ROMANIAN JOURNAL OF INFORMATION

SCIENCE AND TECHNOLOGY

Volume 10, Number 3, 2007, 291–300

A Power Spectral Density Study of Thin

Films Morphology Based on AFM Profiling
R. GAVRILA1 , A. DINESCU1 , D.MARDARE2
1
R&D Institute for Microtechnology – Bucharest, Romania
E-mail: [email protected]
2
Faculty of Physics, Al. I. Cuza University, Iaşi, Romania
E-mail: [email protected]

Abstract. Atomic Force Microscopy (AFM) is extensively being used for

characterizing thin film surfaces. In addition to direct imaging, AFM profile data
enable to derive the Power Spectral Density (PSD) of the surface roughness and
thus provide useful information on characteristic features which compose the
microstructure of the films. In the present work PSD spectra computed from
AFM data were used for studying the morphology of three different titanium
oxide thin films obtained by dc magnetron sputtering onto glass. The derived
PSD curves were fitted with an appropriate analytic function and characteristic
parameters were deduced and discussed in order to compare film morphologies
in conjunction with the deposition conditions.

1. Introduction

Currently there is a great interest in designing and engineering thin films with
morphologies tailored to specific requirements for various application fields. The
connection between surface morphology of the films and functionality is especially
important for optical applications, as surface microstructure generates scattering and
stray light in optical components.
The most popular parameter characterizing the morphology of surfaces is the rms
roughness σ, which represents the root mean square height of a surface around its
mean value. However this statistical description, though simple and reliable, makes
no distinction between peaks and valleys and does not account for the lateral dis-
tribution of surface features. A more complete description is provided by the power
292 R. Gavrila et al.

spectral density (PSD) of the surface topography, which performs a decomposition

of the surface profile into its spatial wavelengths and allows comparison of roughness
measurements over different spatial frequency ranges. For optical surfaces the PSD
function is especially important and represents a very popular way to describe the
surface quality as it is connected with the amount and angular repartition of scattered
light. Thereby scattering theories predict that in the smooth-surface limit the angular
distribution of light scattered from a surface fits the power spectral density (PSD) of
the surface topography [1]:

IS(θ, φ) = C(θ, φ) · PSD(fx , fy ) (1)

where IS(θ, φ)is the scattered flux per solid angle unit in the (θ, φ) direction, PSD
(fx , fy ) is the 2-D PSD of the surface topography, fx , fy are the the spatial frequencies,
related to the incident (θi ) and scattering (θ, φ) angles, C(θ, φ) is an optical factor
which depends on the refraction indexes of the incident medium (generally air) and
surface material and on the illuminating and observation conditions (wavelength,
geometry, polarization).
This study focuses on titanium oxide, a material currently used as thin film coat-
ing in optical applications due to its notable dielectric and optical properties. Ow-
ing to their good durability, stability and high refractive index, TiO2 thin films are
used as antireflection coatings [2], optical filters (in multilayer coatings) [3], optical
wave-guides [4] and recently, due to its semiconducting properties, as photocatalytic
coatings in anti-fogging mirrors [5].

2. Definitions

Different methods for computing PSD from profiling data are found in literature,
most of them differing by normalization procedures. This fact encumbers the reference
and comparison of results and urges for the specification of the PSD expressions used
in each case.
In this work we have adopted the following definition for the 2-D PSD of a surface
described by its topography z (x, y):

 2

 1/2L
Z 1/2L
Z 

1
S2 (fx , fy ) = lim 2 dx dy · z (x, y) · exp [2πi (fx · x + fy · y)] (2)
L→∞ L  

−1/2L −1/2L

where L is the scan length, assumed to be equal in x and y directions, fx , fy are the
spatial frequencies
√ (inverse of spatial wavelengths) for x and y directions, respectively,
and i = −1.
Strictly speaking, the integral in expression (2) is not exactly the Fourier transform
of the height function z(x, y), as the existence of the limit depends on the properties
of the function and moreover the limit L → ∞ is physically unattainable. In practice
A Power Spectral Density Study of Thin Films Morphology 293

a finite N number of values are measured, assumed to be equidistantly spaced in x, y

directions at distances ∆x, ∆y = L/N . The 2-D PSD is then computed as:
" N N
#2
2 1 XX
S2 (fx , fy ) ≈ L z(xm , yn ) exp (2πi · xm · fx ) · exp (2πi · yn · fy ) (3)
N 2 n=0 m=0

L L
where xm = m · , yn = n · , and the spatial frequencies fx , fy take the discrete
N N
range of values:
1 2 N
fx , fy = , , ... .
L L 2L
The analogous PSD definition for the 1-D case, corresponding to the topography
of a profile described by its height z(x) is:
" N
#2
1 X
S1 (fx ) ≈ L zk · exp (2πi · xk · fx ) , (4)
N
k=1

L 1 N
with xk = k · and fx = j · , j = 1, 2, ..., .
N L 2

3. Experiment
Three titanium oxide films were deposited onto glass substrates by d.c. magnetron-
sputtering method, using water vapors as reactive gas, with the same deposition rate,
r = 0.02 nm/s. After deposition all samples were submitted to a thermal treatment
within a certain temperature range ∆T . The experimental arrangement and deposi-
tion technique are described in [6].
The samples, denoted L31, L48 and L27, differ by substrate temperature during
deposition and film thickness, as shown in Table 1.

Table 1
sample L31 L48 L27
substrate temperature [◦ C] 310 480 270
film thickness [nm] 450 260 230

The investigations described in [6] showed that the deposited titanium oxide films
have a polycrystalline and generally multiphasic (anatase and/or rutile) structure.
The samples L27 and L31 have a mixed crystalline structure with an equal proportion
of anatase and rutile, while pure rutile structure has been observed for the sample L48.
The morphology of the films, together with that of the uncoated glass substrate
were measured with a home-built Atomic Force Microscope (AFM) operated in con-
tact mode. For each sample 1 µm × 1 µm and 10 µm × 10 µm scans were performed,
using silicon nitride cantilevers with sharpened pyramidal tips, nominal tip radius of
20 nm and nominal apex angle of 18◦ . The tip shape was checked after each measure-
ment with the aid of reference samples and SEM imaging. These checks have shown
294 R. Gavrila et al.

that tip radius increased during the 10 µm × 10 µm measurements. Actually it was

almost impossible to maintain the initial tip shape intact over a full 10 µm scan, be-
cause of the strong tendency of the tip to wear, especially in the scanning direction.
In Fig. 1 is purposely shown the SEM image of an initially sharp tip, severely worn
in the direction parallel to the scan after a 10 µm × 10 µm measurement.

Fig. 1. 157 000× magnification SEM image of the apex region

of a tip, severely damaged in the direction parallel to the scan;
the tip sharpness was quite unaffected in the direction normal to the scan.

Due to that phenomenon each sample was measured several times with different
new tips and the image acquired with the less affected tip was retained for subsequent
processing.

4. Data processing

An important step for obtaining consistent results is the correction of the instru-
mental AFM artifacts generally due to the piezo scanner, artifacts which could mask
the real surface morphology. For that purpose, all raw data sets were subjected to
a similar procedure meant to remove plane artifacts (tilt and bow) and occasional
noise or contamination that can cause the scan lines to be leveled differently. This
procedure consisted in a global flattening correction by a three order polynomial fit
and a line by line correction by least mean square and histogram alignment methods.
Image and data processing were done with SPIPTM software.
Although with great value in optical applications, the 2-D PSD of the surface
morphology is difficult to work with and unpractical for comparing purposes. In
order to condense the acquired information and facilitate the comparison between the
samples, a 1-D PSD could be extracted from the 2-D function. It is this approach
that we have taken in the present work.
A Power Spectral Density Study of Thin Films Morphology 295

Scans made by rotating the samples in different directions were practically iden-
tical, so it could be assumed that the films are isotropic; the natural procedure for
1-D PSD extraction would be in this case the transition to polar coordinates in fre-
quency space and angular averaging. But in AFM measurements the data acquired
in y (slow scan) direction are usually distorted by instrumental drift and moreover
in this case by the asymmetrical tip geometry. This slight apparent deviation from
perfect isotropy could be noticed in Fig. 2, an image of the 2-D PSD for the sample
L31 given for exemplification.

Fig. 2. 2-D PSD plotted for the surface morphology of L31.

For that reason we have preferred to compute the 1-D PSD function S1 (fx ) ob-
tained by integrating S2 (fx, fy ) defined by Eq. (3 ) over fy .
It can be shown [7] that S1 (fx ) essentially represents the average over y of the
1-D PSD S1 (fx, y) computed according to Eq. (4) from profiles traced in x (fast scan)
direction:
N
1 X
S1 (fx ) ≈ S1 (fx , yn ). (5)
N n=1
In order to increase the frequency range of the resulting PSD, two sets of calcu-
lations were performed, based on 10×10 µm2 scan size, respectively 1×1 µm2 scan
size. For a better statistic two to four data sets obtained from different locations of
the samples were processed and then averaged for each case. Because of the tip pre-
disposition to blunt in the much wearing 10 µm scans, the shorter wavelength region
(magnitudes comparable to estimated tip radius) of the PSD curves obtained from
these scans were set aside in subsequent calculations. Finally the PSDs computed
from 10×10 µm2 scans and 1×1 µm2 scans were properly combined for every sample
to give the extended PSD.
296 R. Gavrila et al.

5. Results and disscution

Figure 3 a–d displays the surface morphology of the uncoated glass substrate and
respectively of L31, L48 and L27 thin film samples.

(a) uncoated glass substrate (b) sample L31

(c) sample L48 (d) sample L27

Fig. 3. AFM images of the surface morphology of substrate and titanium oxide
thin films. Scan area 10×10 µm2 .
In Fig. 4 the fine-grained structure of thin film samples is more explicitly revealed
in the 1×1 µm2 images. The larger grains in the structure of L31 film compared with
L48 and L27 films could be easily perceived.
A Power Spectral Density Study of Thin Films Morphology 297

substrate L31

L48 L27
Fig. 4. 1×1 µm2 AFM images of the substrate and titanium oxide samples,
displaying the granularity of the films.

Table 2 contains the rms roughness values for the films and glass substrate, com-
puted for the 1×1 scans in order to avoid the influence of the large casual defects
which appear in the 10×10 µm2 images (see Fig. 3).

Table 2
uncoated glass L31 L48 L27
rms roughness 1.0 7.0 3.3 4.4
(nm) for 1 µm2

Figure 5 displays in a log-log plot the resulted PSD profiles computed by the
technique described in the previous section for the three samples, together with that
for the substrate.
As one can see, all PSD curves essentially present the same characteristic shape,
consisting in a flat response in the lower part of the spatial frequency spectrum and
a power law roll-off with frequency in the upper part of the spectrum. The strong
spikes in the mid and high frequency range of the substrate PSD were neglected in
the presents study, as additional work is needed to elucidate if they correspond to real
periodicities of the surface morphology or to measurement artifacts and/or noise.
298 R. Gavrila et al.

Fig. 5. PSD profiles of thin film samples and substrate, calculated

from AFM measurements, displayed together with their ABC-model fits.

The curves depicted in Fig. 5 clearly indicate that over the entire spectrum of
frequencies the thin film morphology dominates the substrate morphology for all the
samples. It is to be mentioned that in the range of spatial frequencies which come into
play for optical (visible and UV) scattering, for all the films the roughness is quasi
uniformly distributed with frequency (white spectrum); the slight deviation of L27
at low frequencies will be discussed below. Considering the topographic scattering
in optical applications, it is obvious that this would not be influenced by improving
the substrate finish and scatter reduction could be achieved only by changing the
deposition conditions, both for visible and UV regions.
The steadiness and relative smoothness of the thin films PSD curves, their quasi-
alignment in the low spatial frequency zone and their monotonic ranking with respect
of the roughness (in descending order L31, L27, L48) over the whole range of frequen-
cies suggest a similar overall mechanism underlying film growth for all samples.
For a condensed description of the thin film and substrate morphology, we made
use of the so-called ABC of k-correlation model, which proved to be applicable to
a large range of sample morphologies and allows quantitative comparison between
samples over large length scales [1]:

A
S (f ) = h iC , (6)
2 2
1 + (B · f )

where A, B and C are model parameters.

A is the value of the spectrum in the low-frequency limit, the “shoulder parameter”
B is a “correlation length” which sets the point of the transition between the low and
high-frequency behavior, and C is the exponent of the power-law fall-of f at high
frequencies. At low spatial frequencies (f ¿ 1/B) the PSD is constant and equals A;
at high f values, the surface is fractal, its PSD function scaling as 1/f C . This model
A Power Spectral Density Study of Thin Films Morphology 299

can be seen as a generalization of the fractal case (pure roll-off of PSD with increasing
frequency), with a flat response below a cut-off frequency.
As shown in Fig. 5, the ABC model succeeds in describing quite satisfactory the
morphologies for both the substrate and all thin film samples over the entire spatial
frequency range. The A, B and C parameters of the respective fits are given in Table 3:

Table 3
substrate L31 L48 L27
A (nm3 ) 200 3500 482 1066
B (nm) 595 123 82 96
C 1.9 3 3.4 3.5

Several observations could be drawn from the above results:

– In the mid and high frequency regions, the glass substrate behaves as an ideal
fractal surface (C∼2).
– The value of A parameter is related to the low frequency component of surface
roughness, while the value of B parameter (the correlation length), is linked to grain
size.
– The roughness value increases with the grain size. For L31 and L27 films,
which have about similar phase composition, grain size (and roughness) enhances
with increasing film thickness and substrate temperature. Slightly smaller grain size
(smaller B) and consequently smaller roughness seem to characterize pure rutile phase,
obtained with the greater substrate temperature (sample L48) in comparison to mixed
rutile and atanase phases of L31 and L27.
– While at high spatial frequencies the roughness values of L48 and L27 tend
to converge, their PSD curves move away as the frequency lowers. This behavior
is reflected by the similarity of B (grain size) and C (power exponent) and by the
considerable difference of A values. The low frequency upper deviation of L27 SD
from the ABC model may be attributed to the presence of large scale structures
(aggregates) [8] in a greater number in L27 than in L48, which is consistent with the
lower substrate temperature during deposition for L27.

6. Conclusions

The main goal of this work was to establish and test a procedure for computing
and combining PSD spectra from AFM data acquired on different length scales on
optical surfaces. This procedure was applied for studying the morphology of three
different thin titanium oxide films obtained by dc magnetron sputtering. This type
of study provides a more comprehensive understanding of the influence of the obtain-
ing conditions on morphological features of the films and could help in tailoring the
deposition parameters according to surface morphology requirements for a specific
application. Particularly, as for optical thin films PSD is closely linked to the bidirec-
tional reflectance distribution function (BRDF), AFM data can support optimization
of the obtaining processes with the view of reducing scatter losses in thin film optical
coatings.
300 R. Gavrila et al.

References
[1] CHURCH E. L., TAKACS R. Z., 7. Surface Scattering, in Handbook of Optics (2nd ed),
vol. I, Fundamentals, Techniques and Design, McGraw Hill, New York, 1995.
[2] BANGE K. et al., Thin Solid Films, 197, p. 279, 1991.
[3] SAWADA Y., TAGA Y., Thin Solid Films, 116, p. 155, 1984.
[4] SIEFERING K. L., GRIFFIN G. L., J. Electrochem. Soc., 137, p. 1206, 1990.
[5] Annual Report 2003, Fraunhofer-Institut für Silicatforschung ISC.
[6] MARDARE D., BABAN C., GAVRILA R., MODREANU M., RUSU G. I., Surface
Science, 507, pp. 468–472, 2002.
[7] MARX E. et al., J. Vac. Sci. Technol. B, 20(1), pp. 31–41, 2002.
[8] SENTHILKUMAR M. et al., Applied Surface Science, 252, pp. 1608–1619, 2005.