In situ ellipsometric monitoring of complex
multilayer designs
Current developments in optical multilayer design and computation make it possible to calculate filters that
satisfy the most demanding optical specifications. Some of the designs are highly sensitive to manufacturing
errors and require accurate monitoring and control during thin film deposition. Ellipsometric monitoring
enables the accurate deposition of any thickness, including very thin layers, and in situ measurement of
both refractive index and thickness of the layers during deposition, which facilitate the subsequent real-time
design reoptimisation. In this letter, a number of complex multilayer designs with the aid of ellipsometric
monitoring are presented, including a laser notch plus band-blocker filter, dichroic filter, beamsplitter, and
a wide-range broadband multiplayer antireflection coating.
OCIS codes: 240.2130, 310.1860, 120.2130.
doi: 10.3788/COL201008Sl.0044.
In recent years, significant progress has been made in the
field of multilayer optical coating design[1?5]. Nearly any
optical filter specification can be designed theoretically,
and a number of solutions have become available for very
complex and challenging structures. Unfortunately, the
required accuracy and reproducibility of the properties of
a material for such structures are sometimes so high that
mass production may be impossible, even in state-of-theart
deposition systems. Hence, the accuracy of in situ
monitoring and effective feedback control of the deposition
process remain significant issues to be addressed.
A number of well-established optical and non-optical
techniques allow the monitoring and control of the layer
deposition process[6]. The applicability of a particular
method is dependent on the deposition technique and
the design requirements. For a deposition process being
able to maintain a constant rate and stability of the optical
constants of the material, satisfactory results can
be achieved by using a simple time-termination process.
Deposition for a specific period is often adequate for the
sputtering techniques when the required accuracy of film
thickness is within a few nanometres[7?9]. However, in
many coating methods, including evaporation, the material
refractive index, and deposition rate resulting from
simply timing the deposition are not sufficiently stable
to produce precise optical filter coatings. Quartz crystal
monitoring, another frequently used non-optical method
for thickness monitoring, indirectly measures the deposition
rate and film thickness. It is simple, easy to install,
and relatively cheap. However, the random thickness errors
of the crystal monitoring systems used in production
are in the order of 4%[10], which makes it inadequate for
the production of certain complex optical structures. Optical
in situ monitoring can produce much better results
since the monitoring is conducted using the optical parameters
of the structure, which are more directly related
to the end performance. Optical monitoring techniques
can be subdivided into two major categories: photometric
methods (e.g., reflection, transmission, and others) and polarisation-dependent methods (e.g., ellipsometry,
polarimetry, and others). Many optical designs can be
successfully produced using single or multi-wavelength
photometric monitoring. Single wavelength optical monitoring
is the most widely used production technique.
Turning point[11] and level monitoring[12,13] are the popular
single wavelength monitoring methods suitable for
designs based on periodic quarterwave optical thickness
(QWOT).
Broadband transmittance and/or reflectance have been
widely used for deposition monitoring of non-quarterwave
designs over the last three decades[14]. The enhancement
of the computational power of modern computers
has boosted the practical application of photometric
methods and has led to the increased capability in
data processing and deposition control. For example,
sub-nanometre accuracy in thickness control has been
reported for single layer deposition[15]. Real-time reoptimisation
based on broadband reflectance and transmittance
data has been implemented for the manufacture
of high-performance non-quarterwave designs[16]. Moreover,
advances in broadband photometric monitoring
have allowed nanometre-level thickness control in multilayer
production.
Ellipsometry is one of the most widely used
polarisation-dependent methods. In situ ellipsometry
has been successfully applied to a diverse range of applications
in both research and industry. It is relatively
simple, reliable, and provides highly accurate real-time
measurements of both the thickness and refractive index
of growing layers[17?20]. It also provides robust information
for the reoptimisation of the design according to
the measured properties of the deposited layer. In many
cases, this can be accomplished without interrupting of
the deposition process. Owing to its advantages, ellipsometry
has grown in popularity. It has been extensively
used in the study of optical thin film material properties
and in the monitoring of multilayers with demanding
specifications[21?26]
.
Ellipsometry is a century-old technique, and its theory
and applications are covered at length[17?20]. Progress
in the automation of ellipsometric instruments started in
the early 1960s and has accelerated significantly during
the last decade. The introduction of fast and inexpensive
computers has allowed the development of broadband
spectroscopic ellipsometers with the latest commercial
instruments, enabling their expansion into both the vacuum
ultraviolet and mid-infrared ranges[27]. Currently,
high-accuracy data can be acquired over broad spectral
ranges within seconds[28]. Improvements in ellipsometric
instrumentation have advanced research and industrial
capabilities in traditional areas such as multilayer optical
coatings, and have triggered the development of a new
range of applications[19]
.
Ellipsometry has a number of advantages compared
with conventional photometric monitoring. Unlike photometric
monitoring, where only one value is obtained
per measurement, ellipsometers measure two ellipsometric
values, Ψ and ?. As a result, two parameters of a film
can be simultaneously determined for each measurement
point: refractive index n, and thickness d. In the case of
a bare substrate, the optical constants n and extinction
coefficient k can be directly derived from Ψ and ?.
The ellipsometric angles, Ψ and ?, are defined (for
reflection) in Eq. (1). In this equation, ρ is defined as
the complex ratio of the Fresnel reflection coefficients for
p-polarised light to that for s-polarised light[17]
.
where rp and rs are the Fresnel reflection coefficients,
|rp| and |rs| are the amplitude terms, and δs and δp
are the phase change on the reflection terms for p- and
s-polarised light, respectively.
Traditionally, ellipsometers are divided into four major
groups: early single wavelength nulling and principal
angle[29,30], division-of-amplitude[31,32], phasemodulated[33],
and rotating element[34]. The merits and
limitations of different configurations are discussed in
detail[18?20]. All the multilayers presented in this letter
were manufactured using a multi-laser line (up to five
wavelengths from the visible to near-infrared (NIR) region)
rotating analyser ellipsometer (RAE) constructed
in-house, as described in detail in the reports of Netterfield
et al. and Hauge et al.
[35,36]. The basic setup of
the instrument is shown in Fig. 1.
An incident laser wavelength is selected by the
computer, then polarised on transmittance through a
computer-controlled Glan-Thompson polariser that can someter.
be set to any angle within 0.02?
. To increase the accuracy
of the measurement of ?[36], a quarterwave plate
compensator is used when ? is approximately ±10? of
either 0? or 180?
. The compensator optical axis is set
at approximately 45?
to the plane of incidence. Light
then passes through the vacuum window at normal incidence
and onto the sample. The angle of incidence is
~70? and all angles can be measured up to 0.01? accuracy.
The reflected beam exits the vacuum system
through another window (which is also orthogonal to the
light beam), passes through the analyser, and then onto
a silicon detector. The analyser rotates at a constant
frequency of up to 10 Hz. The signal intensity is measured
using a 16-bit analogue-to-digital (A/D) converter
at 225 equally spaced angular positions over a full revolution.
The measured intensity of the signal I(θ) is related
to the azimuth of the transmitting axis of the analyser θ
by the following equation[36]:
where I0 is the average intensity for a full rotation of the
analyser. The coefficients a and b are calculated from a
Fourier analysis of the intensity variations as a function
of the analyser azimuthal angle, and related to Ψ and ?
by
where P and C are the azimuths of the fast axes of the
polariser and compensator, respectively, and ρc is the
fast-to-slow complex relative to the transmittance of the
compensator
where Tc is the ratio of the transmittances of the compensator
along its fast and slow axes, and ?c is the
relative retardation along the axes.
A number of high-precision optical coatings have
been successfully produced using the real-time, multiwavelength
ellipsometry. Crystal monitoring has been
used to maintain the stable deposition rate, and a spectrophotometer
has been used for broadband transmission
(reflection) monitoring. In this letter, several types of
multilayer optical filters are presented, including broadband
anti-reflection coatings, dichroic mirrors, beamsplitters,
and colour-corrected laser protection filters.
All the multilayer filters presented here were produced
using an ion-assisted deposition system[37?39] built for
the purpose. The high-refractive-index oxide layers and
magnesium fluoride were deposited by ion-assisted electron
beam evaporation. Likewise, SiO2 layers were deposited
by the ion-assisted reactive thermal evaporation
of silicon monoxide in the presence of oxygen.
To date, one of the most difficult optical designs we
have fabricated was a colour-corrected, near-infrared
blocking with a 532-nm notch laser protection filter[25]
.
The multilayer optical thin film coating consisted of 79
layers of Ta2O5 and SiO2 of various thicknesses (Fig. 2).
The filter design was extremely sensitive to errors
in thickness and deviations of optical properties of the
constituent films. Its specifications required an optical
density > 4 at 532 nm, optical density > 3 over
the 690–1100 nm range, photopic transmittance > 55%
(wherein 60.4% was achieved within 1% of the design
value), and colour saturation < 10% (wherein less than
5% was achieved). The results are shown in Fig. 3.
Laser protection filters require the thickness of each
layer to be accurate within 0.5 nm in order to meet the
required optical performance specifications. To achieve
the high level of control required during the fabrication of
such a multilayer optical coating, real-time ellipsometric
monitoring is paramount.
The advantage of ellipsometric monitoring over the
single or multi-wavelength photometric monitoring was
evident in the monitoring of layer 48 in this filter. The
calculated spectral transmittance of the multilayer at the
end of the layer and the corresponding layer, which was 5 nm less than the correct thickness, is shown in Fig. 4.
Note that the figure shows little difference between the
curves over the wavelength range 450–650 nm, a typical
range for a multi-wavelength transmittance monitoring
system.
Considering the uncertainty in the spectrum caused
by the errors in the previous layer (i.e., wavelength shift
and absolute value), the monitoring substrate run-out
due to the rotation, and the drift in the absolute value
of the monitoring response, such a method is unlikely
to positively resolve the monitoring difference of layer
48, which requires better than a thickness error of approximately
5 nm. Using ellipsometry, there are many
wavelength options where the sensitivity at the termination
layer is significantly better than a nanometre. For
the same layer, Layer 48, the change in ? over the last
5 nm was 18.7?
, and the change for Ψ was 0.28? at the
633-nm monitoring wavelength. Even for an inaccurate
ellipsometer, the termination of the layer to << 1 nm
was assured.
Broadband antireflection filters and dichroic coatings
present other design challenges. Here, a broadband antireflection
coating was designed to achieve a reflectivity
of < 0.5% in the wavelength regions of 400–550 nm and
600–900 nm, and a reflectivity of < 0.01% at 1319 nm.
The calculated and measured optical performance of the
broadband anti-reflection coating is shown in Fig. 5. Re-
flectivity was measured using a Cary-5 spectrophotometer.
The reflectivity at 1319 nm was measured separately
using a 75-mW laser to verify that the filter satisfied the
specified requirement (< 0.01%). The coating was about
1.5-μm thick and consisted of 26 alternating layers of
Ta2O5 and SiO2, with MgF2 as the outside layer[26]
.
An example of a dichroic filter that requires the subnanometre
accuracy during fabrication is shown in Fig.
6. This coating was designed to provide high reflectance
(>98%) at 650–900 nm and 1319 nm, and high transmittance
(>85%) at 400–550 nm. The beamsplitter coating
consisted of 24 alternating layers of Ta2O5 and SiO2.
The total thickness of the coating was ~1.8 μm[26]
.
Neutral beamsplitter coating is another example of a
multilayer design that requires sub-nanometre accuracy
in order to meet specifications, as shown in Fig. 7. The
beamsplitter coating is sandwiched between two glass
flats with an anti-reflection coating on both the outside
faces. Here, this component was assembled by optical
contacting. The multilayer coating was designed with
the following specifications: 0.95 < Tp,s/Rp,s < 1.05 in
the range of 400–900 nm and 1319 nm (Tp,s and Rp,s
are the transmittance and reflectance for the p- and spolarisations,
respectively). The coating was ~3.8-μm
thick and consisted of 36 layers of TiO2 and SiO[26]
2
.
All the multilayer filters presented here were very sensitive
to fabrication errors, in both thickness and refractive
index, and required a high level of control of the deposition
process. Thus, the given examples indicate that
ellipsometric monitoring is a very versatile method and
yields good results in a wide range of applications.
In order to achieve a high level of deposition control,
the monitoring strategy must be adjusted for each particular
design. To monitor the deposition, pre-calculated
ellipsometric curves were used for each layer. During
the modelling process, it was vital to select the optimum laser wavelength to achieve sufficient gradients in ? and
Ψ at the end of each layer. These monitoring curves were
used to automatically terminate the deposition when the
ellipsometric parameters have reached the target values
(Fig. 8).
All minor variations of the refractive index during deposition
and/or deviations from the targeted layer thickness
were computed using the in-house-developed ellipsometric
analysis software, and simultaneously taken into
account. In our deposition system, the largest refractive
index variations were observed in the high refractive index
materials at the beginning of a deposition run (Fig.
9). This behaviour is thought to be due to the change in
the substrate temperature.
The observed variations in the refractive index and/or
thickness of the layers were used in the reoptimisation of
subsequent layers.
There are a number of instrumental challenges to overcome
in order to obtain precise and accurate ellipsometric
measurements in real-time and in a production environment,
especially for multilayer coatings. One of
the problems is the monitored beam wobble caused by
the rotation of the substrate. In most deposition systems,
substrate rotation is essential to achieve the required
uniformity of the growing film. The rotation of
the substrate creates a challenge in alignment, and in
maintaining the orthogonality of the rotating substrate
to the ellipsometer’s plane of incidence with high accuracy
(preferably within < 0.01?
). A number of strategies
have been reported to minimise the effect of substrate
wobble and improve the signal-to-noise ratio of the ellipsometric
measurements[40]. In our system, alignment
and calibration procedures were carried out on a rotating
substrate (~1 Hz). Time averaging of the measurements
was performed to reduce the periodicity induced
in the ellipsometric data by the substrate wobble. Computer
simulation with all the available wavelengths can
be used to determine the best monitoring wavelength for
each layer in order to achieve the highest possible sensitivity
and lowest error.
Interpretation of in situ ellipsometric data requires
fitting real-time measurements to a model. To create
such a growth model, knowledge regarding the optical
properties of the growing film is necessary. Extensive ex
situ measurements of the optical properties of the deposition
materials were carried out and a library of optical
constants for the materials was created. In order to reduce
the “operator” effect and maintain the optical properties
close to those of the design model, a computerised
feedback system was developed to control the deposition
process.
In order to minimise the coating deviation from the design,
special attention was given to the layers with the highest relative sensitivity to errors in the layer thickness
during deposition. As a result, deposition processes capable
of producing a whole array of broadband optical
coatings to sub-nanometre accuracy were developed.
In conclusion, we have demonstrated that in situ ellipsometry
can be utilised under real-time production conditions
to manufacture a range of multilayer structures
with high accuracy. Furthermore, future works aimed
at extending the capabilities of the in situ ellipsometric
system to monitor mixed and graded materials whose
optical properties are extremely dependent on the deposition
conditions are in the planning stages. This improved
ellipsometric system would allow small deviations from
the targeted parameters to be detected immediately, and
conditions to be adjusted to achieve the required optical
properties.
Additional improvements in the ellipsometric data processing
would include the simultaneous measurement at
several wavelengths and the development of more “intelligent”
software that will enable more efficient assessment
of the sensitivity and error limits for each wavelength.
The author gratefully acknowledges the valuable assistance
and vital inputs of Dr. M. Gross, Dr. A. Bendavid,
and Dr. A. Chtanov.