See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/229263562
Derivation of the complex refractive index of ITO and ITON films in the
infrared region of the spectrum by the analysis of optical measurements
Article in Thin Solid Films · September 2008
DOI: 10.1016/j.tsf.2008.04.014
CITATIONS READS
9 5,168
3 authors, including:
E. Aperathitis Mircea Modreanu
Foundation for Research and Technology - Hellas Tyndall National Institute
98 PUBLICATIONS 1,230 CITATIONS 191 PUBLICATIONS 2,449 CITATIONS
SEE PROFILE SEE PROFILE
All content following this page was uploaded by E. Aperathitis on 14 December 2017.
The user has requested enhancement of the downloaded file.
Available online at www.sciencedirect.com
Thin Solid Films 516 (2008) 8073 – 8076
www.elsevier.com/locate/tsf
Derivation of the complex refractive index of ITO and ITON films in the
infrared region of the spectrum by the analysis of optical measurements
A. Kondilis a,b,⁎, E. Aperathitis a , M. Modreanu c
a
Microelectronics Research Group, Institute of Electronic Structure & Laser, Foundation for Research and Technology-FORTH-Hellas, P. O. Box 1527,
Heraklion 71110, Crete, Greece
b
Department of Materials Science & Technology, University of Crete, P. O. Box 2208, Heraklion 71003, Crete, Greece
c
Photonics Group, Tyndall National Institute, Lee Maltings, Prospect Row, Cork, Ireland
Available online 10 April 2008
Abstract
We use the Newton–Raphson method to analyze the optical spectra of indium–tin-oxide and indium–tin–oxynitride films, by deriving the
complex refractive index and, thereby, the plasma wavelength and the relaxation time. Reflectance and transmittance as well as partial derivatives of
them, necessary for the application of the method, are introduced in analytical form reducing complexity and improving speed. The films we have
investigated were deposited on glass by sputtering at different RF power levels. After deposition, they were subjected to annealing at 600 °C and
were measured prior to and after that treatment. The results obtained are physically meaningful and lead to useful inferences about the quality of the
material in the two different types of the investigated films.
© 2008 Elsevier B.V. All rights reserved.
Keywords: Indium tin oxide; Indium tin oxynitride; Optical properties; Refractive index; Relaxation time; Plasma wavelength
1. Introduction work, not only transmittance and reflectance but also partial
derivatives of them needed in the calculations are introduced in
The problem of finding the complex refractive index in the analytical form thus providing the user with a relatively simple
infrared region from the analysis of optical spectra is an old one tool of improved speed. It should be mentioned that there have
and has been tackled in various ways [1–6]. Many of the pro- been previous attempts aiming at reducing complexity by the
posed algorithms treat the problem numerically, a fact that adoption of analytical expressions. In Ref. [4] the authors use
results in lengthy and complex calculations. Commercially analytical expressions for transmittance and reflectance. How-
available tools have rather a limited success as, according to the ever, their assertion of using analytical expressions for the partial
authors of ref. [1], one cannot state unambiguously that the pa- derivatives too is rather unclear [8]. Once the real and imaginary
rameters inferred from the fitting procedures are physically parts of the complex refractive index are derived, the plasma
meaningful. The aforementioned factors are most likely among wavelength and the relaxation time are obtained easily in the
the reasons for which literature is poor in providing the plasma framework of the Drude model. By the used algorithm, the
wavelength and the relaxation time through the analysis of optical properties of indium–tin-oxide (ITO) and indium–tin–
optical measurements. In the present work, we derive the com- oxynitride (ITON) films are investigated in terms of the
plex refractive index by an algorithm which makes use of the aforementioned Drude parameters.
Newton–Raphson method [7]. Contrary to already published
2. Derivation of the plasma wavelength and the relaxation
time
⁎ Corresponding author. Microelectronics Research Group, Institute of
Electronic Structure & Laser, Foundation for Research and Technology-
FORTH-Hellas, P. O. Box 1527, Heraklion 71110, Crete, Greece. Fax: +30 2810 For the cases we study, the wavelength of light is long com-
39 4106. pared to the thickness of fabricated films and, consequently, long
E-mail address: [email protected] (A. Kondilis). compared to the size of the different types of non uniformity [9]
0040-6090/$ - see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.tsf.2008.04.014
8074 A. Kondilis et al. / Thin Solid Films 516 (2008) 8073–8076
present in the films. Then, it is expected for light to see an overall using the well known Drude equation for the dielectric function.
medium and the followed treatment that assumes a perfect- The relevant expressions that ensue are:
homogeneous film to be a good approximation. vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
u
Each wavelength λ is associated with a specific pair value g2l g2 þ k 2 u g2l g2l g2 þ k 2
s¼ k and kp ¼ t 2 k: ð6Þ
(η, k), where the first element in the pair is the real part of the 4pcgk 4g2 k 2 þ g2l g2 þ k 2
complex refractive index and the second one is the imaginary
pffiffiffiffiffiffi
part. At each specific λ, in order to derive η and k the following gl ¼ el , where ε∞ is the high-frequency dielectric cons-
equations must be solved: tant and c the speed of light. As seen, in order to compute λp
and τ the only missing parameter is η∞. We have adopted the
Texp T ¼ 0 and Rexp R ¼ 0: ð1Þ
reasonable choice of putting in its place the value of η cal-
Texp and Rexp denote the experimental transmittance and the culated at that extremum which is the first met at the onset of the
experimental reflectance, respectively, whereas T and R the cor- band to band transitions (see graph (b) in Fig. 1).
responding theoretical quantities. In practice, the system of
equations (1) is considered to be satisfied if the following
conditions are met: 3. Results and discussion
jTexp T j jRexp Rj Graph (a) of Fig. 1 shows the experimental optical spectra of
Vd and Vd: ð2Þ an ITO film. Graph (b) shows the real and imaginary parts of the
Texp Rexp
δ denotes the accuracy by which the experimental data are
reproduced. Generally, at each λ, the initial (η, k)-input pair does
not meet the requirements of the relations (2) and a next input
(η′, k′) needs be introduced. That input is provided by the
Newton–Raphson method:
gV g Q dT
¼ þ ; ð3Þ
kV k Det dR
where,
dT ¼ Texp T ðg; k Þ; dR ¼ Rexp Rðg; k Þ ð4Þ
and
0 1
AR AT
B Ak Ak C
Q ¼ @ AR AT A ð5Þ
Ag Ag
In Eq. (3), the matrix Q is evaluated at (η, k). Det is the
determinant of that matrix. T and R as well as their partial deri-
vatives are introduced in analytical form. To find the partial
derivatives in analytical form, it is convenient to start with the
analytical expressions of R and T given in the appendix of ref.
[10]. One can recognize that they are written in terms of the
elements of the system transfer matrix one obtains if the subs-
trate is assumed semi-infinite. In fact, what is needed is to find
the partial derivatives of these matrix elements with respect to η
and k. The task is somewhat tedious, however, straightforward.
The procedure we follow is iterative. At the start wavelength,
we introduce an initial guess for η and k and Eq. (3) is repeatedly
Fig. 1. In graph (a), solid lines show the measured reflectance and transmittance
called until the conditions (2) are met. The successful parameter of an ITO film prepared at RF power 450 W and annealed at 600 °C. Symbols are
values derived serve as an initial input for the iterative procedure the theoretical reproduction. In graph (b), squares or solid lines illustrate the real
to start at the next wavelength. Again, Eq. (3) is recalled until the part η and the imaginary part k of the complex refractive index which was used
relations (2) are met and so on. In this way we finally get η(λ) for the reproduction. The specific way η and k were derived is reported in the
and k(λ) [11]. In our search for (η, k)-solutions, the expected text. The dotted lines serve as a guide to the eye. In the case of η, they are
extended down to wavelength zero in an attempt to give an outline of the
shape of η(λ) proves a useful guide. This point becomes clear in expected behavior if band to band transitions were totally absent. In order to
the next section. Once η(λ) and k(λ) have been obtained we can reproduce the experimental R- and T-curves in the short wavelength range below
easily derive the plasma wavelength λp and the relaxation time τ 600 nm, surface roughness was taken into account. Its rms value was σ = 10 nm.
A. Kondilis et al. / Thin Solid Films 516 (2008) 8073–8076 8075
complex refractive index derived from the analysis of these
spectra. In order to have a complete view we have chosen to
cover the whole wavelength range starting from the ultraviolet
and proceeding down to the infrared region. We quote a few
words on the way analysis was performed in the ultraviolet and
visible region of the spectrum: The basic interference equation
[12] provided η (solid squares), while k (thin solid line) was
found following the method given in ref. [13], this method,
however, appropriately modified so as to take surface roughness
into account as well. The inclusion of surface roughness in the
calculations proved indispensable in order to obtain the excellent
reproduction of the experimental curves in the short wavelength
part of the spectrum below 600 nm. In the infrared region, η and
k, shown by bold solid lines, were derived by the analysis pre-
sented in Section 2. They are seen to exhibit the overall expected Fig. 3. Plasma wavelength at different RF power levels. As-prepared cases are
behavior and solid symbols in graph (a) of Fig. 1 leave no doubt shown by open symbols while post-annealing cases by solid symbols. Stars
about the successful reproduction of the experimental curves in correspond to ITON while squares to ITO. The data of the as-prepared samples
this part of the spectrum too. As seen in graph (b), in the infrared grown at 150 and 250 W are missing because in these specific cases the plasma
excitation signal was out of the range of our measurements.
region we have two solution-branches. In between there is a gap
where the algorithm does not return any solution [14]. Regarding
η at the ends of that gap, the two branches show a bending in For 1 ≾ λ ≾ 1.5 μm, except the aforementioned drawbacks
opposite directions which is incompatible with the expected observed in the neighborhood of points where Det = 0, the algo-
trend of a smooth connection between them. The gap as well as rithm often returned solutions-artifacts. Such solutions exhibited
the false bending at its edges are among the characteristic physically unacceptable wavelength dependence and were
drawbacks of the developed algorithm met in the neighborhood rejected. Within the same wavelength range, another possibility
of points for which Det = 0. In Fig. 2, the derived plasma wave- consisted in solutions with a peculiar behavior possibly relating,
length and relaxation time are shown for both solution branches. as simulations showed, to spatial inhomogeneity inside the film.
Disregarding the false behavior at the edges of the solution gap, One must bear in mind that as the wavelength of light becomes
that behavior originating from a similarly false behavior in η, τ shorter, light ceases seen an overall medium and starts disti-
shows an increasing trend with the shortening of wavelength. nguishing more and more details in a possible inhomogeneous
Such dependence, which was systematically observed in all structure. On the other hand, for 1.5 ≾ λ ≾ 2.5 μm, many trials
cases considered here, has been reported in earlier investigations with different initial guesses have returned, in all cases examined
too and has been attributed to electron scattering by ionized here, one unique-well behaving solution in the sense of
impurities [2]. Contrary to τ, λp is rather wavelength indepen- exhibiting the expected features for the peculiarly shaped
dent and shows not to be appreciably affected by the false de- η(λ). These features are observed in graph (b) of Fig. 1 and
viation of η at the edges of the solution gap. consist in a steep descent followed by a shallow trough, the latter
ob served in cases the range of the measurements allowed it be
detected. All of the aforementioned reasons led us restrict pre-
sentation in the region between ∼1.5 and 2.5 μm.
It is likely that drawbacks relating to solution gaps or the
appearance of physically non realistic solutions could be par-
tially cured by adopting an approach different from that based on
the Newton–Raphson method. Such an approach would consist
in the minimization of appropriate error functions [5]. However,
as essentially no extra information is gained from the point of
view of the physics, the simpler Newton–Raphson method is
preferred at the present stage against a relatively more complex
minimization procedure.
In as much as an almost wavelength independence of λp was
a repeatedly visible trend in all of the analyzed samples, in Fig. 3
we show wavelength-mean values of λp at different RF power
levels. The data of the as-prepared samples at 150 and 250 W are
missing because in these specific cases the plasma excitation
Fig. 2. The upper curve corresponds to the relaxation time of the film whose signal was out of the range of the measurements. The depen-
spectra are shown in Fig. 1, while the lower curve to the plasma wavelength of
the same film. In the gap the algorithm does not return any solution. The dotted dence on RF power is poor and the only exception is the prior to
lines serve as a guide to the eye in an attempt to show the expected behavior annealing case of the ITO prepared at 550 W. In this case, λp
inside the solution gap. shows a sudden drop to 2 μm. One should note that at this
8076 A. Kondilis et al. / Thin Solid Films 516 (2008) 8073–8076
appreciably lower than that of the ITO. Calculations based on
Eqs. (7) and (8) associate the resulting ITON’s lower quality of
structure with an increased electron effective mass and an in-
creased impurity density as compared to those of the ITO. It is
likely that the thermally-induced release of nitrogen, observed
for T N 550 °C [17], could provide an explanation for the dam-
age of the ITON structure. However, even if so, the issue of why
this damage seems, as seen in Fig. 4, not to be effective for RF
power lower than 250 W remains open.
4. Conclusions
We have adopted the Newton–Raphson method to derive the
plasma wavelength and the relaxation time of ITO and ITON
films in the infrared region of the spectrum. In applying the
Fig. 4. Post annealing-long wavelength relaxation time at different RF power
method, the use of analytical expressions gives the advantage of
levels. Stars correspond to ITON while squares to ITO.
simplicity against the complexity of the otherwise numerical
work as well as the advantage of improved speed in obtaining
specific power, annealing has no effect on λp. However, in all results. In the wavelength range between 1.5 and 2.5 μm we
other cases it systematically results in a decreased λp, a behavior notice the following: For either ITO or ITON, the plasma wave-
which is consistent with the increase in carrier concentration length is practically constant. However, the relaxation time
observed by electrical measurements [15]. Prior to annealing, the exhibits an increasing trend with the decrease of wavelength, a
ITONs exhibit a λp that is higher as compared to that of the ITOs. behavior consistent with impurity-induced electron scattering.
In the post annealing stage this difference persists albeit con- Thermal annealing at 600 °C results in a shorter plasma wave-
siderably reduced. In fact, it would be justified to assert that, after length. The post annealing reduced value is 2 μm, approxi-
annealing, λp drops practically to 2 μm for all samples and RF mately, irrespective of film type and RF power level. On the
power levels. overall, after annealing ITON exhibits a relaxation time that is
Contrary to λp, τ exhibits a non negligible wavelength de- shorter as compared to that of the ITO. The lower structure
pendence particularly pronounced for λ b λp. To account for this quality implied by the shorter relaxation time is likely to be
behavior one has to go beyond the Drude model and consider the associated with thermally-induced release of nitrogen.
scattering of electrons by ionized impurities incorporating the
effect of screening in the calculations [2,16]. In this more rea- References
listic model, one observes that the expression that corresponds to
the Drude relaxation time τ depends on the wavelength, λ, the [1] T.J. Coutts, D.L. Young, X. Li, Mater. Res. Bull. (August 2000) 58.
electron effective mass, m⁎, the electron density, ne, the impurity [2] I. Hamberg, C.G. Granqvist, J. Appl. Phys. 60 (1986) R123.
[3] I.A. Rauf, Mater. Lett. 23 (1995) 73.
density, ni, and the charge of the ionized impurities, Ze, where e [4] O. Stenzel, V. Hopfe, P. Klobes, J. Phys. D: Appl. Phys. 24 (1991) 2088.
is the charge of the electron. If we restrict ourselves to the long [5] O. Stenzel, R. Petrich, J. Phys. D: Appl. Phys. 28 (1995) 978.
wavelength limit and, following [2], we put Z = 1 and ne = ni, we [6] A.B. Djurisic, T. Fritz, K. Leo, Opt. Commun. 166 (1999) 35.
get a wavelength independent expression for τ: [7] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical
recipes in Fortran, Cambridge University Press, Cambridge, 1992.
s ¼ sðm⁎ ; ne Þ: ð7Þ [8] In Ref. [4] the authors claim the use of analytical expressions for the partial
derivatives of reflectance and transmittance. Such expressions are not
provided in [4]. On the other hand, in a subsequent publication [5], the
Once τ and λp are known, Eq. (7) along with the well known
need of evaluation of the same partial derivatives is covered by numerical
expression for the plasma wavelength: calculations, thus leaving the aforementioned claim unsupported.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi [9] A. Kondilis, Phys. Rev., B 62 (2000) 10526.
2pc eo el m⁎ [10] A. Kondilis, in: S.G. Pandalai (Ed.), Recent Res. Devel. Physics, vol. 4,
kp ¼ ð8Þ Transworld Research Network, Kerala, India, 2003, p. 891.
e ne
[11] In deriving η and k, the thickness is a necessary input parameter. For the
can provide m⁎ and ne. Aiming at gaining information from cases considered, it was measured by a profilometer and its values ranged
Eqs. (7) and (8) we confine ourselves to λ N λp, the region where between 250 and 350 nm.
τ is practically wavelength independent. Taking into account [12] R. Swanepoel, J. Phys. E: Sci. Instrum. 16 (1983) 1214.
[13] A. Kondilis, E. Aperathitis, M. Modreanu, Thin Solid Films 515 (2007)
that the higher wavelength bound of our measurements is 8586.
2.5 μm, approximately, Fig. 3 indicates that the region in ques- [14] This is observed in other investigations too [2,5].
tion cannot be reached for the as-prepared cases save one. [15] E. Aperathitis, M. Modreanu, M. Bender, V. Cimalla, G. Ecke, M.
However, it can, for all of the post annealing cases whose τ- Androulidaki, N. Pelekanos, Thin Solid Films 450 (2004) 101.
values are shown in Fig. 4. As observed, when the RF power is [16] C.G. Granqvist, A. Hultaker, Thin Solid Films 411 (2002) 1.
[17] M. Himmerlich, et al., Surf. Sci. (2007), doi:10.1016/j.susc.2007.04.061.
higher than 250 W, the ITON exhibits a relaxation time that is
View publication stats
