Full text of DOI:10.1007/BF02665848

ImprovementoftheAccuracyoftheIn-situEllipsometricMeasurementsofTemperature
Journal of ELECTRONIC MATERIALS, Vol. 30, No. 6, 2001
Special Issue P6a3pe7r
and Alloy Composition for MBE Grown HgCdTe LWIR/MWIR Structures
Improvement of the Accuracy of the In-situ Ellipsometric
Measurements of Temperature and Alloy Composition for
MBE Grown HgCdTe LWIR/MWIR Structures
M. DARASELIA,1,4 J.W. GARLAND,1 B. JOHS,2 V. NATHAN,3
and S. SIVANANTHAN1
1.—Microphysics Laboratory, Department of Physics, University of Illinois at Chicago, IL 60607.
2.—J. A. Woollam Co., Inc., 645 M. Street, Suite 102, Lincoln, NE 68508. 3.—Air Force Research
Laboratory/Space Vehicles Directorate, Kirtland Air Force Base, NM 87117.
4.—e-mail: mdaras1@uic.edu
Spectroscopicellipsometry(SE)hasproventobeaveryreliabletechniqueforthe
in-situ monitoring of the substrate temperature and alloy composition during
the HgCdTe epitaxy. In this work, the influence of the variations in the angle of
incidence and the spectral wavelength shift on the measured accuracy of the
growth temperature and alloy composition are studied, and a method for
precisely determining these variations independent of the modeling of the SE
datahasbeendeveloped. Itisshownthatthestabilityofthefittingsoftheoptical
models for in-situ applications increases and that the couplings between model
parameter decreases upon either eliminating the angle of incidence as an
independent model parameter or correcting for the shifts of the wavelength
offset. The variations in the angle of incidence and wavelength shift, which arise
in the M88 ellipsometer from reflected beam deflections, were precisely cali-
brated in two dimensions as a function of alignment parameters, using a thick
thermally grown SiO2/Si sample and were parameterized for our experimental
geometry. A new extension of the WVASE software was developed to correct the
raw SE data in real time for wavelength shift and the angle of incidence drift. A
comparison of the corrected and uncorrected results of in-situ temperature
measurements for HgCdTe and CdZnTe(211)B/Si(211) clearly demonstrates
that the proposed method significantly enhances the accuracy of temperature
and composition readings over a broad range of values in these parameters.
Key words: Hg1–xCdxTe, LWIR, MWIR, CdTe, CdxZn1–xTe, spectroscopic
ellipsometry, temperature, alloy composition, in-situ monitoring
[mid-wave infrared (MWIR)] and 8 mm and 14 mm
INTRODUCTION
[long-wave infrared (LWIR)] in wavelength, making
this material widely used in military and space appli-
Recently Hg1–xCdxTe (MCT) became the preferred
material for the fabrication of infrared photo detec-
torsandarraysbecauseofitshighquantumefficiency
and wide operating temperature range.1 The alloy
composition x for this ternary alloy and hence the
band gap can be tuned in order to access the atmo-
spheric transmission windows between 3 and 5 mm
cations.2,3
Molecular beam epitaxy (MBE) has demonstrated
to be a flexible manufacturing technology for the
growth of MCT layers and heterostructures,4 particu-
larly on lattice matched CdZnTe substrates.5 Growth
of high quality lattice matched MCT structures re-
quires accurate control of the substrate temperature
and alloy composition. Remote sensing techniques,
(Received November 21, 2000; accepted January 16, 2001)
637

638
Daraselia, Garland, Johs, Nathan, and Sivananthan
alloy composition, substrate temperature, and layer
thickness parameters, unless properly accounted for
during data analysis.
Table Ia. Correlation Matrix for MCT Alloy
Composition Model Parameters Obtained for
Free Angle of Incidence
For our system consisting of a Riber 3200 MBE
reactor and an M88 spectroscopic ellipsometer,10 we
have performed a detailed study of the problems
caused by sample alignment variations. Analysis of
the reasonably large number of MCT growth runs
showed that the AOI varies by ~0.15∞ during a single
run and has run-to-run non-repeatability of ~0.35∞.
These values are considered to be large if one consid-
ers the high sensitivity of SE data to the angle of
incidence.
MCT
Comp.
Oxide
Angle
Comp.
Oxide
Angle
1.000
0.838
–0.774
0.838
1.000
–0.849
–0.774
–0.849
1.000
Table Ib. Correlation Matrix for MCT Alloy
Composition Model Parameters Obtained for
Fixed Angle of Incidence
We have studied the influence of the angle of inci-
denceparameterontheaccuracyofMCTalloycompo-
sition determination. The two-layer ellipsometric
model used for this study employed a composition
dependent MCT optical constant library and an effec-
tive medium approximation for surface roughness.
The surface roughness thickness, the angle of inci-
dence and the MCT alloy composition were set as free
parameters. The stability of the model with respect to
any pair of these parameters is determined by the
corresponding cross-correlation coefficient in the cor-
relation matrix. The comparison between correlation
matrices obtained with and without a free AOI pa-
rameter is shown on Table Ia and b. The exclusion of
the AOI as a free parameter clearly leads to reduced
correlation coefficients and therefore smaller error
bars. The increased fit stability in this case is
observed as reduced noise in the time-dependent
composition, temperature, and thickness plots. How-
ever, thequestionthatisraisedhereiswhetherornot
we can keep the AOI constant during the entire
growth monitoring.
MCT
Comp.
Oxide
Comp.
Oxide
1.000
0.540
0.540
1.000
such as spectroscopic ellipsometry (SE), have re-
cently attained remarkable success in the real time
monitoringoftheseimportantparameters.6–8 Certain
experimental configurations involving indium-free
sample holders almost completely rely on SE tem-
perature measurements due to the lack of efficient
thermocouple control.9
The use of the epitaxial CdZnTe(211)B/Si(211) or
CdTe(211)B/Si(211) structures as composite sub-
strates for HgCdTe growth requires a substantially
higher accuracy in the SE analysis, owing to larger
deviations in the surface morphology and bulk prop-
erties of CdTe and CdZnTe epilayers as compared to
those of the corresponding bulk materials.
The desired layer thickness, composition, and tem-
perature are obtained in real time through the fitting
of an optical model to experimental SE data. The
accuracy of the determined parameters is primarily
dependent on both the degrees of approximation of
the data in the model and the ability of the model to
provide a unique solution with respect to any pair of
free model parameters. The overall model stability
and parameter error bars are mainly defined by the
number of free parameters and the degree of correla-
tion between them. Therefore, one of the ways to
achieve better stability of the fitting process and
to improve the overall accuracy of the analysis is to
reduce the number of free model parameters involved
in the fit.
Another significant source of inaccuracies in the
ellipsometric analysis arises from the raw spectral
data itself. Unfavorable experimental conditions as-
sociated with in-situ experiments include sample
surface vibrations and shifts out of alignment due to
thermal displacements of the manipulator compo-
nents. These problems are generally responsible for
drifts in the angle of incidence (AOI) and spectral
disturbancesoftherawSEdata.Insomeellipsometer
setups the individual spectral channel central wave-
length shifts occur as a direct result of reflected beam
deflections. These spectral shifts may have a dra-
matic influence on the accuracy of the determined
Fig. 1. Variation of the SE measured alloy composition correspond-
ing to a DQ = 0.15∞ variation in the angle of incidence obtained for
different MCT alloy composition settings. The linear regression is
applied to data to estimate the composition error for LWIR MCT
structures.

ImprovementoftheAccuracyoftheIn-situEllipsometricMeasurementsofTemperature
and Alloy Composition for MBE Grown HgCdTe LWIR/MWIR Structures
639
the typical variation in the AOI produces up to ~28∞C
uncertainty in the ellipsometrically measured sur-
face temperature, a value significantly higher than
acceptable limits for MCT processing.
In addition to the problems created by uncertain-
ties in the angle of incidence, we also studied the
wavelength shift arising in the M88 SE family. This
shift of the central energy in each spectral channel of
theellipsometeractsinthesamefashionastheactual
shift of the semiconductor critical point (CP) energy
with temperature and composition and hence it con-
tributes to significant errors in the determination of
these parameters. This effect will be discussed in
more detail in the following section together with a
method for maintaining misalignment-independent
real time SE analysis.
WAVELENGTH SHIFT CALIBRATION
Thewavelengthcalibrationproposedinthissection
is based on the unique ability of the M88 SE to sense
smalldeflectionsinthedirectionofthebeamentering
the analyzer unit. The M88 detector design is shown
on Fig. 3.
Fig. 2. VariationintheSEmeasuredCdTetemperaturecorresponding
to a DQ = 0.15∞ variation in the angle of incidence obtained for the
temperature range of 200∞C to 500∞C. The vertical dashed line
separates simulation results performed on SE data sets obtained at
different spectral ranges.
After passing through the rotation analyzer and
the pinhole, the beam is doubly dispersed on the
positive and the negative orders by the diffraction
grating. Two ultra-violet-visible (UV-VIS) and near
infrared (NIR) detector arrays provide two parts of
the spectrum with 44 wavelength channels each and
matched at l ª 500 nm. The undispersed zero-order
reflection from the grating is being picked by a quad-
rant detector, which is used to assist in the alignment
of the detector unit by balancing the slit image on the
detector center. Any deviations from perfect align-
ment produce non-zero values for the alignment pa-
rameters AlignX and AlignY in the commercial
Several authors proposed keeping the angle pa-
rameter fixed at the best-fit value obtained at the
beginning of the growth run and have reported a
significant improvement in alloy composition repeat-
ability.11 However, in our configuration, the long-
term angle variations were found to cause significant
systematic errors in composition and temperature if
neglected.Theseerrorswereestimatedbypost-growth
analysis of several MCT real time SE data sets.
Figure1showsthesesimulations, obtainedbyforcing
AOI to deviate 0.15∞ from the best-fit value and
recording the corresponding composition variation.
The SE data were modeled in the spectral range from
1.8 eV to 3.3 eV. Each data point on the plot corre-
sponds to the different x-value used in LWIR and
MWIRstructures.Thissimulationdemonstratesthat
the variation in the AOI of dQ = 0.15∞ produces
dx ª 0.012 uncertainty in the alloy composition for
MWIR structures. Extrapolation of this result to
LWIR structures using linear regression gives a
slightly higher prediction of dx ª 0.014.
A substantially less satisfactory conclusion is asso-
ciated with CdZnTe or CdTe temperature control.
Figure 2 shows a simulation where the CdTe tem-
perature variations for dQ = 0.15∞ are plotted in the
temperaturesetpointrangefrom200∞Cto500∞C.The
actual surface temperature in our configuration is
substantially lower, than the set point, so that the
region of typical MCT growth temperature is fully
covered. The spectral range was selected to be from
2.5 eV to 4.5 eV to cover the high dispersion region of
the CdTe dielectric function. The three lowest tem-
perature experimental points in Fig. 1 were obtained
from the SE data analysis in a shorter spectral range
(from 3.0 eV to 4.2 eV), and showed higher angular
dependence. As we see from this figure, at T = 200∞C
Fig. 3. M88 detector unit design. (with permission of. J. A. Woollam
Co., Inc.).

640
Daraselia, Garland, Johs, Nathan, and Sivananthan
Si temperature, the SiO2 thickness, and AOI were
obtained in-situ at nearly perfect beam alignment.
The fit yielded a thickness of 2581.8 Å with a unifor-
mity of 4.55%. These parameters were fixed as best-
fit values, with only the wavelength shift left as a fit
parameter.
Table II. Best Fit Coefficients for Equation 1
Obtained During the Wavelength Calibration
Procedure*
A
E
B
F
C
MSE
D
The actual calibration was performed in-situ by
adjusting the detector mounting so that the align-
ment parameters scanned the 2D square area and by
fitting the corresponding SE data for the wavelength
shift. We obtained a total of 81 calibration points with
AlignXandAlignYcoveringuniformlytheentirearea
of typical beam deflections. To decrease the random
noise, a total of 10 experimental points was averaged
for all pairs of alignment parameters. The quadratic
form
E < 2.48 eV
–4.75E-2
E > 2.48 eV
4.16E-2
4.83E-5
9.54E-4
–1.89E-5
–1.46E-3
2.35E-5
2.452
1.16E-5
0.254
6.94E-6
1.11E-7
9.01E-7
1.10E-7
* The two rows correspond to the NIR and UV-VIS detector array
wavelength shift calibrations.
Table III. Best Fit Coefficients for Equation 1
Obtained During the Angle of Incidence
Calibration Procedure*
Dl[nm] = A ◊ x2 + B ◊ y2 + C ◊ x ◊ y
+ D ◊ x + E ◊ y + F
(1)
A
E
B
F
C
Theta
D
MSE
was used to parameterize the results for each of the
two spectral ranges, and the A–F parameters were
obtained using a least-square-error fit of the data.
TableIIshowsthebest-fitcoefficientsfortwospectral
ranges: E < 2.488 eV and E > 2.488 eV. An analysis of
the coefficients shows that the wavelength shift de-
pends primarily on the displacement in the X-direc-
tion and that the two spectral regions exhibit wave-
length shifts in opposite directions. The non-zero
values of the F parameters indicate small inaccura-
cies in the initial detector calibrations.
Theobtainedwavelengthcalibrationcanbeusedin
virtually any type of in-situ SE experiments with the
use of our newly developed extension of the WVASE
software,whichmakesreal-timecorrectionoftheraw
SE data before the actual SE analysis is performed.
8.02E-7
1.26E-6
70.89
–7.95E-7
8.88
1.14E-3
4.37E-6
–9.39E-5
* The Theta parameter represents the rotation of the detector coordinate
system corresponding to the detector azimuth change between the ex-situ
and in-situ configurations.
WVASEsoftware.Ourwavelengthcalibrationmethod
is based on the fact that these alignment parameters
uniquelydefinethedeflectionoftheenteringdetector
beam away from perfect alignment. The same deflec-
tion is responsible for the shift of the central wave-
length of each spectral channel from its calibrated
value.
The grating dispersion equation suggests that the
change in the incident angle should result in a uni-
form shift of the central wavelength for all detector
array channels. Therefore, only a single parameter is
sufficient to describe the wavelength shift for each
detector array.
We were able to obtain in-situ calibrations for the
wavelength shift separately for two parts of spectra
corresponding to UV-VIS and NIR detector arrays.
This calibration was easy to perform using the wave-
length shift parameter implemented in WVASE soft-
waretoaccountformonochromatorcalibrationerrors
and to improve the portability of material optical
constant data obtained from different sources. This
parameter can be fitted and acts in the same way as
theabovementionedmisalignmentshiftsifonelimits
the spectral range to only a single detector array
channel.
WechosetousethermallygrownSiO2 onSi(211)for
this calibration. The SiO2 layer was thick enough
(~2500 Å) to produce at least one full interference
fringe in the two spectral ranges: from 1.5 eV to
2.488 eV and from 2.488 eV to 4.5 eV. This interfer-
ence pattern in the SE data allowed one to effectively
reduce the coupling between the layer thickness and
the wavelength shift parameters, making this cali-
bration more reliable. First the best-fit values for the
Fig. 4. The measured CdTe surface temperature dependence on
surface misalignment, obtained with and without wavelength and AOI
correction procedures. Solid triangles correspond to uncorrected SE
data.SolidsquarescorrespondtoappliedwavelengthandAOIcorrec-
tions.

ImprovementoftheAccuracyoftheIn-situEllipsometricMeasurementsofTemperature
and Alloy Composition for MBE Grown HgCdTe LWIR/MWIR Structures
641
rotation j was found to be 8.88∞. Every time the
detector is reinstalled on the reactor this parameter
mustberecalculatedforabettercalibrationaccuracy.
The parameter F in the AOI representation is of
special concern. Representing the undisturbed angle
ofincidence,thisparameterprimarilydeterminesthe
run-to-runreproducibilityofthecalibrationandmust
be optimized specifically for the in-situ configuration.
We propose two means of F parameter determina-
tion with sufficient accuracy. One is based on its
direct single time in-situ determination by fitting the
AOI for a perfectly aligned system using any reliable
material system such as a SiO2/Si sample. The
obtained F parameter can be fixed for all successive
growth runs together with the detector unit align-
ment. The alignment for any new sample in this case
should be limited only by the manipulator position
adjustments, leaving the detector and polarizer un-
touched. However, a better accuracy can be achieved
using an alternative approach, which involves the
alignment of the sample and detector before the
growth run and fitting for the AOI using the current
substrate. Nevertheless, great care must be taken in
the interpretation of the fit results for over-etched
CdTe or CdZnTe substrates.
Usually MBE-ready CdTe substrates have signifi-
cant scatter in the overlayer thickness (from 10 Å to
70 Å depending on sample preparation details). This
is especially true for composite CdTe/Si(211) and
CdZnTe/Si(211) substrates, the etching rate of which
also depends on material quality and surface mor-
phology. Inaccuracies in the model representation of
theCdTeandCZToxidesandlargevariationsintheir
thickness result in a large scatter in the fitted AOI
parameter. However, a precise determination of the
AOI still can be performed for such substrates at the
expense of reduced accuracy at the initial stages of
the growth run, if the analysis is done after the
Te desorption is completed. The successful Te re-
moval process leaves less than a few Å of overlayer,
which is well represented by surface roughness in a
two-layer optical model.
CALIBRATION OF THE ANGLE
OF INCIDENCE
It has already been mentioned that significantly
higheraccuracyincompositionandtemperaturecould
be achieved if the AOI were excluded as a free fit
parameter. On the other hand, keeping it fixed is not
justified for our setup. Under the conditions when the
polarizer and analyzer units are maintained at fixed
orientations, any deflections in the light beam enter-
ing the detector result only from changes in sample
surface orientation. We were able to relate the varia-
tions in the measured alignment parameters AlignX
andAlignYtovariationsintheAOIandtoparameter-
ize this relation using a calibration process quite
similartotheoneforwavelength, exceptthatonlythe
sample position would be changed, leaving all other
components fixed.
Because the control on the sample alignment is
very limited in the in-situ configuration, we have
performed our calibration on the ex-situ stage. The
arm length and the angle of incidence were chosen to
beclosetothoseforthein-situconfigurationtoensure
nearly the same degree of beam defocus and spatial
intensity distribution. The detector unit azimuth was
alsochosentobeclosetotheoneforreactormounting.
The same type of thick thermal SiO2/Si sample was
used for our calibrations because the SE data from
this specimen shows the highest sensitivity to the
AOI, due to the relatively high thickness and low
refractive index of the SiO2 film.
The alignment parameters were scanned in a
zigzag pattern by adjusting the sample tilt while the
angle of incidence was fitted for every sample surface
position. A wavelength correction also was applied to
the raw SE data before processing. The AOI was
parameterized as a function of the alignment param-
eters using of a formula similar to Eq. 1. As in the
wavelength calibration case, the coefficients A to F
were obtained using a regression analysis and are
presented in Table III. In both the ex-situ and in-situ
configurations, thedetectorunitwasmountedsothat
its X-axis is almost parallel to the system p-plane.
Therefore, as expected, the Y-sensitivity of the AOI is
significantly smaller than the X-sensitivity.
It is important to note here that the AOI obtained
inthecalibrationisexpressedasafunctionofthelocal
detectorcoordinates. Relocationofthedetectoronthe
MBE reactor generally involves changing of its azi-
muth.Thusinordertomaintainabettertransferabil-
ity of our ex-situ results to the in-situ configuration
we introduced an extra parameter, j, defining the
degree of rotation of the detector coordinate system
around the beam axis. The parameter, j, allows the
conversion of the actual alignment parameters of
real-time experiments to the x and y parameters used
in Eq. 1. It can be found as the difference between the
ex-situ and in-situ analyzer offset parameters As
produced by the ellipsometer regression calibration
procedure12 and accessible within the WVASE soft-
ware. For our experiment, the coordinate system
Theangleofincidencealignmentdependentcorrec-
tion procedure complements the wavelength shift
correction and is handled in real time by the above-
mentioned extension of the original WVASE soft-
ware. This extension communicates with the original
software through the TCPIP link and uses the elec-
tronic position indicator (EPI) command interface
implemented in WVASE.
SUMMARY AND CONCLUSIONS
We quantitatively studied the degree of error sup-
pression in the determined temperature and alloy
composition parameters by wavelength and AOI cor-
rection procedures.
Two sets of CdTe SE data were obtained during
two independent in-situ experiments in which after
Te desorption both samples were held at T = 200∞C
while their alignment was changing. The wavelength
and AOI corrections were applied only during one of

642
Daraselia, Garland, Johs, Nathan, and Sivananthan
these experiments. We maintained good temperature
controlduringbothexperiments.Thesampleposition
waschangedtoproducevariousalignmentparameter
values, and the calculated CdTe surface temperature
was obtained as a function of the alignment param-
eter AlignX, as shown in Fig. 4. We decided to express
the results as function of AlignX parameter only,
since the sensitivity to AlignY in our setup is signifi-
cantly lower and adding this parameter to the plot
will not provide any additional information.
Figure 4 clearly shows that our corrections signifi-
cantly reduce the alignment dependence on tempera-
ture from DT ª 28∞C to DT ª 3∞C. A linear regression
was applied to both data sets to obtain an exact
rejection coefficient of 9.5. At AlignX = 0 the tempera-
ture difference was found to be only 0.2∞C, a value
remarkably small considering the fact that the two
data sets were obtained in two different in-situ ex-
periments.
introduced into the model. Until then, the AOI cali-
bration is accurate only under the conditions that the
detector azimuths are the same for both ex-situ and
in-situ cases.
Finally, a significant improvement of the AOI fit
accuracy can be accomplished by developing an opti-
cal model capable of the precise determination of the
angle parameters from the SE data from “as-loaded”
CdTeandCdZnTesubstrates.Areliableellipsometric
analysis must produce good fits independent of the
overlayer thickness. We are currently working on a
three-layer optical model in order to improve the
modeling accuracy for intensively etched samples.
ACKNOWLEDGEMENTS
This work was funded by the Air Force Research
Laboratory under the contract number F29601-98-C-
0053 through a subcontract from EPIR Ltd., and is
coordinated by Dr. Y. Chen.
Even if we did not show explicitly the similar
comparison for MCT composition, the x dependence
on alignment parameters is expected to drop at the
same rate after the correction procedure is intro-
duced.
Eventhoughtheobtainedcalibrationssignificantly
reduce the uncertainties in the determined param-
eters, their quality still can be improved. Therefore, a
few sources of calibration errors are worth mention-
ing here.
First,thediscrepancybetweentheparametricform
(Eq.1)forAOIorwavelengthshiftandtheactualdata
points increase with increasing misalignments. This
is not surprising, because the ellipsometer quadrant
detector is expected to be highly nonlinear with re-
spect to slit image shifts. The introduction of the
higher order polynomial forms, such as cubic, is ex-
pected to reduce these discrepancies.
REFERENCES
1. P.W. Kruse, Semiconductors and Semimetals, ed. R.K.
Willardson and A.C. Beer, (New York: Academic Press,
1981).
2. W.E. Tennant, S. Cabelli, and K. Spariosu, J. Electron.
Mater. 28, 582 (1999).
3. A.I.D’Souza,L.C.Dawson,C.Staller,P.S.Wijewarnassuriya,
R.E. Dewames, W.V. Mclevige, J.M. Arias, D. Edwall, and G.
Hildebrandt, J. Electron. Mater. 29, 630 (2000).
4. T.J. de Lyon et al., J. Electron. Mater. 27, 550 (1999).
5. T. Skauli, T. Colin, R. Sjolie, and S. Lovold, J. Electron.
Mater. 29, 687 (2000).
6. J.D. Benson, A.B. Cornfeld, M. Martinka, K.M. Singley, Z.
Derzko, P.J. Shorten, J.H. Dinan, P.R. Boyd, F.C. Wolfgram,
B.Johs,P.He,andJ.A.Woollam,J.Electron.Mater.25,1406
(1996).
7. M.J. Bewan, W.M. Duncan, G.H. Westphal, and H.-D. Shih,
J. Electron. Mater. 25, 1371 (1996).
8. J.E. Jenson, J.A. Roth, P.D. Brewer, G.L. Olson, J.J. Dubray,
O.K. Wu, R.D. Rajavel, and T.J. de Lyon, J. Electron. Mater.
27, 494 (1998).
Another problem is related to the rotation of the
detector coordinate system in the AOI calibration
method. It was found that the sensitivities of the
quadrant detector in the X and Y directions are not
equal. Therefore, the interpretation of the detector
unit axial rotation as the rotation of the coordinate
system is not completely valid until ellipticity is
9. L.A. Almeida, N.K. Dhar, M. Martinka, and J.H. Dinan, J.
Electron. Mater. 29, 754 (2000).
10. J.A. Woollam Co., Inc., 645 M. Street, Suite 102, Lincoln, NE
68508.
11. F. Aqariden, W.M. Duncan, H.D. Shih, L.A. Almeida, and
M.J. Bevan, J. Electron. Mater. 28, 756 (1999).
12. B. Johs, Thin Solid Films 234, 395 (1993).