You are on page 1of 13

PROCEEDINGS OF SPIE

SPIEDigitalLibrary.org/conference-proceedings-of-spie

Determination of residual stress in


low-temperature PECVD silicon
nitride thin films

Martyniuk, Mariusz, Antoszewski, Jarek, Musca, Charles,


Dell, John, Faraone, Lorenzo

Mariusz P. Martyniuk, Jarek Antoszewski, Charles A. Musca, John M. Dell,


Lorenzo Faraone, "Determination of residual stress in low-temperature
PECVD silicon nitride thin films," Proc. SPIE 5276, Device and Process
Technologies for MEMS, Microelectronics, and Photonics III, (2 April 2004);
doi: 10.1117/12.523327

Event: Microelectronics, MEMS, and Nanotechnology, 2003, Perth, Australia

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use


Determination of residual stress in low temperature PECVD silicon
nitride thin films
M. Martyniuk*, J. Antoszewski, C.A. Musca, J.M. Dell, and L. Faraone
The University of Western Australia, Crawley, WA 6009, Australia

ABSTRACT

Two experimental techniques have been investigated to examine residual stress in low temperature plasma enhanced
chemical vapour deposited (PECVD) SiNx thin films: one that measures the stress induced substrate curvature, and the
other that takes advantage of the stress induced deformation of freestanding diagnostic microstructures. A general linear
dependence of residual stress on deposition temperature is observed, with the magnitude of stress changing linearly
from circa 300MPa tensile stress to circa 600MPa compressive stress as the deposition temperature is decreased from
300°C to 100°C. However, the results deviate from the linear dependence by a different degree for both measurement
techniques at successively lower deposition temperatures. The stress values obtained via the substrate curvature method
deviate from the linear dependence for deposition temperatures below 200°C, whereas the values obtained via the
diagnostic microstructures method deviate from the linear dependence for deposition temperatures below 100°C. Stress
uniformity over the deposition area is also investigated.

Keywords: silicon nitride, residual stress, thin films, MEMS

1. INTRODUCTION

The ability to control internal stress in thin films is an essential factor in the realization and subsequent performance,
reliability, and long-term stability of almost any micro-mechanical, electronic, magnetic or optical device, where thin
films are used. It is often the case that this stress control needs to be attained while adhering to restrictions that are
directly imposed by the processing technologies associated with the device application area. An example of the above is
the integration of silicon based micro-electromechanical systems (MEMS) technology and compound semiconductor
based infrared (IR) detectors, with which severe restrictions on the maximum processing temperatures are introduced.
Typical MEMS technologies allow significantly higher film deposition temperatures, which are compatible with silicon
integrated circuit technology, but incompatible with most compound semiconductors. This is especially the case
wherever silicon nitride structural elements are part of a MEMS device. High quality SiNx thin films are usually
deposited using low pressure chemical vapour deposition (LPCVD)1-4. However, this process requires high temperatures
(∼700°C) and, generally, the resulting films are under considerable tensile stress with limited opportunity for stress
tuneability.
Currently, the highest performing photon detectors operating in the IR spectral range are obtained utilizing mercury
cadmium telluride (HgCdTe)5,6. The device technology based on this semiconductor employs relatively low temperature
processes (<100°C), with higher processing temperatures causing irreversible damage to devices.
The HgCdTe IR detectors available today are sensitive to a broad spectral band with no ability to discriminate for
wavelength of the detected photons (Fig. 1). The resulting detector output signal represents the total power of the
incident IR radiation within the broad spectral range of the detector. Consequently, the quantitative information
contained within the spectrum of the received IR radiation cannot be extracted. Future IR systems require more than this
envelope detection, with new and emerging applications requiring the detection of narrow spectral lines in the IR
spectrum that can allow unambiguous material/object detection, chemical analysis, and/or accurate temperature
measurement. To satisfy the requirements of these applications, future IR detectors need to be capable of discriminating
within a narrow spectral band that is tuneable over large portions of the IR spectrum.

*
mariusz@ee.uwa.edu.au; tel. +61 8 9380 3745; fax. +61 8 9380 1095

Device and Process Technologies for MEMS, Microelectronics, and Photonics III, J.-C. Chiao, 451
A. J. Hariz, D. N. Jamieson, G. Parish, V. K. Varadan, Editors, Proceedings of SPIE
Vol. 5276 (2004) © 2004 SPIE · 0277-786X/04/$15 · doi: 10.1117/12.523327

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
typical sensitivity
band of present
HgCdTe Figure 1: Sensitivity band of present broadband HgCdTe detector
detectors
with a cutoff of 5.7µm, and the narrow band tuning of the proposed
device superimposed on the background of a sample IR
transmittance spectrum. The narrow spectral features associated
with various chemical species cannot be detected with the present
tuning of the narrow spectral sensitivity bands
of proposed next generation IR detectors day detectors. However, the tuneability of the narrow sensitivity
band of next generation detectors would allow for chemical species
identification.

At present, IR spectral data are typically obtained using grating or Fourier transform spectrometers. These are bulky
instruments that employ sensitive and expensive optics. It is proposed that their performance, both in terms of detector
sensitivity and spectral resolution, can be matched by an inexpensive microscale device that is realized through the
integration of HgCdTe IR detection technology with silicon nitride based MEMS. The concept, as shown schematically
in Fig. 2, is based on the HgCdTe IR detector sensing only IR photons that have been discriminated for a particular
wavelength by a tuneable Fabry-Perot (FP) interferometric filter. This could be accomplished by placing the IR detector
between two IR reflectors separated by a cavity length that directly determines the detected (or resonant) wavelength.
Electrostatic actuation can be used to vary the position of the top reflector, thus providing a means of controlling the
cavity length and resulting in wavelength tuneability. It is important to note that the length of the optical cavity must be
modulated by a relatively large amount. Applications in the mid-wave (MWIR, 3µm−5µm) and long-wave (LWIR,
8µm−12µm) IR regions require approximately 1.0µm and 2.0µm displacements of the movable reflector, respectively.
To fabricate the tuneable FP cavity, the top reflector must be supported by a suspended membrane, which is a very
critical element that must satisfy many requirements simultaneously. The membrane must be strong enough to support
the top reflector, and should not introduce unwanted deformation under the reflector as to keep the optical cavity stable.
However, at the same time the membrane must have a high degree of flexibility in order to allow large and reliable
displacement of the top reflector. Accurate control of the residual stress in the SiNx support membrane is essential in
order to obtain the desired characteristics. If the SiNx thin film is deposited with a resulting high compressive residual
stress, the released membrane will collapse and fail to produce a cavity. On the other hand, high tensile residual stress
renders the membrane to be mechanically very stiff, thus preventing the large displacements required for IR wavelength
selectivity in the FP optical filter. Minimizing the residual stress is of paramount importance in the realization of the
proposed microspectrometer device. Moreover, such characteristics must be attained under the strict restriction of low
temperature (<100°C) processing that is imposed by HgCdTe IR technology.
SiNx deposited by the LPCVD technique cannot meet the low temperature deposition requirements imposed by
HgCdTe. More promising is the use of plasma enhanced chemical vapour deposition (PECVD), which allows lowering
of the SiNx deposition temperatures, however, this may compromise film quality. This study explores the influence of
decreasing the deposition temperature on both the residual stress and stress uniformity in PECVD SiNx thin films. Stress
measurement techniques are also compared, since the residual stress values have been obtained via optical
measurements of stress induced substrate curvature and via stress induced deformation of diagnostic microstructures.
The next section introduces the basic understanding of stress and its effects in thin films, and provides the background
for stress measurement in thin films. Subsequently, the experimental details (including the fabrication process used to

IR radiation

top reflector
SiNx membrane Figure 2: The proposed structure of the next generation HgCdTe
detector. The narrow band sensitivity is obtained by resonance
air support
support
d cavity HgCdTe structure
phenomena of the Fabry-Perot interferometric filter. Only the
structure
IR detector
bottom reflector
resonance wavelength of the optical cavity reaches the HgCdTe
electrostatic force-plate electrodes
detector, and wavelength tuning is performed through electrostatic
substrate actuation of the optical cavity length.

452 Proc. of SPIE Vol. 5276

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
fabricate the diagnostic microstructures) are presented, and followed by the presentation and discussion of the results
obtained.

2. THIN FILM STRESS

The major influence of stress in thin films can be visualized by considering the schematic presented in Fig. 3. During
the deposition of a thin film the constituent particles adhere to the substrate and subsequently tend to be separated from
each other by an equilibrium distance, schematically denoted a0, that corresponds to the system lowest energy state
allowed (Fig. 3a). This separation is associated with the film being in a state of lowest (minimal) stress. Figure 3b
illustrates the case where the thin film is deposited with a resulting compressive stress. This corresponds to the
constituent particles ordering themselves closer to each other as compared to the minimal stress scenario (acompressive<a0).
In order to relieve the stress, the constituent particles will tend to spread apart so as to approach and possibly reach the
equilibrium, minimum stress separation a0. Consequently, if adhesion of the film to the substrate is not compromised,
this stress relaxation will result in bending of the substrate concave up (as shown in Fig. 3b). Alternatively, if the
deposition of the thin film results in tensile stress, the constituent particles’ separation is greater than that resulting from
the minimum stress deposition (atensile>a0). In this case, approach of the constituent particles to their equilibrium
separation positions, that results in the relaxation of this stress state, causes bending of the substrate concave down (as
shown in Fig. 3c). The more stressed the thin film, the more prominent is the bending of the substrate. Thus, by
measuring the substrate’s radius of curvature one can determine the presence and magnitude of the stress in the
deposited thin film.
The quantitative determination of the thin film stress, σf, via the thin film’s stress induced substrate bending is
performed with the aid of the Stoney formula7,
Es d s2 ,
σf =
6 R(1 − vs ) d f

where Es, vs, and ds are Young’s modulus, Poisson ratio, and thickness of the substrate, respectively, R is the
experimentally obtained substrate’s radius of curvature, and df is the thickness of the thin film. This relationship is
conditional on ds >> df.
This measurement method, however, is only able to provide a value of the stress averaged over a large (typically several
cm2) deposition area. For meaningful stress values the film’s properties should be uniform over this large deposition
area, which is not a trivial task in a PECVD process.
An alternative concept that can be utilized for the measurement of stress in thin films relies on the introduction of a
sacrificial layer between the substrate and the deposited thin film (Fig. 4). Selective removal of the sacrificial layer
renders in a suspended thin film that is free to deform under residual stress, without introducing any substrate bending.
Thin films, deposited with a resulting compressive residual stress, would buckle upwards (or down) as the constituent
particles tend to approach their equilibrium separation distance to relax the stress (Fig. 4a). In a similar fashion, the
constituent particles of a film that is deposited with a resulting tensile stress would tend to decrease their separation

(a) a0 Figure 3: Visualization of the influence of thin film stress on the


curvature of the deposition substrate. Deposition that results in near
zero thin film residual stress is characterized by the preferred
thin film near zero stress separation, a0, of the constituent particles (a). Since the thin film’s
acompressive atensile
(b) (c) constituent particles separations are smaller (larger) for thin film
deposition events resulting in compressive (tensile) stress, the
relaxation of stress through the displacement of the constituent
a0 particles towards their preferred separations causes concave up (b)
a0 (down (c)) bowing of the underlying substrate. This is provided
that the thin film to substrate adhesion is not compromised. The
nomenclature of concave up and concave down is as presented in
thin film compressive stress thin film tensile stress (b) and (c), respectively, and is referred to in the text in this way.

Proc. of SPIE Vol. 5276 453

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
compressive stress near zero stress tensile stress Figure 4: Visualization of the influence of thin film stress on the
deformation of thin film suspended layers. Deposition that results
in a near zero thin film residual stress is characterized by the
preferred separation, a0, of the constituent particles (b). Since the
thin film’s constituent particles separations are smaller (larger) for
thin film deposition events resulting in compressive (tensile) stress,
the relaxation of stress through the displacement of the constituent
particles towards their preferred separations causes buckling up (a)
(possible rupture (c)) of the suspended thin film. This is provided
that no stress relaxation is associated with the sacrificial layer and
(a) (b) (c) substrate.

distance. Since the thin film is anchored to the unremoved portions of the sacrificial layer, tensile forces are created
within the film, which may even lead to its rupture (Fig. 4c).
One can take advantage of this stress induced deformation to extract quantitative information on the thin film stress by
fabricating diagnostic microstructures as shown in Fig. 5. The doubly clamped beam or a microbridge (Fig. 5a) would
be sensitive to compressive stress. Relaxation of the thin film compressive stress would tend to expand the structure.
Since the beam is clamped at both ends, compressive forces arise at the two anchor points resulting in buckling of the
beam upwards (or down). However, the presence of tensile stress in the thin film would stretch rather than buckle the
microbridge, therefore one must resort to other geometries. The ring-and-crossbeam structure, shown in Fig. 5b, would
convert tensile forces that are present in the film into structural compressive forces that cause structural buckling of the
crossbeam. If the ring-and-crossbeam structure were completely detached from the substrate, relaxation of the thin film
tensile stress would uniformly shrink the structure. Since the ring is clamped at two diametrically opposite points, the
tensile forces that are present at these anchor points elongate the ring, such that compressive forces are consequently
developed on the crossbeam, hence buckling the crossbeam.
The quantitative information of the stress in the thin film can be obtained in two ways. The first approach makes use of
the fact that the deformation amplitude increases with an increase in the deposited film stress; therefore, a measurement
of the buckling amplitude will provide a stress value. In the case of compressive stress measurement, the vertical
displacement of the microbridge (of original length L) can be translated into the expanded length of the microbridge by
simple geometry. The magnitude of the length increase ∆L, is then related to the strain in the deposited thin film, ε0,
through
∆L = ε 0 L ,
which can then be translated into stress if the thin film’s Young’s Modulus is known. It is important to note, however,
that it is the uniaxial stress that is measured, whereas the original stress in the film is biaxial.
As depicted in Fig. 6, the tensile strain, ε0, elongates a freestanding ring-and-crossbeam structure (of radius R), so that it
receives an additional radial separation,
∆Lexpansion = 2ε 0 R ,

between its anchor points, as compared to an unclamped structure. This expansion induces a perpendicularly directed
radial contraction (i.e. parallel with the crossbeam)
∆Lcontraction = 2ε 0 Rg (R ) ,
where g(R) (the ratio of contraction to expansion) is the conversion efficiency of tensile strain to compressive strain, and

Figure 5: Schematics of the diagnostic microstructures, whose


vertical deformations are directly related to the magnitude of the
thin film compressive (a) and tensile (b) stress. Thin film
compressive stress results in the formation of compressive forces
on the doubly supported beam at the anchor points (a), whereas thin
film tensile stress is translated into structural compressive forces on
(a) (b) the crossbeam in the ring-and-crossbeam structure (b).

454 Proc. of SPIE Vol. 5276

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Fcompession

2R−2ε0 Rg(R)
Felongation

2R
Figure 6: The deformation of the ring-and-crossbeam structure of
radius R. The elongation of the ring by the outward (tensile) force
2R is translated into perpendicularly directed contraction of the ring,
2 R+2ε 0R effectively buckling the crossbeam. See text for additional details.

is given by

g (R ) = −
2br f 2 , with
2br f1 + bb f12 − bb f 22

R  π 2  2e 4 π 1+ v ,
f1 =  − − + − + πk f
e  4 π  πR π 4 2

R  1 2  2e 1 4
f2 =  − − − + − k f (1 + v ) ,
e  2 π  πR 2 π
br
e=R− , and
ln (Ro Ri )

Ro + Ri ;
R=
2
where kf is a cross-section form factor (equal to 1.20 for a solid rectangle), v is Poisson’s ratio, br is the width of the
ring, bb is the width of the crossbeam, and Ro (Ri) is the outer (inner) radius of the ring8. This conversion efficiency
approaches a value of 0.918 if one is dealing with a very slender ring (small eccentricity e). By the use of simple
geometry, the vertical displacement of the crossbeam can be translated into ∆Lcontraction, allowing for the determination of
tensile strain in the original thin film. With the knowledge of the thin film’s Young’s Modulus a value of tensile stress
can be assigned.
An alternate approach in the thin film stress determination from the deformations of diagnostic microstructures is to use
the critical point buckling geometry. A structural member, in the case of the determination of the thin film compressive
stress using the doubly clamped beam of a given length, will not buckle until a critical amount of compressive stress is
applied. As a corollary, a given amount of applied compressive stress will only buckle beams whose length is greater
than a critical value. In other words, there exists a critical length that can directly reveal the magnitude of compressive
strain present in the as deposited film. This is known as Euler buckling and for a doubly supported beam the
compressive strain, ε0, is given by
h 2π 2 ,
ε0 = −
3lcr2
where h is the beam’s thickness and lcr the critical length at which buckling occurs9.

Proc. of SPIE Vol. 5276 455

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
A similar scenario holds for the ring-and-crossbeam geometry under the presence of tensile stress in the deposited thin
film. The structure, of given dimensions, will deform only if the tensile strain in the original film exceeds a critical
value. Consequently, there exists a critical radius, Rcr, which marks the boundary of crossbeam buckling. Once it is
determined the original film tensile strain, ε0, is given by

ε0 =
(2.224 )2 (h R cr )2 ,
12 g (R )
for the case where the ring and the crossbeam are of the same width (br=bb), which is much larger than its thickness, h8.
In order to determine the critical dimensions, the critical point buckling technique requires fabrication of increasingly
smaller structures with increasing strain magnitudes. This requirement can be a limiting factor in the measurement,
when the fabrication technology restricts structures to a minimum size and prevents measurements of large stress
magnitudes.
The main advantage in the use of diagnostic microstructures for stress measurements is that the probed dimensional
scale is of the same size as the microstructures, compared to the substrate curvature method which gives average values
across the sample. At the same time, diagnostic microstructures provide a means of mapping the local stress field, thus
obtaining information on the stress uniformity over the deposition area. However, fabrication of the diagnostic
microstructures adds complexity (of critical importance is ensuring that no stress is introduced by the sacrificial layer)
and is more time consuming as compared to measurement techniques utilizing the thin film’s stress induced substrate
bending.

3. THE EXPERIMENT AND THE FABRICATION OF THE DIAGNOSTIC MICROSTRUCTURES

Two series of samples were prepared to investigate stress in SiNx films. Stress measurements, via the thin film stress
induced substrate bowing, were performed on a series of samples where PECVD SiNx layers were deposited directly on
100µm thick silicon <100> wafers. The SiNx depositions were performed using an Oxford Instruments PECVD 80
system. The deposition temperature was varied between 50°C and 300°C with all other deposition conditions being
constant and summarized in Table 1. After the deposition, the samples were cleaved into strips with a length to width
radio of 10:1.
The radius of curvature of the SiNx coated silicon substrates was obtained from the angular changes of initially parallel
laser beams, that arise as a result of reflection from the substrate surface. To calibrate samples the radius of curvature of
bare silicon wafers was also measured, confirming that the uncoated substrate bowing was negligible. The experimental
setup is presented in Fig. 7. Three parallel laser beams were created by reflecting a laser beam from an optically flat
half-silvered, 9.78mm thick glass plate. The parallel beams were reflected off the substrate, and any substrate bowing
induces divergence (or convergence) from the parallel of the outer beams with respect to the central beam. As depicted
in Fig. 7 for the case of reflection from the thin film’s (substrate’s) side, the presence of compressive (tensile) film stress
causes concave up (down) bowing and divergence (convergence) of the initially parallel beams. The angular change of
the laser beams was determined by using a CCD camera, mounted on a micropositioning stage to measure the interspot
distance. A variable ND filter was inserted in the path of the laser beam (before the beam splitter) in order not to
overload the CCD camera. Sample to sample comparison accuracy was ensured by the use of an alignment beam that
was reflected off the sample onto an alignment screen, where (through sample rotation and tilt) the samples’ positioning
was affirmed to be in the same plane.

Table 1: The deposition conditions of PECVD SiNx.

Gas chemistry Pressure RF SiNx thickness Temperature


SiH4 : 5sccm
Power : 100W
NH3 : 45sccm 450mTorr 500nm 50°C − 300°C
Frequency : 13.6MHz
N2 : 100sccm

456 Proc. of SPIE Vol. 5276

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
laser
ro lle g
d
in

alignment screen
e c ition
on t
s tag r opos

r
irro
m
mic

ND filters
CC
cam D
era

r
it te
spl
am
Figure 7: The experimental setup for stress evaluation of the

be
PECVD SiNx thin films by the stress induced substrate curvature.
The bowing of the presented sample is understood as concave up
measured sample and causes the initially parallel beams to diverge.

The substrate radius of curvature, R, was determined from the interspot distance, using the relationship
2L D ,
R=
cosθ ∆D
where L is the substrate to CCD camera distance (68cm), θ is the angle of incidence (52.5°), D is the interspot distance
of the parallel beams (6.40mm), and ∆D is the change of the observed interspot distance that is reflected from the
measured sample as compared to the interspot distance of the parallel beams.
The second series of samples were prepared under identical SiNx PECVD deposition conditions as series one (Table 1).
However, in this case the SiNx deposition took place on the polyimide sacrificial layer and the deposited thin films
served as the structural material for fabrication of the diagnostic microstructures, whose deformation was to be used in
order to determine the thin film’s stress. It has been confirmed by measurements that introduction of the polyimide layer
did not affect the SiNx stress. Doubly supported beams and ring-and-crossbeam structures (as shown in Fig. 5) were
fabricated on each sample for the determination of the compressive and tensile stress, respectively. To obtain
information on the stress uniformity over the deposition area, the microstructures were distributed over a sample area of
1.5×1.5cm2. The microbridges varied in length from 10µm to 104µm in increments of 2µm, and from 105µm to 460µm
in increments of 5µm. The ring-and-crossbeam structures ranged from 50µm to 120µm radii in 5µm increments, and
from 130µm to 240µm in 10µm increments. Both kinds of diagnostic structures were fabricated with 10µm and 20µm
SiNx widths.
The process for fabricating the diagnostic microstructures is presented in Fig. 8. An HD Microsystems product, PI-2616,
low stress polyimide was chosen to serve as the sacrificial layer due to minimal stress introduction and the availability
of a dry process (i.e. oxygen plasma ashing), with which the freestanding microstructures could be released. The
polyimide was spun on a silicon substrate, which was previously cleaned using a standard RCA cleaning process10, to
produce a final polyimide thickness of ∼1.2µm. Subsequently, the polyimide was cured to obtain stable low stress
characteristics and to prepare its surface for succeeding SiNx thin film deposition. Next, the structural SiNx film was
deposited, on which a photoresist layer was spun and patterned. The photoresist layer pattern was transferred onto the
SiNx layer using CF4 plasma reactive ion etching (RIE). The RIE chemistry was changed to oxygen in order to strip the
photoresist pattern and to selectively remove the polyimide sacrificial layer, without etching the SiNx. The release of the
suspended microstructures was attained by prolonging the exposure to the oxygen plasma, so as to produce an undercut
in the polyimide layer. The amount of this undercut was carefully controlled in order to achieve full removal of the
polyimide material from under the microstructures, with enough material left to anchor the support structures. The
residual stress has been determined using the values of Young’s modulus which were obtained by nano-indentation of
SiNx thin films deposited under similar PECVD conditions11.
The vertical deformation of the microstructures was measured using an interferometric technique. As shown in Fig. 9,
light that is incident on the freestanding structure will reflect off the suspended thin film as well as off the substrate

Proc. of SPIE Vol. 5276 457

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
SiNx thin film
(a) (b) polyimide
polyimide
silicon substrate
silicon substr ate

CF4
(d)
(c)
photor esist
photoresist SiNx thin film
SiNx thin film polyimide
polyimide silicon substrate
silic on substrate

(e)
(f)
O2

photoresist
SiN x thin film
SiNx thin film
polyimide supports
polyimide
silicon substrate
silicon substrate

Figure 8: The fabrication process of the diagnostic microstructures. (a) Polyimide spin coating of silicon substrate, to serve as the
sacrificial layer (∼1.2µm). (b) PECVD SiNx deposition (∼500nm), the structural layer of the microstructures. (c) Spin coating and
patterning of photoresist to the desired geometry. (d) Photoresist pattern transfer to the SiNx layer using CF4 plasma RIE. (e) O2
ashing of the photoresist layer and selective removal of the polyimide sacrificial layer to produce (f) a freestanding SiNx
microstructure.

surface, with the resulting light waves free to interfere with each other. If the incident light is monochromatic, an
interference pattern will be created on the suspended thin film with destructive interference fringes formed every time
the separation between the suspended thin film and the substrate surface changes by λ/2. Counting the number of the
interference fringes on the suspended structures revealed the vertical deformation. The measurements were performed
using an optical microscope fitted with a green (λ=532nm) interferometric filter.

suspended
thin film
λ

Figure 9: The illustration of the interferometric technique used to


determine the vertical deformation of the diagnostic
substrate ∆d = λ/2
microstructures. Destructive interference fringes form every time
the distance between the suspended thin film and the substrate
surface changes by λ/2, as a result of optical interference of the
reflected monochromatic light waves (λ=532nm).

458 Proc. of SPIE Vol. 5276

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
4. RESULTS AND DISCUSSION

The fabricated freestanding diagnostic microstructures are shown in Fig. 10, which presents the SEM micrographs (Fig.
10a&b) as well images obtained using an optical microscope fitted with an interferometric filter (Fig. 10c&d). The
interferometric fringes are clearly visible and reveal not only the presence, but also the amplitude of the vertical
deformation of the doubly supported beams (Fig. 10c) and the vertical deformation of the ring-and-crossbeam structure
(Fig. 10d). Figure 11 presents the room temperature experimental results for the residual stress as a function of the
PECVD SiNx deposition temperature. For comparative purposes, this figure presents results obtained via thin film stress
induced substrate bowing, as well as results obtained via the deformation of the diagnostic microstructures. It is clear,
that in the case of results obtained by the diagnostic microstructures method, the measurements obtained via the critical
point buckling method agree with those obtained via the deformation amplitude method. In addition, it is clearly seen
that the room temperature residual stress of PECVD SiNx depends strongly on the deposition temperature. By
considering the results obtained via the diagnostic microstructures, the magnitude of the residual stress changes linearly
from circa 300MPa tensile stress to circa 600MPa compressive stress as the deposition temperature is decreased from
300°C to 100°C. These results compare favorably with previous reports12. It should be noted that the slope of this linear
relationship is PECVD chamber dependent. This was determined by analyzing the thin film stress on samples that were
placed in different radial positions of the PECVD chamber.
Minimal stress introduction by the polyimide layer was verified by measuring the radius of curvature of two reference
samples. The first sample, where a polyimide layer was spun on a silicon wafer, confirmed that the polyimide did not
introduce noticeable substrate bowing. The aim of the second reference sample was to observe any stress effects related
to the polyimide layer, simultaneously mimicking the experimental conditions as close as possible. This sample
experienced the same processing steps as the sample used to determine the substrate curvature induced by the stress in
the SiNx thin film deposited at 200°C, with the exception of the polyimide layer introduction between the silicon
substrate and the SiNx thin film. It can be clearly seen in Fig. 11 that the substrate curvature method reports the same
residual stress value for both cases (i.e. SiNx thin film deposition at 200°C on bare silicon wafer, and on the silicon
wafer on which polyimide was spun), confirming that the polyimide layer does not introduce any noticeable stress.
It is also clearly visible (Fig. 11) that the residual stress values obtained via the substrate bowing technique are in good
agreement with the residual stress values obtained via the diagnostic microstructures for deposition temperatures above
200°C. Below this deposition temperature the substrate bowing method reports values that progressively deviate from
values obtained by the diagnostic microstructure method, and seem to approach minimal stress values with decreasing
deposition temperature. Such a discrepancy between the two measurement methods could be accounted for if the
adhesion of the SiNx thin films deposited below 200°C was increasingly compromised with decreasing deposition
temperature. However, it was confirmed by an “adhesion tape” test that this is not the case, and that the adhesion of the
SiNx film is not compromised to such a degree that it could explain the discrepancy between the measurement methods
used. It is also well known that a low temperature SiNx deposition process results in moisture absorption by the
deposited thin film. If this absorption of water was associated with the relief of compressive stress, it could provide an
explanation for the progressively decreased thin film’s substrate bending with decreasing deposition temperature. The
lower the deposition temperature, the more moisture could be absorbed, and as a result more stress could be relieved by

(a) (b) (c) (d)

Figure 10: The freestanding diagnostic microstructure images obtained using SEM (a&b) and an optical microscope fitted with a 532
nm interferometric filter (c&d). Vertical deformation (buckling) is clearly visible with the help of the interferometric fringes present
on the doubly supported beams (c), and on the crossbeam of the ring-and-crossbeam structure (d).

Proc. of SPIE Vol. 5276 459

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
400

200
Residual Stress [MPa]

-200

-400 buckling amplitude − microstructures

critical buckling − microstructures

-600 SiNx layer − substrate bending

SiNx/polyimide layer − substrate bending

-800
0 50 100 150 200 250 300 350
Depositon Temperature [oC]

Figure 11: The room temperature residual stress of the PECVD SiNx thin films as a function of the deposition temperature. The open
(full) diamonds and full squares represent data obtained via the diagnostic microstructures buckling amplitude (critical point
buckling) method and via the stress induced substrate curvature method, respectively. Due to processing restrictions on the structure
sizes, the diagnostic microstructures critical point buckling method provided values for relatively small magnitudes of stress only.
Each point corresponds to the curvature measurement of a separate sample (for substrate curvature method) or to a different location
of the diagnostic structure on the sample (for diagnostic microstructures method). The full grey circle corresponds to the substrate
curvature measurement of the second reference sample (as defined in the text). The true SiNx thin film stress is provided by the
diagnostic microstructures method in the deposition temperature range of 100°C−300°C and the substrate curvature method in the
deposition temperature range of 200°C−300°C.

the moisture absorption process rather than through the bending of the substrate. However, such reasoning does not
explain the fact that the thin film has not been affected in a similar fashion at the microscopic level. The absorption of
moisture should affect both the substrate curvature method and diagnostic microstructure method in a similar fashion, as
the deposition temperature is decreased in the PECVD process.
Another issue, that needs to be considered during the comparison of the stress results by the two measurement
techniques, is that the SiNx layer on the second series of samples (that were used in microstructure fabrication)
experienced prolonged exposure to the RIE oxygen plasma during the sacrificial layer removal. It is reasonable to
assume that any RIE-related impact on residual stress would have the same effects on every sample from the second
series of experiments. Since the two measurement techniques report identical SiNx residual stress values for high
temperature (>200°C) depositions, it is reasonable to conclude that exposure to the oxygen plasma is not correlated with
the discrepancy between the two measurement methods that is found for the low deposition temperature (<150°C)
region.
It is believed that the increasingly porous nature of the SiNx films deposited at lower deposition temperatures is
responsible for the discrepancy between the two measurement methods. As the deposition temperature decreases, the
SiNx thin film becomes increasingly porous and the deposition-induced residual compressive stress is relieved not
through bending of the substrate, but by expansion into the pinholes/micropores, which causes the stress relaxation to be
averaged over the substrate area. However, in order to be consistent with the obtained results, this effect cannot occur
during the microstructure fabrication process using SiNx material deposited at deposition temperatures as low 100°C.
This seems to be confirmed by the fact that the device yield of the successful microstructures remained relatively
constant (above 90%) for deposition temperatures greater than 100°C. However, the yield of successful microstructures
decreases significantly when the deposition temperature is below 100°C such that, eventually, no suspended

460 Proc. of SPIE Vol. 5276

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
microstructures were successfully fabricated for a SiNx deposition temperature of 50°C. At these low deposition
temperatures, the extent of the micropores/pinhole density reaches the microscale and the integrity of the thin film is not
sufficient to serve as the structural layer for the production of suspended microstructures. This discussion is further
supported by the fact that the results obtained by the diagnostic microstructure method for SiNx deposition temperatures
below 100°C, deviate from the linear dependence with the deposition temperature. For the substrate curvature samples,
the presence of an underlying (i.e. supporting) material, maintains the average integrity of the low temperature deposited
PECVD SiNx thin film, and results in samples that are suitable for measurement of the substrate bowing. That is, it is
only during subsequent processing and patterning of the SiNx thin film, that the poor integrity of the film becomes
apparent.
It has to be noted that the SiNx thin film layers deposited below 100°C become partially permeable to TMAH based
photoresist developer. This is not observed for SiNx layers deposited at temperatures above 100°C, and suggests the
presence of a significant density of micropores/pinholes. However, inspection of the SiNx film quality under an optical
microscope does not provide evidence for the presence of pinholes beyond a reasonable doubt. Therefore, there may be
further factors that have an influence on the discrepancy between the measurement of the residual stress via the stress
induced substrate bowing and via deformation of the diagnostic microstructures, and await further investigation in the
future. However, this is an additional factor that confirms that a decrease in the deposition temperature of the PECVD
process of SiNx thin films renders films of poorer quality.
Each data point presented in Fig. 11 corresponds to a different location of the diagnostic structure on the sample area.
An indication of the stress uniformity across the deposition area can be estimated by considering the measured stress for
one particular deposition temperature. In Fig. 12, the residual stress values that correspond to the PECVD of SiNx at a
deposition temperature of 150°C are plotted versus the position of the diagnostic microstructure on the wafer. For
clarity, the position of the diagnostic microstructures is plotted in one dimension only. It is evident that the stress varies
smoothly by approximately 60MPa over a 10mm change in the position of the diagnostic microstructure along the
wafer. However, if one accepts a measurement error of ±1 interference fringe, which corresponds to circa ±30−40MPa
(depending on the length of the microbridge used in the measurement), this variation in the measured stress value is
within the experimental uncertainty.
It needs to be noted that the investigated residual stress has two major components. The first contribution arises due to
the difference in thermal expansion coefficients between the SiNx thin film and the Si substrate, i.e. the thermal stress.
The other is the intrinsic stress associated with the thin film deposition process itself. At present, the experimental set-up
is being modified to permit measurements at temperatures from 150K to 400K, which will allow separation of the
thermal and the intrinsic components of the residual stress.

-300

-320
Residual Stress [MPa]

-340

-360

-380

-400
0 2 4 6 8 10 12 14
Diagnostic Microstructure Relative Position [mm]

Figure 12: The room temperature residual stress of the PECVD SiNx thin films as a function of the position of the diagnostic
microstructure along the wafer for deposition temperature of 150°C. The data were obtained via the diagnostic microstructures
buckling amplitude method. For clarity, the position of the diagnostic microstructures is plotted in one dimension only.

Proc. of SPIE Vol. 5276 461

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
5. SUMMARY AND CONCLUSIONS

It is proposed that next generation HgCdTe IR sensors, which will be characterized by a narrow wavelength band which
is tuneable over a specific range is the IR spectrum, can be achieved by the integration of present day HgCdTe detectors
with SiNx-based MEMS technology. Attaining accurate control of the residual stress in SiNx thin films deposited at
temperatures below 100°C is identified as a critical step in the successful development of the technology. For this
purpose, two stress measurement methods have been investigated for assigning thin film residual stress. Namely, the
substrate curvature method and the stress induced deformation of diagnostic microstructures method. Measurements of
the residual stress in PECVD SiNx thin films deposited in the deposition temperature range of 50°C to 300°C have been
performed, and exhibit a linear dependence of stress with deposition temperature such that the stress becomes
increasingly compressive at low deposition temperatures. However, the results deviate by a significant degree from the
linear dependence for deposition temperatures below 200°C. Stress uniformity over the deposition area is found to be
within the range of ±30MPa, with a monotonic variation of the stress values having been found over a 10mm linear
distance of the diagnostic microstructure position along the wafer.
Currently, the experimental set up is being modified to extend the measurement temperature range to 150K−400K and
allow separation of the thermal and intrinsic components of the residual stress. It is also apparent from the obtained
results, that the quality of the SiNx thin films is significantly compromised for deposition temperatures below 100°C.
Therefore, successful development of the next generation HgCdTe IR sensor may require other thin film deposition
methods, such as the inductively coupled plasma chemical vapour deposition (ICP-CVD) technique.

REFERENCES

1. A. Stoffel, A. Kovacs, W. Kronast, B. Muller, “LPCVD against PECVD for micromechanical applications”, Sixth
European Workshop on Micromechanics (MME’95), pp. 1-13, Copenhagen, Denmark, 1996
2. M. Shenasa, M. Moinpour, B. O’Toole, B. Stueve, R. Wilkes, J. Reece, D. Borror, J. Glarneau, “Optimization of
LPCVD silicon nitride process in a vertical thermal reactor: use of design of experiments”, Advanced
Semiconductor Manufacturing Conference and Workshop, ASMC 92 Proceedings. IEEE/SEMI 1992 , pp. 216-219,
1992
3. M. Modreanu, N. Tomozeiu, P. Cosmin, M. Gartner, “Physical-optical properties of LPCVD amorphous silicon
rich-nitride and oxynitride”, Semiconductor Conference, CAS ‘98 Proceedings. 1998 International, Volume 1, pp.
201-204, 1998
4. I. Kobayashi, T. Ogawa, and S. Hotta, “Plasma-enhanced chemical vapor deposition of silicon nitride”, Jpn. J.
Appl. Phys., 31, pp. 336-342, 1992
5. J.M. Dell, J. Antoszewski, M.H. Rais, C.A. Musca, J.K. White, B.D. Nener, L. Faraone, “HgCdTe mid-wavelength
IR photovoltaic detectors fabricated using plasma induced junction technology”, J. Electron. Mat., 29, pp. 841-
848, 2000
6. J. Antoszewski, C.A. Musca, T. Nguyen, J.K. White, R.H. Sewell, J.M. Dell, L. Faraone, “Small 2D arrays of
MWIR diodes fabricated by reactive ion etching”, J. Electron. Mat., 32, pp. 627-632, 2003
7. G.G. Stoney, “The tension of metallic films deposited by electrolysis”, Proc. R. Soc. London, Ser. A, 82, pp. 172-
175, 1909
8. H. Guckel, D. Burns, C. Rutigliano, E. Lovell and B. Choi, “Diagnostic microstructures for the measurement of
intrinsic strain in thin films”, J. Micromech. Microeng., 2, pp. 86-95, 1992
9. H. Guckel, T. Randazzo, D.W. Burns, “A simple technique for the determination of mechanical strain in thin films
with applications to polysilicon”, J. Appl. Phys., 57, pp. 1671-1675, 1985
10. W. Kern, D.A. Poutinen, “Cleaning solutions based upon hydroperoxide for use in silicon semiconductor
technology”, RCA Review, 31, pp. 187-206, 1970
11. K. J. Winchester, J.M. Dell, “Nano-indentation characterisation of PECVD silicon nitride films”, Proc. Conf.
Optoelectronic, Microelectronic Materials and Devices, pp. 117-120, Melbourne, Australia, 2000
12. K. J. Winchester, S.M.R. Spaargaren, J.M. Dell, “Tunable Fabry-Perot cavities”, Internat. Symp. on
Microelectron. and Assembly (ISMA 2000), Singapore, SPIE Proc. 4230, pp. 198-209, 2001

462 Proc. of SPIE Vol. 5276

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 22 Oct 2020


Terms of Use: https://www.spiedigitallibrary.org/terms-of-use

You might also like