See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/304348200

An Introduction to Modeling in Spectroscopic Ellipsometry, Focusing on

Models for Transparent Materials: the Cauchy and Sellmeier Models

Article · July 2016

CITATIONS READS
2 2,192

5 authors, including:

Cody Cushman Nicholas James Smith

Brigham Young University - Provo Main Campus Corning Incorporated
36 PUBLICATIONS   174 CITATIONS    42 PUBLICATIONS   354 CITATIONS   

SEE PROFILE SEE PROFILE

Massoud Kaykhaii Nikolas Podraza

University of Sistan and Baluchestan University of Toledo
195 PUBLICATIONS   1,161 CITATIONS    230 PUBLICATIONS   3,223 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Multi-instrument Characterization View project

Solid-state permanent digital data storage View project

All content following this page was uploaded by Matthew R Linford on 23 June 2016.

The user has requested enhancement of the downloaded file.

 By Cody V. Cushman1, Nicholas Smith2, Massoud Kaykhaii1, Nikolas J. Podraza3,
and Matthew R. Linford1; 1Department of Chemistry and Biochemistry, Brigham Young
University, Provo, UT, 2Science &Technology Division, Corning Incorporated, Corning,
NY, 3Department of Physics and Astronomy & WrightCenter for Photovoltaics Innovation
and Commercialization, University of Toledo, OH, Contributing Editors

An Introduction to Modeling
in Spectroscopic Ellipsometry,
Focusing on Models for Transparent Materials:
the Cauchy and Sellmeier Models
Introduction

I
n the Linford lab at BYU, we synthesize and characterize device fabrication—most frequently thin film solar cells. These
new materials with a special focus on surfaces and interfaces. areas are all interconnected, as each device structure consists
As a result, many of the characterization methods we em- of one or more thin film layers, the properties and structure of
ploy operate under moderate to ultrahigh vacuum. These include which impact device performance. We use this ability to probe
X-ray photoelectron spectroscopy (XPS),1 time-of-flight second- the optical response of layers in the device structure, determine
ary ion mass spectrometry (ToF-SIMS),2 low energy ion scatter- other material properties from the optical response, and connect
ing (LEIS),3 and scanning and transmission electron microscopy variations to device functionality.
(SEM and TEM). These tools are well established and proven, Here we discuss a powerful technique for characterizing thin
but they also have some drawbacks. Vacuum pumps and cham- films that provides rapid data acquisition, makes measurements
bers add expense and complexity to a system, waiting for cham- without contact, typically collects data under ambient condi-
bers to pump down to perform analyses can be time-consuming, tions, yields chemical and morphological information, and can
and beams from these instruments can damage some samples. probe deeply into many materials. Ultimately, it can yield film
We also frequently use atomic force microscopy (AFM) and thicknesses, reveal gradients in films and materials, give surface
contact angle goniometry, which also provide valuable informa- roughnesses, show whether materials are anisotropic, provide
tion. These techniques are not generally performed under vacu- band gaps, etc. This technique is spectroscopic ellipsometry
um, although AFM may be on occasion. However, AFM does (SE). In essence, SE probes the surface and near surface of a
not usually provide a lot of chemical information about a surface/ sample with light of a known polarization state, and detects how
material; we usually just use it to measure surface roughness and that polarization state changes upon interaction with the sample.
film thickness/step heights. Contact angle goniometry is similar- These measurements are usually taken in reflection mode and
ly very sensitive to just the outer few Ångstroms of a material, al- data are often collected at multiple angles of incidence, but ellip-
though it also responds to surface roughness and morphology. In sometry data can also be obtained in transmission mode. Modern
some cases, the drop of liquid used in contact angle goniometry spectroscopic ellipsometers can provide data over the infrared
(usually water) can perturb or react with a surface, and regardless (IR) to ultraviolet (UV) spectrum with high spectral resolution.
it must be removed after the measurement. In the Podraza lab at SE instruments are generally cheaper than the vacuum systems
the University of Toledo, we work primarily on optical charac- mentioned above (XPS, ToF-SIMS, LEIS, SEM, and TEM),
terization of materials, thin film deposition, and opto-electronic comparable in price to AFMs, and more expensive than contact

2 [email protected]  June 2016 • Vacuum Technology & Coating

angle goniometers. SE instruments are available from a number Figure 1, we change the orientation of the electric field in the
of companies. We obtained our instrument from the J.A. Wool- resulting linearly polarized light (compare Figures 2a and 2b).
lam Company, and the data shown herein were obtained with If we have a beam of light as in Figure 2b, we can describe it as
this instrument. We like their software, their instrument has been the sum of two orthogonal components that run along the x- and
reliable, and we’ve been grateful for their good customer service y- directions. Clearly, the two component waves here must be
over the years. Ellipsometers (spectroscopic, single wavelength, oscillating in phase, i.e., the maximum and minimum amplitudes
multiwavelength, and scanning/imaging) can also be purchased in the x- and y- directions are reached at the same time. However,
from Accurion, Film Sense, Gaertner, Horiba, neaspec, Opto- if two beams of equal amplitude along the x- and y-axes are 90
sense, Semilab, and Sentech. This list is no doubt incomplete. degrees out of phase, we obtain circularly polarized light (see
We don’t know as much as we probably should about these ellip- Figure 2c). In this scenario, the amplitude of the beam on the
someters – these companies are listed in alphabetical order here. x-axis is a maximum when the amplitude of the beam along the
y-axis is zero, and vice versa. The direction of the precession of
The Raw SE Data this light (the sum of these two beams that make a circle) de-
Most often, the raw SE data for isotropic samples is present- pends on the phase relationship between them. The more general
ed as two quantities, Ψ and ∆, or other related representations. case, elliptically polarized light, results when we add vertically
To understand their meaning, a basic understanding of polarized and horizontally polarized beams of light with arbitrary phase
light is necessary. Light waves consist of oscillating electric and and amplitude relationships (see Figure 2d). For these cases, the
magnetic fields. For interactions with non-magnetic matter, the circularly or elliptically polarized light can be described as either
electric field is the most important component – we can generally right- or left-handed, depending on the adopted convention.
ignore the magnetic part. Most light we deal with in our everyday We need to define one more bit of terminology here with re-
lives is not polarized, meaning that the electric fields of the pho- gard to light reflecting from a surface. In Figure 3, we see a plane
tons are randomly oriented. However, when the electric fields in defined by two grey triangles. This is the plane of incidence,
a beam of light are aligned/specifically oriented relative to each which contains the surface normal and the directions of the light
other, the light is polarized. Linearly polarized light is the easiest impinging on and reflecting from the surface. If the electric field
to visualize. Figure 1 shows monochromatic, unpolarized light of our light is in this plane, it is called p-polarized light, where ‘p’
passing through a device called a polarizer, which yields linearly stands for ‘parallel’. However, if the electric field is perpendicu-
polarized light – all the electric fields are parallel to each other. lar to the plane of incidence, it is referred to as s-polarized light.
If we could see the electric field of the beam of light coming We get the ‘s’ here from the German word for perpendicular,
towards us we would see something along the lines of what is which is ‘senkrecht’. In ellipsometry, we generally probe a sur-
depicted in Figure 2a. Obviously if we rotate the polarizer in face with a beam of light of known polarization and examine its

Figure 1. Representation of the polarization of unpolarized light by a wire grid polarizer. Light with an electric field perpendicular to the wires
interacts less with the electrons in the wires and is transmitted.

Vacuum Technology & Coating • June 2016    www.vtcmag.com  3

 polarization state after reflection. Very often, the reflected beam
of light is elliptically polarized (see Figure 3), which gives the
technique its name.
We now define Ψ and ∆, which are important in ellipsometry
– the results of ellipsometric measurements are often provided
in terms of these parameters. In ellipsometry, we are interested
in the change in the state of polarization of light, which can be
written in terms of the complex amplitude reflectance ratio, ρ,
given by Equation 1,

, (1)

where rp and rs are the amplitude reflection coefficients for p- and

s- polarized light, respectively, E is the amplitude of the electric
field, the subscripts r and i denote the reflected and incident light,
respectively, and the subscripts p and s indicate the polarization
state of the light. For the case where the incident p and s com-
ponents of the light are of equal amplitude, Ei,p = Ei,s, Equation 1
reduces to

. (2)

Okay, great. Now we need to separate the terms containing Ψ
and ∆ to see what they mean. If we write rp and rs in their polar
coordinate representations, we obtain
rp = | rp | e iδr,p, and (3)
Figure 2. Representations of (a) linearly polarized light with the elec-
tric field parallel to the y-axis, (b) linearly polarized light at an angle rs = | rs | e iδr,s, (4)
not aligned to an axis, (c) circularly polarized light, and (d) elliptically
polarized light. where the δ values are the phases of the waves. It follows that

Figure 3. A depiction of linearly polarized light impinging upon a surface (from the left), and becoming elliptically polarized upon reflection. The
grey triangles on either side of the figure define the plane of incidence of the light.

4 [email protected]  June 2016 • Vacuum Technology & Coating

 for experimental, and σ represents the standard deviation of a
. (5) measurement of Ψ or ∆ at a certain wavelength. The unweighted
form does not consider error in the denominator.
While the weighted MSE gives a measure of how well the
Therefore, tanΨ= |rp ||rs |, Ψ= tan-1 |rp ||rs|, and ∆ modeled data matches the experimental data, a low MSE, in and
of itself, does not guarantee that a model is good. For example,
δr,p–δr,s. Thus the quantity ∆ tells us about the phase relationship low MSEs can usually be generated by inserting a large number
between the reflected p and s waves, and the angle Ψ tells about of fit parameters into a model, where the resulting model may
the amplitude ratio between the two components. Measurements make no sense physically or produce nonsensical results. Deter-
of these quantities with high precision will depend on the angle mining whether or not a model is appropriate is a topic best left
at which we collect data and the optical and structural properties for a future article, but a lot of pitfalls can be avoided by follow-
of the materials we are measuring. A more complete discussion ing some simple guidelines and critical thinking. First, base your
of this topic can be found in Prof. Hiroyuki Fujiwara’s outstand- model on your knowledge of your sample. Ideally, you will start
ing book on ellipsometry.4 out knowing the nature of your substrate, what materials were
deposited on it, the order in which they were deposited, and their
From Raw Data to a Model nominal thicknesses. Information obtained from SEM, TEM,
So, what’s so great about Ψ and ∆ that we want to measure AFM, SIMS, and other optical techniques can help you set the
them? The truth is that quite often we are not interested in them starting guesses in your model, or at least serve as a reality check
directly. More often than not we want to know about material to see if your model is reasonable. The information obtainable
properties, which might include film thickness, optical constants, by ellipsometry can and should be confirmed by other methods
surface roughness, gradients in films, the bandgap of the mate- when possible. If you have a complex sample and you start out
rial, etc. However, as an aside, we should point out that chang- knowing nothing about it, your analysis will probably be much
es in Ψ and ∆ can be useful by themselves as indicators of a more challenging. The second thing to remember is to keep your
change in a material.5-6 In any case, the work up of ellipsometric model as simple as possible. If there are adjustable fit param-
data usually consists of developing a model that is supposed to eters in your model that are contributing little to the goodness
mimic one’s experimental system. Often the data acquisition can of your fit, they should probably be fixed at reasonable values
be accomplished relatively quickly compared to the user time or eliminated. Remember here that you have a limited quantity
spent developing and evaluating many models. Using the laws of ‘knowns’ from which to extract your ‘unknowns’. The fewer
of physics/optics, Ψ and ∆ are then predicted for this model. For fit parameters required to do the job, the higher your likelihood
example, Fresnel’s equations describe the reflection and trans- of success in obtaining a physically reasonable model. One way
mission of light at interfaces for the p- and s- components of of testing for correlation between parameters in ellipsometry is
light (for a detailed treatment see Fujiwara, Chapter 2).4 These through uniqueness tests, i.e., good fits don’t have unexpectedly
equations depend on the optical constants of the materials: n, the correlated parameters.7 When we say that parameters are cor-
refractive index, and k, the extinction coefficient, and the angle related we mean that we can change a given parameter and get
at which the light interacts with the material. Chemists prefer the the same low error for the fit because the remaining parameters
n/k notation, while physicists usually like to express optical con- can vary in a way that compensates for the change in the first
stants in terms of the complex dielectric function, i.e., in terms one. When this takes place the parameter in question has little or
of ε1 and ε2. Both are quite interchangeable, as are the spectral no statistical meaning. We recently took the concept of unique-
units applied, depending on fields and sub-fields (photon energy, ness plots in ellipsometry and applied it to peak fitting in X-ray
wavelength, frequency). The modeled values of Ψ and ∆ are then photoelectron spectroscopy.8 We think that this type of tool may
compared to the measured ones. In general, there will be one or be useful to the XPS community to quickly identify bad peak fits
more adjustable parameters in the model, such as a film thick- that come from using too many unconstrained parameters.
ness. These parameters are usually varied to minimize the dif-
ference between the modeled and measured values of Ψ and ∆. The Cauchy Model
The resulting optimal value of the parameter(s) in question, e.g., There are a variety of approaches for modeling the optical
thickness, is often taken as the value of that parameter for the constants of a film or material. The simplest case, and the one
film. This is how we measure a film thickness by ellipsometry. we will focus on in this article, is when the material is trans-
The goodness of fit between the modeled and experimental data parent. For transparent materials, the extinction coefficient, k, is
is often calculated using a weighted or unweighted mean squared zero over the measured wavelength range, and we only need to
error (MSE), with the weighted form given by Equation 6, fit the refractive index, n, over this range. This greatly simplifies
the modeling. Of course we must remember that materials have
dispersion, i.e., n varies with wavelength, which we signify by
, (6) writing n(λ). Sometimes n(λ) is fit with a curve-fitting tool, like
a B-spline, a point-by-point fit, or other numerical inversion of
where 2N is the number of Ψ and ∆ values collected, M is the experimental data. Depending on the goals of the analysis and/or
number of fit parameters, ‘mod’ stands for modeled, ‘exp’ stands the nature of the samples, these can work well. The other option

Vacuum Technology & Coating • June 2016    www.vtcmag.com  5

is to use a parametric model, where a few parameters are fit in an to understand and to implement – it only requires 2 – 3 adjustable
equation that describes the optical constants over a broad spectral parameters and it is mathematically simple.
range. The advantages of these models over curve fitting tools
are that (i) they typically require fewer parameters, i.e. they are The Sellmeier Model
simpler, (ii) they eliminate noise from the final optical constants
Over an expanded wavelength range, a more physically rea-
of the material (the optical constants of a material can’t change
sonable model is required to accurately describe the optical con-
discontinuously from point to point; it is unphysical), and (iii)
stants of transparent materials, and this is where the Sellmeier
some parametric models have a physical basis that the other fit-
dispersion relationship comes in. One form of the Sellmeier
ting tools lack.
equation is given as
The first parametric model we will discuss is the Cauchy dis-

persion relationship, . (8)


. (7)
Here, n is the refractive index of the material at a given value of
The Cauchy relationship is a polynomial equation and an em- λ, AUV is the amplitude of the UV pole, AIR is the amplitude of the
pirical model with no physical basis, i.e., unlike a good NFL IR pole, λuv and λIR are the positions of the UV and IR poles, re-
lineman, it lacks physicality. In reality, it’s just a function that spectively, and ε ∞ is the low-frequency offset. Here we are mod-
happens to match the shape of n(λ) in transparent materials over eling the real part of the dielectric function through the placement
certain wavelength ranges. Sometimes you will see the Cauchy of two infinitely-narrow absorbing features (the poles) outside
model expressed as a power series with different powers of λ in (above and below) the spectral range. This gives us the shape
the denominator, but Equation 7 is its most common embodi- we are looking for. For example, while the Cauchy dispersion
ment. You should know that, because it does not have theoreti- relationship looks kind of like a ski – flat at longer wavelengths
cal underpinnings, some researchers who shall remain nameless (lower energies) and turning up at shorter ones (higher energies),
consider it to be nothing short of an abomination (sorry for the the Sellmeier dispersion relationship also has a downturn at lon-
strong language – CVC must have written this part of the docu- ger wavelengths. The IR, UV, and constant additive parts can be
ment). That is, we are discussing the Cauchy model here not be- used together or singly—depending on the material and wave-
cause we recommend it, but because it shows up fairly frequently lengths studied. To reduce the number of fit parameters, the λIR
in the literature so you should understand it when you see it. To term is usually fixed at a very long wavelength.
better understand Equation 7, we will examine it in its limits. As
λ approaches infinity, i.e., the wavelength gets really long, the B The Cauchy and Sellmeier Models in Action
and C terms divided by powers of wavelength in Equation 7 go To demonstrate the Cauchy and Sellmeier models, we mea-
to zero, and the refractive index approaches some constant val- sured a thick film of thermal oxide (SiO2) on a silicon wafer by
ue, i.e., ‘A’. Accordingly, as we move from the visible towards SE at a 75° angle of incidence. Thermal oxide is transparent over
the IR, the Cauchy model dictates that the refractive index of a a wide wavelength range, so it would be appropriate to consider
material is modeled as a horizontal line. On the flip side, as λ ap- the models discussed herein. Figure 4 shows the resulting Ψ and
proaches zero, the B and C terms approach infinity, so the refrac- ∆ data. The oscillations here are consistent with a thick film – it
tive index becomes infinitely large. Real optical constants of ma- is useful to remember that the thicker the film, the higher the fre-
terials do not go to either limit or do not do so that simply. As the quency of these oscillations. These types of guidelines are useful
wavelength that probes a transparent material becomes shorter,
one will eventually encounter absorptions in the material. This
will certainly cause n to rise through the Kramers-Kronig (K-K)
integral, but not to infinity. For similar reasons, n will dip down
as one goes to longer and longer wavelengths. At these longer
wavelengths, phonon modes may also be encountered, which
also increases absorption. Finally, we note that in modeling some
materials, the first two terms in the Cauchy relationship suffice,
so ‘C’ can be set to zero/ignored. And when there is very little
dispersion in a material one might even opt to model it as a con-
stant over a limited wavelength range – one could think of this as
setting ‘B’ and ‘C’ to zero in the Cauchy equation. Fujiwara did
this recently in a paper in which he modeled a transparent SnO2
layer using a constant n value of 1.86.9 He reported that he did
this to simplify his SE analysis – you can’t get much simpler. So
given its lack of physicality, why is the Cauchy model used so
often? There are probably two basic reasons for it. First, it tends Figure 4. Raw ellipsometric data in degrees for a ca. 500 nm film of
to work reasonably well for many transparent materials across thermal oxide on a silicon wafer. Data taken at a 75° degree angle of
the visible range and sometimes beyond, and second it is simple incidence.

6 [email protected]  June 2016 • Vacuum Technology & Coating

 same data were fitted to a Sellmeier model. This model gave a
film thickness of 489.73 nm and an MSE of 1.531, a modest
improvement over the three-parameter Cauchy. Figure 7 shows
the acquired and modeled data from this approach. Interesting-
ly, the differences between Figures 6 and 7 are not large – both
parametric models give similar fits, although one can see that the
Sellmeier does indeed appear to be more effective. Finally, Fig-
ure 8 shows the optical constants of thermal oxide obtained with
the three-parameter Cauchy and Sellmeier models plotted with

Figure 5. Schematic representation of the model used to describe our

thick layer of thermal oxide on silicon.
– it is powerful to be able to look at raw data and get a sense for
the type of material that generated it. A schematic representation
of the model used to analyze this sample is provided in Figure
5. While it is common, and often essential, to include roughness
layers on sample surfaces and intermix layers between adjacent
materials, we have opted to use a simple two-layer model for
the sake of this tutorial. Accordingly, this model consists of (i)
an upper layer of thermal oxide with unknown thickness and
optical constants, and (ii) a silicon substrate for which we have
pre-derived optical constants that have been published and are
conveniently available in the instrument software. The silicon is
Figure 6. Raw ellipsometric data from a ca. 500 nm film of SiO2 on
opaque to a wide range of the wavelengths that probed it (probed Si plotted with data generated from a three-parameter Cauchy model
from ca. 190 – 1700 nm, and opaque from ca. 190 – 1100 nm) fit. The raw data is shown with symbols, and the model fit is shown
and the back-side of the wafer is roughened to scatter any trans- with lines.
mitted light, so its thickness is effectively infinite as far as the
model is concerned. In practice we will often set the thicknesses
of such substrates to 1 mm in a model. We first tried fitting the
data with a Cauchy model. We tried two different approaches –
one with three parameters (A, B, and C) and the other with two
(A and B). The two-parameter approach gave a film thickness of
489.2 and an MSE of 2.343, while the three-parameter approach
yielded a film thickness of 489.5 nm and an MSE of 1.659. We
see here that the Cauchy model with two parameters appears to
be sufficient for measuring the thickness of the thermal oxide, al-
though adding the third parameter does significantly improve the
fit to the raw data. As a guideline, if an additional fit parameter
does not improve your MSE by at least 10%, you might consider
leaving it out of the model. Based on the MSE values of these
fits, we selected the three-parameter model. The results of these
models are shown in Table 1. Figure 6 shows the acquired and Figure 7. Raw ellipsometric data from a ca. 500 nm film of SiO2 on Si
modeled Ψ and ∆ data for the three-parameter model. Overall plotted with data generated from a Sellmeier model fit. The raw data is
the agreement between them is good. In a similar fashion, the shown with symbols, and the model fit is shown with lines.

Table 1. Fit results for the 2- and 3-parameter Cauchy models and the Sellmeier model.

2-Parameter Cauchy 3-Parameter Cauchy Sellmeier

Parameter Value Error Parameter Value Error Parameter Value Error
Thickness (nm) 489.6 0.3 Thickness (nm) 489.5 0.2 Thickness (nm) 489.7 0.5
A 1.452 4 × 10-4 A 1.455 4 × 10-4 AUV 198.3 0.5
B 4.07 × 10-3 1.5 × 10-5 B 3.38 × 10-3 3 × 10-5 EUV (eV) 13.32 0.01
C 2.8 × 10 1 × 10-6 AIR 6.7 × 10-3 8 × 10-4
-5

MSE 2.336 MSE 1.659 MSE 1.531

Vacuum Technology & Coating • June 2016    www.vtcmag.com  7

 Figure 9. Raw ellipsometric data from a ca. 500 nm film of SiO2 on Si
Figure 8. Optical constants of thermal oxide obtained with our three-pa- plotted with data generated by using two different Sellmeier models to
rameter Cauchy model (top) and a Sellmeier model (middle) compared describe the optical constants of the absorbing substrate (Si) and film
to the results obtained by Herzinger and coworkers (bottom).10 (SiO2). As expected, extremely poor agreement is obtained.

the optical constants of thermal oxide obtained by Herzinger et the spectral range, more complex models are required. Often,
al.10 Of course Herzinger and coworkers used a slightly differ- absorbing features are fit with Gaussian or Lorentzian based
ent material to obtain their optical constants so we would not functions. In addition, special models have been developed for
expect exact agreement between the results in Figure 8. Never- specific applications, like the Tauc-Lorentz and Cody-Lorentz
theless, it is of interest to note that, as expected, the Cauchy and models for absorptions in amorphous semiconductors (these
Sellmeier models are in good agreement at shorter wavelengths models yield a band gap), or the Drude model for free-carrier
(below about 1000 nm), but diverge above this point – we recall absorption in conducting materials. Models also exist for dealing
again that the Cauchy has no way of further decreasing at its with mixtures of materials, surface and interfacial roughness, and
long wavelength end. In addition, it is gratifying to see that the gradients, e.g., the Bruggeman effective medium approximation.
Sellmeier model more closely approximates the Herzinger opti- We will probably write more about these in a later article.
cal constants. As an aside, when we add roughness and interface
layers to the model in Figure 5, we match Herzinger’s optical Conclusions
constants very closely with a Sellmeier model. Ellipsometry is a powerful surface/material analytical tech-
nique. By modeling Ψ and ∆, a tremendous amount of useful
Using the Wrong Model – Ouch!
information can be obtained about materials, including their
Now we’re going to show you something that you should not optical constants, layer thicknesses, and surface and interfacial
do in SE data modeling. In particular, we are about to make the roughnesses. Ellipsometers are rather easy to use and only mod-
significant mistake of modeling both the substrate (silicon) and erately expensive. Transparent materials are among the sim-
the film (SiO2) of our sample using two Sellmeier models. This plest to model because only their refractive indices must be fit
is a really bad idea. Don’t try this at home. Of course, this ap- to describe their interaction with light. We have discussed the
proach is entirely inappropriate because the Cauchy and Sellmei- basics of modeling ellipsometry data from transparent materials
er models are designed for transparent materials, and the silicon and shown how the Cauchy and Sellmeier models perform. The
is not transparent for most of the range of wavelengths here. That Cauchy model is simple, but empirical. Nevertheless, it can work
is, a given parametric model will have conditions under which quite well over limited wavelength ranges. The Sellmeier model
it should be applied. One should be aware of these constraints. has a solid physical basis and is more suitable for wider wave-
Nevertheless, throwing caution to the wind, we tried it. The re- length ranges. Applying the wrong model in SE data fitting can
sults were awful and are shown in Figure 9. The MSE for this lead to disastrous results.
fit was 33.953. We also tried this unreasonable fitting approach
with two Cauchys and, as expected, again got very bad results. References
The lesson here is clear. You can make a mess out of your SE
1. Gupta, V.; Ganegoda, H.; Engelhard, M. H.; Terry, J.; Linford, M.
modeling by using the wrong model. R., Assigning Oxidation States to Organic Compounds via Predic-
tions from X-ray Photoelectron Spectroscopy: A Discussion of Ap-
Note On More Complex Materials proaches and Recommended Improvements. Journal of Chemical
The parametric models we have discussed thus far are great Education 2014, 91 (2), 232-238.
for modeling the optical constants and thicknesses of transpar- 2. Jensen, D. S.; Kanyal, S. S.; Madaan, N.; Hancock, J. M.; Dad-
ent materials, but in cases where there are absorbing features in son, A. E.; Vail, M. A.; Vanfleet, R.; Shutthanandan, V.; Zhu, Z.;

8 [email protected]  June 2016 • Vacuum Technology & Coating

 Engelhard, M. H.; Linford, M. R., Multi-instrument characteriza-
tion of the surfaces and materials in microfabricated, carbon nano-
tube-templated thin layer chromatography plates. An analogy to
‘The Blind Men and the Elephant’. Surface and Interface Analysis
2013, 45 (8), 1273-1282.
3. Cushman, C. V.; Brüner, P.; Zakel, J.; Major, G. H.; Lunt, B. M.;
Smith, N. J.; Grehl, T.; Linford, M. R., Low energy ion scattering
(LEIS). A practical introduction to its theory, instrumentation, and
applications. Analytical Methods 2016, 8 (17), 3419-3439.
4. Fujiwara, H., Principles of Spectroscopic Ellipsometry. John Wiley
& Sons, Ltd: Chichester, England, 2007.
5. Bagley, J. D.; Dennis Tolley, H.; Linford, M. R., Reevaluating the
conventional approach for analyzing spectroscopic ellipsometry
psi/delta versus time data. Additional statistical rigor may often be
appropriate. Surface and Interface Analysis 2016, 48, 186-195.
6. Diwan, A.; Linford, M. R., The ‘No Model’ Model (Just Moni-
toring Psi and Delta). Vacuum Technology & Coating November,
2014.
7. Hilfiker, J.; Synowicki, R.; Tompkins, H. In Spectroscopic ellip-
sometry methods for thin absorbing coatings, Society of Vacuum
Coaters. Proceedings 51st Annual Technical Conference, 2008; pp
511-516.
8. Singh, B.; Diwan, A.; Jain, V.; Herrera-Gomez, A.; Terry, J.; Lin-
ford, M. R., Uniqueness Plots. A Simple Graphical Tool for Identi-
fying Poor Peak Fits in X-ray Photoelectron Spectroscopy. Accept-
ed, Applied Surface Science 2016.
9. Akagawa, M.; Fujiwara, H., Optical characterization of textured
SnO2:F layers using spectroscopic ellipsometry. Journal of Applied
Physics 2012, 112 (8), 083507.
10. Herzinger, C.; Johs, B.; McGahan, W.; Woollam, J.; Paulson, W.,
Ellipsometric determination of optical constants for silicon and
thermally grown silicon dioxide via a multi-sample, multi-wave-
length, multi-angle investigation. Journal of Applied Physics 1998,
83 (6), 3323-3336.

Vacuum Technology & Coating • June 2016    www.vtcmag.com  9

View publication stats