCo layer. The
errors of the silicon nitride thicknesses are from 2% to 3%,
relative. Similar results were also obtained by measuring the
Fe-La line instead of the N-Ka line to determine the
thicknesses of the silicon nitride layers.
10. References
1. T. Shiraiwa and N. Fujino, Adv. X-ray Anal., ~ (1968),
p.446
2. T.C. Huang and W. Parrish, Anal. Chem., 49 (1977), p.1152
3.T. Arai, Adv. X-ray Anal., 30 (1987), p.315
4. J. Willis, Adv. X-ray Anal., 31 (1988), p. 175
5. H. Kohno, M. Murata, Y. Kataoka and T. Arai, Japan Adv.
X-ray Anal., 19 (1988), p.307
6. Y. Kataoka, Rigaku Journal, Q (1989), p.33
7. T. Arai, Adv. X-ray Anal., 30 (1987), p.213
8. R.L. White and T.C. Huang, in Proceedings of the 37th Denver
Conference, 36 (1988)
FUNDAMENTAL PARAMETER-BASED X-RAY FLUORESCENCE ANALYSIS
Abstract
The Fundamental Parameter-based calculation of x-ray fluorescence intensities for thin and
multilayer samples is performed with new formulas. By the calculations it is shown that both illlra- and
inter-layer secondary fluorescence can cause significant enhancement. With a method based on the
discussed theory, it is possible to determine the composition and layer thicknesses of multilayer samples.
Analyses are discussed of samples from IC technology and magnetic recording. Furthermore, the angular
dependence of intensities is calculated, in good agreement with experiments performed with a novel
spectrometer.
Introduction
For the analysis of samples consisting of multilayers or thin layers on substrates, X-Ray
Fluorescence (XRF) is an attractive technique, as it enables both chemical composition and
layer thicknesses to be determined in a fast and in general nondestructive way. Examples
from our laboratory are: optical coatings (e.g. Ag/glass), protective coatings (e.g. WCfstecl),
magnetic recording material (e.g. CoCr/Ge/polyimid), optical recording material (e.g.
InShTe/PMMA), superconductive ftims (e.g. YBa2CuJO'-x/SrTiOJ), as well as sample~ from
Ie technology (e.g. metal/Si and BPSG/Si). Some of these examples will be discussed in detail
below.
Just as in the case of hulk samples, quantitative analysis can be accomplished using
calibration standards with composition close to that of the sample. lIowever, since the layer
thicknesses are additional unknown quantities, more standards arc needed than in the bulk
case, where it is difficult (if possible) to produce accurate thin-film standards. Thus, more
than in the case of bulk samples, Fundamental Parameter-based analysis is demanded for thin
and multi-layer samples where reference standards are only required to determine the
spectrometer sensitivity for an x-ray line, although the accuracy of the results may he
improved using standards close to the sample composition.
Theory
Here the formulas for the calculation ofXRF intensities J from multilayer samples 2 are
used in the closed form derived by De Boer3. Besides fluorescence caused directly by the
incident x-rays (primary fluorescence) cnhancement due to secondary fluorescence, caused
hy (primary) fluorescent radiation, will be considered. Tertiary and highcr effects are
neglected.
First the formulas are given for monochromatic irradiation with wavelength A. Thcy
will be extended to calculate the intensity for polychromatic irradiation. The symhols used
are defined in Table I. The primary fluorescence intensity of (the considered line with
wavelength A, of) element i in a homogeneous layer (I) with thickness d1 (rf fig. I) is
The factor between braces describes the intensity change as a function of layer thickness.
The last exponential accounts for absorption of radiation in layers a which may be prcsent
above the considered layer I. For more layers one has to sum over a in the exponent.
As for the secondary fluorescence, one has to distinguish betwecn intra-Iaycr and
inter-layer contributions. The intra-layer secondary fluorescence of i due to fluoresccnce
radiation fromj (i.e. within the same layer I) is
(2)
For bulk samples (dl ---> 00) the factor L is given by the well-known 1 formula
L =!.L In
/1/
II £L I
+
Ilj
+ I ~I
In I + Il) (J)
a b
lda tda
d1 t d2 t j
tdb 1db
d2 t d1 t
Fig. I. Multilayer sample with in layer 1 atom i, emitting the detected x-rays, and in layer 2 atom j ,
emitting x-rays which can cause secondary fluorescence of i . One or more top layers a and one or more
intermediate layers b may be prescnt. (it) layer 1 above layer 2. (b) layer I below laycr 2.
where
(Ill +/12)f1j
1- C-(I'I +IJ 2 )d
PI (PI
I
+ 112)
In 11 + ~
Ii}
I +
112 (PI
I
+ P2 )
e-(I'I +112)d In II - f'~J~ I) (5)
Here
(6a)
and
-,- (6b)
n.n.
x n=l
Et(x) is the exponential-integral function and D(x) can be evaluated rapidly by a computer,
using fast converging series.
For the inter-layer secondary Ouorescence one has to distinguish the cases that the
radiation emitted hy j is directed upward (Fig. la) and downward (Fig. I b). The intensity
of the corresponding inter-layer secondary Ouorescence S./ CA.) and S'/ CA) can he written as
.1/ ()"J') t pc r ii
S..
lj
u,d
(A)= Y2
"
c.JJ.(A)w,gj
J.
--
.liU)
. M
'uI12i-. u,d
P (A)
/
(7)
where
(9)
In Eqs. (8) and (9), if more than one intermediate layer is present, one has to sum over h.
The factor X in (8) and (9) is defined as
X(p,qA ,d2 ) = V( dl A) - V( dl ,0) - V(OA) + V(O,O) ( 10)
240 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS·
Ill} q
+ ----,----''-- lJ{(I--)(llIJA +llbidb+112id2)}
q (pIlI} + qll?,i ) 11 \j J"
where the summation is over the x-ray lines of aJl kinds of atoms with Aj < AabS.;, both within
and outside the layer with atom i. (Of course, if i is present in more than one layer, one has
to sum the contributions from the different layers.)
If the incident radiation is not monochromatic, but has an intensity distribution 10 (,1),
the total intensity of fluorescence from i is
(13)
Calculation of intensities
As an illustration we will consider the calculation of Cr Ka intensities from
Coo.sCro.z/Ge films (analyses of which will be discussed in the following sections). In this
section we will use monochromatic irradiation with A = 0.2 A (and 1/" = 64°, "'Z = 400). In
general, the enhancement by secondary fluorescence is less for polychromatic irradiation.
Fig. 2a shows, on a double-logarithmic scale, the Cr Ka fluorescence intensity as a
function of the thickness of a Coo.sCro.z film on an infinitely thick Ge substrate. The dotted
curve indicates the primary contribution (Eq. I). The dashed curve includes the enhancement
due to intra-layer secondary fluorescence (Eq. 2) from the Co K lines (which grows, as a
function of thickness, from zero to the constant value given by Eq. 4). The full curve also
includes the enhancement due to inter-layer secondary fluorescence (Eq. 7) from the Ge K
lines (which decreases gradually as a function of thickness).
Fig. 2b shows the Cr Ka fluorescence intensity for a ca.235 nm thick Co o.SerO.2 film as a
function of Ge substrate thickness. For such a thin film the intra-layer secondary
f1uorescence (difference between dashed and dotted line) is only slight. The full curve
includes inter-layer secondary fluorescence (Eq. 7) due to Ge K lines, which incrcases as a
function of substrate thickness.
Fig. 2c shows the same, but now for back irradiation: Cr Ka intensity for a ca. 235 nm
thick Con.RerO.2 film as a function of the thickness of the Ge layer on top of it. The
pnmary contribution (dotted) is a simple exponential decrease. Again, dashed and full curve
include intra- and inter-layer seconclilry fluorescence, respectively.
D. K. G. de BOER AND P. N. BROUWER 241
"..
-'
.,'
.. '
, .'
".---:-:'. "
,.,.'
- - -.-
Fig. 2. Calculated Cr Ka intensity for various CoCr/Ge samples. Dotted: primary fluorescence only,
dashed: primary + secondary fluorescence due to Co, full: primary + secondary fluorescence due to Co
+ secondary fluorescence due to Ge. (a) Cr Ka intensity as a function of coating thickness for a
COROCr20 coating on an infinitely thick Ge substrate. (b) Cr Ka intensity as a function of substrate
thickness for a 2 x 10 -4 g/cm 2-thick CO&OCr20 coating Oil a Ge substrate. (c) Cr Ka intensity as a function
of coating thickness for a Ge coating on a 2 x 10- 4 g/cm 2 thick COROCr20 substrate.
Examples of analyses
Of more practical interest than intensity calculations from known composItIon is the
calculation of sample composition (i.e. both concentrations and layer thicknesses) from
measured intensities. For this purpose a Fundamental Parameter-based (1'1') computer
prognun was written in which. for each x-ray line measured, (at least) one reference standard
is required (which could be a bulk sample). The program takes an initial estimate of the
composition from which the intensities (related to those of the reference) arc calculated, using
the formulas given above. Then the thicknesses and concentrations are updated (with the
condition that the sum of the concentrations equals one inside each layer) and this step is
repeated until the root-mean-square differencc between measured and computed intensities
is minimal.
In this section, some thin-layer analyses are considered (which, unless stated otherwise,
were done in a standard way using a Philips PWl404 spectrometer).
As a first example we will discuss CoCr films (on polymer substrates), which can be used
for perpendicular magnetic recording. The use of a Ge layer between CoCr and polymer
gives rise to a better crystallization of the CoCr layer, resulting in a higher perpendicular
magnetic anisotropy and better recording properties 6 • First a series of CoCr films sputtered
on a polyester substrate (without Oe interlayer) were analysed. The nominal composition
of all films, determined from the sputter target compo~ition, was 80.55 mass% Co, 19.45
mass% Cr. Neutron Activation Analysis of the films yielded 80.4 mass% Co, 19.6 mass%
Cr. The thicknesses d were determined from sheet resistance measurements.
The results of XRF measurements worked out with the FP program, using bulk Co and
Cr as standards, are as follows (assuming a density of 8.50 g/cm3 for the CoCr layer):
242 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
Many thin-layer samples which arc analysed by XRF in our laboratory originate from
Integrated-Circuit (lC) technology. Examples are metallization layers (e.g. AI, Ti, Co) on
Si wafers. We will consider (oxygen-contaminated) cohalt films on silicon. In advanced IC
technology silicide (in a self-aligned silicide process) is formed by the reaction of such a thin
metal film with the silicon substrate 1. The reaction is extremely sensitive for contamination
(C, N, 0) in the metal film. In order to study the inOuence of oxygen contamination in
sputter-deposited Co films on the silicide formation process, cobalt deposition was performed
not only in pure argon, but also in Ar/0 2 gas mixtures. The amount of 0 incorporated in
the deposited Co films (with a thickness of ca. 38 nm) was measured using XRF and the FP
method, with bulk quartz as a reference. It was found that the relative contribution of Si
substrate nuorescence to the 0 Koc signal is 21 % throughout the whole concentration range
of interest. In Fig. 3 the oxygen content in the Co layer is plotted against the O 2 content in
the argon during sputtering. The offset at the vertical axis corresponds to a native oxide layer
of 3.3 nm. At increasing O 2 content in the Ar/02 mixture the 0 content in the Co film
increases linearly up to 0.8% O 2 , At larger O 2 contents the oxygen incorporation in the Co
U
t
o
.!:
o
Cr tube
20
(60 keY)
~ Si detector
(width=50 pm)
10
o 2 . bent crystal
%02,nAr-
layer increases rapidly. This is ascrihed to the fact that then, during the sputtering process,
the Co target surface is oxydized to such an extent that Co-O is sputtered instead of Co.
(Similar effects have been ohserved at Ti sputter deposition in M/N2 gas mixtures R .) The
discontinuity was also found in the formation of CoSi 2 from the sputtered films: at levels up
to 0.8% O 2 silicide formation occurs, whereas at higher levels hardly any continuous CoSi 2
film could he formed.
Another example from IC technology is the analysis of horophosphosilicate glass
(BPSG) planarisation layers on Si wafers. The properties of these layers, which mainly
consist of Si0 2 , strongly depend on their boron and phosphorus content. We will express
the last two as B20 3 and 1'2 0 5 (although some P may be trivalent). These measurements were
done with the Philips XRF Wafer Analyzer (which can handle whole wafers) using an Rh
end-window tuhe at 40 kV. The Band P contents were determined by Inductively Coupled
Plasma/Optical Emission Spectrometry (ICP/OES) and the thickness d was measured by a
taly-step method, ellipsometry or weighing before and after glass-layer removal 9 • (From these
measurements, the layer density was found to be 2.13 g/cm 3 .) With XRF, the intensities of
the B KIX, P KIX and Si Kf3 lines were measured and the results worked out by the FP method.
A selection of the results is as follows, using sample S as calibration standard:
sample mass% B20 3 mass% P2 0 S mass% Si02 d (nm)
ICP XRF IeI' XRF ICP XRF measured XRF
S 7.5 11.0 81.5 GOO
T 14.6 12.7 6.6 8.1 78.9 79.2 290 270
U 7.0 7.5 11.2 10.2 81.8 82.3 920 1020
V 7.1 6.4 12.9 12.1 80.0 81.5 590 620
W 7.4 6.6 13.6 12.4 78.9 81.0 590 610
The average relative deviation in both H20 3 and P20S content is about 10%; the average
relative deviation in layer thickness is ahout 5%. This probahly arises from the uncertainties
in the absorption coefficients and in the fluorescence yields (influencing the calculated
enhancement, cf C'q. (7». It was found that the elTect of enhancement due to substrate
fluorescence on the B K('( intensity is as high as 8%.
a b
o 30
Fig. 5. Intensity vs. detection angle for Ti Ket. measured with (113) reflection of an SiC 111) crystal. (a)
Ti layer on Si. (b) li bulk. +: measured (using a Cr tube at 60 kV) with q, > 0 (Vtl + V/2 = 42.57"), 0 :
measured (using a Cr tube at 50 kV) with <p < 0 (1//1 + V/2 = 101.56°). Full line: theory.
244 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
t I c)
I
.~
t front
en
c
\ ,Sl
I £;
I
\
\
x'\
t
CrKa
a) x\
COQ,e CrO,2 I
~
'iii Ge x\
c front y-back
,Sl polyimid
£; I
I
x
30
~
\x 1j>2(')-
\
+ I
+ + '\
+ +\~ ,,,
+'~')(
+ '<t~"'... x GeKa
/'+.....x
I
front ~if--.
I ~-;-if-~
o ~ 00 (0) o 30 60 1j>2(')_
1j>2 -
Fig. 6. Intensity V.f. detection angle for a CoCr/Ge/polyimid sample (see inset), measured (using a Cr tube
at 60 kV) with an Lifo(IOO) crystal. +: measured at front side, x : measured at the back, full curve:
calculated for front side, dashed: calculated for the back. (a) Co KG( intensity measured using (400)
reflection (VII + 1/12 = 77.69°). (b) Cr KG( intensity measured with an AI filter (0.1 mm) using (200)
reflection (V" + 1/12 = 49.65°). (e) Cie KG( intensity measured using (600) reflection (1/11 + 1/12 = 84.11°).
e isthe Bragg angle of the monochromatising crystal and 4> the asymmetry angle (i.e. the
angle between crystal surface and reOecting crystal planes), which can be positive or negative.
In FIg. 5 results are shown for the Ti KG( intensity of a titanium layer on silicon and of
bulk titanium as a function of 0/2 in two experimental configurations. Measured values are
shown, as well as values calculated with the for method (which are scaled in such a way that
they fit the experiments at intermediate 0/2)' As a general trend, it is seen that the intensity
decreases as a function of 0/2, due to the factor sin 0/1/ sin V/2 . For a thin layer (Fig. 5a) the
decrease is much greater than for a bulk sample (fig. 5b). So the surface sensitivity can be
increased by measuring at a small detection angle. As indicated above, for 0/2 below 2.4° the
experimental intensity decreases. for the bulk sample a deviation from theory is already seen
for slightly larger 0/2, because the (foil) sample was not completely flat.
Another example is shown in Fig. 6, giving both experimental and theoretical results for
a CoCr/Ge/polyimid tape (sample G of section 4), measured from both the front (CoCr) side
and the back (polyimid). It can be seen that for such a three-layer sample the results are also
good (except for small 0/2, because the sample was not completely flat).
Conclusions
For thin and multilayer samples, Fundamental Parameter-based calculation of XRF
intensities from the composition, and vice versa, is possible using the proposed method. In
many cases it is important to take into account enhancement due to inlra- and inter-layer
secondary Ouorescence. Also, the angular dependence of XRF intensities can be calculated
and agrees well with experiment.
Analytical results for both concentrations and layer thicknesses can be obtained with a
relative accuracy of at least 10%. It is expected that in the near future the accuracy can be
improved by the use of better values for the Fundamental Parameters, based for instance on
a comparison of many independent analytical results. Because of the high precision of the
XRF measurements, the method is already well suited to monitor changes in sample
composition.
D. K. G. de BOER AND P. N. BROUWER 245
Acknowledgments
We should like to thank O. Elgersma, J..J.M. Borstrok and C. Nieuwenhuizen for helpful
discussions and for providing results of analyses, TJ.M. Verspaget and M.G.M. Meulendijk
for performing analyses, J.P.c. Bernards for providing the CoCr layers and R. Wolters for
providing the data on Co(O)/Si.
References
\. 1. Sherman, Spectrochim. Acla 7 (1955) 283.
2. M. Mantler, Analytica Chimica Acta 188 (1986) 25.
3. D.K.G. de Boer, submitted to X-Ray Speclr.
4. D.K.G. de Boer, accepted by Spectrochim. Acta B.
5. P.A. Pella, L.Feng and 1.A. Small, XRay Spectr. 14 (1985) 125.
6. c.P.G. Schrauen, J.P.c. Bernards, RW. de Bie, GJ.I'. van Engelen, HI!. Stel, V. Zieren and S.B.
Luitjens, IEEE Trans. Magn. MAG-24 (1988) 1901; .T.I'.c. Bernards, c.P.G. Schrauwen, S.B.
I,uitjens, Appl. Phy,r., submitted for publication.
7. L. van den Have, R. Wolters, K. Maex, R.F. de Keersmaecker and G.J. Dcclerk, IEEE Trans.
Electron Devices ED-34 (J 987) 554.
8. R.C. Ellwanger and 1.M. Towner, Thin Solid Films 161 (1988) 289.
9. O. Elgersma and J..T.M. Borstrok, to be submitted to XRay. Spectr.
10. D.K.G. de Boer, X-Ray Spectr. 18 (1989) 119.
NON-DESTRUCfIVE CHEMICAL-STATE ANALYSIS OF THIN FILMS AND SURFACE
LAYERS (1-1000 NM) BY LOW-ENERGY ELECfRON INDUCED X-RAY
SPECTROSCOPY (LEEIXS)
ABSTRACf
1. INTRODUCTION
2. EXPERIMENTAL
Two different types of electron source have been used to excite x-ray
emission from surfaces, one novel and one quite conventional. The former
utilises the electron flux present in a gas-discharge plasma, the latter is an
ordinary electron gun. In both cases the analysis of the x-rays from the
sample is by standard procedures. In the experiments described in this
paper single flat crystals were used to disperse the x-rays which were then
detected in gas-flow proportional counters fitted with thin windows [~I 11m
polypropylene].
A A
,
,,
I
,,
,,
I
,. I
I
r--, L--- "-
"-
~ [ HT
r-- "-
"'-
-L
Figure 1. Diagram of essential features of the gas-discharge source. A:
anode, C: cathode, hatched areas: insulators, dashed line: electron
beam.
This type of source has the great advantage that it can be used with
insulating samples. It is not necessary to have a conducting sample to
complete the electronic circuit. A further advantage is that the source can
be configured to replace an x-ray tube in a conventional x-ray spectrometer.
This means that ordinary x-ray fluorescence spectrometers, provided only
that they are capable of achieving a vacuum of a little better than one tenth
of a torr within the spectrometer chamber, can be easily adapted for this
type of surface analysis.
b) Electron-Gun Source
c) X-ray Spectrometer
The angles of incidence of the electron beam on the sample and also the
take-off angles for the characteristic Xrays were different for the two
spectrometers so that both the penetration depth of the electron beam, as a
function of electron energy and also the importance of x-ray absorption
effects were different in the two systems. This means that a direct
comparison of results from the two spectrometers is not possible although
overall trends should show the same form (but perhaps at different beam
energies)
a) Alumina/Aluminium
Fig. 2 shows how the oxygen Ka x-ray intensity varies with electron-
beam energy for the natural film of alumina which covers metallic
aluminium. In such a very thin film [5 nm or less] the absorption of oxygen
K a x-rays will be negligable so that the form of the curve in figure 2 should
be closely related to the variation of ionisation cross -section (cr) with energy.
That this is indeed so can be seen by comparing figures 2 and 3. Figure 3 is a
theoretical curve showing the ionisation cross-section as a function of
S 10
o 2. 4- kV Eo / EK
x/
/
•
/ >
/+ ----
)k
~
,;'
+
/
/
/
/T
y
/
-,
!o
2. 4
Figure 4. Variations with electron beam energy in the intensities of (i) OKa
(upper curves) and (ii) Al K (lower curves), X-rays from an
alumina sample. Open or solid circle data points (2-5 keY) were
obtained using the gas-discharge source. The other data. 4-11
ke V (x or +) were measured using the Telsec Betaprobe. As the
geometries of the two spectrometers are different, differences in
absorption effects can be expected and so the solid and dashed
curves have been displaced from each other. It should also be
noted that the solid circles are measurements of Al Ka intensity.
the crosses of Al K~; the change was made for experimental
convenience. Whilst absolute intensities are very different no
variation due to self-absorption is anticipated so that the form of
the curve should not be affected.
252 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
Z" La(
iI
M
3'-0 3go 400 e.1/
A thin film of zinc oxide on a soda-glass cover slip was grown by low-
pressure metallo-organic vapour deposition using basic zinc acetate
(Zn40(CH3COO)6) as a precursor (ca. 0.4 Ilm, Shrs, 320°C, 10- 2 -10- 3 Torr). The
results of an examination of this film by LEEIXS are shown in figure 6. By a
happy circumstance it proved possible to study many of the X-ray lines of
interest over a small angle -range, because of the proximity of second -order
Zn La, ~ and Na Ka with first order oxygen Ka. As the beam energy is
increased from 4 through S to 7 ke V the corresponding increases in both Zn
La,~ and 0 Ka x-ray signals are consistent with increasing electron
penetration into the zinc-oxide layer. In going to 9 keY, however, striking
changes are seen in the intensities of both the oxygen Ka signal and the
sodium Ka (2nd order). This indicates that between 7 and 9 keY the electron
beam starts to penetrate the underlying glass to a significant extent. This
was confirmed by a direct measurement of the silicon Ka intensity which is
negligible up to 9 keY and then increased rapidly after 9 keY. The last
spectrum shown in fig. 6, at 11 keY is of interest in that not only is the
increase in Na Ka signal maintained but the Zn La,~ actually falls a little, as
would be expected once the exciting electrons can pass through the whole
film. It should also be noted that although a larger oxygen signal is found at
11 keY that at 9 keY the increase is slight. This is of course in keeping with
the behaviour seen in figure 4. (i). In this case of zinc oxide on glass both
film and substrate contain much oxygen : the OKa yield thus behaves as if it
came from a 'thick' sample.
A. K. GY ANI ET AL. 255
•
•
•
•
•
2- to KV
of- +
+
+
+
.-
0
+ 0
0
"
0
'0
ltv
2- 4 6
•
• •
2 4
Figure 8. Variation of X-ray intensities with electron beam energy from an
aluminium - nitride film.
i) top curve, nitrogen Ka
i i) middle curves, open circles Al Ka, crosses Al Kp.
The data points have been separated because they were
measured on different spectrometer. The
complementary nature of the curves is apparent.
iii) bottom curve, oxygen Ka.
256 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
I
1530 1!;7o
as an interfering peak in the 3rd- order 5.5 keY. This interference prevented
an extended study of the N Ka intensity as a function of beam energy. The
range 2-5 ke V is however shown in Figure 8 (i). Clearly a plateau is being
reached consistent with the notion of the nitrogen being confined to a
surface layer. A similar conclusion can be drawn from the curve shown in
figure 8 (ii), for aluminium x-rays. Note that the two curves [one for Al Ka,
at low beam energies, the other for Al Kp at higher beam energies] were
measured on different spectrometers. Thus a direct anion is not possible, as
explained above, however the form of the curves is clearly complementary.
Above 9 keY the intensity of the aluminium X-rays is effectively constant
showing that aluminium, too, is present only in a surface layer. Figure 8 also
shows (curve (iii» the behaviour of oxygen Ka. The rise and then dramatic
fall in intensity even at low beam energies indicates that oxygen is only
present very near the surface of this sample [which had been carried in the
air from the laboratory where it was made to the spectrometer].
A. K. GYANI ET AL. 257
J I I
170 1'ao 190
Figure 10. Boron Ka emission spectra from silica films on silicon (top three
curves), from B203 and [rom BP04 in descending order. Beam
energies for the silica films were 9,7 and 5 ke V.
258 III. X-RAY SPECTROMETRIC CHARACTERIZATION OF THIN FILMS
CONCLUSIONS
The above examples show that considerable qualitative data about the
thickness and chemical structure of relatively thick surface films can be
obtained using low-energy electron-induced x-ray spectroscopy.
Comparable results can also be obtained, as Sza. sz 11 has shown at even longer
wavelengths using a grating spectrometer. It is also possible to carry out
non-destructive chemical state analysis of surface layers by x-ray-
fluorescence spectroscopy, using conventional x-ray sources
[Andermann ]14 and paper Willis presented at this conference .15 In the
former case the analysed depth is controlled by varying the angle at which
the exciting beam strikes the sample, at near-grazing incidence. As the
LEEIXS results described here can be obtained by the introduction of an
electron source into a conventional X-ray spectrometer it may be that this
will prove to be the simplest and most cost-effective solution for the analysis
of samples that range from tens to hundreds of nanometers in thickness.
ACKNOWLEOOEMENTS
The authors wish to thank the Royal Society, the Science and
Engineering Research Council [UK] and the Central Research Fund of
London University for their support of some of the research reported here.
They are also grateful to the European Economic Community for a 'twinning'
contract [it SJ2J -0282-C] which has facilitated the collaboration between the
London and Lyon Groups. They would also like to express their gratitude to
Dr. Elgersma, Philips Research, Eindhoven for the provision of the silica
samples discussed in section (e).
A. K. GYANI ET AL. 259
REFERENCES
INTRODUCTION
Two strategies are at present commonly used in studying and refining crystal
structures from powder diffraction data: the total pattern refinement proposed by
Rietveld (1969) and the two-step method originally proposed and applied by Will (Will
et aI., 1965). The latter one works by separating the intensity determination of the indi-
vidual peaks from the actual structure refinement, structure analysis or any structural
calculation (for example based on line broadening). Both methods have their merits,
and their drawbacks.
The merits of Rietveld's method are found especially in straight forward and con-
ventional structure refinements if we are dealing with i) single-phase substances; ii) a
simple and easy to be described background; iii) if the profile is known; iv) if the cell
metric is known approximately. The last two requirements are conditions sine qua non.
In the total pattern structure refinement a model must be presented to the program, and
this model then describes every data point in the diagram. This is useful and welcome
if the diagram is loaded with peaks, which becomes especially true if we have low-
symmetry materials or large unit cells. In other cases we have a great number of un-
necessary data points, basicly this is a ballast with no connection to the structure.
The two-step method has shown itself very usefu I and superior to total-pattern
methods if we have unconventional diffraction techniques with large amounts of con-
taminating peaks. This is found for example in in-situ high-pressure x-ray diffraction,
where there are always peaks from the sample holder, from the pressure marker and
from other containments. A second case is found when using position sensitive detec-
tors, especially in neutron diffraction, where there are quite commonly errors in the
linearity. Other fields of application and preference are cases where the background is
high and complex, where the profiles are not known and especially if the nonlinearity
disturbs peak positions. Also magnetic satellites fit into this list because such
incommensurate models are not implemented in Rietveld programs ..
information, which means only the peaks belonging to the structure in question can be
sorted out and if needed, they can be varied specific to the problem (for example, if
there is a need to analyze for strain or stress; compare for example the presentation
by Delhez et al. (1989) at the Petten conference).
In the light of the progress made in powder diffraction it seems worthwhile to have
an unbiased comparison between the two methods on the same material. We have
therefore taken synchrotron diffraction data of quartz collected by Dr. W. Parrish, IBM,
at the Stanford Synchrotron laboratory, and in addition selected neutron diffraction data
from our daily measurements.
For Rietveld refinements we have used the program version DBW 3.2 (Wiles,
1982). This program is well documented and needs no further description. In the two-
step refinement we have used for the actual crystal structure refinements the program
POWlS-80 (Will, 1979; Jansen et aI., 1988) which also is widely used, well known and
well documented (Will et aI., 1983).
Attention is required, however, for acquiring integrated intensities for the POWlS
calcu lation. For a decomposition of the pattern, several programs are available world-
wide. One was published some time ago by Pawley (1981). In our laboratory we have
developed two programs, PROFAN and FUlFIT (Jansen et aI., 1988). PROFAN works
interactive, where we can tell the program by a crosshair on a graphic's terminal what
reflections we expect under an unresolved peak. Different profile functions, available
in a catalogue in the program, can be tried and the best one can be selected for final
analysis. If the peak function is known and we have an approximate knowledge of the
cell metric. e.g. cell constants and angles, it is convenient to use the program FUlFIT,
which is quite similar to the Rietveld program and stops just one step earlier after re-
finement of the peak positions, however, listing integrated intensities belonging to each
reflection (rather than positional parameters x y z). and a complete refinement of the
lattice constants.
The samples were prepared from oscillator plate quality single crystals. After they
were sifted in a micro mesh sifter the final particle size used was 1-5 f.!. The sample
was rotated during the experiment around the normal to the sample plane. In a first
data set the wavelength was). = 1.0020 A, the step with ~(20) = 0.020 and the angu lar
range of the measurement 20 = 45 to 1350 (Smax = 0.92 A-1; S = sin 0/ ). ). The peak
halfwidth was 0.1r. All peaks could be analyzed and separated by profile fitting with
PROFAN. With this very short wavelength, 157 resolved peaks coming from, in many in-
stances,intrinsically overlapping 280 Miller planes of nonidentical intensities could be
obtained. (Due to the trigonal symmetry of quartz I(hkl) -+ I(khl).) Many of the peaks
were at high angles and had quite low intensities and consequently poor counting sta-
tistics; this necessarily affects the least squares calculations. Despite the low intensi-
ties these peaks are clearly visible in the diffraction diagram and well above the back-
ground.
The refinement calculation with POWlS was begun with only a limited data set to
20 = 75° (s = 0.47 A-1) with 31 peaks from 47 often intrinsically-overlapping planes.
We then proceeded in the refinement by extending the angular range, e.g. by increasing
the number of reflections to 20 = 45° to 105°, and finally to the full data set up to 20
= 135° (Table 1). All the results obtained are in excellent agreement with the results
from single crystal data by Levien et al. (1980), who used 602 reflections. The standard
deviations from the powder data are, as to be expected, about 3 times larger than the
ones from the single crystal data. For obtaining temperature factors, however, the high
G. WILL ET AL. (1) 263
FUlFITI
POWlS
Rietveld FUlFITI
POWlS
Rietveld FUlFITI
POWlS
Rietveld I
) (A) 2 Theta - sections (deg)
1.0020 45 - 135 45 - 105 45 - 75
angle data are needed, as can be seen very clearly when comparing results in Table
1.
Because in this first experiment the data below 20 = 45° were missing (due to a
failure in the electron accelerator), and also in order to increase the spatial resolution
and hence the separation of the peaks, we repeated the experiment with the same
sample a year later. Now the wavelength was set to A = 1.2823 A, the step width
t1(20) = 0.01 ° and the angu lar range of the measurement from 15 to 115°
(8 m" = 0.66 A-1). The halfwidth FWHM was 0.08°. This resulted in 72 well resolved
peaks. Large sections of well defined background were observed in this diagram with
only occasional small overlaps at the tails of some peaks. All peaks could be analyzed
and separated very well by profile fitting. The peak heights measured range from I(max)
= 35494for(101)/(011)tol(min) = 5 for (222), e.g. a range of 7000:1. (See Table 2). The
individual profile R-factors obtained with PROFAN are in the range of 1.4% to 2.0%,
except for the very weak peaks.
Also here we began the analysis with a smaller, low angle data set using only the
first 60 observations with 150 hkl-planes measured in the angular range up to
20=90°. The best results were obtained with 64 planes and 40 observations with
R = 1.41 % (Rw = 2.22%). (Not shown here). With 8 variables, the system is already 5 fold
overdetermined, therefore the limitation of the data set is well justified. This is clearly
demonstrated in Table 1. With the fu II data set up to 20 = 115° we have 72 observations
and 119 hkl planes and the system is 15 fold overdetermined. The R-values are
R = 1.68% and Rw= 3.21%.
The resu Its, e.g. the positional parameters, the temperature factors and even the
standard deviations did not change or improve, when we extended the data set to more
and finally all observations. In routine measurements for crystal structure refinements
264 IV. ANALYSIS OF DIGITAL DIFFRACTION DATA INCLUDING RIETVELD
.,
0
F ' L FIT
6
"C
-'
:J 4
c
u
2
-,. .
, , .. .', . .; " -,
M
,....
0
Ri e tveld
6
--
VI
C
;J 4
0
u
Fig. 1: Selected part of a quartz diffraction pattern measured with synchrotron radiation
(points). Profile refinements (lines) have been performed using the FUlFIT (above) and
the Rietveld routine (below) (A = 1.0020 A) .
it suffices from our experience to collect data out to s = sin 0/A = 0.7 A-1 or with our
wavelength out to 20 = 80-90°. (Remember the fu II Cu-sphere ranges only to s = 0.65
A-1).
In a subseqent analysis the same diagrams have also been refined a) with the
Rietveld method and b) with FUlFIT + POWlS. These are the values listed in Table 1.
The time spent for analysis was about the same for both methods, in real time approx-
imately one of the authors sitting one day in front of the terminal. Fig. 1 depicts the
quartz diagram after analysis by FUlFIT and by Rietveld. For the profile we used a
Pseudo-Voigt function, which gave the best Rpro/values. As can be seen from Fig. 1, we
have about the same quality in the difference diagram.
Table 1 summarizes the results, giving the structural parameters and the R-values
for both methods. We also demonstrate the influence of the size of the data set, where
we have refined the diagrams with different intervalls. There is no significant difference
in the actual values of the parameters. The real structural R-values (RsRAGG) are slightly
better with POWlS for the 1.2823 A data and with Rietveld for the 1.0020 A data. We
suspect that the differences originate from a different handling of the background with
regard to the R-value determinations. In both programs. Rp,o",. is slightly higher using
the Rietveld program, however, we have to keep in mind that this includes the struc-
tural constraints, whereas FUlFIT is run without structural constraints. For both wave-
lengths we can conclude that we get good and reliable atomic parameters already with
"short" diagrams, which means with a limited data set. On the other hand for reliable
G. WILL ET AL. (1) 265
Rietveld developed and published his method originally for neutron diffraction data
collected on polycrystalline samples. Because of the Gaussian peak shape the Rietveld
Material Reference
CaMn(CO,), N. Jahrb. Mineral. Mh. 12, 539 (1989)
CsTaN, J. Less-Common Metals 146, 147 (1989)
DyFe,AI, Mat. Sci. Forum 27/28, 243 (1988)
ErFe.AI, Mat. Sci. Forum 27/28, 243 (1988)
HoFe,Al, Physica B 156/157, 751 (1989)
Li,CuO, Z. anorg. allg. Chemie, in print
Li,ZnCl, Mat. Res. Bull. 23, 1805 (1988)
NpCo,Ge, J. Less-Common Metals 149,243 (1989)
NpCu,Ge, J. Less-Common Metals 149, 243 (1989)
TbNiC, J. Magn. Magn. Mat., submitted 1989
TbFe,Al, Mat. Sci. Forum 27/28, 243 (1988)
ThFe,Al, J. Less-Common Metals 149,237 (1989)
T/- Fe - S J. Phys. Chem. Sol. 50, 297 (1989)
UFe,Al, J. Less-Common Metals 149,237 (1989)
YBa,Cu,O" Mat. Res. Bull. 23, 1439 (1988)
266 IV. ANALYSIS OF DIGITAL DIFFRACTION DATA INCLUDING RIETVELD
Table 4: Crystal structure data from three recent neutron diffraction measurements ac-
cording to Rietveld (R) and POWlS (P) refinement calculations
routine works extremely well. We use the Rietveld technique on almost a daily basis,
and we have refined about 15 structures in the last 12 months. A list of some sub-
stances and references is given in Table 3. It is our experience that the Rietveld method
is superior in cases of structures with strongly overlapping peak clusters because of
the intrinsic structural constraints; as an example we may refer to a series of structure
refinements of the superconducting 1-2-3 compound (Schafer et aI., 1988).
For this presentation we have selected some recent diffraction patterns and re-
peated the analysis with both methods. Some results are listed in Table 4. But also in
neutron diffraction there are many cases where the two-step method is to be favoured.
One such case is when the position-sensitive detector technique is used, where we
have deviations in the linearity, where the observed and calculated positions of the re-
flections are not always where they should be according to calculations, C:'.nd which
therefore give necessarily errors in the Rietveld refinement.
In the determination of magnetic structures the Rietveld method often fails com-
pletely, only the two-step method can lead to a solution. Such an example is TbAs0 4 ,
where we found an incommensurate magnetic structure below 1.5 K. The neutron
diffraction pattern is characterized by a complex scheme of sattelite reflections, which
cannot be interpreted by a simple spiral. A section of the low temperature diffraction
pattern is depicted in Fig. 2. The successful profile analysis of the satellite split re-
flections has been performed with PROFAN. The real incommensurate structure has not
been unscrambled at this date.
It is the aim of all researches in this field to push powder diffraction analysis to a
point where it is fully competitive with single crystal data, this means where unknown
crystal structures can be solved ab initio from the data. The way most wanted is of
course to have separated intensities for every reflection and then to go through the
common techniques like Patterson synthesis, heavy atom method, direct methods etc.
This can be done only if one refers to a two-step approach where the diffraction dia-
gram is first decomposed before crystallographic work begins.
G. WILLET AL. (1) 267
-.
'"o I I I I I I J
6 - r-
5 -
(f)
c:
2
::J
'c:"
4 -
.3
I
..: •...• .-~ .
I
.Ii ~<. ~
I
It
I
JI:,
"IJV~
I
J ,n
I I
l~.,. r-
5 10 15 20 25 .30 .35
2theto
There are, however, some rare cases where the old method of trial and error has
to be used. For such crystal structure determinations the Rietveld method is very use-
ful, simply because it it so easy to imagine a model, take the atomic parameter set and
have the program to refine iL We have done such a procedure on tetragonal NaPO•.
The Rietveld analysis of the neutron diffraction is based on space group P42 1c with a
= 10.80 and c = 6.81 A (Lissel et aI., 1989).
ACKNOWLEDGEMENT
This work has been funded by the Ger-man Federal Minister for Research and Tech-
nology (BMFT) under contract number 03-WI1BON.
REFERENCES
Delhez, R., de Keijser, Th.D., Langford, J.L, Lauer, D., Mittemeijer, E.J. and Sonneveld,
E.J., 1989, Crystal Imperfection Broadening and Peak Shape in the Rietveld Meth-
od, I nL Workshop on the Rietveld Method Petten, Netherlands Energy Research
Fou ndation.
Jansen, E., Schafer, W. and Will, G., 1988, Profile Fitting and the Two-Stage Method in
Neutron Powder Diffractometry for Structure and Texture Analysis, J. AppL CrysL
21:228.
Jansen, E., Schafer, W. and Will, G., 1989, Background and Reliability of Pattern Profile
Refinement in Powder Diffractametry, in: InL Workshop on the Rietveld Method
'Petten, Netherlands Energy Research Fou ndation.
Levien, L., Prewitt, C. and Weidner, D.J., 1980, Structure and elastic propertiesof quartz
at pressure, Am. Mineral. 65:920.
Lissel E., Jansen, E., Jansen, M. and Will, G. to be published
268 IV. ANALYSIS OF DIGITAL DIFFRACTION DATA INCLUDING RIETVELD
Pawley, G.S., 1980, Unit-cell refinement from powder diffraction scans, J. Appl. Cryst.
14:357.
Rietveld, H.M., 1969, A profile refinement method for nuclear and magnetic structures,
J. Appl. Cryst. 2:65.
Schafer, W., Jansen, E., Will, G., Faber Jr., J. and Veal, B., 1988, Crystal Structure Anom-
alies of YBa2Cu30S9 at the Superconducting Transition Temperature, Mat. Res. Bull.
23:1439.
Wiles D.B., 1982, "Program for Rietveld Analysis of X-ray and Neutron Powder Diffract-
ion Patterns: DBW 3.2." Georgia Institute of Technology, Atlanta, USA.
Will, G., 1979, POWlS: A powder least squares program, J. Appl. Cryst. 12:483.
Will, G., Frazer, B.C. and Cox, D.E., 1965, The crystal structure of MnS04 , Acta Cryst.
19:854.
Will, G., Jansen, E. and Schafer, W., 1983, "POWlS-80. Program for calculation and re-
finement of powder diffraction data." Report JOI-1867, KFA JOlich, F.R. Germany.
X-RAY J?CloIDER DIFFRACTlOO QPA BY RIF:IVEI.D PATl'ERN-FITl'm:i
B. Jordan
Deakin University, Geelong
Victoria, Australia
INI'ROOOCTION
R = I Wi (Yi-Yci) 2 (1)
for intensity YLin the neasured diffraction pattern for profile point i,
Yci is the Rietveld-calculated intensity and Wi is the weighting
factor. '!he Yci are nodelled with the theoretical expression
PREFERRED ORIENTATION
where 0CJc is the angle between the preferred orientation direction and
the relevant reciprocal lattice vector. Pararreter r is a measure of the
degree of preferred orientation according to the pole density
distribution ranging fran 1.00 (randon) to 0.00 (fully oriented).
2000 ,-~----------------------------------~
GIB.B5 : MEASURED
>
!::
III
Z 1000
W
I-
Z
CALCULATED (MARCH)
>
I-
iii
z
W
1000
I-
Z
CALCULATED (NON·MARCH)
~
iii
zW 1000
I-
~
1/i 25 35 45 55 6S 7S 85
1.01---1.-;:-=,...".-----------------,
[[
w
I-0.9
w
:;
oct
[[
oct
a.
Z
0
i=
oct 0.8
I-
Z
w
a:
0
c
w
[[
[[
W
"- 0.7 .GIB-A5
w
[[
a.
GIB-A1D
• GIB-A15
•
GIB-B5
0.6
•
5 10 15
MORPHOLOGICAL RATIO, MR
CRYSTALLINITY
s = K Ck/J.!.. v~ (4)
using expression
(5)
B. H. O'CONNOR ET AL. 273
PHASE .ABUNDAOCE
• Actual values
• Actual values
• Rietveld, coarse Si02
80
• Rietveld values
80
o Rietveld, fine Si02
a
60 • 0c
60
N" N
..,e.
•
..,e.
40 40
20
• 20 o
20 40 60 80 100 20 40 60 80 100
% a-Si0 2
Fig. 3. Phase abundance assays for Fig. 4. Phase abundance assays for
oc-A1 20 3 /ZnO binaries. oc-Al 20 3 /oc-Si02 binaries.
274 IV. ANALYSIS OF DIGITAL DIFFRACTION DATA INCLUDING RIETVELD
100 100~---r-------....,
_ Actual values
• Rietveld, coarse Si02 I~ II-oActual values
Rietveld values,
80
o Rietveld, fine Si02
80 €)
coarse Si02 (constrained)
0'" 60 0'" 60
• I
.
I
1j 40 1j 40
20 - 20
20 40 60 80 100 20 40 60 80 100
REFERENCES
INTRODUCTION
Texture is a technical property, which carries importance in the every day pro-
duction processes in industry, for example of metals or ceramics. The anisotropy of
many physical properties originates from inhomogeneous orientation distributions of
the crystallites. X-ray diffraction is widely and commonly used for texture analysis. The
classical procedure is to keep the detector fixed at the 2,9 position of a possible Bragg
reflection and to turn the sample, positioned on an Eulerian cradle, stepwise through
the whole space about the two angles <p and X. With modern computer controlled tex-
ture goniometers and high-intensity X-ray tubes this procedure is fast. Nevertheless
only one pole figure is measured at a time, the procedure has to be repeated for each
hkl of which a pole figure is needed or wanted. For example if the three-dime.nsional
orientation distribution, called ODF (orientation distribution function) is requested there
are different modes of operation: equal-step scanning or equal-area scanning. We are
using equal-area scanning (in neutron diffraction) with generally more than 621 angular
positions <p and X.
for X-rays and neutrons, however in neutron diffraction this progress is available today
routinely. The evaluation of such a measurement yields a set of pole figures.
DATA AQUISITION
Neutron beams are large, in our case up to 100 x 50 mm ; neutrons are in general
little absorbed in the material and as a consequence large samples several cm in di-
ameter can be measured. Inhomogeneous samples or samples with coarse grains
ranging up to millimeters cause no difficulties; we can penetrate and therefore explore
the whole sample containing a sufficient number of grains with respect to statistics, we
can measure in transmission and therefore record and calculate complete pole figures.
Neutrons measure the "global" texture; in comparison,X-rays measure the "local" tex-
ture at the sample surface.
X-rays and neutron are complementary in many cases because of significant dif-
ferences in the scattering powers of the elements. In combination they are therefore
an excellent and powerful technique. This was demonstrated in a study on calcite 5,
CaC0 3, where some reflections like (00.6) are too weak in the X-ray diagram, however
very strong in the neutron diagram. For (11.0) and (20.2) the reverse is true.
In multiphase samples, like rocks or ores often containing low symmetry crystals,
but also in modern ceramic materials,the diffraction patterns contain a large number
of reflections which in general are highly overlapped. For a successful texture analysis
of such samples two requirements have to be fulfilled:
600
I II
;; n~
~N
"0
N
~
~O
I
I I IOJ I I I
n
0
0
0
~
N
0
N
"0
I I
400 N
;; "
0
'"c I OJ
2 OJ OJ
~
OJ
n
0
n
z 0
.0
ON
200
15 20 25 30 35
2Theto
Fig. 1. Individual di,ffraction diagram with q> = 20 0 and X = 90 0 • The time spent was 3
minutes.
G. WILL ET AL. (2) 279
s:. 4
~
I I
'"00 '"
~
~
""0
N
I
Im m
0
N
f: i::N
3 m m
m I I m I ml
0
N
0
;'
0
~
0
0
N
N
<0
0
0
'""" i::""
I I
Ul
C
;:> 1m m I 0] 10]
,. ,.:::'
:; 0] Q] Im m I
(l)
2
·0
;'
'"00 ;' '" o~ 0 '"'" "" '"
0 "-
z ~ I
~O
~ NO]
""
0
~
I
N
I "" 0
0
,.
~
I I I I I;::: I~ ~ ~I ~I h
NO
ON :::' 0
N
N
N
I I
2Theto
Fig. 2. Sum-diagram generated by adding up 621 individual diagrams like that shown in
Fig. 2. Included is the result of the profile analysis. The indices refer to the two
phases, H = hematite, B = biotite.
The crucial point in the data analysis of multiphase and/or low-symmetry materi-
als is the separation of overlapping peaks. This can be done by profile analysis and
pattern decomposition. I n the authors' laboratory powerfu I programs have been devel-
oped 7 analysing the data in 4 successive steps:
I
n
I m
C'J
3 ,0
C'J
I m
"
0
~
o
C'J
C'J
m m
:r: m m
2
n n
n
I
C'J
o
N
24 26 28 30
2Theto
4. Based on the known backgrou nd-shape-fu nction, the 29- positions and the FWHM
values the orientatioll-dependent intensities of the individual patterns are now de-
termined in an automated procedure by applying LSQ routines. In the given exam-
ple this took about 420 minutes for 636 individual diagrams using an IBM PC AT.
This procedure has been applied for the texture analysis of a hematite ore using
the neutron texture diffractometer of Bonn University at the FRJ-2 in the KFA Julich.
This diffractometer is equipped with the PSD JULIOS. The wavelength used was
1.09 A. The detector covered a 29 region from 12.5° - 37.5°. Fig. 4 gives a sequence
of such diffraction diagrams of varying x-angles. The highly developed texture of the
sample can be seen easily by the appearance and disappearance of peaks. The sample
16000
14000
20 30
2 Thela
Fig. 4. Sequence of hematite diffraction diagrams of varying x-angles
G. WILL ET AL. (2) 281
(a) (b)
Hematite Hematite
hkl =003
(c) (d)
Hematite Biot ite
hkl= 113 hkl=003
,
-'
:..
--, . - '
" "
, /
Fig. 5. Representat ive pole figures calcu lated from the integrated intensit ies after pro-
file analysis of all individual diagrams. Maximum and minimum pole densities
are (a) 12.47 and 0.02 (b) 5.37 and 0.09 (c) 3.03 and 0.15 (d) 4.42 and 0.42.
investigated consisted of two phases - hematite and biotite, a mica with monoclinic
symmetry. The measurement yielded a total of 31 reflections (Fig . 2), of which 15 were
considered to have enough intensities for pole figure representations. Fig. 5 depicts
four out of the 15 pole figures. The biotite content is quite small, about 5 vol.%, therefore
the further discussion concentrates on hematite.
DDDD
FMAX = 13.53
FM1N = -0.53
CONTOUR LEVELS
Z.O,
4.0,
DDDD
6.0,
8.0,
lD.O,
lZ.0X RANDOM
LSDCSu
uuDD
D 'Pl'
90.0' 97.5' 105.0' 112.5'
I <1> _ _
120.0'
Fig. 6. Orientation distribution fu nction (OOF) of hematite, projection ({J1 + ({J2 = const.
ocean. In this case all sailboats are concentrated around one point with a ±30° devi-
ation in E-W-direction. If we are interested in the direction the boats are heading we
must study the a-axis [1010J distribution, or equivalent (crystallographic) directions,
like [1120J (in Fig.5 we omit the hexagonal index i = -(h + k) for brevity); 110 of Fig. 5b
are the prismatic planes; Fig. 5c with 113 the pyramidal plane distribution.
ODF-CACULATIONS
ACKNOWLEDGEMENT
This work has been funded by the German Federal Minister for Research and Tech-
nology (BMFT) under the Contract Number 03-WI1BON.
REFERENCES
1. A. March, "Mathematische Theorie der Regelu ng nach der Korngestalt bei Affiner
Deformation.", Z. Krist., 81:285 (1932)
G. WILL ET AL. (2) 283
5. H.R. Wenk, H. Kern, W. Schafer and G. Will, "Comparison of Neutron and X-Ray
Diffraction in Texture Analysis of Deformed Carbonate Rocks," J.Struct.Geology,
6:687 (1984).
6. W. Schafer, E. Jansen, F. Elf and G. Will, "A Linear Position Sensitive Scintillation
Detector for Neutron Powder Diffractometry," J.AppI.Cryst., 17:159 (1984).
8. E. Jansen, W. Schafer and G. Will, "Profile Fitting and the Two-Stage Method in
Neutron Powder Diffractometry for Structure and Texture Analysis," J.AppI.Cryst.,
21 :228 (1988).
9. G. Will, "paWLS: A Powder least Squares Program", J.Appl. Cryst., 12:483 (1979).
10. G. Will, E. Jansen and W. Schafer, "POWlS-80. A Program for Refinement of Powder
Diffraction Data," KFA-Report Jul-1867 (1987).
11. S. Matheis, H.R. Wenk and G.W. Vinel, "Some Basic Concepts of Texture Analysis
and Comparision of three Methods to Calculate Orientation Distributions from Pole
Figures", J.Appl. Cryst., 21:285 (1988).
PROBLEMS IN THE DERIVATION OF d -VALVES FROM
Ron Jenkins
JCPDS-ICDD
Swarthmore, PA
Monte Nichols
ABSTRACT
recent round-robin test [1] in which participants were asked to measure samples of the
binary mixture CaC03/ZnO. Here the multi-person! multi-laboratory precision
(SIGMA) is about 0.048° . More often than not, the problem arises not so much from
the quality of the peak data as from the number of peaks found. Three questions should
be considered:
a) Were all of the "true" peaks identified?
b) Were additional (real) peaks found?
c) Were any false peaks found?
One major source of confusion is often the recognition of peaks by a given
automated system on a certain specimen, especially where data collected by manual and
automated methods are being compared (as is often the case in quahtative phase
identification!). The extreme sensitivity of, for example, the second-derivative
peak-hunting method may give significantly more lines than would be found manually.
Additionallx, the number of peaks found is very sensitive to the method of defining
what a sigmficant signal above background is. Many of these "peaks" may not be true
peaks at all but simply statistical variations in the background or on the shoulders of
real peaks. As an example, a recent round-robin [1] showed that for a well
characterized specimen of a-AI 20 3, the number of peaks reported by different users,
representing a range of instrument types, over the same 29 range varied from 25 to 53,
whereas the calculated number is 42.
- l~' J-
Compton
Seitter
COK«l
NIK«l
Cu K« 1
W L«
CuK« 2
CoK« 2
CoK{3
.J ~!e~lLro~~d
NI K« 2 CUK{3
NIK{31
7.0 7.5
I 8.0 8.5
____ ...J
9.0
level
(keY)
diffracted and detected may be more than simply the Ka doublet. The eu Ka emission
is actually made up of six a lines, but for most practical purposes it can be considered
bichromatic [5]. Figure 1 shows the spectral distribution in the energy region around
copper K radiation. The Cu Ka and Kp doublets are shown along with other lines
which could fall within the acceptance range of the monochromatizationldetection
device. There are three methods that are commonly employed to render the radiation
monochromatic (in actual fact, bichromatic because the Kal/Ka2 doublet is generally
employed). These techniques are: a) use of a p filter; b) use of a diffracted-beam
monochromator, or c) use of a solid-state detector, typically Si(Li). The Si(Li) detector
is now being used as a replacement for the conventional scintillation detector in some
powder diffraction instruments. While the advantages of wavelength selectability and
high quantum counting efficiency are somewhat counteracted by large dead time
problems, it is clear that the Si(Li) detector has an important role to play, especially for
the detection/measurement of very weak peaks. However, of the three
monochromatization devices, the diffracted beam monochromator is the more
commonly employed in automated powder diffractometers.
Figure 1 also shows the relative band pass regions for the three
monochromatization systems discussed. Because intensity is at a premium in most
diffraction experiments, most diffracted-beam monochromators employ a pyrolytic
graphite crystal, which has high reflectivity but poor dispersion relative to a "perfect"
single crystal. The angular dispersion of the pyrolytic graphite crystal is less than 100
eV, but because its mosaic structure allows diffractIon over a wide angular range, it has
an effective band pass of around 500 to 1,000 eV, which, if correctly employed, is still
sufficient to remove the Kp radiation. The actual band-pass value for a given
monochromator will depend upon the widths and positions of limiting apertures which
are placed somewhere on the focusing circle of the monochromator. It should also be
appreciated that a slight mis-setting of the monochromator can displace its energy
acceptance window to either lower or higher energy. From the foregoing, it is clear that
even with correctly set-up monochromatizing devices, it is possible that undesirable
radiation can pass through the monochromator to the counting circuits. When the
monochromatizing device is incorrectly set, even more radiation may be passed.
288 IV. ANALYSIS OF DIGITAL DIFFRACTION DATA INCLUDING RIETVELD
12~---------------------------,
'"
~
C>
"<D" 0.6
'"
U
0.4
0.2
/'
a)
Maximum -2.L ; 0
d(28)
b)
c)
Most powder diffraction work is carried out using the Cu Ka.JaZ doublet, and
one of the greatest experimental inconveniences arising from this chOice IS the variable
angular dispersion of the diffractometer. The angular dispersion of a diffractometer
de/dA is e<I,ual to (n/2d CoSe). Because the angular dispersion of the diffractometer
increases With increasing 2e, the a doublet is unresolved at low 2e values, while at high
2e values it is completely separated. In the mid-an~ular ran~e the lines are only
partially resolved, leading to some distortion of the diffracted hne profile, as well as
giving difficulties in numerical techniques used to separate a 1 from a2. The absolute
value of the angular dispersion varies from about 100 eV at low 2e values to about 2eV
at high 2e values. Thus the Ka 1/Ka2 doublet typically starts to be resolved at about 50°
2e;Figure 2 shows the angular dispersion as a function of 2e and indicates where the
al/a2lines are resolved at the half-maximum J?osition for different values of the line
Width. When usin~ manual peak-finding techmques, it is common practice to use the
weighted geometnc average of the al/a2 wavelength as the experimental wavelength, at
least until a 2e value is reached where the a 1/a2 doublet is sufficiently resolved to allow
accurate measurement of the alline. Where automated methods are used, a2 stripping
is generally employed over the whole range of the measured pattern.
More than 90% of all new powder diffractometers now sold in the USA are
automated. Many excellent computer programs are provided with these instruments to
aid the diffractionist in the reduction of powder data. Unfortunately, there can be
pitfalls in the use of some of these programs, especially for the newcomer to the
diffraction field. A popular misconception is that automated diffractometers are easier
to use than manual systems. What in fact is true, is that automated powder
diffractometers can give many hours of unattended and reliable data collection and can
greatly assist in the tedious and routine tasks of data analysis provided one is careful to
apply the programs in the correct manner.
crystallite size and/or strain, one should increase the step width accordingly. The
situation is clearly different where profile fitting techniques are employed because, in
this case, much more of the profile data are being used.
Essentially two computer-based methods of peak hunting are in use at the
present time - profile fitting and a combination of smoothing, a:2 and background
stripping, and a derivative method of peak location. The latter of these options is by far
the most common, even though the success of derivative methods has a strong
dependence on random counting errors. To minimize such errors, a combination of raw
data smoothing and differentiation of data by the least squares technique may be
employed [7]. The second-derivative peak-location method is commonly employed
[8,9], and this technique yields information about both the peak width and peak
position. Typical steps which may be employed in the data treatment process [10]
mclude the collection of the digitized data, smoothing and background subtraction of
the data, a:2 stripping [11] and finally, use of a peak-location method. This peak
location is typically done using a second-derivative method in which the negative space
is fitted to a parabola and the minimum calculated. A 2e calibration may then be
R. JENKINS AND M. NICHOLS 291
USE OF FIGURES-OF-MERIT
Figure 4 illustrates the case of a simple peak in which the full width at half maximum is
given by W, the width at the base as 2W, the average background is B, and the
maximum peak intensity above background is given as M.
The area under the peak is equal to MW and that under the background is 2BW.
It will be clear from this that as W increases (due to, for example, a specimen of small
crystallite-size), whereas MW does not change appreciably, 2BW increases
significantly. Application of the figure-of-merit to a wide range of data from different
diffractometer configurations and types has shown that the larger the value of FOM,
the statistically better are the net count data.
R. JENKINS AND M. NICHOLS 293
REFERENCES
ABSTRACT
A precise and practical method for the determination of d-values and lattice parameters
from digital diffraction data is described. Systematic errors are corrected mathematically
during a d-spacing (lattice-parameter least-squares refinement process making it unneces-
sary to use internal standards. X-ray and synchrotron diffraction data of an ICDD alumina
plate obtained with a wide variety of experimental conditions and analysis parameters were
used to study the precision in the derivation of d-values and the accuracy in the determi-
nation of lattice parameters. Results showed that the precision in determining d-values was
high with I~d/dlava ranging from 2x 10- 5 to 4x 10- 5. Using the results obtained from the high
precision XRD analysis as a reference standard, the accuracy in the lattice parameter de-
terminations from the synchrotron diffraction data reached the 1-2xlO-6 range. Lattice
parameters, with an accuracy in the high 10- 5 range, were also obtained using parameters
commonly used in a routine XRD analysis such as a wide RS (0.11°) for high intensity,
peaks only in the front reflection region, no Kcx 2 stripping, and a single 28 0 parameter for
systematic error corrections.
INTRODUCTION
The use of digitized data eliminates the time-constant errors associated with the analogue
strip-chart recordings and permits the data to be analyzed by computers. The data col-
lection method most frequently used is step-scanning with selectable fixed increment ~28
and constant counting time t per step, with the diffractometer under computer control.
There are a number of factors which limit the precision of the d's and the intensities. I As
in most experimental data there are two major types of errors: random and systematic. The
random errors can usually be reduced by improving the experimental conditions, e.g. using
longer counting times and reducing the background in the intensity measurements. The
systematic errors are generally inherent in the instrument geometry and require a knowledge
of their origin to properly handle them. In practice, most can be corrected in the least-
squares refinement and with standards.
The quality of the pattern is often limited by the quality of the specimen. Studies of the
effect of particle size statistics in polycrystalline specimens show that small particle sizes
(< lO,um) and/or specimen rotation are necessary to obtain reliable experimental data. 2 ,3 In
conventional X-ray tube focusing geometry each reflection is a KIX-doublet, and the ge-
ometrical aberrations such as specimen surface displacement, flat specimen, etc. vary with
28 causing systematic errors in the peak positions. 2 In parallel beam synchrotron radiation
powder diffractometry, the radiation is a single narrow wavelength which is selectable to
obtain high peak-to-background ratios, and the geometrical aberrations are absent, thus
eliminating most of the systematic errors. In addition, the resolution is high and the peak
shapes are virtually symmetrical. 3 - 5
The derivation of d-values from digitized data is one of the most basic data reduction
processes in diffraction analysis. The peak search (PS) method can 'be used for precision
peak determination of non-overlapped peaks, while the profile fitting (PF) method is needed
to resolve overlaps.6-8 The resolution, peak shape, and step size are important factors in
deriving d's from overlapping and tight clusters of reflections by both methods. A recent
JCPDS-ICDD round robin study on d-spacing accuracy reported large spreads of J128 or
J1d for analyses with no internal-standard corrections for systematic errors. 9
EXPERIMENTAL
Diffraction data obtained directly from the ICDD alumina plate with no specimen prepa-
ration and a wide range of experimental conditions were used in this study. As shown in
Table I, data were collected from a conventional Cu-target X-ray source at the IBM
Almaden Research Center and from synchrotron radiation at Stanford Synchrotron Radi-
ation Laboratory (SSRL). The resolution calculated from full width at half maxima
(FWHM) of the observed alumina diffraction profiles are plotted in Fig. 1. The top two
curves are obtained from XRD, and the bottom curve from SSRL synchrotron diffraction
data. Values of FWHM and the resolution of the XRD data vary depending on the aper-
ture of the receiving slit (RS) used. The larger the RS aperture, the larger the FWHM and
the lower the resolution. The synchrotron diffraction data obtained at SSRL have higher
resolution, and were obtained with an incident-beam SiC Ill) channel monochromator and
a set of long horizontal parallel slits to define the diffracted beam.
Table 1. Experimental Conditions for the Analysis of the ICDD Alumina Plate
0.2'
Fig. I.
2
I Plots of FWH\1 vs. 28.
~
u.
RS=O.ll' The increasing width at
the higher 28's is caused
• by wavelength dispersion.
0.1'
RS=0.03' 0
0
SSRL
O~~~--~~~~~~~~--~~~--~~~~
DATA ANALYSIS
The procedure for deriving the d-values is given in Fig. 2. The analysis can be either done
directly from experimentally observed Cu-Kcx] 2 diffraction peaks or from the net Cu-Kcx]
peaks with the Kcx 2 components removed mathematically. The Kcx 2 overlap may have a
significant effect on observed peak positions. The smaller the peak 28, the larger the effect
of overlap. For example, an analysis of the alumina XRD pattern obtained with
RS = 0.03° shows that the 28 displacements between observed Kcx],2 and pure Kcx] peak
positions were 0.009°, 0.004°, and 0 for peaks at 25.6°, 52.6°, and 70.4°, respectively. The
effect of overlap is more severe for peaks recorded with a wider RS.
The diffraction peak positions are determined either by peak search based on derivative
methods or by profile fitting. Systematic errors are then corrected by simultaneous least-
squares refinement of the d-spacings and the lattice parameters. ID •II Mathematically, all
systematic errors including zero-angle calibration, specimen surface displacement, flat
specimen, specimen transparency, axial divergence, etc. which may affect the peak posi-
tions can be included in the refinement. It has been found that the largest errors are due
to zero-angle calibration and specimen surface displacement, and can be represented
mathematically by 28 0 and Clcos8, respectively. The solution of a least-squares refinement
is reached and the values of 28 0 , C I and the lattice parameters are obtained when the sum
of the squares of the differences between the observed 28s, corrected for the systematic er-
rors, and the calculated 28s derived from the lattice parameters reaches a minimum.
Values of the observed d-spacings corrected for systematic errors (dear) by the least-
squares refinement and the theoretical d-spacings calculated from the refined lattice pa-
rameters (deal) are then calculated using the Bragg law. (The value of wavelength used in
the analysis of XRD data was 1.540562A.) The overall precision of the analysis is esti-
mated from the value of I~d!dl, where ~d = dear-deal' averaged over all peaks.
A typical XRD pattern of the ICDD alumina plate obtained with RS=0.03° is plotted in
Fig. 3. It should be noted that the intensities of the back reflection peaks which are plotted
in the the lower half of Fig. 3 have been expanded by a factor of 10.
The diffraction profiles of alumina reflections shown in Fig. 3 are relatively sharp, with
no significant line broadening, indicating that the material is well-crystallized. Diffraction
profiles obtained with RS = 0.11-" are slightly broader because of the wider RS. Typical
results of a least-squares refinement using PS data of the RS = 0.11 ° pattern are listed in
Table 2 for the front reflection region. Values of the 28 cor and dcor (corrected for systematic
errors), and deal calculated from the refined lattice parameters are given in columns 3, 4 and
100
~
Q)
.~
iii
E
0 0
c: 30 40 50 60 70 80 90
....> 10
'c;;
c:
....Q)
.5
o~~~~~~~~~~~~~~~~~~~~
90 100 110 120 130 140 150
Fig. 3. XRD pattern of the ICDD alumina plate.
T. C. HUANG ET AL. 299
Table 2. XRD Data of the ICDD Alumina Plate Obtained with RS = O.ll 0
and 0.013 respectively. To evaluate the accuracy, completeness, and reliability of the
0
,
measured d-values, the figures-of-merit I2 ,13 of the PS data were calculated using the PC
version of the 1\BS* AIDS program l4 : F 25 = 406 (0.0025,25) and M20 = 606 (1.1589, 20).
The XRD patterns obtained with RS=0.03° and O.llo were analyzed using a wide range
of analysis parameters. Results of four selected analyses are discussed below. The param-
eters used in these analyses and the results of determining d- values and lattice parameters
are summarized in Table 3. Both analyses #1 and #2 use the same RS = 0.03 pattern, 0
Kill peaks with KIl2 components removed, PS method for peak-position determination, and
28 a + Clcos8 for corrections of systematic errors. However, analysis #1 was refined using
300 IV. ANALYSIS OF DIGITAL DIFFRACTION DATA INCLUDING RIETVELD
Analysis # 2 3 4
I dd/d lavg 2.2x 10- 5 2.6x 10- 5 3.9x 10- 5 3.9x 10- 5
a (A) 4.75845 (3) 4.75851 (3) 4.75883 (7) 4.75882 (6)
c (A) 12.99190 (12) 12.9920 I (9) 12.99281 (30) 12.99283 (26)
Ida/al l.3x 10- 5 8.0x 10- 5 7.8x 10- 5
Idc/cl 0.9x 10- 5 7.0x 10- 5 7.lx 10- 5
only the high-angle peaks between 90-160°, and analysis #2 using the entire pattern from
20-160°. The precision in the derivation of d-values as estimated from the average value
of Idd/dl were 2.2xlO- 5 and 2.6xlO- s, respectively. Results also show that the a- and c-
lattice parameters obtained from these two analyses are practically the same.
As shown in Table 2, the maximum values of dd/d and d28 are 19.5xlO-s and -0.013°,
respectively, for the weak (122) reflection overlapped by the strong (OIS) reflection. The
overlaps of (122) and (0 IS) reflections can be resolved by PF. The results obtained from
the PS data of the non-overlapping peaks and the PF data of the overlaps of (122) and
(018) reduce the maximum values of Idd/dl and Id281 to 13xI0-5 and 0.009°, respectively.
The figures-of-merit also improve to F25 = 426 (0.0025, 25) and :\120 = 654 (1.15S9, 20).
Results obtained from the PS (analysis #3) and the PS+ PF data (analysis #4) are listed in
columns 4 and 5 of Table 3. Values of Idd/dl avg ' Ida/ai, and Idc!cl of both analyses are
slightly higher than those of analyses #1 and #2 probably because of lower precision for
peaks in the front reflection region.
As noted in the Introduction, synchrotron peaks have the advantage of single symmetrical
narrow profiles, which is important in deriving the values of 28's and d's with high preci-
sion. If desired, the number of peaks and the d-range can be greatly expanded by using
T. C. HUANG ET AL. 301
100
'0
Q)
.!::!
"iii
E
oc:: O~~~~~~~~~~~~A-~~~~~~~~~~
20 30 40 50 60 70 80
~
.;;;
20
c::
~
c::
O~~~~~~~~~~ __~~~~~~~~~
80 90 100 110 120 130 140 °20
The parameters used in the synchrotron data analysis are listed in Table 4. Analysis #5
was from data recorded with 0.01°28 steps, 1 second per step, and analysis #6 was a fast
Analysis # 5 6 1*
run recorded with 0.02° steps and 0.5 second per step. Both synchrotron diffraction pat-
terns have very high figures-of-merit with F Nand MN above 600 and 999.90, respectively.
To derive the values of d-spacings and lattice parameters, the actual wavelength used in
the synchrotron experiment must first be determined. Unlike the radiation obtained from
a conventional X-ray source for which wavelength tables of the characteristic spectral lines
are available, there are no lines in the synchrotron-radiation spectrum. Different methods
for experimentally determining the value of synchrotron wavelength have been used, e. g.
absorption edges and a silicon powder standard. 6
In this study, the wavelength was determined by making the sum of the lattice parame-
ters obtained from the calculation equal to that of the high precision XRD analysis # I; the
values are listed in Table 4. The precision in determining 28's and d's were good (i.e.,
1~2elavg:S:;0.00 16° and I~d! dl avg :S:;3.1 x 10- 5). Excellent agreement in lattice parameters were
also obtained. The discrepancies are negligible and well within the standard deviations es-
timated by the least-squares refinements (see numbers in the parentheses). Using the values
of analysis # I as a reference standard, the accuracy of the synchrotron lattice parameters
reaches the 1-2x 10-6 range (see values of I~a/al and I~c!cllisted at the bottom of Table 4).
It should be noted that the diffraction patterns collected using t~e same instrument at
SSRL but slightly different wavelengths and operating parameters both give practically the
same values of lattice parameters.
SUMMARY
It has been demonstrated that precise d-values and accurate lattice parameters can be ob-
tained from digitized diffraction data by proper corrections for systematic errors during the
d-spacing / lattice parameter least-squares refinement without using internal standards.
An analysis of the XRD and synchrotron data of the ICDD alumina plate showed that
results with l~didlavg;::::2.2xIO-5, l~a/al;::::2.lxlO-6, and l~c/cl;::::l.5xlO-6 were obtained.
ACK~OWLEDGEMENTS
We thank SSRL for providing the synchrotron radiation facilities for this study, R. Cernik
and M. Hart for the Daresbury pattern, R. Jenkins of JCPDS-ICDD and W. Y Schreiner
of the Philips laboratories for supplying the ICDD alumina plate, and R. G. Garvey of the
1\0rth Dakota State University for modifying the NBS*AIDS program.
REFERENCES
II W. Parrish, M. Hart, T. C. Huang, and M. Bellotto, Adv. X-Ray Anal. 30,373 (1987).
John R. Burleson
INTRODUCTION
Processing time and disk storage penalties are paid as the step/chop
size is decreased. See Table I. The processing times listed are those
required to execute the background subtraction and peak finder routines and
TABLE I - MEMORY AND DATA PROCESSING PENALTIES WITH DECREASING STEP SIZE
to draw the pattern four times. As with most of the data presented here,
the weight of the positive or negative effects of an item will depend on the
capabilities of the individual system and the sample through-put of your
laboratory. Your situation may be more or less sensitive to computer memory
limitations or time spent to draw a particular pattern.
The detection of weak peaks is, to a great extent, the ability to dis-
tinguish the weak peak from the background noise. If the total scan time
is held constant and the step size is decreased, background noise is
increased. The noise is generated by the deteriorating counting statistics
caused by the dwindling counting time. A typical situation is displayed in
Table II. The effect is easily seen by examining the size of one standard
deviation (one sigma) and the total number of extra counts needed to produce
a peak 0.1 degrees wide and three sigma above the background. Note that the
average count rate in counts per second remains rather constant for all
seven data sets. Remembering the background data, examine the data from a
weak and a strong peak. The average intensity of both peaks generally in-
creases as step size is decreased. However, for the weak peak collected
with a small step size, the intensity above background is very low but in-
creases for moderate step sizes.
The data for Table II was developed from patterns collected during con-
tinuous scanning. For step scanning, the counting statistics deteriorate
when the scan is collected during the same fixed time. The reason is that
during step scanning, a larger portion of the time required to collect the
scan is spent doing bookkeeping, not counting. The total time required to
collect a single scan may be expressed as the summation of: (1) the time re-
quired to change the sample, setup and close the computer files, (2) the
time actually spent counting, (3) the time spent stepping the diffractometer
to the next location and (4) the time spent reading and clearing the counter
after each step. When a well designed diffractometer is used in the contin-
uous scanning mode, the third term is zero and the fourth term approaches
zero. The summation (T ts ) may also be expressed in terms of an initial
constant (Ci), the angular range scanned (A), the step size moved (S), the
preset or counting time at each step (Tp ), the speed at which the
diffractometer moves (R) and the delay time spent for bookkeeping before
and/or after each step (D).
To obtain an estimate of the delay time D and the stepping rate R for
various diffractometers, I asked several of my x-ray diffraction colleagues
around the country to make two scans on their automated units. The request
was for one eleven-degree scan with 0.1 degree steps and a preset time of
one second. The other was to be stepped at 0.01 degrees with a 0.1 second
preset time. With two sets of data, the two unknowns could be determined.
308 IV . ANALYSIS OF DIGITAL DIFFRACTION DATA INCLUDING RIETVELD
A 37 0.056 90 .035
B 1 19 1. 24 40 .075
B 0.1 19 1.24 7 .045
C 8 0.078 72 .040
D 1 32 0.24 76 .040
D 0.1 32 0.24 28 .0lD
Some of the instruments could not count for preset times less than one sec-
ond. For these instruments another scan with different parameters was sub-
stituted. The data from instruments of four different manufacturers is
shown in Table III. R ranges from 8 to 37 degrees per minute and D ranges
from 0.056 to 1.24 seconds per step. The efficiency of data collection av-
eraged 70% with a preset time of one second and declined to an average of
18% with a preset time of a tenth of a second. With the specified 40 minute
total scan time and Ci set at 3 1/2 minutes, the smallest possible step
size averaged around 0.04 degrees. On the other hand, for the two instru-
ments in our laboratory capable of continuous scanning, one commercial and
one home brewed, the counting efficiency is 99.9+%. This is based on our
designed and the OEM's specified delay times D in the range of 5 micro-
seconds. With these very short delay times, there is no practical lower
limit on the chopper size.
THE POSITIVE POINTS OF DECREASED STEP SIZE - MORE ACCURATE PEAK INFORMATION
Table IV displays three values for many combinations of slit width and
step size for the most intense peak of alpha alumina. As a measure of the
peak profile, the first entry for each combination is the full width of the
peak at half its maximum intensity (FWHM). The second line for each combin-
ation is the maximum intensity of the peak in counts per second. This is
followed by the location of this maximum count.
The smaller step sizes do improve the profile of the diffraction peak.
However, the noisy statistics of the low count rates are exaggerated by the
very short preset times of the very small step sizes. A point of minute
gain in either intensity, location or peak width occurs when the step or
chopper size is less than the angular width of the receiver slit. The ang-
ular widths of the receiver slits (at 230 mm) are included in this table.
Asterisks mark the approximate location of these slit widths. Diffraction
patterns from combinations to the left of the asterisks follow the same pro-
file as the one just right of them, only they are noisier. The profile of
the peak also gradually degrades with increasing step size, but even at
0.125, the peak is still very evident: Note that as the step size in-
creases, the probability of a data point falling close to a peak maxima
decreases. Therefore, peak maxima move around to the location of the near-
est data point. For stepping, the maximum intensity decreases due to the
position of the data point on the side of the peak. For continuous scan-
ning, the decrease is due to the larger contribution of lower intensity data
from the sides of the peak.
A NEUTRAL POINT OF DECREASED STEP SIZE - THE PROPOSED USE OF THE DATA
One must consider the end usage of the pattern before scanning
parameters are selected. The previous discussion has shown that when the
task is simply to check for the existence of major phases, large step sizes
are satisfactory. If, on the other hand, information about a peak profile
or the location of a peak is sought, smaller step sizes are Ln order.
900.0
750.0
600.0
450.0
J
300.0
150.0 f-
0.010
44 54 64
Figure 1. Three half patterns with decreasing chopper size, originally
displayed on high-resolution color graphics equipment.
very large step sizes were included in the various tables. For many color
CRT's, that is all that can be displayed. If full scans to 70 degrees are
only to be displayed on a low or medium resolution CRT, step sizes less than
0.12 are a waste of resources and generate noise. One note on thin-line
hardcopy color devices, if the line is too thin, the human eye can't dif-
ferentiate the various colors. On AC's high-resolution color copier, with B
size paper, it is necessary to double the line width to allow the observer
to distinguish the different colors. Color plotters have a very high number
of addressable points, but the fine tip pens are twelve times wider than the
point addressed.
The fact that small step size has very little effect on the display of
even half of a full pattern is shown in Figure I. Very little difference is
detected between the three patterns collected with chopper sizes of 0.01,
0.03 and 0.05 degrees. Note that the three patterns have all been normal-
ized to 500 counts per second full scale.
A NEUTRAL POINT OF DECREASED STEP SIZE - BALANCING SLIT AND STEP SIZES.
Peak profiles are a function of both step size and optics. However,
not all combinations enhance the pattern and may actually reduce the quality
of the pattern. Also, changing a slit does have an effect on the alignment
of the instrument. If more precise data is needed, just going to finer
slits and longer counting times may not yield any better information. Try
changing just the antiscatter slit that is ahead of the detector slit.
While intensity is increased greatly with larger slits, the location of the
maximum intensity and the peak centroid shifts toward lower angles.
CONCLUS IONS
* Select the scanning parameters by considering the end use of the pat-
tern.
* Experiment with your instrument. Determine how the slits are related
to each other and how, in conjunction with changing the step size, they
affect:
Peak intensity,
Peak profile and
Peak position.
* For routine phase analysis, step sizes in the range of 0.03 to 0.04
degrees are fine if:
A good quality CRT is used for viewing the pattern and
The pattern is to be viewed in sections.
INTRODUCTION
51al""' ..
51..IFrb
Boo Window
diffractometer and opening the incident beam shutter, it was found that
the beam, in fact , grazed the frame at lowe angles . The cell's Al
reference surface parallel to the face plate was found to be too far
forward on the diffracting circle. The x-rays were thereby occulted at
very low angles, producing an anomalously low background and also
generating interfering aluminum and aluminum oxide diffraction peaks, when
exposed to the high-intensity x-ray beam. This was corrected by machining
back the aluminum frame surrounding the Be window so as to make the
reference surface coplanar with the inside surface of the Be window, i.e . ,
the front surface of the contained specimen. Following such modification,
the cell was again checked with phosphor on the reference surface . With a
1/2 0 incident beam slit for diffraction angles from 10 0 to 25 0 , and
a 1 0 slit for diffraction angles of 25 0 and above (see next section),
the x-ray beam no longer struck the frame, and the aluminum/aluminum oxide
lines were no longer present in the diffraction pattern .
DISCUSSION
1000
925
~
.~ 850
oS!
..5
775
700
21.5 22.0 22.5 23.0 23.5 24.0 24.5 25.0
2-Theta
Figure 2 -100 Second Count Time
4000
3125
~
.~ 2250
CI)
\
];
\"-~~
1375
\ 4 Second Scan
500
40.000 40.714 41.429 42.143 42.857 43.571 44.286 45.000
2-Theta
2250.0
1812.5
~
'iii
1375.0
gs::::
.E
937.5
0.4 Second Scan
500.0
41.0 41.5 42.0 42.5 43.0 43.5 44.0 44.5
2-Theta
Figure 4 - Comparison of 100, 4 and 0.4 Second Scans
SUMMARY
With the use of a high intensity x-ray source, the extraction of weak
peaks under adverse conditions is possible. Success, however, depends
both on control of experimental parameters and on the use of the more
evolved analytical tools which provide profile - fitting capabilities.
Among the experiment factors of prime interest are: control of the
background; precise and accurate assessment of the background through the
318 IV. ANALYSIS OF DIGITAL DIFFRACTION DATA INCLUDING RIETVELD
3000
2250
Fitted Data
~
.~ 1500
oS!
.E
Synthetic Profile 1
750
o
29.2 29.6 30.0 30.4 30.8 31.2 31.6 32.0
2-Theta
Figure 5 - Doublet in the 4c Sample
ACKNOWLEDGEMENT
REFERENCES
Zenjiro Yajima
Yukio Hirose
and
Keisuke Tanaka
INTRODUCTION
EXPERIMENTAL PROCEDURE
The material used was sintered silicon nitride ceramics with additions
of Y203 and A1203 as bonding compounds. Young's modulus was 276 GPa.
The bending strength was 399 MPa. The three-point bending specimen was
used for the fracture tests. 5 The radius of the notch-tip was varried
between 0.10 and 0.50 mm. The notch was made with diamond blades and
All the specimens were fractured under three-point bending. The onset
of crack extension was detected from the load-displacement record and
the acoustic emission method. 5 The fracture toughness value was expressed
in terms of the stress intensity factor which was calculated using the
initial notch length as the crack length for notched specimen. 3
X-Ray Observation
X-ray equipment used for the measurement of the residual stress was a
stress analyser ( Rigaku MSF). The diffraction profile of Si 3 N4 (411)
plane was obtained by Cr-Ka X-ray from the area of 2 mm in diameter.
The X-ray stress constant was first determined. Fig. 1 shows the loading
apparatus used for the measurement of the stress constant. The conditions
of X-ray observation are given in Table 1. The incident angles of X-rays
weneO°, 10°, 15°, 25°, 30°, 45°. The residual stress in the crack-growth
direction on the fracture surface was determined by the sin2~ method.
The distribution of the residual stress beneath the fracture surface was
obtained by irradiating X-rays on the new surface revealed by successive
chemical polishing. Fig. 2 shows the schematic illustration of the X-ray-
irradiated area on the fracture surface.
(1)
125.4
Ea (xl0-6)
--------
---0-
- -fr-- 400
0
125.3
7-.'-----z
--0-- 800
- .. ~ 1200
01
T----.Q._~,~
-e _-1:.--_
QI
~ 125.2 .n.
CD ----~-- -
0
"~--
N
1251 "----
125.0
o 0.2 0.4 0.6
0.1
o
01
QI
'0 -0.1
~I~
CDN
N C
~iii
ttl -0.2
o
-0.3
d ( 28 )
Fig. 4. vs. Ea'
d sin2~
322 V. X-RAY STRESS ANALYSIS
where EM is the mechanical Young's modulus ( 276 GPa). The obtained value
was K = -932 MPa/deg. By using K = -932 MPa/deg. the residual stress oR is
determined from the measurement of M through
(2)
Fig. 5 shows the record of load versus crack opening displacement for
p = 0.40 mm. The change in acoustic emission total count and total energy
with displacement are also shown in the figure. The relation was linear
up to the point of the fracture. The fracture toughness, Kp , of bluntly
notched specimens is plotted against the square root of the notch radius in
Fig. 6. As seen in the figure, the relation could be roughly approximated
by two straight lines. Kp was proportional to IP at large p values.
It approached a constant value Kp, i.e., the fracture toughness of the pre-
cracked specimen, as p became smaller.
Fig. 7 shows the distribution of the residual stress near the fracture
surface. The residual stress measured on the fracture surface was
compression. As the depth increased, the compressive residual stress
slightly increased and then gradually diminished. At a certain depth, the
residual stress approached a constant value. The residual stress was cause,
by the local plastic deformation near the crack-tip.
o.5.----~--.--....._-.__-~2.0 1.0
p = 0.40mm
0.4
N
----.. 0.8
1.5
N
>
z 0.3 Z O.Sw
"" E C
1.0 ~
u
'0
"o
-'
05
02
L--~~--~~--~---~O o
10 20
Displacement u (fJm)
10
~0
Q.
~
8
Co.
:.::
;;
:.::
..
tf
6
Ci"
>
..
In
In
I:
.c / I'( p (mm)
4
'"
/
0 0
2"
/
()
.. I
/ 0.10
t> 0.15
•
I
I
~U 2 I
I 0.25
a I l::. 0.40
tt I 0 0.50
I
I
I
I
0
0 0.2 0.4 0.6 0.8
Square root of notch radius .fjf (Iffiiii)
50r------.-------r------,-------r-----~
0
Q.
~
~
In
In
~
iii -100 ""-
0
p(mm)
a
C
t> 0.15
"
.. •
u
iii -150 0.25
a:
0 0.50
-200
0 10 20 30 40 50
Depth from fracture surface y (fJm)
The plastic zone size, Wy , can be defined as the depth where the
residual stress approaches zero. In Fig. 8, the depth was plotted against
the toughness divided by the bending strength aBo The value of Wy was
proportional to the square of Kp/aB' i.e.
324 V. X-RAY STRESS ANALYSIS
(3)
40
E
:1
20
3'"
..
~
0.
..
"0
c
0
10 ~ p(mm}
N
0 0
u
V;
() 010
a::" t) 015
5
•
t:.
o 25
o 40
0 050
02 0.5 10
Kpl crB ( Imffi" )
Table 2. Values of a[= Wy /(KP/cry)2] and yield stress in the plastic zone.
Bending Bending
strength strength
a in plastic zone
a H (MPa) a H • (MPa)
Si, N. 399 0.067 597
TiC-Cr,C 2 689 0_ 1 08 812
Al 2 O, 263 0.048 465
Z. YAJIMA ET AL. 325
.Si3N4
.. TiC- Cr3CZ
a 1500 o
Q. • Alz03
~ o Ductile cast iron
"t) o Quenched-and-
.c:
0
tempered steel
t:. Normalized steel
N
1000
.~
111
0
a.
c:
111
111 ay'=- 5396 .15130'y
~ 500
iii
'0
Q;
>
o~--~----~----~----~--~
o 500 1000
Monotonic yield stress cry (MPa)
The result of the present study agrees with this equation. From equations
(5) and (6), the a value is given as a function of Oy.
CONCLUSION
(1) The fracture toughness was proportional to the square root of the
notch-tip radius for the cases of notch-root radius larger than 0.10 mm.
For the cases of sharper notches and pre-cracks, the fracture toughness
was nearly constant.
(3) The plastic zone size wy was determined on the basis of the
distribution of the residual stress beneath the fracture surface.
The size was related to the fracture toughness value Kp and the bending
strength 0B through
(4) The published data on the a value determined for various kinds of
steel and ceramics are related to yield stress 0y by the following function:
REFERENCES
Zenjiro Yajima
and
Keisuke Tanaka
INTRODUCTION
In the present paper, the distribution of the residual stress beneath the
fracture surface of a high strength steel (AISI 4340) made by stress-corrosion
cracking was measured with the X-ray diffraction technique. Stress-corrosion
cracking tests were conducted by using compact tension specimens in 3.5% NaCl
solution environment at various temperature between 277 and 343K. Based on
the results, the mechanisms of stress-corrosion cracking were discussed.
EXPERIMENTAL PROCEDURE
Polyethylene ball
Environmental solution
J Water bath
P
Filter V
Figure 3 shows the relation between the crack growth rate and the stress
intensity factor. The value of KISCC is the value at which no crack growth
could be detected after 240 hr. The value of KSC is the value of K above
which the crack becomes unstable. The curves are divided into three regions:
Region I is near KISCC' Region ill is near KSC' and Region II lies between
these two. The crack growth curves can be characterized in terms of the
following four quantities: (1) KISCC' (2) KSC' (3) KII is in the middle of
Region II, (4) crack growth rate is (da/ dt) II corresponding to KII •
The change of the crack growth rate with temperature at several K values
is shown in Fig.4, together with the diffusivity change with temperature. At
each stress intensity factor, the crack growth rate decreases linearly with
the inverse of temperature. The diffusivity shows a similar change with
temperature.
C
E
E
S
U 10°
u
a
!:
2
:5
~
0
~ 10- 1 (> 277 K
-'"
u
v 283 K
2 0 293 K
u
b. 313 K
0 323 K
~ 343 K
o 20 40 60
Stress Intet1sity factor K (MPav'iTi)
u
c: a.
10 2
~
~
E 10° N
E
E u
E
~
-
-0
0
"0
~
0
~
a. C
a.-
~ 10-' 10' ~
.r; *.
a.
~ 0
0 0 K=124MPav'ffi u
0, c:
.><
6 K=186MPaVm ~
u Ul
~ 0 K =24 8MPav'ffi ::J
U
oMPav'ffi \ '-:
•
'V K=31 0
10- 2
• 0 lO v
28 30 32 34 36
1 IT (xlO- 3K-')
(
~--
50llm
50 101m
• 1-------1
•
40
E
::l...
3"'
30
/
M
.r:
J;V.:==~
1/
u
20
~
CJ1
c "-ij/
.r:
u
c
0
<>
~
co 10 277 K
0 293 K
'"
0
313
323
K
K
o '* 343
L - J -_ _ _ _ _ _- L_ _ _ _ _ _- L_ _ _ _ _ _
K
~
10 20 30 40
Stress Intensity factor K (MPalm)
313 K
~ 200
'"
"'"
100
"'"
0:: 0
o 10 20 30 40 50
Depth from fracture surface ( I'm
K =18.6 MParm
300 <> 277 K
o 293 K
6. 313 K
o 323 K
~ 200 tr 343 K
'"''""
o 10 20 30
Depth from fracture surface (I'm)
The plastic zone size, wY' is defined as the distance at which the
residual stress approaches the initial value. Figure 9 shows the relation
between plastic zone size, wY' and the stress intensity factor K divided by
the yield strength, Oy. It is noted that the plastic zone size, wY' is
proportional to the square of K/oy. The relation between the plastic zone
size, wY' and the stress intensity factor K can be approximated by the
100
[7
1 /
E
:l.
50
2
Q
/~/
/il
>-
3 ~ 0
s=
20
Q
/J
Q.
'"
,1/
"0
'"
c
0
10
N
a 0 277 K
U1 5 / 0 293 K
c
Cl. 6. 313K
o
2
* 323 K
343 K
0.2 0.5 10
K/CJ,. (rmm)
Fig.9 Relation between plastic zone depth and stress intensity factor
divided by yield strength.
M. TSUDA ET AL. 333
150 10'
a 10 3 E
a.
a.
10 2 0
U
1.00
10'
c:
~
10° c:
0
Co u
050 10-' c:
III
Ol
0
10- 2 -0>-
I
10-3
o
280 300 320 340
Tempera t ure T (K)
where Oy is the yield strength. The I). value is dependent on the temperature.
The proportional constant I). in eqn (1) was calculated to be 0.15 by Levy
etal. 8 with a finite element method for elastic perfectly plastic material.
The experimental value obtained in the present study was smaller than 0.15 and
varied depending on the temperature. Two factors can be responsible for the
decrease of ~ from 0.15. The one is the weakening of the stress concentration
of a crack due to micro-branching, and the other is the hardening of the
material within the plastic zone. Figure 10 shows the change of ~ with test
temperature. The ~ value was minimum at 293K and maximum at 343K.
CONCLUSION
(1) The residual stresses measured on the fracture surface were all
tensions. The residual stress in the vicinity of the fracture surface depends
on the test temperature. The residual stress increases gradually and then
decreases with increasing distance from the surface. The peak tensile stress
in the subsurface increases with increasing K value and the test temperature.
(2) The relation between the plastic zone size, WY' and the stress
intensity factor, K can be approximated by the following fracture mechanics
equation:
W Y = ~ ( K/ 0 y ) 2
334 v. X-RAY STRESS ANALYSIS
where ay is the yield strength. The a value depends on the test temperature.
The a value took minimum at 297K and maximum at 343K.
(3) The change of a value was related to the micro-branching width and
solubility, CO. At 277 and 343K, the width of micro-branching is small, the
a value is large. There is an inverse relation between a and C •
REFERENCES
INTRODUCTION
o 256(0)
Encoder number n (Time)
Amp.
Diffraction pattttrn
Start
or
CPU
(NEC PC-980IVX)
INSTRUMENTATION
The MCA used could be computer controlled for the start and end of
sampling of the diffracted x-rays. The MCA and the rotary encoder were
connected to a personal computer (PC980lVX) to measure time-resolved
diffraction patterns. This computer also handled the x-ray incident angle
settings, LPA correction, determination of peak position and stress
calculation.
155.10
1
tp' = o'
t>
~ ~ ~.
~ 155.05
!!Ff:]]~~~
• ~_ ,1(.1,. t ~:
., ....
f . 'J.6-;-
0
.a
.iI:,
I 00
~
~ 0
!
OJ
~<.J 156.00
o •
1...
...Ci"
~ t.
0..
J '--
155.950 1024 2048 3072
Peak counts of diffraction profile (counts)
EXPERIMENT
The specimen used in this experiment was a low carbon steel with high
tensile strength. The yield strength was 853 MPa while the tensile strength
was 902 MPa. The specimen was 18 mm in width, 120 mm in length, and 1 mm in
thickness. The specimen had little initial residual stress.
~ 200 '.
,£"
.... \~
4 215 ± 42* 8603 V)
r
-·t
the ordinary measurement. When sampling time was less than 8 ms, the stress
value did not agree with that of the ordinary measurement and the confidence
limit was more than twice as large. In addition the sampling cycles
necessary for one stress measurement increased in inverse proportion to the
decreases of sampling time.
Minimum (80MPd)
200 IIL-_~II_ _ :&
ti
~
"" 0
<J)
<J)
.......
Q)
() 2Hz
V)
(J 8Hz
-200 o Static
• Residual Stress
Width of specimen •
-9 o 9
Distance from center (mm)
Fig.5 Stress distributions at maximum and minimum applied stress.
340 V. X-RAY STRESS ANALYSIS
CONCLUSION
REFERENCE
David L. Milam
Material Science Research
The Timken Company
Canton, Ohio
ABSTRACT
An investigation was conducted of the potential for detection of the
contact stress distribution in tapered roller bearing components made of
carburized steel. Various raceway stress distributions were created by
control of the profile and alignment of the rolling surfaces. Determination'
of residual stress in the rolling direction and peak breadth were made
using a multiple tilt sin~ w technique. The measurements were made at the
surface and at depths below the surface. It was found that the distribution
of residual stress and peak breadth perpendicular to the rolling direction
in the layers below the surface of the inner race correlated highly with
the imposed contact stress distribution. The correlation was less in the
case of the outer race. The residual stress distribution measured on the
surface of the inner race was observed to be inversely proportional to the
contact stress distribution.
INTRODUCTION
Tapered roller bearings consist of four basic components. Shown in
Figure 1 are the cone (inner ring), the cup (outer ring), tapered rollers,
and the cage (roller retainer). Under normal operating conditions, the
cone, cup, and rollers carry the load while the cage separates and retains
the rollers on the cone. For best performance and life of tapered roller
bearings, the cup and cone should be aligned such that their axes are
coincident. The angular deviation from coincidence is the measure of
misalignment. The causes of misalignment include shaft deflection, housing
deformations, and inaccuracies in machining of the shaft or housing.
The contact stress at the surface of an aligned bearing is a maximum
at the middle of the contact width, Figure 2. Subsurface shear stresses are
created by the contact stress distribution at the surface. The maximum
shear stress is the unidirectional shear stress acting on a plane
inclined 45° to the rolling surface, -r45; -r45 is a maximum beneath the
center of contact. A second important shear stress is the orthogonal shear
stress acting simultaneously on planes parallel and perpendicular to the
surface, -rOo The maximum value of -r0 is located below the edge of contact.
The contacts and stress distributions analyzed deviate from the line
contact concept to various degrees, but the general concepts still hold.
~ Rollers
Po
to ot
- -LOmax +LO max -
Table 1
Nominal Composition of Bearing Components weight percent
C Mn Si Cr Ni Mo
0.20 1.00 0.25 0.60 0.30 0.13
Table 2
Crown Radii of The Two Groups
The roller bodies of both groups had a crown radius of 175 inches.
Crown
Radius, inches
Group A Cone 25
Group A CUp 47
Group C Cone 280
Group C CUp 100
PROCEDURE
Bearings made from a carburized low alloy steel were subjected to an
induced misalignment condition. The nominal composition of the material is
shown in Table 1. The outer diameter of the bearing is 65 mm; the bore is
35 mm. A 0.00333 radian misalignment condition was created within a bearing
fatigue test machine by making the end cup and center cup adaptors off-
square. A peak radial stress of 2,570 MFa (assuming line contact) was
imposed on the bearings for 6 x l07revolutions. The lubricant was automatic
transmission fluid maintained at an inlet temperature of 63°C.
There were two treatment groups of bearings in this test, groups A and
C, differing in raceway contour. Group C bearings had a flatter crown on
the cup and cone than did group A, see Table 2.
344 v. X-RAY STRESS ANALYSIS
,,
O+T--------+---------+---------+-~r_~--~
-1000
,,
\
Cd
\
~-2000 \
I
ts
\
\
\ /
/
/
\
,
/
~ -3000 /
\
U .,. ./
A
-4000
-~+---------~------~---------+--------~
1 2 3 4 S
Large Small
End Position on Raceway End
Fig. 3 - Cone-roller contact stress distribution across
the raceways of Group A and Group C cones.
0+---------+---------+---------+---------+
-800
......._•.....•...._........ _--_.._--_._-_...._..•.__...._....._--_..._._..._.........
-1000 +---------+---------+---------+---------4-
1 2 3 4 5
Large Small
End Position on Raceway End
Fig. 4 - Residual stress measurements of untested cones.
6.7 +---------'---------~---------i---------_I_
6.6
6.5
Depth in mm
bO - _.. Surface
-8 6.4
.-----. 0.05
f
~
6.3
- - - 0.10
- - - 0.15
- - 0.20
~
---- 0.25
~ 6.2 - - 0.30
6.1
6.0
~. . . . . . . . . . . . . ~......
---_.........-........-....... a •••••
-_
..... .....
~............ . . . .
5.9 +---------I--------~---------+---------I-
1 2 3 4 5
Large Small
End Position on Raceway End
Fig. 5 - Peak breadth measurements of untested cones.
346 v. X-RAY STRESS ANALYSIS
-BOO -h-----+----+-----+-""7---+ 0
-- __ ..... , Tested
,,,
--
/C
-900 -1000
os
~
~
----- Untested -
-2000 t'"
(I)
tl
~
U
-1100 -3000
RESULTS
-800 +-----+-----I-----+----r--__+_ 0
-1000
-900
-1100
-4000
-1000
-200
os os Depth in mm
~.,; ~ - 0.05
2000 • ..----- 0.10
'" -300
~
- ~ _ . - 0.15
fIl ~ --- 020
fIl _ _ 0:25
~
u _ ... _ 0.30
~ -3000 ~ --_.._.. Contact
~ u
-400
-4000
-500 +-----+----+-----+-----+
1 2 3 4 5
Large Small
End Position on Raceway End
Fig. 8 - Subsurface residual stress measurements
of tested Group A cones.
-1000
-200
Depthinmm
~.,;
~
~ - - 0.05
-2000 1i .------ 0.10
15'" 15 _.- 0.15
fIl -300 fIl - - - 0.20
~ t) - - 0.25
::I
:sa -3000 U~o ---- 0.30
............ Contact
'"
Q.)
I:x:
-400
-4000
-500 +-----+----+-----+-----+
1 2 3 4 5
Large Small
End Position on Raceway End
Fig. 9 - Subsurface residual stress measurements
of tested Group C cones.
D. L. MILAM 349
6.8+-----01-----1-------1-----+
6.7
6.6
" ...
~
....""',\;- DeptbinDUD
--_...... Surface
-- 0.05
---~:. .-----. 0.10
"" _.-
---
--
0.15
0.20
0.25
---- 0.30
6.1
6.0
5.9+-------11------+----+-----+
1 2 3 4 5
Large Small
End Position on Raceway End
Fig. 10 - Peak breadth measurements
of tested Group A cones.
6.9
6.8
6.7
~..--- ....~
6.1
6.0 -t------+------+-----+------+
1 2 3 4 5
Large Small
End Position on Raceway End
Fig. 11 - Peak breadth measurements
of tested Group C cones.
350 v. X-RAY STRESS ANALYSIS
DISCUSSION
The residual stress distribution across the raceway of the cones, both
on and below the surface, was observed to have been altered by the contact
stress. In the case of the cups, however, there was no significant change
in the state of residual stress (neither on nor below the surface). This
difference in observed residual stress differences is caused by the
difference between the cone - roller and the cup - roller contact stress
distributions. The magnitude of cup - roller contact stress distribution
was less than that of the cone - roller contact.
A contact stress greater than the yield stress is necessary for
plastic deformation and creation of residual stress. It has been reported
that a minimum contact stress of 3,200 MFa is necessary for alteration of
the pre-existing state of residual macros tress in through-hardened bearing
steel.[4] Because the pre-existing residual stress state in carburized
bearing steel is significantly different than that existing in through-
hardened bearing steel, it may be that the minimum threshold contact stress
for alteration of residual stress is different, too.
In the case of the Group A cup - roller contact, the maximum contact
stress is calculated to be 2,800 MFa at position 3. Because there was no
significant alteration of residual stress in the cups of Group A, the
minimum contact stress required for alteration of residual stress is
therefore greater than 2,800 MFa.
It is not possible to estimate the minimum contact stress necessary
for alteration of residual stress from the correlation between residual
stress and contact stress observed in this investigation because the
spatial resolution on the raceway surface is insufficient. The diameter of
the incident beam was nearly equal to the distance between positions. The
diameter of the irradiated area is too large relative to the large slope of
the cone - roller contact stress distribution at position 3 (Figure 3).
There exists a direct correlation between the subsurface residual
stress distributions and the contact stress distributions in both groups of
cones (Figures 8 and 9). The compressive contact stress generated a more
compressive increment to the pre-existing compressive residual stresses
shown in Figure 4. It is believed that the correlation can be improved by
the use of a smaller irradiated area. A 0.5 x 3.0 mm source collimator has
been used in more recent investigations with the long axis aligned in the
circumferential direction of the raceway. The width of this collimator is
one quarter of the diameter of collimator used in the present
investigation. The 3.0 mm length of the slit permits a tolerable intensity
of diffracted beam, but tests with this collimator have not yet been
completed.
There exists an inverse correlation between the residual stress
distribution at the surface and the contact stress distribution (see
Figures 6 and 7). On the area of the raceway where the contact stress was
most compressive, the residual stress became less compressive. The residual
stress became more compressive in the area of the raceway surface where the
contact stress was least compressive. It is suggested that this relation
is a result of accommodation between tension and compression at the
surface.
The correlation between the contact stress distributions and the peak
breadth distributions is not consistent. In the case of Group A cones, the
correlation is good only below the surface. In the case of Group C cones,
the correlation is good only at the surface. This inconsistency is
intriguing and cannot be explained.
D. L. MILAM 351
ACKNcmLEDGEMENTS
thanks The Timken Company for permission to present the results of this
work.
REFERENCES
1. H. Swahn, P. C. Becker, and O. Vingsboro: Metal. Trans, A, 7A
(1976) 1099-1110.
2. O. Zwirlein and H. Schlicht: Rolling Contact Fatigue Testing of
Bearing Steels, ASTM 771, J. J. C. Hoo, Ed., Amer. Soc. for
Testing and Materials, 1982, pp 358-379.
3. J. J. Bush, W. L. Grube, and G. A. Robinson: Trans. ASM 54 (1961)
390-412.
4. A. P. Voskamp, R. Osterlund, P. C. Becker, and O. Vingsboro:
Metals Technology 7 (1980) 14-21.
5. H. Muro and N. Tsushima: Wear 15 (1970) 309-330.
6. A. J. Gentile, E. F. Jordan, and A. D. Martin: Trans. AlME 223
(1965) 1085-1093.
7. H. Muro, N. Tsushima, and K. Nunome: Wear 25 (1973) 345-356,
8. H. Schlicht, E. Schreiber, and O. Zwirlein, Ball and Roller
Bearing Engineering, 1987-1, FAG, 14-22.
9. R. H. Marion and J. B. Cohen: Advances in X-Ray Analysis 18
(1975) 466-501.
RESIDUAL STRESS DISTRIBUTION OF CERAMIC-METAL JOINT
Akira Saito
Hi tachi, Ltd.
Chiyoda-ku, Tokyo, 101 Japan
ABSTRACT
INTRODUCTION
residual stress are not well defined and cannot ~e taken into account in
calculating the stress. Therefore, it is greatly desired in engineering and
industry to measure the residual stresses in the joints in order to develop
techniques of bonding that do not introduce large residual stress.
TEST PROCEDURES
Specimens
Carbon steel
II
Irradiated area
18 3
0.2 0.2
18
Thickness 3
(mm)
Fig. 1. Specimen.
M. KURITA ET AL. (1) 355
The residual stress on the surface of the silicon nitride was measured
by x-ray diffraction with copper Ka radiation, which may have eliminated the
influence of the work hardened surface layer by grinding and lapping because
of the deep penetration of the CuKa radiation. The (323) plane of silicon
nitride, having a diffraction angle of 141.5 0 , was measured by using an
automated x-ray diffractometer developed in our laboratory. 9 The effective
penetration depth that diffracts 90% of the total diffracted x-rays is 55 to
80 ~m for the (323) plane of silicon nitride with copper Ka radiation. The
conditions of x-ray stress measurement are given in Table 1. Both residual
stresses, ax and 0y' in the directions of x and y shown in Fig. 1 were
measured. The x-ray irradiated area was 1 x 3 mm. To measure both stresses
in the same irradiated area, the stresses ax and a were measured by the
iso- and side-inclination methods, respectively. the residual stress of the
silicon nitride adjacent to the bonding interface was measured by delimiting
the x-rays so that the irradiated area adjoined the interface. In both
methods, peak positions were determined by the Gaussian curve method using
the fixed lji method (the lji angle is fixed in determining a peak position);
the stress was determined by the sin 2 lji method where lji angles between the
specimen and diffraction plane normals were chosen as 00 , 24 0 , 35 0 , and 45 0 .
To reduce the measurement time, the background subtraction was omitted in
determining a peak position because it does not affect the stress value in
the Gaussian curve method. 6 , 8 A peak position was determined from x-ray
counts at eight to twelve data points. In calculating the stress, the
stress constant -779 MPa/deg obtained in a previous paperlO was used.
LD
~
zt
~~SbN=4~:::=jl~-;
:OffillllffilllllllR=-=\=1=1:f=1
Steel 1:15,. 02 Cu 20 ,I
Fig. 2. Eight nodal solid elements of Specimen W30 used in FEM.
stress concentration may occur there. The numbers of the elements and nodal
points are 357 and 576, respectively, for Specimen WlO, and 612 and 936 for
Specimen W30. The residual stress was calculated which was generated in
cooling the joints from 923 K, at which the brazing alloy begins to sustain
stress, to 293 K with a decrement of 25 K. In the calculation of the
stress, the silicon nitride was treated as a linear elastic body, and the
steel and copper were treated as elastoplastic bodies, the temperature
dependence of Young's modulus and yield strength being taken into account.
The two-dimensional thermoelastoplastic stress analysis by FEM was also
carried out in order to compare with the three-dimensional analysis.
300r---~--~=-------~
Ground Si3N4
8: 200
2
(/)
(/)
<1>
L-
Ui
~ -100
'0
~ -200
a: Measured
3-D FEM
0
•
-300 ~2--D~F~EM~----~--~1
-15 -10 -5 0 5 10 15
Distance y • mm
300
Lapped SiJN4I
~ 200
~
100 ~£~2
V
:J
Ul
Ul i ~~
~ 0
U;
~ -100
!(.- --!/J'~h'<!<
I !----
u
&-200 Measured
cr,
a
cry
1 •
13-0 FEM 1-----
-300
I
-15 -10 -5 0 5 10 15
Distance y, mm
Fig. 4. Residual stress distributions on the lapped surface of
the silicon nitride along the ceramic-metal interface of
Specimen W30.
On the other hand, all the stresses a y are compressive and are lower
at the center and at both ends. The measured stress values approximately
agree with the values calculated from the three-dimensional analysis by FEM.
However, the stress values calculated by using the two-dimensional analysis
by FEM, assuming a plane stress state, do not agree with the measured stress
values. This is because the stress distributes three-dimensionally at the
interface, as will be shown later, and a plane stress state cannot be
assumed even in the specimen of thickness of 3 mm. The stress concentration
appearing in the calculated stress ax at the interface on both sides of the
specimen does not appear in the measurements because the measured stress
value was an average for the irradiated area of 1 x 3 mm.
x
0 , ~
0 120 80 40
(a) Distribution of stress ax
Steel Cu Si3N4 Stress O'y. MPa
Yt
15 O~
40-
80 -
120
E 160-
E 200-
10 240-
» 0
<!J 280
u
e 300-
<1l
Ul
i:5 5 320-
x
0 0
~
I I I I I I I I I I I I I
0 5 10 15 20
Distance x. mm
(b) Distribution of stress a y
the interface at the center, (y=o) , and edge of the specimen. At the
distance of about 10 mm in the y direction from the center, the sign/ of the
stress Ox changes and the stress 0y takes the maximum compressive value, as
shown in Fig. 5(a) and (b).
Figure 6 shows the residual stress distribution along the center line
C (x-axis) on the surface of Specimen W'30. The calculated residual stress
Ox on the surface of the silicon nitride reaches a maximum value at a short
distance from the interface, and decreases rapidly with increasing distance
from the interface. The stress 0y takes small compressive value and varies
gently along the x-axis. The residual stresses Ox and 0y calculated from
the three-dimensional FEM, shown by the two lines, approximately agree with
the measured values. However, some measured Ox values do not agree well
with the calculated stress values. This is considered to be attributed to
M. KURITA ET AL. (1) 359
100 \QQQOQQ
III
III
~ 0 ~--"
iii !_U_!_!~"!!---·r --,
'iii
:l
-100
'U _f-- Steel
-200
~ Cu~ -Cu
-300 :-_~_~_--=--_~-"---:'__.l-_-'::-_-:'.
-20 -15 -10 -5 0 5 10 15 20
Distance x, mm
Hg.6. Residual stress distributions along the center line of
Specilen W30.
the effect of the work-hardened surface layer, the size of the irradiated
area in the measurement by x-rays, the uncertainty of the elastic constants
and yield stress at high temperature, and errors accumulated in each
temperature decrement in calculating the residual stress by FEM.
E
E
- zt Stress O-x. MPa
N
2l 0 .-49 120 80 4.0 20 10 0
~ ~t?t22\ I I
QI
u
c 11-1
'"
iii
(5 Steel Cu Si3N4
(a) Distribution of stress ax
Stresso-y. MPa
.:.- -'-
E
E Steel Cu Si3N4
N
2 '~~60 I
1 l 360 - =~
-40 -20 -10 0
I
QI
u
c 1 1 12
'(5" o
iii
0
Z.OO -80 -60
'
I I I I I I I I I I I I I I
0 5 10 15 20
Distance x, mm
(b) Distribution of stress a y
.
300
Lapped SbN4
cr. cr
200 Measured
3-D FEM
0
Ul Q
Ul
100 5)
...
<l1 Q Q
iii
-4 -2 0 2 4
Distance y , mm
Fig. 7. Contour I ines of the residual stresses in the cross
section through the center line of Spesimen W30.
300
Lapped Si3N4
~ 200
~ Measured 0 •
/
3=-DXEM - -
Ul 100
Ul
~
U; 0
"iii
.g -100
Ul
8: -200 Steel
Cu
-300
0 5 10 15 20
Distance x, mm
Fig. 9. Residual stress distributions along the center line of
Specimen WlO.
:>.
5C~1
-80
-40
<l1
u -20 0
C
rt! X
U;
i:5
o 0
40 20
-7
5
Si3Nt.
E
E
:>.
<l1
u
c
f 40-
80-
100
rt! X
U; 0 -;>-
i:5
I I I I I I I I I I I I I I I I I I I I
0 5 10 15 20
Distance x, mm
(b) Distribution of stress a y
Fig. 10. Contour lines of the residual stresses on the surface of
$pecilen WlO.
M. KURITA ET AL. (1) 361
E
E t Stress O'x. MPa
2l
Z
N
o -40\ 40 0
1-1
I /
(l)
~ 40~~
u
C
<U \
iii
o Steel Cu Si3N4
(a) Distribution of stress a x
Figures 10 and 11 show the contour lines of the stress on the surface
and in the cross section through the center line C for Specimen WIO,
respectively. The stress values in Figs. 10 and 11 are slightly smaller than
those for Specimen W30 that has a larger width (as seen in Figs. 5 and 7),
and the stress concentrates in a narrower region near the interface, showing
the effect of specimen width on the residual stress distribution.
CONCLUSIONS
ACKNOWLEDGMENTS
REFERENCES
ABSTRACT
The 95% confidence limits of the x-ray elastic and stress constants
of a -a I um i na were determi ned from seven kinds of diffract ion planes by
the Gaussian curve method in order to investigate the diffraction plane
dependence of the constants. No difference in the elastic constants larger
than their 95% confidence intervals was observed for most diffraction
planes. Also, the measured elastic constants for most planes were closer
to the values calculated from the Voigt model than those from the Reuss
model. Since the diffraction line of the (410) plane measured with cobalt Ka
radiation by using an automated x-ray stress analyzer locates at the highest
diffraction angle of 168.4·, the use of this plane will allow the most
accurate stress measurement. Also, the measured x-ray elastic constants
for the (410) plane almost agreed with both values calculated from the Voigt
and Reuss models. Therefore, the (410) plane is the most appropriate plane
for x-ray stress measurement of alumina.
INTRODUCTION
X-ray diffraction technique can be considerd to be the most appropriate
method for measuring residual stress in ceramics because it permits a
rapid and nondestructive measurement of residual stress in a small area
on the surface of polycrystalline materials. In this method, the stress
value is calculated selectively from strains of a particular diffraction
plane in the grains which are favorably oriented for the diffraction; we
must first choose a diffraction plane to be measured. Then, we need to
determine the x-ray stress constant experimentally in order to measure the
stress accurately.
This method is based originally on the theory of elasticity for mac-
roscopically elastically-isotropic polycrystalline materials1. 2. In genera!,
however, the elastic constants of a single crystal depend on the plane of
Advances in X-Ray Analysis. Vol. 33 363
Edited by C.S. Barrett el 01.
Plenum Press, New York, 1990
364 v. X-RAY STRESS ANALYSIS
where, E and 11 are Young's modulus and Poisson's ratio for isotropic
materials.
Transforming the strain e in Eq. (1) to the peak position p (in 28)
of the diffraction line by using the Bragg law, is given approximately by
p = -,8s2('l"+'l"~)sin2¢-2,8s1('l"+'l"~+'l"~)+280 (3)
where 8 0 is a Bragg angle for a stress free specimen, and ,8 is a constant
given by
,8 = 180tan8 0 (4)
7l
From Eq.(3), we get the slope M of the sin2¢ diagram, a plot of the peak
position p versus sin2¢ , as
M= a a,P2¢
sin
= -,8s2('l"+'l"~) = ('l"+'l"~)/K (5)
B-
_---a-r
a M -_ -,8 S2 _ 1
- -K- (8)
Equation (8) shows that the compliance S2 and the stress constant K can be
determined experimentally from the slope B of the linear relationship
between the slope M of the sin2¢ diagram and the applied stress 'l".
Also, the intercept N of the sin 2 ¢ diagram can be obtained by sub-
stituting zero for sin2¢ in Eq.(3) as
*
N= -2,8S1('l"+'l"~+t"~) + 28 0 (9)
From Eq.(9), we can get
C= = -2,8S1 (10)
TEST PROCEDURES
Specimen
JS21 0
~_ _ _ ~
"'<:0--_ _--=::.:55'-_ _ _;;oIJ
(mm)
Fig.!. Specimen.
Table 1 Conditions for measuring stress constant and x-ray elastic constants.
25
c
...
'jij
20
-
Vi
'c
~ 15
lii
0.
<D 10
N
.£
4> 5 ( 1.2. 10.)
Cl
c:
ttl
.s: 0
u
100 120 140 160 180
Diffraction angle x. deg
is given by
aK = K2t'i= a1
a P 12
;; 1 1
2 (12)
where, a1 are constants, and a p 1 and n are the standard deviations and
the number of the peak positions used in determining the stress constant
K. From EQs.(7) and (12), we obtain
aKK a: K a: cot8 (13)
Equations (11) and (13) show that it is important to choose a diffraction
line having an diffraction angle as high as possible in order to determine
the stress constant and strain precisely.
Figure 2 shows the sensitivity (the change in 28 per unit strain) of
the strain measurement as a function of diffraction angle 28 for the
seven kinds of diffraction planes measured in this study. The (410) plane is
approximately 4.6 times as sensitive as the (1.2.10) plane as shown in Fig.2.
TEST RESULTS
Figures 3 and 4 show the diffraction lines measured with CoKa radi-
ation b~ using the automated x-ray diffractometer and the x-ray stress
4000,----------------------,
AJ,03
III
0. Co Kex radiation
u 3000
>. S?
~
N
;;;
~ 2000 o 0
o ~ a;
c: ~ N
N
>.
?
x
1000
-'"
iii
c 2000
<11
~
C
III
:l
0
u
0;
C N ~
e J: 2
'-
!:::
1000
x
I
-g
0
~0
u 0
150 160 170 166 167 168 169 170
Diffraction angle x, deg Diffraction angle x, de<]
Fig. 4. Diffraction lines measured with Fig. 5. Approximation of cobalt Kat
cobalt Ka radiation by using diffraction peak of (410) plane
an automated x-ray stress by a Gaussian function.
analyzer.
g'
"0
0.2
°)
AI 20 3 ( 4 1 plane l 0.6 A12~ (1.2.10) plane
CoKa radiation 0.4 CoKa radiation
~ 0.0
E
~ -0.2
~
'5 -0.4
<7
c:
'iii
-0.6
'0 -0.8 '0 -0.4
CI> CI>
a. a.
0
iii -1.0 ~ -0.6 '-'-_L---'-_L---'-_L---I
0 50 100 150 200 250 300 o 50 100 150 200 250 300
Appl ied stress t. MPa Applied stress t, MPa
01 01
CI> CI>
"0 AI 20 3 ( 41a)plane "0 A12~ ( 1. 2.10 ) plane
168.65 CoKa. radiation z· 127.45 CoKa radiation
Z
E E
~ 168.60 t!! 127.40
~ ~
'5
~
'5
168.55 03- 127.35
03- N
N
c: c:
'iii 'iii
168.50 '0 127.30
'0 195% confidence
a. I 95% confidence
interval
a. interval
CI>
~
168.45 ...CI>fl 127.25
.s
CI>
0 50 100 150 200 250 300 .s 0 50 100 150 200 250300
Applied stress t ,MPa Applied stress t, MPa
Fig. 10. Linear dependence of intercept Fig. 11. Linear dependence of intercept
of sin 2¢ diagram on appl ied of sin 2 ¢ diagram on appl ied
stress for (410) plane. stress for (1.2.10) plane.
370 V. X-RAY STRESS ANALYSIS
(lOTPa)-l
I
Compliances, s2/2
s,
28.34±2.85 32.54±4.05 30.05± 1.88 28.50±1.71 27.43±1.21 28.36±0.73
4.68±0.88 5.63±1.09 5.31 ±O.64 5.19±O.57 5.02±0.43 5.52±O.28
27.98± 1.93
5.47±0.65
Young's modulus ~, 423±39 372±46 404±24 429± 24 446 ± 18 438± 10 444 ±30
(GPa
Poisson's ratio v 0.20±O.03 0.21 ±O.03 0.22±O.02 0.22±0.02 0.22±0.02 0.24 ±0.01 0.243± 0.02
Stress constnt K, -872± 88 -1325± 165 -879± 55 -704±42 -548±24 -314±8 -977± 67
(MPa deg)
VI
<lJ 4
u
c ---(121)
(221)
.~
a. (321)
(311)
E (hOI)
u
0 2
0
0 2 4 6 8
Compliance SI, (10TPa)-1
4 ~----------------------~
I----()--I ~:;~~~e Y~tl~;Va1Wi th 95 '/0
Reuss model
N
......
Jl 2
-(121)
<lJ (221)
U
c --(321)
.!:!1 (311)
(hOI)
~1
o
u
2 3 4
Compliance 52/2. (TPa)-1
Fig. 13. Diffraction plane dependence
of compliance s2/2.
CONCLUSIONS
(1) The 95% confidence limits of the x-ray elastic and stress constants
were determined by the Gaussian curve method from seven kinds of diffraction
planes of alumina in order to investigate the diffraction plane dependence
of the x-ray elastic constants.
372 v. X-RAY STRESS ANALYSIS
(2) No difference larger than the 95% confidence intervals was observed
in x-ray elastic constants for most diffraction planes.
(3) The measured x-ray elastic constants for the (1.2.10), (220) and
(410) planes were close to both values calculated from the Reuss and Voigt
models. The measured elastic constants of the other planes were closer to
the values calculated from the Voigt model than those from the Reuss model.
(4) The relati ve error O'K/K of the stress constant K is proport ional
to cot8. Therefore, it is important to choose a diffraction plane having
a diffraction angle 28 as high as possible in order to measure the stress
constant precisely.
(5) Since the diffraction line of the (410) plane measured with
cobalt Ka radiation by using an automated x-ray stress analyzer locates
at the highest angle of 168.4', the use of this plane would allow the most
precise measurement of the stress and stress constant. Also, the measured
x-raY elastic constants for this plane almost agreed with both values
calculated from the Reuss and Voigt models. Therefore, the (410) plane is
the most appropriate plane for the x-ray stress measurement of alumina.
(6) The peak position of the (410) plane can not be measured by the
half-width method because the diffraction line of this plane locates at so
high angle that its background needed in this method can not be determined.
However,the Gaussian curve method can determine the peak position because
it does not need the background subtraction.
The authors would like to thank NGK Spark Plug Co. ,Japan, for preparing
the specimen.
REFERENCES
ABSTRACT
INTRODUCTION
where 82 and 80 are the positions of the high angle limit and the peak
maxima, respectively. From this equation, the "span" of a diffraction
profile in lid space depends upon its a3' value.
1.14 ...------,-----,---..,.----r---,
..
CONSTANT a ~ = 25 'A
1.02 _._._.--~,.-.-- ••• -
o
o
0
0.90
~
N
;:z
..... 0.78
~
III
0
0 0.66 0
~ 0
calculation, all eight profiles in these two data files were generated
with a symmetric Gaussian type function:
jC4ln2) X 2
I(x) • exp[-4ln2·(--) 1 (2)
J'/T·fJ fJ
where fJ is the full width at half maximum (FWHM) of the diffraction line.
It is clear that the primary step of the data generation for a Gaussian
profile is to determine the FWHM of the peak. Once the FWHM is obtained,
the diffraction data may be easily calculated at any desired diffraction
angle. Note that no background function was used. Thus, the generated
profiles extend to zero intensity on each side. The reference or
instrument profiles, get), were generated assuming that the profiles are
free of size and strain broadening effects. Therefore, the simulated
profiles of this hypothetical Si powder should contain only the broadening
due to instrumental effects. Langford et al. 7 have shown that for such
profiles a simple quadratic relationship may be used to determine the
FWHM as a function of 28: fJ2 = u(28)2 + v(8) + w. In that paper, the
u, v, w values were given as 0.1538, -0.0941, and 0.0977, respectively.
In our reference-profile generation, the quadratic relationship and these
coefficient values were used to determine the FWHM values at various
diffraction angles of the eight instrumental lines.
(3)
where DG and EG are the crystallite size and strain values for the
powdered material. Thus, for a known size and strain value, the FWHM of
the diffraction line at any given diffraction angle 20 may be calculated.
This relationship is adequate for our computer modeling study but is not
universally applicable for many real materials.
In the generation of our broadened data, it was assumed that the size
of the Si crystallite size is 100 A and the strain is 0.001. The FWHMs of
the profiles f(t) at each of the different line positions were then
calculated using equation (3). For each of the convoluted broadened
profiles, h(t), the FWHM was obtained by the summation of the squares of
the FWHM values of the corresponding reference profile (using Langford's
parameters; see above), get), and the pure profile, f(t), at the same
diffraction angle.
For both the reference and broadened data, the Cu-Kal wavelength
1.5405981 A was used. The simulated diffraction data contain contri-
but ions due only to Ql radiation and were calculated at a step interval of
0.005° 20. As a result, the simulated profiles are free of background,
statistical noise, and contributions from noncharacteristic radiation.
The profile fitting was carried out with the recently developed
computer program, XRAYL.10 This computer program is particularly useful
as a preprocessor program in profile analysis as it can be fitted by
least-squares refinement of a variety of profile models to the input
data, and then a smoothed profile based on the refined model parameters
and a user-specified a3' value can be calculated.
Our two generated Si data files were fitted by XRAYL using both
Gaussian and Pearson VII profile models. In each pair of refinements, a
different a3' value was used. Two types of profile widths were used: (1)
a constant a3' was used for all the peaks; (2) a3' values were randomly
chosen for each of the peaks. In the profile -fi tting procedure, each
generated Gaussian profile was treated as "raw" data and fitted by XRAYL.
Since the "raw" data were ideally generated, perfect fits were obtained in
the profile fitting: most fitting errors are zero, and the worst is 0.001%
due to computational rounding error.
Each pair of the XRAYL calculated profiles were then used to compute
the size and strain by the W-A and the G-S "plotting" methods. The
calculations for the size and strain were carried out using the computer
program, CRYSIZ. ll As customary with the W-A analysis, individual
crystallographic directions for which higher orders were available were
examined (the [111] and [100] directions in our case). In addition,
following a procedure we have used in the past, the entire set of lines
was also examined as a single group. This is accomplished by employing
Y. ZHANG ET AL. 377
1/d2 rather than the conventional order of the index. This results in
average values of size and strain parameters (we refer to these values as
"Overall Average Values" hereafter). CRYSIZ determines a background
intensity correction from the reference profile and subtracts it from both
the reference and broadened profiles. Since the generated reference
profile always reached zero intensity, no error in background level was
introduced and no "hook" effects appeared, as are common with
overestimated background corrections.
The assumption that the diffraction lines are Gaussian does not
affect the generality of the subsequent profile analysis. This is because
the Til-A method does not assume any specific functional form for the
profiles being analyzed. The G-S plotting method employs exactly the same
relationship as that assumed in the original data generation, equation
(3) .
Figure 1 shows two G-S curves for size and strain separation for our
simulated diffraction data using 100 A and 0.001 for size and strain,
respectively. The black dots (upper line) are for the data set that were
refined by using a constant a3' of 25 A, for all the profiles; while open
squares (lower curve) are for the data set where a3' values were randomly
.chosen (20-50 A). The peak width measurement, ~, used in these two curves
was the second moment calculated from the profile f(t) after Stokes cor-
rection (back Fourier back-transformed profiles). The curves in Figure I
show that random a3' values result in a scattering of points on the G-S
line and the slope and intercept of the least-squares line through these
points do not correspond to the "known" size and strain values. On the
other hand, a small constant a3' value gives a straight line that is
consistent with the predication of profile analysis theory and the "known"
values.
130
120
0«
ci
UJ
I- 110 GAUSSIAN-SQUARED
UJ PLOTTING METHOD
:::i:
<{
a: 100
~
UJ
N
iii 90
80~----~----~----~--~~--~
1.07
1.01
0.95
"0
~ 0.89
z
«a:: 0.83
I-
(f)
0.77
0.71
0.65
20.0 40.0 60.0
aj VALUES. (Al
Fig. 3. Strain versus a3'. Simulated Gaussian diffraction profiles
calculated on the basis of 100 A size and 0.001 strain values
plus a typical instrumental contribution are employed to back
calculate strain values for particular a3' values. Open
squares: the results of the G-S method using second moment as
the measure of peak widths; Solid circles: the results of W-A
method in the [100] direction.
Y. ZHANG ET AL. 379
We have also examined the effect of altering the values of a3' within
a data set on W-A results. When a constant a3' value of 25 A was used for
all three calculations (two individual directions and the overall
average), the Fourier size-coefficient and root-mean-squared strain curves
are smooth and well-behaved (not shown). In the case of random a3' values
(20- 50 A), however, both the Fourier size -coefficient curves and the
strain curves versus domain length were severely scattered and the
corresponding size and strain estimates were incorrect. This arises
principally from the use of inconsistent a3' values which affect the
Fourier transformation process and later the Stokes correction. These
results also suggest that size and strain values from W-A method are
reliable only when a3' is small and equal for both of the paired peaks for
a crystallographic direction.
Figures 2 and 3 present results obtained by the usual W-A and G-S
methods and show the crystallite -size and strain results versus a3',
respectively, using our simulated diffraction profiles. Two curves are
shown in each figure. One curve is the result of the G-S plotting method
employing all eight lines, and the other curve is the W-A calculations in
the [100] direction. While the a3' values used for the data sets in size
and strain calculations varied from 20 to 50 A, the a3' used for all the
peaks within each data set was the same value. Therefore, the variation
of size and strain values are primarily due to the difference in a3'
values. From Equation 1 the width of the data collection region depends
inversely on the a3' value. The curves in Figures 2 and 3 show that the
use of different spans for data collection affects the crystallite- size
and strain results and this is particularly significant when a3' is
large. For example, using a3' values of 55 and 40 A, the calculated size
values using the G-S method are 127 and 106.5 A, respectively (similar
large differences are noted for the W-A method). However, as a3' becomes
smaller than a certain value, a further decrease of a3' (increase of span)
will not further alter the size and strain results (the flat parts of the
two curves). These results (together with other model calculations with
various size and strain values) suggest that for both the W-A and the G-S
plotting methods, consistent and reliable size and strain values will not
be achieved unless the a3' values of the diffraction lines are small.
Further, for our simulated diffraction data based on a 100 A size and
O. 001 strain, only when a3' values are smaller than 30 A will the G-S
calculation give back the original assumed size and strain values (the
effect of a3' values with respect to the degree of broadening will be
discussed below). In the present example, when the a3' values were not
appropriately chosen, an overestimation up to 27% in the size value and an
underestimation up to 33% in the strain value were observed. Moreover, it
should be pointed out that in Figures 2 and 3 that the back-calculated
size and strain results by the W-A method are smaller than the assumed
values. This is not surprising because the "raw" data were generated
using the G-S relationship and the physical significance of the size and
strain deduced from the W-A and G-S approaches is different. The W-A
method gives "volume-weighted" size and strain while the G-S plotting
method gives "area-weighted" size and strain. 12
How small an a3' value is "proper" for a given sample depends upon
the degree of broadening. What is well known, and confirmed in this
study, is that the profiles must reach background on both sides or Fourier
380 VI. DETERMINATION OF CRYSTALLITE SIZE AND STRAIN
ACKNOWLEDGMENTS
REFERENCES
Frank E. Briden
INTRODUCTION
The Air and Energy Engineering Research Laboratory (AEERL) has been investigating the injection
of dry calcium hydroxide into combustion streams for the removal of sulfur oxides. It has been known, for
some time, that the reactivity of Ca(OH)2 with sulfur oxides varies over a wide range for different batches of
calcium hydroxide. The reactivity is a critical factor in the economics of the process. Because of this it is
desirable to have a method to predict the reactivity of calcium hydroxide batches. The reactivity can be
affected by many factors including the origin of the calcium carbonate mineral from which the hydroxide is
produced, the production conditions, and the concentration of additives used for reactivity enhancement. It
has been theorized that all of the preceding contribute to such physical characteristics as surface area, particle
size, and porosity. The examination of these physical characteristics has provided the basis for
understanding the reaction processes but has not provided all the factors necessary to correlate intrinsic
sorbent characteristics with reactivity.
The application of microstructure factors derived from X-ray Diffraction Line Profile (XLP) for the
characterization and reactivity prediction of calcium-hydroxide sorbent materials was tirst reported by Briden
and Natschke (1). In that study, the XLP derived microstructure factors from several calcium-hydroxide
factors were found to be related to their reactivities. In the study, the microstructure factor mean column
length eCL) was seen to be inversely proportional to reactivity while the strain at the mean CL was seen to be
directly proportional to reactivity. When the method was applied to more widely varying batches, the
relation was not as consistent. The original data had been generated, using bulk sample holders and zincite
as the instrument line-broadening standard. At this point, it was decided to investigate the effects of the
choice of standards, the use of thick samples versus thin, and the simultaneous use of more than one
microstructure factor in reactivity prediction.
EXPERIMENTAL
The experimental data were taken on a Siemens D-500 diffractometer. The x-ray source was a
copper-target tube running 50 kV and 40 rnA. The tube had a fine focus with a beam size of 0.4 x 12 mm
and a take-off angle of 60 resulting in an effective beam width of 0.040 mm. The collimation was a 10
divergence slit, a 2.4 0 incident-beam Soller slit, a 10 scatter slit, and after the sample another 10 scatter slit, a
20 Soller slit, a 0.05 0 receiving slit, a 0.15 0 monochrometer slit, and a graphite monochrometer. A
scintillation detector was used.
The bulk sample holders were end loaded and about 2 mm deep. The thin sample holders were zero
background quartz plates. The thin-mount samples were slurried in amyl acetate and then the slurry was
spread over the plate and allowed to evaporate. The samples were not ground in order to preserve the, as
received, crystallite size and strain conditions. In general, the samples would pass a 44 /lm mesh sieve.
The data were processed on the system DEC 11/73, with an RSXIIM operating system using the
Siemens DIFFRAC 500, version 1.1 software. For the Warren-Averbach method (Z) used to separate
crystallite size and strain effects, it was necessary to use diffraction peaks of at least two orders of the same
reflection. For Ca(OH)z (JCPDS No. 4-733), the only pair of reflections readily available are the 101 and
ZOZ because of low intensities and interferences. The standards used to compensate for instrument line shape
effects were LaB6 (SRM 660), Si (SRM 640B), CaFZ (Spex TMI-IO, Lot 01841 P), a-AIZ03 (Norton 871),
and ZnO (Fisher Reagent, Lot 80186).
In order to use the Siemens computer program CRYSIZE to determine microstructure factors with
the Warren-Averbach method, it is first necessary to fit the diffraction peak with one of several functions
available in the program FIT. This process improves the Warren-Averbach analysis by the reduction of
random noise and the elimination of interference from foreign peaks. In all cases, the Pearson 7 or Split
Pearson 7 functions were used. Six microstructure factors from the CRYSIZE output were examined. The
first three were the mean CL (the mean distance across the crystallites in the 101 direction), the mode CL
(CL occurring with the greatest frequency in the 101 direction), and the maximum calculated CL (the highest
CL determined at the iteration limit of the CRYSIZE algorithm in the 101 direction). Also the root mean
square strains (RMSS) for the three respective CLs were examined. The six microstructure factors were
determined for six commercial Ca(OH)z sorbent products [Linwood, Marblehead, Kemidol, Snowflake,
Modified Marblehead, and Illinois State Geological Survey (ISGS) Mixture] plus a more crystalline, reagent
grade material from Alfa Chemical Company (Lot 8171481). The six microstructure factors and the full
width at half maximum (FWHM) for the 101 and ZOZ peaks for the products are listed in Table I. The
products are listed from left to right in order of increasing reactivity [weight of Ca(0H)2 converted to
CaS04 divided by the total weight of Ca(0H)2 used times 100]. The reactivities were obtained on the EPA
innovative furnace reactor with a flow rate sufficient to give a residence time of 0.75 s between 538 and
4Z7 °C with an SOZ concentration of 3,000 ppm (3). In general, it can be seen that the CL factors decrease
with increasing reactivity while the root mean square strains increase with increasing reactivity.
It had been anticipated that the microstructure factors would be very dependent on the mass
absorption coefficient of the standard materials. When the mean CLs for the different Ca(OH)2 products
were plotted versus the FWHM of the second order standard peaks, a fairly good progression of increasing
mean CL with increasing FWHM was observed (Figure 1). However, it was also observed that the CLs were
not falling in order of the standard mass absorption coefficients. It appears that the mass absorption
coefficient effects are outweighed by whatever other factors influenced the FWHM. The lowest mean CLs
were obtained with CaFZ. However, it is noted that, even though the absolute values are different, the
relative rankings for a given standard are the same except in the case of Kemidol and Snowflake which are
extremely close. Of course, these Ca(OH)z samples all had very broad peaks and the difference between
them and the standards was very large.
The RMSS values at the mean CLs were also plotted versus the FWHM of the second order
standard peaks (Figure Z). The data point for the Kemidol sample with AlZ03 standard is obviously in error.
In general the RMSS is seen to decrease with increasing FWHM of the second order standard peaks. The
CLs obtained with Si are much lower than predicted. This, quite likely, indicates that the Si has a relatively
high RMSS. Some of the strains are slightly higher for CaFZ while some are slightly higher for LaB6
indicating that, within experimental error, their strains are small but equivalent. As with the CLs, the
absolute strain values vary with the FWHM but the relative order for the samples is constant.
The previous work (1) on relating microstructure factors to reactivity was limited to the
consideration of one factor at a time. As more widely varying sorbent prodUCl~ have been investigated, it has
become quite obvious that fairly wide deviations from the regression may occur. Because of this, it was
attempted to perform the regression with more than one microstructure factor. Of course, the more factors
considered the better the agreement that may be obtained. However, because of the limited size of data sets
that become available for testing, it was considered that no more than two factors at a time should be used.
F. E. BRIDEN 385
l .. M. ·
25
E 1 1 1 I
oS
.£ 20 ........................... :......... ·t.. · .. ·· ...... ·.... ·· .. ··
Cl
c
......... .................. ~l.. .. .............. ..................~
Q)
~
15
· rl
c
· · · · · · · r=~ · · · ·· ·
E
::;)
(5
U 10
c
!.... !........... .
cu 1 I 1 1 1
Q)
:E 5 ·.... ·...... ·.......... ·.... !" ............ ·t.. ................. .................... 1" ......
CaF2 LaBS AJ203 Si ZnO
0
0.05 0.06 0.07 0.08 0.09 0.1 0.11 0.12
FWHM Second Order Sid Peak (Oeg)
Figure 1. Mean CLs for Seven Ca(OH)2 Products Calculated With Five Standard Materials Versus the
FWHM of the Second Order Standard Peaks.
30
J}~~'~'.
25
<f ...
UJ 20
x
c: 15
.~
U5 I
(/)
~
a:
10 .... 1
I ~
II I
5 I I I
CaF2 LaB6 AI203 Si ZnO
0
0 .05 0.06 0.07 0.08 0.09 0.1 0 .1 1 0.12
FWHM Second Order SId Peak (Oeg)
Figure 2. RMS Strain at the Mean CL for Seven Ca(OH)2 Products Calculated with Five Standard Materials
Versus the FWHM of the Second Order Standard Peaks.
386 VI. DETERMINATION OF CRYSTALLITE SIZE AND STRAIN
Tables 2 through 4 show the linear correlation coefficients for the six factors regressed, one and two at a
time, over the seven products, for the standards CaF2, LaB6, and Al203. In these tables, the coefficients on
the diagonals are for the single factors alone. The balance of the coefficients are the result of regressions of
the factors taken two at a time. For all of these standards, the mean CL and the maximum CL are the best
single predictors of reactivity. Of the strain factors, the strain at the average CL is the best predictor. It is
not fully accurate to compare the magnitude of these strain coefficients to the size coefficients because it has
been seen that the functions relating strain to reactivity are not linear and these coefficients came from linear
regressions of size and strain. Preliminary calculations have shown that, when strain regression is treated as
a second degree polynomial, the coefficients will improve. It would be logical to assume that any scheme to
reliably predict reactivity should be based on both size and strain components. The mean CL and the strain
at the mean CL are consistently high for all three standards.
The original work had started with bulk sample holders. It was desired to compare the results from
the bulk sample holders with those from the thin mounts on zero background quartz plates. Table 5 presents
the mean CL factors determined for CaF2 and LaB6 standard with bulk and thin samples. The absolute
values for LaB6 for bulk and thin samples are, in general, larger than those for CaF2. The order for both thin
samples is almost identical and likewise the order for both bulk samples is almost identical. However, the
order for the bulk samples is quite different from the order for the thin samples. Table 6 gives the correlation
coefficients for the regression of the microstructure factors for CaF2 standard calculations for the thin
samples. It is seen here that the single factor correlations are significantly lower than those for the bulk
samples (Table 2). The mean CL correlation is only 0.61 for thin samples where it was 0.77 for the bulk
samples. Because of the low correlations for the single factors (those on the diagonals) it was surprising to
find that several of the regression pair correlations were significantly higher than those for the bulk samples.
Of particular interest is the pair of factors maximum CL and RMSS at maximum CL. The first had a single
factor correlation coefficient of 0.69 and the other 0.03 but when regressed together they had the highest
correlation of 0.93. A theory that this evidence could support is that the better resolved thin sample peak
allows the separation of the peak profile parameters which contribute to the two factors. In cases where the
separate correlations are low but the simultaneous regression for the pair is high, the two factors could be
synergistic so that the contribution of each factor depends directly or inversely on the contribution of the
other.
CONCLUSIONS
1. When different standards were compared for Ca(OH)2 analysis, the CLs varied in a similar manner
to the standard peak widths. The CaF2, LaB6, Si, and AI203 standards all gave quite comparable
results. The absolute values obtained varied a little but the relative order in the series was hardly
affected.
F. E. BRIDEN 387
Factor ModeCL MeanCL MaxCL RMSS Mode RMSS Mean RMSS Max
TABLE 5. COMPARISON OF CRYSTALLITE MEAN CLs (nm) IN THE 101 DIRECTION OBTAINED
WITH CaF2 AND LaB6 STANDARDS WITH BULK AND THIN FILM MOUNTS
Standard! Mod
Mount Reagent Linwd Marblhd Kemidol Snowflk Marblhd ISGS Mix
2. The present data indicate that crystallite size is inversely proportional to the first order of reactivity.
The root mean square strains appear to be directly proportional to reactivity with a second degree
function. When the microstructure factors are used to predict reacti vity of solids for gases, the
results are improved considerably by regressing two factors at a time. .
3. The use of thin samples gave a different order of mean CLs than those obtained for a series of bulk
samples. The single factor correlation coefficients were lower for thin samples than bulk samples.
However, when the thin sample results were regressed two at a time, higher correlations were
achieved for some pairs than for those obtained for bulk samples. It is hypothesized that the thin
samples gave better resolved peaks allowing the separation of factors which were synergistically
related.
The assistance of David Natschke and Shawn Karns of Acurex Corporation, in collection and
reduction of data, is deeply appreciated.
REFERENCES
1. Briden, F.E. and Natschke, D.F., The Characterization of a Solid Sorbent With Crystallite Size and
Strain Data From X-ray Diffraction Line Broadening, Advances in X-ray Analysis, 32,
437-441,Plenum Press, NY, (1989).
2. Warren, B.E., "X-ray Diffraction," Addison Wesley Co., (1969), Chapter 13.
3. Bruce, K.R., "Furnace Sorbent Reactivity Testing for Control of S02 Emissions From Illinois
Coals." Acurex Corporation, Durham, NC, Personal Communication, (1989).
THE USE OF 2-D DETECTOR UTILIZING LASER-STIMULATED LUMINESCENCE FOR
INTRODUCTION
PRINCIPLE OF DETECTOR3
Imaging Plate
Photostimurable
phosphor crystals
Support
~!l)j E'p",",'
L_a_:~~_~~~",-_~_~~~~~ng ~
Read out
Erase
The IP has the following properties which are suitable for obtaining
high-quality images with low x-ray doses:
Therefore, the IP can be used in almost the same way as an x-ray film
except for the repeated use, and the properties except for the spatial
resolution are superior to those of x-ray film.
In this study, the digital data in the magnetic tape are again sent to
a magneto-optical disk (594MB) connected to a personal computer and the
image processing, such as subtraction of background noise and binarization
etc., is performed with assistance of an image processor.
PMT
~-<J
-1LOgAMP~
h-c:=J
!
Imaging Plate
a) b)
Experimental
The material used was a 0.16%C carbon steel which has a yield stress
of 267 MPa. ,A fatigue test of an ASTM compact tension-type specimen was
conducted under a maximum load of 700 Kgf and a stress ratio of 0.05.
position relative to the notch root, and a Debye-Scherrer pattern from (211)
crystal planes was recorded in the IP. The distance between the IP and the
specimen was 80 mm and the diameter of radiated area on the specimen surface
was about 0.9 mm. Several patterns were obtained at stated intervals in the
crack propagating direction, and a thin layer of fracture surface was
electropolished away. This process was repeated for measuring the plastic
zone size beneath the fracture surface.
/ ,
.'.
a) f
a)
) "
I
/
a .. _ _ ; _
'/
., " -" ...
.,
•
.N •
,..
b) b)
/
/ "
..,.
.,
- ...... ~ ~
"......... - --:. -
~
/ ' -. ~ ...
c) I
f \ c)
\
\
/
"
"
I
"
..... - . . - ...
E:
SI5C. Pmax=700Kgt. R=0.05
15 • -150 '"
~
Kmax=20MPa/m. 100l'm
~
~
• <..>
"- 10 f--
o Arc length
• Total area
V)
• 2>
~
"
.....
0
0
;: 10 ~ ~
"
""
c:: • 0 0 100
~ o ~
<..> 0
"- f--
'" t\J
")
o
5 0 - 50
I
130
I
140
I
150 o '--'_-;-,3-!,-0:;::-...I--:-,47 0"'-...1--:-,5;"0::' o
Threshold level Threshold level
Fig. 7. Changes in number of arcs, average length of arc and total area
of diffraction ring due to threshold value for binarization.
a) at 20 ~m in depth from fracture surface.
b) at 100 ~m in depth from fracture surface.
arcs which contain many small spots in a DS rin! and can measure the
sub-grain size by counting the number of spots.
a DS ring will be countable with the help of the imaging processing system.
In order to measure the sub-grain size, two or more diffraction patterns
which were taken under different exposure times are necessary.4 When x-ray
film was used for this measurement, however, it was difficult to count the
number of spots in a pattern of low density because of the superposition of
background noise caused by the chemical fog. As the background noise of the
IP is lower than that of the x-ray film because of the wide dynamic range of
the IP, such difficulties with weak patterns are reduced.
CONCLUSION
ACKNOWLEDGEMENT
REFERENCES
INTRODUCTION
In this study, the cause of such convex type background line was
analyzed and we proposed an experimental correction method to linearlize
the background line.
ANALYSIS
I(a)/Imax=cos 2a exp[-~R(1/cosa-1)]
When the x-ray beam is obliquely entering the PSPC, the counting
efficiency is outwardly decreased as the result of the spread of spatial
resolution. Such decrease has to be experimentally measured because the
decrease is dependent on the PSPC used. The PSPC used in our work had an
effective length of 100 mm and the setting length was 200 mm i ; an
angular range of 28° could be covered for measurements. The counting
efficiency was measured for the x-ray source of 50 ~m width, which was
vertically placed at a distance R=200 mm with angle of a from the PSPC.
S. OHYA AND Y. YOSHIOKA 399
..
.
1',
~ 1 0
•.....,
•
"-
..... Fig.2 Distribution of intensity
with oblique incidence of
/' \, x-ray beam to PSPC used in
/ -9 ~ this study. (50 wm width
- - X-ray Cr-Ka x-rays)
R=200mm
o~~~~~~--~~~~~
-10 0 10
Incident angle a (deg)
The result is shown in Fig.2. The ratio I'(a)/I'max was measured as the
counting efficiency; the decrease at a= 14° was 55-60%. The total
relative intensity ratio at angle a, I(a)/Imax, resulting from the x-ray
path length was then computed.
EXPERIMENTAL
-16 -8 0 8 16
Angle from peak position a (deg)
400 VI. DETERMINATION OF CRYSTALLITE SIZE AND STRAIN
(j
.,:.
Smoothed
.........
•J.
1
'iii
c:: r " pc 'T" .,
The correction factor for linearizing the background line was deter-
mined from such distribution. The value of Imax!I(a) at each channel was
smoothed by a polynomial curve fitting reported by Savitzky-Golay2. The
inverse of this value was taken as the correction factor at each channel.
The upper part in Fig.5 shows the distribution of Imax!I(a) after being
smoothed by the data at 21 points. The intensity at each channel was
multiplied by the correction factor on that channel. This correction
wasapplied to another distribution measured under the same condition. The
lower part in Fig.5 shows the result of the correction. The statistical
fluctuation is still observed but a straight-line distribution is seen
over the entire effective length.
Fig.7 shows the change in both the half-value breadth (FWHM) and the
peak position of profile due to changes in the choice of the width,' W.
The peak position on the corrected profile differs in absolute value from
that on the original. However, the shift of the peak position by applying
the correction is slight. On the contrary, the half-value breadth of the
original profile is evidently increased with increase of the width,
W,but the FWHM dependence on W is smaller after the corrections are
applied. In the range below channel-width 310, since the background
equation is determined from the 42 data including the skirts of the
diffraction profile, the level of background line id increased; therefor
the FWHM decreases. This phenomenon is more noticable on the broadened
profiles than the narrow ones. The same measurement was carried out on a
diffraction profile which showed a FWHM of only 2.3 0 and low background.
However, differences of peak position and half-value breadth before and
after being corrected were insignificant.
.......
~
~ 155.7
:;:::
'iii 000 0 0 0 0 0 0 0
C>
<:1.155.6
~
Cb
Q 155.5
4.4
~
Cb 0
.....
00
~ 4.3
~
~
l.&..
4.2
..
4./ L-~_ _ _-'--~_ _ _- ' - - '
The result shows that the correction was not necessary for the stress
determination even with these broadened diffraction profiles.
CONCLUSION
REFERENCES
ABSTRACT
We have determined the strain and particle size for several samples of
palladium powder by time-of-flight nrutron powder diffraction on two
different diffractometers and by x-ray powder diffraction. The results are
compared and found to be in fair agreement. The time-of-flight method gives
good enough precision to reveal deficiencies in the simple models used for
strain and particle size line broadening.
INTRODUCTION
* deceased
EXPERIMENTAL PROCEDURE
The Rietveld method analyzes the entire diffraction pattern and makes
a simultaneous fit of crystallographic parameters (lattice constants, atomic
positions, etc.) and materials parameters (strain, extinction and particle
size). In the simplest model, one that we used for the present study, the
effect of strain and particle size is assumed to be a Gaussian broadening of
each diffraction peak. 3 This Gaussian broadening is convoluted with the
intrinsic peak shape (which is generally quite unsymmetric for a TOF neutron
source).
Figs. 1 and 2 show Rietveld refinements of "alloy" data for GPPD and
HIPD. In our Rietveld plots, the crosses represent the observations, the
line through the data is the Rietveld fit, and the lower curve is the
residual. The difference in resolution between the two diffractometers is
clearly evident. Figs. 3 and 4 show GPPD data for the "old MB" and "Metz"
palladium samples. The "Metz" sample shows considerable line broadening.
Results of the Rietveld refinements are shown in Table 2. We have also
included results from x-ray diffraction analysis when available .
.
:::.
Pd o.• Rho
GPPD
t ..... eo
iii
~ "0
>--
Z ~ ".0
:J
oo 8
0
w
20 5: 2.0
!:::!
N
::;
« __-,....._____---J]L
:l;
....__
,:
.;
t .o----'-.,----~, ~~----------~. .------
~ 00
2
,.
oz
11 10 2" 2'2
.t 1.' 20 21 22
D-SPACING (AI O·SPACING (AI
Figs. 1 and 2. Rietveld refinements of alloy data for GPPD and HIPD,
respectively.
406 VI. DETERMINATION OF CRYSTALLITE SIZE AND STRAIN
Generally the results from the two neutron diffractometers and the
x-ray results are in good agreement. However, the discrepancies exceed the
reported experimental errors, and it is evident from the residual curves of
the Rietveld plots that the fits are not perfect. We tentatively attribute
these discrepancies to the approximate nature of the models used for strain
and particle size, which assume that there are single values (or at worst
Gaussian distributions) for particle size and strain. These models are of
course too simple for the real world, and some discrepancies are bound to
occur. The inclusion of Lorentzian as well as Gaussian line broadening
would make the model more realistic, but only with a severe penalty of
increased calculation time. Future development is likely to be in that
direction.
..
::L
1S
OLOMB
GPPO
.,
::s. O.l
ill
METZ Pd BLACK
GPPO
z~ " !Z u
:>
8 os 50 0.1
oW
~
_ ': .~''....... "-----',l.;l____
C
w
N
~ QO
' u........
J_
. ~ o.o
~
II:
~
a:
o ---. ~-'----"i---ii'_~- - - - -
oz
Z
05 ,0 \5 2.Q
O·SPACING (A)
Figs. 3 and 4. GPPD data for "old MB" and "Metz" Pd samples,
respectively.
A. C. LAWSON ET AL. 407
CONCLUSIONS
ACKNOWLEDGEMENT
This work was supported under the auspices of the United States
Department of Energy. The Manuel Lujan, Jr., Neutron Scattering Center is a
national user facility funded by the United States Department of Energy,
Office of Basic Energy Science. The Intense Pulsed Neutron Source is
operated as a National User Facility by the United States Department of
Energy Basic Energy. Sciences - Materials Sciences, under contract No.
W-31-109-ENG-38.
REFERENCES
ABSTRACT
INTRODUCTION
Many of the problems arise in the spectrum of the x-ray tube. The
angular separation of the Kal 2 doublet and its variation with sin e is the
major source of diffractomete~ peaks' asymmetry and may cause errors in
peak-angle measurement.
EXPERIMENTAL
The basis of this method is that the measured peak profile shapes pee)
result from the convolution of the smearing function [W(e)*G(e)]
(instrumental window function) of the diffractometer and the pure intrinsic
diffraction of the specimen.
where wee) is the spectral distribution of x-ray such as the CuKa doublet
received at the detector
G(e) is the convolution of all aberrations arising from the
diffraction geometry and process, BG is the background, and *
represents the convolution operator
K. OHNO ET AL. 411
pew)
sew) = (2)
W( w)xG( w)
Therefore, the intrinsic peak shape reflected from the specimen can easily
be calculated by the use of a fast Fourier-transform algorism. The first
step is to determine experimentally the convolution W(w)*G(w) by accurate
measurement of the line profiles in an angular range of interest. This
provides the smearing function, which is the convolution of the instrumental
window function as one function and the spectral distribution of the x-ray
source, at the reflection angle common to all powder patterns under a given
experimental condition.
A modified pseudo-Voigt function (mp-V) gave the best fitting for the
raw data shown as Fig. 1.
s(L)2 ]-1 [
+ (l-x).I p ex p -4 In(2)
FWHM(L)2
with
s(L) = s(G) =
1+a(28 - 28p )/FWHM(G)
where x, I p ' a, and 2e~ are, respectively, the fractional factor relating
the ratio of the peak lntensities of G to L, the maximum peak intensity, the
asymmetrical factor (skewing factor) and the peak angle. FWHM(L) and FWHM(G)
refer to the Lorentzian and Gaussian components.
calculated using the non-linear least squares method. Some examples of them
are shown in Fig. 1, since all profiles were similar in the shape and
components, with the exception of peak position and skewing factors. The
optimized parameters of these functions were determined by minimizing the
sum of the squares of the differences between the unsmoothed experimental
points.
Si (111) Si(400)
1.0
>-
....In
.j.J
~ 0. 5
.j.J \
<!
) j\~1.
H
Si (333) Si(620)
1.0
& Si(511)
..,>-
....
~ 0. 5
..,
QJ
C
H
1""".....=----.k-----"--==-"'I-+0.05
r--_~~~~~F~...-+ 0.00
\---...,.---.---.----.---+-0.05
95 96 127 128 129
2e(deg . ) 2e(deg .)
!.:. 1. 0
..
: .
: .
0.5
1.0
>-
........
<II
; 0.5
.u
c
H
67 &8 69
Ze(deg . )
~
.....
o
N
1.0
~ ~ ""
o
M
N M
.... 0
>- -a-.
C\I
x: •
N C\I I
.....u
.
N
""
III
; 0.5 ~
....
c
H
66.0 66 .5 67 .0 67 68 69
2e(dcg . ) 28 (deg . )
Fig. 3 Quartz triplet peaks measured with P~PD(Left) and with X-ray
tube focusing diffractometer(Right) together with profile-
fitting showing relative intensities of the CuKa doublet.
414 VI. DETERMINATION OF CRYSTALLITE SIZE AND STRAIN
1 Ilm
Negative ..."'~----- Lattice misfit- - - -- -... Positive
1.0
1.0 0.5
-------j 0.0
1.0
1.0 \ 0.5
'~"'>':'~.: 'r",.o! Mea sur e d
"v...
~ 0,51---- ------10,0
+-'
r:
H
Deconvoluhd
0,0 I - - - - - - - - - - - - - - - - - j
small shoulder in the lower-angle side of the main peak in Fig. 5 was not a
ringing peak because the shoulder was also observed in the pattern measured
with the PFPD.3 The reflection peaks from the superlattice of the Y'
precipitate in the Y-matrix were not measured because these reflection peaks
were very weak.
CONCLUSION
ACKNOWLEDGMENTS
REFERENCES
G.P. Hamill
GTE Laboratories, Incorporated, Waltham, MA
R Jenkins
JCPDS-Internationai Centre for Diffraction Data, Swarthmore, PA
W.N. Schreiner
Philips Laboratories, Briarcliff Manor, NY
ABSTRACT
In planning for PDF-3, the International Centre for Diffraction Data's full pattern
database of raw diffraction data, it is evident that a standard format for storage and
exchange of diffraction data is necessary. An evaluation of the JCAMP-DX protocol
[1] by a task group" of the International Centre for Diffraction Data has resulted in a
set of format codes specific to X-ray diffraction. The proposed structure of the data is
divided into four parts: the minimal component set required by the JCAMP-DX
definition (name, data, owner, sample identification, data type, etc.), a minimum item
set required to define the X-ray diffraction data, an open selection of requested but
not required information on the sample, its preparation and the instrument, and
finally the data itself in one of several specified formats. All information stored in
JCAMP-DX format is in ASCII characters. Therefore, these data are printable, easily
read by the user and compatible with almost any computer or media storage device.
Codes defining the information are primarily in shortened, but readable, English.
The task group is completing the work on this project and will be presenting its
proposals to JCAMP.
29·1496
Ke.,A1 z(Si,A1)40 11(OH)z dA lal dA !at dA lal
10.1 35 2.242 12 1.536 4b
PotassIUm Aluminum Silicate Hydroxide mite·IM 5.0 30 2.216 10 1.496 40
4.43 100 2.151 12b 1.312 4b
4.33 30 2.<l93 8 1.341 6b
RH. CuK .. l. 1.5418 FIller Ni IMp 4.11 14 2.023 8 1.292 12
Cal"" lal. DilfraClometer ......... 3.66 40 1.989 12
Ref. Brindley, G .. !'eM State University, University Park, !'eDDSyl·
varna, USA .. JCPDS G,QIIJ·iII·Aid Report. (I9TI) 3.31 35 1.951 8
3.06 40 1.903 3
Sy., S.G. 2.931 16 1.820 3
• 5.2 b 8.98 c A C 2.615 12 1.688 10
II y Z
"Ref. 2.560
2.445
85
16
1.663
1.631
12
12
2.386 25 1.585 4b
D.
Specimen from Hungary. K possibly kaolinite. Mica
group. dioctabedral subgroup.
pattern to a set of d-spacings and relative intensities. Such a list is easy to archive and
to search. However, there are several disadvantages in the use of the reduced pattern.
The most important of these is that information such as profile shapes and presence
or absence of amorphous background peaks is ignored in the data reduction process
and, because only the reduced pattern is archived, this and other information is lost.
In an attempt to retain some of this information, the International Centre does
include a portion of analog patterns on the PDF card in certain cases, notably, clay
minerals. An example of such a card is given in Fig. 1.
Unfortunately, such information is difficult to store in the analog format, and these
data cards are expensive to produce because they must be generated by manual
methods. Although identification by search/match methods may not have been
significantly affected in the present database, information about the crystalline
quality, crystallite size, the analytical procedure for calculating d-I pairs, and phase
purity of the material is irretrievable. A second disadvantage which stems from the
use of the reduced pattern is that data reduction techniques change over a period of
time and there may be problems in comparing old d-I lists with those obtained by
state-of-the-art methods.
Problems such as those indicated above could be avoided if the full experimental
pattern were archived along with suitable reference patterns to characterize the
Instrumentation used. Such archiving would allow better data treatment methods to
be used in the future to obtain d-I data. Other benefits that would accrue through the
availability of the complete experimental pattern include the potential use of pattern
recognition techniques in qualitative analysis, and the use of full pattern fitting for
quantitative analYSIS [2,3]. With the recent advent of low cost, high storage capacity
systems, based on CD-ROM technology [4] now available on all levels of computers,
fully digitized diffraction patterns can be saved for future review and use.
Round robin tests [5] have indicated that the angular effects of sample and instrument-
dependent parameters are difficult to correct for on a purely theoretical basis. The
next best alternative is to "model" the intensity sensitivity of individual diffractometers
by use of an instrument intensity-calibration standard. Such effects as divergence slit
aperture, detector dead time, etc., can each introduce systematic errors into the
experimental intensities which are very instrument dependent. Efforts are now
underway to accumulate reference data on as many different diffractometers as
possible in order that the problem can be better quantified and recommendations
made where appropriate. We anticipate the need to store, alon~ with each (set of)
diffraction pattern, a similarly recorded diffraction pattern of the mtensity calibration
standard.
corrected, and annotated with a text editor. Until recently, the main focus has been on
infrared spectra, but JCAMP-DX can easily accommodate Raman, UV, NMR, mass
and other types of spectra, X-ray powder patterns, chromatograms, thermograms, and
other data which require the capability of representing contours as well as peak
positions and intensity. JCAMP-DX also provides for combining adequate
mformation about the sample and method of observation with its spectrum.
Fig. 2. Straight Flow Data Base Format for the storage of X-ray diffraction data
G. P. HAMILL ET AL. 421
B. **X-RAY SOURCE=
1. Tube:
a) Type = (Sealed Beam, Rotating Anode, .•• )
b) Anode = (Cu, Co, Fe, W, ..• )
c) Manufacturer = (Philips, Rigaku, Siemens, ••. j
d) Model = (LFF, FF, BF, cylinder, Disk ..• j
e) Focus Used
1) Line:
a> width = 0.5 [mm]
b> Length = 13 [mm]
2) Spot :
a> Diameter = 0.5 [mm]
f) Serial Number = 40983
g) Power Rating = 1800 [w]
h) Take-off Angle = 6.2 [deg]
2. Generator
a) Type = (Half Wave, Full Wave, Constant Current, Regulated,
••• j
b) Manufacturer = (Philips, Rigaku, Siemens, ••. j
c) Model = XRG-2000
d) Serial Number = 120-00309
e) Settings
1) Voltage = 45 [kV]
2) Current = 35 [mAl
f) Stabilization
1) Voltage = 0.01 [%]
2) Current = 1.0 [%]
Fig. 3. Hierarchical Data Base Format for the storage of X-ray diffraction data (A
portion of the section defining the "X-ray Source")
CONCLUSIONS
The JCAMP-DX format seems well suited for the storage of digitized X-ray powder
patterns. While the straight flow format for data entry provides the simplest system to
utilize, the greater experimental detail definition offered by the hierarchical system
may make its use preferable. The need to store intensity calibration data along with
each experimental pattern (or at least a reference to a full instrument calibration data
file) seems essential.
REFERENCES
[1] McDonald, RS. and Wilks, P.A,Jr., JCAMP-DX: A Standard Form for
Exchange of Infrared Spectra in Computer Readable Form, Appl. Spectrosc., 42
(1988) 151-162.
[2] Smith, D.K., Johnson, J.J.,Jr. and Wims, A, Use of full diffraction spectra, both
experimental and calculated, in quantitative powder diffraction analysis,
AustrJ.Phys., 41 (1988) 311-321
[3] Taylor, J.C. and Pecover, S.R, Quantitative analysis of phases in zeolite bearing
rocks from full X-ray diffraction patterns, AustrJ.Phys., 41 (1988) 323-335
[4] Jenkins, R and Holomany, M., PC-PDF - A search/display program utilizing
the CD-ROM and the complete powder diffraction file, Powder Diffraction, 1.,
(1987) 215-219
[5] Jenkins R and Schreiner, W.N., Intensity Round Robin Report, Powder
Diffraction, 1.. (1989) 74-100
[6] PC-Outline - available through Softworks Development, 750 Stierlin Rd., Suite
142-A, Mtn. View, CA 94043
X-RAY ANALYSIS OF THE STRUCTURE OF WHOLLY AROMATIC
COPOLYAMIDES AND COPOLYESTER CARBONATES
ABSTRACT
This paper summarizes our work on the structures of two wholly aromatic
copolymers: the copolyamide prepared from terephthalic acid, phenylene diamine, and
3,4' -diaminodiphenyl ether (mole ratio 50/25/25) and the copolyester carbonate prepared
from p-hydroxy benzoic acid and hydroquinone, with the introduction of carbonate units
(mole ratio 60/20/20). X-ray fiber diagrams of both copolymers exhibit a number of
aperiodic meridional maxima that are similar to those reported for a series of wholly
aromatic copolyesters. The latter systems have been shown to consist of arrays of parallel
chains of completely random comonomer sequence: the aperiodic maxima arise due to
structural correlations along the extended random sequences. The structures studied in the
present paper are more complex due to the presence of kinked units, i.e. the 1,3 phenylenes
and ether linkages of the copolyamide, and the carbonate groups of the copolyester
carbonate. The results obtained show that the x-ray data can be explained by completely
random monomer sequences. However, the chains are significantly distorted in order to
produce more extended conformations than might be expected for isolated chains. These
distortions minimize the effects of the kinked units and lead to better chain packing.
INTRODUCTION
In recent years there has been extensive x-ray work on the structure of a group of
wholly aromatic copolyesters and similar copolymers, including copolyamides and
copolyester carbonates. These polymers are finding commercial applications as high
strength fibers and high performance moldings (eg. Vectra (Hoechst-Celanese), Xydar
(Amoco) and Technora (Teijin».
Analysis of these structures has focussed mainly on one of the simplest copolymers,
the copolyester prepared from p-hydroxybenzoic acid (HBA) and 2-hydroxy-6-naphthoic
acid (RNA). The homo polyesters are crystalline infusible materials, but copolymerization
lowers the melting point to below 300°C in the middle of the composition range. The melts
are nematic due to the inherently extended chain conformation, which occurs as a result· of
the 1,4 and 2,6 aromatic linkages. The x-ray patterns of melt spun fibers indicate a high
degree of orientation along the direction of draw, and also that three dimensional order
crystallinity is present. This order increases further upon thermal annealing, with the
development of polymorphic crystalline structures. A most interesting feature of the data is
that the scattering in the molecular axis direction, along the meridian of the fiber diagram.
consists of a series of aperiodic maxima, i.e. they are not orders of a single repeat. The d-
spacings also shift with the monomer ratio and could be used analytically to determine the
composition.( 1)
It has been shown in this (1-4) and other laboratories that the above aperiodic
diffraction arises from a structure consisting of parallel chains of random sequences of the
two monomers. The meridional intensity is calculated as the Fourier transform of the
autocorrelation function of the chains, which define the probability and spatial relationship
of the nearest neighbors in the copolymer sequences. This calculation is performed most
effectively by treating the copolymer chain as a one dimensional paracrystal with bimodal
coordination statistics.(5) The intensity is calculated from the first nearest-neighbor
probability function, Ql(z). Ifwe treat the monomers as points, the Fourier transform of
Q 1(z) can be written as the product of three matrices P, M and X:
where
exp 2m ZZB 0
I o PN I
P = PB 0
M = IMBBMBN
MNBMNN
X=
o exp 2m ZZN
Here Z is the coordinate in reciprocal space in the chain axis direction and the subscripts B
and N designate HBA and RNA respectively. The P matrix defines the monomer ratios for
HBA and HNA, PB and PN; the M matrix defines the probabilities of forming the four
possible monomer pairs, and the X matrix contains phase terms defined by the lengths of
the two monomers. The meridional intensity I(Z) is given by
I(Z) = 1 + ~ ~ 2PARe[TAB(Z)]
M·X(Z)
T(Z) = 1- M.X(Z)
in which I is the unity matrix. When we use atomic models for the monomers, the equation
becomes
FAB(Z) is the Fourier transform of the cross correlation of monomer A with monomer B,
given by
Here f is the atomic scattering factor of the atom with axial coordinates z, and the
subscripts A,i and B,j designate the ith atom of monomer A and the jth atom of monomer
B.
Excellent agreement was obtained between observed and calculated data for models
consisting of completely random sequences of monomers.( 1,5) Modification of the M
A.-I. SCHNEIDER AND J. BLACKWELL 425
a) ~L@-A~ot-©-i.~N-@-0-@jN~,
b) ~~-@-~~%oJ(J)
60 20 20
Fig. 1 Chemical structure of a) the copolyamide of terephthalic acid,
phenylene diamine and 3,4' -diaminodiphenyl ether and b) the
copolyester carbonate of p-hydroxy benzoic acid, hydroquinone
and carbonate units.
matrix allows us to consider non-random sequences. This results in shifts and splitting of
the peaks predicted for random sequences, and the agreement with the observed data is
soon lost, such that we can eliminate all but minimal deviation from randomness.(5) This
conclusion has recently been confirmed by NMR work on partially deuterated
copoly(HBMINA) which pointed to a random sequence distribution for the polymerized
50/50 composition.(6)
The above methods have also been applied successfully to other copolyesters
containing three or more monomers.(7 ,8) In the present paper we describe analyses of still
more complex structures: a copoly amide prepared from terephthalic acid (TPA),
phenylene diamine (PDA) and 3,4' -diamino-diphenyl ether (DPE), and a copolyester
carbonate prepared from HBA and hydroquinone (HQ) with the insertion of carbonate
units. The monomers for these systems are shown in Fig. 1. The key point here is that the
presence of the 1,3 phenylene and ether linkagesdnthe copolyamide and the carbonate units
in the copolyester carbonate lead to "kinks" in the chains, i.e. they are not so straight as in
the copoly (HBMINA) system, which must affect the conformation and packing of the
chain.
EXPERIMENTAL
Copoly (TPNPDNDPE)
Fig. 2 shows x-ray patterns for solution spun fibers of copoly (TPNDPFJPDA) with
a 50/25/25 mole ratio. The x-ray fiber diagram shows a high orientation and a perhaps
surprisingly high degree of crystallinity when one considers that this is a copolymer. Fig.
3 (curve d) shows the diffractometer data recorded along the fiber axis (meridional)
direction: five maxima are observed, and two more weak maxima are seen in the film data,
and their positions are indicated by arrows. The d-spacings are indicated above the peaks
and are seen to be nonperiodic.
426 VII. PHASE IDENTIFICATION, STRUCTURAL AND QUANTITATIVE ANALYSIS
Models for the monomers were constructed based on standard bond lengths and
angles. The lengths of the monomers are defined between the nitrogens of the amide
group. Thus the length of the PDA unit is set at 5.64A. The phenyl-amide inclination of
the TPA unit was set at 30°, approximately that found in poly(phenylene
terephthalamide)(9), and depending on the actual combination of torsion angles selected,
the TPA length can vary from 6.75 to 7.26A. The most variable structure is the DPE unit,
where changes of the phenyl-ether torsion angles lead to a range of nitrogen-nitrogen
lengths from 7.39 to 10.02A.(10)
In initial calculations, using both point and atomic, all combinations of monomer
lengths were explored, and the best fit obtained was for the maximum length in each case,
corresponding to a fully extended chain.(10) It is significant that when the DPE unit is at
its maximum length, the zig due to the 1,3 phenylene is compensated by the zag of the ether
linkage, such that linear progression of the chain is maintained. Curve a in Fig. 3 shows
the calculated meridional intensity for an atomic model, where the good agreement between
the observed and calculated d-spacings can be seen. In addition there is a reasonable match
for the intensities.
Nevertheless, an obvious discrepancy between curves a and d in Fig. 3 is the width
of the peak at d ~ 2.1A, where the calculated peak is much sharper than that observed.
Note that both peaks are truncated, and the actual areas under the peaks are approximately
equal. The sharpness of the calculated peak at this position has been seen previously in
work on the copolyesters, and it has been shown that this arises from the assumed linearity
of the chains.( 1,5) Stereochemically acceptable models for the chains are highly extended,
but not fully linear, so that the advance per monomer is only approximately consistent for
each monomer type. Some idea of the non-linearity can be obtained by surveying a number
of models and measuring the axial advance for each monomer. Fig. 4 shows projection of
typical chains, together with the results of a survey based on 100 such chains of 15
A.-I. SCHNEIDER AND J. BLACKWELL 427
monomers, where the distributions of lengths are presented as histograms.( 11) This data
can be incorporated into the calculations via the X matrix, and the results for an atomic
model are shown as curve b in Fig. 3.
It can be seen that the peak at d 5! 2.IA has been broadened by incorporation of the
length distributions presented in the histograms. However, we still have not reproduced
2.14
d)
3.32
.1
c)
o 10 20 30 40 50
29 (degrees)
a)
'0 100.
~
Q)
E 100.
..0
:l
Z OPE
00.
O.~rr~----~~~--~~~-+~~~~~
0.0 5.5 6.0 6.5 7.0 9.5 10.0
Length of residues (A)
Fig. 4 a) Projections of models of typical chains of 15 monomer units.
b) Histograms of axial advance for the three monomer types
derived from 100 chains of 15 monomers.
the width of the observed peaks: to do this a broader distribution of lengths is necessary,
and we have found that we can match the peak widths if the length distribution is
broadened by a factor of 4. This means that the chains are rather more non-linear than
might have been imagined based on the models in Fig. 4. In addition, the length
distribution implies non-linearity which can only be achieved if we have a less extended
conformation. This results in shifts of all calculated peaks to lower d-spacings, since their
positions depend on the average axial advances for each type of residue, which are now
reduced from the fully extended values.
The only way to have agreement between the observed and calculated data is for the
average axial advance for each monomer to be approximately equal to the fully extended
lengths used previously (to generate curve a) and to incorporate broadened histograms.
This results in the calculated intensity curve shown in Fig. 3, curve c, which is in excellent
agreement with the observed data. ( 11) This leads to the following conclusions: the chains
are highly extended but also significantly non-linear, such that some of the monomers are a
little stretched beyond the lengths expected based on acceptable models for isolated chains.
This distortion probably arises because of the need to pack random copolymer sequences:
the kinks on identical sequences are easily accommodated but when random sequences
pack there will need to be some "flattening" of the chains, otherwise there will be
unacceptable spaces in the structure. Perhaps the most likely distortions will be in the
torsional angles of the amide groups, but some distortion of the bond angles and non-
planarity of the phenyl rings can also be expected.
A.-I. SCHNEIDER AND J. BLACKWELL 429
Copoly(HBMIQlCO)
The above analyses have been extended to consider copolyester carbonates prepared
from HBA and hydroquinone, with the introduction of carbonate linkages by the addition
of biphenyl carbonate in the reaction mixture. Five different monomer ratios have been
examined, but only our results on the 60/20/20 composition are presented here. Fig. 5
shows the x-ray pattern of melt spun fibers arranged a) perpendicular to the beam and b)
tilted by 21 0 from the perpendicular. Four non-periodic meridionals are observed, and
these can be seen on the meridional diffractometer scan presented as Fig. 6, curve c, where
the d-spacings are indicated above the peaks.
Fig. 7a shows four random sequence chains in stereochemically acceptable extended
conformations. The presence of the carbonate groups leads to kinks in the chain, which
cannot immediately be compensated for, as was the case for the copolyamide. Hence there
is significant non-linearity, and axial progression becomes difficult unless we constrain the
kinks to alternate, as has been done for the chain in Fig. 7a. It seems obvious that there
3.22
2.15
b)
would be major problems in packing chains like this, when the kinks occur randomly along
the monomer sequences. Nevertheless, we proceeded to treat the data in the same way as
for the previous copolymer, deriving histograms of monomer lengths and calculating the
meridional intensity, which is presented as Fig. 6, curve a. The agreement obtained is
very poor, in terms of position, intensity and profile of the peaks. The first predicted peak
is at d = 6.IIA, deviating significantly from the observed peak at d = 6.4 7±O.1 oA. This
first peak is a measure of the average advance per phenylene ring, and it is clear that the
chain must be more extended than those shown in Fig. 7a. The most likely way of
extending the chains would be by distorting the torsion angles of the carbonate unit. In
constructing the models in Fig. 7a, we assumed that the carbonates, like the ester units, are
planar due to delocalization of the carbonyl oxygen lone pairs. Torsion distortion of the
carbonate involves much less energy than similar distortion of the ester, and it is possible to
adjust this group so that we have a "crank shaft" structure in an otherwise straight chain.
Examples of chains incorporating this type of distortion are shown in Fig. 7b, and the
calculated meridional intensity for a nematic array of chains is shown in Fig. 6, curve b. It
can be seen that excellent agreement is obtained. It is also clear that introduction of
distortions to linearize the chains will also relieve most of the problems inherent in packing
the undistorted chains. We can conclude therefore that the distortions necessary to achieve
good agreement between the observed and calculated x-ray data occur because of the
requirements for chain packing of random sequences. The energy necessary to distort the
carbonate groups will be more than compensated by the better packing of the molecules.
Although we have only distorted the carbonate groups in our model, it is reasonable to
expect the distortions to be spread over the rest of the molecules to involve the ester and
phenyl groups, just as we proposed for the copolyamide discussed earlier. It should also
be noted that studies of deviations for random sequence show that unacceptable deviations
occur between the observed and calculated data for all but minimal non-randomness. The
results presented here for copoly(HBMIQ/CO) will be discussed in more detail in a
forthcoming publication.( 12)
432 VII. PHASE IDENTIFICATION, STRUCTURAL AND QUANTITATIVE ANALYSIS
CONCLUSIONS
X-ray analyses of the structure of random wholly aromatic copolyesters have been
extended to consider a copolyamide and a copolyester carbonate which contain "kinked"
units. In both cases, the results favor random monomer sequences that are highly
extended, and show that the requirements of packing lead to distortion of the chains away
from conformations that might be considered optimal for the isolated molecules.
Assemblies of these chains are now being considered by computer molecular modeling
methods, and the results of these studies will be presented at a future date.
ACKNOWLEDGEMENTS
REFERENCES
1. Gutierrez, G.A.; Chivers, R.A.; Blackwell, J.; Stamatoff, J.B.; Yoon, H. Polymer,
1983, 24, 937.
2. Blackwell, J.; Gutierrez, G.A.; Chivers, R.A. Macromolecules 1984, 17, 1219.
3. Mitchell, G.R.; Windle, A.H. Colloid Polym. Sci., 1985,263,230.
4. Blackwell, 1.; Biswas, A.; Bonart, R. Macromolecules, 1985, 18,2126.
5. Biswas, A.; Blackwell, J. Macromolecules, 1987,20,2997.
6. Muhlbach, A.; Johnson, R.D.; Lyerla, J.; Economy, J. Macromolecules 1988,21,
3117-3119.
7. Blackwell, J.; Gutierrez, G.A. Polymer 1982, 29, 671.
8. Blackwell, J.; Cheng, H.-M.; Biswas, A. Macromolecules 1988,21,39-45.
9. Northolt, M.G. Eur. Polym. J. 1974, 10, 799.
10. Blackwell, J.; Cageao, R.A.; Biswas, A. Macromolecules 1987,20,677.
11. Cageao, R.A.; Schneider, A.-I.; Biswas, A.; Blackwell, J., Macromolecules in
press.
12. Schneider, A.-I.; Blackwell, 1.; Pielartzig, H. in preparation.
POLYMORPHISM IN SYNDIOTACTIC POLYSTYRENE
INTRODUCTION
ISOTACTIC
R R R R R
ATACTIC
,
R "
"\
R
"\
-1C'!~/'!""/'!·~l'!""/C,
"\
R
"\
"
different crystalline forms. In any given sample the crystalline form whicl
develops depends strongly on polymerization, crystallization and processing
conditions. In addition, exposure of SPS to selected solvents can induce
changes in the crystalline structure (8,9). This behavior has also been
reported for IPS (10,11).
EXPERIMENTAL
SAMPLE Mw %SYNDIOTACTIC Wc
API 86,000 90 51
AP2 354,000 >95 63
MCI 354,000 >95 52
MC2 354,000 >95 70
QU 354,000 >95 <5
Structural data was obtained from several SPS samples, each possessing
a different thermal history. Two as-polymerized resins were characterized.
These resins were polymerized using different catalyst systems and are
labeled API and AP2. Structural data was also gathered on two melt-
crystallized materials. The first melt-crystallized sample (MCl) was
prepared in a DSC cell by holding -100 mg of AP2 at 3l0'C for 1.5 minutes,
then cooling at 5·C per minute to room temperature. Sample MC2 was prepared
by heating -100 mg of AP2 in the DSC cell to 3l0'C for 1.5 minutes. The
premelted sample was placed in a second DSC cell for 120 minutes at 263'C,
then cooled to 25'C at a rate of 0.5·C per minute. Finally, a quenched SPS
sample (QU) was produced by holding -100 mg of AP2 in a DSC cell for 1.5
minutes at 3l0·C. The premelted sample was removed from the DSC cell and
immediately submerged in liquid nitrogen, then held there for one minute.
Mass spectrometry
Real·Hm" monltorlng
&conlrol
Real·tlme monitoring
&cOnlnll
Atmospheric x-ray
control diffraction
To
compulel
H.
R..I-Ume monitoring & control
Both lattice type and unit cell dimensions were determined using DrCVOL
(14) and TREOR-4 (15) trial and error indexing algorithms. Program inputs
included observed d and intensity values from the SPS diffraction data.
Experimentally determined unit cell dimensions were refined using the
Appleman least squares method (IG).
POLYMERIZATION CONDITIONS
\~~ CATALYST I
--........---..
~ CATALYST II
...... """.........-~
10 20 30
28 (DEGREES)
CRYSTALLIZATION CONDITIONS
MELT CRYSTALLIZED
29 (mm)
SYNDIOTACTIC PS
>
I-
u;
Z
W
I-
~
29 (mm)
The XRD pattern for sample MC2 is pictured in figure 5. The high
crystallization temperature and slow cooling rate of this sample produced a
material of large crystallite size and substantial crystallinity (table 1).
A unique feature of this set of structural data is the large number of
diffraction peaks which possess d values greater than 6 A. Based on this
data one would expect that a large unit cell (>1000 A3 ) would describe the
crystalline structure of sample MC2 . Unfortunately, no choice of unit cells
(up to 2500 A3 ) could account for more than 80% of the observed diffraction
B. G. LANDES ET AL. 439
SIMULTANEOUS DSCIXRD
>-
I-
in
z
W
I-
~ .l
10 20
28 (DEGREES)
peaks from this material. This suggested that more than one crystalline
form may be present in the sample.
TEMPERATURE
TEMP. ('C)
>-
I- 150' C
in
z
W
I- 200'C
~
250'C
Cooled
100' C
10 20 30
28 (DEGREES)
SIMULTANEOUS OSCIXRO
TEMP. (·C)
10 20 30
29 (DEGREES)
The XRD pattern which is observed for sample APl above l80·C is similar
to that for the AP2 sample. A plot of the XRD patterns for sample AP2 as a
function of temperature is pictured in figure B. Note that the XRD pattern
which is observed for this sample at 2S·C persists until the material is
heated through its melting transition. No structural transformation is
observed for this sample under these experimental conditions.
Data from DSC/XRD experiments on several different SPS samples were used
to identify the major scattering peaks of the different crystalline phases.
This data assisted us in defining the contributions of each crystal form to
the observed scattering pattern of sample MC2 (figure 5). Two separate sets
of d values and relative intensities (one from each structure) were
identified and used independently as input to the Treor-4 and DICVOL
B. G. LANDES ET AL. 441
ORTHORHOMBIC MONOCLINIC
a=8.8 A a=19.8 A
b=28.9 A b=12.9 A
c=s.1 A c=s.1 A
P=9s"
442 VII. PHASE IDENTIFICATION, STRUCTURAL AND QUANTITATIVE ANALYSIS
CONCLUSIONS
REFERENCES
John J. Tomlinson
Oak Ridge National Laboratory
Oak Ridge, TN 37831
ABSTRACT
INTRODUCTION
EXPERIMENTAL
(B) Instrumentation
X-ray diffraction results for PE with 10%, 20% and 30 mol% NPC show
that there is little solubility of NPC in PE and there are two distinct
phases present in the system at 19 0 C; these are monoclinic (8) and
tetragonal (a) phases. The PE-rich a phase is more abundant in this two-
phase mixture than in the NPC-rich 8 phase. These estimations have been
made based on the peak heights from the x-ray diffraction patterns by
monitoring the intensity of the (011) reflection of the monoclinic phase and
(002) of the tetragonal phase. It was observed that the intensity of (011)
of the monoclinic phase increased as the concentration of NPC increased,
suggesting little solid solubility of NPC in PE at approximately l7°C in
these melted and crystallized PE-NPC powders. These results are quite
different from results with the PC [CSH12H3l - NPC [CSH12H2l system, where a
large solubility of NPC in PC12 was observed, even though the structure of
PC is the same as that of PE (space group 14).
290 co
X-RAY DIFFRACTION PATIERNS
280 LlOUID ry'(200)
270
-210·CT y'
250 ~ --G_ y'+ LlOUID -197·Ci...:...-
260 ~ - - 173 OC!J!..±. y'
240 --e....!-_ w
a.. ~
, UJ -151·Ct ex
--127OC :5
m ---- (fl. ..., __ 105°C
220 o _ 92°C '"0
:r:
210 l>
y'-- l>
0'>
(31-.......
a.. ••
.. •
-.....l
'.
790C
680C o.+y l>
en
200 l> l> 51°C m
Z .~ 440C
IT 190:""'" e e o
~ 1W e *" ..~ -- 41·C m
0. + Y' -
_ - 36"C
31"C Z
~ 170 o " .,""
""II o.+~ :j
- - 250C 'T1
1= 160 o
0. -- 17°C n
~ 150 o l>
-l
~ 140
~ 130
oz
o Gl a + ~ phases
I- 120 en
o a + yphases 1'(111)1
110 -l
o a phases :0
100 __ 205"C~' C
o V a + y' phases --195OC (")
90 cx+y l> y'phases ~ E:::::::::~y~'(:I~I~I):-~, EIi:!, ~.:L::===~~~u.1I --182OC 0.+1' -l
C
80 --170OC :0
70 (fl. " ,_ --l58OC 0. l>
g . .... --145OC r-
60 o (3
. .Y.J
(I'I) ...._ _1IIiIIJ --l33OC l>
50 --122OC z
a.. __ 970CI o.+y o
40 -- 6O"C o
30 Z '. C
e cf!. I I -- 400C
Gl o .11i. -- 27"Ct A
l>
20 o.+~ (\I z
-- 19OC~1' -l
10
I!(OI :::j
OQ__--~--.---~--~--~---J l>
o 10 %NPG 20 30 -l
<:
m
Figure 1. Phase diagram of PE-30 mol % NPG constructed by using X-ray diffraction and differential l>
scanning calorimetric analyses. Guinier X-ray powder diffraction patterns obtained from z
l>
PE-10 mol % and PE-20 mol % are presented to show the phase stability ranges. (These X-ray r-
-<
en
diffraction results are important to determine the phase transitions, in particular, where
the enthalpies of transitions are very low and the DSC data is difficult to interpret.) en
D. CHANDRA ET AL. 449
from the PE-rich Y' phase. These results show that as the PE+10 mol% NPC is
heated, first a low temperature Y is formed and then this Y phase transforms
to a phase which in turn transforms to a high temperature Y' phase.
The lattice parameters of a and Y phases for PE+20 mol% NPC with
increasing temperatures are plotted in Figure 2. The lattice parameters for
the a phase in the a+B region below ~380C show a slight growth in the
c-direction with very little increase in the a-direction. (The monoclinic B
phase is not plotted in this graph.) As the temperature is increased the
a + B phases transform to a+ yphases and an increase in the c-parameter of
the a phase is observed while the a-parameter does not increase signifi-
cantly. In the temperature range of 40-97°C there is a gradual nonlinear
increase in the c-parameter associated with the changes in composition in
equilibrium at different temperatures. The solubility of the NPC in PE
increases with the increases in temperature. The composition change (~c)
per °c increase varies depending on the temperature range because of the
shape of the solubility curve. Note the anisotropic behavior of the a phase
in which the c-parameter increases significantly as compared to the
a-parameter throughout the two phase region. In the PE-rich single y' phase
region, increases in lattice parameters are linear because the composition
remains unchanged. The lattice parameters of the high-temperature phase of
the pure NPC (y) and PE (y') are comparable to those of the PE-NPC binary,
just above B + y and a +y' transition temperatures.
9.1.,-----------------,
9.0
en
a::
w
I-
W
::iE
c:(
a::
c:(
,:
a. 8.9
w
()
~
c:(
...J
8.8
if
~: a+~
8.75 .Y a+y a y'
6.1 ,'\'..
:01": J Iii i, 'H
~
'
: : '" a+y'
a(a)
6.05 t---.--.,---..-,.---.--,--.-...,.--.---1
o 50 100 150 200 2~0
TEMPERATURE
ACKNOWLEDGEMENT
REFERENCES
1. E. Murrill and L. Breed, Annual Summary Report No.1 NASA Contract No.
NAS 8-21452 (1969).
2. D. K. Benson, J. D. Webb, R. W. Burrows, J. D. McFadden, and
C. Christianson, SERI Report, Task Nos. 1275.00 and 1464.00 WPA 304
Mar. 1985 (available through NTIS, SERI/TR 255-1828 Category 62e).
3. D. Chandra and C. S. Barrett, Effect of Interstitial Dopants on the
Thermophysical Properties of Solid State Phase Change Materials.
Draft of Final Report DOE Contr No. DE-AC03-84SF-12205, Jan. 30 (1986).
4. D. Chandra, C. S. Barrett, and D. K. Benson, Adjustment of Solid-Solid
Phase Transition Temperatures of Polyalcohols by the use of Dopants,
"Advances in X-Ray Analysis," Plenum Publishing, 29: 305-313 (1986).
5. F. J. Llewellyn, E. G. Cox, T. H. Goodwin, J. Chern. Soc. 1883, (1937).
6. I. Nitta and Z. Krist, 112,234 (1959); and I. Nitta and T. Watanabe,
Bull Chern. Soc. Japan, Vol. 13, 28 (1938).
7. D. Chandra, J. J. Fitzpatrick and G. Jorgenson, Adv. in X-Ray Analysis
Vol. 28: 353, Plenum Publishing, (1985).
8. Timmermann, Solid State Phase Transitions in Pentaerytritol and
related Polyhydric Compounds, J. Phys. Chern. Solids, Pergamon Press,
v:18, No.1 (1961).
9. J. G. Aston, "Physics and Chemistry of Organic Solid State," Vol. 1,
Chapter 9, Interscience Publishers Inc., New York, NY, (1963).
452 VII. PHASE IDENTIFICATION, STRUCTURAL AND QUANTITATIVE ANALYSIS
Ceramics Division
National Institute of Standards and Technology
Gaithersburg, Md 20899
ABSTRACT
INTRODUCTION
The electronic structure, stable valence state and the size of the
lanthanide elements play important roles in governing compound formation.
Among these factors, the progressive reduction in size of these elements,
as one goes across the lanthanide series, known as the lanthanide
contraction, allows one to study the trend of the chemical and physical
properties and phase formation of compounds. Since knowledge of crystal
chemistry and phase equilibria is essential for controlling processing
parameters and understanding material properties, a systematic study of
the phase relationships in the BaO-Rz 0 3 -CuO systems has been initiated at
the National Institute of Standards and Technology (NIST). Our goal was
Detailed phase diagram studies for the BaO-Y2 0 3 -CuO and BaO-La2 0 3 -
CuO systems have been prepared by Roth et al. 2 and Kilbanow et a1 3 (Fig.
1a and 1b). A comparison of these two diagrams reveals substantial
differences. For example, along the binary BaO-R20 3 edge, the La system
has only one compound while the Y system has four. Although both the
R2 0 3 -CuO systems contain only one binary phase, they are of different
composition; namely, the La system forms a 2:1 binary phase La 2 Cu0 4 while
the Y system forms a 1:1 binary phase Y2 Cu 2 0 S ' Furthermore, a few more
ternary compounds and solid solution series are found in the La system as
compared to the Y one. Presumbably the size of the lanthanide R has an
important effect on the phase diagrams. We have investigated the CuO-rich
region of the phase diagrams of several selective lanthanide systems with
R = Nd, Sm, Eu, Gd and Er with the size of R mostly between La and Y in
order to determine if a trend exists among them. Several series of binary
and ternary oxide compounds as well as a variety of compositions in the
BaO-R20 3 -CuO systems have been studied in order to be able to find out
the phase relationships. This report summarizes our current findings4-11.
EXPERIMENTAL
Solid state reaction techniques were used for all sample preparations.
Similar heat treatment procedures were adopted for all series of
Ba2-zRl+zCu306+x materials. Most compositions were prepared from
stoichiometric mixtures of CuO, R20 3 and BaC0 3 . In the case of R = La,
La(OH)3 was used instead of La 2 0 3 . Before each firing and annealing, the
powder was pressed into pellets and placed on MgO single crystals. The
pellets were heat treated at both 850°C and 900°C for one day, then fired
in air at 950°C. Several regrindings and annealings took place until a
single phase material was confirmed by .the X-ray powder diffraction.
R3+ La Nd Sm Eu Gd Dy Ho Y Er Tm Yb Lu
RZ0 3 -CuO:
Rz Cu0 4 1Z T T T T T
°T °T °T °T °T °T °T
Rz CUz Os 1 4 , 1 S
°°°°°
BaO-Rz 0 3 :
BaRz 0 4 8 T T T T T T T T T
°°°
Ba 3R4 0 9
°° T T T
BazRz Os
°°° T
Ba 4Rz 0 7
°°° T
BaO-Rz 0 3 -CuO:
Baz RCU 30 6+x zo T T T T T T T T T T T T
BaRzCuO s
green phase 7
°
Baz +z x R4 - Zx Cuz - x 10 - Zx
°° T T T T T T T T T T
brown phase 7
zo
T T
°° ° ° ° °T °T °°°
Ba3+xR1-xCuZ-z06.4-z
°T ° ° °° ° ° °°
BaR4CuS013+x16,17,18
Ba3R3Cu6014+X 10 °T °T °T ° ° ° °°
Ba 1+x Rz -x Cuz 0 6 - x /z Zl
T
° °° °°
°°° °°°° °°°
a b
Figure 3. Polyhedral environment of (a) R0 10 and (b) ROB found in the
structures of Baz+zxR4-ZxCuZ-x010-Zx' R=La and Nd.
W. WONG-NG ET AL. 457
o-+.~~-()-I--Z =}
eCu fl)Nd 00
Nd.Cu 0. -Type
Figure 4. Crystal structures of RZCu0 4 , (a) R=La, (b) R=Pr, Nd, Sm, Eu
and Cd.
While all the RZ 0 3 :CuO systems contain only one binary compound at
ambient atmosphere, there exist two distinctive compositions. Binary
compounds of 2:1 composition with general formula RZ Cu0 4 can be prepared
with the lighter and larger size of R, for example, with R=La, Pr, Nd,
Sm, Eu and Cd, whereas oxides in the second half of the lanthanide series
tend to form a 2(~Rz03):2CuO binary compound RzCuzOs. Not all Rz Cu0 4
phases are isostructural. For example, while La Z Cu0 4 is orthorhombic
with space group Cmca, and has the distorted Kz NiF 4 type structure as
shown in Figure 4, Pr Z Cu0 4 , Ndz Cu0 4 , SmZ Cu0 4 , EUZCu0 4 and Cdz Cu0 4 are
tetragonal with space group P4/rnrnrn. These tetragonal structures do not
resemble the Kz NiF 4 type but have a coplanar Cu-O layer similar to that
found in CaF z type structure 1Z . These features are illustrated in Figure
4. Doping the La ZCu0 4 structure with a small amount of a 2+ ion such as
and Srz+ changes the structure from orthorhombic to tetragonal and
results in a superconductor phase.
(e) z=O.3
>--
....
~o
~
.
~*
....
~ 0 0
b 0
A 0
0 0
0
•
/}.
0
0
• 0
.Hol OHoD .Cul /}.cu D OOI-V
0
2000 . . , - - - - - - - - - - - - - - - , 485.0 , - - - - - - - - - O - T - b - - - - - - ,
§ 1950
E
C
475.0
¢ lly
o " Pr o
o o flo
~ 1900
o "'0
" Nd Y
x X 465.0 0 Er
ill ill
E 1850 E 0 Tm
:>
o
>
" Sm g 455.0 0 Yb
1800 - " Eu
" Gd 0 Lu
175 a
10.5
.,. .- -r- -r- -r- - r- -.- -.~'~I~~~-r-r ,~J
11.0 11.5 12.0 12.5
445.0 +----~-__.__-~--r___-..,--~
10.0 10.5 11.0 11.5
Ionic Radius x 100 nm Ionic Radius x 100 nm
Figure 7. Size dependence of the unit cell volume of (a) R2 Cu0 4 , R = Pr,
Nd, Sm, Eu and Gd and (b) R2 CU2 Os , R = Tb, Dy, Ho, Er, Tm, Yb and Lu.
are connected via corners in a wavelike chain along [010]. The Ho 3 + ions
are surrounded by octahedra which form I-dimensional infinite chains
connected via edges.
Figures 7(a) and (b) are plots of the cell volume V of R2 Cu0 4 and
R2 Cu2 0 S versus the ionic radii of Shannon. The monotonic decrease in
crystallographic volumes, as the ionic radii of the lanthanides decrease
across the series, follows the well-known lanthanide contraction. The Gd
compound which deviates from this linear dependence may be due to the
special stability associated with the half-filled Gd 3 + (4f7) f-subshell.
Figure 8. Crystal structure of the compound BaLa4 CuS 0 13 . The full and
open circles represent Cu atoms belonging to the CuOs pyramids and the
Cu06 octahedra, respectively. La 3 + and Ba2 + ions are located in the
hexagonal and pervoskite tunnels, respectively. The z-axis points out of
the plane of the paper.
460 VII. PHASE IDENTIFICATION, STRUCTURAL AND QUANTITATIVE ANALYSIS
103
110
2~. •.
TWO - THETA (DEGREES)
Figure 9. X-ray diffraction patterns of Baz-zLal+zCu306+x' The appearance
of the BaLa4Cu5013+x is indicated by the arrows for z=0.8 and 0.9.
The La system appears to be the only one among the lanthanide systems
that forms this black metallic conductor. The BaLa4CuS013+x compound is
tetragonal with space group P4/m. The lattice parameters of a sample
which was prepared in air at 950°C were found to be a= 0.86602(5) nm and
c = 0.38629(3) nm 16 , which are closely related to that of the cubic
perovskite: a~ap j5, and c=a p=.38594 nm. ~ichel et al. 16 . 17 described the
framework [Cus 013 1 to be bunt up from corner- sharing CuO s pyramids and
Cu0 6 octahedra forming hexagonal tunnels and perovskite cages where the
La 3+ and Baz + ions are located in an ordered manner. Figure 818 shows
the structure of BaLa 4 Cu S0 13 in which each CuO octahedron shares four
corners with four pyramids, and the two remaining corners with two other
octahedra. Each pyramid is then connected to four other pyramids and one
octahedron. Oxygen vacancies,which are found to be between pairs of
CuO s , form l-dimensional channels along the z-axis. The lanthanum ions
are located in the hexagonal tunnels whereas the barium ions are located
in the perovskite tunnels.
R z Ionic Radius
(nm)
..
II!
"green"
103
110
BaO(BaCO)
BaO(8aC03)
BaO(BaC03)
BaO(BaCO)
CONCLUSION
ACKNOWLEDGEMENT
REFERENCES
10. W. Wong-Ng, C.K. Chiang, B. Paretzkin and E.R. Fuller, Jr., Powd.
Diff., accepted for publication, 1989.
11. J.E. Blendell, W. Wong-Ng, C.K. Chiang, R.D. Shull, and E.R. Fuller,
submitted to the Proceedings of the Annual Meetings of The Metallurgical
Society, Las Vegas, Nevada, Feburary, (1989).
13. J.G. Bednorz and K.A. Muller, Z. Phys. B-Condensed Matter 64, 189
(1986) .
16. C. Michel, L. Er-Rakho, and B. Rave au , Mater. Res. Bull. 20, 667
(1985) .
R. A. Newman
Analytical Sciences
Michigan Applied Science & Technology Laboratories
The Dow Chemical Company
Midland, Michigan 48667
INTRODUCTION
Bround Surface
Fracture Surface
EXPERIMENTAL
In the case of the fracture surface of the PSZ parts, where the area
to be analyzed was small, the incident x-ray beam was collimated to
eliminate diffraction from other regions of the part surface (see figure
R. A. NEWMAN 469
Huber Incident
Beam Monochromator KOIl Radiation
STABILIZED ZIRCONIA
295.3 Sample A ~~
p$.<f
,\"
246.15
196.92
,..
0 (j~<QI'
..
4<
1/1
C
147.69 #'.~~>f.
~I
::I
0
0 98.46 Polished Surface
Ground Surface
49.23
26 28 30 32 34 36 38 40 42 44
2-Theta (deg)
Figure 2. Typical diffraction patterns obtained from the different surfaces
of interest on test part A. Note the near coincidence of the
cubic and tetragonal phase 1 1 1 reflections.
tThe intensities and FWHM values for the Tetragonal phase are less
reliable due to overlap with the 1 1 1 reflection of the cubic phase.
R.A. NEWMAN 471
II: 6c- A
r-
4r-
i
I L I
Polished Ground Fracture
Surface Surface Surface
,
"tl - 0
:i 0.8 - &.
J:
- <>
~ - ~
•
0.6 0 A
.lO:
10 -
-
II)
A
D.. 0.4
-
0.2 -
-
0.0 I I I
Polished Ground Fracture
Surface Surface Surface
CONCLUSIONS
Even though several limiting factors were present, the analyses yield-
ed the expected trend in T --> M phase transformation for the polished
surface and the fracture surface. The low values and wide range (from 5 to
>10 %) for the M phase concentration obtained for the ground surface of the
samples (see fig. 3) are due to the combination of preferred orientation
and the non-uniform grinding technique. This was clearly evident by visual
inspection of the samples. The degree of martensitic transformation to the
M phase in the fracture zone as observed by X-ray diffraction (ca. 10 - 14%
by intensity ratio) was less than expected for a high quality fabricated
PSZ ceramic part. Overall, the analyses were compromised by the small
dimensions of the part surfaces, which restricted the sampling of the phase
composition by the incident x-ray beam. The presence of interferences
(primarily preferred orientation) further degraded the quality of the
results. Better phase composition sampling and analysis results could be
obtained if larger test parts were provided, with more uniform surface
treatment areas.
REFERENCES
INTRODUCTION
is provided below, along with several sets of results which confirm the
accuracy of the technique both in the evaluation of ~ (for several
compounds of reagent grade purity) and for the measurement of the pyrite
abundance in each of the eight Argonne Premium Coals. 8 Quantitative
EXPERIMENTAL
Each analyte was finely powdered and then lightly packed into a
conventional x-ray diffraction sample-holder (aluminum) in such a way as
to produce a "flat" surface for the powdered sample. The aluminum
sample-holder used in these experiments had an inside diameter of 2.54
cm and was machined to provide a sample depth of 0.05 cm.
10r-----------------------------------------------------------,----,
311
III
200
I
I .,ii, I
i~~,,---~J~I
A" I
,I
,I
I!
iI
i ______________~~
'L~l_. I
~
: ! I 1~IIIIIIIIIIIIIIIIIITillllilllillll'lllill
20 30 40 50 60 -;0 60
TWO THETA
p(2e) (2)
where 0 is 1/2 of the Bragg diffraction angle, and Ap and AS are the
integrated areas of the diffraction peaks of aluminum in the absence of
and in the presence of m grams of the powdered analyte, respectively.
For each of the three large diffraction peaks of aluminum, Twas
determined from the Ap/AS ratio (see TABLE I). For all subsequent
evaluations the arithmetic value of T (which agreed quite well with the
value for T determined by the AgBr experiments described above) was
used.
For this sample of amorphous silicon dioxide, or for other samples
whose absorption is less than "critical", the mass absorption may be
calculated from the decrease of the peak area(s) of the crystal
substrate peaks by:
~ = (T·sin0/2·m)·ln(AS/Ap). (3)
478 VII. PHASE IDENTIFICATION, STRUCTURAL AND QUANTITATIVE ANALYSIS
AL PEAK Jl
(cm 2Jg)
111 35.06
200 35.87
311 34.69
MEAN VALUE: 35.21
RELATIVE MEAN DEVIATION: 1.26%
With stronger absorbers, the transmission of the 111 and the 200
are diminished much more than is the transmission of the 311 peak of
aluminum. While this effect provides less information from which the
mass absorption coefficient of the absorber may be calculated, it also
causes less of the diffractogram of the absorber to be complicated by
the diffraction peaks of the crystal substrate.
Coals have been examined for many years by XRD methods. 10-14 Coals
represent complex "real life" samples which contain a carbonaceous
480 VII. PHASE IDENTIFICATION, STRUCTURAL AND QUANTITATIVE ANALYSIS
2.6 - , - - - -
2.4
2.2
1.8
.r-. 1.6
u ~
W"U
c
'-0 1.4
~
~ ~
+'
".c
C
01-
°
"
1.2
U'-'
0.8
0.6
0.4
0.2
0
76.4 76.8 77.2 77.6 78 78.4 78.8 79.2 79.6
Two-Theta
A. MINERALS
B. ORGANIC COMPOUNDS
10 20 30 40 50 60 70 80
2 Theta
Fig. 4. The x-ray diffractogram of a 0.2356 g sample of Argonne
Premium Coal 301, a high volatile bituminous coal (Illinois
No.6). The coal reportedly contains 2.47% pyritic sulfur
and 2.01% organic sulfur. The 1.63 A diffraction peak of
pyrite is darkened. An * indicates the 311 peak of AI.
/
35
("')
'0 30
~
"; 25 -
/
Q)
Ul
o
j 20
<=:
;:l
o
u 15
<>:
'"<>:
~
10
:.:
;;j
p., 0······0······· ··0
5
{ji
O~~-------------------------------
0.00 6.00
% Pyrite
Fig. 5. 1.63 A Pyrite Peak Area (Qpy) Measured from IM(26) and the
1.63 A Pyrite Peak Area (Qopy) Measured from from 10 (26) vs.
Pyrite Abundance Reported for the Argonne Premium Coals.
482 VII. PHASE IDENTIFICATION, STRUCTURAL AND QUANTITATIVE ANALYSIS
The measured intensity for each of the Argonne Premium Coals was
converted to the corresponding absorption corrected intensity by eq. 1,
and the latter were used in subsequent analysis for the pyrite (FeS2)
abundance in each coal. Also shown in Figure 5 is a graphical
representation of area under the 1.63 A pyrite peak in the absorption
corrected diffractogram, QOpy, plotted versus pyrite abundance for the
eight Argonne Premium Coals. The linearity of the relationship is
expressed by R2 > 0.997.
(5)
CONCLUSIONS
REFERENCES
INTRODUCTION
TECHNIQUE
TABLE 1
Properties: Properties:
Powder density, gm/cc 0.6-0.8 Powder density, gm/cc 0.6-0.8
Properties: Properties:
Powder density, gm/cc 0.6-0.8 Powder density, gm/cc 0.6-0.8
Until recently this information could not be put to practical use when
analyzing the detergent for total composition due to the complicated
matrix. Today, with an understanding of detergent formulary and good
subtraction software, the diffractionist can often rapidly distinguish the
various species of builders and bleaches. Table 4 details the analytical
scheme used in our laboratory to speciate the various builders that may be
present.
TABLE 2
ALKALINE BUILDERS
(e.g. sodium carbonate, sodium silicate, etc.)
SEQUESTERING AGENTS
(e.g. EDTA, NTA, etc.)
BLEACHING COMPOUNDS
(e.g. sodium perborate, -chlorinated TSP, etc.)
TABLE 3
38-241
TABLE 4
ANALYTICAL SCHEME
I
SOLUBLE MATERIAL
C02 BY
>= 27- ACTIVE & TOTAL iON SELEClM:
I
ALKALI TITRATION
E1£CTROOE \
SPECTROSCOPY
SPECIES
IDENTIFICATION ION BY XRD
TOTAL CHROMATOGRAPHY SODIUM
BY X-RAY OlFfRAC11ON BLEA I CARBONAT£.
Oxidation/Reduction P, P SPECIES,
I
ZEOLITE An
Titration (COMFlRI! BY XRO)
SODIUM
BlCARIIONAT£.
Si de
I (SPECIES BY XRO) NTA,
SODIUM
SESQUICARIIONATE
SPECIES BY XRD
yls SODIUM PERBORATE
B
CITRATE,
SODlUII ISOCYANURATE
CHLORiNATED TSP
S04 de CI
QUANTIFY BY XRD (SPECIES BY XRO)
DETERGENT A
X10 31
~,110
4.05
3.20 lIater 10%
2.4:5 Total Surfactant 16
1.88 Insoluable Material 30
1.2:5 Sodium Sul fate 19
0.88
0.4:5
0.20
8.05
10.0 20.0 30 •• 41 •• sa ••
:5.88
4.05 Sodium Perborate 8%
3.28 Sodium Carbonate 11
2.4:5 Sodium Si l i cate 2
1.88
Mi sce II aneous _5_
1.25
100%
8.80
8.4:5
0.20
8.9:5
L DETERGENT A
190.8
88.8
68.0
48.8
28.0
190.0
88.8
60.0
48.8
28.8
DETERGENT B
X18 31
~.i10
4.85 lIater 5 %
3.20
Total Surfactant 18
2.4:5
Insoluable Materi al 7
1.89
1.25 Sodium Sul fate 32
0.se
0.4:5
0.20
0.0:5
10.0 29.0 30.0 40.0 :50.0
5.90
4.85 Sodium Tripol yphosophate 16
3.29 Sodium Pyrophosphate 11
2.4:5
Sodium Sil i cate 3
1.88
Miscellaneous _5_
1.2:5
0.se 100%
0.4:5
0.20
0.0:5
L
198.9
89.9
60.9
49.9
28.9
198.9
89.9
60.0
40.8
28.8
LS~lel Tri~hloroisoc~i~
198.9
89.0
60.8
48.8
28.8
18.8 28.9
I I
39.8
I
40.9
I, I I I I
5&.8 60.8
I I I II I I I I 31-1956
198.8
89.9
60.9
40.8
28.9
Both the diffractionist and the detergent chemist must be aware that
the "jargon" used in each others' field is different for the same
compound; for example, the PDF lists Trona while detergent literature
refers to this as sesquicarbonate. The PDF also refers to chlorinated
isocyanurates as chlorinated-S-triazine trione. (See Figure 5.)
REFERENCES
Breck, Donald W. "Zeolite Molecular Sieves", John Wiley & Sons, New York,
(1974) .
ABSTRACf
Existing analytical techniques for the determination of crystalline silica at levels of less
than 1.0% were found to be inadequate to meet the new requirements. An improved
technique based on x-ray diffraction has been developed specifically to analyze perlite for
crystalline silica, which occurs largely in the form of quartz, at the 0.1 % level. The
technique employs long counting times and improved sample preparation and mounting to
increase both precision and accuracy, and to lower the detection limit to less than 0.1 %.
The technique was tested on a large number of samples from a variety of sources and
proven to give excellent results for all types of expanded perlites and perlite ores. The
procedures developed are applicable to a wide variety of materials in addition to perlite.
INTRODUCTION
The publication of IARC Monograph 42 which assigned a 2A or "probable carcinogen"
classification to crystalline silica triggered an immediate need to accurately determine the
weight percent of crystalline silica in any material at the 0.1 % level. A wide variety of
procedures exist for determining total silica at levels well below 0.1 %, however, a review of
the literature, indicated that procedures for determining only the crystalline portion of the total
silica were generally limited to levels of 1% or greater. As a result of the need for analysis
and the lack of an acceptable procedure, research was undertaken to determine if satisfactory
procedures could be developed to analyze products which contain 0.1 to 1% crystalline silica.
Among the products studied was perlite and it is the results of this work which is presented
in this report.
DISCUSSION
Perlite is a hydrated volcanic glass, generally of rhyolitic composition (Table 1).
Mineralogically it is 95% plus glass with minor amourtts of micas, feldspar, and crystalline
silica.
Crystalline silica in perlite may occur in a number of polymorphic forms the most
common of which are quartz, cristobalite, and tridymite. Experience has shown that
cristobalite and tridymite are very rare in perlite hence the research was focused entirely on
quartz. Because perlite is very high in total silica, any determination must be based on a
technique sensitive only to the crystalline portion since the results cannot be checked by an
analysis for total silica. Techniques which can discriminate between crystalline and
amorphous silica include infrared spectroscopy, x-ray diffraction, polarized light
microscopy, differential solubility, and heavy mineral separations. Combinations and
variations of all these techniques had either been tried in the past or were tried in the course of
the current research. X-ray diffraction was found to be the most satisfactory for perlite
despite a number of limitations. Diffraction analysis is the most cost effective, reliable, and
precise method available. It is the least sensitive to operator skill and variations in the nature
of the perlite. The combination of density separation, optical microscopy, and x-ray
diffraction is the most sensitive but, is slow, expensive, dependent on technique, and yields
results of lower precision.
Perlite is a difficult matrix in which to determine quartz by XRD. On one hand the
linear absorption coefficient is within 5% of that of quartz hence matrix absorption effects can
be ignored. On the other hand, the glassy material which makes up 95% + of the bulk
produces a very high background signal. The high background leads to a very poor peak to
background ratio which greatly reduces the accuracy of the measurement and raises the
detection limit. The other crystalline phases which are usually present in the perlite especially
micas and feldspar cause interferences with the quartz peak because of peak overlap. The
poor peak to background ratio can be compensated for by accumulating a large number of
counts. The peak overlap from other minerals can only be treated on a case-by-case basis by
a complete diffraction analysis of the sample.
In addition to analyzing the perlite ore it is also necessary to analyze the expanded
product. The expanded perlite has lost most of its water and has a bulk density of about 0.1
grams/cc. Fortunately the glassy foam produced by expansion is very brittle and can be
easily ground to resemble the ore in terms of particle size and bulk density.
PROCEDURE
Sample Preparation
A uniformly fine particle size is essential for accurate XRD analysis. The ideal
procedure would be to grind the sample to be analyzed to 100% <20 ]lm particle size.
Perlite ore and especially expanded perlite is difficult to grind and sieve. The ore tends to
cake while grinding and the expanded material clumps because of electrostatic charge. Under
the circumstances it is best to grind the sample as fine and unifonnly as possible and not
sieve it.
No standards for quartz in perlite or perlite-like material are generally available hence it
is necessary to prepare calibration standards. Standards for expanded perlite were prepared
by spiking quartz free expanded perlite ground as described above with -325 mesh quartz in
levels of 0.5 to 5 weight percent. The spikes are mixed by hand in an agate mortar and x-
rayed several times to verify their homogeneity. Standards for 0.1 and 0.2 weight percent
were made by diluting the 2% standard with quartz-free perlite.
Two approaches were used in generating ore standards. Because of concerns about the
effects of crystal size and degree of crystallinity,ore standards were generated from mixing
quartz.free ore with an ore of known quartz content. Samples were ground and mixed as
described above. A second set of standards was prepared by spiking a quart~-free ore in the
same way the expanded perlite samples were prepared. All three sets of standards produced
identical calibration curves. Because of the finer and more homogeneous particle size, the
standards prepared with expanded perlite were adopted for general use.
XRD Analysis
All samples were analyzed using a Siemens D500 diffractometer with a SIEMAC
interface and Siemens-supplied software. The D500 was configured as shown in Table 2.
The samples were mounted in custom-made aluminum holders. The holders are
machined from 1/8" thick aluminum which is cut into square 1-31/32" on a side with a 1"
hole cut in the center. The upper face is lapped flat and care is taken not to round the edges
of the central hole. The special holders are used because their greater dimensional stability
and flatness compared to the normal plastic holder minimizes displacement errors and
produces better peak intensity precision.
The powder is back packed in the holder against a glass plate by hand. To assure
uniform packing, a brass piston which is .004" smaller than the sample hole is used to pack
the sample. The packed sample is backed by pressing in a cardboard disk which is secured
with masking tape. Because preferred orientation is not a problem but peak intensity is, the
samples should be packed as firmly as possible. Experience shows that with uniform
grinding, firm packing greatly improves the reproducibility of peak intensities.
All samples are manually loaded into the diffractometer rather than using the sample
changer. Manual loading insures more accurate sample positioning and improves the
precision of the results.
Each sample is scanned from 26.0° to 27.5° 29 with a step of .05° and a count time of
40 seconds per step. Integrated peak intensities corrected for background for the quartz
(101) peak at 26.65° 29 are extracted from the raw data using the program provided by
Siemens.
Calibration
Table 3. Calculated and Observed Precision of Peak Intensities (d counts) for Quartz-Perlite
Standards Based on 40 sec/O.05° 29
% Quartz
a) d c = (lOOJR-l){R(R+ l)/N} 1/2 where R =peak to background ratio and N =total peak
counts.
b) multiple runs on the same mounted sample.
c) sample repacked between runs.
oto 5% quartz. With proper background correction the curve will intersect the origin. Any
of the standards which consistently falloff the nominal curve should be reform ed.
RESULTS
The procedure developed has been tested on over 100 perlite samples. To evaluate the
sensitivity,samples were checked using a combination of heavy-mineral separation and optical
microscopy. This method has a very high sensitivity but a low precision. The checks
verified that quartz was consistently detected at the 0.1 % level as long as no peak overlap
from other phases occurred. Despite the persistent presence of trace amounts of mica and
feldspars in most perlite ores, peak overlap was not found to be a significant problem.
Originally two calibration curves were used, one for ore and one for expanded perlite.
Theoretically the two curves should be identical if the powder can be packed to the same
density in both cases. Based on observed intensities the grinding and packing techniques
used did produce samples with identical densities.
40s
8s
0.5s
Precision
The 40 sec/step counting time was selected as a compromise between precision and
analytical time. The desired detection limit and precision was obtained. Taking into account
the inhomogeneity of geologic materials it is difficult to justify any attempts to increase the
precision of the determination. With a 40 sec count time the statistical variation in the
background is small enough to permit the peak fmder to reliably locate the quartz peak at the
0.1 % level. Figure I illustrates the improvement in peak discrimination resulting from longer
counting times.
Accuracy
At this point the accuracy can only be inferred from the precision, the linearity of the
calibration curves and checks with synthetically prepared material and verification by a
combination of heavy mineral separation techniques. Only when well characterized material
is available for standards will a better assessment of the accuracy be possible.
The procedure works on both ore and expanded product and does not appear to be
sensitive to either the origin or nature of the ore or the variation in the expanding process.
Although the method was developed for a Siemens D500 diffraction system it should
be easily transferable to any system as long as the critical points outlined are followed. When
attempting to use the techniques with an older Philips system two problems were
encountered. First the holders with a rectangular sample cavity were impossible to pack
uniformly and second, without sample rotation the precision dropped considerably.
Whenever this method is employed care should be taken to document the precision and
accuracy for the particular system in use.
Longer counting times should improve the precision but at the expense of productivity.
Each sample now takes about 20 minutes to run. To reduce the statistical counting error by
50%, the run time would have to be increased to about two hours. This appears to be a
situation where a position sensitive detector would greatly improve the precision and the
detection limit.
THE CONCEPT OF PATHLENGTH DISTRIBUTIONS
INTRODUCTION
a)
b)
c) flo.
d)
e)
8 8.
I
15 . 811· to 2S . 811·
35. 811· to
.a . 1I38
.a . 828
.a . 818
.a.88II~~----~--------r-------'-------~--------'---
.a.8811 .a. 1811 .a .2811 .a . 3811 .a .4811 .a . 5811
,......u..UOft 1entrU! l_)
8 . 838 _
haUwldth :
8.811861 _
.8.838
8 . 8763
ion (ength [1'",,] I
8.1817 8.
.e.816
.e.812
+8.888
.e.884
+8.888 ~------,-------~------.------,------~-------r-------
.e.888 +8.388 .e.688 .e.988 +1.288 +1.588 +1.888
penetration length [NN]
.l/o
I
REFERENCES
1. J.W. Criss and L.S. Birks, Anal.Chem. 40, 1080 (1968)
2. J.V. Gilfrich and L.S. Birks, Anal.Chem. 40, 1077 (1968)
3. J. Sherman, Spectrochimica Acta 7, 283 (1955)
4. J. Sherman, Spectrochimica Acta 11, 466 (1959)
5. T. Shiraiwa and N. Fujino, Jap.J.Appl.Phys. 5, 886 (1966)
6. R.G. Gardner and A.R.Hawthorne, X-Ray Spectrometry 4, 138
(1975)
7. H. Ebel and Ch. Pohn, X-Ray Spectrometry 18, 101 (1989)
8. M.F. Ebel, G. Moser, H. Ebel, A. Jablonski and H.
Oppolzer, J.Electron Spectrosc.Relat.Phenomena 42, 61
(1987)
ACKNOWLEDGEMENT
The research program has been supported by "Fonds zur
Forderung der wissenschaftlichen Forschung in Osterreich".
Project P 7012-PHY
A VERSATILE FUNDAMENTAL ALPHAS PROGRAM FOR USE WITH EITHER TUBE OR
B. J. Cross
Kevex Instruments
San Carlos, CA 94070-1308
INTRODUCTION
m1
Fig. 1.
~ I s( A)
Geometry for Calculation of Scattering From Secondary Targets
L. FENG ET AL. 511
(2)
~(A)(l/sin~ + l/sinw)
In the same manner, we can obtain expressions for both coherently and
incoherently scattered tube characteristic line intensities as follows:
(3)
~(ACh)(l/sin~ + l/sinw)
(4)
(5)
and
Na [ a2 (l-COSIJ)2]
uincoh = A(Z) r!/2(l+a(1-coslJ)]-2 1+cos 2 1J + S(X,Z) (6)
1+a(l-cos/J )
where
/J = Scattering Angle
X = Momentum TransferParameter = Sin(IJ/2)/A
Z = AtomicNumber of Secondary Target
A(Z) = Atomic Weight as a function of Z
re = Classical Electron Radius = e 2 /mc 2 = 2.81794 x 10- 13 cm
a = hv/mo c 2 = E(keV)/511.006
e =
= Electron Charge; m, mo Electron Mass, Electron Rest Mass
c = Velocity of Light
Na = Avogadro's Number
F(X,Z) = Atomic Form Factor (Coherent Scattering Factor)
S(X,Z) = Incoherent Scattering Function
The terms F(X,Z), and S(X,Z) are tabulated by Hubbell et al. (7). We
applied a fourth degree log-log fit to the tabulated data to provide a
convenient way to calculate these quantities. The fitting equations are
as follows:
(7)
512 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
A B C
CDA360 61.68 34.85 62.18 +0.80 34.00 -2.44 61. 92 +0.40 34.47 -1.10
CDA464 60.22 39.00 60.52 +0.50 38.01 -2.50 60.22 0 38.67 -0.85
CDA482 58.55 40.10 59.01 +0.80 38.97 -2.80 58.66 +0.19 39.62 -1.20
CDA485 59.67 38.00 60.24 +0.96 37.12 -2.30 59.95 +0.47 37.71 -0.76
CDA675 58.46 39.70 58.98 +0.90 38.62 -2.70 58.68 +0.38 39.30 -1.01
*Data taken using Kevex microfluorescence spectrometer with Mo tube direct excitation
(50 kV, 1 rnA, 100 p aperture); standard used: CDA 836 (Cu 84.13%, Zn 5.78%).
A - Certified Value
B - ZnKp is not considered for enhancement of CoRa.
C - ZnKP is considered for enhancement of CoRa.
R.E. - Relative Error
and
S(x,Z) =Exp(a' + b' ·logX + c' ·log2 X + d' ·log3 X + e' ·log4 X)Z (8)
where a thru e, and a' thru e' are fitting constants for a certain atomic
number region. Over a range of X from 0.005 to 5, and an atomic number
range from 13 to 83, the average agreement between the tabulated and
calculated values is within ±5% relative. We are currently testing this
extended algorithm in the computer program for bulk analysis of selected
samples and will report these results in a future publication.
Ouantitation by Difference
(9)
where
where
oi j - aid
(13)
and
(15)
Rf
It can be seen from equations 12-15 that the terms associated with the
eliminated constituent are incorporated in ofj' ofjk' and Rf·
514 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
CONCLUDING REMARKS
REFERENCES
INTRODUCTION
In a first approach. the program calculates the total mass per unit
area riC i (mg'cm- 2 ) of the high-Z elements from the area of their charac-
teristic peaks in the spectrum. by applying the well known equation 1.
and considering that the absorption of the sample matrix is negligible
(t E ,=1) .
(2 )
Ri n c (3)
where: Reoh and Rine are the measured count rates of the coherently and
incoherently scattered radiation. teoh and tine are the absorption cor-
rection factors. (pO) is the sample mass per unit area (mg·cm- 2 ) and Seoh
and Sine the coherent and incoherent scattering factors (counts·s-'.
mA-'.mg-'.cm 2 ) . These are calculated as the weighted sum of the contribu-
tions for all the elements constituents of the sample:
a L.t Wt Zi
O El.1
Sine (5)
< Z >
bo
= [---- ------Rcoh 1 1/(a..1-b1)
(6)
ao Rinc:
Combining equations (6). (5) and (3). the sample mass per uni t area
is given by:
Rinc
pD (7)
IoGt i n c bo <Z>b1
EXPERIMENTAL
pDCweighing) pD(scatter)
The net peak areas of the characteristic lines were evaluated using
a non-linear least-square fitting procedure, implemented in the software
package AXIL-PC '2 . Scatter intensities were evaluated by a summation over
the channels within the range 18.80-19.80 keV for the incoherently scat-
tered Ka and 19.98-20.lfO keV for the coherently scattered Ka after sub-
tracting the background. This was calculated over the energy range of
15-21 keV by an algorithm that iteratively eliminates the channels
containing peak information '3 .
The agreement between the calculated and the certified values of the
elemental concentrations of the standards are generally very satisfac-
tory, usually within one standard deviation. The only systematic
F. ARAUJO ET AL. 519
deviations observed were the lower values for the low-Z elements (AI and
Si), probably because for such low atomic numbers the fundamental
parameters are less accurate and more dependent on the instrumental
specifications. Also, there are large uncertainties in the evaluation of
the characteristic lines, due to the high Bremsstrahlung in this spectral
region obtained with a direct excitation.
CONCLUSIONS
The total sample absorption due to both light (Z<12) and heavy ele-
ments is calculated by an iterative procedure that yields self absorption
corrections. It utilizes the coherent and incoherent X-ray scatter peaks
to calculate an average atomic number <Z>, representative of the bulk
light element sample composition.
REFERENCES
I. INTRODUCTION
In the spectroscopic analysis of composite elements by x-
ray fluorescence (XRF) , it is the fundamental feature of this
method that the background x-ray intensity is lower than that
with electron excitation. However, the background x-rays of
this method, which consist of Thomson (Rayleigh) and Compton
scattered x-rays from the primary radiation, impair the
analytical performance at the ppm level. In order to
investigate the intensity of the background x-rays precisely,
the study was conducted in two parts. The first part compared
the measured and theoretically calculated x-ray intensities for
Rh Ka and Rh KB peaks from various materials. The second part
examined the determination of low concentrations of lead,
arsenic and selenium in steel samples. The variation in the
background x-ray intensities of the analyzed elements was found
to be caused by the variation of the major elements and a
correction equation for it is derived.
II. OUTLINE OF CALCULATION OF TOTAL ATOMIC SCATTERING FACTOR
The formula that was used for calculating intensities was
derived for the x-ray intensity based upon the estimation of
Thomson and Compton scattered x-rays of an atom in previous
papers.' The formula is the product of the intensity
distribution term of the primary radiation (I), the combined
term of scattering and absorption for an atom, and the
instrumental term, in which a main part of this term is the
reflection of the analyzing crystal (c). The scattering term
in the formula is calculated with respect to the individual x-
rays, which consist of scattering terms in the numerator and
absorption terms in the denominator. The product of I and C
is numerically a constant for each x-ray for simplification.
From the calculated results the characteristic items shown
in Fig. 1 can be derived as follows: In each case of Thomson
I Absorption effects
olow Z - high intensity
o high Z - low Intensity $CoHered
I
Fl Material condition
o Crystalline
o Amorphous
o Liquid
A: Absapllon onenuallon
8
3070 -Ci.-17.633
LiF(2C01 cleawJ. .~
Fine soller slit u
50KV SOmA Rh 1
I
J:
0: Pb-L seri
I
J:
aE
c
UF10167
>. 8
...
0
I
~,
X E7
~J:
... 0:
-:2-
, AI 5.105
J:
0:
Pb
3.569
10 20 30
r-~
3.5
VI
Q.
u Rh-Kp, Rh-Kp.
oX
1.0 Rh-Kp, 2D
~
..
VI
C
C
-;, 0.5
~
I Compton .Conllton
X
Caluculated ThOG! 11.60 2.15 2.01 Lilli 1.162 2.00 - 1.82 x 10-'
intensity Ka COGIp 27.8 8.01 5.53 0.794 0.580 0.197 - 0.lA6
COGIp/Tholll 6.011 3.73 2.75 0.551 O.A99 0.099 - 0.080
Measured peak ThOCll 7.95 11.23 3.79 2.63 2.11 3.69 3.68 3.A2
intensity Ka COGIp 15.6 3.65 2.9A 0.372 0.25A 0.116 0.065 0.075
(kcps) COGIp/ThOCll 1.96 0.863 0.776 O.llil 0.120 0.031 0.018 0.022
2, the intensities from Pb, AI, and LiF powder samples are
shown. In the case of a LiF powder sample, both the Thomson
and Compton scattered x-ray intensities are higher than those
from Al and Pb metals and the Compton x-ray intensity is higher
than that of the Thomson.
with the Aluminum metal sample the intensities of the two
scattered x-rays are roughly equal. with the lead metal sample
only the Thomson x-ray intensity can be seen clearly; the
Compton x-ray intensity is too low.
In Fig. 3 Thomson and Compton x-rays of Rh-Ka and Rh-KB
radiation scattered from Pb, Au and W metal specimens are
shown. It is clearly seen that the Compton x-ray intensity is
very low. Table 1 shows the calculated and measured
intensities for some typical materials.
I
RhK«
IQ LIF
/
'"ua.
.x
51
PI> ·AI
MY·
. ·~r
~
iii
.,c: ~I• ••
F.
"'
RhKpl
5
/
CU
.,o
.x
0 L·C:I
a.
.,
/
:b:'-.t:
t;;.. .~"11
."
Ni 51
Fe
~
/
o
'"
::!
.
LIF
~ 10~--------r-------~~~----~
a.
<.J
.x
PI>
/
0.0 I L,-,-:------=-=1::-,------.:-I~----~.
0.000 I 0.00 I 0.0 I 0.1
Calculated intensity
Figure 5. Comparison between calculated and measured
intensity of Compton scattered X-rays of Rh-Ka and Rh-KB
using polycrystalline materials
3.0 W__________~-+---H--~03070
°50KV SOmA
035-15 Rh
h
~}fMO
III
Co
u
oX
2.0
~
..
III
c:
.s
,.,
e
I • Mo
X
1.0
QO~2~~~--~---L~~~----~--~40~0~----
29 UF(200)
Figure 6. Background X-ray intensity from binary alloys
of iron
Element Mo W Cr Hn f"e(NBS1265) Co N1
Spec1men
!ItS 7.07 19.8 30.3 9.9 Pure 20.0 79.b
Heas. 1ntens1ty (kcps) 1.261 1.220 1.112 1.08b 1.086 1.072 1.001
~ ""
I
,
X
~ 29 Content
wtOfo • (01 2 + bI +c ) ( 1+ LdljWj ) - I:lljWj + I: mijWj
i matrix over~ background change
Figure 7. Definition of correction factors for
background intensity change
Interfering elements
Ni + 0.cxx)!3/0.cxx> 19 a 0::033/0.0::032 ot:rot:Jl:T 1 0
Co + /0.0::0091 /0.00015 IClCXXX)6
Mn - 0DJ)15/ 0 0.0::031/0.0::013 0.0::0018/
Cr - 0.00016/000012 O.ax>:o /0.00032 0.CXXlO31/0.cxxxl37
Ti - 0.0X24/ 0 0.0::067/0.0023 OJ:OX)52/000042
Nb /0.0018 /0.0053 10.r0::J38
Mo - 0.0010/0.0018 0.0031 /0.0054 0.Q)026 / CCJ:X:S)
* overlapping cal./mea.
528 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
1264
1.20
.2
'§ 1.15
1.10
~
iii 1161
c 1.08
.s 1.10
II
'1162 1161
1.06
~
'-
I .'155
X
1.05 1155
'1161
• ·,263
1.04
-
1261
1.Q2 1117.'11"
1263
, gradienl ~ 1,ItA
,, 1.00, I I • I ' I •
12" 0002 QOO4 0.006
I I
001 002 003
Pb wI OJ.
Figure 8, Relationship between X-ray intensity and
chemical content in low alloy steel
IV. CONCLUSIONS
The basic concept for the intensity and distribution of
background x-rays has been clarified using Rh-Ka and Rh-KB x-
rays. The intensity ratio of Thomson versus Compton x-rays
which are scattered from analyzed samples is dependent upon the
atomic number and the absorption of x-rays. It was also proven
that the intensities and distributions can be predicted.
In practical application for trace element analysis in steel
samples, the intensity change of background x-rays has been
studied and a correction formula has been proposed.
V. REFERENCES
1. T. Arai, Adv. X-ray Anal., 31 (1988), pp. 507-514 and T.
Arai, Adv. X-ray Anal., ~ (1989), pp. 83-87
AN ARTIFICIAL INTELLIGENCE SYSTEM FOR XRF DATA
ON A PERSONAL COMPUTER
SUMMARY
INIRODUCTION
The value of good results depends upon the accurate and timely transfer of this
information to a customer in a format that the customer can assimilate readily.
This is a problem with many analytical instruments, including XRF systems. For
example, currently available XRF systems incorporate a high degree of automation.
The automation has focussed on instrument control and mathem:atical manipulation
of raw data. Such systems will therefore produce good results for well designed
experimental methods. However, the tools that allow the chemist to convey results
effectively are sadly lacking.
The commonality of this problem across a wide range of analytical instruments led
us to develop a consistent, modular system that could be adapted easily to convey a
variety of information from different analytical chemistry instruments. This system
has been adopted for a wide range of analytical instruments within Chevron
Research Company. This paper discusses the first implementation of this system, to
XRF. Since then, the system has been applied successfully to mass spectrometry,
thermal analysis, ICP and M spectrometry, and small chemical analyzers (e.g.,
sulfur analyzers, CHN analyzers, ... ).
A number of factors drove the design of this system. These factors all relate to the
customer need for receiving accurate, readily interpretable information. These
factors are:
• Quality Control
• was this the appropriate test?
• are the results out of range (above or below method reporting
limit)?
• were there significant interferences?
• were there other boundary problems (e.g., mass balance,
concentration of an added element)?
• Transcription errors
• did the analyst transcribe the results correctly?
• did the customer transcribe the results correctly?
• Costs
• transcription costs.
• rework due to transcription or QC errors.
• cost of having "bad data" in customer data base.
• Commercial systems deficiencies
• limited implementation of "multi-tasking."
• limited, if any, control over reporting formats.
• limited capabilities for data archival and retrieval.
In addition, any system designed to correct these deficiencies should also make it
easy for the analyst to implement and audit the process.
EXPERIMENTAL
The PC forms the central hub for our in-lab computing. It is connected to both a
Philips PW-1400 XRF and a Philips PW-1404 XRF through their PDP-ll based data
systems. It is further connected to a variety of customer systems via an in-house
voice-over-data network. A switch box selects the appropriate connection.
E. P. WITTIG AND C. E. RECHSTEINER 533
Table 1
List of Commercial Software and uses in PC system
The system blends commercial software with our modular software. A menu
system which clearly states each task in a logical sequence assists the user. The
commercial programs and their uses are listed in Table 1.
SYSfEM
The system was designed by examining the steps that happen between sample
receipt and data reporting. The transformation of data into the final form for the
customer is a critical step. The key steps of this process are outlined in Figure 1.
The data transfer is handled readily with command files on the instrument's data
system that are initiated from the terminal emulation program. Data archival is
important for Good Laboratory Practices, environmental work, and patent positions.
In the archival, the data is preserved in the original format and can be retrieved
readily.
Figure 1
Functional Flow Chart of Program
534 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
Table 2
Key method requirements
Valid JD
Valid Analyst
Method's QC is in control
All elements are within the appropriate concentration range
Interferences are not significant
Mass balance is in control
Intensities from added elements are in range (e.g., internal standard)
The data manipulation is the heart of the program. Key information, listed in Table
2, is checked. The data is compared to a rule-based system for checking the data
validity. Checks include things such as the mass balance on catalysts, the presence
of known interferences that may cause significant matrix effects, and validating that
the concentrations of the elements are within the range that the method can
handle. For some samples, dilutions are required to bring the instrument
concentrations into the method's range. Dilution data can be entered and the
appropriate concentrations determined.
All of the information about the test method was programmed to form the decision-
making engine. If, after analysis, the test is inappropriate, the system suggests
alternative analytical tests.
The data manipulation program creates output files which contain the data in the
appropriate format. This data is transmitted readily to other computers with the
same terminal emulation program used to gather the data from the instruments.
RESULTS
The overall system design includes the internal validation of data, reformatting, and
transmitting data. The system was running within 3 weeks after detailed
specifications were written. The data manipUlation program currently runs methods
with 25 results per sample. The upper limit is controlled by the array size which is
set at 200.
Some of the lab's tests are quite simple -- only one result being determined. For
example, the determination of low levels of Pb in gasoline closely follows ASTM
03229. In our implementation, the statistics are good enough to go down to 0.005
g/gal and the upper limit is 0.50 g/gal. The instrument output is shown in Figure
2. The checks done on the method verify that the data is within the method's
upper and lower concentration limits. If the result is less than the lower limit, it is
E. P. WITTIG AND C. E. RECHSTEINER 535
PB
0.011
Figure 2
Instrument Output for Ph in Gasoline
reported <0.005 glgal. If the result is within the range, it is reported to the third
place beyond the decimal point. If the result is greater the upper limit, it is
reported >0.500 glgal and the test method for higher level lead samples is
suggested, in this case ASTM D2599 Method B. The customer's computer to which
the results are normally transmitted expects a special packed format. Figure 3
shows the format for a series of samples and identifies the various fields. Each
field is separated by an asterisk and the last field can accept a comment if
appropriate. If the result is too high an alternative test is recommended. All such
recommendations are printed in an exception log.
Another example is the lubricating oil method (1) in which a number of items are
checked or computed. In addition to checking the concentration for each element
as in Pb in gas method, the program computes the dilution factor (if used) and
checks several additional elements for possible interference. For example, calcium
sulfonate is a frequently used additive with no Zn or P. If the additive is diluted
into the appropriate concentration range, the Ca and S concentrations can readily
be recovered by multiplying the results by the dilution factor. For reporting P and
Zn, the appropriate result is the lower limit of the method multiplied by the
dilution factor. If, for example, the dilution is insufficient and the Ca level is too
high, the data system will report the Ca as the dilution factor times the upper limit
and suggest a larger dilution. It will also report the other elemental results as not
available. They are meaningless because the matrix corrections cannot be applied
accurately. Included in the method are checks for other elements found in some
formulations for lubricating oils but not handled in this method. If the upper limit
of the apparent concentration of these interferences exceed a predetermined level,
then the results for Ca, P, S, and Zn will be biased and should not be reported by
this method. The data system reports which element was interfering and prints an
exception report which includes a recommended alternative method.
6101290*1618*.011 *G/GAL*404
6101454*1618*.399*G/GAL*404
6101344*1618*<.OOS*G/GAL*404
Figure 3
Compressed format for a Single element analysis
As a class, catalyst samples have their own particulars. Most of the hydroprocessing
and catalytic cracking catalysts are prepared with a lithium borate fusion. A check
on the mass balance provides a sample preparation check and also a check for
interfering elements not analyzed for in the method. Additionally, the amount of
non-wetting agent retained in the sample influences the total mass absorption and
needs to be within certain limits so as not to affect adversely the quality of the
results. Both of these checks operate similarly to the interfering elements check in
the lubricating oil method. For example, the mass balance check for fluid cracking
catalysts is 98-102 percent and samples outside that range are flagged.
The data system has been used for about a year and a half for over 19,000 samples
and pays for itself about every two months. It is also the model for other similar
systems -- mass spectrometry, ICP, MS, UV, Leco S, and Carlo Erba CHN. In
addition, it is being added to the 6 thermal analysis instruments, 3 viscosity
instruments, sorption-desorption instruments, HPLC and ICP-MS. Currently about
40% of the division's samples (>4000 samples/month) are transformed and
reported electronically though these systems.
CONCLUSIONS
A data handling system for XRF is reported. The system checks the data internally,
reformats, and sends it on to the client. A low-cost PC runs the user-friendly
system. It reduces the time spent transcribing data, reduces the transcription
errors, improves reporting consistency, and reduces rework for results reported
incorrectly.
The system is easily transferred to other instruments and methods because of its
modular design of commercial software and structured data manipulation program.
REFERENCES
ANALYTICAL LABORATORY?
P. B. DeGroot
ABSTRACT
Statistical Process Control (SPC) methods have become extremely popular in quality control and
process improvement in the manufacturing environment, first in Japan and more recently in the U.S. They
are gradually being introduced into the research and development setting as well, where they offer some
advantages over more traditional statistical approaches. The greatest advantage of the SPC approach is that it
ensures that the statistical description generated is valid, i.e., that the data on which it is based incorporate
only random error. This is verified both initially and on a continuing basis. This means that the results
provide valid predictions of future performance, not just a description of an historical set of replicate samples.
SPC statistical evaluations of analytical results and instrument performance require only simple calculations
and are .easily applied in a consistent, valid manner by statistically unsophisticated analysts. Besides the
numerical statistical analysis tools, the SPC approach provides other problem-solving techniques for
improvement of the entire analysis process. These aspects of the SPC approach will be described and
illustrated with examples from our x-ray fluorescence analysis laboratory.
INTRODUCTION
Statistical Process Control, or SPC (sometimes also called Statistical Quality Control, SQC), is a
methodology developed out of conventional statistical principles. Unlike conventional statistical methods
which primarily aim to describe the capabilities (accuracy and precision) of measurement, SPC has as its
objectives both the statistical description of the measurement and the improvement of the measurements
and the process that generates them.
SPC was pioneered by the American statistician Deming! in the 1950's, who found a receptive climate
for his ideas in Japan as they tried to rebuild their industrial capacity after the war. Ironically, primarily
because of competition from Japan, these ideas recently have been reintroduced into the United States. They
have been widely adopted in manufacturing with great success. Acceptance in the research and development
area has been much slower, in part, perhaps, because of the previous emphasis on manufacturing and the
name "Process Control", which has a decidedly engineering connotation.
I will try to show that SPC has some distinct advantages in the R&D setting as well, first by
discussing an area already familiar to analytical chemists; the statistical description of measurement
capabilities. Then I will briefly introduce the other (perhaps more powerful) aspect of the SPC philosophy;
the improvement of measurements and measurement processes.
DISCUSSION
The greatest advantage of SPC in evaluation of measurement capabilities is its emphasis on continuous
monitoring of measurement capability. This results in greatly increased predictive value of SPC results
compared to the sort of "one-shot" statistical analyses to which analytical measurements are often subjected.
Some significant secondary benefits are also incorporated into the SPC approach. Since SPC methods and
calculations were designed to be used on the factory floor by the workers themselves, they are easy to apply
in a consistent manner by mathematically and statistically unsophisticated analysts. They are generally also
robust in the statistical sense. That is, the successful application of SPC principles does not depend on the
measurements in question having an approximately normal or Gaussian distribution. Finally, there is a great
emphasis on graphical presentation of results, which are much easier to interpret than numerical results.
The basic tool for measurement evaluation in SPC is the control chart. This document is a continuous
record of the performance of an analytical (or other) process, along with notations of external events and
conditions which mayor may not affect this process. A typical control chart is initiated by running 20 or
more replicate analyses of one or more standards or reference materials. These initial measurements are then
converted by means of simplified formulae 2,3 into a mean and upper- and lower control limits., as shown in
Figure 1. The initial measurements are represented by open circles.
Thus far, the SPC approach is quite similar to what one might commonly see in the literature as a
statistical evaluation of a newly-developed analysis, except that often the conventional evaluation is based on
far fewer replicates, and the results are usually presented in numerical rather than graphical form. The mean,
upper and lower control limits (UCL and LCL) are entirely analogous to the mean and ±3·Sigma limits of
conventional statistics.
Here the similarity usually ends. The next step in the SPC approach is to determine whether the initial
standardization data are, in fact, statistically valid (or "In Control" in the parlance of SPC). That is, whether
1.04,-----------------------------,
Upper Control Limit
.•
0 0
•
1.02
.,
0
"#. 0
• 5>
.., 0 0
c- 0
- •., • •
0 0 0 0 0
'';::; 00 0 000
~
n OO
uu v v 00
•• • •
c 0 Q)0
Q) 0 000 0 Q)o
(.) Q)
c 0 0
0 00
••
0 0 0 0
u
0
•
0
0.98 0 0 0 0
5 10 15 20 25 30 4 9 14 19 24 29
MAY JUNE
Figure 1. An example of a control chart for x-ray fluorescence analysis results. • - initial data used to
establish centerline and control limits, 0 - continuing monitoring of analytical process.
P. B. DeGROOT (1 ) 539
*There are additional, more complicated tests for the Out of Control condition, but these
three simplest rules will detect nearly all real problems and exclude spurious ones.
it is probable that no systematic, non-random errors are incorporated in this crucial initial data. This is easy
to do in SPC, by means which we shall learn shortly. It is, of course, also possible to make this evaluation
by appropriate conventional statistical methods, but I have almost never seen it done in the analytical
literature, regardless of the fact that it is ~ absolutely necessary first step in establishing the predictive
value of the statistical analysis. Despite this, such incomplete conventional statistical analyses are almost
universally used to predict future measurement capability.
Although this initial determination of the "In Control" (no non-random errors) nature of the analytical
process is necessary, it is not sufficient. One must assure that the analysis stays In Control. This is often
not done at all in the usual statistical treatment of a chemical analysis. The initial statistics are assumed to
prevail forever. At best, precision and accuracy are checked at arbitrary intervals with a few replicate
analyses. In Spc, this assurance is obtained by continuing to plot periodic standard or reference analysis
results on the Control Chart. The filled circles in figure 1 illustrate the continuing monitoring of the
results. Inspection of these results and comparison to a few simple criteria (Table 1) readily determine
whether the process is still In Control.
A very important additional benefit of the Control Chart is the elimination of "Wild Goose Chases".
This is the needless expenditure of corrective effort on perceived problems, when they are in fact part of the
natural statistical fluctuation in the results. Without the perspective of the control chart in figure 2, for
example, the sharp drop in the Si x-ray fluorescence intensity of a reference material on the morning after our
XRF spectrometer was subject to a power failure might initiate a needless search for an instrument problem.
3.05~--------------------------------------------~
en
••• •••
•• ••
0.
U
:..: 2.95
i
.u;
•• • • ••
• • •
c:
Q)
.E
•
<tl 2.9
:..::
(j)
----------------------------------------------------------~------
.....
2.85
Lower Control Limit :::J
0
Q;
~
0
0-
2.8 -t--,---,--r--,----r--r--r--r---,--,---,--,---,----r--r--,--J
12 14 16 18 20 22 24 26 28 30 1 3 5 7 9 11 13
JANUARY FEBRUARY
1988
Figure 2. Control chart to monitor general performance of an x-ray fluorescence spectrometer. Si K-alpha
and other intensity measurements are made on a special reference material every morning.
540 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
This brings up the importance of the notation of external events on the Control Chart. For papers and
presentations, a neat computer-generated representation such as figure 1 or 2 may be appropriate. For really
useful day-to-day SPC analysis, a hand-written chart with voluminous notes on external conditions is far
preferable. This ensures that important observations won't be missed because it is not convenient to log on
to the computer and run the software (most of which doesn't have provision for extensive note-taking
anyway). It also promotes the active participation of the analyst in the maintenance and improvement of the
quality of the results. Quality becomes more than just something that the computer or the supervisor does
with mechanically generated numbers.
Why are these Control Chart marginalia important, anyway? They are the beginnings of the second,
more powerful, aspect of SPC, the potential for correcting problems and improving the process. Some
systematic SPC tools for doing this will be introduced briefly in the next section, but often improvements
can be made or problems solved simply by correlation of external observations with Out of Control
conditions on the Control Chart.
One other benefit often claimed for the SPC approach to monitoring process performance is the
elimination of needless instrument recalibration, which can actually increase the scatter in analytical results.
This is the subject of a separate paper4. It turns out in practice to be a relatively minor consideration.
Process Flow Charts. An example of this formal analysis of the steps of a process or sub-process is shown
in figure 3 for a sample preparation process for XRF analysis. Preparation of a formal process flow chart on
even a simple operation that one has taken for granted for a long time can be a real revelation. It can be very
helpful in suggesting steps in the process where the two other procedures described below can be used to
pinpoint a problem more exactly. Typically, every step in the overall process can be further resolved into
process flow charts of finer steps, to whatever level of detail seems to be needed.
CauselEffect CFishbone) Diagrams. Figure 4 shows a portion of such a systematic listing of effects
pertaining to a potential problem or cause of error, and the postulated possible causes. It is also sometimes
STRRT
Return to
~""T""'''ISubmitter
Select Window
Material
FRIL
OK
Ready for
Analysis
Figure 3. An example of a process flow chart for acceptance and preparation of samples for XRF analysis.
P. B. DeGROOT (1) 541
IRREPRODUCIBLE
PRESSED POWDER
PLANCHETS
Figure 4. An example of a cause and effect diagram applied to an XRF sample preparation process.
called a Fault Tree. The actual cause is usually pinpointed by collecting appropriate data and displaying it on
a Pareto Chart.
Pareto Charts. This is nothing more than a simple bar chart or histogram of the frequency of occurrence of
events (usually problems) in a process. This tells you where to concentrate further efforts at improvement to
have the biggest impact.
Although these process improvement tools have been treated very briefly, they are probably more important
than the Control Chart, which primarily serves to assure consistent, acceptable performance. The biggest
pay-off in any competitive situation is not generally from maintaining the status quo, but from improving
upon it. There are also a number of different types of control charts which have not been dealt with here.
Some of them are more effective than others in dealing with certain kinds of data, or in detecting subtle
changes such as slow drift in the data. I recommend further investigation of these SPC tools (2,3).
CONCLUSIONS
Although the application of SPC techniques to analytical procedures requires somewhat more effort than is
usually expended on the assay of the capabilities of the measurement process, it is worth it. SPC methods
assure valid predictions of the method's accuracy and precision and provide tools for the correction of
problems and the improvement of the method.
REFERENCES
1. Deming, W. E., Ouality. Productivity and Competitive Position, M. I. T. Press, Cambridge, MA,
(1982)
2. Wheeler, D. J. and D. S. Chambers, Understanding Statistical Process Control. Statistical Process
Controls, Inc., Knoxville, IN, (1986)
542 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
3. AT&T Technologies, Statistical Ouality Control Handbook_ AT&T Technologies, Indianapolis, IN,
(1958)
4. DeGroot, P. B., "SPC Analysis of Optimal Strategies for Restandardization of X-ray Fluorescence
Analyses", Adv. in X-ray Anal. 33:543 (1990)
SPC ANALYSIS OF OPTIMAL STRATEGIES FOR REST ANDARDIZA TION OF
P. B. DeGroot
ABSTRACT
Quantitative XRF analyses almost always depend on measurements of x-ray emission intensities of
standards of known composition. The usual approach is a careful initial standardization with multiple
standards that are checked for accuracy, stability, homogeneity, and internal consistency. This is followed by
limited periodic restandardization or recalibration with one or two of the initial standards or a reference
sample. The frequency and protocol of these periodic restandardizations tend to be rather arbitrary; for
example adjustment on the basis of the average of a triplicate measurement of a standard at the beginning of
the work day. There are certainly a large number of other possible restandardization strategies. They range
all they way from no restandardization at all, through single or n replicate restandardization measurements
before analysis of each group of m unknowns, to cumulative or moving averages, again with variable
replicate size, frequency, and now averaging range. This paper examines the outcomes of some of these
restandardization strategies applied to synthesized, realistic XRF analysis data which incorporate both abrupt
and gradual changes in instrument response. SPC (Statistical Process Control) control charts are shown to
be a very useful tool for determining optimum restandardization strategies. Optimal strategies (least scatter
in calculated concentrations with fewest standards ) which emerged from this study included not
restandardizing at all except when indicated by an SPC chart of a reference standard intensity run concurrently
with each unknown, and a moving range of 3, with standards run every 5 unknowns, coupled with SPC
monitoring of the standards used for this restandardization.
INTRODUCTION
Chemical analyses are usually carried out by initially calibrating or standardizing the analytical
instrument or procedure against materials of known composition, then periodically restandardizing against
one or more of these standards. The recent emphasis on statistical quality-control methods has lead to this
consideration of the optimum procedures for restandardization. These procedures were tested by statistically
evaluating the results of simulated XRF analyses using computer-generated normally-distributed standard and
unknown intensities.
Several possible restandardization strategies are considered, the simplest being no restandardization at all.
More complicated strategies include restandardization with n replicate standards before every m unknowns
and moving averages of the last n standards run individually before every m unknowns with the values of n
and m according to Table 1. As recommended by many SPC advocates! restandardization only when a
reference sample shows the analytical process to be out of control is also examined, with the reference
sample being run before every m unknowns as also designated in Table 1.
EXPERIMENTAL METHODS
Simulated XRF analysis data were genemted by a BASIC computer program. This data consisted of
normally-distributed standard and unknown intensity data with the following characteristics:
Modified sets of standard and unknown intensities simulating non-random instrumental malfunctions or
unplanned changes in analytical conditions were then generated from these sets of standard and unknown
intensities by superimposing a step function of 2,4, or 6 times the standard deviation of the intensity in
question, or a linear drift beginning at zero for sample 1 and ending at 2 or 4 sigma at standard or unknown
100. Seveml sets of 20 unmodified standard intensities were also genemted and used for initial calibration of
the method and restandardization according to the SPC protocol discussed later.
The relationship between the standard and unknown intensities and concentrations was assumed to be
linear with a zero intercept: that is, the unknown concentration is unknown intensity divided by standard
intensity times standard concentration. The unknown concentration thus calculated should avemge 1.00 from
the synthesized data.
It is instructive initially to examine the advantages and disadvantages of two fairly extreme cases: no
restandardization at all, and restandardization with a single standard before every unknown. The results of
calculations for the concentrations of 100 unknown measurements by each of these strategies are shown in
figure 1. Note that restandardization with a single standard before every unknown actually increases the
variability of the calculated concentrations. This increase in scatter due to needless restandardization is often
stressed in the SPC approach. This additional scatter can be reduced by restandardizing with replicate
standards, and by restandardizing less frequently, as shown in figure 2. Of course, restandardizing with
replicates greatly increases the amount of work that has to be done. On the other hand, restandardizing at less
frequent intervals reduces this work (at the risk of letting a few bad analyses slip by between
restandardizations). Another mechanism for reducing the number standards that must be run, while retaining
P. B. DeGROOT (2) 545
1.04
1.02
c
o
'.p
~
+"
C
Q)
()
c
o
u
0.98
~- ----- . --------------------------------8 ---------------. ----_.
•• •
0.96 --'-------,-----.----,--------,--------,
20 4060 80 100
Analysis Number
Figure 1. Comparison of scatter in calculated concentrations with no restandardization ( [] ) vs
restandardization before every unknown ( .). Control limits are calculated from initial twenty
points with no restandardization.
1.5 J
1.4 13.~~~~!'!g!!~g!?!l.~i~~. _~Y_~~Y. _I,J_~!<,~~~!,!_ .. __ .
Rest~ndardjzatio~ Every 5th 9nkn~~
E 1.3 Restandardization Every 10th Unknown
OJ
U5
Q) 1.2
>
'';=
<0
~ 1.1 \-.. '.
....... .
~------------~~ ... -- .. -- .. -.... --- . -.... ----- .. ------------ .. -- ...
I I I I I I I I I I I
o 2 4 6 8 10 12 14 16 18 20
Number of Replicate Standards
Figure 2. Relationship between replication of standards, frequency of restandardization, and the net scatter in
the calculated concentration.
the advantages of decreasing scatter by averaging multiple standards, is to use a moving average of the last n
standards. This does introduce some correlation of results over time, however.
No restandardization at all has the advantage of lowest scatter in the result of interest and certainly the
least work. However, this is true only if the analysis is well-behaved, and few analysts are willing to take
this on faith. This means some additional work to determine under what conditions no restandardization is
justified, and how to restandardize when it is not. The approach suggested by most SPC advocates is to run
546 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
one or more reference samples (either a standard of known concentration or simply a stable material of
unchanging intensity), and restandardize only when this intensity indicates that the analytical process is out
of control according to SPC criteria.
The problem, then, reduces to just how much benefit (accurate concentration results with minimum
scatter) can be obtained at the least cost (number of standards or reference materials that must be run) under
conditions in which non-random perturbations to the analysis may be occurring. Before attempting to
quantify this benefit/cost relationship, however, a coarse screening of the strategies to eliminate clearly
unacceptable ones seemed in order.
The initial evaluation of the various combinations of the strategies according to Table 1 with the non-
random perturbations introduced above revealed several strategies which had severe problems. These
unsuccessful strategies produced excessive numbers (defined as 4 or more) of calculated unknown
concentrations outside the 3-sigma limit of the original unperturbed data. An example of such a failure using
the replicate-standards strategy is given in figure 3. In this figure, as in all other similar figures in this
paper, the data are presented in the manner of an SPC control chart, but the control limits are those of the
original. unperturbed data. This shows clearly where and by how much the various strategies deviate from
this best-case situation.
The pure SPC restandardization strategy also produced similar successes and failures. Standard SPC
criteria were used to determine at which point the analysis was out of control. Restandardization. again with
20 standards as initially, was done at this point. In general, failure of this strategy was due to the out-of-
control decision coming too late unless a reference sample was run before every unknown.
With the SPC strategy, a drift rather than a step perturbation caused the most difficulty. In the other
strategies, the greatest difficulty was generally with the large step discontinuity.
To further evaluate the remaining strategies which dealt more or less successfully with the simulated
analytical problems, a benefit-to-cost ratio was defined. Benefit was defined as the reciprocal of the standard
deviation of the 100 calculated values of concentration averaged over the results for 2 and 4 sigma step and
drift perturbations. Cost was defined as the number of standards that must be run, including the initial 20,
per 100 unknowns.
...••
1.06
.
1.04
..... ...
c -------------ii ----------------,-------. ----. ----------------.
..
... .
0
',p 1.02
• •
.. .
~ ~
...,
+'
C
Q)
(,)
c ~.-~-- ~----=--.----~~~~
0
U
0.98
------- -~-~- -- -- -- -- ------ -- -- -- -- ---- -- -~- ---------------------_.
0.96
2040 60 80 100
Analysis Number
Figure 3. An example of an unsuccessful restandardization strategy: Triplicate standards run before every
tenth unknown, data incorporate 4 sigma step discontinuity.
P. B. DeGROOT (2) 547
Moving Range 3
Every 5th Unknown
Single Standard
Every 5th Unknown
Triplicate Standards
Every 5th Unknown
Moving Range 3
Every Unknown
Triplicate Standards
Every Unknown
Figure 4 shows a bar graph of the benefit/cost ratio of some of the more promising strategies. The
most cost-effective strategy appears to be the moving range of 3 standards with one standard run every 5
unknowns. This is followed closely by a single standard restandardization every 5 unknowns, which has the
same cost in number of standards which must be run, but has slightly more scatter to the calculated
concentrations. The least cost-effective group, at about half to one-third the benefit/cost ratio of the two best
strategies includes the only "pure" SPC strategy that did not fail one or more of the step or drift simulations.
This is the strategy in which a reference sample is run before every unknown. This generates the lowest
scatter in the concentration results, but has a high cost in reference measurements. Running replicate SPC
reference standards did not noticeably improve the performance of the strategy, and, of course, increased the
cost still further.
Combined strategies
While the moving range of 3, standard every 5 unknowns seems the most attractive, there is a danger in
using it if the rate of change of a step or drift in the data is even a little larger than the maximum of 4 sigma
investigated thus far. In fact, a step of six sigma produces very unacceptable results.
This suggested combining the SPC strategy with the moving average as an optimal strategy. The
moving range is very good at compensating for small drift or step functions which are very difficult to detect
by SPC methods unless reference materials are run very frequently. On the other hand, SPC charts readily
detect large fluctuations in the data which cause problems with the moving-range strategy. This was, in fact,
quite successful, as figure 5 shows. The net cost of the combined moving-range-SPC strategy is only
slightly greater than the moving-range strategy alone. The moving range-standards can also serve as the SPC
reference intensities, and additional standards need be run only when SPC restandardization is necessary.
Incorporating an SPC control chart into the strategy has an additional benefit over relying only on
periodic restandardization. The SPC approach tells you when something is going wrong with your analysis,
while periodic restandardization merely compensates for whatever is occurring. It is much better, in my
opinion, to strive to operate with the best analytical technique rather than relying on compensation after the
fact.
548 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
....>-
'iii
cQ)
192
•
• • •••••
•
.. -•••• . -.
188
....
.E 184
...:
Q)
180
a:
----------------------_.- --------------------------------.
U 176
a..
en 172
SPC
1.08
RECALl8RATION
c 1.06 ~
•
0
.." 1.04
....C~ ----------------------------iii --------------------------.
-"'I.rI.';;" ,,~wt-
1.02
Q)
0
C
0
U 0.98
------------------------------------.-----------------_.
0.96
o 20 40 60 80 100
Analysis Number
Figure 5. A strategy combining the moving range method with the SPC approach is successful in dealing
with data incorporating a 6 sigma step perturbation.
CONCLUSIONS
The optimal restandardization strategy depends to some extent on the requirements imposed on the
analysis. If there is some leeway between the precision and accuracy capabilities of the analysis and outside
constraints such as specifications or the onset of undesirable effects in a product, then either a single standard
run about every 5th unknown, or the moving average of 3, with a standard run about every 5th unknown
combined with an SPC chart of these standard intensities is optimal. This allows a few analyses outside the
3 sigma limit of the unperturbed analysis to slip through if a perturbation in the analysis such as a step
change or continual drift occurs. The number of unknowns before repeating a standard measurement can be
adjusted to allow for the worst-case number of "bad" analyses one is willing to tolerate in case of a step
change.
On the other hand, if specifications or other outside constraints are very close to the capabilities of the
analysis, then the extra work in running a pure SPC strategy with reference standards run before every
unknown, or at least very frequently, is recommended. This will yield the least scatter in the results, and the
greatest assurance of being within the external constraints. Further, the SPC approach can lead to insights
into the analytical process which may allow an improvement in the precision of the process which will give
more leeway between the process capabilities and the external constraints.
In either case, including the SPC approach is recommended because it can tell the analyst when
something is going wrong in the analysis, and can yield clues as to how to fix or improve it if the SPC data
is correlated with other observations of factors which can potentially affect the analytical process.
REFERENCE
INTRODUCTION
It was thought that the digital circuit which replaced the sin (theta)
potentiometer fitted to earlier models might have been malfunctioning, or
that it was imprecisely defined because its steps for conversion were too
large and the settings irreproducible. Because of the sine function the
situation had always been more critical at low two-theta (higher
energies).
18.~-------------------------------------------'
7.
4.
1.
-2.~----r---~--~---r--~----r---~--'----r--~
428 462 496 538 564 598 632 . 666 788
2.S~~---------------------------------------------------------------------~
'.
..
.. t.. ; .. .. .. +.
·-·-·-·-·-.-.-·-.-;-~.r·-·-·-·.!~-;·-'~·:·-·-·-·-:~·-·r·-.-;-:-.;:-,-;.-~_._._._._._._._._._,_._ .•. _._.
. .... . +... .
I.25&-
.... ..
. .. ..~... . , ........ ...
. .. +. : .... " .. " ".-+" .. - ..
.. : ..
"t.. .. .. " ..
.. .... .. .... " .... +•
-2.5&+w----,---~~--_r----~--~----_r----._---.----_r--~
rocks it was found that there was also a sporadic error of about 10%
relative in magnesium when determined using a PX1 multilayer. A pulse
distribution (Fig. 3) shows silicon fluorescence from the multilayer at
about 75% of the maximum energy of the window encroaching on the analyte
peak position at approximately 50%. Adjustment of the upper level to 60
or 65% reduced this effect but the sporadic errors continued.
e.le9TTTTT--------------------------------------~~--~
kCPS
background it was possible for the peak and background to be measured, one
wi th the corrector in, and one out. This was the case for low - level
determinations of antimony and tin using their Ka lines_ In addition, the
fault moved from element to element as countrates changed with tube
emission and detector response_ Secondly, the precision of description of
goniometer angle was inadequate (insufficient bits were allowed for in the
random access memory of the microprocessor). This had the effect of
"quantising" these low angles into large steps in an energy range where
the values were critical in determining the amount of correction to be
applied.
At low energies the energy profile is broader in relation to the
energy width of the window such that the tails may be close to the upper
and lower levels and correction lead to a different proportion being
counted. In the case of magnesium determined with a PX1 synthetic
multilayer there is a further complication in that silicon fluorescence
from the PX1, excited largely by calcium present in samples, is situated
just to the high -energy side of the magnesium peak in the energy
distribution. This necessitates ~etting of the upper level to a lower
value than normal (perhaps 60-65%) and the upper level is then on a
steeply sloping part of the distribution with a high derivative of
intensity with respect to energy leading to statistical unsoundness and,
if the correction circuit is faulty, a further change in intensity on
switching. If the upper level is set to 75% there will be sporadic
partial capture of silicon radiation as the countrate changes.
SOLUTION OF PROBLEM
ACKNOWLEDGEMENTS
The authors would like to thank A S Robertson for assistance with the
experimental work. This paper is published with the permission of the
Director, British Geological Survey (NERC).
CORRECTIONS FOR THE EFFECT OF SCATTERING ON XRF INTENSITY
ABSTRACT
INTRODUCTION
In this work we write the exact expressions for the corrective terms of
the XRF intensity due to enhancement by scattered radiation. We also show
how these expressions may be easily evaluated in the particular geometrical
case of normal incidence.
THEORY
1)
af(Z,~,A)
8z ~
f(Z,W,A) + J
dA' J~ ~ ~
dw' k(W,A,W',A') 'U(z) f(z,~, ,A') +
4n
I o(z) O(~-~ ) O(A-A ) (1 )
o 0 0
the Eq. (1) can be transformed into a set of non-coupled equations. The sum
J. E. FERNANDEZ ET AL. 555
of these equations for n = 0, 1, ... is identical with eq. (1). The solu-
tion of this set of transport equations gives us both the source-dependent
zeroth-order flux
I
o
(2)
co
{ (l+sgn
2
1)
exp -TilT
(zll)
exp(W) J~A'Jd~' k(~,A,~',A') f(n-l)('t,~"A')
41£
co co
+ (l-~gn 1) Jd't exp(- 'til) JdA'Jd~' k( w,/\,w
~ ~ ~, ,/\~') f(n-l) ('t+z,w~'~')}
,/\ (3)
o TilT 0
41£
I (4)
"1 = cos1}
"1o=cos*o
z
Fig. 1: Irradiation arrangement of a homogeneous specimen of infinite
thickness under the excitation of a plane monochromatic X-ray
source, represented with the photon-transport equation (1).
556 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
4n:
{( l+S g n n')
'f
Til' IT
exp(- ~)
T'II'
dT'exp( ~) g
(0) -7
(T' w' A') +
2 0 a TJi'T
" TJi'T
2
n n')
+ (1 -Sg'f IOO
dT'exp(-
0 WI
T' II'
~) g ( 0) (T'+Z W'
a
-,
A') }
"
(5)
where
00
-7, A')
ga(0) ( z,w, = IdA"Id~" ka (~, ' A' , ~" , A") flO) (z,~" ,A")
o 4n:
(l+sgn '1)0)
(6)
2
To obtain Eq. (5), the total kernel has been split into the terms cor-
responding to every process of interest. A couple of them have been named
generically a and b. Eq. (6) is valid for positive z and depends directly
on the shape of the kernel for the interaction a. It includes all the in-
formation on the source.
Replacing (6) in (5) we get the second-order intensity for the thin-
beam model resulting from the chain of interactions (a,b)
(l-sgn '1)) (l+sgn '1)0 ) I 1
0
I~) (~,A) x
2 2 11)0 1 fl flo
+
11) 1 11)0 1
00
1
IdA' Id~' kb(~,A,~"A') k a (~',A',~,A)
0 0 TJi'T x
0 4n:
1
{
1
(l+sgn
2
1)' )
11 11'
+
(l-sgn
2
'1)' )
110 11'
} (7)
+ +
1'1)1 1'1)' 1 1'1)01 1'1)' 1
Eq. (7) is valid for every type of photon-atom interaction in
place of a and b. For the energy range we are considering in this work, the
possible interactions are the coherent and the incoherent scattering and
the photoelectric effect. The possible cross-interactions between them are
shown in Figure 2.
Photoelectric-effect kernel
1
[l-'I1(A'-A )] (8)
41l e
The line can only be emitted when A' is lower than the threshold wavelength
of the absorption edge (A ei ) of the series to which this line belongs.
Coherent-scattering kernel
~ Photoelectric
effect
Coherent
scattering
Incoherent
scattering
~ XRF intensity scat- Modification of the
Photo- Secondary tered from other symmetry of the XRF
electric
XRF intensity scatter centers line due to the lower
effect
towards the detector energy Compton emiss.
Coherent scatter-
Coherent
ing enhancement to
scattering
the XRF intensity
Second-order
Incoherent scat- scattering
Incoherent tering enhancement
scattering to the XRF inten-
sity
Fig. 2. Physical meaning of the chain of interactions (a,b) for the main
processes affecting soft X-rays. In this work we analyze the lower
two chains of the first column.
558 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
2)
2 -+ -+
-? -? (1 -+ -+ F (A' ,w·w' , Z)
kCOh(w,A,W',A') = ~ 8(A-A') + (w'w') (9)
Z
F (u) 0.388 -3
---Z--- = exp(-1.49 u ) (1 + exp(-0.3)) + 3.27xl0
where
AC ( 1- ~.w'
-+ -+ ) 1/2
u = 2
A
This expression will be used in this work to simplify the computations.
Experimental data and refined theoretical computations for many electrons
are reported in a very profuse bibliography6.
where
3
J. E. FERNANDEZ ET AL. 559
4n I11' I Z
1
}
1
{ (l+sgn 11' )
2
Il Il'
+
(l-sgn 11' )
2
Ilo Il'
(13)
+ +
1111 111' I 1110 1 111' I
x
560 VIII. X-RAY SPECTROMETRY OAT A ANALYSIS
1 1
->
a (1 - w"w 0+
->
\~CA') {
(1+sgn 1)' )
2
+
(l-sgn 1)' )
2
M M' Mo M' }
+ +
11) 1 11)' 1 11)0 1 11)' 1
(14 )
0- QA (\) [1-'U(A -A )]
(l-sgn 1)) I a(A-A, ) o e,
0 1 i
1(2) (i3,A)
coh-ph
1) ~l 2 M 2
0
+ Mo
11) 1
+ (1)' ) 2 F2 (A ,1)', Z) 1 F2 (\,-1)' ,Z) 1
{ }
1 1
0
Jd1)' + (15)
1)' Z M Mo Z Mo
+ Mo +
11) 1 1)' 1)'
-7
I,(2) h h (W,A) I 0 1
x
lnco -p 1) ~1
2 2
0
/-I
+ Mo
11) 1
KKN(i\',\) QA(A') [1-'U(i\'-i\e)]
i
[1 +
A - i\'
~
1--- + M'M
C 11) 1
The integrals in Eqs. (15) and (16) must be numerically evaluated, be-
cause the total cross sections are better known experimentally as a func-
tion of energy and the form factors are not easily integrable functions.
Since the factor (J' in these equations is not the scattering cross section
taking part of the total cross section, these expressions must be fully
evaluated in order to be compared with the secondary-XRF intensity.
EqS. (15) and (16) have been studied for some pure elements of low and
medium Z. A Romberg quadrature algorithm generalized to improper
J. E. FERNANDEZ ET AL. 561
integrals 10 has been used in the computations. The integrals have been
evaluated with a precision of four figures.
As follows from Eqs. (15) and (16), the cross-term intensities depend
on the incident radiation wavelength, the atomic number of the irradiated
element and the take-off direction. Because of the azimuthal symmetry, the
intensity is the same along all the directions in the cone of aperture-
cosine 1/' .
The first thing to point out for low energy x-rays is that the
coherent-scattering correction decisively has a greater importance than the
incoherent one. Beyond 100 KeV the incoherent cross-section grows and
overcomes the coherent 9ne, but since the photoelectric probability for
these energies falls rapidly, we cannot wait for an increase in the
importance of the XRF intensity due to incoherent enhancement.
CONCLUSIONS
The corrective terms to the XRF intensity due to both coherent and
incoherent scattering have been deduced using the photon-transport theory.
Complete kernels including atomic form factors have been used to obtain the
562 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
,~
i \
c \
\"----.
~ i
c
I
~
1:.
o
u
o I (a)
I
~
J II
U
r..
"-
x I ,
ENERGY [KeUI
10 -5 ~
~
\
(?
C
~
u
JIF \
1
~ \
.j
c
~ 1
2 1 (b)
J \ '\
1:.
o
U E
\
\
I
~
J
1J
r..
"x ~~
r , ,
r\
ENERG'! [KeUI
\
\
\
\
\
\
~
~
(c)
J
1J
r..
"x
, J
ENEDGY [KeUI
1 ~,
(
I \.\\
I
I (a)
\
\
\
\
c \
•
l
I \.\
'"'-"~"
~
.c
~I
8c
,I
~
J
1)
1-______....
L
C!
X
ENEQG'/ (KeUJ
"
c". I
I ""~'-
~
I
I '~
'-,
/
~
c
• I (b)
I I
l
~
.c
I
Q 2
u
C I
/
I
~
1)
L
C!
X 0
I(
f- , ,
/--------~ ~
///
/
I
I
I
c
•
l
I /
~ I
.c I
Q
U I
C I
I
/ (c)
I
!
Q I
ENEQGY (KeUl
Fig. 4. Like Fig. 3. but due to incoherent scattering. The smooth growth
starting at the absorption edge energy is due to the partial con-
tribution of the wavelength-integrals of Eq. (16), for increasing
energy.
564 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
1 .---.
/ ~,
/
"-
"-
(
(a)
~""- , ..... ....
~-'-,.
0
I I
/
/
t-
a
x
t /
\
t-
o 2
(b)
/
/
X
I
1:
o
U
c
1
//
/'
J ~
------/
I ,
./
ENEQG'/ [K.UJ
L
a
x
;
./
/1
(c)
I //
I
\
t-
o E
",
3"
"
C'
C E /
/
c
~ //
",; 2
//
u
"'
-
I
1
/./
o / i--
I!
I--
0
I
ENERG'/ [KeUJ
JQ
-,
II
~ //
//. "
j/ '-" 1
" I
L " \ rI
1 \
I a) \\ til
\1
\ JI
\I
1I! .1
i
r ~
~
1
~ (b)
\ \, !
/ ~\ /1
\ /
// II
\ I
'\
f'" //
/ !
'""- , '-,
I
Fig. 6. This polar graphs show the geometrical behavior of both the cohe-
rent (a) and the incoherent (b) corrections for a fixed energy.
Given the azimuthal symmetry, the full geometrical behavior can be
obtained as a revolution solid about the incidence direction axis.
// - --I----
--
'-"- 10 -of
/' "
---- - -".--.
/
/ '\,
/' \\, ,./-- -~
~
..
I
"
I ",
/' ",
! \
\ , ;/,.
"
"
(a)
I / \
/
I ~
\. I
\ !
/ (b)
\,\\
'-, //
0
'-, ~/ -- - - ---i
-------- \0 -4
Coh-XRF tXRF (POLAQ) Inc"h-tPr,xp, (:lOLAC!
2P SiitlPLE. INCiOENCE=NOQtlAL, 1:='10.0 K.. V
20 SAMPLE. INCIO(NCE:NOQHAL. (:40.0 KeU
Fig. 7. Polar graphs for both the coherent (a) and the incoherent (b) cor-
rections measured in units of the primary-XRF. The graphs show the
importance of the corrections as functions of the take-off angle.
They measure the differences with an isotropic process (like the
photoelectric one) also. The coherent enhancement is anisotropic.
The incoherent is almost isotropic.
566 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
intensity enhancement formulas of Eqs. (13) and (14). For the particular
case of normal incidence we have obtained the simpler expressions given by
Eqs. (15) and (16). These have been evaluated for some pure samples of low
and medium atomic number as functions of the incident energy and the
take-off angle.
ACKNOWLEDGEMENTS
One of us, J.E.F., would like to express his gratitude to the Argenti-
nean Scientific Research Council (Consejo Nacional de Investigaciones Cien-
tificas y Tecnicas de la Republica Argentina) which supports him with a
postdoctoral fellowship at the University of Bologna.
REFERENCES
J. E. Fernandez*
INTRODUCTION
XRFPC uses five files to compose its internal data base. Four of them
are ASCII files which can be edited and modified easily with a text editor.
PARAM contains the atomic number, the chemical symbol, the atomic
weight, the density and the conversion constant from
barns/atom to cm 2/g.
MCMASTER contains the McMaster l et al. coefficients for the photo-
electric series K, Land M, and for coherent and incoherent
scattering.
EDGES contains the absorption-edge related information as spectral
symbol, absorption-edge energy, absorption-edge jump, and
fluorescence yield. 2 ,3,4
LINES contains data on every single emission line as the s~ectral
series to which it belongs, its spectroscopic symbol, its
energy, its line fraction. 6 ,7,8
The fifth one is the file with the spectra. Since each spectrum
contains a high number of data points the original ASCII file needs to be
preprocessed by a program that transforms it to floating-point internal
format. XRFPC reads in the transformed file interactively.
PROGRAM ORGANIZATION
The query of the data base is very simple. There is not a utility
built into the program to change the data. The only allowed operation is
the reading of the data base, the parameters by atomic number (see Fig. 1),
and the spectra by classification number. At the end of a spectrum reading,
a plot of the spectrum is shown on the screen (see Fig. 2).
Fig. 1. A screen with the parameters for Zr as given by the data base query
utility. Parameters on the right are the McMaster coefficients.
,
.-~
! '; 1
\\
1
I 1
i
/ '\1 "1 I
i ,
\,
.~
I.~ I \
\
I
I
C .I
(J, \
\. I f\
\\ I!\
:I \
Jl
~ \ Ii
I' \,
a \ ,J
q
\
I
\ \
x Ii
\
--", Ii ,- ....
'\,
t
!
r ~
~-----"--.---,
....-.....
'-- I"--_~-
n! I
I I II I I Ii I I! I I I i I II I I
3 5
Fig. 2. A tube spectrum from the data base is plotted as a function of the
wavelength.
570 VIII. X-RAY SPECTROMETRY OAT A ANALYSIS
(see Fig. 3). After that, asking for single energies allows the user to
select his own energies of interest.
10i~------------+-------------+--------------
103 I=------>,,--------+-------------+-------------~
2~--------~~~~------------_+--------------~
~ 10
C.
~
U
::l 10 I~------------~------~-------+--------------~
r::
I~~-------------r----------~~,-------------~
10 -II:---------+--------+---------'=--....:::::-----l
I 2 3
10 10 10
ENERGY [KeUl
Totdl mass-attenu.3tion coefficient
Fig. 3. The total-mass attenuation coefficient can be plotted for multicom-
ponent mixtures providing a valuable tool to analyze complex
absorption chains.
J. E. FERNANDEZ 571
10 -2 8 SAMPLE
7
r
/1 FE 0.500
HI 0.300
CR 0.200
II I
I
E
Ii 1= J
,/
I
I
J'
,"'
5 ,I II
I II
'ri
~
[
~ ,l
III
'["' II LIliES I)
I
E ,1 II FE Kotl Ii
L-
:: II II I
a
x
2
~ /
I
i I
I
I
1 ;:
§ I
! II I
nJF=
~/ II
~ I I II I i I' I I i I I I I
3
/--, ,
II \ i\\
,
!•. '~
.l I! '\
\,
2r---~~----~r-+-------~---------+--------~
/
I
L-
I
I
a
x i
t !
~/
II
I II I II I I
Fig. 4. The integrated emission spectrum gives the XRF-intensity when the
excitation is polychromatic. How may this spectrum be affected
by the shape of the x-ray tube continuous spectrum? How is this
spectrum modified by changes in the excitation-detection geometry,
or in the sample composition? Answering these questions is easy
with the graph interface built into the XRF-intensity computation
module. The plots show the same intensity emission spectrum
(a) without x-ray tube interference, and (b) modulated by the
x-ray tube intensity.
572 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
CONCLUSIONS
ACKNOWLEDGEMENTS
REFERENCES
ABSTRACT
INTRODUCTION
In this work we write the exact expression for the correction term of
the XRF-intensity of a single line due to the multiple enhancement of
three other lines in the sample. We also show how this expression may be
numerically evaluated.
THEORY
af(Z'~'A)
f(Z'~'A) JdA'Jd~' k(~'A'~"A')
. ~
the Eq. (1) can be transformed to a set of non-coupled equations. The sum
of these equations for n = 0, 1, ... is identical with eq. (1). The solu-
tion of this set of transport equations gives us both the source-dependent
zeroth-order flux
I
o
(1 + sgn ~ sgn z) (2)
f(n) (z,~, A) = k x
00
{ (1+sgn
2 ~) exp (zM)
-~
Jdz1: exp (~
1:M) JdA'Jd~' k(~'A'~"A') f(n-l)(1:,~',A')
I~I 0 I~I
o 41l
The line may only be emitted if i\' is lower than the threshold wavelength
of the absorption edge (Aei ) of the series to which this line belongs.
Iterating Eq. (3) with the kernel k (it,i\,it',i\') on fro) (1,it,i\) , then
with k (it,i\,it',i\') on fIll (tit,i\), then w:th k (it,i\,it' ,i\') on f(2) (tit,i\) ,
k
(3) -7 -7 j .
and finally with k, (it,i\,it',i\') on f (r,w,i\), we obta~n the XRF-emission
l
of the line with wavelength i\ which is enhanced by the line i\" which is
J
576 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
x
41£ 2 2 I1J1)O/ + 11 0 /1)/
1 1 1
J ~J~J~x
a 1)' a 1)" a s
(a/ a) (a/ a o ) (a'+ a) (a'+ a o ) (a'+ ail (a"+ a) (a"+ a o) (a"+ ail (a"+ a')
(6)
where
a a a' a" a
a
1)' 1)" s
COMPUTATION
Eq. (6) has been studied for some few mixtures with elements of low
and medium Z. A ten-point Gauss quadrature algorithm generalized to
multidimensional integrals B has been used in the computations.
Other interactions among these lines have been ignored, such as the
following
Our main scope has been to compare the fourth-order with the lower orders,
without entering in details about the complexity of the emission for a
particular class of material. It is worth to be noted that in a real
mUlticomponent material, the possibilities of crossed interactions are
many, since the number of interacting lines is very high.
CONCLUSIONS
The corrective term to the XRF intensity due to the fourth-order
XRF-intensity has been deduced using photon transport theory. We have
obtained eq. (6) by using an isotropic scattering kernel to describe the
photoelectric interaction.
Eq. (6) has been evaluated for some mixtures of low and medium atomic
number as functions of the excitation energy for fixed incident and
takeoff angles. Only simple interaction chains have been studied. In
such cases the fourth-order term does not exceed 10 % of the tertiary
correction in the same chain.
578 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
-~
I
10 - ~
t- i
~
~
i\ I
i
I
Jo
~
.~
~
~
i
i,
J
I
I
3[
L~
[ I
IJ! i
C
~ \.\
II
E i
L
a
x
~
I-
!I
, I
I
I
I
I- \ I
t: \1
.:::
II (a) Ii
~
~
I
F 11
1
~ i"'.
-t1 -----------,--
C! I
U
I
I
I, I
~ i ',._-
f'-~. ''-,
I 1111 II I II I I I II I I i I I
Energy [1~eUl
10 -8 et.
~ ~-\
E 11 \ I
\
;,
II
E Ii'I
Jo 6
~
II
\\ , I
I!
.~
tJl
C "
il \
Gi
jl \ (b)
c \j
Ii ~'"
L
ii
a
x ~ i\
'I
'\'\,
\..,
I
~
11
2,;:
;:
IiII'I
Ii
I \,
\
,
~
!E~------tr--------r-------+----~~~----~
Qr~-------4------~~------+--------r------~
I Iii i 1 Iii I II I I I 1
Energy [I~eUl
ACKNOWLEDGEMENTS
One of us, J.E.F., would like to express his gratitude to the Argenti-
nean Scientific Research Council (Consejo Nacional de Investigaciones Cien-
tificas y Tecnicas de la Republica Argentina) which supports him with a
postdoctoral fellowship at the University of Bologna.
REFERENCES
APPENDIX 1
+ ai (8 ai + 10 aD + 10 a) + aD (4 aD + 6 a) + 2 a 2 )
+ ai «4 aD + 4 a) ai + aD (2 ao + 6 a) + 4 a 2 )
+ aD (2 a aD + 2 a 2 )) a" + a'
+ aD (8 aD + 13 a) + 5 a 2) + ai
(ai (8 ai + 21 aD + 21 a) + aD (15 ao + 30 a) + 15 a 2 )
+ aD (aD (4 aD + 11 a) + 9 a 2 ) + 2 a 3 )
580 VIII. X-RAY SPECTROMETRY DATA ANALYSIS
+ ai (ai «4 ao + 4 a) ai + ao (5 ao + 13 a) + 8 a 2 )
+ ao (ao (2 ao + 9 a) + 11 a 2 ) + 4 a 2 )
+ ao (ao (2 a ao + 4 a 2 ) + 2 a 3 »
+ 2 a 2 ) a' + ai (ai (8 ai + 20 ao + 20 a)
+ ao (15 ao + 28 a) + 13 a 2 ) + ao (ao (4 ao + 11 a) + 8 a 2 )
+ 15 a 2 ) + ao (ao (5 ao + 21 a) + 21 a 2 ) + 5 a 3 )
+ ao (ao (5 a ao + 8 a 2 ) + 3 a 3 »
+ ai (ai «ao (2 ao + 6 a) + 4 a 2 ) ai
+ ao (ao (ao + 8 a) + 11 a 2 ) + 4 a 3 )
+ ao (ao (3 a ao + 8 a 2 ) + 5 a 3 » + ao 2 (2 a 2 ao + 2 a 3 »
+ ao (2 a ao + 2 a 2 » a' + ai
(ai «4 ao + 4 a) ai + ao (6 ao + 12 a) + 6 a 2 )
+ ao (ao (2 ao + 9 a) + 9 a 2 ) + 2 a 3 )
+ ao (ao (2 a ao + 3 a 2 ) + a 3 »
+ ai (ai ( (ao (4 ao + 6 a) + 2 a 2 ) ai
+ ao (ao (2 ao + 9 a) + 9 a 2 ) + 2 a 3)
+ ao (ao (3 a ao + 6 a 2 ) + 3 a 3 » + ao 2 (a 2 ao + a 3 »
+ ai (ai (ao (2 a ao + 2 a 2 ) ai
+ ao (ao (a ao + 3 a 2 ) + 2 a 3 » + ao 2 (a 2 ao + a 3 » (A1)
where a, ao' a,1 , a' and an are defined in the main text.
TXRFSPECTROMETERFORTRACEELEMENTDETECTION
Introduction
A W-anode tube, operated with 55kV, was used as the primary source. Mainly the W-L rays
(8.4 - 9.9 keY) have induced K lines of elements with Z<28 and L lines of elements
with Z<65. The bremsstrahlung has induced K lines of elements with Z<58 and L lines
of elements with Z>65. By these means a wide range of elements (l2<Z<83) was
measurable.
Fig. 1 shows the set-up of the spectrometer. The finely collimated beam (0.05 mm) was
directed on the synsil sample substrate with an incident angle of 0.5 mrad.
w
ANODE
COLLIMATOR
CHAMBER
~
I
/+
to Co Zn
>-
Mn~
l- Ni
0.5
ca/+ ;:r-\
(J)
z / E. ~
w o Au
l-
Sc Gd 0
z S~ I~
>-
<t
a:
01
K
/
/
+ 0
I
Cs
0
Ba
Ce
Re Hg
I
><
0.05
Several calibration solutions containing 7-8 elements with equal concentrations (1.0 or 10.0
ng/ul) were prepared. Each solution contained two internal standards: Co and Mo. The
measured K- and L-line intensities below 10.5 keY were normalized to the Co K- and above
10.5 keY to the Mo K-line intensities. Two internal standards, Co and Mo, were used
because it was observed that the relative intensity of the lower energy lines compared to the
higher energy lines can vary from one measurement to another. This effect could be due to
the variation of: the incidence angle, the roughness of the substrate and the thickness of the
sample. Fig. 2. and 3. show the measured K- and L-line intensities of 30 elements.
>-
f- Rb
Se
to 0
Sr
(J)
Z
W 0.5 In
f-
Z ~b
>- ~I
~
<I:
a:
I
ru
0.1
>(
I
x
Cs
,
10 20 30
~ ~~~--~--~~---r--~~--~-----r----~~~
C
w
l-
S -5.0
o
1:
<I:
W
...J
OJ
<I: 2.0
0::
::J
Ul
<I:
W
1: to
1:
::J
1:
....
z 0.5 40
.... 20 60 80
I:
ATOMIC NUMBER
FIG. 4. Minimum Measurable Amount VS. Atomic Number
To define the minimum measurable amounts for this experiment two criteria are introduced.
The first is that the net Kil or La peak intensity is 20% of the total counts in the inverval of
the region of interest. The second criterion is that the number of counts is higher than 3000.
Fig. 4 shows the minimum measurable amount vs. atomic number. It is to see that with one
short measurement (typically 2-5 min.) a sample could be analysed for a wide range of
elements. However for certain elements, e.g. Z<20 or 2>80, the detection limit is high. To
lower the detection limit for these elements, the sample could be remeasured with another
spectrometer set-up.
The spectrometer was successfully used for trace element detection in various
environmental samples, e.g. aerosol and water.
References
1. Advances in X-Ray Analysis, Vol. 30. 85-88,1987
D. J. Kalnicky
PGT/Outokumpu, Princeton, NJ
INTRODUCTION
ANALYZER COMPONENTS
Electronics Unit
o 0
o 0
Printer o 0
Measurement
Head
¢
II
..
User
Terminal
I
II
MEASUREMENT OPTIMIZATION
APPLICATION EXAMPLES
Several application examples will be discussed below to
illustrate the excellent performance of the Courier 10 for a
variety of sample types.
ppm S
o 9
0.8
o 7
~n
'"'19
t? c: o 6
'"
.-t;~
wg 0.5 0
Of-
'-v
U
o 4 0
0.3
0.2
0
o 1
0
0
0 o 2 o 4 o6 o 8
CThousands)
Assay Value
Figure 2: Sulfur in Gasoline with Courier 10
D. J. KALNICKY ET AL. 589
240 Mg/L Pb
220
200
180
160
...~
f 140
...
Ul
w 120
...0
u 100
80
60
40
20
0
0 40 80 120 160 200 240
Assay Value
Figure 3: Lead in Gasoline with Courier 10
Calib,ation Results
160
150
ppm Ca
140
130
...al 120
~
...
Ul
w 110
0
..-
u
100
90
80
70
70 90 110 130 150
Assay Value
Figure 4: Calcium in Polymer Pellets with Courier 10
o 9 ~NI
0.8
o 7
o
...al 0.6
o
~
... 0.5
8l
o
..-
0
u
o 4
o 3
0 2
o 1
0
0 o 2 o 4 o 6 o 8
Assay Value
Figure 5: Ni in Metal Treatment Bath with Courier 10
D. J. KALNICKY ET AL. 591
SUMMARY
REFERENCE
1. M. Hietala and J. Viitanen, "A Radioisotope On-Stream
Analyzer for the Mining Industry", Advances in X-Ray
Analysis, Vol. ,21, Plenum, New York (1978).
XRF MACRO PROBE ANALYSIS OF GEOLOGIC MATERIALS
E.A. Erslev
Department of Earth Resources
Colorado State University
Fort Collins, Colorado
INTRODUCTION
An x-ray fluorescence macroprobe was built for
intermediate-scale compositional mapping to bridge the
gap in spatial resolution between bulk x-ray fluorescence
and electron beam methods. The macroprobe was optimized
for quantitative whole rock mapping on a millimeter scale
to evaluate changes in bulk composition of fine-grained
mineral aggregates.
Previous studies of changes in bulk composition have
been hindered by the limited volume of material excited
by electron beam methods. X-ray excitation was chosen
over electron beam excitation because of the order of
magnitude greater effective penetration of the x-rays,
the greater stability of x-ray tubes and the relative
absence of surface effects such as charging.
X-ray "milliprobes" have been used to analyze small
areas of geologic materials in the past (1,2,3,4). These
wavelength dispersive instruments were designed for
single spot analysis, not automated for the mapping of
large areas. Recent advances in x-ray microfluorescence
aim for resolutions close to that obtainable with an
electron beam (5,6). Our goal was to develop an instru-
ment which complements existing electron beam and micro-
L~ DEWAR
INSTRUMENT
The Colorado State University XRF macroprobe consists
of a vacuum-tight sample chamber, computer-driven x-y
translational stage, a micro focus x-ray tube and an
energy dispersive Si(Li) detector (Figure 1). The sample
chamber was designed allowing maximum stage movement
while mounting both the x-ray tube and the detector
window as close to the sample as possible. The 3/8 inch
thick stainless steel walls of the sample chamber have a
door on the front for sample introduction and a leaded
glass window to allow viewing of the sample with a
binocular microscope.
The x-ray tube is a Kevex (Scotts Valley, CAl 50
watt, Rh anode, water-cooled Kevex microfocus tube. The
electron beam diameter at the anode is 0.10 x 0.15 mm,
yielding a more concentrated x-ray beam than conventional
tubes. The tube was mounted on the chamber at 45 degrees
N. L. GILFRICH ET AL. 595
PRECISION
The overall precision of the macroprobe was evaluated
and the contributions of collimator size and stage pos-
itioning were determined. Three sets of 10 replicate
measurements were performed on a single area of a slate
sample. The relative standard deviations (RSD) in
percent were calculated from the intensities for each of
nine elements (Table 1). The first set of measurements
was performed using the 0.7 x 1.0 mm beam spot size. The
RSD ranges from 0.26% to 5.45%.
STANDARDIZATION
The fundamental parameters computations developed by
Criss and Birks (7) and implemented by the Tracor soft-
ware was used to quantify the analyses. Because the
fine-grained rocks to be analyzed show little similarity
to traditional standards (e.g., fluxed disks or pressed
pellets), homogeneous slates of known composition were
used as standards.
A fine-grained, optically homogeneous slate (CSU1)
was chosen as a standard to analyze other slates. Part
of this rock was sent to the u.S. Geological Survey
(USGS) and analyzed by WDXRF (8). Sulfur was determined
with a LECO sulfur analyzer. The other part of the rock
was mounted for analysis with the macroprobe. In order
to standardize for an analysis, several spots on this
slate were analyzed in the macroprobe. The peak intens-
ities for the elements from each spectrum were extracted
and averaged. These average intensities were combined
with the weight percent composition from the USGS for the
standardization.
ACCURACY
The macroprobe accuracy using the above standard-
ization procedure was evaluated by analyzing two other
slate samples also analyzed by the USGS (W91 and W92). A
group of spectra were collected from different points on
these samples and the weight percent composition of each
spot was determined. Table 2 compares the average macro-
N. L. GILFRICH ET AL. 597
W91 W92
~ SAMPLE CM- 3A
A
60
o 00
o o~ 0
....
0 0 0
.: "
II 7 II 50
~ 6 ," , :Q
5 40
,. ,,
t:.t;l> t:.
5 ,
~ 4 t, ~,
"
~ 30
3
20
2
x
X
xxx xxt)( 10
1 o· } x x
0 "" 0 + .+
9 11 13 15 17 19 21 23 25 9 11 13 15 17 19 21 23 25
Wt%AI203 Wt%Al203
+ Fe203 6 K20 x TI02 v MnO o SI02 + cao
c Oxides/Al203 Versus Distance
D
Volume-Loss Versus Distance
70
0.8
0.6 60
g
0.4 50
C/l
0.2+-_ _ _ _ _ _ _ _ _ _ _ _---1 g40
i . . . . - . . . . . .- -. - -.-
w
~ ~.2
~ 30
~::
~ 20
.J -0.8 "f-
-1 lO
-1.2
0
-1.4
-1.6 +-.---.-_._r-+_,_~___.',.....,r_r__._r__r_,_~___,__1 ·10
o 2 6 8 10 12 14 16 18 20 0 2 8 10 12 14 16 18 20
Distance (mm) Distance (mm)
o Si02 I- cao 0 Fe203 6 K20 y Ti02 Avg Vol. Loss = 42%
CONCLUSION
The XRF macroprobe has provided a good method for
volume-loss determination in slates. The technique shows
promise for further studies of millimeter-scale element
fractionation. Tests of macroprobe precision on a rela-
tively homogeneous slate yielded relative standard
deviations for different components ranging from 0.17% to
3.02%, with poorer precision obtained using a smaller
spot size. The accuracy obtained for the slates was
limited by the use of inhomogeneous standards and un-
knowns yielding total absolute errors after normalization
of 2.59 and 4.42 weight percent for two different
samples.
ACKNOWLEDGEMENTS
The authors would like to acknowledge the National
Science Foundation (Grant EAR-8720824) for funding the
construction of the macroprobe. We would also like to
extend appreciation to D. Ward for sample preparation, J.
Taggart of the USGS for the WDXRF analyses and A. Harding
from Tracor Xray, Inc. for his assistance with the pro-
ject.
REFERENCES
1. I. Adler, J.M. Axelrod, American Mineralogist, 41,
524, 1956.
2. I. Adler, J. Axelrod, J.J.R. Branco, Proceedings
Seventh Annual Conference of Industrial Applications of
X-ray Analysis, 167, 1958.
3. K.F.J. Heinrich, Advances in X-ray Analysis,s, 1,
1962.
4. H.J. Rose, Jr., R.P. Christian, J.R. Lindsay, R.R.
Larson, U.S. Geological Society Professional Paper 650-B,
128, 1969.
5. M.C. Nichols, D.R. Boehme, R.W. Ryon, D. Wherry, B.
Cross, G. Aden, Advances in X-ray Analysis, 30, 45, 1986.
6. D.A. Carpenter, M.A. Taylor, C.E. Holcombe, Advances
in X-ray Analysis, 32, 115, 1989.
N. L. GILFRICH ET AL. 601
Lo I Yin
Code 682, Laboratory for Astronomy and Solar Physics, NASA Goddard
Space Flight Center, Greenbelt, MD 20771
Stephen M. Seltzer
ABSTRACT
INTRODUCTION
..,
O~r-~~~~~-r-r~~~~~~~
o
o
co
o
ID
o
2~
c
:J
or-
U 0
ID
o
n
o
~
o
..,
o
N
o
(1)
Table I. Normalized correlation coefficients (NCC's) derived from the XRF spectra of 14 standard
pellets of geological samples. All spectra were taken under identical conditions and with comparable
counting statistics. The pairwise correlation coefficients form a triangular matrix, and have been
multipled by 100 to convert them to percentages. Pairs of spectra with very similar patterns have
NCC's near 100%.
AGV BCR BHVO BIR ONC G1 G2 GSP PCC RGH ST1A STH W1 W2
AGV 100.00
BCR 95.79100.00
BHVO 92.35 95.97 100.00
BIR 89.15 91.04 98.24 100.00
ONC 91.06 92.45 98.17 99.76 100.00
G1 82.79 65.41 57.81 55.09 58.16100.00
G2 87.58 71.38 64.58 61.69 64.63 98.88 100.00
GSP 91.44 78.32 70.46 66.69 69.63 97.78 98.87 100.00
PCC 78.68 87.24 73.79 67.08 70.36 51.74 55.35 63.49100.00
RGM 82.42 64.51 56.44 53.67 56.93 98.65 99.26 97.04 50.98 100.00
ST1A 93.51 98.58 98.47 96.02 96.85 58.42 65.02 71.98 82.98 57.39 100.00
STH 93.54 88.49 76.62 71.55 75.18 87.11 88.75 93.19 85.27 85.72 82.34 100.00
W1 92.83 96.51 99.12 98.45 98.95 58.44 64.77 71.01 78.16 57.00 99.30 79.57 100.00
W2 93.48 94.99 99.10 99.27 99.63 61.31 67.84 73.07 73.35 60.29 98.18 78.59 99.51 100.00
If two XRF spectra have a NCC of 99%, are their chemical compositions
really very similar? To answer this question, we took the published
chemical compositions of these samples, treated each set of chemical
concentrations as a vector in pattern space, and applied Eq.(l) to these
vectors. In this way, we obtained another set of NCCs based on chemical
concentrations. This set of "chemical" NCCs were arrived at by attenuating
the concentrations of A1203 by a factor of 4, and Si02 by a factor of 15 in
order to make them comparable in magnitude with the other components. The
results are shown in Table II. For easy inspection, we have also set those
NCCs below 97% to 0 to highlight the strongly correlated samples. It is
seen that, with only three exceptions, high NCCs in XRF spectra also reflect
high similarities in chemical concentrations. In fact, when all the
"chemical" NCCs are plotted against the "spectral" NCCs, they follow each
other very well as shown in Fig. 2.
Table IJ. C~rison of the NCC matrix derived from the XRF spectra with that derived from the
chemical c~sitions. for the same set of 14 standard sa~les. The matrices list only those NCC
values above 97% to select those sa~le pairs which have either highly similar spectral patterns or
highly similar chemical c~sitions. The two matrices produce very similar selections.
AGV BCR BHVO BIR ONC G1 G2 GSP PCC RGH ST1A STH 111 112
AGV 100.00
BCR 0.00 100.00
BHVO 0.00 0.00 100.00
BIR 0.00 0.00 98.24 100.00
ONC 0.00 0.00 98.17 99.76 100.00
G1 0.00 0.00 0.00 0.00 0.00 100.00
G2 0.00 0.00 0.00 0.00 0.00 98.88 100.00
GSP 0.00 0.00 0.00 0.00 0.00 97.78 98.87 100.00
PCC 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
RGH 0.00 0.00 0.00 0.00 0.00 98.65 99.26 97.04 0.00 100.00
ST1A 0.00 98.58 98.47 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
STH 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
111 0.00 0.00 99.12 98.45 98.95 0.00 0.00 0.00 0.00 0.00 99.30 0.00 100.00
112 0.00 0.00 99.10 99.27 99.63 0.00 0.00 0.00 0.00 0.00 98.18 0.00 99.51 100.00
AGV BCR BHVO BIR ONC G1 G2 GSP PCC RGH ST1A STH 111 112
AGV 100.00
BCR 0.00 100.00
BHVO 0.00 0.00 100.00
BIR 0.00 0.00 98.72 100.00
ONC 0.00 0.00 98.26 99.65 100.00
G1 0.00 0.00 0.00 0.00 0.00 100.00
G2 0.00 0.00 0.00 0.00 0.00 98.50 100.00
GSP 0.00 0.00 0.00 0.00 0.00 0.00 98.47 100.00
PCC 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
RGH 0.00 0.00 0.00 0.00 0.00 99.38 98.92 0.00 0.00 100.00
ST1A 0.00 98.91 98.69 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
STH 0.00 0.00 0.00 0.00 0.00 0.00 0.00 97.48 0.00 0.00 0.00 100.00
111 0.00 0.00 99.48 99.47 99.43 0.00 0.00 0.00 0.00 0.00 98.15 0.00 100.00
112 0.00 0.00 99.39 99.54 99.57 0.00 0.00 0.00 0.00 0.00 97.86 0.00 99.99 100.00
o
O;-__. -__. -__. -__, -__, -__, -__, -__. -__. -__. -__.-~~
)~
U')
OJ O~
o O~ 0
OJ
o o 0
0 0 0
U')
co o o
o o
lo...
""0
o 00
~co
a. o@
U'lU') o
Er-- o
o
lo...
o 00
.... 0
or--
o 8ft; 0
o 0 0 0
'B@
ZU')
to
o o
to
U')
U')
0
U')~
0 __<;)
~ __ ~ __ ~ __ ~ __ ~ __ ~ __4 -__4 -__ ~ __ ~ __ ~~
40 45 50 55 60 65 70 75 80 85 90 95 100
NCC from Compositions
Fig. 2. NCCs derived from XRF spectra of 14
standard geological samples versus
those derived from tabulated
chemical concentrations for the same
set of samples.
samples which poses a very different kind of problem. We obtained from R.A.
Livingston of the University of Maryland a set of pigment samples which
simulated those used by ancient Egyptians for their wall paintings. The
various pigments were painted as round spots of about 0.5 cm in diameter on
a large limestone block. For each pigment, painted spots were prepared
using pigment concentrations of 100%, 50%, 25%, 12.5% and 6.25%. The XRF
spectra were collected in air using a small portable low-power x-ray
generator with a tungsten anode operated at 45 kV and 25 pA, and a Si(Li)
detector. Because the size of the x-ray beam is larger than the painted
spots, each spectrum contains considerable contribution from the limestone
background. Furthermore, this background does not remain constant. As the
concentration of the pigment decreases the relative intensities of the
background features increase. Under these circumstances, we are interested
in whether the pattern-recognition technique could still sort and identify
the pigments, and if so, down to what concentration level.
Fig. 3 shows the spectra of some of the pigment samples. The bottom
spectrum is from the limestone background without any pigment. We see that
although the spectra contain many features, most of them are from the
background, i.e., Ca, Fe, W (L lines) and Sr. The W L lines originated from
the W anode which were coherently scattered by the sample into the detector.
The pigments represented in Fig. 3 are: malachite green (Cu carbonate),
cobalt blue (Co), ochre red (Fe), realgar orange (arsenic sulfide), and
cinnabar brown (HgS) , all at 100% concentration. The predominance of the
background features tend to make all of these spectra highly correlated.
Furthermore, in the case of cinnabar, there is strong overlap among the Hg
and W L lines. Faced with this complexity, we implemented a very simple
routine to subtract the background spectrum. We normalized the intensities
of the background spectrum against each pigment spectrum using the three
L. I. YIN AND S. M. SELTZER 609
major peaks common to all spectra: Ca Ka (3.7 keV) , one of the W L lines at
8.4 keV, and Sr Ka (14.1 keV). The largest normalization factor among the
three was chosen to multiply the background spectrum which was then
subtracted from the sample spectrum. Negative counts due to over
subtraction were set to zero. The largest normalization factor was chosen
because residual background counts could be mistakenly interpreted as
spectral features by the pattern-recognition algorithm, especially when the
signal was low, whereas zeroes could not contribute to any artifacts.
"0
•
'0
,.,
'0
N
'0
0
'0
..
0
2C ~
"'10
0 0
u
....
0
.0
"0
•0
~
,.,
0
N
0
Table III. Normalized correlation coefficients (NCC's) derived from the XRF spectra of 6 paints, diluted to the indicated concen·
trations. The spectra were measured with the samples painted on a plaster surface. The Nee's were calculated after subtracting
from each spectrum the plaster·only background spectrum. In the matrix below, all Nee values below 90% were set equal to 0 to
to facilitate the recognition of highly correlated pairs.
100% 25% 6% 100% 25% 6% 100% 25% 6% 100% 25% 6% 100% 25% 6% 100% 25% 6%
100% 100.00
25% 98.46 100.00
6% 96.59 97.11 100.00
100% 0.00 0.00 0.00 100.00
25% 0.00 0.00 0.00 0.00 100.00
6% 0.00 0.00 0.00 0.00 0.00 100.00
100% 0.00 0.00 0.00 0.00 0.00 0.00 100.00
25% 0.00 0.00 0.00 0.00 0.00 0.00 99.94 100.00
6% 0.00 0.00 0.00 0.00 0.00 0.00 98.42 98.58 100.00
100% 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
25% 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 99.91 100.00
6% 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 96.76 97.00 100.00
100% 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00 X
25% 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 93.70 100.00
6% 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 91.84 100.00 X
:0
"'T1
100% 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00 Z
25% 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 99.94 100.00 en
6% 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 99.75 99.79 100.00 -!
:0
c:
:s:
m
Z
-!
~
-!
0
Z
L. I. YIN AND S. M. SELTZER 611
PROPORTIONAL-COUNTER SPECTRA
'"o~-r'-~-r'-~-r~~rT~~~-r~
.o
,...
o
.,
o
2c
::J
o
U
OJ
o
•o
,.,
o
N
o
Table IV. C~rison of the NCC matrix derived from the proportional-counter
XRF spectra with that derived from the chemical compositions, for the same
set of 9 standard geological s8fl1)les.
AGV 100.00
BCR 95.21 100.00
BIR 0.00 0.00 100.00
DNC 0.00 0.00 99.94 100.00
DTS 0.00 0.00 0.00 0.00 100.00
G2 0.00 0.00 0.00 0.00 0.00 100.00
GSP 0.00 0.00 0.00 0.00 0.00 99.81 100.00
olO 95.25 0.00 0.00 0.00 0.00 95.11 0.00 100.00
112 0.00 96.27 99.81 99.79 0.00 0.00 0.00 0.00 100.00
AGV 100.00
BCR 97.86 100.00
BIR 0.00 0.00 100.00
DNC 0.00 0.00 99.79 100.00
DTS 0.00 0.00 0.00 0.00 100.00
G2 0.00 0.00 0.00 0.00 0.00 100.00
GSP 0.00 0.00 0.00 0.00 0.00 99.61 100.00
olO 98.51 0.00 0.00 0.00 0.00 99.14 99.21 100.00
112 0.00 0.00 99.72 99.40 0.00 0.00 0.00 0.00 100.00
K x-ray peak. Although this procedure helped to subtract out the escape
peak of Fe from the Ca region, it also eliminated the Fe peak as a
distinguishing spectral feature. The NCC matrix, computed after spectrum
subtraction, is shown in Table IV based on a 97% threshold. For comparison,
we also show the "chemical" NCC matrix using a 95% threshold and based only
on those constituents which contributed to proportional-counter "spectral"
NCCs (i.e., K Ca, Ti, Cr, and Mn). It is seen that even under such
unfavorable e~perimental conditions the agreement between the two matrices
of NCCs is still quite acceptable.
CONCLUSIONS
ACKNOWLEDGEMENTS
REFERENCES
James M. Thorne
Department of Chemistry
and
Center for X-ray Imaging
Brigham Young University, Provo, Utah 84602
INTRODUCTION
X-ray windows are used in sources and detectors to separate the neighborhood
of the x-ray generation or detection from the use environment. While each use has
its own requirements, there are some principles that should be used in designing an
optimal x-ray window. Because x rays are absorbed to some extent by all materials,
minimizing absorption is one criterion in preparing windows. Also, for most uses
there is a pressure difference across the window so that ensuring pinhole-free structure
and sufficient mechanical strength to support the differential is another criterion for
window design. Traditionally, absorption is minimized by fabricating the window with
lower atomic number (low Z) elements such as Be, B or C. However, the wavelength
(energy) region of interest becomes a very important complicating factor.
Over the years many different materials have been used as windows for soft x-
rays. A primary use has been for proportional counters. More recently interest has
developed for Si(Li) detectors. The materials used to date for proportional counters
include: polypropylene, lexan, and formvar l - 4 . Materials used for Si(Li) detectors
have been Be, BN x , formvar, a-C:H (diamond-like carbon), and mylar 5 . The purpose
of this paper is to discuss the design of x-ray windows with particular emphasis on
windows for detectors in the soft x-ray region. This is because existing windows can
typically be made thin enough to have very good transmission for the harder x-rays.
and this is the approach taken in this study. The second is more subjective and will
be treated in a descriptive fashion. While materials can be ranked according to bulk
moduli and yield strengths, the integrity and mechanical properties of thin films are
very dependent on preparation techniques and microstructure.
The calculations were done using a computer code called UM which was designed
and written by us. It is written in FORTRAN 77 and was designed to do a wide variety
of calculations involving x-rays, thin films, and crystals. It accesses a data base of
atomic scattering factors. For the problem at hand these are used to calculate optical
constants.
The atomic scattering factors were obtained from Henke's tables and expanded at
BYU and Lawrence Livermore National Laboratory. Beer's law was used in optical
calculations. This is justifiable since the films are always much thicker than the
wavelength of soft x rays and hence interference effects are negligible. In addition,
the reflection of x rays from front and back suriaces of the windows at near normal
incidence was calculated to be much smaller than 10- 4 and thus could be neglected
over the whole range.
Calculations were made over the range lOA. to 70A. (170 eV to 1200 eV). X rays
of this range are too soft to be transmitted through the beryllium windows used in
typical Si(Li) detectors. This ranges includes the Ka radiation of all elements with
atomic numbers from Z = 5 to 11 (boron to sodium).
RESULTS
It will be noted that the polypropylene is much more transmissive than is the
typical Be window for the soft x ray region, particularly on the low energy side of
the C absorption edge. One can see the effect of the carbon content by the rather
dramatic absorption edge; the nitrogen Ka radiation is highly attenuated. In this
respect, polypropylene resembles a carbon K-edge filter (allows carbon x rays to pass
but blocks more energetic, particularly nitrogen, x rays). This is an important design
point since any material which is predominantly a single element or a single element
with hydrogen will act as an edge filter. Polypropylene is not a single element but
acts as such for soft x rays since the hydrogen is transparent and can be ignored. The
transmission curve for 400 nm of diamond is shown in Figure 3. Like polypropylene
it acts as an edge filter, extinguishing nitrogen x rays. However, because of its higher
density and low hydrogen content, 0.4 J.lm of diamond is equivalent to about 1.9 J.lm
of polypropylene.
There is a variety of materials that incorporate various low-Z elements into their
structure. The transmission of 400 nm of formvar (polyvinyl formal), which has been
R. T. PERKINS ET AL. 617
1.0
0.9
I
0.7
0.6
0.5
~ 0.4
~ 0.3
0.2
,
,
0.1
,
0.0 '____'__...L.:..~"--__'_......:..:..:...w..._..L.-;:J.....L_ _l..._...L__'____'__...J
10 15 20 25 30 35 .0 45 50 55 60 65 70
WAVELENGTH ( ANGSTROMS )
Figure 1. Transmission for 1 Jlm (solid line), 2 Jlm (dotted) and 6 Jlm (dashed)
thick polypropylene. '
0.9
0.6
I:
~ 0.5
o~ 0.4
~ 0.3
0.2
0.1
10 15 20 25 30 35 40 45 50 55 60 65 70
WAVELENGTH ( IN ANGSTROMS )
Figure 2. Transmission for 8 Jlm thick beryllium.
used in the past for proportional counter windows, is shown in Figure 3. Another
material which may be used for soft x-ray windows is a boron containing material
such as BN. Also shown in Figure 3 is the 'transmission for 400 nm of hexagonal BN.
Note that for equivalent thicknesses of 400 nm , BN windows, and especially formvar
windows, are much more transparent than diamond for F, 0, and N x rays. In this
case the BN falls below diamond for carbon x-ray transmission but is still useable at
this thickness.
The superiority of the compound materials comes about for two reasons. First,
formvar has a lower density than diamond, and second but more important than
618 IX. XRF INSTRUMENTATION
1.0
0.9
.\
0.8
\
0.7 \
0.8
0.5 , ...
0.4 \
\
0.3
0.2 , ,,
0.1
0.0 ':---:':-~---:':--l...--':''''-'-''........L-~_l.--......L_...L..---l'-----.J
10 15 20 25 30 35 40 45 50 55 60 65 70
WAVELENGTH ( IN ANGSI'ROMS )
Figure 3. Transmission for 400 nm of formvar (solid line), diamond (dotted), and
BN (dashed).
.... :
0.0 l---'-_~_.L..---'-_-'-_..L---'_--'-_...L-_L----'-_-'
10 15 20 25 30 35 40 45 50 55 60 65 70
WAVELENGTH ( IN A.~GSI'ROMS )
density, it is not a single element. It contains a mixture of carbon and oxygen. The
importance of this second point can be seen in Figure 4, which is the transmission
of the same three materials as in Figure 3 but for the same mass per unit area
instead of equal thickness. This suggests that the best window material from the
point of view of high transmission for a given mass thickness would be a mixture of
C, N, and O. Hydrogen would be used as required to tie up the dangling bonds. A
transmission curve for such a material is shown in Figure 5. This is in contrast with
R. T. PERKINS ET AL. 619
10
09
08
z
0 0.7
Bl
Si
CIl 0.6
~ 0.5
~
z 0.4
0
~ 0.3
0.2
0.1
0.0 70
10 15 20 25 30 35 40
WAVElENGTH
t5)
A
50 55 60 65
1.0 .....
0.9
i
0.6
0.7
0.6
0.5
~ 0.4
~ 0.3
0.2
0.1
0.0
10 15 20 25 30 35 40 45 50 55 60 65 70
WAVElENGTH ( IN ANGSTROMS )
Figure 6. Transmission for 10 (solid line), 20 (solid line) and 40 nm of Al (dashed).
single element windows like diamond which are "colored" in that they transmit the
certain wavelengths better than others. Windows with 1:1:1 mixtures of the elements
will transmit roughly equivalently the wavelengths that correspond to each element.
Such windows would be approximately neutral density for elemental soft x rays and
can be termed "gray windows". Producing a "gray window" would be a noteworthy
achievement. Probably it will not be possible to produce such a window with a 1:1:1
mixture exactly, but in practice it may be possible to develop a compound with nearly
these ratios. It should be noted some of the families of materials which contain C, N,
and 0 are the polyimides, nylon and Kevlar.
Nonmetallic x-ray windows are normally metallized. In the case of proportional
counters to prevent charge build up. In the case of Si(Li) detectors this is to block the
620 IX. XRF INSTRUMENTATION
transmission of infrared and visible light which would otherwise generate unwanted
noise. The metallic coating should be as thin as possible to to minimize x-ray absorp-
tion. The metal almost invariably chosen for this task is· aluminum. It shows about
the highest ratio of visible light extinction to soft x-ray extinction of any metal and
is self passivating in air. It is also a safer material to work than beryllium. We have
calculated the transmittance of light as a function of metal thickness using Airy's
formulae 6 and assuming normal incidence. The calculations show that Al is superior
to Be for blocking light. For example, at a wavelength of 578 nm a pair of 20 nm
Al layers have a combined transmittance of 0.504%. This is to be compared with a
transmittance of 43.04% for similar Be layers. Even though Be has a low mass absorp-
tion coefficient for x rays, the thickness required makes it inferior to Al for blocking
transmission of visible and UV light to Si(Li) detectors. We have not found a metal
superior to Al in this regard. The soft x-ray transmission of various thicknesses of
Al are shown in Figure 6. In the visible range 40 nm of aluminum extinguishes more
than 95% of the incident light. The overall transmission of the x rays is the product
of the film and the aluminum transmission.
The x-ray transmission is only one issue. The window must possess mechani-
cal properties commensurate with the task of maintaining a pressure differential of
approximately one atmosphere for long periods of time and/or under cycling condi-
tions. The properties which go into achieving this include film mechanical strength,
film perfection, and the support structure. It must have sufficient strength to survive
a pressure differential of over one atmosphere and sufficient rigidity not to flow. For
an aluminized polymer such as polypropylene, flow or stretching can cause cracking
of the aluminum thus compromising the window.
As has been discussed above, it is desirable for transmission purposes to make
a window as thin as possible. As a material is made thinner, there comes a point
at which it is not mechanically robust or reliable. This will depend upon the area
being spanned. The larger the area, the greater is the practical limiting thickness.
It was long ago realized that with some kind of support grid, effectively larger areas
could be covered with thin films. Recently our group at BYU developed a method
of making a very close-spaced grid for the support of thin films for the purpose of
soft x-ray windows for Si(Li) detectors. See Figure 7. A patent is pending for the
method and resulting structure. The method consists of micromachining a grid from
silicon. Typically the structure has ribs 25 11m wide separated by 100 11m, resulting
in a fractional free area of 80%. Windows with such a support structure have been
made using formvar, Si 3 N4 , a-C:H, and other materials.
50
20
'"
G1
"0
>-
0
10
'0
'"c:
"tJ
III
5
'"
::l
0
.t::
t-
85 80 80 70 80 50 40 30 20 10 5
Cumulative Percent Remaining
Figure 8. Pressure cycle results for 400 nm of aluminized formvar on 80% transmis-
sion grid.
With such a structure, film strength rarely is a serious issue. Calculations show
that as little as 1 nm of SbN4 , if the material were to possess bulk properties 111HI
were free of defects, when stretched over a 200 {tm opening could support a pressur.·
differential of one atmosphere. Such a thin film is not practical since thin films rarely
possess bulk properties and a single weak spot could render worthless an otherwise
acceptable window. Window integrity is thus usually more important than bulk yield
strength as a window selection criterion. It is well known that film perfection depends
on film formation. Such factors include care, cleanliness, and choice of formation
techniques.
Film structure must fulfill another role with Si(Li) detectors. It must hlock the
diffusion of gases into the region of the detector. this puts further constraints on
the microstructure of the window membrane and particularly discriminates against
polycrystalline materials such as diamond. For polymer windows for Si(Li) detectors
then, gas permeability is an additional reason for aluminization.
CONCLUSION
Criteria for designing soft x-ray windows have been discussed. It was shown
that a thin material containing several elements, most notably 0, N, and C will
622 IX. XRF INSTRUMENTATION
APPENDIX
where log = 10g10, T is the independent parameter; in our case it is the number
of cycles to failure, and T50 is the number of cycles for 50% of the samples to fail. It
is very difficult to obtain sufficient data to "divine" the probability density itself, but
it is easy to get the cumulative probability distribution.
The cumulative probability distribution is
F(T) = l x =T 1
-exp{--(
x=o J'Ei 2
1 log
a
x
T5°?}d(
log
a
x
T5o ),
REFERENCES
3. G.D. Borley, et aI, "A Method of Producing Mounted Thin Windows for X-Ray
Spectrometer Flow Counters", X-Ray Spectrometry, Vol. 5, pp 13-15, 1976.
4. F. Williamson, et aI, "Thin Films for X-Ray Astronomy", Rev. Sci. Instrum.,
vol. 46, pp 50-52, 1975.
B. Stocklassa
National Laboratory of Forensic Science
Linkoping, Sweden
ABSTRACT
INTRODUCTION
In the last decade there has been a rapid development within the field
of SEM (Scanning Electron Microscopy) and microbeam PIXE (Protron Induced
X-Ray Emission). These techniques, which utilize the characteristic x-rays
from different elements, have enhanced the possibility to determine the
spatial distribution of elements in a sample on a microscale level. However,
in order to determine trace amounts of an element from the intensity of its
characteristic radiation, particle-induced x-ray emission methods might be
inadequate due to the severe background of x-rays generated by secondary
electron emission.
x-ray tube and uses a capillary optical system for generating microbeams
down to a few microns.
CAPILLARY OPTICS
X-ray optics have been developed for over thirty years and much effort
has been devoted to constructing intense x-ray microbeams. The basic
difficulties in constructing an x-ray microbeam are related to the problem
of focusing a divergent x-ray beam into a focal point. As the attenuation
of x-rays is normally very strong, x-ray optics are normally based on mirror
optics. However, as the refractive indices of x-rays are very close to
unity the reflectivity for x-rays is normally extremely small and reasonable
reflectivity can only be achieved below the critical angle. As the critical
angle for x-rays typically is in the range of a few seconds of are, mirror
optics for x-rays will generally have a very large focal length and low
efficiency.
f.
R - -- -
---
Fig. 1. Micrograph of a capillary cross section with an
inner diameter of 19 ~m. (The grid size is 25 ~m)
S. LARSSON ET AL. 625
THE SPECTROMETER
4 300
'i
';:'3
::>
.r:.
%2
I
tU
0
200
rJ)
100
RESULTS
Hairstrand crossection
30
• a I •
20
•
g 10
Q)
E
•
0
.~ -1 0
E
-20
• • • •
-30
0 20 40 60 80 100 1 20
micrometer
Fig . 5. Hairstrand cross section as Fig. 6. Micrograph of the 19 ~m
determined from the intensity capillary end and a hair
of scattered primary radiation . strand (grid size 25 ~m) .
Corrections for the attenuation
of x-rays inside the sample
material have been taken in
account.
S. LARSSON ET AL. 627
Fig. 7. Micrograph of a leaf sample. The graph in the figure shows the
sample thickness profile as determined from the intensity of
scattered primary radiation. Scan increment is 20 11 m.
Note!- the inorganic particle incapsulated inside the leaf rib.
This particle contains high concentrations of Ca, Ti and Fe.
Leaf-Scan Analysis
A birch leaf was scanned along the line indicated in Fig. 7 with the
use of a 19 11m microbeam Cr characteristic x-ray. The stepsize was 20 11m
and the intensity of characteristic radiation as well as scattered primary
Titanium Sulphur
Q) Q)
~
If)
~
If)
.ri
~
.e~
c: c:
o 0
~ ~
~o 'E
c:
~
c:
o 0
o 0
o 200 400 600 800 0 200 400 600 800
Distance (~m) Distance (~m)
CONCLUSIONS
REFERENCES
ABSTRACT
As the Environmental Protection Agency adds ever more sites to the National
Priorities List for Superfund clean-up, there is a tremendous need for a fast method of
screening for contaminants. In the case of inorganics, we can apply field-portable X-ray
fluorescence technology.
NOTICE: Although this research was funded in part by the U.S. EPA through
Contract 68-03-3249 to Lockheed Engineering & Sciences Company, it has not undergone
Agency review and does not necessarily reflect Agency policy.
INTRODUCTION
As the number of hazardous-waste sites continues to grow, the EPA has a growing
need for reliable analytical data In 1980 the EPA established the Contract Laboratory
Program, CLP, to specify standardized analytical methods and strict quality-assurance/quality-
control (QAlQC) protocol for the participating laboratories (U.S. EPA, 1986; U.S. EPA, 1989).
The QAlQC protocol includes intensive data and lab auditing by teams independent of the
labs. The resulting data is of known quality, and is ideally defensible in a court of law.
Under this system the cost of analyses is considerably higher with a much greater turnaround
time than for unregulated laboratory analyses.
Lockheed Engineering and Sciences Company contracts to the EPA in Las Vegas,
Nevada to provide technological expertise in analytical chemistry, remote sensing and
logistical support. Under this mandate, we have been developing methodology for the use of
field-portable X-ray fluorescence in characterizing metal contamination in soils on hazardous-
waste sites (Raab, et ai, 1989; Raab, et ai, 1987). The goal of this work is to minimize both
time and expense, while providing data of known quality.
The traditional soil sampling and analysis pathway (Figure 1) is as follows: the sample
is taken, packaged, decontaminated, and sent by overnight courier to a CLP analytical lab for
chemical analysis. The results are then delivered to a data audit group for review. The
verified data are sent back to the EPA project manager, who meets with his contractors to
make the decisions on how to proceed. This process is lengthy and expensive with sample
turnaround of 21 to 45 days, at a cost of $200.00 to $4oo.00/sample for 24 analytes.
Field portable XRF (FPXRF) technology can be employed in site evaluation to save
time and money by meeting the following objectives:
o Develop an initial database from rapid on-site in-situ analysis, which requires no
intrusive sampling.
o Provide further analysis for questionable areas, after evaluation of the initial database.
o Allow optimized selection of samples for laboratory analysis, after evaluation of the
enhanced database.
~, EPA
.O-z:_ ::o>I--.Y
JI II
EP A
CONTRACTons
ffi
OUI.~======v
Figure 1. Traditional Pathway for Sampling and Analysis
G. A. RAAB ET AL. 631
EQUIPMENT
180
170
160
150
140
130
120
C>
"""-
C>
110
100
E
.; 90
II
0
80
::l
iii 70
> 60
SO
40
30
20
10
0
Cu Zn A. Ph
Elemenls
[22] lCP/AA [SS) EDXRF fiLZl 'CP/AA CS§j EDXRF
In-situ measurements are acquired by placing the probe directly on the soil, opening
the shutter, and taking a measurement of predetermined duration, usually 15 or 30 seconds.
The fluorescence intensity data is converted to concentrations by previously developed
calibration memories, or models. The electronic unit has 32 such calibration models. Each
model can be independently calibrated for as many as six elements. These can be used to
measure elements from silicon to uranium assuming the proper isotope source is available.
The routine data is recorded by hand as this unit has no memory for analyses.
PROCEDURE
The crux of FPXRF analysis is in the instrument calibration. Our instrument has no
fundamental parameters capability, so compensating for matrix effects must be done
empirically using calibration standards.
~
SITE SPECIFIC
MATCHES MATRIX WITH RESPECT TO:
The instrument is first calibrated with a set of spiked soil samples, or site typical
standards. Numerous measurements are then taken over the site to obtain a wide range of
values typical of the site. This screening In-situ allows us to get a relative range of
concentrations for the analyte(s) of interest, from which we can better select samples to be
characterized and used as site-specific calibration standards. Physical samples are then
taken, dried, sieved through a 2mm mesh, homogenized and split. One split undergoes total
digestion in a Paar bomb (aqua/hydrofluoric regia/boric acids) and is run on AAS or ICP.
The total digest is used rather than the standard EPA extraction procedure, which is a nitric
acid/hydrogen peroxide leach, to avoid potential discrepancies between XRF totals and EPA
extraction results. Another split of the same soil is then measured on the XMET 880, and the
intensities are compared to the values obtained by the total digestion process. In this
manner a calibration curve using these site specific standards is constructed, and matrix
effects are minimized. Figure 3 illustrates a calibration curve used on a recent waste-site
analysis.
G. A. RAAB ET AL. 633
c
o
1.5
III
!::
c
1Il.Q
llCl.
c
O/fl
~i 1.0
III *
III
E *
II)
UJ 0.5
=
Correlation Coefficient: R 0.995
O.o~--------.--------.--------.--------,---------.----
0.0 0.4 0.7 1.1 1.5 1.9
Actual CLP Concentration
wt% Pb
Figure 3. Calibration Plot for Pb
Once the calibration curves for the analytes are generated from the site-specific
standards, the In-situ analysis can be performed.
In this step measurements are taken across the site on a grid pattern. The data is
recorded and transferred to a computer, and through the use of geostatistical software and a
plotter, concentration isopleth maps are constructed. These can be utilized to guide further
sampling or remediation efforts. For example, during the course of the remediation activity,
realtime maps can be produced on a daily basis to chart progress and delimit areas where
further work may be necessary.
Standard Pb Cu Cr
Deviation All values in mg/kg
1 sigma 25 50 15
3 sigma=ID 75 150 45
r-
LIMIT OF
aUANTITA TION
10 a
- ,-
INSTRUMENT
DETECTION LIMIT
1 a
o
-
Sample
A
Sample
8
Sample
C
n
Sample
D
Figure 4. Visualization of Instrument Detection Limit
and the L1mH of Quantltatlon
At high concentrations, 10 sigma and above (Figure 4, Sample A), our measured
values show excellent correlation to laboratory values. Physical samples from this range are
not really necessary (ct. Table 1). The questionable measurements are those below 10 sigma
(Figure 4, Samples B, C, and D), and here corroboratory samples need to be taken and sent
to a laboratory.
Once the data are collected, processing begins. In general the data points are
assumed to be spatially autocorrelated, so that data analysis utilizing the theory of
regionalized variables, or geostatistics, can be employed to derive maps of element
distribution over the site. If this is not the case, then alternative interpolation methods are
used to contour the data.
AN EXAMPLE
The following three figures represent data from an hazardous waste site analyzed for
Pb. Figure 5 is a map of the site, showing the sampling locations and buildings on site.
Figure 6 is a spatial model of the Pb concentrations generated from the sample data. Using
this model, the software constructs a contour map of Pb concentrations, Figure 7.
Using the grid that was previously established, the 119 sample points were measured
in less than two days. At each sample point, field analysts made three measurements and
G. A. RAAB ET AL. 635
•
~ • • •••., •
~
• • • •
:::&r(~
•.....
• •••
•••• *~ -
....
~ •
~
:' - L,
•
•
• \
~
•
•
,--
•.,
., \
1
•
•
.,•
•
•
•
...•
•
., .,
D
., ., .,., .,
.,
• ., .,
., ., •
., ., .,., •
.,
• ., .,
•
., .,
••
., ~ ., ., f
0 ~ ,n
'-mi-'
ILKHC :ILL'LI
'.soo
1 .000
3.500
~
~ 3.000
,.
.'"
w 2.500
u
w
2.000
'"
u
w
1.500
" 1.000
0.500
0.000
nnn
t.~~.
\tJ~o
used an average value calculated in the data processing stage. The three measurements
help to compensate for variability in the soil, and increase sample representativeness.
CONCLUSIONS
Field portable XRF has been demonstrated to play a useful role in the hazardous
waste industry. It is economic of both time and money, as well as being a valuable tool to
aid in assessment planning. With the number of contaminated sites growing steadily, it offers
exciting potential for future applications for several reasons:
o We can effectively use in situ analyses to routinely analyze soils on hazardous waste
sites providing valuable information in realtime with defensible quality assurance
procedures.
REFERENCES
Columbia Scientific Industries Corporation (1988). X-MET 880 Portable XRF Analyzer
Operating Instructions.
Raab, G. A., M. L. Faber, and S. J. Simon, (1989). Development of a Field Portable X-Ray
Fluorescence System for On-Site Hazardous Waste Screening. !n: Proceeding of the
Thirteenth Annual British Columbia Mine Reclamation Symposium. Mining Association of
British Columbia, pp. 137-172.
U.S. EPA (1986). User's Guide to the Contract Laboratory Program. Office of Emergency
and Remedial Response, Washington, D.C., 20460.
U.S. EPA (1989). Contractor Laboratory Program Statement of Work for Inorganic Analyses,
SOW No. 788, Attachment A.
XRF TECHNIQUE AS A METHOD OF CHOICE FOR ON-SITE ANALYSIS
Stanislaw Piorek
Princeton Gamma-Tech
A Member of the Outokumpu Group
1200 State Road
Princeton, New Jersey 08540
ABSTRACT
Major difficulties still inhibiting wider use of the XRF for hazardous
waste screening stem from the lack of suitable calibration samples and
proper protocols, as well as from the low awareness of the industry about
the technique.
INTRODUCTION
While the solid state detectors can completely resolve K x-rays from
neighboring elements, they mostly still require cooling for optimum
performance; this often contradicts the idea of portability. Gas filled
proportional detectors have been recognized traditionally as devices of
mediocre resolution (20% for the Fe K-alpha line). New developments in
proportional-counter technolog~ have yielded detectors with much better
energy resolution (10 to 12%). This, coupled with modern microprocessor
technology, has resulted in the commercial availability of a hand-portable
x-ray analyzer capable of simultaneous multielement analysis. 3
CALIBRATION
o
o
o i
I
i
~~~----,_---r----r_--,_--~----,_--_r----,_--~--_,-~
4 6 8 10 12
(Thousonds)
mg/kg Arsenic (ci,em.)
o
Calibration for Arsenic in Soil
S. PIOREK 643
Element: Cr Cu Zn As Pb (Lf3) Cd
Source: 6 mCi Cd-109 or 100 mCi Cm-244 6 mCi 100 mCi 30 mCi
Cd-109 Cm-244 Am-241
RMS
Error +/-100 +/-180 +/-150 +/-200 +/-180 +/-220 +/-130
Detection
Limit 200 100 100 120 100 200 50
Table 4. Results for Total Chlorine in Waste and Fuel Oil. mg/kg
About 500 million gallons of used oil are burned each year in boilers
and industrial furnaces in the U.S.A. Used oil is often mixed with spent
chlorinated solvents; therefore, a real danger exists that boilers may emit
hazardous levels of solvents to ambient air, creating a cancer risk to
humans.
In 1985, EPA published the rules which ban sale of used oil for fuel
to nonindustrial users if the total halogen content of the oil exceeds 1000
mg/kg. Since it is not easy to determine a total halogen content, EPA has
agreed to interpret "total halogen" as total chlorine. As a consequence,
the affected community will have to determine total chlorine in used oils
and fuels.
CONCLUSIONS
REFERENCES
A.R. Harding
Tracor Xray, Inc.
345 E. Middlefield Rd.
Mountain View, CA 94043
J.P. Walsh
James P. Walsh and Assoc.
Boulder, CO 80306
INTRODUCTION
Effective assessment and remediation of hazardous waste sites
dictates that analytical methodologies be developed which
assist in the evaluation of site contamination and
simultaneously make efficient use of sampling time and
resources (1). Optimally, a technique would provide on-site
personnel with immediate and accurate information concerning
the identity and concentration of inorganic soil contaminants
(2) •
EXPERIMENTAL
The field mobile EDXRF spectrometer used in this work was a
Tracor Xray Spectrace 6000 (Tracor Xray, Inc., Mountain View,
CA). The EDXRF system consists of three modules: the
spectrometer, the control/pulse processing electronics, and
the data analysis computer. The compact size and weight (90
lbs.) of the modules permits installation of the system in a
laboratory trailer or van.
The bench-top spectrometer module, which can accommodate a
single soil sample, is powered by 110 V line or generator
feed. The excitation source used is a low-powered Rh-anode X-
ray tube (50 kV, 0.35 IDA (17 W) maximum output) positioned at
a 45° incident angle to the sample. Three primary radiation
filters permit optimum spectral acquisition conditions to be
computer selected.
The thermoelectrically cooled Si(Li) X-ray detector is
mounted at a'45° take-off angle in an inverted geometry with
respect to the sample. The 20 mm 2 Si(Li) crystal, which is
protected by a 0.5 mil Be window, is reduced to -90°C for
operation using a mUlti-stage thermoelectric (Peltier effect)
cooler. The 300 watts produced at the detector heat sink are
dissipated by forced ambient air. Thermoelectrically cooled
detectors provide typical resolutions of 185 eV (Mn Ka) •
A card-cage module is interfaced between the spectrometer and
an IBM PS/2 or PC/AT series computer. The microcomputer
serves as the contour mapping computer as well as the
spectral data acquisition/analysis module. The card-cage
components include the detector high voltage supply, the
pulse processing electronics, and the control circuit board
for the EDXRF spectrometer.
Sampling of the suspected waste site was performed using EPA
approved protocols in a Q500'x 3500' rectangular area. Forty
three (43) cores were collected and were partitioned into
four depth levels: surface to 2"; 2" to 6"; 6" to 12"; and
12" to 18", and designated levels 1 through 4, respectively.
At the site, samples were first homogenized and then split
into two fractions. One was submitted for EDXRF analysis and
the other sent to an independent lab for AA/ICP analysis.
The independent laboratory'used EPA SOW 846 methodology to
determine contaminant concentrations in the soil sample
split.
Sample preparation for the EDXRF analysis consisted of drying
the sample for 4 min. in a microwave oven followed by sieving
the dried sample to 2 mm. Sieving removes large foreign
objects such as pebbles and sticks. Drying the sample was
required due to the variable moisture content in the
submitted soils; some surface samples had the consistency of
mud. The sieved soil was then ground in a Spex shatterbox
grinder (Spex Ind., Edison, NJ) using tungsten carbide cups
for 2 min. The grinding cups were cleaned using soap and
tapwater. The cleaned cups were rinsed with
distilled/deionized water followed by isopropanol. The
A. R. HARDING AND J. P. WALSH 649
,
e Z
r
I
I 'I~
I~ l ,/,,1
~.
to contain 125 ppm As, 1100 ppm Pb, and 729 ppm Zn. A
multiple linear least-squares peak fitting routine was used
for deconvolution of overlapped peaks.
The soil characterization method was standardized using four
standard reference materials (SRM): NBS 1648 (urban
particulate); NBS 2704 (river sediment); SO-l and SO-3, two
soil standards available from the Canada Centre for Mineral
and Energy Technology. Standards labeled NBS are available
from the National Institute for Standards and Technology
(NIST). These SRMs have certified concentrations of Fe, Mn,
Cu, Zn, Pb, and Cd.
A fundamental parameters (FP) method was employed as the data
treatment scheme and used the certified concentrations of Fe,
Mn, Cu, Zn, Pb, and Cd in the four standard materials. To
compute the instrumental sensitivity, the balance of the
standard was assumed to be comprised of SiOS to account for
the contribution of the matrix to the measured analyte X-ray
intensity. Arsenic sensitivity was determined using a
theoretical calculation based on the computed Zn sensitivity.
The selection of an FP method for standardization allows well-
characterized SRM's to be the basis of analyte sensitivities
and eliminates the need for site-specific soil standards.
The FP method permits accurate analyte concentrations to be
determined in samples with wide matrix variation without
restandardization. Theoretical analyte sensitivities can be
computed by the FP routine when a standard with a certified
analyte concentration is not available.
RESULTS AND DISCUSSION
Table 2 lists the calculated (3jB) lower limits of detection
for the two sets of spectral acquisition conditions used.
For the analytes of specific environmental concern, the LLD's
are at the 10 ppm concentration level. Calculated LLD values
could be improved with longer spectrum-acquisition livetimes
and alternate excitation conditions.
1600 0 0
§
~ 1200
>-
CD
Ea. BOO
-5
CD
a..
400
1600
PB (ppm) BY EDXRF
350 0
0
280
a..
~
0
>- 210
CD
0
Ea. 0
-5 140 0
0
u
0
70
50-1 Pb 41 14 21
Zn 129 147 146
50-2 Pb 19 17 21
Zn 55 123 124
A As 45 ± 4 41 ± 3
Cd 20 ± 2 31 ± 3
Pb 286 ± 28 312 ± 12
Zn 185 ± 15 134 ± 10
B As 17 ± 3 14 ± 1
Cd 80 ± 6 58 ± 4
Pb 141 ± 15 158 ± 3
Zn 556 ± 39 529 ± 46
C As 17 ± 1 19 ± 4
Cd 10.0 ± 0.9 9 ± 0
Pb 117 ± 8 142 ± 14
Zn 173 ± 26 128 ± 3
A. R. HARDING AND J. P. WALSH 653
CONCLUSION
Field-mobile EDXRF analysis of soils suspected of being
contaminated provides information concerning the nature,
extent, and magnitude of the contamination. Due to the
minimal sample preparation necessary for EDXRF analysis,
sampling-to-result turnaround time is relatively short so the
most effective use of sampling resources is realized. The
plume of contamination is illustrated to on-site
assessment/remediation personnel in the form of concentration
contour plots. Accuracy and precision for the analytes of
interest using the EDXRF method was shown to be comparable
to results obtained with AA/ICP analysis. comparable results
for Cd, As, Pb, and Zn for the two methods validates the use
of EDXRF analysis for hazardous waste site investigations.
ACKNOWLEDGEMENTS
The authors would like to acknowledge L. Gilfrich and D.
Leland for assistance in the EDXRF analysis of the soil
samples, G. Bogart for data analysis work and K. Proctor for
assistance with graphics.
REFERENCES
(1) H. Vincent, Field screening Methods for Hazardous Waste
site Investigations Symposium Proceedings, 61, (1988).
(2) R. Perlis, M. Chapin, IILow Level XRF Screening Analysis
of Hazardous Waste Sites ll , Field Screening Methods for
Hazardous Waste site Investigations symposium
Proceedings, 81, (1988).
(3) N.W. Madden, G.H. Hanepen, B.C. Clark, IIA Low Power High
Resolution Thermoelectrically Cooled Si(Li)
Spectrometer II , IEEE Trans. Nuc. Sci., 33, (1), 303,
(1986) .
(4) Data Quality Objectives For Remedial Response Activities,
U.S. EPA, EPA 540/G-87,003, Appendix A, (1987).
SCREENING OF HAZARDOUS WASTE WITH AN ENERGY DISPERSIVE
N.C. Jacobus
HNU X-RAY
1003 Larsen Drive, Oak Ridge, Tennessee 37830
INTRODUCTION
INSTRUMENTATION
SAMPLE PREPARATION
All samples were analyzed as collected in 1.25-inch diameter
polyethylene cups with < 4 J,£m Mylar film windows. wet
samples were air dried. Selected samples were ground for
comparison. All samples were analyzed at 100 seconds live
time.
Cu Zn
Pb
300
Pb
....
III II
200~
Z
:J
o
U
I
Fe
4 6 8 10 12 14 16 18
keY
j
a, 3000
S 3
...J w
~ 2000 c
iii
! t;
c 1000 w
~
...J ~
5 4 3 2
HOUSE SIDE 251eet
i.... 3000
l
5 2000
1/1
!
c 1000
cW
....I
5 4 3 2
HOUSE SIDE
251eet
a.>t
....
-~--R
g) 3000
§.
g 2000
oJ
-Fil--lJ)
--UJ--
:!!:
~ 1000
w
oJ -~-~--g-/
SOUTH SIDE OF LOT
20 feel
SOIL CONTAMINATION
CHROMIUM-COPPER ARSENATE (CCA)
300
As
260
220
180
~
c:
"
0
0 140
100
Cu
60
20
3 5 7 9 11 13 15
keV
Figure 6. The ~-ray energy spectrum for a
soil sample contaminated with chromium-copper
arsenate at an abandoned wood preservative
plant using 10 mCi Cd-lOg radioisotope source.
N. C. JACOBUS 661
""""-''-I'"""-f-+-+--+--+-+-+-+--IB
10 9 8 7 6 5 4 3 2
~
50 feel
Fiqure 7. 3-D graphic representation of EDXRF
quantitative analyses of arsenic in soil
samples at an abandoned wood-preservative
plant.
5000
III
'"....
III
E
.!i
c
3000
GI
!!
<
1000
Depth in Feet
CONCLUSION
REFERENCES
1 Data Quality objectives for Remedial Response
Activities, Example Scenario, united states Environmental
Protection Agency, EPA/540/G-87/004, March 1987, p. 4-7.
National Technical Information Service, PB88-131388.
2 Furst, G.A., V. Tillinghast, and T. Spittler, Screening
for Metals at Hazardous Waste sites: A Rapid Cost-Effective
Technique using X-Ray Fluorescence. Proceedings of the
National Conference on Management of Uncontrolled Hazardous
Waste sites. Hazardous Materials Controls Research
Institute, washington, D.C., p. 93-96, 1985.
3 Raab, G.A., D. Cardenas, S. J. Simon, and L.A. Eccles,
Evaluation of a Prototype Field-Portable X-Ray Fluorescence
System for Hazardous Waste Screening, Environmental
Monitoring Systems Laboratory, Las Vegas, NV, EPA/600/4-
87/021, National Technical Information Services, PB87-
227633, August 1987.
4 The standards and samples were provided by the
California Department of Health EETS. The wet chemical
analytical results were obtained by AA spectroscopy (HN0 3
digestion).
5 All the 3-D graphic plots were made with a "3-D
Graphics," enhanced graphics for Lotus users Copyright (c)
4-5-6 World Ltd/Ross Research 1987. Software Engineering by
Ross Research PS2 Version 1.02.
6 Quantitative results were obtained using an IBM AT
compatible personal computer, HNU's data acquisition software
based on Lotus 1-2-3/Lotus Measure, and HNU's ATAC regression
analysis software program. IBM AT is a trademark of
International Business Machines, Inc.
X-RAY FLUORESCENCE ANALYSIS OF TRACE METALS IN THE ANNUAL
GROWTH LAYERS OF FRESHWATER MUSSEL SHELLS*
INTRODUCTION
*Work supported in part by a grant from the Center for Reservoir Research,
Murray State University, Murray, Kentucky.
2nd year
\
peripheral nacre
PROCEDURE
Each shell was thoroughly cleaned and separated into two separate valves.
Each valve was baked in a furnace at 3S0 oC for approximately 30 minutes.
This baking process destroyed the protein material holding together the layers
of CaC03 crystalline matter thus allowing the layers to be easily separated
using a sharp blade and a pair of tweezers. Baking the shells for shorter
periods of time or at lower temperature make it more difficult to separate layers
while over-cooking the shells caused them to crumble when handled.
Once the shells were separated into their annual growth layers, the material
was pulverized in a ball-mill grinder then pressed into a standard 1 1/4" pellet
suitable for XRF analysis. The XRF spectrometer and the method of analysis
have been described in a previous article (5).
RESULTS
The mussel shells were collected from a site at the 44.S-mile marker on
Kentucky Lake in Western Kentucky at a depth of approximately 7 meters. Ten
121111~---------------'
71 73 75 77 79 81 83 85 87
Growth Yeor
Fig. 2: Layer concentrations of Sr in ten different Megalonajas gjganJ.e.a
shells.
W. E. MADDOX ET AL. 667
1280~---------------'
71 73 75 77 79 61 63 65 67
Growth Yeor
Fig. 3: Layer averages of Sr concentrations in each of the three species.
1~8;----------------'
n n ~ n ~ ~ ~ ~ ~
Growth Yeor
Fig. 4: Layer concentrations of Fe in ten different Megalonajas ~
shells.
668 X. XRF TECHNIQUES FOR HAZARDOUS WASTES AND OTHER APPLICATIONS
1280..-------------------,
KENTUCKY LAKE SITE 113
QQ, AP AND MG AVERAGES
-a 968
.9-
c.
~ 728
T-
o
L.
T-
&
u
488
c.
o
u
~ 240 00 /AP
MG ..... . -- --. -- '>"':"-.". :.::,::.,. . -~----t
71 73 75 77 79 61 63 65 67
Grollth Yeor
Fig. 5: Layer averages of Fe concentrations in each of the three species.
1280..--------------------.
KENTUCKY LAKE SITE 113
MEGALONAIAS GIGANTEA
a 968
.9-
c.
0 728
T-
o
L.
T-
c.
IV 488
u
C.
0
U
c. 240
~ ~_v-
71 73 75 77 79 61 83 65 67
Growth Yeor
Fig. 6: Layer concentrations of Mn in ten different Mega!onajas ~
shells.
1280 . . - - - - - - - - - - - - - - - - - - - ,
KENTUCKY LAKE SITE 113
QUADRULA QUADRULA
a968
a.
c.
~ 728
T-
o
L
&
T-
488
u
o
c.
u
c. 248
~
71 73 75 77 79 61 63 65 67
Grollth Yeor
Fig. 7: Layer concentrations of Mn in 10 different QuaQ.w!a ~ shells.
W. E. MADDOX ET AL. 669
1288 . . . . - - - - - - - - - - - - - - - - - ,
KENTUCKY LAKE SITE 113
-a
AMBLEMA PLICATA
968
a.
c
~ 728
t-
o
L.
t-
iu 488
c
o
u
c 248
:l:
71 73 75 77 79 61 63 65 67
Growth Yeor
Fig. 8: Layer concentrations of Mn in ten different Anl.b.lwna lllil:ala shells.
1288 , . . . - - - - - - - - - - - - - - - - - - .
KENTUCKY LAKE SITE 113
QQ, AP AND MG AVERAGES
a
a.
968
c
~ 728
t-
o
L.
t-
iu 488
c
o
u
c 248
:l:
MG --
71 73 75 77 79 61 63 65 67
Growth Yeor
Fig. 9: Layer averages of Mn concentration in each of the three species.
DISCUSSION
Clearly, the most interesting result of this work is the pattern of the
layer-to-Iayer concentrations of Mn in the shells. The higher Mn concentrations
in the more recently developed layers could indicate an increase in the Mn
concentration in the sediment of the lake where the mussel feeds, or it could be
a natural process that takes place in the shell development. Further studies
will have to be done on the relation between sediment concentrations and Mn
uptake by the mussel before definite conclusions can be drawn concerning this
Mn pattern. This work will continue with shells gathered from three additional
sites on Kentucky Lake. The Mn concentrations in the sediment of these sites
has been determined to be different from site to site (6) and an effort will be
made to correlate the shell concentrations with the sediment concentration.
ACKNOWLEDGEMENTS
The authors wish to thank Chris Blewett and Mike Keeling, undergraduate
students at Murray State University, for their many hours spent cooking,
separating and grinding the mussel shells used in this work.
REFERENCES
6. Owen, David A. and York, Keith A., "A Study of the Forms, Concentrations,
and Interphase Dynamics of Iron and Manganese in Northern Kentucky
Lake Benthic Waters, Elutriates, Sediments, and Other Lacustrine-Based
Matrices", Report to Center for Reservoir Research, Murray State University,
Murray, Kentucky. 39 pp., 1988.
APPLICATION OF PIXE METHOD FOR ENVIRONMENTAL
PROTECTION IN POLAND
T. Zoltowski*, W. Ratynski**
M. Jaskola**, J. Parus*, L. Wawrzonek*
sot-.
gaseous S02 and during gas-to-particle conversion it may become oxidized to
As sulfates it can occur first as sulphuric acid, depending on the
environmental conditions, and later can be neutralized by ammonia to
ammonium bisulphate and finally ammonium sulphate. The concentration of
sulphur dioxide and nitric oxides in air over the territory of Poland is
excessive, (expected emission of S02 and NO x in Poland: 1990-(S02-4.9, NO x
-1.7); 1995-(S02-5.1, NO x -2.0); 2000-(S02-5.5, NO x -2.3) [min t./year]. Due
to the country regulations and international agreements until 1993 Poland
should limit emission of the mentioned pollutants about 30%.
in the 1990's. The experience of Polish researchers with using the 2.8 MeV
Van de Graaff accelerator in Warsaw Center and the 5 MeV cyclotron in Cracow
Center during the last years will certainly be utilized. Actually, apart
from using the C-30 cyclotron for PIXE measurements, this machine will be
utilized for Proton Elastic-Scattering Analysis.
The chamber for beam preparation and holding a sample was constructed
in two versions. A first one has been used during PIXE experiments with the
Van de Graaff accelerator 4 , the second was designed according to Japanese
suggestions. Electronics consist of an ORTEC Si/Li detector located in the
backward direction of 135 0 to reduce bremsstrahlung background and an ORTEC
MCB 800A system with an IBM/PC computer for spectra evaluation.
The PIXE method will be devoted in the first place for quick
determination of sulphur species pollution in Poland territory. However the
concept to distinguish between PIXE spectra according to the sulphur
oxidation state could lead to a method for the determination of acidic forms
of H2S04 and the NH4HSO. The PIXE method will be used for the elemental
constituency measurement of coal waste products, due to an urgency to
utilize coal waste material which is in great quantity in Poland.
REFERENCES
INTRODUCTION
The assessment of the potential environmental and toxi-
cological effects of particulate material emitted to the at-
mosphere requires detailed physical and chemical characteri-
zation of the particles. One of the most widely studied types
of pollutant particles is coal fly ash as a byproduct of coal
combustion. These particles are inhomogeneous, highly varia-
ble, span a broad range of sizes and have diverse morphologies.
It has been shown that numerous toxic trace elements tend
to increase in bulk concentrations with decreasing particle
size (1).
The emission of black coal-fired power plants has been
discussed in several studies but the chemical formation of the
toxicological impact of brown coal or lignite-fired power
plants, typical for the geographical area of Middle-East Eu-
rope, is much less clear. In the mentioned previous studies
different analytical techniques were applied to bulk samples
and an average chemical composition was obtained for both the
larger particles and the smaller ones.
In a few studies, the morphology of individual fly-ash
particles was (2) also analysed with electron microprobe X-
ray analysis EPXMA (3). LAMMA technique also is commonly used
for such purposes but has some limitations with regard to
quantitative work because of the occasionally poor reprodu-
cibility of the laser power density and the heterogeneity of
the individual particles (4).
The aim of the present paper is to explore the feasibi-
lity of using EPXMA and SRXRF as highly sensitive and non-
destructive multielemental techniques to analyse trace and
especially toxic elements in individual particles combined
with the conventionally used bulk XRF techniques.
EXPERIMENTAL
Sample
origin Cu Zn As Pb Sb
8- fly ash
A - fly ash
EPXMA SRXRF
25 kV
15 keV 40 keV
CONCLUSION
SRXRF as analytical technique can be used efficiently to
measure toxic elements in small environmental samples of
different size fractions. Efficient focusing techniques should
be used for analysing individual particles of 10-100 )lm dia-
meter where lower concentrations could be determined than by
EPXMA for elements with Z>40.
ACKNOWLEDGEMENTS
This work was partially supported by IAEA Coordinated
Research Project No.4959/Rl/RB and by Hungarian National
Scientific Research Fundation under contract numbers 1030 and
1028.
REFERENCES
1. R.L. Davison, D.F. Natusch, J.R. Wallace. and J.R. Evans,
Environ.Sci.Technol. 8: 1107 (1974).
2. R.J. Lauf, L.A. Harris and S.S. Rawiston, Environ.Sci.
Technol. 16: 218 (1988).
3. Y. Mamane and T.G. Dzubay, Water, Air and Soil Pollu-
tion 37 :389 (1988).
4. E. Denoyer, D.F. Natusch, P. Surkyn and F. Adams,
Environ.Sci.Technol. 17:457 (1983).
5. V.B. Baryshev, G.N. Kulipanov, E.Z. Zaytsev, Ya.V.
Terekhow and V.J. Kalynzny, Nuc.Instr.Meth.
A251:279 (1987).
6. P. Van Espen, R. Nullens and F. Adams, Nucl.Instr.Meth.
142:243 (1977).
678 X. XRF TECHNIQUES FOR HAZARDOUS WASTES AND OTHER APPLICATIONS
I NTRODUCTI ON
Zeolites-based fluid cracking catalysts are used in petroleum
refining. Zeolites are crystalline alumina silicates with a framework
based on extensive three-dimensional network of oxygen, and cations
which occupy nonframework positions. The distribution of silicon and
aluminum atoms within the framework determines many important properties
of zeolites, most notably catalytic, absorptive capacity, acidity,
stability, etc. For the zeolite scientist, information obtained from
the determination of silca:alumina ratio (SAR) and soda:alumina ratio
(NAR) during crystallization, dealumination and cationic exchange is
important for the optimization of preparative techniques and development
of new products.
Classical analytical wet techniques such as complexometric
titration, gravimetry and flame photometry are used for the
determinations. Recently application of atomic absorption spectrometry
(AAS) (1), and inductively coupled plasma-atomic emission spectrometry
(ICP-AES) (2), also have been reported. These techniques, though
accurate, are time consuming and therefore can't be applied when a
rapid analysis is required for a large number of zeolite samples.
Neutron diffraction, nuclear magnetic resonance (NMR) (3), infra-red
(IR) (4), Raman spectroscopy and X-ray diffraction (5), have been applied
for obtaining silica:alumina ratio within the framework, for elucidating
crystal structure and for locating cations and molecules within the
framework.
Whenever the primary objective is a determination of chemical
silica:alumina ratio and soda:alumina ratio, then the X-ray fluorescence
becomes the method of choice for the analysis because it is rapid,
simple to use, inexpensive and nondestructive. Multielements with
a wide variety of concentrations can be analyzed simultaneously with
high precision and accuracy. Application of borate fusion technique
for samples preparation eliminates grain-size influence and reduces
matrix effect, thus accurate analysis can be obtained for conventional
NaY zeolites as well as modern-day USY zeolites.
TABLE 1 STANDARDS
# 2 3 4 5 6 7 8 9 ID 11 12 13 14 15
%A1 203 24 23 20 30 28 18 21 19 22 17 31 19 16 21 13
%Si02 61 52 68 67 64 . 75 69 64 73 81 69 62 83 77 65
%Na 20 9 12 7 2 4 5 7 10 3 0 11 0.5 1 13
EXPERIMENTAL
Standards are prepared using precalcined high-purity alumina,
silica and sodium tetraborate. The sample in the slurry form is filtered
and washed with 01 water and then calcined at 300° for 15 minutes and
at BOO°C for 45 minutes. One gram of the calcined sample is fused
with 0.1 gram KI, 1.6 gram lithium rretaborate and 4.B gram lithium
tetraborate using Claisse-fluxer ® (6).
... 7
.~
V')
l-
I
.....
<.!l
w
3:
ACCR .06664
5 7 8 9
WEIGHT %5;02 (CORRECTED)
o
...t
!
!
-";r .... ... .
ON
ACCR .01105
1£1· - -
2 3
WEIGH T S A1
Z03 (CORRECTED)
scatter for the data points (Fig 1, Fig 2 and Fig 3). Better intensity
counts are desirable for the analysis of soda; it is possible to obtain
an accurate analysis even at a lower concentration of the analyte (Fig
2) .
1677 Y
1342
1006
II p
y
p
I
I P
Y
5
I
[I
y '(
I
670
II
I
y
III
y?
Y
5 S
334 P
YS
-1
Frank R. Feret
Alcan International Limited
Arvida Research and Development Centre
Jonquiere, Quebec, Canada G7$ 4K8
INTRODUCTION
Alumina is produced in the Bayer process in which crushed bauxite is
digested with hot sodium hydroxide solution. One of the products is
aluminum trihydrate which is then calcined [1]. Depending on the
degree of calcination, metallurgical or ceramic grade aluminas are
produced. They differ with respect to the alpha alumina content,
grindabi I ity, particle shape and size, impurity content, etc. [2,3].
Appl ication of X- ray fluorescence for the analysis of alumina in the
Alcan system dates back to the sixties. However, the instrumental
technology existing at that time al lowed only the analysis of inter-
mediate atomic number elements beginning with Ca. With time the X-ray
methods for alumina characterization were upgraded and the I ight ele-
ments were included. In this paper two recently developed sample
preparation techniques for XRF elemental analysis wi I I be presented and
the analytical results discussed.
EXPERIMENTAL
Instrumental Conditions
This method provides very rei iable results for metallurgical grade
aluminas as shown for Na in Figure 1. Ceramic grade aluminas do not
give the same I inear response. It might be thought that longer grind-
ing and a separate cal ibration curve could be used. However, the grain
size is already small (in the range 1-10 #m) and further grinding is
not effective. The longer grinding time of abrasive material such as
ceramic alumina causes its inevitable contamination. At present, the
briquetting technique gives inacceptable results for the ceramic alu-
mina. To el iminate these inconveniences, get rid of the particle size
distribution effect and avoid shortage of cal ibration standards, new
attempts involving fusion sample preparation techniques were tried.
METALLURGICAL CERAMIC
ALUMINA ALUMINA
* +
1 . 5 . . - - - - -_ _ _ _ _ _ _ _ _ _ _ _ _ _ _--,
lZ
<:}
~ 1.0 +
~
lZ
<:}
+
g 0.5
+
0.0 +
0.0 0.5 1.0 1.5 2.0
INTENSITY FUNCTION
1 2 3 4 5 6 7 8 9
Na2
MgO
° 0
0
0
.003
0
.006
0
.012
0
.018
.25
0
.50
0
1.0
0
1.5
0
Si O2 0 0 0 0 0 .15 .30 .60 .90
P20S 0 .001 .002 .004 .006 0 0 0 0
S03 0 .0018 .0036 .0072 .011 .06 .12 .24 .36
CaD 0 .06 .12 .24 : .36 0 0 0 0
Ti O2 0 0 0 0 0 .003 .006 .012 .018
V2 0 S 0 .003 .006 .012 .018 0 0 0 0
,Cr 203 0 0 0 0 0 .0015 .003 .006 .009
iMnO 0 .0005 .001 .002 .003 .004 0 0 0
I
Fe2 03 0 0 0 0 0 .01 .02 .04 .06
NiO 0 .003 .006 .012 .018 0 0 0 0
CuD 0 0 0 0 0 .001 .002 .004 .006
InO 0 .003 .006 .012 .018 0 0 0 0
Ga 2 03 0 0 0 0 0 .003 .006 .012 .018
Mo0 2 0 .0013 .0025 .0038 .005 0 0 0 0
Sn02 0 0 0 0 0 .002 .004 .008 .012
PbO 0 .005 .010 .020 .030 0 0 0 0
automatic Claisse fluxer a four (4) g alumina sample is now fused with
nine (9) g of flux. The 4 g sample mass ensures a representative
sampl ing of trace elements. The flux is composed of 90~ Li 2B407 and
10~ LiF. Fluxes marketed by JMC, Aldrich, Spex, Laborlux ana Claisse
Corporation were tested with respect to their impurities content, and
the latter one was found to be the most adequate. The glass disk
diameter increase from 30 to 40 mm resulted in a 20-30~ intensity gain
for the elements concerned.
The cal ibration procedure was carried out using a set of synthetic
glass standards made of high purity element oxides and the flux. The
reagents used to prepare the synthetic standards were dried for a mini-
mum of one hour prior to weighing in order to obtain a known stoichio-
metry. To bui Id the set of synthetic standards, two "master" glass
discs were first cast. They were broken into sma I I pieces and ground
separately in a Shatterbox grinder. Smal I powdered portions of two
master discs (ranging in mass from 0 to 0.5 g), supplemented with high
purity alumina and flux were used to make nine synthetic standards.
Their composition in weight percent is given in Table 1.
Results
Na2
MgO
° 0.005-1.5
0 -0.012
210
-
100
8
S i 02 0.001-0.1 13 10
P2 0 S 0 -0.006 2 3
S03 0.01 -0.4 170 750
CaO 0.001-0.3 20 6
Ti 02 0 -0.02 7 6
V2 0 S 0 -0.02 5 5
Cr 2 03 0 -0.01 - 3
MnO 0 -0.001 3 3
Fe2 03 0.001-0.05 15 15
I NiO
I
0 -0.01 ,
- 4
! CuO 0 -0.01 - 4
0 -0.01 5 3
0 -0.025 7 4
0 -0.01 3
0 -0.03 6
Table 4. Detection Limits (2 sigma) in ppm for NBS 699 Alumina Pre-
pared as Briquette and Analyzed on Different Spectrometers
Na 2 0 0.59 80 38 15 13 26
Si 02 0.014 40 1 0.5 1.8* 6*
S03 0.012 40 0.5 0.2 0.2 -
CaD 0.036 40 0.3 2.2 2.8 3.5
Ti 02 0.0020 40 4 4.4 2.8 4
V2 0 S 0.0005 40 1 0.9 0.9 -
MnO 0.0005 40 0.3 0.3 1 0.7
Fe2 03 0.013 40 1.3 1.3 1.5 2
ZnO 0.013 40 1.5 0.7 0.7 1.6
Ga 2 03 0.010 40 1.6 0.8 0.7 1.3
Discussion
In the past the performance of the fusion method was I imited by in-
adequate instrumental sensitivity for Na due to matrix di lution.
Recent hardware improvements helped upgrade XRF analytical performance.
With the Phi I ips PW 1400 spectrometer, modified through the use of the
Sc anode tube and the multi layer PX-l crystal, the analysis of Na
became very rei iable. Yet better performance, in terms of lower limit
of detection and shorter analytical time, could be obtained using
spectrometers equipped with a Rh end window tube, as wei I as using
fluxes of higher purity than presently.
ACKNOWLEDGEMENT
The author would I ike to thank Dr. F.M. Kimmerle, Alcan's chief analy-
tical chemist, and Dr. G.F. Giasson for reviewing the manuscript and
fruitful discussions. He also wishes to thank Mrs. C. Gagnon, Mr. C.
Boulanger and Mr. J.M.N. Tropia for their assistance at the experimen-
tal stage.
REFERENCES
1. Wheat, T.A., Journal of the Canadian Ceramic Society, Vol. 40, p.
43, 1971.
690 X. XRF TECHNIQUES FOR HAZARDOUS WASTES AND OTHER APPLICATIONS
5. Van Zyl, C., X-ray Spectrometry, Vol. 11, No.1, 29, 1982.
AUTHOR INDEX
691
692 AUTHOR INDEX
695
696 SUBJECT INDEX