The refined EUV mask model

I.A. MAKHOTKIN1*, M. WU1, V. SOLTWISCH2, F. SCHOLZE2, V. PHILIPSEN1

1.IMEC, Kapeldreef 75, 3001 Leuven, Belgium
2. Physikalisch-Technische Bundesanstalt, Abbestr. 2-12, 10587 Berlin, Germany

Abstract.

A refined model of an extreme ultraviolet (EUV) mask stack consisting of the

Mo/Si multilayer coated by a Ru protective layer and a TaBN/TaBO absorber
layer was developed to facilitate accurate simulations of EUV mask performance
for high-NA EUV photo-lithography (EUVL) imaging. The model is derived by
combined analysis of the measured EUV and X-ray reflectivity of a state-of-the-
art mask blank. These two sets of measurements were analyzed using a combined
free-form analysis procedure that delivers high-resolution X-ray and EUV optical
constant depth profiles based on self-adapted sets of sublayers as thin as 0.25nm
providing a more accurate description of the reflectivity than obtained from only
EUV reflectivity. “Free-form analysis” means that the shape of the layer-
interfaces in the model is determined experimentally and is not given a priori by
the structure model. To reduce the numerical effort for EUV imaging simulations
a low-resolution model of the multilayer and absorber stack with sublayer
thicknesses larger than 2nm, that fits to only the EUV reflectance, was derived
from the high-resolution model. Rigorous high-NA EUVL simulations were done
to compare the performance of the new model to our previous work 1.

*Corresponding author
[email protected] (Igor A. Makhotkin)
This article has been submitted to the Journal of Applied Physics.

1. Introduction

The EUV mask is one of the key components of the photo lithography setup that
can be designed and optimized by the end-user. The mask design largely
determines the EUV lithography performance. The development of new masks is
based on the large-scale numerical optimization of the EUV imaging quality by
varying the parameters of the absorber patterns. Such mask optimization, as for

1
example discussed in 2, can be used for the mitigation of mask 3D effects in order
to improve EUV imaging.
A comprehensive understanding of the EUV mask stack (multilayer and
absorber) is required to explore EUV imaging at high NA using rigorous mask 3D
lithography simulations and to support EUVL at current NA 0.33 using full-field
1
design modeling software. Current mask model was presented in 2013 and is
calibrated to the EUV reflectivity measured from the, at that time, state-of-the-art
mask blank. The recent developments in mask making process as well as in EUV
multilayer metrology calls for the model update.
The detailed study of the periodic multilayer EUV reflectivity (EUVR)
analysis 3 showed that the single-wavelength EUV-only reflectivity measurement,
although being very sensitive to minor structural changes of the multilayer,
generally cannot be used for accurate determination of the sample structure
because of the highly correlated influence of multilayer structural parameters, for
example multilayer thickness ratio, densities and stoichiometries of layer
materials. A minor change of the ratio between layer thicknesses in the multilayer
Comodel will change the simulated EUVR curve, however this change can be
compensated by the change of total bi-layer thickness and layer density, making
it impossible to determine accurate structural parameters from EUVR fit only 3.
To solve such correlation a combination of X-ray reflectivity (XRR) and
EUVR measurements 3 was proposed, as well as the more complex combination
of EUVR with scattering and X-ray fluorescence measurements 4. In this paper
we will present a mask stack model – that simultaneously fits to both EUVR and
XRR measurements of the actual mask. This combination allows us to accurately
describe the structure and optical properties of the mask stack for EUV imaging.
The implementation of the presented mask stack model in mask 3D aware
simulation tools will enhance their predictive and pre-compensation power.

2. Mask measurements and data analysis

The presented model is built based on analysis of two sample structures from
current state-of-the-art EUV mask blanks. The first sample (referred to as MLM)
is a periodic Mo/Si multilayer mirror containing 40 Mo-Si bilayers with the period
thickness of 7nm deposited on SiO2 glass substrate. The multilayer is covered
with a 3nm protective Ru layer. The second sample, referred to as absorber, is
TaBN-TaBO absorber bi-layer with a thickness of 58nm and 2nm respectively
deposited on the glass substrate coated with a 3nm thin Ru layer. In this set both
samples have identical Ru layers and therefore the obtained models can be merged
by overlapping the identical Ru layer.

2
 Figure 1. Scheme of the a) Mo/Si multilayer and the b) TaBN/TaBO absorber
layer with corresponding measured EUV reflectivity maps.

The grazing incidence XRR is measured using Malvern Panalytical X’pert MRD
XL multipurpose diffractometer equipped with a Cu X-ray source. For XRR
measurements the Cu K X-ray beam was conditioned, using the combination of
parallel beam mirror and 4x Ge 220 monochromator enabling measurements over
the large dynamic range of 107 counts per second with high resolution of 0.012
degrees, determined by the beam divergence. The EUV reflectivity measurements
were conducted at the soft X-ray beamline of the Physikalisch-Technische
Bundesanstalt at the electron storage ring BESSY II, which covers the photon
energy range from 50 eV to 1900 eV. The beamline is equipped with a ultra-high
5, 6
vacuum lubrication free Ellipso-Scatterometer . The EUV reflectivity curves
were measured using s-polarized radiation for the full accessible angular range
from grazing incidence to near normal and mapped over the incidence wavelength
from 10nm to 18nm. The experimental uncertainty is considered for every angle
of incidence scan separately to take into account for beam instabilities during the
mapping. The schemes of the two measured samples together with EUV
reflectivity maps are presented in Fig. 1.
For the analysis of the sample structures, only the incidence angle scan at
13.5nm wavelength was used from the measured EUVR map. The data used for
the analysis of XRR and EUVR are shown in Fig. 2a, 2b and 3a, 3b.
The analysis was performed by following steps. The high-resolution profile of
the MLM and absorber. samples is obtained using a modified free-form approach
similar to discussed by Zameshin et.al. in 7. For a free-form analysis of X-ray and

3
EUV reflectivity curves, the analyzed film is modeled as a set of thin sub-layers
where the optical constants of each sublayer are determined by sub-layer chemical
composition and density 3. The stoichiometry values and densities were coded
using the array of integers P as described in 8. These values of P and total layer
thicknesses were the only fitting parameter. In this way, a set of consistent optical
constant profiles can be calculated for various wavelengths having equal sub-layer
densities and stoichiometries and changing only wavelength-dependent atomic
scattering factors 9. The extended description of applied data analysis procedure
is beyond the scope of current paper and therefore will be published elsewhere.
The maximum thickness of the sublayer defines the in-depth resolution of the
optically constant profile of the thin film and is determined by the measurement
with the highest resolution. For both samples the X-ray reflectivity determined
the resolution of the high-resolution model to be 0.25nm corresponding to the
measurement range of 9 degrees (Fig. 2a) at 0.154nm wavelength (see 7 for more
details). For the XRR analysis of the absorber sample a sub-layer thickness of
0.5nm was used as the XRR measurement is informative only till 4 degrees (Fig.
3a). The profile steps in the Ru layer were kept as small as in the MLM model for
the ease of the later merger of the models. Following the same logic, the minimal
steps in both profiles to fit EUV reflectivity can be as large as 3.3nm assuming
measurements at 13.5nm till 88 degrees grazing (2 degrees normal) incidence.
Regardless of their low in-depth resolution, the EUVR data contribute to the
combined analytical accuracy of the determination of densities and chemical
stoichiometries of absorber and MLM models3.
However, the fact that the optical constant model builds up from sublayers
with thickness of ~3.3nm can be accurate enough to fit the measured EUVR data,
means that for EUV lithography simulations a low-resolution model can be build
based on EUVR-only fitting, that produce as accurate as high-resolution model
simulation results. Consequently, as a second step of the analysis we obtain here
the low-resolution models for MLM and absorber layer by fitting only the EUVR
data. The initial guess model here was build based on the high-resolution model
by combining its thin sublayers to thicker ones with averaged stoichiometries and
densities.

4
 2.1 The Mo/Si multilayer model

a) b)

1E+06
Measurements
High-resolution model 0.6
1E+05

EUV reflectivity at 13.5 nm

X-ray reflectivity, CPS 1E+04

1E+03 0.4

1E+02

1E+01 0.2 Measurements

low-resolution model
1E+00 high-resolution model
2013 model
1E-01
0 2 4 6 8 10 0.0
70 72 74 76 78 80 82 84 86
Grazing incidence angle, degrees
Grazing incidence angle, degrees

c)

Figure 2 (a) Measured XRR of the Mo/Si multilayer and its best fit solution. (b)
Measured EUVR of the Mo/Si multilayer and the simulated EUVR from the
low- and high-resolution multilayer model, as well as the simulated EUVR
using the 2013 multilayer model. (c) The reconstructed low- and high-
resolution -profiles for a single Mo/Si bi-layer calculated for EUV wavelength
of 13.5nm.

The measured and simulated (using best-fit results) X-ray and EUV reflectivity
curves of the MLM sample are shown in Fig. 2a and 2b, respectively. The best
fitted optical constants profiles, calculated for 13.5nm wavelength are shown in
Fig. 2c by , the decrement of the real part of optical constant n, =real(1-n),
while the fitted complex value of of the optical constants n,  can be found in
Table 1.

5
 In the -profile of the MLM sample (Fig. 2c) the Ru layer is shown between
0 and 7nm and the Mo/Si bilayer structure between 7nm and 14nm. We show the
structure of only one Mo-Si bi-layer, because in our model we consider all 40 bi-
layers to be identical. This assumption is supported by the good fit of the measured
data, especially by the excellent agreement between the width of the measured
and simulated Bragg peaks on the XRR curve.
The high-resolution profile shows the well-known interface asymmetry for the
Mo/Si multilayer, namely the Mo-on-Si is larger than Si-on-Mo interface 10. The
low-resolution profile of the periodic MLM part just roughly describes the high-
resolution shape. However, as was mentioned afore, both high- and low-resolution
profiles are equally good for fitting of the EUVR measurements. Figure 2b shows
that both new high- and low-resolution models fit much better to the new EUVR
measurements than the previous model from 2013 1, possibly due to the
modification of the multilayer structure.

2.2 The absorber model

The analysis of the absorber layer was a bit more complicated, since the same
approaches we used in the analysis of the multilayer mirror with tabulated CXRO
atomic scattering factors did not yield a consistent fit for X-ray and EUV
reflectivity, although the solution for X-ray was relatively straightforward. The
reasons may be the uncertainty of the exact chemical composition and the
inaccuracy of atomic scattering factors for mask materials. The manual correction
of both  and  of optical constants for -4% for TaBN and +50% for thin surface
TaBO layers of the EUV optical constants were necessary to obtain the good
combined agreement between EUVR and XRR fitting. This assumption about
such correction of optical constants for TaBN layer was confirmed by the rigorous
analysis of the optical constants for the TaBN/TaBO structure in the EUV
wavelength range performed at PTB (not presented here).

The figures 3a and 3b show that the high-resolution model reasonably

well fits to both X-ray and EUV reflectivity measurements. Figure 3b also shows
that the low-resolution model fits even better to EUVR measurements than the
high-resolution model. This is expected as the low-resolution model was obtained
by fitting of EUVR-only measurements and presents one of the deep local
minima, while high-resolution model can be seen as a Pareto optimal solution for
the combined X-ray and EUV reflectivity data sets. The comparison between
high- and low-resolution -profiles for 13.5nm wavelength and high-resolution
profile for 0.154nm wavelength are shown in Fig. 3c.

6
 a) b)

107 1

106 EUVR measurements

XRR measurements high-res. fit
high-res. fit low-res. fit
105 0.1

X-ray reflection, cps

EUV reflectivity
104

103 0.01

102

101 0.001

100

10-1 1E-4
0 1 2 3 4 0 20 40 60 80
Grazing AOI, degrees
Gazing angle of incidence, degrees

c)

Figure 3 Measured and best fit solutions for Ru/TaBN/TaBO thin film.(a)
measured and best-fit high-resolution model calculated curves for XRR; (b)
measured best-fit high- and low-resolution models for EUVR; (c) real part of
decrement of the constants profiles () shown for 0.154nm and 13.5nm
wavelengths for high-resolution model together with low-resolution 
calculated only for 13.5nm wavelength.

From the high-resolution model we can conclude that only 55nm of the nominally
60nm thick absorber layer has a constant density while the rest is consumed by
interfaces and transition layers. Fig. 3c shows that the top 2nm TaBO layer most
likely forms only oxygen rich interface transition regions. The underestimation of
O content in TaBO model can explain the necessity for manual increase of surface
density as 13.5nm light scatters much more effectively on oxygen atoms than X-
rays due to proximity of O L absorption edge. The remarkable drop, visible on the
 profile calculated for X-ray wavelength at z=0nm, can be an indication of the

7
presence of an oxidized Ru layer, what might be caused by exposure of the sample
to ambient between depositions of Ru and Ta-based absorber layers. It should be
noted that, as the uncertainty of EUV-only data analysis are quite large, the high-
and low-resolution models can coincide within the uncertainty corridors of low-
resolution profile.
For EUV lithography imaging simulations we have combined the low-
resolution models of the multilayer and absorber layer stack. During the
combination we fixed the Ru layer as it was determined for the MLM sample. The
combined multilayer and absorber low-resolution model of the analyzed EUV
mask is presented in Table 1. In the following part we compare the EUV
lithography imaging simulations performed using the low-resolution model
shown in Table 1 and presented in 1 in order to analyze the influence of the mask
model on high-NA imaging.

Table 1. Low-resolution model for TaBO/TaBN absorber on a Ru/Mo/Si MLM mask.

Layer Layer optical constants for 13.5nm EUV light (1--i )

thickness,nm

TaBO/TaBN absorber layer

1.44 0.99375-0.00228i

1.44 0.96463-0.01841i

1.44 0.95139-0.03037i

55.02 0.95056-0.03163i

2.57 0.94800-0.03026i

Ru protective layer

1.55 0.93925613-0.011132225i

2.465 0.89243499-0.016227441i

1.61 0.94843207-0.0088824872i

40x Mo/Si multilayer

1.435 0.99889336-0.001918392i

1.215 0.96323059-0.0041547321i

1.807 0.92539394-0.0063108367i

1.422 0.96379023-0.0041027391i

1.151 0.9976784-0.0019250528i

3. Impact on key parameters in EUVL simulations

The EUV lithography simulations presented in this paper are performed with the
11
rigorous mask 3D simulation software S-Litho EUV (Synopsys) . Lines and

8
spaces through pitch are imaged with a dipole leaf shape illumination at NA 0.55
using 4x/8x reduction system and 20% central obscuration of the projector pupil.
The unpolarized EUV light is incident on the mask at 5.4 degree chief-ray angle12.
The lines and spaces are evaluated over a pitch range from 16nm to 40nm, where
the target critical dimension (CD) is the half-pitch value for pitches until 32nm
and fixed to CD 16nm at larger pitches as can be seen on the curve target CD in
Fig. 4a. The aerial image threshold is fixed to print the smallest pitch on target
and the mask CD (MCD) required to print the other horizontal pitches to target at
the fixed threshold is plotted in Fig. 4a. Lithography metrics such as best focus
(BF), depth of focus (DoF), exposure latitude (EL) and telecentricity error (TE)
are presented in Fig. 4.
The simulations denoted as “Low-res. model” are performed using the mask
model presented in Table 1 and results denoted as “2013 model” are obtained with
the mask model presented in [1].
a) b)

c) d)

e)

9
 Figure 4 Comparison between EUVL metrics through pitch at high NA 0.55
calculated using the mask model presented here (denoted as Low-res. model)
and presented in [1] (denoted as 2013 model).

Fig. 4 shows that regardless of the difference in the mask models there is not much
difference between key EUVL metrics. The small differences can be explained by
the relatively small influence of the EUVR for grazing incidence angles lower
than 75 degrees (i.e., higher than 15 degrees from the normal) where EUVR
curves differ most (see Fig. 2b).

4. Conclusions

We have analyzed the internal structure of the current state-of-the-art EUV mask
blank using XRR and EUVR measurements. The high- and low-resolution models
of the optical constant profiles from the mask blank are reported. The comparison
of the mask model from 2013 to the newly proposed mask model is performed for
next-generation high NA EUV simulation settings on lines and spaces. The
simulations show that the slight difference in measured EUVR at high incidence
angles does not change the key EUVL metrics dramatically.

The high-resolution model gives an accurate description of the internal

multilayer and absorber layer structures. The combined XRR and EUVR
measurement of EUV mask blanks can be used to analyze changes in the
multilayer and absorber structure in future mask R&D using the model described
here as a reference.
Finally, we recommend the use of the new low-resolution mask model for
next-generation EUVL simulations, as it fits better with the current state-of-the-
art masks. The implementation of the presented low-resolution mask model
should increase the simulation accuracy of more complex designs.

Acknowledgements

We acknowledge dr. Sergey N. Yakunin for the assistance development of free-

form analysis code of combined EUV-XRR data sets and dr. Joachim Woitok and
Malvern Panalytical for the help with measurement of EUV mask blanks.

1. V. Philipsen, E. Hendrickx, R. Jonckheere, N. Davydova, T. Fliervoet

and J. T. Neumann, Actinic characterization and modeling of the EUV mask

stack. (SPIE, 2013).

10
2. A. Erdmann, P. Evanschitzky, H. Mesilhy, V. Philipsen, E. Hendrickx

and M. Bauer, Journal of Micro/Nanolithography, MEMS, and MOEMS 18 (1),

1-13, 13 (2018).

3. S. N. Yakunin, I. A. Makhotkin, K. V. Nikolaev, R. W. E. van de

Kruijs, M. A. Chuev and F. Bijkerk, Optics Express 22 (17), 20076-20086

(2014).

4. A. Haase, S. Bajt, P. Honicke, V. Soltwisch and F. Scholze, Journal of

Applied Crystallography 49 (6), 2161-2171 (2016).

5. C. Laubis, A. Barboutis, M. Biel, C. Buchholz, B. Dubrau, A. Fischer,

A. Hesse, J. Puls, C. Stadelhoff, V. Soltwisch and F. Scholze, presented at the

SPIE 8679, Extreme Ultraviolet (EUV) Lithography IV, 2013 (unpublished).

6. F. Scholze, C. Laubis, K. V. Luong and V. Philipsen, Update on optical

material properties for alternative EUV mask absorber materials. (SPIE, 2017).

7. A. Zameshin, I. A. Makhotkin, S. N. Yakunin, R. W. E. van de Kruijs,

A. E. Yakshin and F. Bijkerk, Journal of Applied Crystallography 49 (4), 1300-

1307 (2016).

8. I. A. Makhotkin and S. N. Yakunin, (2016).

9. B. L. Henke, E. M. Gullikson and J. C. Davis, Atomic Data and

Nuclear Data Tables 54 (2), 181-342 (1993).

10. S. Braun, H. Mai, M. Moss, R. Scholz and A. Leson, Japanese Journal

of Applied Physics 41 (Part 1, No. 6B), 4074-4081 (2002).

11. https://www.synopsys.com (2019).

12. E. v. Setten, G. Bottiglieri, L. d. Winter, J. McNamara, P. Rusu, J.

Lubkoll, G. Rispens, J. v. Schoot, J. T. Neumann, M. Roesch and B. Kneer,

Edge placement error control and Mask3D effects in High-NA anamorphic EUV

lithography. (SPIE, 2017).

11