Jn situ test benches have been designed and fabricated allowing to applied elementary solicitations traction, bending and torsion to representative samples.. In this study, in situ test
Trang 13E103.P
MODELING OF FAILURE MECHANISMS FOR OPTIMIZED MEMS CAD:
DESIGN, FABRICATION AND CHARACTERIZATION OF IN SITU TEST BENCHES
Olivier Millet®, Vincent Agache’, Bernard Legrand’, Dominique Collard”, and Lionel Buchaillot®
Institut d'Electronique, de Micro-électronique et de Nanotechnologie IEMN*
and Center for International Research on MicroMechatronics CIRMM?
IEMN UMR CNRS 8520 Avenue Poincaré Cité scientifique 59652 Villeneuve d’ Ascq, France
Tel.: (+0033) 3 20 19 78 38, Fax: (+0033) 3 20 19 78 84, E-mail: lionel.buchaillot@isen fr
ABSTRACT This work considers the reliability of surface
microfabricated structures and particularly the dynamic
response of structural layers during operations, in order to
develop a statistical modelling of failure mechanisms for
micro-actuator Jn situ test benches have been designed and
fabricated allowing to applied elementary solicitations
(traction, bending and torsion) to representative samples
Gold, in situ doped polysilicon and polysilicon doped by
diffusion are used as structural layers These devices are
useful to study the fatigue phenomenon Characterization
and fatigue tests have been performed in a vacuum
chamber under different environmental and stimuli
conditions Moreover, a theoretical analysis using Finite
Elements Method has been achieved
INTRODUCTION The next generation of CAD programs for MEMS will
have to ensure a precise evaluation of the system lifetime,
by anticipating the failure mechanisms The principle is to
develop a statistical modeling, indicating the probability
that a failure mechanism occurs, via the determination of
statistical rules describing probability of failures generated
in each elementary structure (beam, etc.) during the
operation of the whole system [{1, 2, 3] Then, this work
aims at understanding the fatigue phenomenon for
elementary structures and to describe it statistically
according to the design, the use and the environment of the
microsystem
Our activity deals with the design and the realization of
micro-actuators This type of structure involves a lot of
motions, and therefore stress and friction The study of the
failure mechanisms can be divided into two parts, which
are the reliability of the fabrication process and the
reliability of the microsystem itself In this study, in situ
test benches have been developed allowing the analysis of
the fatigue phenomenon of structural layers (reliability of
There is more and more reliability work published on
micro-actuators From gear transmissions to pop-up
mirrors, micro-actuators are used to drive many different
types of devices [4, 5] Brown has investigated the
actuator reliability based upon crack propagation in
polysilicon (1] Sandia Lab has worked on the lifetime of
polysilicon micro-engines [2] Muhlstein and Brown have
analyzed the fatigue with a crack initiation [3] In this
paper, we will first explain the principle of our in situ test
benches Secondly, modeling of the structures will be
presented After, the fabrication process will be presented,
followed by the characterization method and the results
PRINCIPLE Each displacement generates the combination of elementary mechanical solicitations (bending, traction, torsion) Then, in order to quantify the fatigue phenomenon, it is interesting to analyze the evolution of properties (Young’s modulus, etc.) of materials, which are submitted to stress due to elementary movements, versus the number of functioning cycles With this type of device, we can characterize the material every n cycles Since electrostatic actuation is mostly used in the MEMS‘s field, each structure will be electrostatically actuated in order to apply bending, torsion and traction solicitations, under different external conditions
In order to investigate the effects of stress, different types
of structures have been designed For example, in the case
of torsion test, torsion solicitations have been applied to the middle of a beam (Fig 1) or to the ends of a beam (Fig 2) Such an approach allows the application of different stress levels for the same type of elementary
solicitation As other example, some structures are
designed with the aim at performing either torsion or bending test (Fig 1), allowing us to compare the effects of each solicitations on a same structure
Comb-drive structures (Fig 2) can be used as sensor,
allowing us to detect the movements and the frequency of movements during test From previous observations, polysilicon (structural material commonly used) and gold
applications) samples have been designed and fabricated Then, arrays of in situ fatigue test benches have been designed for each elementary solicitation (Fig 2) In the next sections, the characterization method being the same for all tests, only the torsion tests are presented as example (Fig 1, 2)
test test
>> ® 8+7
Figure 1 A torsion/bending test with solicitations applied
at the middle of the beam The structural material is in situ doped polysilicon
TRANSDUCERS ‘03
The 12th International Confarence on Solid State Sensors, Actuators and Microsystems, Boston, June 8-12, 2003
ñ 7ana_7744_1/n4/417 ña @20ñ 1FEE 1578
Trang 2
Torsion
samples
Tested
Figure 2 Array of gold/polysilicon torsion (est benches
with electrostatic actuation The comb-drives are used as
sensors
FABRICATION PROCESS
The in situ test benches have been fabricated by surface
micromachining technology, using a six mask levels The
smallest dimension of the mask pattern is 5 um A 5-20
ohm.cm’' P type (100) silicon wafer was used as the
substrate Figure 3 depicts the process fabrication of one
polysilicon structural layer for the sake of clarity The
structures fabrication requires two structural layers
Sacrificial
layer Bushing Contact
Buried
Sid, layer Ta Diolectric layer
Dieiectric Ez==m Structural Sacrificial
[li layer sio, layer Ey layer
Figure 3 Thin film micromachining of one polysilicon
structural layer
Polysilicon process
A 0.35 um oxidation is performed followed by the first
Low Pressure Chemical Vapor Deposition (LPCVD) n-
doped polysilicon layer After electrode patterning, the
polysilicon was oxidized and covered by a LPCVD low
stress SixN, layer (Fig 3a) A 24m Low Temperature
Oxide (LTO) was deposited followed by the Reaction Ion
Etching (RIE) of bushing and contact (Fig 3b) A 2 um
LPCVD polysilicon layer was deposited This layer is
doped either during the deposition (i situ doped
polysilicon), or by diffusing dopants in the polysicon via
the use of a PhosphoSilicate Glass layer (removed in
Hydrofluoric —HF- after diffusion) The structural pattern
was defined by SFs, CF, and O, RIE (Fig 3c) Then, a
2Hm LTO was deposited followed by the etching of dimples and contacts 2 {1m polysilicon was deposited and doped (as explained previously), and the structural pattern
was delineated by SF,, CF,, Oz etching Next the structure
was annealed at 1000°c for 60 min to relax residual stress
Finally, the structure has been released using HF (see Fig
1, 2, 3d)
Gold process
A 0.35 um Plasma Enhanced Chemical Vapor Deposition (PECVD) oxide is deposited followed by the evaporation
of Titanium (0,1 um) and Gold (0,25 pm) After electrode patterning, the metallic bilayer is covered by a PECVD Si3Ng layer at 200°C (Fig 3a) A 2 um PMGI photoresist was deposited followed by the etching of bushing and contact (Fig 3b) Aluminum mask (0,1 pm) was used during etching A 0,6m gold was sputtered The structural pattem was defined by chemical etching (Fig 3c) Then, a 2 um PMGI was deposited followed by the etching of dimples and contacts 0,6):m gold was sputtered, and the structural pattern was delineated by KI (Potassium/lode) etching Finally, the structure has been released using O, RIE and EBRPG etching (Fig 2, 3d)
SIMULATIONS
CoventorWare™ has been used to analyze the systems;
via Finite Element Method (FEM), the natural frequencies
of the structures have been found in the non-damping case (Fig 4) Moreover, simulations have allowed the determination of the pull-in voltage and the level of applied stress in the pull-in configuration (Fig 4) The following data are obtained in the case of a polysilicon
structure showed in figure 2
Stressp 2 sw i7 †_, vemsge
Voltage (Vv)
Ệ as [ener
28
tụ voltage
5 Voltage (v) Pull-in voltage™~s 7
‡
Resonancer |_——*£% Frequency (Hz)
Figure 4 FEM analysis of stress during torsion test with electrostatic actuation (in the case of polysilicon)
Determination of the pull-in voltage and of the resonant
CHARACTERIZATION
In order to determine the natural frequency of the structure, the wafer is glued on a piezoelectric ceramic
The excitation of the piezoceramic -and therefore the wafer- is controlled by a network analyzer An Optical Beam Deflection (OBD) method [6] (Fig 5) is used to determine the exact value of first natural frequency OBD
is a well-known non-destructive optical method: a laser
TRANSDUCERS ‘03
The 12th International Conference on Solid State Sensors, Actuators and Microsystems, Boston, June 8-12, 2003
1579
Trang 33E103.P
beam is focused on the sample by means of a microscope
objective The reflected beam reaches the four-quadrant
photodiode used as a vibration detector The photodiode
signal is compared to the excitation frequency leading to
the exact value of the first natural frequency of the
microstructure
Tested wafer
TT
Sd oe a he
Figure 5, Optical beam deflection method applied in the
After the previous operation, devices are actuated in a
vacuum chamber (Fig 6) The actuation voltage
corresponds to the pull-in voltage and the excitation
frequency is equal to the first natural frequency The
applied pull-in voltage and the resonant frequency are
obtained via FEM analysis Every cycles, the first
natural frequency is measured It allows to observe the
evolution of the rigidity vs the number of the functioning
cycles, and to determine the maximal number of
operations before the destruction of the structure
Moreover, the external conditions (pressure, vibrations,
etc.) have been changed in the vacuum chamber, allowing
then to study the influence of environmental parameters
system
Figure 6 Vacuum chamber Close-up views show
heater/cooler system and visualization module
EXPERIMENTAL RESULTS
In order to correctly interpret the experimental results, the
measured first natural frequency has been normalized to
the first natural frequency of the structure before any
actuations For the first experiment, the tested structures
have been actuated under vacuum; the voltage and the
excitation frequency were those determined via the FEM
analysis; the corresponding stress being known too The
results show us that there are three phases in the failure
mechanism (Fig 7), First, an adapting phase (Fig 7A), corresponding to the increase of the natural frequency occurs This adaptation is explained by the creations of microstrains due to stresses, jamming microdisplacements
by the multiplication of dislocation nodes, and then
increasing the rigidity Secondly, the apparition of
microcracks involves the decrease of the resonance frequency (Fig 7B), up to the rigidity stabilization and the final destruction of the system (Fig 7C)
108 5
% Ð% 108
6 š 14
z = 102
Ss oss
Ê Ễ oss
FB oss
Z8 94
032
oR
Number of functioning cycles (Millions)
Figure 7 Fatigue evolution for a polysilicon clamped- clamped beam with the three phases of the fatigue phenomenon
400 6a
The simulations with CoventorWare™ show us that for a same clamped-clamped beam, the stress is bigger with
torsion applied to the middle (Fig 1) of the beam than
with torsion applied to the ends (Fig 2) By actuating these two types of torsion test benches at 10 mTorr, we
can observe that the higher the induced stress, the quicker
the fatigue phenomenon occurs (Fig 8)
Conceming the influence of the pressure, two different
structures have been actuated (Fig 1), at 10 mTorr and at
the atmospheric pressure As expected, the decrease of the pressure (and then of the damping) involves the acceleration of the fatigue phenomenon (Fig 8)
The next tests deal with the external mechanical vibration, which is an important part in the reliability field A torsion test bench (Fig 2) is glued on a piezo-ceramic and is electrostatically actuated under vacuum (10 mTorr); | during this actuation, a voltage is simultaneously applied
to the piezoceramic with an excitation frequency equal to
the resonant frequency of the tested structure We observe
a weak acceleration of the fatigue phenomenon in the vibrations case (Fig 8)
The effects of thermal shocks on fatigue phenomenon
have also been studied in the case of polysilicon structures To investigate these, reshaping technology is used [7] Joule heating is applied in the air during a short
time to induce plastic deformation of the beam As a result
of the annealing effect, the plastic deformation of polysilicon occurs We assume this joule heating as a
thermal shock, and fatigue tests are performed on torsion benches (Fig 2) As a result, the fatigue evolution is as
fast as a non-annealed structure, but the maximum rigidity increase is more important Moreover, the lifetime of the annealed structure is very short (Fig 8)
TRANSDUCERS ‘03
The 12th International Conference on Solid State Sensors, Actuators and Microsystems, Boston, June 8-12, 2003
1580
Trang 4By repeating experiments, statistical rules -describing
probability of failures generated in each elementary
structure (beam, etc.) during the functioning of the whole
system- can be determined (Fig 8)
With external Under high
16 a vibrations vacuum
With thermal shocks
Waxtmol rigdity obtained duting the fafigue phenomenon
(Nonnatized on the Young's Modatus before actuation)
Acceleration of the fatigue phenomenon compated to a
simple scinution th the alr
Figure 8 Statistical modeling of the influence of external
parameters on the fatigue phenomenon in the case
of a polysilicon clamped-clamped beam with a torsion
solicitation
Concerning the visual observations, they are performed at
each measurement via a x100 optical microscope Most of
observations concern the clamped ends, where the induced
stresses are very high Figure 7 shows us the evolution of
a clamped end of a structure (Fig 2) vs the number of
functioning cycles Microcracks can be observed This
area is well defined It corresponds to the step covering of
the first structural polysilicon layer (edge of the contact)
by the second structural polysilicon layer (tested beam)
performed and have confirmed a change of the surface
topology [9] This experiment demonstrates the brittleness
of microstructures where step covering occurs
The stiction phenomenon is a_ well-know failure
mechanism which has often been analyzed No in situ test
benches have been fabricated for this study The
phenomenon has been studied on a whole system and an
analytical modeling has been created, based on the
observations [10]
CONCLUSION This study presents behavior of polysilicon/gold
elementary microbeams, which are submitted to
elementary movements (bending, traction, torsion) Jn situ
test benches have been designed and fabricated in order to
determine material properties These devices are used to
analyze the fatigue phenomenon The measurements allow
us to confirm that the natural frequency is a good indicator
of the fatigue It appears that this phenomenon can be
divided in two different phases Finally, the influence of
induced stresses, thermal shocks, external vibrations, and
pressure has been investigated
This work allows a new approach of failure mechanisms
in a whole system (resonator, optical devices) by a better
elementary mechanical solicitations The statistical
modeling of failure mechanisms is an efficient way to
optimize the design and the fabrication Integrated in CAD program, this tool improves the mechanical quality
of next MEMS Acknowledgements
The authors would like to thank the valuable work of the IEMN technical staff, advices of P Bigotte, and SEM
micrographs of C Boyaval The authors would like to thank in the same way the ‘Delegation General de l’Armement’ for the financial support in 2002
References {1] S.B Brown Materials Reliability in MEMS devices
Technical Digest, 1997 International Conference on solid-state sensors and actuators, Transducers 97,
pp 591, 593 [2] D M Tanner “First reliability test of a surface
Proceedings SPIE Symposium on Micromachining and Microfabrication, Vol 3224, Austin, 1997, pp 14-
23
(3] C Muhlstein, S Brown Reliability and fatigue testing of MEMS Tribology issue and Opportunities
in MEMS, November 1997 Klewer Acad Public
[4] J.J Sniegowski Multi-level polysilicon surface micromachining technology: application and issues
Proceeding of the ASME Aerospace Division, Vol
52, 1996, pp 751-759
[5] http:/www.mdl,sandia.gov/Micromachine/
[6] L Buchaillot, E Farnault, M Hoummady and H
Fujita Silicon nitride thin films Young's modulus determination by an optical non destructive method,
Jpn J Appl Phys Vol 36 (1997) pp L794-L797
[8} E.H Yang and H Fujita, ‘Reshaping of single-crystal silicon microstructures’, Jpn J Appl Phys Vol 38
(1999), pp 1580-1588
[9] O Millet, B Legrand, D Collard and L Buchaiilot
Influence of the Step Covering on Fatigue
Phenomenon for Polycrystalline Silicon MEMS Jpn
J Appl Phys Vol.41 No.11B pp.L1339 - L1341 {10] V Agache, E Quevy, D Collard and L Buchaillot Stiction-controlled locking system for three- dimensional self-assembled microstructures : Theory and experimental validation Applied Physics Letters,
Vol 79, No 23, pp 3869-3871, 2001
The 12th International Conference on Solid State Sensors, Actuators and Microsystems, Boston, June 8- 42, 2003
1581
3E103.P