26

Ultraviolet, Vacuum Ultraviolet, and Extreme

Ultraviolet Spectroscopic Re¯ectometry and
Ellipsometry

J. L. Stehle, P. Boher, C. Defranoux, P. Evrard, and J. P. Piel

SOPRA, Bois-Colombes, France

I. INTRODUCTION

This chapter describes various aspects of the application of spectroscopic ellipsometry in

the UV range. After a short introduction on the electromagnetic spectrum and the de®ni-
tion of UV, VUV, PUV, and EUV ranges, general features of the optical indices versus
wavelength are presented. Speci®c instrumental problems related to the wavelength range
are discussed, in particular the problem of sources, polarizers, and detectors. The optical
mounting selected to work in the PUV range is described in detail with the problems of
stray light rejection and carbon contamination. Various measurement results at 157 nm
concerning oxynitrides, antire¯ective coatings, masks, phase shifters, and photoresists are
also presented. Finally, the speci®c problem of EUV lithography and the associated
metrology is introduced.

A. De®nitions
One of the last regions of the electromagnetic spectrum to be routinely used in the IC
industry is that between ultraviolet and x-ray radiation, generally shown as a dark region
in charts of the spectrum. From the experimental point of view, this region is one of the
most dif®cult, for various reasons. Firstly, the air is absorbent, and, consequently, experi-
mental conditions need vacuum or purged glove boxes. Secondly, the optics needed in the
range are generally not available. In fact, exploitation of this region of the spectrum is
relatively recent, and the names and spectral limits are not yet uniformly accepted.
Figure 1 shows that portion of the electromagnetic spectrum extending from the
infrared to the x-ray region, with wavelengths across the top and photon energies along
the bottom. Major spectral regions shown are the infrared (IR), which can be associated
with molecular resonances and heat (1 ev ˆ 1.24 mm); the visible region from red to violet,
which is associated with color and vision; the ultraviolet (UV), which is associated with
sunburn and ionizing radiation; the vacuum ultraviolet (VUV) region, which starts where
the air becomes absorbent; the extreme ultraviolet (EUV) region, which extends from 250

Copyright © 2001 Marcel Dekker, Inc.

 Figure 1 Electromagnetic spectrum with the different regions of interest. The Cu K-alpha line at
0.154 nm and the Si L-alpha line at 12.5 nm are indicated.

eV to about 30 eV; the soft x-ray region, which extends from about 250 eV (below carbon
K-alpha edge) to some keV, and the hard x-ray region, where the air again becomes
transparent.

B. Spectroscopic Ellipsometry Technique

Spectroscopic ellipsometry is now well established as a key metrology method for char-
acterization of thin ®lms and multilayers. In ellipsometry, one measures the change of
polarization state after nonnormal re¯ection from the sample. The polarization state is
de®ned by two parameters so that each ellipsometric measurement provides two indepen-
dent parameters. The term ellipsometry derives from the fact that linearly polarized light
after oblique re¯ection becomes elliptically polarized due to the different re¯ection coef®-
cients for s- and p-polarized light. By measuring the polarization ellipse, one can determine
the complex dielectric function of the sample. The fact that two independent parameters
are obtained at each wavelength is an important advantage compared with photometric
measurements like re¯ectometry. The measurement is also self-calibrated, with no need for
baseline measurement, and the accuracy is generally better than for re¯ectometry.
Spectroscopic ellipsometry traditionally works in the visible and UV region (1).
Some extension is possible into the IR region using a Fourier transform spectrometer
(2). Special attention to the optical components is needed to reach the limit of the air
absorption around 190 nm (3). Below this value, two possibilities can be used to work in
the VUV region. Using a purged glove box to reduce the oxygen and water content to the
parts per million range, it is possible to reach 135 nm (4). Below this, vacuum systems are
needed, as, for example, the system mounted on the BESSY synchrotron (5). The EUV
region requires a special experimental setup (see Section IV, New Applications).

C. Optical Indices
Optical indices of all the materials give a picture of the energetic structure of the material
under study. From the hard x-ray region to the infrared, the photons interact successively
with the electrons in the core levels of the atoms, with the electrons involved in the band
structure of the material, with the vibration modes of the molecules or phonons in a solid,

Copyright © 2001 Marcel Dekker, Inc.

 and with the weak bonds of the material. Some optical indices measured for different
reference materials are shown in Figures 2, 3, and 4. These indices are taken from Ref. 6
and have been measured with various methods, depending on the wavelength range. They
are associated by type of materials, such as semiconductors (crystalline silicon, crystalline
germanium, and gallium arsenide in Fig. 2), insulators (silicon carbide, silicon nitride,
aluminum oxide, and silicon dioxide in Fig. 3), and metals (copper, molybdenum, and
tungsten in Fig. 4).
So in the hard x-ray region, the soft x-ray region, and the EUV region the optical
properties of all the materials depend only on the atoms of the material. The refractive
index of the materials can be written as (7)

l2re X

nˆ1 d ‡ ib ˆ 1 Za ‡ fa0 ‡ ifa00 Na
2p a

Here, re is the classical electron radius and is a measure of the scattering strength of a free
electron, and l is the wavenumber of the x-rays, and l is the x-ray wavelength. The atomic
number of the atomic species a is denoted Za , and the correction for anomalous dispersion
is fa0 and fa00 . d and b are always in the range 10 5 ±10 6 in the hard x-ray region, and all the
materials are transparent to x-rays in this range. This can be seen for all the materials
reported in Figs. 2, 3, and 4. Against wavelength, the absorption increases as a function of
l2 , with some discontinuities due to the absorption edges of the atoms (Si L-alpha edge at
12.4 nm, for example, as reported in Fig. 2 for silicon and in Fig. 3 for SiO2 ).
In the VUV and UV range the insulators become rapidly transparent (cf. Fig. 3). The
interesting part is located in the VUV (the case of Si3 N4 and Al2 O3 ) or the UV (the case of
the SiC, which can be taken as a semiconductor). The decrease of the extinction coef®cient
is followed by the appearance of optical transitions, well de®ned in particular on the
refractive index. For semiconductors like silicon, germanium, and GaAs (cf. Fig. 2), the
interesting part corresponding to the conduction band is located in the visible range, and
this is why standard ellipsometry has long been used to characterize such materials. For
both the insulators and the semiconductors, there is a range in the infrared where the
material is completely transparent, which is useful to measure the layer thickness of thick
®lms. This does not apply to the case of metals, as shown in Fig. 4. In this case, the
material stays completely absorbent in the entire wavelength range from VUV to IR. This
is called the Drude tail for conductive materials.

II. OPTICAL MOUNTING FOR PUV

A. Air Transmission
Owing to air's strong molecular absorption band, transmission in air below 190 nm is very
poor. It is therefore necessary to install measurement tools such as ellipsometers in a
carefully controlled purged environment where the oxygen and water vapor levels must
be maintained at the parts per million range.

B. Adsorbed Carbon Layer

Contaminants adsorbed on substrate or ®lm surfaces can strongly affect measurement at
157 nm. It is estimated that 1 nm of organic contaminants such as alkanes reduces
transmission by 1%. Similarly a 1-nm-thick layer of water reduces transmission by 2%.

Copyright © 2001 Marcel Dekker, Inc.

 Figure 2 Optical indices of some semiconductors in the range from soft x-rays to infrared. (From
Ref. 6.)

Copyright © 2001 Marcel Dekker, Inc.

 Figure 3 Optical indices of some insulators in the range from soft x-rays to infrared. (From
Ref. 6.)

Copyright © 2001 Marcel Dekker, Inc.

 Figure 4 Optical indices of some metals in the range from soft x-rays to infrared. (From Ref. 6.)

Copyright © 2001 Marcel Dekker, Inc.

 Contamination can occur during brief exposure to the ambient or as a result of outgassing
from chamber walls and samples as photoresists in the measurement. The surface con-
taminants play an increasing role in transmission measurements as the wavelength is
reduced further, below 157 nm. To minimize contamination it is possible to clean the
surface prior to the measurements using a 157-nm excimer laser beam.
Measurement of ambient is also very important, and it is reported (9) that after 30
minutes in vacuum, the transmission of CaF2 samples drops signi®cantly across the mea-
sured spectrum from 90.2% to 87% at 157 nm. Using laser-cleaning in N2 atmosphere, it
is possible to restore the transmission to its original values. Moreover purging the chamber
with nitrogen rather than maintaining it under vacuum reduces the level of surface con-
tamination. It is clear that this effect strongly depends on the pumping arrangement used.

C. Stray Light Rejection

All light that reaches the spectrometer exit slit from anywhere other than the disperser
(prism or grating) by any means other than diffraction is called stray light. All components
in an optical system contribute to the stray light, such as apertures, partially re¯ecting
surfaces, and even the grating itself. In this last case, it could come from imperfections in
the shape and spacing of the grooves and micro-roughness on the surface of the grating
(diffuse stray light). We can also observe what are called ghosts, which are caused by
periodic errors in the groove spacing. Re¯ection from instrument chamber walls and
mounting hardware also contributes to the redirection of unwanted energy toward the
image plane. Light incident on the detector may be re¯ected backward to the grating and
rediffracted. Since the angle of incidence may now be different, light rediffracted along a
given direction will generally be of a different wavelength from that of the light originally
diffracted along the same direction. Stray light can be reduced greatly by the use of a
double spectrometer, in which light diffracted from one grating is incident on a second
disperser.
The signal-to-noise ratio (SNR) could be de®ned in this case as the ratio of diffracted
useful energy to unwanted light energy. Stray light usually plays the limiting role in the
achievable SNR for a grating system.

D. Calibrations
We distinguish four kinds of calibrations for the spectroscopic ellipsometer (SE). The ®rst
one consists of the spectrometer calibration using well-known spectral lines (from a Hg
lamp). The second calibration is used to evaluate the nonlinearity of the photomultiplier
tube. The third one consists of the calibration of the residual polarization induced by the
photomultiplier window. The last one allows evaluation of the offset of the polarizer and
analyzer main axes versus the plane of incidence, which is the reference plane of the
instrument. This procedure is applied at installation and when the user changes the
lamp, so the stability of the calibration can be checked every quarter, as for visible
ellipsometers. Before a daily use, the user can verify the level of the signal in the critical
range to be sure that nothing special occurred as gas absorption or no counting occurred
because of a lamp or detector failure.

Copyright © 2001 Marcel Dekker, Inc.

 1. Wavelength Calibration
First of all, it is necessary to determine the mirror position of the prism (4% of re¯ection
on the surface of the prism). To perform this calibration, it is necessary to scan the prism
around this mirror position for a ®xed position of the grating. The maximum of signal
corresponds to the mirror position (Figure 5). This position is then used for the calibration
of the grating, independent of the prism. The grating is then calibrated using a Hg lamp
with the prism ®xed in mirror position. The different emission lines and the zero order of
the grating are selected. For each theoretical position, a scan of the grating is made. The
difference between the theoretical position and the experimental one is ®tted with a poly-
nomial of degree 2 and the parameters saved (Figure 6). Finally, it is necessary to perform
the calibration of the prism. For this purpose, different wavelengths are selected by the
grating, and the prism is scanned around each of these wavelengths (Figure 7).

2. Calibration of the Nonlinearity of the Photomultiplier Tube

The aim of this calibration is to evaluate the nonlinearity and the residual polarization of
the detector. In the ®xed polarizer and rotating analyzer mounting used, we need to take
into account their effect in terms of residual polarization.
The software measures the residual and the phase of the signal versus the intensity in
a straight line at different wavelengths and analyzer positions. The intensity of the signal
measured versus the polarizer position is given by the following equation when there is no
nonlinearity effect:
Iideal ˆ K 1 ‡ a cos 2A ‡ b sin 2A†

Figure 5 General optical layout of PUV-SE to be installed in purged box.

Copyright © 2001 Marcel Dekker, Inc.

 Figure 6 Grating calibration window with calibration already made.

Figure 7 Prism calibration window with calibration already made.

Copyright © 2001 Marcel Dekker, Inc.

 When there is a nonlinear effect of the detector, the measured intensity can be assumed to
be given by
2
I ˆ Iideal ‡ pIideal ˆ K 0 1 ‡ a 0 cos 2A ‡ b 0 sin 2A†

The Hadamard detection measures the a 0 and b 0 , and the corrected parameters a and b
can be recalculated using the following equations (Figure 8):
2 f ‡ 1†a 0
aˆ h i1=2
0
2f ‡ 1†2 2f f ‡ 1† a 2 ‡ b 2 †
0
2f ‡ 1† ‡
a
b ˆ b0
a0
The f parameter is characteristic of the nonlinearity. It is given by:
2Z 2
f ˆ
4 3Z
q
0
where Z ˆ a 2 ‡ b 2
0

3. Calibration of the Residual Polarization of the Detector

If the photomultiplier MgF2 window induces anisotropic light with a small linearly polar-
ized contribution of amplitude e with an angle f with regard to the reference plane, the
electric ®eld is given by
E ˆ Ep 1 ‡ ex cos f ‡ ey sin f†
Then the intensity measured by the Hadamard detection is given by

Figure 8 Nonlinearity correction of the photomultiplier tube.

Copyright © 2001 Marcel Dekker, Inc.

 I ˆ K 1 ‡ ex cos A ‡ ey sin A†2 1 ‡ a cos 2A ‡ b sin 2A†
The measured Fourier coef®cients can be expressed by:
2 a ‡ px † 2 b ‡ py †
a0 ˆ and b0 ˆ
2 ‡ p x a ‡ py b 2 ‡ px a ‡ py b
where
e2 e2
px ˆ cos 2f and py ˆ sin 2f
2 ‡ e2 2 ‡ e2
The px and py values can be calibrated versus the wavelength using the variation of the
phase with the analyzer position. It is given by
   
b0 b ‡ py

y ˆ arctan 2A ˆ arctan 2A
a0 a ‡ px
In practice, we make a regression on the experimental points at each wavelength using a
Levenberg±Marquard procedure (Figure 9).

4. Polarizer (P0 ) and Analyzer (A0 ) Offset Calibration

In the measurement procedure, precise calibration of the position of the plane of incidence
is a direct part of the measurement. Two methods can be used to calibrate the position of

Figure 9 Calibration of the polarizer PO and of the Px residual polarization of the signal induced
by the detector.

Copyright © 2001 Marcel Dekker, Inc.

 the plane of incidence: the residual method (3), described next, and the regression method
(Figure 10).
The idea is to make a scan around the expected position of the zero of the analyzer
and to determine precisely its zero position and the corresponding phase using the mea-
sured parameters. In this respect, we use the residual value R obtained by

Rˆ1 a2 b2

where a and b are the Fourier coef®cients of the measured signal. In the ideal case, the
residual value is given by

tan2 c tan2 P† ‡ 2 cos
tan P†

Rˆ1
tan2 c ‡ tan2 P†2

where P is the polarizer position. The minima of the residual values are then obtained
when P ˆ kp=2 with k an integer value.

E. Instrumental Function
In order to check the accuracy of the SE instrument, it is very useful to perform tan c
and cos
measurements without sample. To do this, the system is set in a straight line
such that the polarizer arm is at 90 from the normal to the sample (same for the angle
of the analyzer). In this con®guration, the theoretical values for tan c and cos
must be
1. The measured values give useful information about the goodness of the system itself
(Figure 11).

Figure 10 A0 and P0 calibration using residual method.

Copyright © 2001 Marcel Dekker, Inc.

 Figure 11 Measured Cos delta without sample in straight line.

For the evaluation of wavelength accuracy and resolution, it is necessary to use

another approach, such as a calibrated thick sample or an interference ®lter or an atomic
spectral lamp.

F. Limitations
The transmission of the ¯ux is limited by the presence of H2 O and O2 gases in the
glove box. Even for a concentration of few ppm of these gases, the transmission
decreases signi®cantly below 175 nm. Contamination of the mirror surfaces
and the polarizer surfaces can easily reduce the light ¯ux arriving at the detec-
tor.
The sensitivity of the detector itself is optimized for the PUV range but consequently
is limited to a shorter spectral range, 125±750 nm.
The spectrometer resolution is another limitation. Due to the fact that we need to
optimize the light ¯ux arriving on the sample, it is necessary to compromise
between the ¯ux and the spectrometer so its resolution is around 1.0 nm.

III. APPLICATIONS

Applications of UV, DUV, and PUV ellipsometry are numerous. Extensions to the lower
wavelength have been driven by photolithography, but many other applications have also
bene®ted. The determination of optical characteristics such as refractive index of materials
in the UV has permitted the de®nition and calculation of properties of the material using
SE. The UV-SE applications will be presented with the following plan:
A. Photolithography
1. Photoresist, antire¯ective coating, swing ratio simulation
At 365 nm
At 248 nm and 193 nm
2. Mask and phase-shift masks
3. Measurements for 157-nm lithography
B. Thin oxide, thin ONO, and nitrided oxides
C. Optical gaps
D. Roughness and surface effects

Copyright © 2001 Marcel Dekker, Inc.

 A. Photolithography
Ultraviolet spectroscopic ellipsometry for photolithography materials is certainly one of
the most complete and useful applications of ellipsometry. Not only can we measure
thickness or refractive indices of materials like photoresist (PR) and antire¯ective coating
(ARC), we can also predict by simulation the interaction and behavior of the light with the
resist during the photolithography process. Calculating the best thickness for each layer of
a structure, like ARC/PR/ARC, at any stepper wavelength will become easy with the
accurate knowledge of their refractive indices at the stepper wavelength.

1. Photoresist (PR), Antire¯ective Coating (ARC), Swing Ratio Simulation

Spectroscopic ellipsometry is the best optical technique to determine the thickness and
refractive index of materials like PR and ARC at the same time and using the same
instrument. These materials are transparent in the visible range and become absorbent
in the UV part. Calculating the refractive index in the UV range requires a high-precision
instrument and a well-de®ned thickness (the transparent part of the materials will help to
determine N and T, where the k value is 0).
At 248 and 193 nm and 157 nm, the sensitivity of DUV PR requires measurement in
a photo environment and the use of a premonochromator to avoid bleaching the resist.
Figure 12 shows the refractive indices and the extinction coef®cients of two types of DUV
248-nm resists.
Measuring at 193 nm has been a technical challenge. In 1992, a good signal-to-noise
ratio was obtained from measurement of photoresist. Since 1992, the accuracy of the
determination of the optical properties at these wavelengths has reached the quality
required by the PR manufacturers and users (11). Figure 13 presents a measurement

Figure 12 Refractive indices and extinction coef®cients of two types of DUV 248-nm resists.

Copyright © 2001 Marcel Dekker, Inc.

 Figure 13 Measured and ®tted curves of DUV 193-nm photoresist.

and a ®t on a DUV 193-nm photoresist sample. The measurement and ®tted curves are in
good agreement.
Figure 14 shows the optical properties determined using this last measurement.
Accurate determination of n and k can be reached down to 190 nm.
All these measurements and determinations on PR are done without any effect on
the PR by using a special ellipsometer setup. The light source and the monochromator are
before the sample. By using another setup, photoresist behavior versus exposure dose can
be investigated (19).
In today's photolithography the extension to the deep UV (KrF excimer laser line
at 248 nm and ArF excimer laser at 193 nm) leads to very dif®cult problems. The ®lm's
interference effect can lead to periodic behavior of the resist sensitivity swing effect, and
the stability is linked to the minimum of re¯ectance, which depends on the optical
indices. On nonplanar substrates, the effect of re¯ection on the vertical sidewalls of
the steps can cause an increasing coupling of the energy into the resist volume and a
reduction in linewidth (16,17). The best solution that has been chosen so far is the use of
top and/or bottom antire¯ective coatings. One needs a nondestructive technique to
characterize not only the thickness of the layers but also their optical indices. Their
characterization must be done at the working wavelength. And SE has been chosen as
the best method (12,13).
SiOx Ny :H ®lms make appropriate ARCs. Such ®lms are obtained by PECVD using
SiH4 and N2 O gases, whose ¯ow ratio can be adjusted for the deposition process, changing
the optical indices at the same time (5). Figure 15 shows the refractive indices of different
SiOx Ny :H ®lms.
Organic ARCs are also used in the lithography process. Figure 16 presents organic
top ARC and bottom ARC and photoresist refractive index.

Copyright © 2001 Marcel Dekker, Inc.

 Figure 14 Refractive index and extinction coef®cient of DUV 193-nm PR.

Figure 15 ARC: SiOx Ny :H. Refractive indices and extinction coef®cients.

Copyright © 2001 Marcel Dekker, Inc.

 Figure 16 N and k of top ARC, PR, and bottom ARC for DUV 248 nm.

Using the optical indices and appropriate lithography simulation software, such as
WINELLI (from SOPRA S.A.) or PROLITH (from FINLE Technologies), we can simu-
late the re¯ectivity of the material for any unde®ned parameters (12,13).
Figure 17 shows the effect of ARC on the re¯ectivity versus PR thickness.

2. Mask and Phase-Shift Mask

The decrease of the linewidth is also linked to the progress in photomask development.
The need for mask characterization is important, especially in the new design of phase-
shift masks and in all multilayer mask techniques. Materials like TiN, Chromium, Mo,
and SiON are some examples of such layers (Figure 18).

Figure 17 Re¯ectivity calculation of PR/SiOx Ny :H ®lms versus PR thickness.

Copyright © 2001 Marcel Dekker, Inc.

 Figure 18 N and k spectra for Cr and TiN.

3. Applications for 157-nm Lithography

Recently, the increasing interest in 157-nm laser source in projection lithography as suc-
cessor to 193-nm-based systems has pushed SE manufacturers to ®nd technical solutions
to be able to measure the optical properties at this wavelength. One of the ®rst applications
for these ellipsometers is to characterize the refractive indices and extinction coef®cients of
the optics of the future steppers. Primarily lens material for projection systems will be
metal ¯uorides, such as CaF2 , MgF2 , and LiF, and these require intensive characterization
as well as their coatings.
The optical indices of photoresist and antire¯ective coatings also need to be known
at 157 nm with the same accuracy as for 248 or 193 nm. The layer thickness is generally
smaller than for the previous lithographic generations. In this case, ellipsometry is also one
of the best techniques (Figure 19).

B. Thin ONO and Nitrided Oxides

In the DRAM industry, upgrading storage capacity used to be one of the major goals of
R&D. For its achievement, materials with high dielectric constant (high z) became the state
of the art. The related technology now focuses on the use of thin ONO (oxide/nitride/oxide)
stacks, which necessarily tend to be thinner and thinner to enhance capacitive performances.
Until now, most optical techniques could only measure the thickness of the stack as a
whole or determine the different thicknesses by ®xing one of the oxides in the ®t. Since the
thickness of each layer is dif®cult to measure independently, the ONO structures show a
tremendous need for measurement using spectroscopic ellipsometry, especially in the deep
UV range (Figure 20). Indeed, in this part of the spectrum, the extinction coef®cient k for

Copyright © 2001 Marcel Dekker, Inc.

 Figure 19 N and k of photoresist between 150 and 450 nm.

Figure 20 N and k of SiN and SiO2 versus energy (eV ˆ 1:24=mm).

Copyright © 2001 Marcel Dekker, Inc.

 the nitride increases. Thus, the three thicknesses of the ONO stack can be decorrelated. The
refractive index of the nitride has been determined independently on a single-layer sample.
Figure 21 shows the difference between three ONO samples. The main differences
are in cos
. On tan c the differences occur in the UV range. Figure 22 presents the
measurement and the regression curves on the 40/50/40 sample.
As the semiconductor industry continues to follow Moore's law, gate dielectric
thickness has become one of the most rapidly scaled dimensions. The SIA roadmap states
that a sub-0.25-mm process requires equivalent oxide values of less than 50 AÊ. Some
leading-edge manufacturers are already pushing 0.18-mm process technologies with gate
oxide thickness in the 30±40-AÊ range and for 0.13 mm near 20AÊ.
Nitride/oxide gate dielectrics have already been widely used. In this structure the
reliability of a high electric ®eld depends not only on the layer thickness but also on the
nitridation level of the SiOx Ny layer. It is important to characterize precisely the layer
thickness and composition of this structure in order to control the process quality (20).

C. Optical Gaps
By measuring the extinction coef®cient of silicon nitride down to the UV range, we are
able to calculate the optical gap of the material. This property is used in the ¯at panel
display industry to control the deposition of the SiN around 5.2 eV (Figure 20).

D. Roughness Effects
Because silicon and polysilicon are absorbent in the UV range, the only interface of the Si
seen in this range by the re¯ected beam is the top one. On a blank wafer or on a polysilicon

Figure 21 Measurement of three ONO samples. Thicknesses are in angstroms.

Copyright © 2001 Marcel Dekker, Inc.

 Figure 22 Fit on a 40-50-40 AÊ ONO samples.

without any top layers, the quality of surface, the roughness, and /or the crystallinity of
the material are easy to investigate. The following Table compares two similar structures:

SiO2 2 nm
SiO2 2 nm Roughness 2 nm
Polysilicon 150 nm Polysilicon 150 nm
SiO2 100 nm SiO2 100 nm
Si Substrate Si Substrate

We add 2 nm roughness between polysilicon and native oxide. The roughness is composed
of a mixture of polysilicon and SiO2 50%±50% (Figure 23).

IV. NEW APPLICATIONS

A. Extreme UV (EUV) for Lithography

The next-generation lithography may be using x-rays, electron beams, or ion beam projec-
tion and probably will be based on EUV photons with wavelength of 13.4 nm (another
wavelength at 11.34 nm using a MoBe mirror is less probable). If EUV becomes the next-
generation lithography, this will require intensive metrology, including defect inspection,
re¯ectometry, and spectroscopic ellipsometry. The stepper at 13.4 nm will require a source

Copyright © 2001 Marcel Dekker, Inc.

 Figure 23 Roughness effect on ellipsometric spectra of polysilicon/SiO2 =Si.

of photons specially developed to generate the spectral wavelength and the spectral purity
that will match with the mirror's bandwidth.
There are several potential sources for EUV lithography. Because the design of the
stepper is based on a step-and-scan mounting, the source is essential in terms of through-
put, and it must run at very high frequency, such as 6 KHz, because of the scanning of the
slit image on the ®eld. The source may be a plasma of Xe or water created by focusing a
powerful laser beam at a high repetition rate. A capillary electrical discharge in O2 or Xe
may be an alternate, cheaper source than the synchrotron radiation used today. Other
sources are under development as plasma pinch and plasma focus.
The source for metrology is also an issue at this wavelength, but the requirements are
different from those for lithography. It requires a broad emission to cover a minimum
spectral range around the maximum re¯ectivity of the layers, because the matching of
re¯ectance is very important in terms of the total transmittance of the system composed of
at least nine re¯ecting optics. Another characteristic is the stability. The frequency of the
repetition rate can be reduced to 10 Hz if a normalization is used (23,24). The debris
generated by the plasma creates a real production problem . It can decrease the quality of
the ®rst collection mirrors. But for metrology, the number of shots for a laser plasma
source is reduced and the lifetime of the optics is not such an issue.

1. Importance of the Re¯ectance of the Multilayers

In this spectral range, around 13.4 nm, the materials are not transparent, but it is possible
to use catioptric optics, polished mirrors coated with multilayers. The refractive index is
very close to 1, and the absorption coef®cient must not be large because the wavelength is
very short. There are a couple of materials that enable us to achieve these conditions, with

Copyright © 2001 Marcel Dekker, Inc.

 a large enough difference in absorption, Mo and Si. When their respective thicknesses are
2.8 and 4.2 nm, then by coating 40 bilayers or more, re¯ectivity near normal incidence has
been measured at over 68% for one re¯ectance mirror.
The projection optics (four or six mirrors) added to the collimating optics (four
mirrors) and the re¯ecting mask must have the highest re¯ectivity possible and must
match themselves very accurately to avoid losing costly light and to maximize the through-
put of the stepper. The large-aperture optics are illuminated with different angles of
incidence from the center to the edge, requiring the multilayers to gradually adapt their
thickness with the angle of incidence; the tolerance of thickness is less than 1 AÊ.
The instrument must be enclosed in a high-vacuum environment to avoid absorption
from the residual gas. To separate the source vessel from the fragile projection optics, one
very thin membrane is used, losing 50% of the light. Another membrane of SiN or Si may
be used to protect the costly optics from the degassing of the resists of the wafer, again
losing 50% of light. The consortium called EUV-LLC has designed a prototype of the ®rst
stepper, called ETS, for Engineering Test System, to be used as an alpha tool for testing
the concepts and the speci®cations and to prove the manufacturability of the EUV stepper
(Figure 24).
The new metrology developed for lithography will also be used for surface and layer
characterization in semiconductors, as well as to detect defects on the masks. Due to the
wavelengths used and the sensitivity to the surface, very thin ®lms will be analyzed. The
range from 10 to 15 nm needs vacuum and covers the absorption peak of the silicon at 12.4
nm, corresponding to the L line of the x-ray spectrum of Si, so these wavelengths are very
sensitive to layers including Si, such as poly-Si, silicon oxide, and silicon nitride, as well as
silicon oxinitride and new high-k materials for next-generation gate oxides.
The wavelength is 30 times shorter than the visible ones, so the sensitivity to the
thickness is then enhanced 30 times, except that the refractive index is close to 1 for all
materials. But the absorption can be used to characterize the materials.
The roughness will be very large because its amplitude becomes comparative to the
wavelength. The scattering effect of the light is proportional to lambda E-4; the scattering
will be enhanced by E6 compared to visible, so very smooth surfaces must be used and the
scattering induced by very small amplitudes, even the atomic steps, will play an important
role in the complex re¯ectance of light at these wavelengths, which can be considered as
soft x-rays. A model must be used that will not be a layer mixture for the roughness but
may be closer to the Debye±Wahler used in hard x-rays.
The detectors currently used are MCP (microplates) and photodiodes and channel-
trons or photomultiplier tubes or proportional counters. The ef®ciency of these detectors
is usually high, because one such photon has an energy of 100 eV and the noise is low,
particularly when the signal is pulsed. So the signal-to-noise ratio can be very good. The
¯uctuation is due mainly to quantum noise associated with the low number of photons
(1 sigma ˆ number of photons†E-1=2. To reach a ¯uctuation less than 1%, one needs at
least 10 000 photons.
The detection can be digital, by counting the photons one by one, if the source is
continuous, or at very high repetition rate if one collects only one photon maximum at
each pulse, such as, for instance, synchrotron radiation or high-voltage discharges on a
cathode of Si. In the case of pulsed sources, a reference is required to normalize the
amplitude ¯uctuation on a pulse-to-pulse basis. In this case, a 1% ¯uctuation may be
achieved in a single large pulse. The reference detector must be representative of the
signal for the numerical optical aperture, the spectral bandwidth, and the image of the
source.

Copyright © 2001 Marcel Dekker, Inc.

 Figure 24 Optical mounting of the stepper in EUV. (From Charles Gwyn, EUV LLC.)

The resists at this wavelength are the same as for either e-beam writing or DUV. The
sensitivity is below 10 mJ/cm2 , and isolated lines of 35 nm have already been demonstrated
by using an interferometer.

B. Combination of GXR and DUV SE

It is advantageous to combine a very-short-wavelength source, such as Cu K-alpha 1.54 AÊ,
with the UV-visible-wavelength spectroscopic ellipsometer (Figure 25). The models are the
same optically, except for the wavelength, the effects are complementary, and no vacuum
is required.
The refractive indices for hard x-rays are approximately 1 within 10 5 , and the
absorption is very low, about 10 4 , so all materials are transparent at this energy, includ-
ing metals, dielectrics, semiconductors, and organic layers. The optical beams are not
refracted as in the visible because the N values are always the same and equal to 1; the
Fresnel (Snell) law applies in a straight line.
The thickness of the layers is measured by the angle position of the interference's
after-re¯ectance at grazing incidence, between 0.2 and 4 typically, relative to the sample
surface. This measurement is very accurate for the optical thickness, and it is independent
of the material because N ˆ 1, so the GXR interference values, linked to angle positions,
give the thickness without ambiguity. The accuracy is given by the angular position of the

Copyright © 2001 Marcel Dekker, Inc.

 Figure 25 Schematic view of the combined GXR/SE system.

interference, which is very high. When the thickness is known with precision, after grazing
x-rays re¯ectance, it can be used in the data regression in UV-visible spectroscopic ellip-
sometry to extract the refractive indices N and k with the same relative precision as the
thickness measured. This is particularly useful for thin layers, when the thickness and
refractive indices are strongly corelated. In the case of multilayers, an addition of re¯ective
interference will give beats with amplitudes proportional to the respective density or
refractive indices of each layer.
The roughness is measured with precision and without ambiguity by the rapid
decrease of re¯ectance and can be modeled using the Debye±Wahler factor. In the case
of a layer it is even possible to discriminate the top layer roughness from that of the
interface.
The advantage of combining the two techniques in the same equipment is that one
can measure the same area of the same sample on the same day by using only one sample
holder, one goniometer, one electronic rack, and one computer with the same software for
modeling. This is not only a saving of time and money but it also gives a better and more
accurate result with fewer uncertainties. Nontransparent layers in the UV-visible can be
measured with grazing x-rays re¯ectance, as long the thickness is less than 200 nm. For
thicker, ¯at, and smooth layers, the interference pattern is so tight that the contrast
disappears, and the measure is lost. The critical angle of incidence, at which the light
penetrates into the top layer, is related directly to the density of the material; for example,
for SiO2 it is 0:26 , because of its density of 2.8 g/cc
The limitations linked to the GXR are the need for a large and very ¯at surface of at
least 1 cm2 with a very smooth surface. If the roughness value is larger than 5 nm, the
reduction in the re¯ectance is so large that no layer interference can be measured and the
critical angle itself is also affected.
The measurement is traditionally made by a sequential scanning of angles of inci-
dence and of re¯ectance. But recently a multichannel detector has been used to detect the

Copyright © 2001 Marcel Dekker, Inc.

 re¯ectance coef®cients in parallel. This has been made possible by the use of multilayer
graded concave mirrors. In this case it is very dif®cult to measure all the dynamic range
requiredÐ7 decadesÐand the angular resolution is limited to resolving very thin ®lms and
very thick ®lms as well as multilayers. This system is well suited to measure the thickness
of barrier layers such as TiN and TaN, semimetals such as TiSi, and seeds for Cu where
the thickness varies from 10 nm to 100 nm and the densities are very different from the Si
or SiO2 . Monitor wafers are still needed, due to the spot larger than 1 mm. It is not easy to
focus the beam, and if the source size was 40 m, then the grazing image should cover a
length of 2400 m for an angle of incidence of 1 .
The same x-ray source, goniometer and detector can be used to measure x-ray
diffraction. Eventually, with a solid-state SiLi detector, the layers and surface x-ray ¯uor-
escence spectrum can be measured to analyze the materials that ¯uoresce under Cu K-
alpha line excitation. Another source would give better sensitivity and a larger spectrum.
The atmosphere can reduce the range for light elements.

V. CONCLUSION

Ultraviolet, VUV, and EUV are spectral ranges where interesting features in optical
indices are present. These ranges are very sensitive to very thin ®lms for absorption and
roughness, due to the shorter wavelengths compared to visible and infrared. Many dif®-
culties must be overcome, such as the need for vacuum, the degassing of materials, and the
availability of polarizers and sources. Because it is new, there is a need for characterization
of the new materials now available to build a good database. This range will be investi-
gated further in the future for its special attractive qualities for analysis, particularly for
the photolithography application, which always uses shorter wavelengths.

REFERENCES

1. See, for example, the proceedings of the Second International Conference on Spectroscopic
Ellipsometry, Charleston, SC, 12±15 May 1997, Thin Solid Films, Vols. 313±314 (1998).
2. P Boher, JP Piel, P Evrard, JL Stehle. A new combined instrument with UV-Visible and far
infrared spectroscopic ellipsometry: application to semiconductor characterization. EMRS
Conference, paper PVII2/P6 (1999).
3. P Boher, JP Piel, C Defranoux, JL Stehle, L Hennet. SPIE 2729: , 1996.
4. P Boher, JP Piel, P Evrard, C Defranoux, M Espinsa, JL Stehle. A new purged UV spectro-
scopic ellipsometer to characterize thin ®lms and multilayers at 157 nm. SPIE's 25th Annual
International Symposium, 27 February±3 March 2000.
5. J Barth, RL Johnson, M Cardona. Spectroscopic ellipsometry in the 6±35eV region. In: E D
Palik, ed. Handbook of Optical Constants of Solids II. San Diego, CA: Academic Press, 1991,
ch. 10, p 123.
6. ED Palik, ed. Electronic Handbook of Optical Constants of Solids, ScVision. San Diego, CA:
Academic Press, 1999.
7. BL Henke. Low-energy x-ray interactions: photoionization, scattering, specular and Bragg
re¯ection. Proceedings of the American Physical Society Topical Conference on Low-Energy
X-Ray Diagnostics, Monterey, CA, June 8±10, 1981.
8. ED Palik, ed. Handbook of Optical Constants of Solids II. San Diego, CA: Academic Press,
1991.
9. Bloomstein et al. Proceedings of SPIE. 3676:342, March 1999.

Copyright © 2001 Marcel Dekker, Inc.

 10. DE Aspnes, AA Studna. Rev. Sci. Instruments. 49:291, 1978.
11. P Boher, C Defranoux, S Bourtault, JL Stehle. High accuracy characterization of ARC and PR
by SE: a new tool for 300 mm wafer technology. 24th Annual SPIE Meeting on
Microlithography,
12. P Boher, C Defranoux, JP Piel, JL Stehle, L Hennet. Characterization of resists and ARC by
SE in the UV and deep UV. SPIE 2729: ,1996.
13. P Boher, C Defranoux, JP Piel, JL Stehle, L Hennet. Precise ARC optical indices in the deep
UV range by variable angle SE. SPIE's 1997 Symposium on Microlithography,
14. C Defranoux, JP Piel, JL Stehle. Deep UV measurements of SiON ARC by SE. Thin Solid
Films 313±314:742±744, 1998.
15. T Ogawa, H Nakano, T Gocho, T Tsumori. SiOxNy: high performance ARC for the current
and future optical lithography. SPIE 2197:722±732,
16. RR Dammel. Antire¯ective coating: sultans of the swing curve. Semiconductor Fabtech. 2nd
ed., 1995, pp 215±221.
17. JE Lamb III. ``Organic ARC, ARC Application within the Microelectronics industry. Semi
conductor Fabtech. 2nd ed. 1995, pp 223±227.
18. P Boher, JP Piel, P Evrard, JL Stehle. A new Purged UV spectroscopic ellipsometer to char-
acterize 157-nm nanolithographic materials. MRS 1999 Fall Meeting,
19. P Boher, C Defranoux, JP Piel, JL Stehle. A new method of determination of the photoresist
dill parameters using SE. SPIE 3678:126, 1999.
20. P Boher, JP Piel, JL Stehle. Precise characterization of ultrathin nitride/oxide gate dielectrics by
grazing x-ray re¯ectance and spectroscopic ellipsometry. MRS 1999 Fall Meeting,
21. D Attwood. Soft X-Rays and Extreme Ultraviolet Radiation: Principles and Applications.
New York: Cambridge University Press.
22. C Gwyn et al. EUV LLC. Livermore, CA, November, 1998. PO Box 969, MS-9911, Livermore,
CA 94551-0069.
23. M Yamamoto, K Mayama, H Kimura, Y Goto, and M Yanagihara. Thin ®lm ellipsometry at
a photon energy of 97 eV. J Electron Spectroscopy Related Phenomena 80:465±468, 1996.
24. JH Underwood, EM Gullikson. Beam line for measurement and characterization of multi-
layers optics for EUV lithography. Proc. SPIE 3331:52, 1998.

Copyright © 2001 Marcel Dekker, Inc.