Mirror substrates for EUV-lithography: progress in metrology and

optical fabrication technology

U. Dinger, F. Bisert, H. Lasser, M. Mayer, A. Seifert, G. Seitz, S. Stacklies, F.-J. Stickel, M.Weiser

Carl Zeiss (Germany)

ABSTRACT

EUVL, i.e. projection lithography, utilizing actmic wavelengths in the range of 21 1-13 nm, appears to be the natural
extension of optical lithography into the NGL-regime.
All-reflective imaging systems, utilizing aspherical, multilayer coated, near normal incidence mirrors with reflectivities
around 70%, were designed to enable diffraction limited imaging from 7Onm down to 3Onm feature sizes at high
throughput.
However, diffraction limited performance of optical systems calls for surface figures of typically 2J50to 2/1OO. Considering
the operation wavelength of 2l 1-13 nm, this demand corresponds to unprecedented requirements for figure metrology and
fabrication technology.
Simultaneously, low flare levels and high reflectivities of the mirrors have to be ensured for proper system operation. These
properties are related to the substrate roughness in the mid spatial (typically ljim-lmm) and high spatial (typically <1 jim)
frequency ranges respectively. Accordingly these quantities have to be generated and controlled on a few- A-level.

In this paper, the metrology and fabrication concepts at CARL ZEISS will be reviewed. The present status in the fabrication
of specific EUVL mirrors will be reported as well.

1. INTRODUCTION
According to the 1999 edition ofthe International Technology Roadmap for Semiconductors, lOOnm generation ICs will be
in production in 2005, 7Onm ICs in 2008 and 5Onm ICs in 2011.

The resolution of optical microlithography usually is expressed as Res= k1*7JNA. By pushing the numerical aperture NA of
the projection system and the process factor k1 close to their practical limits, F2 laser (?=157 nm) based systems are
expected to meet the 7Onm node.

For even higher resolutions, several Next Generation Lithography (NGL) techniques are possible. These include Extreme
Ultra Violet Lithography (EUVL), Electron Projection Lithography (EPL), Ion beam Projection Lithography (IPL) and X-
ray lithography.

EUVL, using 13 or 1 1 nm radiation, has a high probability to become the NGL of choice because:
. EUVL provides a higher throughput potential down to the 35nm IC generation than EPL and IPL, due to the
absence of stochastic coulomb interaction which couples throughput with resolution
• EUVL is a reduction projection technology, which greatly reduces requirements for the mask technology
• EUVL has the best fit with the current optical technology base, including precision optics and scanning stages.

In the US, a consortium of IC manufacturers and National Laboratories (EUV-LLC/VNL) was formed in 1997 to pursue the
development of the EUVL-technology'. In Japan an ASET program, mainly addressing core technology development, on
EUVL was started2. In Europe, based on the experience in the EUCLIDES program3, ASML has started the development of
an EUVL f3-tool, aimed to print 80 300mm-wafers per hour. CARL ZEISS is addressing the development and fabrication
of the optical system within this program.

Soft X-Ray and EUV Imaging Systems, Winfried M. Kaiser, Richard H. Stulen, Editors,
Proceedings of SPIE Vol. 4146 (2000) © 2000 SPIE. · 0277-786X/00/$15.00 35

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 2. GENERAL CHARACTERISTICS OF AN EUVL OPTICAL SYSTEM
In general the optical column of an microlithographic system consists of a light source, an illumination system, a mask
carrying the IC-information (reticle), an imaging system and the wafer.

EUVL radiation can be obtained from synchrotrons, laser produced plasmas or discharge plasma sources. Whereas
synchrotron insertion devices provide highly collimated light fields, laser plasma sources radiate almost isotropically.
Plasma discharge sources have radiocharacteristics in between these extremes. The illumination system has to efficiently
couple these light fields to the entrance pupil of the imaging system while homogeneously illuminating the reticle. Finally,
the projection system — usually designed as reduction system — transfers the reticle pattern diffraction limited onto the
wafer. Besides providing the required resolution, achieving a system transmission as high as possible is a major system
design and technology issue.

Refractive and catadioptric optical column concepts lack of sufficiently transmissive materials at wavelengths significantly
below 157 nm. Consequently all-reflective systems have to be set up to meet the throughput requirements of a highly
productive NGL-stepper. High imaging fidelity calls for non-obscured mirror systems putting limits on the practicable NA
and resulting inevitably in off-axis configurations with exotic mirror shapes. Fortunately, in the last decade, it was shown
that by drastically reducing the actinic wavelength into the EUV range, obscuration free optical design concepts exist at
comparably small NAs, providing the required resolution at sufficiently low distortion levels4. Since diffraction limited
resolution is required, normal incidence configurations are mandatory for aberration control in the imaging systems.
Grazing incidence mirrors — employed very successfully e.g. in many x-ray astronomical systems — are restricted to the non
diffraction limited EUVL illumination system due their high aberration levels
To enable near normal incidence reflection, 'Distributed Bragg Reflectors (DBR)' are deposited on the mirrors5. At
?=13nm, where several EUV radiation
sources are available, Mo/Si-multilayers
with appr. 40 periods can provide
Reticle reflectivities close to 70% near normal
incidence. At X=1 mm, Mo/Be-multilayers
are favoured. To realize such high
reflectivities close to the theoretical limit
the substrate's roughness at spatial
wavelengths below appr. 1 tm — usually
titled 'high spatial frequency roughness
(HSFR)' - must not exceed a few
Angstroms rms. Otherwise the positive
interference of the partial wavefronts
reflected from the individual layers is
disturbed and the reflectivity drops
drastically5.

In contemporary refractive microlitho-

graphic imaging systems, appr. 60, usually
spherical, surfaces are required for
sufficient aberration control. Nevertheless,
highly transmissive materials and highly
efficient antireflection coatings guarantee
high system throughput.
Since the reflectivities of a normal
incidence EUV-mirror is in the order of
Fig.l:Optical layout of a LPP-based EUVL system: A four mirror 70% only, system transmissions in
illumination system is used to illuminate the reticle with light compliance with the demands of a
originating from a laser produced plasma (LPP) source. A six mirror commercial exposure system call for a
ringfield projection system transfers the mask pattern to the wafer. minimum number of mirrors in the optical
system. To provide sufficient degrees of

36 Proc. SPIE Vol. 4146

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 freedom to control system aberrations the off-axis mirrors will in general have aspherical shapes. To further facilitate
aberration control, the systems are usually laid out as ringfield scanners, utilizing an 30°-60°-segment of a small off-axis
annulus as image field4. Typical field sizes are 22x2mm2. Presently six mirror imaging systems operating at an NA close to
0.25 are the baseline for the realization of a commercial EUVL-scanner within the next decade6'7. In Fig. 1 a typical optical
column is sketched.

According to the Marechal criterion, diffraction limited imaging calls for an rms system wavefront error below AJ14. In a six
mirror systems this requirement constrains the tolerable surface figure error to appr. O.2nm rms. In field of view scattering
(flare) is related to the mid spatial frequency roughness (MSFR) at spatial wavelengths typically between 1mm and lim. A
substrate roughness of a few angstroms rms is needed in this frequency band for sufficient contrast in the image.

The EUV-scanners will most probably be operated with an reflective mask. The all reflective illumination system will
consist of a combination of grazing and normal incidence minors8. Figure specifications on the illumination mirrors will be
significantly relaxed compared to the imaging minors. However, EUVL-compliant roughness levels have to be ensured on
the steeply aspheric normal incidence components.

In contrast to the illuminator, the projection optics designs are in a comparably mature state. Thus recent EUVL-related
core technology development at CARL ZEISS was focused on the imaging minors. Summarizing the considerations in this
section, the general features of these mirrors are:

S Aspherical shape
. Off-axis segments of rotational symmetric parents, usually kidney shaped due to ringfield configuration
. Sharply edged, to avoid obscurations
S Figure specs typically: 0.1-0.2 nm mis
. MSFR-specs typically: 0.1-0.2 nm rms
I HSFR-specs typically: 0.1-0.2 nm mis

Besides these general characteristics

. the aspherical departure from the best fitting sphere,

I the gradient of this quantity,
. the mirror dimensions and shapes and
. the substrate material

will greatly influence the processes needed to fabricate the optics.

Metrology and fabrication technology development for the imaging mirrors will be the major object of the discussion in the
forthcoming sections.

3. THE CHALLENGE OF EUVL ASPHERE FABRICATION

The publication of the first 2-4 mirror designs for EUVL-suited imaging systems dates back into the late 80s.9 During the
last decade a lot of work was spent on the extension of these pioneering concepts towards highly developed systems capable
to meet the requirements of a production system6'7.

It was obvious from the very beginning of these investigations that the fabrication and metrology of the optical components
will be extremely challenging:

• Surface specifications are in the order of atomic dimensions in the whole spatial frequency domain reaching from
the clear aperture of the mirrors down to lateral atomic scales. This is new compared to other categories of high
precision optics: Since the totally scattered intensity strongly drops with increasing actinic wavelength, MSFR
specs are comparably low even in the high quality lenses for DUV lithography. Usually there is no major control

Proc. SPIE Vol. 4146 37

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 of HSFR in these systems. In contrast on grazing incidence x-ray optics for terrestrial and space applications,
where comparable MSFR-specs have to be maintained, figure-specifications usually are much more relaxed, since
no diffraction limited imaging is required and HSFR is —as long as no multilayers are applied- of minor interest.
EUVL-compatible HSFR-levels on test substrates without stringent MSFR and figure specs can be generated with
standardized processes. Thus, the real challenge of EUVL-asphere production is to generate low figure, MSFR and
HSFR simultaneously on an aspheric surface. Further restrictions stem from the choice of material. The highest
overall potential probably has fused silica. Unfortunately, fused silica entering the projection system is rather
unlikely due to it's thermal properties. Thermal management ofthe system calls for low expansion materials as e.g.
ZERODUR® or ULE®.
Any optical fabrication can be only as accurate as the underlying metrology. The reproducibility of the surface
topography with respect to a fixed but unknown reference of the instruments is usually referred to as precision. It
should guarantee a significant signal to noise ratio to ensure the convergence of the fabrication process. Absolute
accuracy errors (deviation of the cited precision-reference from an ideal absolute reference) of the metrology data
will be transferred identically to the surfaces by an ideal fabrication process. To some extend these errors can be
compensated for in a post assembly optimization based on system performance data. Principally, in a centered all
refractive systems all six degrees of freedom of the optical components can be used for compensation. In an off-
axis system the clocking degrees of freedom, useful to compensate certain non rotational symmetric systematic
errors, are no more available. Consequently — at least as far as these error types are concerned - metrology has to
operate on a significantly higher accuracy level as in centered refractive systems.

Two quantities may be used to roughly rate the degree of difficulty of asphere production.

The first quantity is the absolute deviation of the aspherical contour from the best fitting sphere. This number
determines the maximum number of fringes an interferometer or compensation optics has to deal with. Concerning
fabrication, it is a measure of the material that has to be removed with respect to a well corrected sphere. Since this
is usually done by subaperture tools the risk to introduce mid spatial frequency errors due to fluctuations in the
process or integration ofmaterial defects increases with increasing sag.

Even more important is the gradient of the aspherical deformation with respect to the lateral coordinate on the
minor surface. It determines the local fringe density the wavefront detection system of an interferometer has to be
capable to deal with. To generate a certain aspheric gradient, the tool functions have to have sufficiently high
frequency components. The higher the gradients, the smaller the tools will have to be. Fabrication time will
increase with decreasing tool size. The risk to introduce mid spatial frequency errors by shortcomings in the tool
dynamics and positioning increases. Eventually, the effectiveness of MSFR-smoothening tools significantly
decreases with increasing aspheric gradients.

Additionally the overall dimensions and shape of the substrates as well as the details of the aspheric contour will enter the
'figure of difficulty'.
Consequently, the optical design process was always aimed to keep this numbers as low as possible. Table 1 one compiles
typical values for some representative EUVL-mirrors.

System appr. max. ashperical sag appr. max.aspherical gradient

Jewell-Thompson type 3pm O.5jnnlnim
ELT2 2pm O.5jmilmm
ETS-type M2 10pm O.5rmImm
MET Ml 4pm l.2tmImm
MET-M2 6jim O.5pmImm
Alpha-tool 1 5tm (estimate) 2.Ojimlmm (estimate)

Table 1: characteristics of some specific EUVL-aspheres.

38 Proc. SPIE Vol. 4146

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 4. THE MICRO EXPOSURE TOOL (MET)
Carl Zeiss stepped into the field of EUVL-optics fabrication back in 1997. Within the EUCLIDES program technology
development was focused on Jewell-Thompson-class optics. Specific results on the secondary mirror ofthe Sandia NL 3M-
system ELT were performed and presented elsewhere10.

In late summer 1999 CZ started the fabrication of the ZERODUR mirrors for VNLs micro exposure tool (MET). The
MET is an on-axis NA=O.3 two mirror system with central obscuration (Fig.2). This SEMATECH funded project is
predominately aimed to enable defect printability studies with a resolution down to 3Onm.

Ml is a convex, 54 mm diameter, 3 .8tm deep asphere. M2 is a concave, 183mm diameter, 5.6jtm deep asphere. Due to its
smaller diameter Ml has significantly larger gradients than M2 and is the more challenging one to manufacture. ELT2
processes had to be transferred and dedicated interferometers for the two mirrors had to be set up.

Metrology as well as fabrication concepts and results will be presented in the subsequent sections.

M2
Sag=5.6jim
Slope=O.5 jim/mm

Ml
Sag=3.8j.tm
Slope=l .2 jim/mm

Fig. 2: MET-system, courtesy of LLNL The primary Ml and secondary M2 mirror substrates

Proc. SPIE Vol. 4146 39

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 5. PSD CONTROLLED PROCESS
According to the introductory remarks, surface topography has to be controlled over the whole spatial frequency domain
reaching from the clear aperture ofthe mirrors down to atomic dimensions.
Motivated by optical performance criteria this information usually is categorized into three subdomains:

. Figure (spatial frequencies down to appr. one tenth of the clear aperture) will mainly influence aberrations and thereby
the fidelity ofthe image (e.g. contact holes becoming elliptical, asymmetric line edges or unsufficient overlay).
S MSFR ( spatial structures down to appr. 1im) will result in intrafield scattering, thereby decreasing the contrast of the
image. Most problematic in that view is the imaging of isolated dark structures in a bright field.
. HSFR (spatial wavelength down to app. the wavelength of the actinic light) will disturb the proper constructive
overlay of the subwavefronts reflected from the multilayer stack, resulting in a drop of reflectivity. As a consequence
the system throughput in a litho system including up to 10 normal incidence components will suffer dramatically even
if the reflectivity of the individual surfaces drops only a few percent.

The transitions between these three categories are not very distinct. Nor is the response of the optical system uniquely
described by these numbers. Differing surface topographies, although resulting in different optical performance may yield
the same figure, MSFR and HSFR readings.

A more appropriate approach to characterize the surfaces is the concept of the power spectral density (PSD), originating
from system theory and already widely used to specify x-ray grazing incidence optics" . Surface data are decomposed by
Fourier transformation into their spectral components. By normalizing the squared amplitude spectra (the power spectra) to
the frequency sampling intervals the power spectral density ( PSD) is constructed from the readings of various instruments,
each covering a distinct spatial frequency range.

The 2D-PSD of a two dimensional surface map is defmed by:

2
. 1.1. i2tr(fx+fy)
y
S2(f,f)=Lzm—
A- AJjdxdye Z(x,y)
A

with Z(x,y) the surface profile at the position (x,y), A is the sampling area of the instrument. As can be seen from the
formula the 2D-PSD is measured in units of [length]4. By integrating the 2D PSD in polar coordinates with respect to the
angular coordinate one arrives at the 2D isotropic PSD which is due to its reduced complexity better suited for visualization
purposes.

The rms roughness values in a certain frequency band can be calculated by integrating the PSD within the proper frequency
limits. Owing to Parseval's theorem that is equivalent to backtransform the specific frequency band (here the phase
information is needed too) and calculate the roughness by integration in the spatial domain

The one dimensional PSD S,(f), calculated from a line height profile Z(x) of the surface is another quantity, especially
useful for the evaluation of AFM data

The PSD is extremely useful to

• tolerance the surface topography

• predict the scattering behaviour of the system
• analyse the fabrication processes in view of spectral properties
• detect metrology artifacts
• remove artifacts from the data by spectral filtering and backtransforming

40 Proc. SPIE Vol. 4146

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 Fig 3. displays an typical 2D-isotropic PSD of an EUVL substrate

The lowest spatial

!
frequency data origin
from interferometry, the
:.
.
:.i:.. .:' . ... i. . . MSFR-range is covered
by interference
:: microscopes. The HSFR
region is covered by
.. AFM—data. Since the
t! + interierometer—* whole frequency space
has to be controlled on
EUVL-mirrors, it is
Figure extremely important that
I no frequency gaps exist
*— microinterlerometer—* in the PSD. In general
several instruments with
MSFR overlapping bandwidths
have to be provided to
meat this goal. The
4 AFM stitching of the
. .. . I, subspectra and the
interpretation of the data
wavelength / inn
in the roll-off regions
Fig.3 Isotropic 2D-PSD of an exemplary MET Ml surface needs an understanding
of the frequency
response of these
instruments.

6. METROLOGY
1. Interferometry
Gravitational or stress induced deformations usually affect front and backside of a refractive lens in a similar way resulting
in a gross mutual cancellation of the induced wavefront errors. Only higher order contributions have to be dealt with. In
contrast, gravitational and stress induced deformations of reflective surfaces will directly add to the system wavefront error.
As a consequence it is common to all EUVL-metrology concepts, that the mirror substrates have to be measured in their
operational position and in their original mounts. Thereby mechanically induced surface distortions are fed forward to the
fabrication processes.

Figure specs of appr. 0.2mm rms call for a metrology process with precision and accuracy both well below 0. mm rms.
The more challenging quantity here is accuracy. It is strongly influenced by the properties of any ideal and especially as—
build auxiliary optics within the interferometer. Thus interferometers as free as possible from such components were
proposed and developed in the US-EUVL program'2

Another approach to provide the required level of accuracy is to remove the interferometer error by calibration. Calibration
is done by measuring a well known surface with the interferometer to be calibrated. The interferometer error is obtained by
subtracting the figure of the calibration surface from this measurement. Therefore, the systematic part of the residual
interferometer error is determined by the accuracy obtained for the measurement of the calibration surface.

Proc. SPIE Vol. 4146 41

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 In optical interferometry the highest accuracy is achieved on surfaces with spherical shape. For the rotational symmetric
figure components of spherical surfaces, measurement accuracies down to 0.15 nm RMS are obtained today. The non
rotational symmetric components of the calibration surface are of secondary importance, because they can be suppressed by
averaging measurements from different rotational positions of the surface'3.

Following this
approach, we can rely
calibration test on the experience in
Fizeau-type
interferometry at Carl
Zeiss'4"5,which has
Interferometer been applied very
successful in the
fabrication of
ultrahigh performance
optics at CZ.
Transmission
Fig. 4 shows the
sphere metrology concept for
the MET Ml mirror.
Interferometer and
transmission sphere
are calibrated by
compensation measuring an
System absolutely qualified
calibration sphere.
For the measurement
of the asphere, the
calibration sphere is
calibration sphere asphere
replaced by a
compensation lens and
the asphere. The
Fig 4 Metrology concept for the MET Ml minor. compensation lens
transforms the
spherical wavefront from the transmission sphere to the aspherical surface shape of the Ml asphere and vice versa.
Therefore, the interferometer is operated with a
spherical wavefront very similar to the wavefront
it has been calibrated for.

In effect, the interferometer measures the

wavefront formed by the compensation lens and
the asphere. The figure of the asphere has to be
calculated from the measurement and the optical
properties of the compensation lens. It is obvious,
that the measurement accuracy of the asphere
depends strongly on the accuracy of the
compensation lens data. Therefore, we measure
surface figures, curvatures, vertex thickness,
refraction index and material homogeneity of the
compensation lens with the highest accuracies
possible. If we add up the appropriate
measurement accuracies, we can estimate the
figure enor of the asphere introduced by the
compensation system to 0.13 nm RMS. If we
take into consideration the residual enor of the Fig. 5 Precision of MET- Ml interferometer: 0.O6nm rms

42 Proc. SPIE Vol. 4146

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 calibrated interferometer, the total systematic error for the Ml measurement increases to about 0.2 nm RMS.
The random error of the Ml measurement is given by the precision map shown in Fig. 5. We obtain a surface precision of
0.06 nm RMS. Precision here relates to the repeatability ofthe whole metrology cycle, including removal and re-insertion of
the optics into the interferometer.

Due to the high sampling density the interferometric data extend into the MSFR range. PSD analysis of the data helps to
reveal and eliminate spectral components not originating from the surface itself, as e.g. the sharp peak visible at the Nyquist
frequency which is due to the interlaced operation of the camera, or the sharply peaked multiple fringe kernel remnants at
the carrier fringe frequency.

2. Microinterferometry
Commercial microinterferometers are used to qualify the
MSFR. A Micromap 512 with 20x Mireau objective samples 1D-PSD, Comparison 1 D x- and y-direction
a 408 x 311 tm2 field with 640 x 480 pixels, thus providing 102

information in the spatial frequency domain from 408 tm1

to 1 .28 tm1 To bridge the gap to the interferometric data a
ZYGO NewView 200 equipped with a 2.5x Michelson
objective is used. This instrument samples a 3.3 x 2.5 mm2
area with 640x480 pxl resolution. Especially at that large
field size the sag of the substrates becomes relevant. Proper
I
detrending of the surface profiles is essential. Typically third
to fourth order polynomials have to be subtracted from the
profile to remove the contributions by the asphere itself (an
accurate feed forward of the asphere has not yet been
implemented). This detrending process corresponds to a high
pass filter. The fundamental and the first harmonics of the
power spectrum will be most affected. The MSFR-data are
completed at the low frequency side by the Ml wavelength I jim
interferometer which samples the surface with a pixel size of
appr. 6Oim. Thereby it is essential to have sufficient overlap
between the interferometer and the microinterferometer Fig. 6: Comparison of 1D-PSD in and perpendicular to the
PSDs in order to prevent a significant bandwith gap. AFM scan direction

3. Atomic Force microscopy

A Digital Instruments Dimension 5000

AFM operated in tapping mode is used to
qualify HSFR in various frequency bands.
Typically AFM data are taken with two
different scan sizes (1 jtm x 1 tm and
l0im x l0jim), to achieve sufficient
bandwidth overlap in the PSD with the
Micromap data and to cover the whole
range of wavelengths of interest down to
10 nm.
However unavoidably connected with the
scanning measurement principle, AFM
data suffer from a temporal drift of the z-
wavelength I
directional piezos. Commonly this z-drift
of the piezos is corrected by a flattening
Fig. 7 2D isotropic —PSD reconstructed from iD-data (red,lower curve) of the AFM data, which consists in the
compared to the dircetly evaluated 2D data (blue upper line) subtraction of an n-th-order polynom from
each single scan line.

Proc. SPIE Vol. 4146 43

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 As in the case of the microinterferometers, flattening will induces modifications of the PSD in the low frequency range.
Furthermore also destroys the spatial isotropy of the PSD in the range of the short wavelengths. This is demonstrated in
fig.6, showing a comparison ofthe 1-D PSD evaluated in the x- and y-direction.
The flattening correction shifts each single scan line by an arbitrary amount, and therefore the 1D-PSD in the y-direction
exhibits a white noise like character at the very short wavelength end ofthe PSD. This naturally also affects the 2-D PSD.

We therefore choose an indirect way for the evaluation of the AFM-based isotropic 2-D PSD. First the z-drift of the piezos
is corrected by flattening of 2 nd order. In the next step the lD-. PSD is evaluated in the x-direction, avoiding the artefacts
which are generated by the flattening step. From this the 2-D isotropical PSD, is calculated numerically by an inverse Abel
trrm1 This operation is reasonable for samples exhibiting a spatially isotropic roughness.
The overall agreement of the synthesized isotropical 2-D PSD with the directly evaluated 2-D PSD is shown in fig. 7. At the
short wavelength end the 'flattening-noise' reduction is visible.

4. Assisting surface analyses

The response of the surfaces to the specific processes is also monitored by surface analytical techniques as XPS,RBS, REM
etc. Main objective of these studies is to identify potential contaminations and physico-chemical structure changes of the
surfaces during fabrication.

7. FABRICATION
Starting with a sphere fabricated to microlithography standards two different idealized strategies can be thought of to meet
the requirements of EUVL-aspheres.

1) generate the asphere by a suited process and remove afterwards MSFR and HSFR by a smoothening process or
2.) Prepare a superpolished sphere as free as possible from MSFR and generate the aspherical contour by a roughness
conserving process.

In practice especially at the end of the process these two ideals will merge into a compatible iterative process, since every
smoothening process will cause some figure errors to be corrected afterwards- the aspherization in turn will add some power
to the MSFR and/or HSFR to be smoothened away. Both basic approaches have been and still are pursued at CZ to
determine the best starting point for the final conection of the substrates.

Computer controlled processes are used for aspherization, fme correction and smoothening of the mirrors. Some well known
processes ofthis kind are:

. Computer controlled polishing (CCP): Elastic subaperture tools are used to figure and smoothen the surfaces by
means of conventional polishing techniques . The basic tool functions of a linearly moved tool resembles a
sharply edged top-hat. By superimposing rotational movements, the tool functions can be modified, e.g to
generate Gaussian-like footprints. However, the resulting tool function will contain some residual high frequency
components which bear the risk to generate MSFRs if the process is not properly controlled.
The ultimate achievable figure and MSFR is determined by the fundamental statistics of the process, practicable
tool sizes, the capability ofthe elastic pads to adapt to the aspheric surface and the dynamics and mechanical
stability of the supporting machines.

. Magnetorheological fmishing (MRF): The substrate intersects a liquidjet, being composed of abrasive particles,
dispersed in a ferromagnetic fluid. The proper - and highly efficient - tool is formed by application of a magnetic
field in the interaction region. The tool is in close form contact with the substrate. Its size can be controlled by the
magnetic field configuration. Inherent to the sharply edged tool function is the risk to generate MSFRs, if the
process is not properly controlled. As in the case of CCP, achievable figure and roughness are strongly influenced
by the stiffness and dynamics of the machine. The process has a high wear rate and thus a high productivity. The
process is used for figuring only - not for polishing. Roughness is conserved to a certain extent.

44 Proc. SPIE Vol. 4146

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 . Ion beam figuring (IBF): A low energy (appr. lkeV) ion beam is used to figure the surface in an anisotropic non
reactive etching process. The tools are shaped by means of ion optics and can adapt perfectly to the aspheric
profile. IBF-tool functions resemble smooth gaussian type profiles, thus significantly reducing the risk of MSFR
generation. Stiffness ofthe machines is ofminor interest here since no reaction forces occur. Achievable figure is
determined predominantly by the practicable tool sizes and the dynamics ofthe drive and stage systems.
The IBF process is known to be roughness conserving on suited materials. Hence, MSFR and HSFR have to be
controlled by other process steps in advance or after the IBF-steps. As further consequence of this process-feature,
material inhomogeneities encountered during the edging path will be integrated and projected on the resulting
surface. Thus bulk material properties may become important when this technology is applied.

If MSFR and HSFR components are introduced into the power spectrum by the figuring process, efficient smoothening
procedures have to be applied in order to converge to the full specification.

8. RESULTS
The intermediate (Set 1) and end-specifications (Set 2) of the US-ETS-optics may be used as a benchmark for the achieved
surface quality16

spatial frequency band ETS Set 1 nm rms ETS Set 2 nm rms

Figure (CA)'-l mm' 0.50 0.25 Resolution,Distortion
MSFR 1 mni'-l jtm4 0.50 0.20 Flare,contrast
HSFR 1 jirri1-50 tm' 0.40 0.10 Reflectivity

Figure and MSFR results inbetween Seti and Set2 specs.were previously reported for the 1.8jtm ELT2 asphere'°.

Fig. 8.summarizes the results achieved so far on the MET Ml mirror. The cited figures are precision data, stemming from
the original precision-only metrology set-up. Since significant accuracy corrections had to be applied after switching to the
new accuracy-metrology set-up, the substrate had to be reworked. Work on achieving the same numbers on an accuracy
base are in progress. The numbers show that ETS-Set 1 specs had been reached in a very short term.
Present efforts are focused predominately on simultaneously generating figure and MSFR within the MET1 specs. Once
these specs are met, HSFR will be optimized by fine tuning the processes to the material properties.

Figure: 0.34 nm i-ms precision

MSFR: 0.37 nm rms
HSFR: 0.40 nm rms

Figure 8: Filtered surface interferogram of Ml substrate. The cited numbers correspond to the frequency bandwidths
defined by the ETS specs.

Proc. SPIE Vol. 4146 45

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 9. CONCLUSION AND OUTLOOK
Fully PSD-controlled processes were installed to meet the challenging specifications ofEUVL-suited mirror substrates.
High precision interferometry, meeting the EUVL requirements, has been demonstrated. High accuracy interferometiy
concepts are presently qualified. Microinterferometers and atomic force microscopes with overlapping spectral
bandwidths ensure a gap free continuation of the interferometric PSD down to atomic scales.

Surface quality of the alpha-tool class MET Ml on-axis mirror converged rapidly to ETS Set 1 specffications.
Convergence towards the twice as tight Set 2 specs slowed down considerably. Meeting the MSFR-spec. appear to be
the most challenng subtask. Meeting the HSFR specs will require a mutual optimization of material and fabrication
procedures.

Future work will focus on the stabilization and control ofthe processes in order to provide surfaces in set 2 spec.
In parallel the transfer of the process to alpha4ool class off-axis mirrors is started

10. ACKNOWLEDGEMENTS
The authors wish to express their thanks to the MET-team at VNL and to the International SEMATECH EUV Project
Manager Neil Wester for many fruitfiil discussions.
This work has been supported in part by the European Commission within the ESPRIT program (Project EP 28160)
The MET-work was partially supported by 1999-2000 International SEMATECH Project Lith-112

11. REFERENCES
1
R.H. Stulen et al., "Progress in the development of extreme ultraviolet lithography", SPIE 3676 Symposium on
Mircrolithography 1999.
2
Okazaki, "EUV program in Japan", SPIE 3676 Symposium on Microlithography, p 238-245, 1999.
3
.Benschop, U. Dinger, D. Ockwell, "EUCLIDES: first phase completed", SPIE 3997, 2000
4
T. Jewell;.Optical design issues in development ofprojection camera for EUV lithography, SPIE 2437, pp 340, 1995 and
references therein
5
E. Spiller, " Soft x-ray optics, "SPIE Optical Engineering Press, 1994
6
w. Uirich, K.-H.-Schuster, H. Beierl, A. Epple, R.M. von Buenau, H.-J. Mann, U. Dinger, G. Fuerter, "Trends in optical
design of projection lenses for UV- and BUy-lithography", these proceedings
7
D. Williamson, "High numerical aperture ring field optical reduction system", US Pat. 58153 10 (1998)
S
M. Antoni, W. Singer, J. Wangler, I Escudero-Sanz, B. Kruizinga, " illuminastion optics design for EUV-lithography",
these proceedings
9
T. Jewell, J. Rodgers, K.-P. Thompson, Reflective
" systems design study for soft x-ray projection lithography", J. Vac.
Sci. Technology B 8, 15 l9pp (1990) and references therein
M. Suzuki et al., " X-Ray reduction projection exposure system of reflection type", US Pat. 5 153 898 (1992).
10
H. Handschuh et al., "EUV lithography at Zeiss: Manufacturing and metrology ofaspheric surfaces with angstrom
accuracy", Proc. EJPBN 1999
11
F. Church, P. Takacs," The optimal estimation offinish parameters", Proc. SPIE 1530 (1991), '7lpp
"Standard practice for estimating the powerspectral density function and related finish paramaters from surface profile
data", ASTM Designation F181 1-97
12
G Sommargren, "Phase Shifting Diffraction Interferometry for Measuring Extreme Ultraviolet Optics", OSA TOPS on
Extreme Ultraviolet Lithographym Vol.4, 1996, lO8pp
13
C. J. Evans, R. N. Kestner, "Test optics error removal", Appl. Opt. 35, 1015-1021 (1996)
14
M. KUchel, "New Zeiss interferometer", Procs. SPIE Vol. 1332 Optical Testing and Metrology III: Recent Advances in
Industrial Optical Inspection, pp. 655-663, (1990)
15K. Freischlad, M. Küchel, K.H. Schuster, U. Wegmann, W. Kaiser "Real-time wavefront measurement with lamda/10
fringe spacing for the optical shop", Procs. SPIE Vol. 1332 Optical Testing and Metrology III: Recent Advances in
Industrial Optical Inspection, pp. 18-24 (1990)
6 Taylor;"Optical System for EUV Lithography";. Optical Fabrication and Testing, Quebec 2000, OSA Technical digest

46 Proc. SPIE Vol. 4146

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx