Design and Fabrication of Micromachined Silicon Suspensions: © Sunil Kumar, 2007
Design and Fabrication of Micromachined Silicon Suspensions: © Sunil Kumar, 2007
Design and Fabrication of Micromachined Silicon Suspensions: © Sunil Kumar, 2007
Silicon Suspensions
Sunil Kumar
May 2007
This thesis presents the design and fabrication of a low noise lateral silicon suspension
for seismic sensing. Optimizing for the sensor size, and performance, a lateral
ng/√Hz.
Low-noise, low-frequency (ω) and high-sensitivity inertial sensors are used to measure
the displacement of the Earth’s crust. Typical high performance seismometers are few
kilograms in weight with NEA in the sub-nano-g range. The large size and high cost of
current-state of the art seismometers however presents a barrier to its widespread use.
The ability of seismometer to resolve vibrations is set by the self-noise of the system so
to detect low level vibrations the self-noise of the sensor should be lower than the
vibration being measured. The product of mass, period and quality factor is inversely
proportional to the self-noise power spectrum of the sensor. Even though there are
miniaturized vibration sensors like micromachined accelerometers, they have large self-
noise due to the small mass and short period of the sensor.
To realize a micro seismometer with NEA near the low noise seismicity model, a
etching. Frames were placed between sets of springs to increase cross-axis spurious
mode rejection. Analytical and numerical models are developed to simulate the
suspension dynamics.
i
Fabricating high-aspect-ratio structures by through-wafer etching is made possible by
the use of deep reactive ion etching (DRIE). The springs have an overall vertical profile
with some bowing at mid-depth which is accounted for in the dynamical design. Etch
fracture strength and Q. Deleterious artifacts of DRIE plasma processing are analyzed
NEA of 0.4 ng/√Hz at a resonant frequency of 12.65 Hz. Electronic readout of resonant
frequencies shows a frequency difference of 100 Hz between the fundamental and next
mode.
The suspension developed here can be utilized for other sensing applications. Through-
wafer DRIE has been shown to be a reliable process for creating very high-Q, low-
ii
Table of contents
iii
3.9 Conclusion ....................................................................................................106
3.10 Bibliography..................................................................................................108
4 Suspension: Fabrication ........................................................................................110
4.1 Introduction ...................................................................................................111
4.2 Deep Reactive Ion Etching (DRIE)...............................................................112
4.2.1 Applications of DRIE............................................................................115
4.3 Fabrication Process Flow ..............................................................................116
4.3.1 System description ................................................................................117
4.3.2 Mask preparation...................................................................................119
4.3.3 Pattern transfer ......................................................................................122
4.3.4 Through-wafer etch...............................................................................124
4.3.5 Wafer mounting techniques ..................................................................126
4.4 DRIE Process Parameters .............................................................................129
4.5 DRIE Induced Artefacts................................................................................132
4.5.1 Grassing ................................................................................................132
4.5.2 Etch lag .................................................................................................134
4.5.3 Loading effect .......................................................................................136
4.5.4 Top edge cavitations .............................................................................137
4.5.5 Etch masking.........................................................................................138
4.5.6 Scalloping..............................................................................................138
4.5.7 Sidewall roughness................................................................................139
4.5.8 Etch profile evolution............................................................................149
4.5.9 Notching................................................................................................154
4.6 Process Engineering ......................................................................................168
4.6.1 Process description................................................................................171
4.6.2 Fabricated devices.................................................................................172
4.7 Summary .......................................................................................................178
4.8 Conclusion ....................................................................................................179
4.9 Bibliography..................................................................................................181
5 Suspension: Characterzation .................................................................................187
5.1 Measurement Techniques .............................................................................188
5.1.1 Resonant frequencies ............................................................................188
5.1.2 Quality factor ........................................................................................192
5.2 Experimental Work .......................................................................................193
5.2.1 Resonant frequencies ............................................................................194
5.2.2 Quality factor at varying pressure .........................................................197
5.3 Effect of Etch Profile Shape on Suspension Dynamics ................................200
5.3.1 DRIE etch profile ..................................................................................201
5.3.2 Spring profile characterization..............................................................202
5.3.3 Analytical model ...................................................................................204
5.3.4 Compensation for a non-ideal etch profile............................................208
5.4 Summary .......................................................................................................213
5.5 Conclusion ....................................................................................................214
5.6 Bibliography..................................................................................................216
6 Microseismometer: Prototype ...............................................................................217
6.1 Seismometer..................................................................................................218
6.2 Lateral Differential Capacitive Array Transducer ........................................220
6.3 Feedback Controller ......................................................................................222
6.4 Electromagnetic Actuator .............................................................................223
6.5 Components of Microseismometer ...............................................................225
iv
6.5.1 Proof-mass wafer ..................................................................................225
6.5.2 Capping wafer .......................................................................................234
6.5.3 Magnetic unit ........................................................................................238
6.6 Device Assembly ..........................................................................................241
6.6.1 Solder reflow.........................................................................................241
6.6.2 Glass frit bonding..................................................................................244
6.6.3 Vacuum encapsulation ..........................................................................244
6.7 Microseismometer Damping.........................................................................245
6.8 Experimental Result ......................................................................................247
6.8.1 Experimental measurement setup .........................................................248
6.8.2 Microseismometer dynamics measurements.........................................249
6.8.3 Vibration and shock test........................................................................251
6.9 Other Issues...................................................................................................258
6.10 Summary .......................................................................................................258
6.11 Conclusion ....................................................................................................260
6.12 Bibliography..................................................................................................261
7 Conclusions and Future Work...............................................................................262
7.1 Microseismometer Characterization .............................................................264
7.1.1 Alternate designs...................................................................................264
7.1.2 Fully electrostatic geophone .................................................................264
7.2 Modelling Extension .....................................................................................265
7.2.1 Application to other inertial sensors .....................................................266
7.2.2 Application to other devices..................................................................267
7.3 Applications of Through-Wafer DRIE..........................................................267
7.3.1 Other through-wafer devices.................................................................267
7.3.2 Silicon molding .....................................................................................268
7.3.3 Single mask through-wafer devices ......................................................272
7.3.4 SOI replacement through-wafer devices...............................................273
7.4 Bibliography..................................................................................................275
Appendix A: Analytical model .....................................................................................276
A1. Translational Modes...........................................................................................276
A1.1 On-axis compliant stiffness (kx): ..................................................................276
A1.2 Cross-axis spring constant along Y-axis (ky): ...............................................277
A1.3 Cross-axis spring constant along Z-axis (kz): ...............................................278
A2. Rotational Modes ...............................................................................................280
A2.1 On-axis rotational mode (kα): .......................................................................280
A2.2 Cross-axis rotational mode about Y-axis (kβ): ..............................................282
A2.3 Cross-axis rotational mode about Z-axis (kγ): ..............................................284
A3. Bibliography.......................................................................................................285
Appendix B: Numerical model .....................................................................................286
B1. Finite Element Analysis (FEA) ..........................................................................286
B2. Isotropic vs. Anisotropic model .........................................................................287
B3. Modal analysis using ANSYS ............................................................................287
B4. Input files for FEA using ANSYS......................................................................288
B5. Bibliography .......................................................................................................307
Appendix C: Fabrication process flow..........................................................................308
C1. Process Flow for Microseismometer Sensor Unit ..............................................308
C2. DRIE Processing of Lateral Suspensions ...........................................................309
v
List of Figures
Figure 1.1 Wang Chen-To’s visualization of Chang Teng’s 132 A.D. design of a seismoscope. Image
taken from [1.5]. .......................................................................................................................................... 3
Figure 1.2 Principle of seismometer: the mass is almost stationary due to inertia while the frame moves
with the ground [1.1]. .................................................................................................................................. 5
Figure 1.3 State-of-the-art commercial seismometers, (a) Streckheisen STS-2, (b) Guralp CMG-40T, (c)
Guralp CMG-3T without outer housing, (d) Sercel L-4, (e) Geotech KS-2000 borehole seismometer, and
(f) Nanometrics Trillium 120P. These are all big, heavy and expensive...................................................... 6
Figure 1.4 The USGS New Low Noise Model [1.17], here expressed as Noise Estimated Acceleration in
m/s2/√Hz of ground motion relative to frequency. Mechanical noise levels are plotted for the proposed
microseismometer. ....................................................................................................................................... 9
Figure 1.5 Comparison of noise and resonant frequency of seismometers and a variety of commercial and
research accelerometers (datasheet, [1.14, 19, 20]).................................................................................. 12
Figure 1.6 Comparison of noise and volume for traditional seismometer and MEMS accelerometers
(datasheet, [1.14, 19, 20]).......................................................................................................................... 13
Figure 1.7 Commercial MEMS accelerometers with sub μg noise levels, (a) Honeywell Q-flex QA3000
with an exploded view of the sensor and its suspension, and (b) Sercel/Tronics GPU3 with SEM of the
suspension. ................................................................................................................................................. 14
Figure 1.8 MEMS accelerometers developed by research groups, (a,b) A 1.0 μg/√Hz self-noise
accelerometer developed by Bernsetin et al. [1.22] at Charles Stark Draper Laboratory, (c,d) Parylene
based low-noise accelerometer developed by Suzuki and Tai [1.23] at University of Tokyo, and (e,f) one
of the very first miniaturized accelerometer developed at Stanford university by Roylance and Angell
[1.26]. ........................................................................................................................................................ 16
Figure 1.9 (a) Lateral suspension used in Analog Devices ADXL series accelerometers fabricated by
LPCVD deposition of polysilicon and then RIE etch [1.35], (b) Vertical suspension created by
anisotropically etching silicon by KOH (Potassium hydroxide) [1.21], and (c) Leaf spring part of a
lateral suspension fabricated by DRIE of silcon bonded to an underlying silicon substrate [1.36].......... 19
Figure 1.10 DRIE devices (a) Endevco accelerometer [1.41], (b) Extended travel range microactuator
[1.42]. ........................................................................................................................................................ 20
vi
Figure 2.10 Variation of mechanical noise with resonant frequency for varying Q.................................. 46
Figure 2.11 Schematic of the suspension design with one spring and mass which is assumed to be 50% of
the die area for initial estimates................................................................................................................. 47
Figure 2.12 Variation of resonant mode with spring thickness for a 20mm×20mm die size and
10mm×16mm proof-mass........................................................................................................................... 49
Figure 2.13 Metal traces running on the springs connecting the electromagnetic feedback coil and
electrostatic capacitive feedback electrodes. ............................................................................................. 50
Figure 2.14 Quality factor (due to squeeze film damping), Qsqueeze variation for multiple spring suspension
(N). ............................................................................................................................................................. 53
Figure 3.1 Schematic of a lateral suspension illustrating the model geometry and the critical dimensions.
.................................................................................................................................................................... 60
Figure 3.2 Schematic of a single spring set for the lateral suspension showing the geometry and critical
dimensions.................................................................................................................................................. 63
Figure 3.3 Fundamental mode shape of a single spring set lateral suspension. The proof-mass is
resonating along the X-axis........................................................................................................................ 64
Figure 3.4 ωy, Mode shape of a single spring set suspension resonating along Y-axis. ............................ 65
Figure 3.5 ωz, Mode shape of a single spring set suspension resonating along Z-axis. ............................ 66
Figure 3.6 ωα, Mode shape of a single spring set suspension resonating about X-axis............................. 68
Figure 3.7 ωβ, Mode shape of a single spring set suspension resonating about Y-axis. ............................ 69
Figure 3.8 ωγ, Mode shape of a single spring set suspension resonating about the Z-axis. ...................... 70
Figure 3.9 Schematic of a lateral suspension with multiple springs, (a) a lateral suspension with n set of
spring, and (b) a specific lateral suspension with two (n=2) spring sets................................................... 72
Figure 3.10 FEA (Finite Element Analysis) model of a lateral suspension with two set of springs solved
using Ansys. The inset shows the arrangement of the nodes and elements in the model............................ 74
Figure 3.11 The transition elements are designed such that the nodes at the common boundary are
coincident and there is no extreme size change across structures. ............................................................ 75
Figure 3.12 Fundamental resonant frequency for lateral suspension with varying number of spring sets.
The values are calculated using the analytical expression from equation 3.20 and compared against FEA
derived values. The deviation of the analytical model from the FEA model can be attributed to the mass of
the suspension and the effect of end-connector geometry. ......................................................................... 76
Figure 3.13 To model the first on-axis harmonic ωx1, the spring units of the lateral suspension can be
considered as discrete masses separated by massless springs. M is the mass of the proof-mass, the spring
are considered as discrete masses m with massless springs of stiffness k connecting them....................... 78
Figure 3.14 Plot of the first spurious mode along X-axis (ωx1) with increasing number of spring sets (n).
With larger number of spring sets the analytical model approaches the FEA model. ............................... 80
Figure 3.15 Rejection ratio of spurious modes calculated using FEA model for lateral suspension with 1,
2, 3, 4, 5 and 6 sets of springs. The rejection ratio drops significantly for multiple springs except for the
rotational mode about X-axis..................................................................................................................... 81
Figure 3.16 Logarithmic plot of vertical sag of a lateral suspension in horizontal and Galperin
configuration. The sag due to gravity increases with higher number of springs. ...................................... 83
Figure 3.17 Schematic of a lateral suspension with a single intermediate frame between two set of
springs. The intermediate frame decouples the additional spring sets....................................................... 85
Figure 3.18 Plot of ωx1/ωx for lateral suspension with n (one to six) spring sets and (n-1) intermediate
frames. The FEA model used to calculate the rejection ratio is based on the model parameters of Table
3.2 with a frame thickness (f) of 60 μm...................................................................................................... 87
Figure 3.19 Plot of rejection ratio for lateral suspension with multiple springs calculated using analytical
expressions from Table 3.3......................................................................................................................... 88
Figure 3.20 Plot of rejection ratio for lateral suspension with multiple springs calculated using a FEA
model.......................................................................................................................................................... 89
Figure 3.21 Error between the FEA model and analytical expressions for the rejection ratios................ 90
vii
Figure 3.22 Plot of rejection ratio for lateral suspension with three spring sets and two intermediate
frames of varying thickness. Most of the rejection ratio except ωx1/ωx improves with thicker frames
though the gain is marginal beyond a certain thickness. ........................................................................... 91
Figure 3.23 (a) Reinforced intermediate frames to reduce the weight of the frame while maintain higher
rigidity of a thicker frame, as compared to (b) Solid intermediate frame which reduces the on-axis
rejection ratio due to the higher weight of the spring-frame unit. ............................................................. 91
Figure 3.24 Comparison of the rejection ratio for a model lateral suspension with multiple springs
without frames and with frames. (a) Log-log variation as calculated by FEA of the rejection ratio for the
spurious modes as the number of suspension units is increased from 1 to 6. Three fits for the rejection
ratio are shown: for x1, 1/√(n); for α, independent of n; for z, 1/n, and (b) Log-log variation as calculated
by FEA of the rejection ratio for the spurious modes of a suspension incorporating intermediate frames
as the number of suspension units is increased from 1 to 6. Fits for the rejection ratio are shown: for x1,
1/√(n); for α, independent of n. .................................................................................................................. 92
Figure 3.25 Effect of intermediate frames on the vertical sag of proof mass under gravity, (a) multiple
spring set suspension under gravity with equal deflection of all spring beams as shown in the cross-
section view, as compared to (b) multiple spring set suspension with intermediate frame showing a
reduction in vertical sag under gravity due to the intermediate frame. ..................................................... 93
Figure 3.26 Reduction ratio for vertical sag under gravity for multiple spring suspension with
intermediate frames as compared to frameless suspensions. ..................................................................... 94
Figure 3.27 Effect of link width on rejection ratio of the suspension......................................................... 95
Figure 3.28 Maximum stress at the sharp corner of the linkage shown by the red pointer in the globe.... 96
Figure 3.29 (a) Link design with sharp corners, (b) Link design with corner filleting. ............................. 96
Figure 3.30 Schematic of spring element showing the end connector. ...................................................... 97
Figure 3.31 Distortion of the spring beam under an out-of-plane force. The forces acting at the opposite
ends of the elbow causes a torque which leads to an angular rotation of θ............................................... 98
Figure 3.32 Deflection of the spring beams depending on elbow rigidity, (a) completely rigid elbow with c
= 1 where the bending is that for a fixed-free cantilever and (b) completely flexible elbow with c = 4
where the bending is representative of a fixed-guided cantilever. ............................................................. 99
Figure 3.33 Plot of ωz and zsag,g from expression in Table 3.5. ................................................................ 101
Figure 4.1 Passivation and etch cycle in a typical DRIE step. (a) Isotropic etch of silicon in exposed
regions on the wafer, (b) Deposition of polymer CFn on all surfaces, (c) Preferential etching of polymer
from the bottom of the trench by vertically directed positive ions, and (d) Isotropic etching of the trench
bottom by fluorine radicals. ..................................................................................................................... 113
Figure 4.2 Model for inductively coupled plasma etch showing the positive and negative species in the
plasma, their energy angular distribution function (ADF) and the potential drop across the sheath which
accelerates the positive ions..................................................................................................................... 114
Figure 4.3 DRIE devices (a) Actuator-mirror structure fabricated on SOI (Prof. de Rooji, IMT), (b)
MicroSEM deflector structure (Prof. de Rooji, IMT), (c) A MEMS gyroscope fabricated on SOI (silicon-
on-insulator), and (d) Micro-machined intracellular needle [4.15], pillar etched using DRIE and then
post processed to achieve the sharp tip.................................................................................................... 116
Figure 4.4 (a) STS DRIE Kit [4.18], (b) STS ICP Multiplex process chamber Schematic (Source: [4.19]).
.................................................................................................................................................................. 118
Figure 4.5 A halo mask and its various components for a 2D mask designed on a mask plate for
transferring the pattern onto photoresist. ................................................................................................ 120
Figure 4.6 Sidewall quality for varying channel/trench width, (a) 40μm gap, (b) 60μm gap showing a
widening of the gap towards the bottom of the trench as the profile diverges away from vertical, (c) 80μm
gap shows the increase in profile shape divergence from vertical with bigger gap, (d) beams with gap
varying from 40-400μm show the increased breakdown of the sidewall with larger gap........................ 121
Figure 4.7 SEM image of 9 μm thick AZ9260 photoresist. ...................................................................... 122
Figure 4.8 (a) SEM of 5 μm thick oxide etched using CHF3, (b) SEM of damaged photoresist (AZ9260, 7
μm thick) after oxide etch......................................................................................................................... 123
Figure 4.9 Process flow schematic for through-wafer silicon etching using DRIE. ................................ 125
viii
Figure 4.10 Grass formation during silicon etch..................................................................................... 133
Figure 4.11 Reduced grass formation under increasing platen power, (a) 12W: trench is completely
covered with grass, (b) 14W: the centre of the trench bottom is clear but grass is still forming along the
sidewall, (c) 16W: clear trench bottom but the sidewalls have grass curtains, (d) 20W: the trench bottom
and the sidewall is etched cleanly without any grass formation. ............................................................. 134
Figure 4.12 (a) SEM of etch cross section showing etch lag between small and large trench, (b) Etch
depth vs. channel width (all dimensions in μm). ...................................................................................... 135
Figure 4.13 Process parameters optimized to reduce etch lag lead to grass formation in larger width
trench while smaller trenches etch cleanly. ............................................................................................. 136
Figure 4.14 SEM of top edge damage...................................................................................................... 137
Figure 4.15 Micromasking in etch channel resulting in oxford spires..................................................... 138
Figure 4.16 Scalloping (a) Scalloping on a sidewall, (b) AFM scan of top edge of sidewall shows scallops
(courtesy: Sanjay Vijendran), (c) SEM of large scalloping on trench sidewall suggesting high etch rate.
.................................................................................................................................................................. 139
Figure 4.17 SEM image of sidewall quality of 500 μm high sidewall. a,b,c show tapping mode AFM scan
of the top (a), middle (b) and bottom (c) of the trench sidewall. The roughness of the sidewall increases
from peak-to-peak 0.4 μm at the top of the trench sidewall to 0.7 μm at the middle of the sidewall and 1.4
μm at the bottom of the sidewall............................................................................................................... 140
Figure 4.18 Schematic of test structure showing region of interest (ROI). A tapered trench was selected to
examine sidewall surface quality with varying channel width. The individual structures (beams) can be
separated for ease of measurement (courtesy: Werner Karl). ................................................................. 141
Figure 4.19 Sidewall SEM of through-wafer deep reactive-ion etched beam. Zones of varying sidewall
morphology are labelled as: A. Overetched, B. Ion damaged, C. Overpassivated, and D. Striations. .... 142
Figure 4.20 Interferometric measurement of sidewall topography near narrow trench end. The ion
bombardment causes holes in the sidewall near the trench bottom where the re-entrant surface emerges
above the top edge.................................................................................................................................... 143
Figure 4.21 SEM picture showing the damage observed at narrow etch-gaps towards the trench bottom
caused by ion bombardment (Inset: zoomed, rotated view at a tilt angle of 85º) (courtesy: Werner Karl).
.................................................................................................................................................................. 144
Figure 4.22 Passivation build-up observed at wide etch-gaps leading to striations and therefore to rough
sidewalls................................................................................................................................................... 145
Figure 4.23 EDX analysis of the beam shows fluorine presence in the darker areas which were charging
up in the SEM, proving the hypothesis of CF2 deposition. (a) SEM image of the beam section analysed for
Fluorine content showing a higher concentration of carbon and fluorine at the top darker region of SEM
image than at the ligher part towards the bottom of the sidewall, (b) Elemental map of the sidewall
highlighting (red) the carbon and fluorine dominated regions, and (c) Linescan from the top edge of the
beam to the bottom edge showing the concentration of fluorine, carbon and silicon along the etch
direction. .................................................................................................................................................. 146
Figure 4.24 Profile shape of etched trench for various processing conditions in DRIE. (a) Negative
profile, (b) Positive profile, (c) Bowed profile. ........................................................................................ 149
Figure 4.25 Surface potential build-up due to flux imbalance of isotropic electrons and anisotropic ions
leading to lateral etching of the trench sidewalls. ................................................................................... 152
Figure 4.26 Notching damage at the bottom of a through-wafer etched feature. The wafer was bonded to
a handle wafer using photoresist. ............................................................................................................ 155
Figure 4.27 Charge distribution in deep trench and positive charge accumulation at Silicon-insulator
interface resulting in lateral undercutting at the trench foot (notching).................................................. 156
Figure 4.28 Process flow schematic for through-wafer etching using a metal backside coating to
eliminate notching.................................................................................................................................... 159
Figure 4.29 Schematic of etch setup with metal coating directly on the wafer backside and then bonded to
a handle wafer.......................................................................................................................................... 159
Figure 4.30 (a) Notching damage observed at the foot of a through-wafer etched structure; the wafer was
bonded to a handle wafer using thin photoresist. (b) The same structure on a wafer with a thin metal
coating on the backside; no damage at the foot of the sidewall is visible................................................ 160
Figure 4.31 Backside of an etched wafer. (a) Etched trenches with no backside metal coating, showing
extensive notching. (b) Etched trenches with backside metal coating, showing no visible notching. ...... 161
ix
Figure 4.32 Charge dissipation at metal surface in a silicon trench with a metal coating on the wafer
backside.................................................................................................................................................... 161
Figure 4.33 (a) Schematic of test setup with metal coated handle wafer glued to the device wafer at the
outside edges, (b) Schematic of the test setup with silicon handle wafer bonded to the device wafer at the
outside edge using photoresist. ................................................................................................................ 162
Figure 4.34 Schematic of experimental setup, (a) metal layer connected to the plasma at edges, (b) metal
layer completely isolated.......................................................................................................................... 162
Figure 4.35 (a) Damage due to partially floating metal layer, (b) Notching damage due to completely
floating metal layer. ................................................................................................................................. 163
Figure 4.36 Schematic of test setup with electroplated metal on device wafer backside. ........................ 164
Figure 4.37 Optical image of dry release process by peeling the metal layer from the wafer backside.. 164
Figure 4.38 SEM image of trench profile etched for 30 minutes using (a) LF plasma mode showing
thinning down of the structures towards the bottom as compared to, (b) HF plasma mode which
maintains profile anisotropy better and (c) scallop size for LF mode is 260nm × 818 nm compared to, (d)
HF mode scalloping which is 217nm × 770nm........................................................................................ 165
Figure 4.39 Schematic of experimental setup to investigate effect of good thermal contact between the
device and handle wafer, (a) Wafer bonded on the periphery such that when the structures are etched
through-wafer, the structures are thermally isolated from the handle wafer, and (b) the device wafer is
completely bonded to the handle wafer allowing the isolated structures thermal connectivity to the handle
wafer. ....................................................................................................................................................... 166
Figure 4.40 (a) Optical image of burnt photoresist due to thermal sinking problem – the reflective surface
shows normal photoresist, (b) photoresist is burnt in this case on the proof-mass which has bad thermal
connectivity through the long springs – the grey non-refelctive surface is burnt resist........................... 167
Figure 4.41 Full wafer mask design included 20×20 mm dies as well as 10×10 mm and 5×5 mm dies
with lateral suspensions. .......................................................................................................................... 172
Figure 4.42 Mask layout of the horizontal lateral suspension, the inset shows the spring connector and
the linkages with the filleting along with the packing pieces. .................................................................. 173
Figure 4.43 Mask layout of the Galperin configuration lateral suspension, the insets show the spring and
linkage along with the packing pieces...................................................................................................... 173
Figure 4.44 Silicon lateral suspension with six sets of springs and three intermediate frames fabricated by
DRIE. ....................................................................................................................................................... 175
Figure 4.45 Close-up of silicon lateral suspension with reduced mass frame and spring with cross-bar.
.................................................................................................................................................................. 175
Figure 4.46 Close-up of spring sidewall. ................................................................................................. 176
Figure 4.47 Etch profile shape, (a) spring beam cross-section showing an etch profile which is slightly
bowed towards the center, and (b) the profile shape of a cross-section of an intermediate frame showing a
similar bow like the spring beam. ............................................................................................................ 177
Figure 4.48 SEM of a silicon lateral suspension within a 5mm×5mm die............................................... 177
Figure 5.1 Slow-scan image of suspension under excitation, the edges of the spring and proof mass
appear as a sinusoid due to the rastering of the electron beam orthogonal to the direction of motion of the
proof mass. The sinusoidal edges show the fundamental mode of the suspension. .................................. 189
Figure 5.2 Overlap of the first spurious mode along the x-axis (ωx1) on top of the normal x-axis mode
(ωx). The x1 mode is visible as the higher frequency mode on the springs. The x mode is visible as the
low-frequency sinusoid of the proof mass and the springs....................................................................... 190
Figure 5.3 SEM of the suspension with the electron beam scanning along the Y-axis of the suspension, the
insets show the ωz and ωβ modes. Inset 1 shows the see-saw motion of the proof mass due to the rotational
mode (ωβ) about Y-axis of the suspension. The out-of-plane mode (ωz) is clearly visible at the center of
the proof mass in inset 2 as a constant amplitude oscillation. ................................................................. 191
Figure 5.4 Ringdown of suspension in SEM slow-scan. Measuring the time taken for the amplitude of
oscillations to drop to 1/e of its initial value gives the decay time constant (τ) of the system.................. 192
Figure 5.5 Plot of measured quality factor as a function of pressure for 10 Hz silicon suspension. ....... 197
x
Figure 5.6 SEM image of beam cross-section. The mask design was for a spring width of 24 μm which is
reduced to 22 μm at top edge of the spring during the etch and the middle section of the beam is thinner
than the top and bottom............................................................................................................................ 202
Figure 5.7 Undercut measurement using Zygo white light interferometer of spring cross-section. ....... 203
Figure 5.8 Measured lateral undercut for 24 μm and 31 μm wide beam. ............................................... 203
Figure 5.9 Beam cross section of an anisotropically etched beam, (a) rectangular cross-section, (b)
bowed cross-section. ................................................................................................................................ 204
Figure 5.10 Discretization of the spring cross-section into trapezoidal elements to calculate the bending
and torsional constants. ........................................................................................................................... 205
Figure 5.11 Etch profile of the spring - total spring thickness vs. etch depth.......................................... 207
Figure 5.12 Compensation for spring etch profile. (a) Ideal, (b) Actual after fabrication, and (c)
Compensated beam cross-section showing the addition of a rectangular area of width Δw to increase the
second moment of area Ix for the fabricated spring beam to specification.............................................. 209
Figure 5.13 Spring profile for 30 μm wide beam, (a) total spring thickness vs. etch depth, (b) SEM of the
beam cross-section. .................................................................................................................................. 211
Figure 6.1 Two implementation of capacitive sensing, (a) distance (gap) changing: the distance between
the plates is varied and (b) overlap area changing: the overlap area between the plates is varied with
position..................................................................................................................................................... 218
Figure 6.2 Block diagram of the microseismometer. The four major functional units are the mechanical
suspension, capacitive transducer which converts the mechanical motion of the proof mass to electrical
signal, feedback controller which drives the electro-magnetic actuator to keep the proof-mass at
equilibrium. .............................................................................................................................................. 219
Figure 6.3 Cross-sectional view of the microseismometer showing the Capacitive array transducer with
the drive electrodes on the proof-mass and a set of overlapping output electrodes on the top capping die.
.................................................................................................................................................................. 221
Figure 6.4 (a) Configuration of the drive and output electrodes, and (b) alternate drive electrodes are
driven out-of-phase by a +DRIVE and –DRIVE square-wave pulse train. The output electrodes provide a
differential output signal. ......................................................................................................................... 222
Figure 6.5 Perspective view of the magnetic circuit. The planar coil on the proof mass is enclosed by
magnetic field from the pair of magnets on either arm of the coil. The feedback force is generated by
running a current through the coil. .......................................................................................................... 224
Figure 6.6 Proof-mass die showing the mechanical suspension with the circuit fabricated on top of the
suspension to control and measure the displacement of the proof-mass.................................................. 225
Figure 6.7 Process schematic for the complete suspension fabrication with metal and insulator layers.227
Figure 6.8 (a) Contact resistance of pads prior to anneal, and (b) Post-anneal I-V curve for contact pads
showing a contact resistance of 200 ohms. .............................................................................................. 228
Figure 6.9 Contact pad post-anneal showing a rough surface, the brighter islands is the Au layer which
dissolves in the underlying NiCr layer. .................................................................................................... 228
Figure 6.10 Mask layout of the metal-1 layer showing the ground-plane, interconnects for the coil and
traces. Metal 1 is 30nm Cr and 200nm Au and is patterned by wet chemical etching............................. 229
Figure 6.11 Metal-1 interconnect underneath the insulator layer for making the connection between two
segments of metal-2 across the edge seal layer........................................................................................ 230
Figure 6.12 Photoneece, a photo-definable positive polyimide is spin coated on the wafer and
lithographically patterned. The developed and cured photoneece layer has sloping sidewalls suitable for
running thin metal traces across the edge of the insulator pads (courtesy: Werner Karl). ..................... 231
Figure 6.13 Metal-2 mask layout showing the coil, traces, edge seal layer, pickup electrodes and the
interconnect pads for connectivity between the proof-mass die and capping die. ................................... 232
Figure 6.14 (a) Electroplated Cu in the photoresist mold, and (b) Electroplated Cu/Au after removal of
the photoresist mold and etching of the initial Cr/Cu seed layer. ............................................................ 232
Figure 6.15 Optical micrograph of the circuit on top of the proof-mass die prior to DRIE. ................... 233
Figure 6.16 (a) A complete proof-mass die, and (b) Closeup of the metal traces on a spring (courtesy:
Werner Karl). ........................................................................................................................................... 234
xi
Figure 6.17 Exploded view of the capping and proof mass dies. The DT (Differential transducer) die
contains the drive LCAT electrodes. The cavity die is underneath the suspension die and is required for
vacuum encapsulating the suspension...................................................................................................... 235
Figure 6.18 Schematic steps for metallization of the glass wafer and the subsequent sand powder blasting
to etch the cavities.................................................................................................................................... 236
Figure 6.19 (a) Mask layout for the top capping wafer metal layer, and (b) Top capping die with the
metal layer................................................................................................................................................ 237
Figure 6.20 On the top capping wafer, the side containing the circuit is powder sand blasted first to etch
the cavities, then the wafer is turned over and with a different decal dicing lines are etched through the
glass wafer. This allows for easy release of the dies (courtesy: Trevor Semple). .................................... 238
Figure 6.21 Magnetic circuit for the microseismometer comprising of magnets, yoke and coil. The yoke
closes the circuit between the set of magnets on either side of the die sandwich. The yoke under the
magnet smoothes the magnetic flux lines and creates a larger linear region for the coil to move within.
.................................................................................................................................................................. 239
Figure 6.22 Location of the coil within the magnetic field. For maximum deflection force the coil should
be in the linear region of the magnetic field............................................................................................. 239
Figure 6.23 Magnetic flux density at the coil for varying level of bevelling on the yoke. As expected a
sharper bevelling leads to a higher magnetic flux density but the linear range is reduced. .................... 240
Figure 6.24 Cross-sectional view of the microseismometer assembly showing the yoke, magnets, top and
bottom capping dies, proof mass die and electrode pads for signal readout and feedback. .................... 241
Figure 6.25 Schematic of the contact and seal between the top capping die and the proof mass die...... 242
Figure 6.26 (a) Solder electroplating, (b) Solder jetting, and (c) Solder ball placement (courtesy: Trevor
Semple)..................................................................................................................................................... 242
Figure 6.27 Mechanism of solder ball transfer using grooves on wafers, the grooves are etched using
DRIE. The solder balls are transferred to the top capping die. The top capping and proof-mass dies are
then aligned and bonded. ......................................................................................................................... 243
Figure 6.28 Glass frit bonding to form seal between the bottom capping die and the proof mass die. ... 244
Figure 6.29 Schematic of the hermetically sealed cavity enclosing the mechanical resonator. .............. 245
Figure 6.30 A fully assembled microseismometer, for size reference a British penny is placed next to it.
.................................................................................................................................................................. 248
Figure 6.31 Experimental measurement setup for measuring the resonant frequency and Q-factor of the
prototype microseismometer. ................................................................................................................... 249
Figure 6.32 Close-up of the electronic conditioning unit containing the microseismometer................... 249
Figure 6.33 Response plot for the suspension. The coil generates a current when driven under the
external magnetic field. The peak in the response curve corresponds to the normal resonant modes. The
first peak is at a frequency of 12.35 Hz.................................................................................................... 250
Figure 6.34 Ringdown test of the suspension. The current generated in the coil reduces as the vibration
decays....................................................................................................................................................... 251
Figure 6.35 Minimum intensity exposure for transportation of packaged die using a combination of truck,
rail and air transport. .............................................................................................................................. 253
Figure 6.36 Ling Dynamic Shaker at AOPP, University of Oxford used for doing shock and vibration
analysis of packaged suspension.............................................................................................................. 254
Figure 6.37 Fracture of the spring at the linkage. ................................................................................... 254
Figure 6.38 Frame and proof-mass damage due to collision at the corners. .......................................... 255
Figure 6.39 Redesigned linkages to reduce the stress concentration....................................................... 255
Figure 6.40 Damper designs to avoid frames and proof-mass collision, (a) Cavity damper, (b) Comb
finger dampers, (c) Spring constraining tabs, and (d) Frame spring damper. ........................................ 256
Figure 6.41 Damage to the frames and proof-mass at Ariane super vibration levels for packaged
suspension. ............................................................................................................................................... 257
Figure 6.42 Friction damage on proof-mass due to cavity die support fingers. ...................................... 257
Figure 7.1 (a) 2D model of a fully electrostatic inertial sensor, (b) a fabricated geophone.................... 265
xii
Figure 7.2 Damping structures to increase damping when the suspension is at its maximum deflection.
.................................................................................................................................................................. 266
Figure 7.3 Some prototype devices fabricated using the MIMD (Metal-Insulator-Metal-DRIE) process,
(a) geophone sensor, 5mm × 5mm, (b) μSEM components, (c) a tuning-fork gyroscope, (e) comb drive
based suspension...................................................................................................................................... 268
Figure 7.4 Fabrication sequence for the silicon mold, substrate preparation, electroplating, and release.
.................................................................................................................................................................. 269
Figure 7.5 SEM image of a silicon mold after DRIE (inset shows the mold with a 10 μm thick PMMA
release layer)............................................................................................................................................ 271
Figure 7.6 SEM image of a molded metal test structure electroplated to 236 μm height. The sidewall is
vertical except where the structure is anchored to the substrate. Reflow of the release layer forms a hump
at the mold-substrate interface leading to profile variation..................................................................... 271
Figure 7.7 Suspended metal structures molded using the fabrication steps depicted earlier. The plated
structure is 50 μm high and is suspended 14 μm over the base plate. ..................................................... 272
Figure 7.8 Process schematic for single-mask DRIE device fabrication. ................................................ 273
Figure 7.9 Process schematic of a SOI replacement through-wafer process. ......................................... 274
xiii
List of Tables
Table 4.1 Comparison of oxide and photoresist as a masking layer for DRIE. ....................................... 124
Table 4.2 Comparison chart for various wafer bonding techniques. ....................................................... 128
Table 4.3 Effect of process parameters on the etch characteristics. ........................................................ 169
Table 4.4 Etch Parameters for through-wafer etching............................................................................. 171
Table 5.1 Vibration modes of silicon suspension with 24 μm wide springs. The suspension has two set of
springs with one intermediate frame. ....................................................................................................... 194
Table 5.2 Resonant modes for a 24 μm wide spring suspension. Comparison between analytical,
numerical and experimental data incorporating beam profile shape. ..................................................... 208
Table 5.3 Modal frequency for suspension with 24 μm spring which is increased to 30 μm to compensate
for the reduction in resonant frequency due to etch profile shape. .......................................................... 212
Table 5.4 Resonant modes for 1 spring set suspension with 24 μm wide spring...................................... 212
Table 6.1 Suspension dimensional parameters for damping calculation for microseismometer. ............ 246
xiv
Acknowledgements
Even though this thesis bears my name it would not have been possible without the
started with a course taught by Prof. Andrei Shkel at University of California, Irvine. If
it were not for those glossy posters exhorting the exciting field of micro sensors and
actuators I would perhaps still be an Aerospace engineer. The real introduction to the
world of MEMS was while working with Dr. Saroj Sahu, Dr. Chris Kenney and Dr. Jim
Aroyan at Jasmine Networks and later at MuSquared, Inc. Their support, patience and
friendship have been the most crucial in not only expanding my knowledge of
It was perfect timing for me to hop over the pond to London when Dr. Tom Pike offered
at Imperial College after seeing his performance in the comedy play on my first day. I
am extremely grateful for his guidance and support throughout the research. I am also
grateful to Kinemetrics, Inc for sponsoring this research and to Ian Standley for the
The Optical and Semiconductor Devices group is an extremely dynamic group full of
talented people. In particular I would like to thank Dr. Munir Ahmed, who with his
the processing nightmares. Dr. Tom Tate for keeping the cleanroom running and Dr. John
Stagg for his help with processing. I thank Prof. Richard Syms for his tireless effort at
xv
The effort involved in research was made all the more enjoyable by the excellent
company provided by the many friends in the group. London would be very
complicated but for the navigational skills of Werner who always knew the best pub
whatever be the area. I am extremely thankful for his help and guidance throughout the
research. The many evenings I have spent in the company of Ariel and Werner not just
discussing work but pondering over the philosophy of it all has kept me sane. Ariel with
his ability to invent, cook, paint, and still managing to be a doting dad is an inspiration.
The enjoyable dinners at Trevor’s place were always great while getting to see (not
play) the fantastic guitars he makes. Hanna’s parties were always great to test one’s
drinking abilities and her tennis returns can floor the best. Sanjay is always a lesson in
planning. Lunch would not be the same without Robert (and his theories on cave
people), Kai (and his projects) and the discussions which always lead to the two great
topics. It has only gotten better with Michail (and his myths), Rob, Wen, Anshu, Anisha
and Vinita. Economic discussions would not have been same without Anke and her
scepticism of my “economist” data. Michael who was always ready to help and advise
be it research or business. Migel, Nick, Youngki and Lek – members of the lab who
now live in various parts of world and with whom I had some great time and memories.
Yun, Li, and Justin for the fantastic badminton games, and all the other wonderful
members of the lab. It certainly would have taken longer to complete this research if it
Outside of the group, I am deeply thankful to my friends here in London and elsewhere
with whom I have enjoyed so much. Srini, who accompanied me on many trips and
never lost his cool even when I got him stuck in the middle of two glaciers and three
xvi
lakes in Iceland, and who so patiently explains the real economic world. Prabhala for
his constant motivation, and insights. Mr. Alun Parry for his 2 a.m. coffees and endless
discussions. Arnab, Ashish, and Paretha for their company and many others whose
Finally and most importantly I would like to thank my parents who have influenced me
throughout my life and my brothers Onkar, Anil, and Ravi who have always been there
for me and encouraged me all the way. It would not be possible for me to have come
xvii
This thesis is dedicated to
papa, mummi
and to
and to
xviii
One
1 Introduction
In this chapter we present the motivation for designing and fabricating a single crystal
ability to detect extremely low-level vibrations. The microseismometer will fill the gap
1
Introduction
1.1 Background
Mankind has always been interested in understanding all aspects of the Earth and over
the centuries we have expanded our knowledge of the land masses, water bodies and
atmosphere by direct observation and interaction with them. The internal structure
however has always been a mystery which started unravelling only after the advent of
seismology which is the study of propagation of seismic waves through the surface and
to Chinese scientist and philosopher Chang Heng who inventing a seismoscope in 132
A.D. (Figure 1.1)[1.1, 2]. This was a primitive but ingenious device which registered
the arrival and direction of the earthquake. The device consisted of a pot with a
pendulous arrangement within and on the outer surface had eight dragons with spheres
in their mouth and eight toads under them. Depending on the direction of the seismic
tremor the appropriate ball would fall in the toad’s mouth allowing observer to infer the
direction of the tremor. The design influenced mercury and other liquid based
seismoscopes in Europe in the eighteenth century and the first seismograph which could
2
Introduction
Figure 1.1 Wang Chen-To’s visualization of Chang Teng’s 132 A.D. design of a seismoscope. Image
Seismology until the nineteenth century was mostly devoted to the study of earthquakes
which ultimately led to the first systematic study of the Earth’s seismicity. With the
analysis of the seismic wave propagation led to the first understanding of the internal
structure of the Earth. Analysis of the travel time of teleseismic (long distance) body
waves through the Earth enabled R. D. Oldham (Assam, India 1897) [1.2] to propose a
model of the Earth with a liquid core whose outer boundary was calculated by Beno
3
Introduction
in 1909 [1.1] and Inge Lehmann deduced the existence of a solid inner core in 1936
[1.4]. A whole new understanding of the Earth’s internal structure was being made
In the 1920s and 1930s exploration to find oil and gas reserves beginning in Texas and
find and map underground reservoirs [1.6]. This was also the first significant
Seismology received a significant boost in 1950s and 1960s as a means to detect nuclear
explosions [1.1, 7]. This allowed the world powers to keep an eye on their adversaries
by detecting their nuclear tests. The necessity to discriminate a nuclear explosion from
earthquakes and to ascertain the location and strength of the explosion resulted in the
establishment of new world wide network of seismic stations like the WWSSN (World
instrumentations.
Seismometers are (usually) based on the principle of inertia. A heavy mass is loosely
coupled to the ground, for example by suspending from a thin wire (Figure 1.2). The
motion of the Earth is only partly transferred to the mass. The relative displacement
4
Introduction
Figure 1.2 Principle of seismometer: the mass is almost stationary due to inertia while the frame
estimate of ground motion only at the start of the shaking. Emil Wiechert introduced the
viscously damped seismometer in 1898 [1.4, 8] which could record ground motion
continuously over an extended period, thus recording not only the major earthquakes
but also the after-shakes. Significant progress was made at the beginning of the
Prince Boris Galitzin in 1906 in Russia [1.2]. The electromagnetic seismometer has a
coil attached to the mass of the pendulum that moves in a magnetic field created by
magnets attached to the reference frame. Upon ground motion, the reference frame
moves, which creates a varying magnetic field causing a current to flow through the
coil. The electric current generated in the coil is passed to a galvanometer whose
magnification of the ground motion compared to the previous seismometers. The new
5
Introduction
can record ground motion over a large range of periods made possible by a feed-back
circuit that extends the response to very long periods. Some of the commercial
Figure 1.3 State-of-the-art commercial seismometers, (a) Streckheisen STS-2, (b) Guralp CMG-
40T, (c) Guralp CMG-3T without outer housing, (d) Sercel L-4, (e) Geotech KS-2000 borehole
seismometer, and (f) Nanometrics Trillium 120P. These are all big, heavy and expensive.
Streckeisen [1.9] manufactures some of the most sensitive seismometers, their STS-1
and STS-2 designs were some of the first widely available broadband seismometers.
STS-1, now out of production was a single axis seismometer (can measure vibration
only along one axis) with a mechanical period of 360 seconds. STS-2 (Figure 1.3a) is a
three component seismometer which can record vibrations along all three axes; it has a
mechanical time period of 120 second. CMG-40T (Figure 1.3b) and CMG-3T (Figure
6
Introduction
1.3c) are Gűralp’s [1.10] three component seismometers. CMG-3T competes in the
same market as STS-2 and has comparable specifications. Sercel [1.11], Geotech [1.12]
and Nanometrics [1.13] (Figure 1.3d-f) are some of the other key players in the
seismometer market and have a wide range of offerings. Table 1.1 compares a range of
seismometers from various vendors. As we can see from Table 1.1 and Figure 1.3
conventional state-of-the-art seismometers are large (~125,000 cubic cm), heavy (~10
kilograms), and expensive (~$50k-$100k). They are also power hungry (few watts),
tricky to install and expensive to maintain. This presents a major obstacle to their
(cost consideration), building remote large field networks for earthquake monitoring or
oil prospecting (cost and power consideration) and space seismology (size, power and
weight consideration).
7
Introduction
The size and weight restrictions of conventional seismometers limit their usability in
many areas, which has led, in recent times to the development of miniaturized
from Sercel and Oyo-Geospace [1.16]. The key concern while miniaturizing
seismometers is the sensor’s ability to resolve vibrations from the ambient noise. The
noise can have two origins: noise generated in the instrument and the “real” seismic
noise of the Earth vibrations. Conventional seismometers like STS-1, STS-2 and CMG-
3T have instrument noise well below the seismic noise but for miniaturized
seismometers it is important to design the instrument to keep the instrument noise below
the Earth noise in its range of operation. As we see in Table 1.2 the self-noise of the
seismometers are increasing with decreasing size. Self-noise of the instrument gives a
measure of the smallest signal that the sensor is able to distinguish from background
noise.
Table 1.2 Miniaturized seismometers. φ is used to denote the diameter of the sensor.
Many factors contribute to the Earth noise, including human activity, wind, ocean,
temperature fluctuations, solar and lunar interactions. Man made noise is generally from
traffic and machinery and is high-frequency (2-4 Hz). Wind sways structures causing
high-frequency noise though sometimes with very large structures it generates low-
8
Introduction
frequency vibrations. Oceans are the source of most widespread seismic noise seen
globally; it is significantly higher in coastal areas than inland. The waves and tides
cause shorter period vibrations in the 0.1-0.2 Hz range. These factors are constantly at
Figure 1.4 The USGS New Low Noise Model [1.17], here expressed as Noise Estimated Acceleration
in m/s2/√Hz of ground motion relative to frequency. Mechanical noise levels are plotted for the
proposed microseismometer.
The lowest noise levels on the Earth are represented by a USGS (United States
Geographical Society) New Low Noise Model (NLNM) [1.17] (Figure 1.4). The curves
9
Introduction
worldwide. These Peterson-curves are a useful reference for predicting the noise levels
of a seismic station and to quantify the usability of a seismometer to resolve the smallest
seismic motion. The figure shows the target microseismometer mechanical noise floor
An important consideration while designing seismometers is the Earth noise. The key
requirement is creating sensors sensitive enough to detect the smallest seismic signals,
given that noise sets a limit to the level of detection. The instrument noise is due to both
the electronic noise and the mechanical noise of the suspension. The electronic noise is
generally well quantified for the amplifiers and the feedback circuit and can be designed
structure which is due to viscous damping from surrounding air molecules. By vacuum
encapsulating the resonating structure the viscous damping from air can be reduced.
At the lower end, the fundamental limit to the self-noise or noise floor is set by the
Brownian motion of gas molecules hitting the proof mass. Usher [1.18] shows the
ω0 (1. 1)
NEA 2 = 4k b T
mQ
units of m/s2/√Hz and also in terms of gravity as g/√Hz where the relation is 1 g/√Hz =
9.81 m/s2/√Hz.
10
Introduction
From equation 1.1 we can see that the three critical parameters for a low-noise sensor
are large mass (m), long period (1/ω0) and high quality factor (Q). Quality factor (Q) is
the measure of the energy loss in the suspension and the higher the Q, the better is the
very large mass and an extremely long period which reduces the mechanical noise of the
suspension. With miniaturization, the weight of the suspension decreases and it also
becomes harder to fabricate low resonant frequency suspensions. Figure 1.5 compares
the NEA and natural frequency of the suspension for a variety of vibration sensors from
kg mass on leaf springs which gives it a time period of 360 seconds and is installed in
vacuum to give it one of the smallest noise floor. STS-2 and CMG-3T have a 13 kg and
14 kg proof-mass respectively and a time period of 120 seconds allowing them to reach
less than 10-10 m/s2/√Hz NEA. Sercel L-4 and Oyo-Geospace HS-1 with their much
smaller mass of 0.5 kg and 22.7 g and resonant frequency of 1 Hz and 4.5 Hz have
available and from research. Accelerometers are typically high-g vibration measuring
sensors and have significantly higher resonant frequency. The accelerometers shown in
Figure 1.5 are fabricated using semiconductor processing techniques and have
extremely low mass. This leads to extremely high mechanical noise rendering these
accelerometers unusable for detecting small seismic signals. To improve the resolution
11
Introduction
grade accelerometers are being manufactured commercially and have been the subject
of intensive research over the past few years. These accelerometers offer significantly
lower noise floor, some of them finding applications in seismic sensing (ex. Sercel
Figure 1.5 Comparison of noise and resonant frequency of seismometers and a variety of
Figure 1.6 compares the self-noise of seismometers and accelerometers with respect to
their volume. We can see the strong correlation between the size (related to the mass) of
the sensor and its noise floor as predicted from equation 1.1. Different groups of sensors
discussed upto now are typically low-noise low-frequency and high-mass sensors.
12
Introduction
lower self-noise than high-g accelerometers and have generally higher volume and
hence mass. Some navigational grade accelerometers like ones developed by Yazdi et
al. [1.21], Bernstein et al. [1.22] and Suzuki et al. [1.23] have similar or lower volume
than high-g accelerometers – this is due to the high quality factor of the material used to
make the suspension or the extremely low resonant frequency of the suspension as with
Figure 1.6 Comparison of noise and volume for traditional seismometer and MEMS accelerometers
13
Introduction
can reduce the self-noise even further. Most of the current crop of accelerometers are
batch fabricated using semiconductor processing techniques and are collectively known
some of the industry standard ones are presented in Table 1.3. Honeywell QA-3000
(Figure 1.7a) uses a suspension made of quartz which is a very high-Q material. The
low mass of the suspension and the high-frequency of the suspension are compensated
significantly by the high-Q of the suspension to yield a 0.1 μg self-noise sensor. The
Sercel DSU3 has a lateral suspension made of polysilicon yielding an extremely low
mass of the suspension and a high resonant frequency but the high-Q of the vacuum
packaged device reduces the self-noise floor to 10 ng level. The SEM of the suspension
for the Sercel DSU3 is shown in Figure 1.7b. It bears noting that one of the major
vacuum environment reducing the noise-floor to the Brownian noise of the suspension.
Figure 1.7 Commercial MEMS accelerometers with sub μg noise levels, (a) Honeywell Q-flex
QA3000 with an exploded view of the sensor and its suspension, and (b) Sercel/Tronics GPU3 with
14
Introduction
Table 1.3 Specification of MEMS accelerometers. φ represents diameter of the sensor body.
Analog Devices ADXL103 5×5×2 mm <1 gram 1.10×10-3 2.2 kHz ~few μg
[1.25]
Analog Devices ADXL05 5×5×2 mm <1 gram 5.00×10-3 1.2 kHz 1.86 μg
Various research groups have also designed low-noise suspensions for use as an
accelerometer. Figure 1.5 and Figure 1.6 show the noise level, natural frequency and
size of some of the notable accelerometers from various research groups. Roylance and
Angell [1.26] developed one of the first silicon micromachined accelerometers in 1978
at Stanford with a 0.02 mg mass suspended from a silicon cantilever (Figure 1.8e,f) to
noise floor made out of silicon, this was similar to the Roylance suspension in terms of
having a large proof-mass over thin flexures (Figure 1.8a,b). Rockstad et al [1.27]
μg/√Hz and 0.01 μg/√Hz near the resonant frequency of the suspension. Silicon has
been a very popular material of late due to the availability of standard equipments to
machine it easily and also has excellent material properties, though recently a low-noise
Suzuki et al [1.23] which showcases the advantage of using polymer to build very low
15
Introduction
Figure 1.8 MEMS accelerometers developed by research groups, (a,b) A 1.0 μg/√Hz self-noise
accelerometer developed by Bernsetin et al. [1.22] at Charles Stark Draper Laboratory, (c,d)
Parylene based low-noise accelerometer developed by Suzuki and Tai [1.23] at University of Tokyo,
and (e,f) one of the very first miniaturized accelerometer developed at Stanford university by
possible to develop sensors with nano-g or even sub nano-g self-noise by choosing a
A wide variety of materials from silicon to quartz to polymer have been used to
fabricate accelerometers some of which have sub-μg noise. The choice of material is
dictated more from the availability of tools to pattern the material and by the suitability
of the material to create a low-loss suspension. Even though the mass of a MEMS
imperative to maximize the mass to the extent possible. This requires a fabrication
technique which can create thick structures but that can lead to thick flexures increasing
16
Introduction
the resonant frequency of the suspension. High aspect ratio structures can solve the
problem of both a large mass and slender flexures for low resonant frequency. Recent
Deep Reactive Ion Etching based on the Bosch [1.28] process allows the flexibility to
Silicon has excellent mechanical properties [1.29] and a wide variety of standard
fabrication processes which are exploited to make micromechanical devices. Table 1.4
Stainless steel is a popular material used for manufacturing some of the conventional
seismometers, as we can see from the table it has similar modulus of rigidity as silicon
but has three times the density which could be useful for making a high-mass
increases material damping as energy is lost due to grain boundary slips, dislocations,
interface slips etc. resulting in low quality factor (Q) of the suspensions made from
them. Single crystal silicon is a crystalline material with silicon atoms in very stable
energy levels eliminating losses through grain boundary slips etc. The loss mechanisms
internal friction (dependent on defects in crystal lattice) which are much lower
compared to losses through grain boundary slips thus enabling extremely high Q
resonators, at times as high as 600,000 [1.30]. Silicon also has a much higher yield
strength compared to most metals and polymers. Quartz has significantly higher yield
strength and is also a very low-loss material but is harder to pattern due to lack of
17
Introduction
The high Young’s modulus and large yield strength are extremely attractive for
mechanical systems. In addition, low crystal defect allows fabrication of high-Q and
high-reliability devices. Silicon also has a high melting point (1683 C), is radiation
crystal planes. Silicon is a brittle material so failure in silicon devices is generally due to
Utilizing semiconductor processing methods the devices can be batch fabricated out of
Deposition) deposition of polysilicon, RIE (Reactive Ion Etching), wet isotropic and
anisotropic wet etching is supplanted by specialized MEMS tools like DRIE (Deep
Reactive Ion Etching) allowing increased freedom to the suspension designer. Figure
1.9 shows suspensions fabricated from polysilicon and single crystal silicon by various
fabrication methods.
18
Introduction
Figure 1.9 (a) Lateral suspension used in Analog Devices ADXL series accelerometers fabricated
by LPCVD deposition of polysilicon and then RIE etch [1.35], (b) Vertical suspension created by
anisotropically etching silicon by KOH (Potassium hydroxide) [1.21], and (c) Leaf spring part of a
lateral suspension fabricated by DRIE of silcon bonded to an underlying silicon substrate [1.36].
Polysilicon suspensions are fabricated by surface micromachining where thin layers are
deposited by LPCVD and subsequently either reactive ion etched or chemically etched
to fabricate the suspension. These suspensions have extremely low mass due to thin
layers and generally have high resonant frequency, a typical example is the Analog
1.9a) with a resonant frequency of 2.5 kHz and a mg/√Hz noise level.
ammonium hydroxide) is used to create single crystal silicon structure with (111) or
(110) limiting plane. This particular fabrication technique imposes severe limits on the
device geometry. It is mostly suitable for fabricating out-of-plane suspension with large
proof-mass like the accelerometer developed by Yazdi et al. [1.37](Figure 1.9b) which
has 885 Hz suspension with 2.1 mg mass and 0.2 μg/√Hz noise floor.
create high-aspect ratio resist molds for electroplating. The metal structures thus created
19
Introduction
can be used for injection molding plastic components. It is a suitable technique for
creating high-aspect ratio structures in metals (like Nickel) and plastics but the need for
RIE (Reactive Ion Etching) has allowed new freedom in suspension fabrication by
allowing vertical etched structures in silicon. Lately, the requirements for structuring of
silicon as a mechanical structure is moving towards high aspect ratio structures, from
silicon is the Bosch time multiplexed etching technique [1.28] also known as Deep
Reactive Ion Etching (DRIE), where etching and passivating cycles are alternated
during the process. The method has been used to fabricate high-aspect ratio devices like
through-wafer interconnect [1.40]. Figure 1.10 shows some devices fabricated using
DRIE of silicon.
Figure 1.10 DRIE devices (a) Endevco accelerometer [1.41], (b) Extended travel range
microactuator [1.42].
In this thesis we use DRIE for fabricating the lateral silicon suspension which forms the
backbone of the microseismometer. Silicon lateral suspensions have been used in many
20
Introduction
applications like clockwork springs for storing energy, accelerometer, gyroscopes, shear
1.6 Summary
A brief history of seismology and seismic instrumentation was presented and the
such as Streckheisen’s STS-1, STS-2, Guralp CMG-3T were discussed and their key
Analog Devices ADXL05, Honeywell QA-3000 and Sercel DSU3. The ability of the
conventional seismometers to resolve the smallest seismic signals was attributed to low
dependent on the mass, quality factor and time period of the suspension. Maximizing
these three factors leads to a low noise floor for the suspension. With miniaturized
be very high due to the low-mass and high frequency of the suspension. To overcome
high quality factor material like silicon to fabricate the suspension. The availability of
anisotropic fabrication tools like DRIE was shown to make it possible to fabricate
suspensions with high aspect ratio leading to long period, large mass and high-Q thus
measure and control the resonator will enable substantial reduction in size, weight, cost,
21
Introduction
repeatable across devices. These attributes will lead to a bigger envelope of usage of
seismometry in areas that are currently outside its purview due to prohibitive cost, size
these MEMS devices will be lower than the macroscale versions from Streckheisen, and
Guralp, they can still be many orders of magnitude better than comparable miniaturized
alert networks.
3. Oil and gas: Sensor networks for reservoir mapping, Borehole status monitoring, Oil
prospecting.
the internal structure. Seismometer payloads have been sent on missions to Mars
(Voyager) and are planned for future missions to Mars and Moon by ESA and NASA.
5. Explosion detection for anti-terrorist operation and enforcing NPT (Nuclear non
Proliferation Treaty).
Many of these applications are currently either not undertaken or use low-resolution
therefore expected that a MEMS seismometer will open a spectrum of applications for
22
Introduction
Major Contributions:
1. Generic analytical and parametric numerical model for a lateral suspension with
multiple springs and frames. The model incorporates the effects of fabrication process
on structure geometry and is applied to the silicon suspension presented in this thesis,
predicting the modes of the fabricated suspension to within 5% accuracy. The model
can be modified for other suspension designs with folded cantilever flexures.
significant amount of research and development of devices using DRIE, but through
etching of 525 μm to 1 mm thick wafers is uncommon. This work covers all aspects of
deep silicon etching. Wafer thick suspensions with thin flexures have been successfully
fabricated to achieve one of the lowest resonant frequency (10 Hz) single crystal silicon
suspensions.
lateral silicon suspension developed in this thesis has sub-nano-g/√Hz self-noise and a
low resonant frequency (10 Hz). The suspension self-noise and volume are two to three
available and reach the self-noise levels of some of the best conventional seismometers
while still having a volume three to four orders of magnitude lower. The
23
Introduction
actuation modules besides the mechanical resonator. It is not within the scope of this
thesis to cover sensing and actuation though a brief summary is provided to present a
organisations and therefore the thesis also contains work which was carried in a team
effort with major contributions from the author. Wherever this has happened it has been
pointed out in the text. In general, the author’s major field of work has been the
The work carried out in this thesis has been published in various international
24
Introduction
Sweden, 2005.
Microengineering, 2007.
This thesis describes the design and fabrication of a low resonant frequency low-noise
silicon lateral suspension. The design and analysis cycle to hit performance benchmark
The design and feasibility study of a lateral suspension for seismic sensing is presented
suspension and presents a design flow to satisfy the performance criteria. Chapter 4
deals with the fabrication of silicon suspensions. Issues with DRIE and through-wafer
etching are discussed and a fully functional silicon lateral suspension is fabricated.
Chapter 5 analyses the fabricated suspension and compares the experimental results
25
Introduction
behaviour from the model is examined, explained, and the model refined to
accommodate fabrication deviations from the ideal geometry. Chapter 6 looks at the
microseismometer built around the silicon lateral suspension. The processing steps
required to build sensing and actuation components are discussed, and packaging
considerations are presented. Chapter 7 explores the potential use of the demonstrated
silicon lateral suspension, and extends the modelling and fabrication methods described
in the thesis to the manufacturing of other devices. The appendices contain the
suspensions, input files for finite element analysis, and the fabrication process
description.
The content of the thesis has been divided into separate chapters on design, analysis and
fabrication even though in reality they feed into each other and a fully functional device
is the outcome of multiple iterations between the design and fabrication cycles. We start
with a model for the lateral suspension and optimize its design based on the desired
based on some initial presumption; the process is then modified to achieve the etching
characteristics required to achieve the final geometry. The fabricated suspension is then
tested and any discrepancy in results is attributed to either design or fabrication steps.
The models for the design and fabrication are then updated to achieve the final device.
26
Introduction
1.9 Bibliography
27
Introduction
28
Introduction
29
two
2 Suspension Design and Feasibility Study
In this chapter we set the noise floor of the suspension. The parameters affecting the
self-noise of the suspension are then analysed. We then develop a generic lateral
suspension in section 2.3 and explore the design space in section 2.4. The fabrication
constraints are presented in section 2.5. The ability of the lateral suspension designed to
30
Suspension Design and Feasibility Study
Design of the micromechanical suspension for seismic sensing requires us to set a target
noise level for the mechanical suspension. The target noise level for the device should
be below the lowest signal that is to be measured using the device. The overall noise
2
NEAdevice = NEAmechanical
2
+ NEAelectronic
2 (2.1)
Generally the optimal design for lowest noise occurs when the electronic and
mechanical noise are similar [2.1]. The target total noise for the microseismometer
(NEAdevice) is set to 0.5 ng/√Hz (4.9×10-9 ms-2/√Hz) for the seismometer giving a
mechanical noise NEAmechanical of 0.35 ng/√Hz (3.43×10-9 ms-2/√Hz). This allows the
as shown on Figure 1.4. As we discussed in section 1.4 the NEA of the seismometer is
ω0 (2.2)
2
NEAmechanical = 4k b T
mQ
From equation 2.2 it is clear that low resonant frequency, high quality factor and a large
mass are prerequisite for low-noise mechanical suspension. In the next section we will
The important figures of merit for a lateral suspension to be used for seismic sensing are
the resonant frequency, quality factor, and cross-axis sensitivity (the separation of
fundamental and spurious modes). These are a partial set of parameters pertinent to a
suspension design for vibration sensing, a full set of all the relevant parameters for
31
Suspension Design and Feasibility Study
vibration sensors can be found in IEEE standard 528-2001 for inertial sensor
technology.
Resonant frequency is the property of a system where the system retains input energy
with minimum loss. By driving the devices at their resonant frequency the effect of
small forces is enhanced along with the device signal-to-noise ratio (SNR) [2.3].
amplitude vibration even from low-power driving. It is also desirable at times to reduce
the fundamental frequency of a lateral suspension as far as possible. For instance, below
the fundamental frequency, the displacement of the proof mass due to an applied
32
Suspension Design and Feasibility Study
k (2.3)
ω0 =
m
where k is the effective spring stiffness and m is the effective mass of the resonating
system.
x=u–w
w = w0ejωt
x = x0ejωtejφ
( k + jωs − ω 2 m ) x = mω 2 w (2.7)
x0 x mω 2 ω2 (2.8)
= = =
w0 w k + jωs − ω 2 m ⎡⎛ k 2⎞ 2⎤
1/ 2
⎢⎜ m − ω ⎟ + (ωs ) ⎥
⎣⎝ ⎠ ⎦
x0 ω 2 (2.9)
≅
w0 ω02
For a suspension with low ω0 the response will be large for measurements below the
resonant frequency.
33
Suspension Design and Feasibility Study
kinetic and potential forms. Potential energy is stored in compressed springs and when
the spring releases, the mass attached to the spring absorbs this energy resulting in
motion. Each cycle of energy transfer is subjected to loss in some form of energy
dissipation in the damper. At resonant frequency this transfer between kinetic and
ratio.
The response of the seismometer is restricted below the first resonant frequency of the
suspension though it is possible to design equalization circuits which can extend the
velocity amplitude response below the natural frequency [2.4]. Designing a low
frequency suspension thus helps avoid such circuits or increases the effectiveness of the
The quality factor (Q) is a measure of the rate at which a vibrating system dissipates
energy and is defined as the ratio of the energy stored per cycle to the energy dissipated
per radian.
For a vibrating mass-spring-damper system with high-Q, the system responds strongly
34
Suspension Design and Feasibility Study
Figure 2.2 Response vs. frequency for a resonating system showing the bandwidth of the systems
f0 (2.10)
Q=
Δf
where f0 is the resonant frequency and Δf is the bandwidth, which is defined as the
width in frequency when the energy falls to half its peak value (shown in Figure 2.2).
ω0τ (2.11)
Q=
2
where ω0 = 2πf0 and τ is decay time and is defined as the time it takes for the amplitude
35
Suspension Design and Feasibility Study
Figure 2.3 An oscillating system with exponential decay of amplitude in time. Decay time constant,
τ is the time it takes for the amplitude to decay to 1/e times its initial value.
• material impurity and defects (using single crystal silicon reduces material
defects)
• temperature
• anchor loss
The advantage of using single crystal silicon for fabricating low-noise suspensions is
the high-Q attainable in the material. Single crystal silicon has been used to fabricate
36
Suspension Design and Feasibility Study
2.2.3 Proof-mass
The NEA of the suspension is inversely proportional to the mass of the seismic weight
substantially lower than the miniaturized versions due to the advantage of a large mass
small which adversely affects the self-noise so in designing the suspension we will
select one which can create thicker structures. Typical semiconductor processes are
geared towards thin film technology (100’s of nm to 1-5 μm) which leads to extremely
low mass of the suspensions. Recent availability of high aspect ratio micromachining
tools based on reactive ion etching makes it possible to fabricate suspensions 500 μm to
1 mm thick.
Even though cross-axis sensitivity does not directly affect NEA, we will see in section
3.3.2 that this can restrict our ability to design very low resonant frequency suspension
as the rejection ratios for the spurious modes drops with reducing resonant frequency,
thus reducing the bandwidth of the suspension. The proof mass should move only along
This is achieved by increasing the spring compliance along the direction of motion ,
which is also necessary for achieving low-resonant frequency, while increasing stiffness
to translations along the other two axes and rigidity to rotation about all axes. The
37
Suspension Design and Feasibility Study
suspension needs to be designed to have symmetry along the orthogonal axes to the
The sensitivity to measure vibrations along any axis is inversely proportional to the
square of the resonant frequency (equation 2.9), so a low resonant mode along the
compliant axis (X-axis) leads to high sensitivity to motion along that axis. Pushing the
other modes (spurious modes) higher reduces the sensitivity of the suspension to
sets the higher limit for the bandwidth. With the use of a feedback circuit, the
bandwidth of the system is pushed up and is dependent on the gain of the feedback
circuit, which is the reason we have not considered bandwidth as one of the key
parameters here. In the presence of a feedback circuit, the first spurious mode defines
the bandwidth of the system, as the sensor response will not be linear when the
frequencies of vibration to be detected are near the spurious mode. This means one
needs to design a suspension with spurious resonant frequencies much higher than the
fundamental frequency to increase the bandwidth of the suspension. Chapter 3 will look
in detail at design features which are used to increase the cross-axis decoupling of
modes.
Lateral suspensions have been utilized in MEMS to produce resonant structures like
comb drive resonators, inertial sensors like accelerometers and gyroscopes and for
38
Suspension Design and Feasibility Study
lateral translational stages used in optical devices. In all these devices performance will
be enhanced by maximizing the separation of the fundamental and other spurious modes
2.2.3 the ideal suspension would be solely compliant along the direction of motion (X-
axis) of the proof-mass with maximum stiffness in all other axes. In its simplest form
the suspension will consist of two springs either side of a proof-mass connected to an
A 6DOF (Degree of Freedom) system will have six simultaneous equations for the
general case but it becomes separable in their coordinates if there are three planes of
symmetry to the suspension [2.7]. Such a suspension has no coupling between motion in
each coordinates. We will see in section 3.2 how this decoupling greatly simplifies the
analysis by removing off-diagonal terms from the equation of motion matrices for the
6DOF system. Hence, to decouple the motion along the various axes we consider
center-of-gravity systems, so called as symmetry implies that the elastic forces of the
Figure 2.4 A simple suspension with folded cantilever beams as spring on either side of a proof-
mass attached to a fixed external frame. The pair of springs on either side of the suspension only
allows dual planes of mirror symmetry about xy and xz planes as shown by the arrow in the middle
of the proof-mass.
39
Suspension Design and Feasibility Study
Folded cantilever beams have been a frequent design choice to best accommodate these
series of folded cantilevers on either side of the proof-mass (Figure 2.4) which softens
the suspension as well as improves the linearity as the deflection of each cantilever is a
perpendicular to the compliant direction and to avoid the dynamic and analytical
problems of cross-coupling between the axes mentioned earlier, the unit of suspension
is taken as a mirror pair of folded cantilevers [2.13, 14], which reintroduces the plane of
symmetry to the resulting center-of-gravity system. The folded cantilever pairs are
linked at their central attachment points as shown in Figure 2.5. This results in an
increased stiffness of the suspension to motion along the Z-axis. This is due to the
additional constraint of no relative motion in the x-direction for the pair at the linkage.
Figure 2.5 A center-of-gravity suspension design with symmetry along all the three axes. The joint
folded cantilever pairs on either side of the proof-mass provide symmetry about xy, yz and zx
planes. Symmetry in the lateral plane is shown by the arrows in the middle of the proof-mass. The
cantilever pairs are linked at the center to increase the stiffness compared to an unconnected pair.
40
Suspension Design and Feasibility Study
The suspension design shown in Figure 2.5 is the basis of the all the lateral suspension
designed and fabricated in this thesis. Various configurations using multiple joint-
cantilever pairs as springs are considered along with more advanced designs
In addition to symmetry, there are further constraints on the design of the suspension. It
which sets restrictions on minimum feature size and gap. A minimal area should be
required for fabrication to reduce cost. The proof-mass should have sufficient throw in
the compliant direction to cover required motion. Damping should be considered for the
suspension and complete assembly. These issues are considered in later chapters on
Figure 2.6 Homogenous triaxial configuration also called Galperin configuration (Source: [2.15]). u,
v, w are the responses of the sensor along the Galperin axes UVW.
To observe the vibration in all directions, a triple set of sensors are oriented along the
three axes X, Y and Z. For seismic sensing, the standard way is to place a seismometer
41
Suspension Design and Feasibility Study
for detecting ground motion along East, North and Upwards (Z) [2.15]. The designs for
horizontal and vertical vibration sensors are different, and to integrate them into a single
three identical horizontal sensors whose sensitive axis is inclined against the vertical at
same angle (θgalperin = 54.74°) like the edges of a cube standing on its corner
The response along the three axes can be calculated from the response along the
sensitive axis of each of the sensor by using the transformation matrix given by
⎛ x⎞ ⎛− 2 1 1 ⎞⎛ u ⎞ (2.12)
⎜ ⎟ 1 ⎜ ⎟⎜ ⎟
⎜ y⎟ = ⎜ 0 3 − 3 ⎟⎜ v ⎟
6
⎜z⎟
⎝ ⎠
⎜ 2
⎝ 2 2 ⎟⎠⎜⎝ w ⎟⎠
For a seismometer the X axis is normally oriented towards East, and the Y axis points
North. Presently there are two commercial seismometers with homogenous triaxial
the homogenous triaxial configuration. The fabricated suspensions will deform under
gravity when placed in the Galperin configuration. To avoid having to compensate for
lateral deflection the springs are preformed such that under gravity the proof mass is
The lateral deflection of the proof mass under gravity which is compensated by
42
Suspension Design and Feasibility Study
Figure 2.7 (a) A typical single spring suspension with preformed springs, (b) showing the off-
centred proof-mass in horizontal position and (c) Centring of the proof-mass under gravity in
Galperin orientation.
One of the key goals of MEMS device fabrication is to batch fabricate multiple devices
on the same wafer. For our design we have considered a widely available 100 mm
The device die size sets the natural frequency of the suspension. The larger the device
die, the lower will be the natural frequency of the suspension assuming we use the same
fraction of the die as proof mass and the springs are the same thickness and equal in
suspension out of the whole wafer will give us the best performance but the cost will be
high and a single defect can reduce the yield to zero. Taking production and cost issues
into consideration, a reasonable device size might be one which can lead to an
43
Suspension Design and Feasibility Study
Figure 2.8 Schematic of the suspension showing the proof-mass and spring dimensions.
Assuming suspension geometry similar to Figure 2.5 we can look at the relation
between die size (the overall size of the substrate containing a single suspension) and
the resonant frequency of the largest single spring pair suspension which can be
fabricated within that die. The suspension is assumed to have a uniform thickness of t,
considering the spring pair in Figure 2.8 supporting the proof-mass (dimension l×wm) is
made of 8 cantilevers of length l, and width w, the spring stiffness for the suspension is
given by
3EI (2.13)
k = 8×
l3
1 3 (2.14)
I= wt
12
(2.15)
k
ω=
m
Assuming the proof-mass is always the same fraction of the overall die size and the
spring and proof-mass length are the maximum length possible in the die (approximated
44
Suspension Design and Feasibility Study
to the length of the die) and the spring width and thickness are the same across the die
sizes. Doubling the width of the die doubles the width of the proof mass as well. The
24 EI 24 Ew 3t 2 Ew 3t 1 (2.16)
ω2 = = = ∝ 4
ml 3
12 ρl wm t ρl wm t l wm
4 4
From equation 2.16 doubling the die length and width (quadrupling the area of the die)
decreases the resonant frequency by a factor of 4√2. Figure 2.9 shows the variation of
natural frequency with die size. This implies that if we are able to design a 20 mm × 20
mm die with 10 Hz resonant frequency, a similar design with same spring thickness and
proof-mass:die ratio will lead to 56.57 Hz for 10 mm × 10 mm die and 314.75 Hz for a
5 mm × 5 mm die.
Figure 2.9 Variation of natural frequency with die size. Doubling the die length and width
suspension with low resonant frequency and high-Q will have a low NEA as seen from
Figure 2.10. The mass of the suspension is taken as 0.26 grams and the temperature is
45
Suspension Design and Feasibility Study
Figure 2.10 Variation of mechanical noise with resonant frequency for varying Q. The mass of the
We have fixed the die size at 20 mm × 20 mm as this allows for the fabrication of 12
dies on a single 100 mm wafer which means a yield of even 8% will lead to a single
functioning suspension, as we will see later for fabricating a suspension the die yield is
It is possible to fabricate a suspension with a natural frequency of 1 Hz, but when the
suspension is tilted to the Galperin angle the lateral deflection of the proof mass brings
the springs in contact with each other making the suspension useless for measuring the
movement of the proof mass. In these cases the suspension with preformed springs can
not be fabricated as the preforming will not be sufficient to counter-balance the force
46
Suspension Design and Feasibility Study
For initial suspension dimension calculations we will consider a design geometry shown
in Figure 2.11 which is similar to the design in Figure 2.5 and in addition includes space
for an electrical connector pad and an external frame which supports the suspension
within.
Figure 2.11 Schematic of the suspension design with one spring and mass which is assumed to be
As we discussed in section 2.3.1 the intent of the current design is to preform the
springs so that when the suspension is oriented along the Galperin axis the proof mass is
centred (Figure 2.7c). The lateral suspension sag in Galperin configuration is given by
g galperin (2.17)
Δx galperin =
ω 02
47
Suspension Design and Feasibility Study
The dimensional limits on the mass and springs are set by considering some standard
dimensions for frames, connector pads, and proof mass. We will initially assume that
50% of the area is proof mass, in addition the external frame is assumed to be 2 mm
thick with 2 mm connector on one side (Figure 2.11). This allows 16 mm as the spring-
mass length.
Parameter Value
From Figure 2.7a we can see that for a preformed spring we have 4 mm on each side of
the proof mass. The spring element can be seen as composed of a link, two spring
beams and another link. The gap between the spring beams is s and the link width is s/2.
For preforming we can assume the proof mass sag to be 4 mm (neglecting spring
thickness for initial estimate). Using equations 2.17 and 2.18 we calculate the minimum
48
Suspension Design and Feasibility Study
dimensions given above, now we need to calculate the spring width which will be
needed for achieving 6 Hz suspension. We will use equations 2.16 for calculating the
24 EI x , spring 3
2 Ewspring t spring (2.20)
ω0 = 3
= 3
mmass l spring mmass l spring
Using equation 2.20 and dimensional parameters from Table 2.1 we calculate the
30
25
20
ω (Hz)
15
10
5
16 21 26 31 36
Spring thickness (μm)
Figure 2.12 Variation of resonant mode with spring thickness for a 20mm×20mm die size and
10mm×16mm proof-mass.
49
Suspension Design and Feasibility Study
From Figure 2.12 it is evident that for a 6 Hz suspension the ideal thickness determined
theoretically is 18.2 μm. As we will see in section 2.5 this may need to be modified to
The design limitations discussed here are analyzed in detail in chapter 4 on fabrication
but it becomes imperative to handle these parameters here to set the ground work for the
etching of a 525 μm thick silicon wafer. This imposes a constraint on the minimum
feature size which can be etched through the thickness of the wafer.
Figure 2.13 Metal traces running on the springs connecting the electromagnetic feedback coil and
50
Suspension Design and Feasibility Study
The first constraint is on the width of a gap that can be etched through the thickness of
the wafer; this gap is about 40 μm for best etching characteristics with our silicon
thickness.
The second constraint is on the minimum width of a spring which will be able to
support the metal traces running on the top. Eight traces are required to connect the
three set of coils and a pair of pickup electrodes on the proof-mass resulting in two trace
per spring beam. The minimum width of metal trace is 6 μm with a gap of 4 μm
between the traces to satisfy local fabrication facility processing limitations. This leads
to a minimum spring thickness of 24 μm for two traces running on the springs as shown
in Figure 2.13. The two traces on each spring is needed for the differential capacitive
pickup circuit and electromagnetic feedback coil, both these components are discussed
in chapter 6.
We can see that the spring thickness we calculated in section 2.4 of 18.2 μm is below
the minimum spring thickness allowed by the fabrication process. So we can either
increase the number of springs so the thickness is higher for the spring beams or
increase the resonant frequency of the suspension by using a thicker spring. Using
equation 2.9 for spring width of 24 μm the resonant frequency is calculated to be 9.08
Hz. Giving leeways for design dimensions and processing constraints, we set a target
resonant frequency of 10 Hz for the lateral suspensions. We will analyze the suspension
covered in chapter 4.
51
Suspension Design and Feasibility Study
High performance macroscale seismometers have noise floor of 10-10 m/s2/√Hz. Given
our aim to fabricate a sub-nano-g suspension, our target noise floor is 0.35 ng/√Hz.
From the die size, target suspension noise and initial design value for the suspension
resonant frequency we can assess the feasibility of the design in regards to the required
Q and the amount of the die area required for “sag” of the device under Galperin tilt.
ω0 ≈ 10 Hz (2.25)
Using equation 2.2 for calculating the NEA of suspension, we can calculate the required
4k bTω0 (2.26)
Qt arg et = ≈ 330
NEAt2arg et .mmass
So if the fabricated device has a resonant frequency of 10 Hz and Q of above 330, the
We have to check that the suspension damping will be low enough to allow a Q of
above 330. For the suspension without any encapsulation squeeze film damping will be
most dominant, the expressions for squeeze film damping for two surfaces with overlap
mω 0 14ηlt 3 (2.27)
Qsqueeze = , α squeeze =
α squeeze Nh 3
52
Suspension Design and Feasibility Study
For l = lspring, t = tspring, and h = spacing between the spring beam = 2 mm, we can use
equation 2.27 to calculate the Qsqueeze for the suspension with N = 1 springs to be ~7400
which is significantly higher than the required Qtarget. Figure 2.14 shows Qsqueeze vs.
number of springs, N. The Q-factor due to squeeze damping reduces with smaller spring
gap for multiple sets of springs as the damping is higher. For N < 4, Qsqueeze is above
Qtarget.
8000
7000
6000
5000
Qsqueeze
4000
3000
2000
1000
0
1 2 3 4 5 6
Number of Springs (N )
Figure 2.14 Quality factor (due to squeeze film damping), Qsqueeze variation for multiple spring
suspension (N).
2.7 Summary
The targeted noise floor for the suspension was set to 0.35 ng/√Hz. To attain such low
self-noise the suspension should have large mass, low resonant frequency and high
quality factor.
The lateral suspension is covered in US patent 6776042 (Pike et al.). The author’s role
in the design process is to help model various suspensions to set the dimensional
53
Suspension Design and Feasibility Study
The lateral suspension design is based on folded cantilever beams which are commonly
To measure vibration along all three axes and avoid different designs for vertical and
springs are preformed so that under gravity at Galperin tilt, the proof-mass is centred.
This puts constraint on the resonant frequency of the suspension for a specific die size
due to the lateral deflection of the proof-mass under gravity at Galperin tilt.
proof-mass area was 6 Hz. With folded cantilever springs, this leads to a spring
thickness of 18.2 μm which is below the 24 μm minimum feature size constraint of the
Assuming a 10 Hz resonant frequency and a mass of 0.26 grams (50% of the die area)
we can attain the NEA of 0.35 ng/√Hz (3.43×10-9 m/s2/√Hz) if the suspension has a Q
higher than about 330. The squeeze flow damping of a suspension with less than four
2.8 Conclusion
We have demonstrated the feasibility of a lateral suspension which can attain the target
54
Suspension Design and Feasibility Study
fabrication tools like DRIE allows us to design suspensions which are hundreds of
The key parameters for the silicon lateral suspension set from device design and
ω0 ≈ 10Hz
Qt arg et ≥ 330
Devices made of single crystal silicon have been reported in literature to attain Q’s as
high as 100,000 [2.19] and 600,000 [2.5] which increases our confidence that the
suspension design can achieve the self-noise target. We can design suspensions with up
to three sets of springs while still staying within the target Q levels in normal
atmosphere. Vacuum packaging the device will remove all consideration for squeeze
film damping and remove any restriction imposed on putting the springs close together.
In the next chapter we analyze the suspension analytically and numerically considering
an ideal processing environment where the design is transferred into silicon rectilinearly
55
Suspension Design and Feasibility Study
2.9 Bibliography
56
Suspension Design and Feasibility Study
[2.11] J. A. Yeh, C.-N. Chen, and Y.-S. Lui, "Large rotation actuated by in-plane rotary
comb-drives with serpentine spring suspension," Journal of Micromechanics
and Microengineering, vol. 15, pp. 201-206, 2005.
[2.12] G. Zhou, "Tilted folded-beam suspension for extending the stable travel range of
comb-drive actuators," Journal of Micromechanics and Microengineering, vol.
13, pp. 178, 2003.
[2.13] H. Seidel, H. Riedel, R. Kolbeck, G. Mueck, and W. Kupke, "Capacitive silicon
accelerometer with highly symmetrical design.," Sensors and Actuators B:
Chemical, vol. 21, pp. 312, 1990.
[2.14] E. Peeters, S. Vergote, B. Puers, and W. Sansen, "A highly symmetrical
capacitive micro-accelerometer with single degree-of-freedom response,"
presented at International Conference on Solid-State Sensors and Actuators, San
Francisco, CA, USA, 1991.
[2.15] E. Wielandt, Seismometry, in International Handbook of Earthquake and
Engineering Seismology, Part A, W. H. K. Lee, H. Kanamori, P. Jennings, and
C. Kisslinger, Eds.: Academic Press, 2000, pp. 283.
[2.16] E. I. Gal'perin, "Azimutal'nij metod sejsmicekich nabludenij," Gostoptechizdat
1955.
[2.17] B. S. Melton and B. M. Kirkpatrick, "The symmetric triaxial seismometer - its
design for application in long-period seismometry," Bulletin of the
Seismological Society of America, vol. 60, pp. 717-739, 1970.
[2.18] W. A. Clark, Micromachined Vibratory Rate Gyroscope, Ph.D. Thesis,
Electrical Engineering and Computer Science, University of California,
Berkeley, 1997
[2.19] S. Pourkamali, A. Hashimura, R. Abdolvand, G. K. Ho, A. Erbil, and F. Ayazi,
"High-Q single crystal silicon HARPSS capacitive beam resonators with self-
aligned sub-100-nm transduction gaps," Journal of Microelectromechanical
Systems, vol. 12, pp. 487-496, 2003.
57
three
3 Suspension Dynamics
analytical model to explore the design space and verify the results using finite element
sets and look at the pros and cons of the design. Intermediate frames are shown as a way
to decouple the effect of multiple spring sets on spurious modes. Various other design
modifications like reduced mass frames, spring end connector compliance and linkage
geometries are analyzed. Finally, a set of suspension designs fulfilling the specifications
58
Suspension: Dynamics
3.1 Introduction
The behaviour of the fundamental mode of lateral suspensions can be analysed in terms
spurious modes. Finite element analysis (FEA), which is able to extract any number of
spurious modes, has been applied to the uniaxial suspension of the Analog Devices
accelerometer [3.3]. Analytic approaches and FEA have been used to optimize the
design of a dual-axis suspension [3.4]. In this chapter we apply analytical and numerical
Appendix A) is considered in conjunction with a finite element model for the six
degrees of freedom (6 DOF) system. Further modes due to the finite mass of the
rejection ratios on simple multiples of the aspect ratios of the suspensions giving
straight forward heuristic design for suspension optimization. Next, suspensions with
multiple spring sets are studied to lower the fundamental frequency. Finally,
approach for enhancing the performance [3.5]. A full set of analytical equations and
frames.
The analysis is focused on developing a model for the suspension to predict its
dynamical characteristics prior to fabrication. It also offers a test bed to investigate the
design space and improve the cross-axis rejection ratio of spurious modes.
59
Suspension: Dynamics
Figure 3.1 Schematic of a lateral suspension illustrating the model geometry and the critical
dimensions.
extrusion is given by the thickness of the etched layer, which is the full wafer thickness
(t). The suspension consists of two springs either side of a proof mass connected to an
external fixed frame. The ideal suspension would be solely compliant along the X-axis
with maximum stiffness to translations along the other two axes and rigidity to rotation
about all axes. It is assumed that the suspended mass is perfectly rigid. It is further
assumed, at least initially, that the suspension is massless, which is reasonable when
60
Suspension: Dynamics
on the massless solution. Under these assumptions the analysis of such a suspension
involves the solution of the equations of motion for a 6DOF system [3.6]:
(3.1)
&& + [C]u& + [K ]u = 0
[M ]u
(3.2)
u = [ x, y, z , α , β , γ ]T
x, y, z are displacement along the three axes, α, β, γ are the rotational components about
[M] is the inertial matrix, which has the diagonal components (m, m, m, I xx , I yy , I zz ) and
symmetric off-diagonal terms dependent on the shape of the suspended mass, [C] is a
symmetric matrix of the damping terms and [K] is a symmetric stiffness matrix. The six
simultaneous equations for the general case become separable in their coordinates if
there are three planes of symmetry to the suspension [3.7]. Such a suspension has no
coupling between motion in each coordinate implying that all the off-diagonal terms in
M and C are zero. This decoupling not only greatly simplifies analysis but is necessary
if the suspension is used for transduction. Here we consider only such center-of-gravity
systems, so-called as symmetry implies that the elastic forces of the suspension are
directed through the centre of the suspended mass. In addition, the system is further
simplified by neglecting damping as the aim of the analysis is to optimize the design in
61
Suspension: Dynamics
which has the customary normal-mode solution, ui = ai sin(ωi t + φ ) where normal mode
frequencies are
ω i = K ii / M ii (3.4)
with ai and φ set by the initial conditions. The spring constants, and hence the
optimize, the frequency ratio of the cross-axis modes to the on-axis mode, ωi ≠ x ωx
should all be as high as possible. As there could often be opposing effects on the various
design will in general have its two lowest cross-axis-mode frequencies approximately
equal.
order to set a rational design methodology. One obvious approach, optimization through
FEA, has already been applied to this suspension design in order to optimize harmonic
modes for a differential capacitive accelerometer [3.3] but FEA does not produce
general design rules. The analytical formulas developed in Pike et al [3.5] (derivation
presented in Appendix A) and used here should provide a more heuristic approach to
lateral suspension design. In deriving these analytical formulas, FEA (ANSYS, Inc.,
Canonsburg PA), was applied to elements of the suspension to ensure that all the
important modes of deflections were included, and also to crosscheck numerically the
analytical expressions. In order to produce reasonable FEA simulation run times and
node numbers, the suspension was taken as a simple extrusion. In the first instance, a
62
Suspension: Dynamics
minimal suspension consisting of a single spring unit either side of the proof-mass is
The planar geometry (Figure 3.2) of each spring unit can be characterized by the values
of the beam length l, beam width w, beam spacing (from beam axis to beam axis) s,
linkage length width ws, and linkage length, taken here to be (s-w)/2 to give a unit
dimension in the compliant direction of 2s. The detailed elbow geometry is left
undefined. There are two important geometric ratios defining the suspension; the beam
cross-sectional aspect ratio, w/t; the suspension ratio of the beam spacing to the beam
length, s/l. The simplest suspension consists of one unit on each side of the proof mass
(Figure 3.1). Such a suspension satisfies the required symmetry conditions and the six
spring constants can be solved separately for motion along and about each axis.
Figure 3.2 Schematic of a single spring set for the lateral suspension showing the geometry and
critical dimensions.
63
Suspension: Dynamics
The compliant direction for the lateral suspension is taken along X-axis while Z-axis is
spring stiffness constant for the suspension along the compliant direction is given by
24 EI x (3.5)
kx =
l3
where E is the Young’s modulus of silicon, Ix is the area moment of inertia given by
1 w3t and l is the length of the spring as shown in Figure 3.2. The fundamental mode
12
(3.6)
kx
ωx =
m
where m is the mass of the proof-mass. The mode shape is shown in Figure 3.3.
Figure 3.3 Fundamental mode shape of a single spring set lateral suspension. The proof-mass is
anisostropic crystal such as silicon, the beams should be orientated to minimize E. For
64
Suspension: Dynamics
spring constants in the y and z direction which are derived by considering the
deflections of the suspended mass due to a test force applied at the centroid of the mass
The spring stiffness constant in the y direction for the lateral suspension with a single
8EI x (3.7)
ky =
ls 2
ωy l (3.8)
=
ωx 3s
Figure 3.4 ωy, Mode shape of a single spring set suspension resonating along Y-axis.
Next, the deflection in the out-of-plane z direction, often the most important deflection
65
Suspension: Dynamics
spring stiffness along the Z-axis, an axial force impressed in the z direction at the
1 l3 ls 2 (3.9)
= +
k z 24 EI z 8GJ
The two terms in the expression for out-of-plane spring stiffness are due to deflection of
the beam (first term) under an out-of-plane force and twisting of the elbow (second
term) due to the torque generated from opposing forces at the end of the beams. Iz is the
rigidity and J is the torsional constant for the spring beam about Z-axis given by 1 w3t .
3
−1 / 2 (3.10)
ω z ⎛ EI x s 2 I x ⎞
= ⎜3 + ⎟
ω x ⎜⎝ GJl 2 I z ⎟⎠
Figure 3.5 ωz, Mode shape of a single spring set suspension resonating along Z-axis.
66
Suspension: Dynamics
The rotational mode about the X-axis, ωα is the angular deflection of the proof-mass
about X-axis. The angular deflection consists of contribution from twisting occurring
about X-axis of the linkages and the combination of twisting and bending of the spring
beams. The compliance about the X-axis, 1/kα is thus a sum of the compliance due to
torsion of linkages (1/kα,link) and the compliance contribution of the beams (1/kα,beam).
1 1 1 (3.11)
= +
kα kα ,link kα ,beam
The spring stiffness about X-axis due to torsion of the linkages is given by
2GJ s (3.12)
kα ,link =
s−w
where Js is the torsion constant of the linkage given by 1 ws3t . The linkage torsional
3
stiffness can be made very high by increasing the width (ws) of linkage. In that case, the
torsional stiffness of the beam will be dominant in determining the resonant mode about
the X-axis. The compliance due to the bending and twisting of the beam is given by
⎛ ⎞ (3.13)
⎜ ⎟
1 l ⎜ 3 ⎟
= 1−
kα ,beam 4 EI z ⎜ EI z s 2 ⎟
⎜ 4+3 ⎟
⎝ GJl 2 ⎠
The linkage width should therefore be chosen to restrict the contribution of the torsion
of the linkage to the total compliance. The condition kα ,link >> max (kα , beam ) for a
⎡ E ⎛ t ⎞ 2 s⎤
13 (3.14)
w s>> ⎢2 ⎜ ⎟ ⎥ w
⎣ G ⎝ w ⎠ l⎦
67
Suspension: Dynamics
As the beam aspect ratio and suspension aspect ratio will often be comparable, and
certainly no more than fifty each, condition 3.14 will be met for a linkage of width, ws,
The rejection ratio for the resonant mode about the X-axis is given by
−1 / 2 (3.15)
⎧ ⎡ ⎤⎫
ωα ⎪⎪ 4 EI x ( s − w) g 2 4 I x g 2 ⎢ 3 ⎥ ⎪⎪
=⎨ + ⎢1 − ⎥⎬
ωx ⎪ GJ s l 3 I zl 2 2
⎢ 4 + 3s EI z ⎥ ⎪
⎩⎪ ⎣⎢ l 2GJ ⎦⎥ ⎭⎪
Figure 3.6 ωα, Mode shape of a single spring set suspension resonating about X-axis.
The rotational spring constants about the y and z axis, kβ and kγ , is composed of
contributions from the effective translational springs and additional angular deflection
68
Suspension: Dynamics
kβ = kzh2 (3.16)
The rejection ratio of the frequency of the rotational mode about the Y-axis to the
ωβ (3.17)
= 3
ωz
Figure 3.7 ωβ, Mode shape of a single spring set suspension resonating about Y-axis.
Similarly for rotation about the Z-axis, again neglecting torsional effects, the rejection
ratio of the frequency of rotational mode about Z-axis to the translational mode along Y-
axis is given by
ωγ 3 (3.18)
=
ωy 1+ g 2 /h 2
where g, h are the half width and half length of the proof-mass respectively. Therefore
off-axis rotational-mode frequencies, simplifying the design approach. The mode shape
69
Suspension: Dynamics
Figure 3.8 ωγ, Mode shape of a single spring set suspension resonating about the Z-axis.
higher, which means the expressions given by 3.17 and 3.18 underestimate the rejection
ratio and so the design rules developed based on these expressions satisfy the key
expressions for the rejection ratios of all the spurious modes. First, the beam cross
sections are assumed to be rectangular. Second, the ratio of Young’s modulus to the
(for silicon {100} E/G is 2.56 [3.15]). Third, it is assumed that the linkages and elbows
are completely rigid (1/kα, link = 0) and of zero width. In this case the spring mass will be
8ρltw. Fifth, the suspension is assumed to extend the width of the proof mass so that 2l
= g. Last, the proof mass is assumed to be square. Under these conditions, the rejection
ratios are shown in final column of Table 3.1. Further, if it is assumed that the
70
Suspension: Dynamics
suspension and beam aspect ratios are equal, s/l = w/t = R, the simple multiples shown
Mode, ω i2 ω x2 ω x2
for rectangular , with
i ω i2 ω i2
beam cross sections,
s/l = w/t =
E/G = 3, and rigid, zero
width links and elbows, R
and square proof mass
x 24 EI x 1 1
ml 3
x1 m 2 2 w 2 t
π2 ωx R
ms π2 h π2 h
y 8EIx l2 2 s2 3R2
= ωx 3 2
mls 2 3s 2 l
z 1 ωx2 s2 w 2 2R2
= +
mls 2 ml 3 3EI x s 2 I x l2 t 2
+ +
8GJ 24 EI z GJl 2 Iz
α ⎧ ⎡ ⎤⎫
−1
⎛ ⎞ ~ R2
⎪⎪ 4 EI ( s − w) g 2 4 I g 2 ⎢ ⎥ ⎪⎪ 2 2⎜ ⎟
3 w 3
+ x2 ⎢1 − ⎥ ⎬ ωx 4 2 ⎜1− ⎟
x
⎨ 2
⎪ GJ s l 3 I zl ⎢ 4 + 3s EI z ⎥ ⎪ t ⎜ 1 s2 t 2 ⎟
⎪⎩ ⎢⎣ l 2GJ ⎥⎦ ⎪⎭ 4+
⎝ 4 l2 w 2 ⎠
β 3ω z2 1 ⎛ s2 w 2 ⎞ 2 2
R
⎜ + ⎟ 3
3 ⎝ l2 t 2 ⎠
γ 3 s2 2R2
2 2
ω y2 2
1+ g /h l2
Table 3.1 presents the expressions for calculating the modes of a single spring set
dependent on the suspension and beam aspect ratios. In the simplified case, most modes
are linearly dependent on the fundamental. The first on-axis spurious mode ωx1 as a
71
Suspension: Dynamics
applications like an accelerometer for low frequency vibration sensing. For a given
device area this can be achieved by either reducing the width of the beams or adding
additional suspension units on either side of the proof mass. There is a limit to how thin
the beams can be fabricated due to inherent process limitations and so the addition of
Figure 3.9 Schematic of a lateral suspension with multiple springs, (a) a lateral suspension with n
set of spring, and (b) a specific lateral suspension with two (n=2) spring sets.
A series of n folded cantilevers each of spring constant k will have a summed spring
constant of k/n. As well as softening the suspension, the use of multiple units improves
the linearity as the deflection of each cantilever is a lower multiple of the beam
72
Suspension: Dynamics
thickness. For a suspension with n pair of springs as shown in Figure 3.9, the spring
24 EI x (3.19)
kx =
nl 3
The frequency for the fundamental mode for a suspension with n spring sets is given by
(3.20)
kx 24 EI x
ωx = =
nm nml 3
For a set of parameters given in Table 3.2 for lateral suspension and applying equation
3.20 we calculate the resonant frequencies for suspensions with 1 to 6 spring sets. A FE
(Finite Element) model using Ansys (Figure 3.10) is used to verify the analytical
expressions.
program ANSYS [3.16]. A parametric FE model was developed using the input file
capability of ANSYS for a generic spring-mass lateral suspension. This allowed a single
model file to be modified parametrically to set the number of spring sets, spring and
proof-mass dimensions. The suspension was initially designed as a 2-D model and
meshed with quadrilateral PLANE42 elements. The 2-D model was then extruded
vertically 525 μm to form a volume with SOLID45 elements with three degrees of
freedom at each of the eight nodes. The model contains large volumes like the proof-
mass and high aspect ratio structures such as spring beams. Automated meshing of the
problems:
73
Suspension: Dynamics
2. Automated meshing of the structure leads to very high aspect ratio elements with
narrow widths in the spring beams. At the linkages there is extreme size transition from
the larger elements in proof-mass to the smaller elements in spring beams. This leads to
To avoid problems associated with irregular tetrahedral elements, the 2-D model was
initially map meshed with PLANE42 elements with lower density of elements filling
the proof-mass (Figure 3.10) and larger density of elements in spring beams with low
aspect ratio elements at the transition boundary. The extruded 3-D model then contains
Figure 3.10 FEA (Finite Element Analysis) model of a lateral suspension with two set of springs
solved using Ansys. The inset shows the arrangement of the nodes and elements in the model.
The mode shape and frequencies are calculated using Block Lanczos method. FE
simulation allows inclusion of effects of a number of factors that are neglected in the
simplified analytical model like mass of the spring, and anisotropic property of single
crystal silicon. The input files for the ANSYS simulation are presented in Appendix B.
74
Suspension: Dynamics
Figure 3.11 The transition elements are designed such that the nodes at the common boundary are
Table 3.2 Dimensional parameters for suspension design with multiple set of springs.
Parameter Value
75
Suspension: Dynamics
Figure 3.12 shows the fundamental mode calculated using the analytical expressions
along with the FEA calculated values. The FEA model estimates for the resonant modes
are lower than analytically calculated values as the FEA considers the effect of the
spring mass which reduces the fundamental mode slightly. The FEA model also
considers the effect of the end connectors (elbow) of the spring which is neglected for
10
8
6
4
2
0
1 2 3 4 5 6
Number of Springs (n)
Figure 3.12 Fundamental resonant frequency for lateral suspension with varying number of spring
sets. The values are calculated using the analytical expression from equation 3.20 and compared
against FEA derived values. The deviation of the analytical model from the FEA model can be
attributed to the mass of the suspension and the effect of end-connector geometry.
The discrepancy between the analytical and FEA model is under 5% and can be reduced
within 1-2% if the springs are considered massless and the end connectors are made
rigid.
76
Suspension: Dynamics
Up to this point the springs have been considered massless. The mass of the spring
beams will both reduce the frequency of the fundamental mode and produce additional
modes in all degrees of freedom. The frequency will be reduced according to Rayleigh’s
method [3.17] to calculate the effective mass of the spring, leading to an updated
(3.21)
kx
ω*x =
33
m+ ms
140
where ms is the total mass of the suspension. For most suspension designs, m >> ms, and
the mass-loading effect is very small. Corresponding reductions will be seen for the
other modes. If the suspension mass is comparable to the suspended mass, m in equation
33
3.20 should be replaced by m + ms . For a suspension with a single set of springs,
140
incorporating the mass of the spring causes a deviation in the fundamental frequency,
ωx of 0.5%.
The most important additional modes are in the x direction, as they will be excited by
the same signals as the fundamental. In general, an analytical solution is not obtainable.
One can split the suspension into a series of discrete masses separated by massless
springs. To determine the normal modes we have to solve the homogeneous linear
B − ω 2A = 0 (3.22)
where B is the potential energy matrix and A is the kinetic energy matrix [3.18]. Figure
3.13 show a suspension model where the spring beams are considered as discrete
77
Suspension: Dynamics
masses connected with massless springs of stiffness k. The point masses (m) are an
approximation to the distributed masses of the spring sets of the actual suspension.
⎛m 0 0 0 0 0⎞
0 ⎛ 2 −1 0 0 0 0 0⎞ (3.23)
⎜ ⎟ ⎜ ⎟
⎜0 m 0 0 0 0
0⎟ ⎜ −1 2 −1 0 0 0 0⎟
⎜0 0 m 0 0 0⎟
0 ⎜ 0 −1 2 −1 0 0 0⎟
1⎜ ⎟ k ⎜ ⎟
A= ⎜0 0 0 M 0 0 ⎟,B = ⎜ 0
0 0 −1 2 −1 0 0 ⎟
2⎜ 2⎜
0 0 0 0 m 0⎟
0 0 0 0 −1 2 −1 0 ⎟
⎜ ⎟ ⎜ ⎟
⎜0 0 0 0 0 m 0⎟ ⎜0 0 0 0 − 1 2 − 1⎟
⎜0 0 0 0 0 0 m ⎟⎠ ⎜0 0 0 0 0 − 1 2 ⎟⎠
⎝ ⎝
and we solve the equation 3.22 to determine the eigenvalues. Using Mathematica
k
( k
) ( k 2k 2k
) k
(k
) ( )
(3.24)
ω i2 = { , 2− 2 , 2− 2 , , , 2+ 2 , 2+ 2 }
2M m m m m m m
implying
ω1 (3.25)
ω0
= (4 − 2 2 )Mm ≈ M
m
where ω1 is the first on-axis harmonic ωx1 and ω0 is the fundamental mode ωx.
Unless the number of elements is low, this analysis gives little physical insight.
Figure 3.13 To model the first on-axis harmonic ωx1, the spring units of the lateral suspension can
be considered as discrete masses separated by massless springs. M is the mass of the proof-mass, the
spring are considered as discrete masses m with massless springs of stiffness k connecting them.
78
Suspension: Dynamics
distributed mass and spring constant. This approach leads to simpler results, more
self-consistent if the resulting modes have periodicities which are much greater than the
periodicities in the distribution of the mass of the spring along the compliant direction.
For the suspension geometry under consideration this implies that the mode periods
Consideration of the mass of the suspension introduces additional on-axis modes with
the displacement in the x direction of the suspensions on either side either symmetric or
antisymmetric about the suspended mass [3.19]. The symmetric mode has a node at the
suspended mass and is therefore relatively benign to the overall dynamic performance.
However, the antisymmetric mode drives an oscillation of the suspended mass, the
amplitude of which is proportional to ms/m. The derivation for the on-axis symmetric
and anti-symmetric modes are covered in Pike et al. [3.19]. The rejection ratio of the
(3.26)
ω x1 m
≈π
ωx ms
Figure 3.14 shows the drop in rejection ratio for the first spurious mode along X-axis
(ωx1) with increasing number of spring sets (n). The analytical model gets better at
predicting the mode with increasing number of spring sets, approaching the numerically
distributed mass to model the springs and with higher number of spring sets the
assumption is strengthened.
79
Suspension: Dynamics
35
30
Mode ratio, ωx1/ωx
25
20
15
10
0
1 2 3 4 5 6
Number of Springs (n)
Figure 3.14 Plot of the first spurious mode along X-axis (ωx1) with increasing number of spring sets
(n). With larger number of spring sets the analytical model approaches the FEA model.
suspended mass will occur. Both the amplitude and frequency of this mode depend on
the ratio m/ms, linearly and to the square root respectively, and hence it is important to
Figure 3.15 shows the rejections ratios calculated from FEA models for lateral
suspensions with multiple spring sets. The ordering of the rejection ratios for one
suspension unit agrees well with the rules of thumb established in the last column of
Table 3.1. As the number of units increase, the FEA rejection ratio drops as
approximately 1/n for y, z, and β; as 1/n0.6 for γ; as 1/√(n) for x1; and is independent of n
for α.
80
Suspension: Dynamics
The reduction in the rejection ratios of the y, z, and β modes can be explained by
considering that the translational compliance of an elastic member increases with the
cube of the length of the member, so that while from equation 3.20 the fundamental
frequency drops inversely as the square root of n, the cross-axis frequencies fall as 1/n3/2
and so the rejection ratio will scale as 1/n. The γ mode depends more on the twisting
compliance of a member, which increases with the square of the length, and hence the
reasonably close to the 1/n0.6 behaviour seen. For the x1 mode, the mass of the
suspension is proportional to n, and so the rejection from equation 3.22 will fall as the
inverse square root. Like the fundamental, the α mode frequency is inversely
suspension units.
35
30
Mode ratio, ωi /ωx
25
20
15
10
0
1 2 3 4 5 6
Number of Springs (n)
Figure 3.15 Rejection ratio of spurious modes calculated using FEA model for lateral suspension
with 1, 2, 3, 4, 5 and 6 sets of springs. The rejection ratio drops significantly for multiple springs
81
Suspension: Dynamics
Even though multiple spring suspensions achieve lower fundamental modes (Figure
3.12), the dramatic drop in rejection ratio is of serious concern and will affect the
sensitivity of a suspension based on multiple spring design. As we will see in the next
section there is a significant vertical deflection of the proof-mass when using multiple
springs. In section 3.4 we will introduce intermediate frames to stabilize the rejection
ratio.
For a horizontal lateral suspension, the vertical deflection (sag) of the proof mass under
gravity is given by
mg * g * (3.27)
zsag = = 2
kz ωz
where mg* is the weight of proof mass, g* is the acceleration due to gravity and kz is the
For a Galperin configuration (section 2.3.1), the vertical and lateral component of g*
causes the proof mass to deflect both in-plane as well as out-of-plane and is given by
We use equations 3.27-29 to calculate the vertical sag for lateral suspensions with
82
Suspension: Dynamics
100
10
1
1 2 3 4 5 6
Number of Springs (n)
Figure 3.16 Logarithmic plot of vertical sag of a lateral suspension in horizontal and Galperin
configuration. The sag due to gravity increases with higher number of springs.
As is evident from Figure 3.16 suspensions with higher spring sets have larger
deflection under gravity. For the horizontal configuration of lateral suspension the
springs are not preformed and the deflection due to gravity of the proof-mass increases
linearly with the number of spring, n. For the Galperin configuration, the spring is
preformed to compensate for the in-plane deflection of the proof-mass such that when
the lateral suspension is placed in the Galperin configuration the proof-mass is centred.
The vertical deflection of the proof-mass under gravity is slightly better than in the
horizontal configuration but will need to be reduced within reasonable limits to allow
workable capacitive pickup between the proof-mass die and the capping die.
The vertical sag under gravity can be reduced by introducing frames between sets of
83
Suspension: Dynamics
Suspension
As we discussed in section 3.3 there is a considerable price to be paid for reducing the
fundamental frequency by increasing the number of spring sets – the rejection ratio
drops precipitously for most of the off-axis modes and the vertical deflection is
significant.
necessary to decouple the additional spring sets, so the off-axis compliances add
linearly rather than scale as the cube or square of the units added. Intermediate frames
[3.20] can be introduced into the suspension to increase the cross-axis rigidity. Figure
3.17 shows a schematic of a suspension with a single intermediate frame. These frames
are symmetric about the suspended mass and couple together the two sets of springs on
either side of the proof mass. As they join parts of the suspension which move
synchronously in the fundamental mode, they have very little effect on the fundamental
modes the frames can greatly increase the frequencies of these modes. The effect of the
frames is first analysed assuming they are massless and perfectly rigid.
frame for lateral suspension. By using intermediate frames, the suspension is stiffened
in the z direction with a linear dependence on number of intermediate frames (n). Since
the fundamental mode is also linearly dependent on n, the rejection ratio for out-of-
84
Suspension: Dynamics
Figure 3.17 Schematic of a lateral suspension with a single intermediate frame between two set of
corresponding increase in rotational rigidity with the introduction of frames for rotations
about the Y and Z axes. For rotation about the Y and Z axes, there is further stiffening as
the elastic forces are applied further away from the rotation axis as the number of
suspension units increases. This leads to a better rejection ratio for β mode as n
increases. There is similar stiffening about the Z-axis due to the increased moment arm
for in-plane rotations as spring units are added improving rejection ratio for γ mode.
The frames have no effect on rotation about the X-axis as they introduce no additional
constraints for this motion. The analytical expressions for all the modes for n spring sets
85
Suspension: Dynamics
Table 3.3 Resonant frequencies for lateral suspension with n spring sets and (n-1) intermediate
frames.
Mode, i ωi2
x 24 EI x
mnl 3
y 8EI x l2 2
= ωx
mnls 2 3s 2
z 1/ n ω x2
=
mls 2 ml 3 3EI x s 2 2 I x
+ +
8GJ 12 EI z GJl 2 Iz
α ⎧ ⎡ ⎤⎫
−1
⎪⎪ 4 EI ( s − w) g 2 4 I g 2 ⎢ 3 ⎥ ⎪⎪ 2
+ x2 ⎢1 − ⎥ ⎬ ωx
x
⎨ 3 2
⎪ GJ s l I zl ⎢ 4 + 3s EI z ⎥ ⎪
⎪⎩ ⎣⎢ l 2GJ ⎦⎥ ⎪⎭
β ⎡ ( s + f / 2) ⎤ 2
3⎢1 + (n − 1)
⎣ h ⎥⎦ωx
γ ⎡ ( s + f / 2) ⎤
3⎢1 + ( n − 1) ⎥⎦ 2
⎣ h
ωy
1 + g 2 / h2
x1 m + nmspring
3 ωx2
m + n( mspring + mframe )
The frames will have a detrimental effect on harmonics along the X-axis due to the
increased mass of the suspension. The first harmonic will have a frequency of
(3.30)
ω x1 m
=π
ωx ms + m f
86
Suspension: Dynamics
for a total frame mass of mf. Therefore the frames should be made as light as possible
while maintaining sufficient rigidity. Figure 3.18 shows the drop in rejection ratio for
on-axis spurious mode ωx1/ωx for suspension with intermediate frames as compared to
35
Mode Ratio, ωx1/ωx
30
25
20
15
10
5
1 2 3 4 5 6
Number of Springs (n)
Figure 3.18 Plot of ωx1/ωx for lateral suspension with n (one to six) spring sets and (n-1)
intermediate frames. The FEA model used to calculate the rejection ratio is based on the model
The rejection ratio for lateral suspensions with multiple spring sets (with dimensions
similar to the model used in section 3.3.2) but with intermediate frames in shown in
Figure 3.19. By including intermediate frames into the design, all modes except for x1
maintain or, in the case of β or γ, improve their rejection ratio as additional suspension
units are added, in agreement with the analytical formulas of Table 3.3.
87
Suspension: Dynamics
20
Mode ratio, ωi/ωx
15
10
5
1 2 3 4 5 6
Number of Springs (n)
Figure 3.19 Plot of rejection ratio for lateral suspension with multiple springs calculated using
FEA model based on the same parameters as in Table 3.2 is solved using Ansys to
verify the analytical calculations. Figure 3.20 shows similar trend as the analytically
calculated values (Figure 3.19) with improvement in the mode rejection ratio for β and γ
88
Suspension: Dynamics
20
Mode ratio, ωi /ωx
15
10
5
1 2 3 4 5 6
Number of Springs (n)
Figure 3.20 Plot of rejection ratio for lateral suspension with multiple springs calculated using a
FEA model.
The relative error of the analytical model as compared to the FEA model is shown in
Figure 3.21. The error in calculating the mode rejection ratio for most of the modes are
within 10% except for β mode where the error, though high initially, reduces
significantly with higher number of frames. The error is calculating γ mode seems to be
increasing but plateaus at around 15%. Even though the analytical model for calculating
β and γ modes are significantly in error with the FEA model, the analytical model still
predicts most of the other modes with significant accuracy and accurately estimates the
first few spurious modes. Additionally since the analytical model underestimates the
out-of-plane rejection ratio (ωz/ωx) which is the first spurious mode for the suspension,
any design developed using this model will satisfy the criteria to increase the rejection
ratio.
89
Suspension: Dynamics
10
5
Error %
-5
-10
-15
-20
1 2 3 4 5 6
Number of Springs (n)
Figure 3.21 Error between the FEA model and analytical expressions for the rejection ratios.
In Figure 3.18 we discussed how the rejection ratio for the first on-axis spurious mode
gets worse due to the weight of the intermediate frame, but increased rigidity provided
by the frame could be useful in pushing up the rejection ratio for other modes. Figure
3.22 shows the rejection ratio (normalized against a suspension design with 3 spring set
and 2 20 μm thick intermediate frames) for suspensions with multiple spring sets and
intermediate frames. Except for the first on-axis spurious mode (ωx1), the rejection ratio
for most other modes are significantly increased by increasing the thickness and hence
To reduce the weight of the intermediate frame and increase the rigidity, it is possible to
design intermediate frames with slots in them as shown in Figure 3.23. Even though
90
Suspension: Dynamics
fabrication constraints due to minimum etch feature size of 40 μm leaves little room to
1.90
1.70
Mode ratio (normalized)
1.50
1.30
1.10
0.90
0.70
0.50
20 40 60 80 100 120 140 160 180 200
Frame thickness (μm)
Figure 3.22 Plot of rejection ratio for lateral suspension with three spring sets and two intermediate
frames of varying thickness. Most of the rejection ratio except ωx1/ωx improves with thicker frames
Figure 3.23 (a) Reinforced intermediate frames to reduce the weight of the frame while maintain
higher rigidity of a thicker frame, as compared to (b) Solid intermediate frame which reduces the
on-axis rejection ratio due to the higher weight of the spring-frame unit.
91
Suspension: Dynamics
suspension
Figure 3.24 compares the rejection ratio calculated using FEA for multiple spring
suspension with and without intermediate frames. As we can see in Figure 3.24a the
rejection ratio for all modes drop precipitously with increasing number of spring units
whereas the same suspension with intermediate frames between each spring unit leads
to an improvement in the rejection ratio for all modes (Figure 3.24b) except the first on-
Figure 3.24 Comparison of the rejection ratio for a model lateral suspension with multiple springs
without frames and with frames. (a) Log-log variation as calculated by FEA of the rejection ratio
for the spurious modes as the number of suspension units is increased from 1 to 6. Three fits for the
rejection ratio are shown: for x1, 1/√(n); for α, independent of n; for z, 1/n, and (b) Log-log
variation as calculated by FEA of the rejection ratio for the spurious modes of a suspension
incorporating intermediate frames as the number of suspension units is increased from 1 to 6. Fits
for the rejection ratio are shown: for x1, 1/√(n); for α, independent of n.
92
Suspension: Dynamics
The trend line showing the slope of the rejection ratio for the multiple spring
suspensions agree with the analytical model. For frameless suspensions the rejection
ratio varies between 1/n and 1/√n whereas for the suspension with intermediate frames
the rejection ratios are almost flat or improve slightly and x1 mode has a slope of 1/√n.
α mode in both cases stay constant as shown by the flat line in the plots.
increase out-of-plane stiffness. The frames have the effect of changing the boundary
Figure 3.25 Effect of intermediate frames on the vertical sag of proof mass under gravity, (a)
multiple spring set suspension under gravity with equal deflection of all spring beams as shown in
the cross-section view, as compared to (b) multiple spring set suspension with intermediate frame
showing a reduction in vertical sag under gravity due to the intermediate frame.
93
Suspension: Dynamics
The frames increase the torsional stiffness of the suspension along the Z-axis. The ratio
is given by [3.5]
k z , frame 3 (3.31)
=
k z , frameless 4n 2 − 1
kz,frames is the torsional stiffness with n – 1 frames and kz,frameless is torsional stiffness for a
The vertical sag (zsag) under gravity is inversely proportional to the out-of-plane
stiffness, kz. Figure 3.26 shows the reduction in vertical sag with increasing number of
spring units when intermediate frames are used compared to the frameless case. As we
can see from the curve the reduction in vertical sag is significant particularly with
higher number of springs: for six sets of springs the vertical sag with intermediate
60
50
zsag,frame /zsag,frameless
40
Vertical sag ratio
30
20
10
0
1 2 3 4 5 6
Number of Springs (n)
Figure 3.26 Reduction ratio for vertical sag under gravity for multiple spring suspension with
94
Suspension: Dynamics
The linkage affects the twisting about X-axis. A rigid link is therefore important to
Various link thicknesses are investigated and its effects are quantified on α and x1 mode
ωα/ωx ωx1/ωx
14
12
10
Mode Ratio
0
20 70 120 170 220 270
Link width (μm)
With increasing linkage width, ωx1/ωx decreases slightly due to increased mass of the
linkage but the effect is small enough to not merit a design modification. Rejection ratio
of the rotational mode about the X-axis (ωα) improves significantly with increasing
linkage width as the rigidity of the linkage reduces the compliance to torsion. After a
certain linkage width the gain in α rejection ratio does not improve any further as the
95
Suspension: Dynamics
concentration at the linkages with sharp corner. The linkage is filleted to reduce the
stress concentration at the corners (Figure 3.29) for similar applied force. Various
designs were investigated with the aim to reduce the stress concentration. Stress
concentrators are potential regions where fracture happens and shock test of structures
(chapter 6) have shown that most of the suspensions fail near the linkage.
Figure 3.28 Maximum stress at the sharp corner of the linkage shown by the red pointer in the
globe.
Figure 3.29 (a) Link design with sharp corners, (b) Link design with corner filleting.
Numerical simulation using Ansys was carried out to investigate the effect of filleting
96
Suspension: Dynamics
Since filleting the link increases the effective thickness of the spring, there is an
increase in ωx which can be compensated for by increasing the spring beam length. The
stress at the linkage is reduced with larger filleting at the corners as seen by the trend in
Table 3.4. The Von Mises stress is calculated by applying a fixed displacement of 332
μm along X-axis, and 15 μm along Z-axis to simulate maximum deflection of the spring
linkage.
Link design
Radius of curvature
97
Suspension: Dynamics
The analytical expressions in Table 3.1 and Table 3.3 are derived assuming rigid spring
end connectors (Figure 3.30). However, the rigidity of the end connector (elbow) plays
Figure 3.31 Distortion of the spring beam under an out-of-plane force. The forces acting at the
opposite ends of the elbow causes a torque which leads to an angular rotation of θ.
The total deflection in the z direction is the sum of the deflection at the elbow, Δelbow, as
a result of the twisting of the beams from the torque generated from the opposing forces
acting at the beam ends and the two cantilever deflections, Δbeam of the beams as a result
of the impressed out-of-plane force and the resulting torques as shown in Figure 3.31.
This torque will produce a constant twist along each beam [3.13], and produce a
Fz ls 2 (3.32)
Δ elbow =
8GJ
for small loads where G is the relevant shear modulus for the beam direction, and J is
the torsion constant which depends only on the cross-sectional dimensions of the beam,
98
Suspension: Dynamics
Fz is the out-of-plane force acting on each beam, l is the length of the beam and s is the
Figure 3.32 Deflection of the spring beams depending on elbow rigidity, (a) completely rigid elbow
with c = 1 where the bending is that for a fixed-free cantilever and (b) completely flexible elbow
The beam deflection will be given by Fzl3/(48EIz) where the elbow is considered as
completely rigid as shown in Figure 3.32a. The rigidity of the elbow about this axis is
much more difficult to maintain compared to the rigidity about the z axis for deflections
in the compliant x direction. If the elbow has no rigidity, each beam acts as a single
cantilever with a maximum angle of deflection at the elbow (Figure 3.32b) and the
the rigidity of the elbow and varies between a value of 1 and 4, is introduced into the
analysis:
cFl 3 (3.33)
Δ beam =
48EIz
1 cl 3 ls 2 (3.34)
= +
k z 24 EI z 8GJ
Table 3.6 presents this compliance factor for some representative geometries as
determined by finite element analysis. Where present, the elbow cross bar is half the
99
Suspension: Dynamics
beam spacing s from the end on the elbow. It is evident that to minimize the Z-axis
deflection some webbing of the elbow is required. Table 3.5 incorporates the elbow
compliance factor (c) into the analytical expressions for normal modes of the
suspension.
Table 3.5 Resonant frequencies for lateral suspension with n spring sets and (n-1) intermediate
Mode, i ωi2
x 24 EI x
mnl 3
y 8EI x l2 2
= ωx
mnls 2 3s 2
z 1/ n ω x2
=
mls 2 cml 3 3EI x s 2 cI x
+ +
8GJ 12 EI z GJl 2 Iz
α ⎧ ⎡ ⎤⎫
−1
⎪⎪ 4 EI ( s − w) g 2 4 I g 2 ⎢ 3 ⎥ ⎪⎪ 2
+ x2 ⎢1 − ⎥ ⎬ ωx
x
⎨ 3 2
⎪ GJ s l I zl ⎢ 4 + 3s EI z ⎥ ⎪
⎪⎩ ⎣⎢ l 2GJ ⎦⎥ ⎪⎭
β ⎡ ( s + f / 2) ⎤ 2
3⎢1 + (n − 1)
⎣ h ⎥⎦ωx
γ ⎡ ( s + f / 2) ⎤
3⎢1 + ( n − 1) ⎥⎦ 2
⎣ h
ωy
1+ g /h
2 2
x1 m + nmspring
3 ωx2
m + n( mspring + mframe )
effect on the modes. It is evident from Figure 3.33 that rejection ratio for out-of-plane
100
Suspension: Dynamics
(completely rigid) the rejection ratio is 25% better than at c = 4 (completely flexible).
Increasing the stiffness of the spring end connector also reduces the vertical sag under
gravity of the proof mass with 20% improvement from c = 4 to c = 1. There will be a
slight drop in rejection ratio for the on-axis spurious mode ωx1 due to the increased mass
of the spring.
End connector
geometry
(Aspect ratio
~20)
1.2
1.0
Normalized ratio
0.8
0.6
0.4
0.2
0.0
1.0 1.5 2.0 2.5 3.0 3.5 4.0
elbow compliance, c
101
Suspension: Dynamics
The outcome of the investigation of various designs is to select a design which increases
rejection ratio for the spurious modes. Semicircular elbow geometry (design (d) in
Table 3.6) with cross bar, unfilled, represents a reasonable compromise between good
By analysing the various designs we have selected three potential configurations which
seem to satisfy the criteria for a 10 Hz suspension and at least 10 times separation
between the fundamental and first on-axis spurious mode while trying to maximize the
Table 3.7 gives the dimensional parameters and expected normal modes for those
designs. Variance from these characteristics are expected for the final device as the
fabrication methods described in chapter 4 will not yield an ideal geometry, this is
variations.
The lowest spurious mode rejection ratio for one spring set suspension is better than the
two and three spring set suspensions but the superior rejection ratio comes as a result of
the higher fundamental mode of the one spring set suspension. For the three spring set
suspension the first on-axis spurious mode is ten times the fundamental satisfying the
at least 10.
102
Suspension: Dynamics
Material constants
ν 0.23
ρ, kg/m3 2330
Dimensional parameters
t, μm 525
w, μm 24 24 24
n 1 2 3
Resonant frequencies
103
Suspension: Dynamics
3.8 Summary
The author’s key work covered in this chapter is the numerical modelling of the
suspension dynamics and statics. The numerical models are also used to refine the
important parameters while keeping them simple for heuristic examination of the design
(Pike, Stanley, Syms). The dimensional characterization and modelling was performed
by the author.
Analytical and numerical models were used to simulate the dynamics of the lateral
suspensions. The analytical formulas have the advantage of allowing the immediate
The numerical model incorporates effects of factors which are neglected in analytical
model like silicon anisotropy and mass loading effect of the springs, additionally it
The models were used to analyze multiple spring set suspensions which due to their
larger number of springs allows for a lowering of the resonant frequency of the
suspension. The disadvantage of the increasing number of spring sets was the dramatic
axis rejection in the out-of-plane direction. Intermediate frames also improved rejection
104
Suspension: Dynamics
ratio for all other modes. The price to be paid for the introduction of the frames is in the
lowering of the first on-axis spurious mode (x1) due to the increased mass loading. The
mass loading is minimized by using a relieved structure for the intermediate frames,
although at the expense of some loss in torsional strength. The extra cross-axis stiffness
dynamics and frames are incorporated between each suspension unit. In addition to
intermediate frames, webbing at the elbows of the suspensions greatly reduces the out-
The analytical expressions of Table 3.3 and Table 3.5 are used to first determine the
then leads to possible variations on the geometry to push up the rejection ratio of this
mode. In general, as the lowest-frequency spurious mode is thus pushed away from the
fundamental, other spurious modes will be lowered in frequency. The widest separation
between the fundamental and first spurious value occur as the two lowest spurious
The three suspensions presented in Table 3.7 give an example of this approach for an
increase in the number of frames. As the number of suspension sets increases from one,
to two, to three, the lowest rejection ratio is raised compared to non-intermediate frame
suspensions while the other spurious mode rejection ratios decrease. For three
suspension sets several modes have a very similar rejection ratio, representing a near
105
Suspension: Dynamics
3.9 Conclusion
The agreement between analytical and FEA model provides confidence in formulas for
geometries.
Intermediate frames were demonstrated to decouple the rejection ratio from number of
spring sets for multiple spring set suspensions. They improved the rejection ratio for all
suspensions with varying number of spring sets and intermediate frames. The selected
suspensions are near optimum solutions for attaining the requisite resonant frequency
and cross-axis rejection as the mode rejection ratios plateau around 10 for a range of
designs based on the same original configuration of mirrored joint folded cantilever
spring suspensions with intermediate frames. The spring end connectors are designed
minimized to allow for minimum gaps between the suspension and any sandwiching
structures. In this case maximizing the rejection of one specific mode may be preferable
fundamental.
106
Suspension: Dynamics
The suspension designs presented in Table 3.7 are then prepared for fabrication, which
107
Suspension: Dynamics
3.10 Bibliography
108
Suspension: Dynamics
109
four
4 Suspension: Fabrication
This chapter covers the fabrication of the lateral silicon suspension. Processing of the
silicon wafer using DRIE has been analyzed along with issues relating to high aspect
110
Suspension: Fabrication
4.1 Introduction
Subsequent to suspension design analysis using analytical and numerical methods, the
suspension is fabricated by etching single crystal silicon wafers. The full thickness of
the silicon wafer is used to maximize the inertial mass. To etch through the wafer and
maintain a vertical profile, the etching has to be highly anisotropic. Common techniques
for creating highly anisotropic structures from silicon are anisotropic wet etching
or (110) limiting plane is unsuitable for a geometry which has vertical spring elements
with sidewalls along planes other than (111) and (110). Such etching is generally
Plasma processing is the most widely used process in the microelectronics industry. It
has become popular for fabricating high-aspect-ratio MEMS devices after the
development of Deep Reactive Ion etching (DRIE) systems based on the Bosch process
anisotropic etch.
The complete fabrication sequence is discussed in section 4.3 and 4.4. As with most
processing methods DRIE has its own set of problems such as grassing, etch lag,
sidewall roughness, etch profile, notching, etc., which we will discuss in section 4.5.
111
Suspension: Fabrication
Plasma is a largely ionized gas, consisting of positive (ions) and negative (electrons,
negatively charged ions) particles. It is electrically neutral due to the equal number of
positive and negative charges. The plasma generates electrons, atomic and molecular
ions, and atomic and molecular radicals. These plasma species undergo chemical,
physical and chemo-physical reactions with atoms of silicon to form volatile reaction
products. For directional etching the ions can be accelerated using an external electric
field. The accelerated ions remove the target atoms by physical ion bombardment. The
radicals are chemical species which generally cause chemical isotropic etching. Low
pressure inductively-coupled plasma creates a high density of ions and has become the
There are two main established technologies for deep anisotropic plasma etching of
silicon: cryogenic [4.2] and Bosch [4.1] (also known as time-multiplexed or pulsed
etching). In the cryogenic process the wafer is cooled down to below -100 C, which
slows down spontaneous chemical etching by radicals that is isotropic in nature and
only vertically accelerated ions drive the etching process. The need to cool down the
chamber to cryogenic temperatures makes this process expensive and there is a lack of
standard industry cryogenic etchers though some (ex. METlab system of Alcatel, and
The Bosch process works by passivation of the sidewalls in between etching steps (see
Figure 4.1). The process starts by first isotropically etching exposed substrate (Figure
4.1a). After a brief period (typically few seconds) the etching is stopped and a polymer
112
Suspension: Fabrication
passivation layer is deposited (Figure 4.1b). The passivation layer protects the substrate
from further chemical attack by the radicals in the plasma. The process then returns to
the etching cycle where the vertically accelerated ions bombard the bottom of the
previously etched trench removing the passivating layer (Figure 4.1c). The etch then
proceed isotropically to extend the etched trench (Figure 4.1d). Etching therefore is
Figure 4.1 Passivation and etch cycle in a typical DRIE step. (a) Isotropic etch of silicon in exposed
regions on the wafer, (b) Deposition of polymer CFn on all surfaces, (c) Preferential etching of
polymer from the bottom of the trench by vertically directed positive ions, and (d) Isotropic etching
Sulphur hexafluoride (SF6) and octafluorocyclobutane (C4F8) are used for etching and
passivation respectively. SF6 plasma generates fluorine atoms, positive ions SFx+ (x
varying from 0 to 5), positive fluorine ions F+, and negative ions SFy- (y varying from 4
to 6), F- and F2- [4.3, 4]. Rauf et al. [4.5] reports that in inductively coupled SF6
plasmas, the primary ions produced directly from SF6 are SF3+ and SF5+ , however they
113
Suspension: Fabrication
indicate SF2+ to be one of the important ions, where SF2+ is mainly produced due to
Figure 4.2 Model for inductively coupled plasma etch showing the positive and negative species in
the plasma, their energy angular distribution function (ADF) and the potential drop across the
Figure 4.2 shows the model of the plasma etcher. The positive fluorine ions are
accelerated across the plasma sheath and causes the ion bombardment induced etching.
Neutral radicals cause chemical isotropic etching. Electrons and other negative ions
seep across the sheath and are isotropic in nature so are confined to the top surface of
114
Suspension: Fabrication
the substrate whereas positive ions due to their higher energy (higher velocity due to
acceleration by sheath potential) are able to reach the trench bottom. The model shows
how the electrons and ions have similar energy angular distribution function (ADF)
inside the plasma. Due to the presence of a potential drop across the sheath the positive
The etching comprises of physical and chemical etching. Physical etching or sputtering
elastic/inelastic scattering [4.6]. Chemical etching takes place due to the disassociation
of SFx by ion bombardment into fluorine radicals which chemically react with silicon.
To achieve vertical sidewalls or anisotropic etching, during the passivation cycle the
plasma dissociation of CnF2n+2 . The SFx plasma species reacts with the CFx plasma
species deposited on the polymer, forming non-volatile reaction products which guard
C 4 F8 → CFx , x = 1,2,3
+
CFx ⎯⎯→
⎯ (CFx ) n
Ar
DRIE has been used to make a variety of Microsystems devices [4.8], from inertial
sensors such as accelerometers [4.9, 10], gyroscopes [4.11], to optical passive and
active components like switches [4.12], tunable optical filters [4.13], energy conversion
devices [4.14], biomedical devices like needles [4.15], magnetic resonance imaging
Helmholtz coil [4.16], and for creating through-wafer interconnects [4.17] (Figure 4.3).
115
Suspension: Fabrication
Figure 4.3 DRIE devices (a) Actuator-mirror structure fabricated on SOI (Prof. de Rooji, IMT), (b)
MicroSEM deflector structure (Prof. de Rooji, IMT), (c) A MEMS gyroscope fabricated on SOI
[4.11], and (d) Micro-machined intracellular needle [4.15], pillar etched using DRIE and then post
There are multiple commercial DRIE systems on the market by STS (Surface
Technology Systems), Alcatel, Oxford instruments, and Plasma-therm which are able to
thick wafers.
semiconductor etching where etched features are shallower and the silicon substrate’s
116
Suspension: Fabrication
bottom surface is not breached. In this section we will look at these special
section 4.3.1. The process depends on many parameters like gas flow rates, cycle times,
number of cycles or total process times, chamber pressure, chamber temperature, coil
power used to generate plasma, platten power to apply a bias voltage to the ions, and the
gases. In the systems used for fabricating this suspension, SF6 is used for etching
(though a small amount of Oxygen is added to remove any sticking residue in the
chamber or in the residue extraction system) and C4F8 is used for passivating the
The suspensions are fabricated using a Surface Technology Systems multiplex ICP
(Figure 4.4a). The etching chamber is maintained at a low pressure by a turbo pump.
The system has two independent 13.56 MHz RF power sources. As shown in Figure
4.4b, the Coil RF Matching unit is driven by a 1000 W RF power supply. The coil is
RF power source drives a Plasma Matching unit which is used to vary the RF bias
The etching chamber has a single feed on top for the etching/passivating and auxillary
gases (SF6, C4F8, O2, and Ar). The etching chamber is connected to an automatic load
lock which eliminates the need to open the chamber to load the wafer thus allowing the
chamber to be maintained continuously at low vacuum. The wafer is loaded into a twin
wafer capacity loading tray which automatically loads the wafer on the chuck in the
117
Suspension: Fabrication
etching chamber. The chuck is an electrostatic “clicks” plate which uses a high potential
(up to 1000V) across its coating of dielectric layer to grip the wafer. The chuck has 12
μm deep grooves on the top surface and a pressurized helium inlet. The wafer is kept at
a constant temperature of 40°C by helium flowing on its backside during the processing.
The chamber walls are kept at 60°C by water flowing on the outside of the chamber
wall to avoid any residue depositing on the sidewalls of the chamber. The etch chamber
Figure 4.4 (a) STS DRIE Kit [4.18], (b) STS ICP Multiplex process chamber Schematic (Source:
[4.19]).
The silicon etch rate, photoresist selectivity, profile shape of the etched trench, surface
quality and etch rate uniformity is dependent on various parameters such as gas flow
rates, platen and coil power, etch and passivate cycle times and chamber pressure. The
operator can change the etching and passivation times by changing the etch/passivate
time ratio while keeping the cycle time constant or by increasing the cycle time. The gas
flow rate for SF6, O2, C4F8 and Ar can be modified and is usually represented in sccm
(standard cubic centimeters per minute). The total etch time can be given and the
software automatically calculates the number of cycles needed. The coil and platten
power can be varied: coil power refers to RF power used for plasma generation, platten
118
Suspension: Fabrication
power is the bias between the chamber walls and substrate holder to accelerate the ions
vertically to achieve anisotropy. The chamber pressure is varied either by setting the
pressure to a specific milli-torr value or by varying the valve angle, also called APC
(automatic pressure control) angle in which case the pressure is measured by the flow
The basic requirement of successfully patterning the silicon wafer to desired geometry
is to develop a process flow which transfers the pattern from mask to the silicon wafer.
To etch areas of silicon selectively needs other areas on the wafer to be guarded against
the etchants. This needs a suitable masking material which can withstand the etchants
longer than the silicon to be etched. The etching process parameters are then tuned to
achieve the desired etch. An iteration of this cycle is generally needed to achieve the
desired result.
To transfer the design pattern onto silicon, a mask is prepared which transfers the layout
onto the masking layer (oxide, photoresist, nitride etc) so only areas which need etching
A key concern with plasma processes is the loading effect. In DRIE due to the
dependence of etch rate on silicon exposure area, feature rich regions etch slower than
feature scarce regions. This effect is generally known as microloading [4.20] and is due
to the reactant depleted condition in areas with higher silicon exposure (a higher
fresh reactants).
119
Suspension: Fabrication
Additionally smaller areas etch slower than larger areas which is commonly known as
etch lag [4.21]. To avoid both microloading and etch lag we need to design the mask so
that the silicon exposure is similar across the wafer and the etch channels are of similar
dimensions.
To achieve uniform etching throughout the wafer, the features are etched using constant
channel width on either side. This scheme is commonly referred to as halo mask [4.22].
An example is shown in Figure 4.5 which shows a spring and frame (dark grey) feature
Regions which are not structural parts, but are present to minimize silicon exposure and
give a constant channel width for all the etchings, are called “packing” pieces. After a
through-wafer etch, the “packing” pieces which are not attached to any structure on the
Figure 4.5 A halo mask and its various components for a 2D mask designed on a mask plate for
The etch channel width was set to 40 μm after doing test studies with varying etch gaps.
With the same process parameters, gaps wider than 40 μm etch faster but results in poor
120
Suspension: Fabrication
sidewall quality (see Figure 4.6). Smaller gaps take longer to etch through and for gaps
smaller than 20 μm, the etchant flux density is extremely low at the foot of the etch
Another aim of the process engineering is to achieve a single “process parameter set”
for the complete etch run instead of parameter ramping. Automatic process parameter
ramping though a useful method available with DRIE system needs to be constantly
modified with changing mask designs, resulting in a higher process development time.
Figure 4.6d shows etched sidewalls for varying trench width. For trench width above 40
μm we can see significant damage to the sidewall. Additionally Figure 4.6a-c show the
Figure 4.6 Sidewall quality for varying channel/trench width, (a) 40μm gap, (b) 60μm gap showing
a widening of the gap towards the bottom of the trench as the profile diverges away from vertical,
(c) 80μm gap shows the increase in profile shape divergence from vertical with bigger gap, (d)
beams with gap varying from 40-400μm show the increased breakdown of the sidewall with larger
gap.
121
Suspension: Fabrication
Etching silicon needs a masking layer with higher resistance to plasma etching than
silicon. Typical masking materials for DRIE are photoresist, nitride and oxide. The etch
rate for the DRIE system is on the order of 1-2.5 μm/min, though it can be as high as
10-50 μm/min for the latest high etch rate DRIE systems. At 2μm/min etch rate it takes
4 hours 20 minutes to etch a 525 μm thick wafer so we need a mask layer that can last
such a long etch. Thick photoresist and oxide are therefore investigated as appropriate
spin 9 μm (Figure 4.7) and 14 μm thick AZ9260. The selectivity of the resist is 80:1 for
the DRIE process parameters, so requires about 7 μm thick resist coating for etching a
Thick oxide was also investigated as a mask material as oxide has a high selectivity to
SF6 plasma etch, typically 180-200:1 [4.24] requiring approximately a 3 μm thick oxide
122
Suspension: Fabrication
layer for the etch. Silicon wafers with 3 μm thick oxide grown using wet oxidation
process is procured. The oxide is then dry etched using CHF3 plasma. The sidewall and
pattern transfer to oxide is poor due to high-temperature damage to the photoresist used
for oxide etching and the isotropic nature of the etch (Figure 4.8). Etching oxide using
buffered HF (49%) gives poor results due to the isotropic nature of the etch leading to
Figure 4.8 (a) SEM of 5 μm thick oxide etched using CHF3, (b) SEM of damaged photoresist
For through-wafer DRIE we decided to use photoresist as the masking layer, Table 4.1
presents the pros and cons of oxide and photoresist as a masking layer in DRIE. The
photoresist layer needs to be thicker than oxide to withstand the etchants but in all other
respect they are similar or better than oxide when used as a masking layer. Photoresists
are organics and provide an additional supply of carbon during the etch thus helping the
overall passivation.
Even though oxide is removed from consideration as a masking layer for silicon DRIE,
a thin layer of oxide is grown that prevents photoresist mask erosion and acts as a
insulation layer for any electrical circuit fabricated over the suspension. A process is
123
Suspension: Fabrication
developed to grow 200 nm and 500 nm thermal oxide on silicon wafer and etch using
CHF3.
Table 4.1 Comparison of oxide and photoresist as a masking layer for DRIE.
An oxide etch recipe is developed to avoid damaging the photoresist mask, which also
works as mask for the silicon etch. The key reason for the damage to photoresist during
oxide etching (Figure 4.8b) is the high temperature that the photoresist was exposed to
during the etch. The temperature of the wafer exceeded 150°C in just 5 minutes of etch.
To make sure the temperature of the wafer is not allowed to exceed 60°C the etch cycle
is kept short, 1 minute long and the wafer is cooled for 2 minutes before resuming the
etch. This time-multiplexing of etching and cooling cycles results in good etch profile
Availability of high-rate DRIE systems which allows highly anisotropic silicon etching,
makes wafer thick suspensions possible. The process needs to be tuned to perform
wafer thick etching and precautions are taken to avoid etchant contact with the wafer
124
Suspension: Fabrication
holder by using a handle wafer underneath the device wafer. The mask needs to be thick
enough to sustain etchant bombardment for the total etch time. The etch area needs to be
optimized (through mask design) to attain a maximum rate of etch to minimize etch cost
Figure 4.9 Process flow schematic for through-wafer silicon etching using DRIE.
Figure 4.9 shows typical steps in through-wafer etching a silicon wafer. A 200 nm SiO2
layer is first grown on the silicon wafer and then covered with photoresist. The mask
using CHF3 plasma, thereafter the wafer undergoes a partial DRIE short of the etch
breaching the backside of the wafer. The wafer is then removed from the etcher and
bonded to a handle wafer using a temporary adhesive. The etch is then resumed to
completion. The wafer is thereafter released from the handle wafer and the photoresist is
stripped yielding the final device. The need for a handle wafer and the subsequent
release process are the additional steps involved over conventional DRIE. In the next
section we discuss various ways of protecting the etcher substrate holder from plasma
125
Suspension: Fabrication
protective surface on the other side of the wafer so that when the etch gets through the
first wafer it is contained by the underlying surface. In DRIE we need a protective layer
on the backside of the wafer which acts as an etch stop to avoid damaging the chuck
(substrate holder in the etcher). The STS system flows helium on the backside of wafer
stop the helium leaking into the plasma chamber. It is quite common to remove the
various techniques for temporarily bonding wafers, which are mechanically robust and
releasable. In DRIE the added concern of etch variation with temperature makes it
critical to have a bonding method which does not affect the thermal sinking of the
The protective surface is applied either at the start of the etch or after a significant
portion of the etch has finished though always before the etch has breached the wafer
backside. The surface is generally another silicon wafer bonded to the substrate wafer. It
protects the chuck which is used to bias the substrate to vertically direct the ions in the
plasma, and also to flow helium for maintaining the wafer at a low temperature. In the
absence of a handle wafer the ions etch through the substrate wafer causing micro holes
leading to helium leaks into the chamber and premature termination of the etch run
(dependent on the Helium leak rate value set in the etching process parameter). Directly
exposing the chuck to plasma ions reduces its life due to failure of the dielectric coating.
The handle layer is also needed if there are loose structures on the wafer which will
126
Suspension: Fabrication
come off when the wafer is etched through; this is always the case with halo geometries.
Etched structures can also fall off when using the DRIE as a way to dice the wafer;
though using tabs (die connecting structures) can avoid the dies from falling onto the
clicks plate during removal. Etching small wafers or a piece of wafer also requires a
The etch rate, sidewall surface smoothness, trench profile, and mask selectivity varies
with temperature of the substrate wafer. It is normally seen that a higher temperature of
the substrate wafer leads to a fast erosion of the mask layer so it becomes crucial to
have a thermally conductive protective layer on the backside of the wafer which allows
the Helium flow on the backside of wafer to keep the ‘substrate-handle wafer’ sandwich
section 4.5.9) which happens at the silicon-insulator interface, which is a concern if the
This can be avoided by having an electrically conducting layer on the wafer backside.
The handle wafer or the protective layer should be easy to release after etch is
completed, so that the fragile devices can be handled without any mechanically or
Mechanically strong, to sustain the helium pressure difference and hold the
127
Suspension: Fabrication
wafer)
Process cleanliness
We investigated various ways of bonding a handle wafer to the substrate wafer using
various adhesives as well as tested a metal protective layer. The pros and cons of each
It is evident from the comparison chart of wafer bonding technique that the best method
for through-wafer etching is to use a metal layer which is mechanically strong thus
removing the need for a separate handle wafer. The metal layer should be compatible
with the overall process and suitable for electroplating to 10-20 μm thickness. Nickel
plated to 5 μm thickness was found to be suitable for use as a protective layer. The
disadvantage of metal on the wafer backside is the necessity for a chemical etch to
128
Suspension: Fabrication
remove the metal layer to release the features. The next best method is to use a thin
photoresist layer to bond the wafers. It is post-baked to improve adhesion and thermal
For relatively shorter etch, where silicon is already etched through most of the wafer
thickness, a handle wafer is attached using a thin layer of photoresist. Shipley SPR 1813
was spun on a single-side-polished wafer and then bonded to the device wafer. The
photoresist bonded wafer stack was found to perform good thermal sinking for shorter
The effect of each of the processing parameters is covered in detail by Chen et al. [4.26-
28]. The major parameters which can be varied during the etch are etch:passivate ratio,
cycle (etch+passivate) time, coil power, platen power, chamber pressure or APC, gas
flow rates, gas flow rate ratio (SF6 : C4F8), and some minor ones like O2 flow rate with
SF6, and overlap of etching and passivation steps. All these parameters have an effect
on the silicon etch rate, photoresist etch rate, sidewall quality, profile shape of the etch
SF6 and C4F8 flow rates: Increasing the flow of SF6 increases the concentration of the
etchants in the chamber leading to higher etch rate of silicon, as well as higher etch rate
of photoresist decreasing selectivity. At low SF6 flow rates the concentration of fluorine
is limited and so the etch rate plateaus even if other parameters which increase
ionization and dissociation of SF6 are increased, the amount of SiF4 that redeposits also
increases [4.27]. Increasing the flow rate of C4F8 increases the passivation species CFx
129
Suspension: Fabrication
decreasing the silicon etch rate and increasing the selectivity of photoresist. The
deposition of the Teflon like CFx polymer on the sidewall is self-limiting [4.29, 30] but
is dependent on the ion bombardment and with a higher ion density it can be
concentration of the etchant and passivating species, the sidewall quality can vary from
smooth to extremely rough, and trench profile shape can go from positive for a etchant
increasing trench depth, the flow rate of SF6 needs to be increased to counteract the
Etch:Passivate ratio (EPR): Similar to the etching and passivating gas flow rates, a
higher EPR will lead to higher etchant species density compared to passivation species.
Generally for deep etches it is desirable to have a higher EPR as the etch depth increases
as the density of ions reaching the trench bottom decreases with increasing trench depth.
Cycle time: The total time a single set of etch and passivate steps take is the cycle time.
Increasing the cycle time for an etching dominant EPR leads to larger scalloping on the
sidewalls as in each cycle the etchant radicals are active for longer leading to increased
Coil power: The ionisation and dissociation of SF6 and C4F8 is dependent on the coil
power. Higher coil power leads to a higher density of the ions and radicals in the
plasma. Depending on the EPR and gas flow rates, the concentration of the species
130
Suspension: Fabrication
Platen (electrode) power: The platen bias is typically used to accelerate the ions only
during the etching step so increasing the platen power increases the directionality of the
ions allowing for a higher density of ions to reach the trench bottom increasing the
silicon etch rate, and verticality of the trench. As the platen power has no affect on the
radicals which are neutral species, the isotropic part of the etch remains unaffected but
due to an increase in vertical etch rate the scallop size can be reduced by decreasing the
density of the etchants without affecting the vertical silicon etch rate. Increasing the
platen power increases the photoresist etch rate, reducing selectivity, as the photoresist
APC: The silicon etch rate increases initially with increasing pressure as the fluorine
density increases but then as the pressure is increased further it starts decreasing as the
mean free path for the ions decrease with increasing pressure lowering the ion energy.
Etching and Passivation step overlap: The overlap of SF6 extends the etching step and
passivating plasma. This helps maintain a stable glow discharge which leads to
improved uniformity. For electronic devices it is critical at times to have a stable plasma
as a nonuniform plasma can induce current in the electronic circuit damaging it [4.27].
O2 flow during etching: A small amount of Oxygen is added to SF6 during the etching
step, this increases both the silicon and photoresist etch rates. The selectivity is however
unaffected as the etch rate increase is almost similar. Oxygen plasma also removes the
passivating layer though its affect on reducing passivation film from the sidewalls is
131
Suspension: Fabrication
limited due to a lower flux at the sidewall. Oxygen is also added in the system to reduce
sulphur deposition on the chamber sidewall and residue in the extraction pipes.
DRIE is dependent on a large number of process parameters and has a wide process
window allowing the process engineer a large amount of flexibility. The initial goal of
any process engineering is generally to remove or reduce any deleterious effects of the
processes, as we see in the next section DRIE has a number of problems which need to
Plasma induced damage are a common occurrence in any plasma processing. Some of
these effects are inherent to plasma-surface interaction and can be reduced by changing
desirable anisotropic etching are evident as deleterious artefacts which can compromise
the performance of the fabricated structures. In section 4.5.1 to 4.5.6 we will look at
some of the deleterious artefacts which can be easily removed or do not compromise the
wafer etching; smooth sidewalls are important to increase the fracture strength of the
silicon springs as well as to reduce structural defects that can affect the quality factor.
Section 4.5.8 concerns etch profile shapes and ways to control them. We examine a
4.5.1 Grassing
During etching, large open areas develop tall sharp silicon features due to incomplete
removal [4.28], this is particularly common with increasing etch depth due to lack of
132
Suspension: Fabrication
sufficient etchant species reaching the etch channel floor. Figure 4.10 shows grass
Grass formation is generally lower in smaller areas and is reduced by decreasing the
chamber pressure. Enclosing the etched area with packing pieces reduce the effective
etch area thus increasing the etchant density leading to reduced grassing. Decreasing the
APC angle or chamber pressure reduces grass formation. Increasing the platen power
decreases grass build-up by increasing the directionality and energy of the ions. Figure
4.11 shows reduction in grass formation with increasing platen power while keeping all
processing parameters constant. At 12W platen power the complete etch trench is
covered with grass (Figure 4.11a), increasing the power to 14W removes the grass from
the centre of the large trenches but still has grass near trench sidewalls. At 16W there is
no grass in the trench but has a grass like curtain at the sidewall which is completely
133
Suspension: Fabrication
Figure 4.11 Reduced grass formation under increasing platen power, (a) 12W: trench is completely
covered with grass, (b) 14W: the centre of the trench bottom is clear but grass is still forming along
the sidewall, (c) 16W: clear trench bottom but the sidewalls have grass curtains, (d) 20W: the
trench bottom and the sidewall is etched cleanly without any grass formation.
Larger areas etch faster than smaller areas [4.21, 26] due to the presence of higher
etchant species density in the larger area, Figure 4.12(a). The etch rate variation
increases with etch depth as ions find it harder to go down narrow thin channels ending
up somewhere on the sidewall due to slight deviation from vertical directionality of the
ion, Figure 4.12(b). Trenches of similar dimension also have etch rate difference
depending on their location on the wafer. Areas close to wafer edge etch faster than
134
Suspension: Fabrication
Figure 4.12 (a) SEM of etch cross section showing etch lag between small and large trench, (b) Etch
depth vs. channel width (all dimensions in μm) for etch 5 hours long using process parameters from
Table 4.4.
In through-wafer etching “etch lag” results in slower etch rates for smaller trenches
SOI and through-wafer etching this detrimental artifact can lead to lateral silicon
Etch lag can be reduced by using halo mask technique to get a uniform etch area, thus
allowing the process designer to only develop a process for specific trench width.
Reducing the etch chamber pressure increases the mean free path of the ions allowing
them to reach deeper trenches. SF6 flow rate is seen to be the dominant process
135
Suspension: Fabrication
Creating a process situation where the etchant density is optimized to be nearly similar
for both large and small area will reduce etch lag but generally gives rise to grass
formation in larger areas (Figure 4.13). This also requires process parameter ramping to
Figure 4.13 Process parameters optimized to reduce etch lag lead to grass formation in larger width
The loading effect is an interplay between the etch rate and the silicon exposure area. It
macroloading the effect leads to variation in etch rate for similar features at different
positions on the wafer, the effect is due to the dependence of reaction rate on etching
surface area [4.20]. There is better replenishment of etchant species towards the edge of
the wafer than at the center resulting in higher etch rate at the edge of the wafer than the
center. Due to macroloading the features on the edge of the wafer are overetched
136
Suspension: Fabrication
Microloading is the relationship between local etch rate and pattern density. Features in
high pattern density areas of a layout experience more competition for reactants leading
to a negative gradient in reactant flux, and to a slower etch rate. As we saw in section
4.3.2, designing a halo mask geometry reduces microloading by reducing the silicon
Another detrimental artifact common to DRIE is top edge damage or cavernous etch at
the top of the structure (Figure 4.14). This effect is particularly prominent when the
platen power is high. Experiments carried out at 12W platen power resulted in
extremely small top edge damage whereas increasing the platen power to 20W resulted
in significant damage to the top edge. The most probable cause of this artefact is
charging of the mask which deviates the ions off the vertical axis and into the sidewall
[4.31]. Exposing the sample to air after every hour of etch reduces the damage to the top
edge, which suggests charging of the mask as a possible reason for this damage.
137
Suspension: Fabrication
Incomplete development of photoresist can lead to micro masking of silicon in the etch,
which creates oxford spires in the etch channel (Figure 4.15). The micro-masking may
lead to significant constriction of the etch channel resulting in a severe etch lag and
of etch end products from the etch chamber, leads to these materials redepositing on the
de-scum is used after developing the photoresist. The dominant process parameters
affecting micromasking are SF6 flow rate and chamber pressure. Increasing in SF6 flow
4.5.6 Scalloping
Scalloping (Figure 4.16a) is a normal artifact produced by DRIE due to the time
multiplexing of etch and passivate cycles. The size of the scallop depends on the etch
cycle time and etch rate, the rms height of scalloping for a general processing conditions
138
Suspension: Fabrication
is in the range of 5-50 nm. Figure 4.16b shows an AFM scan of scalloping on the
trench sidewall giving an rms roughness of 8 nm for a process with 130 sccm SF6 flow
rate, 85 sccm C4F8 flow rate and a coil power of 600 W, platen power of 20 W and APC
of 78%. Generally faster etch rates will lead to larger scallops (Figure 4.16c). Scalloping
is not a major problem for most mechanical devices unless they need a mirror like
surface. Scalloping can be removed or reduced by either increasing the etch rate or by
performing a smoothing operation after the etch by oxidizing the surfaces and etching
Figure 4.16 Scalloping (a) Scalloping on a sidewall, (b) AFM scan of top edge of sidewall shows
scallops (courtesy: Sanjay Vijendran), (c) SEM of large scalloping on trench sidewall suggesting
The quality of sidewalls resulting from through-wafer DRIE has a significant effect on
the quality factor of the suspension as well as the reliability of the device (as structures
with rough surfaces tend to have lower fracture strength). The sidewalls of the trench as
shown in Figure 4.17 show a surface whose roughness increases with increasing depth
of the trench. It becomes important to understand the process parameters affecting the
fabricating test structures and observing them under SEM, interferometer and AFM. We
also look at the relationship between etch gap and sidewall quality.
139
Suspension: Fabrication
This work was done in collaboration and the full results are published in Pike et
al.[4.32].
Figure 4.17 SEM image of sidewall quality of 500 μm high sidewall. a,b,c show tapping mode AFM
scan of the top (a), middle (b) and bottom (c) of the trench sidewall. The roughness of the sidewall
increases from peak-to-peak 0.4 μm at the top of the trench sidewall to 0.7 μm at the middle of the
140
Suspension: Fabrication
were etched in 525-µm-thick Si (100) wafers (Figure 4.18). The etch pattern was
developed in a 15 µm resist mask and the wafer mounted on a Silicon backing wafer
Figure 4.18 Schematic of test structure showing region of interest (ROI). A tapered trench was
selected to examine sidewall surface quality with varying channel width. The individual structures
After through-wafer etching, the beams were cleaved out of the structure to allow for
The SEM is a LEO 1450 VP (Leica Electron Optics) operated at 30 kV, incorporating a
analysis. A Nanosurf Easyscan AFM was used to image the sidewalls at selected points
along the etched beams, particularly the scalloped regions visible within 15 μm of the
141
Suspension: Fabrication
top surface. All measurements were done in dynamic mode AFM, with Silicon tips.
Scan areas were typically 50×50 μm, at scan speeds of 25 - 50 μm per second.
In order to examine the topography quantitatively over large areas, a scanning white-
light interferometer (ZYGO Newview 200) was used for non-contact mapping of the
etched surface.
Figure 4.19 Sidewall SEM of through-wafer deep reactive-ion etched beam. Zones of varying
sidewall morphology are labelled as: A. Overetched, B. Ion damaged, C. Overpassivated, and D.
Striations.
Figure 4.19 shows the sidewall from one side of a low-angle laterally tapering trench
etched completely through a 525-µm silicon wafer. The trench width in this image
varies from 40 µm to 300 µm, with a taper angle of 6°. Overall, the sidewall is
smoothest at the narrower end of the trench, although puncturing of the sidewall with
narrower end of the trench for the entire depth of the sidewall. As the trench widens to
greater than 100 µm, vertical striations increasingly emerge, resulting in a columnar
appearance for the lower portion of the sidewall. At wider trench width, areas of
complete sidewall breakdown occur from about 50 µm and higher. Optical profilometry
using white-light interferometry shows that at the narrower trench width, bowing of the
142
Suspension: Fabrication
sidewall occurs (Figure 4.20). A cross section of the sidewall shows that the reentrant
surface emerges towards the bottom. In contrast, for wider trenches, the profile is near
vertical although the roughness of the sidewall towards the bottom prevents
interferometric measurement.
Figure 4.20 Interferometric measurement of sidewall topography near narrow trench end. The ion
bombardment causes holes in the sidewall near the trench bottom where the re-entrant surface
For good-quality DRIE, etching and passivation should remain in balance at the
sidewalls throughout the etch, although etching should dominate at the trench floor. We
will analyze these results in terms of this balance. For the sidewalls, if etching
dominates, the sidewall polymer will be removed allowing subsequent Silicon etching
with the slow erosion of the initial scalloping. If passivation dominates, the polymer
layer will thicken with time, being thickest towards the top of the sidewall.
143
Suspension: Fabrication
The loss of scalloping of the sidewalls for the smaller trench width can be seen as a
result of over-etching. Bowing often accompanies such over-etching [4.33]. The major
problem for this portion of the sidewall is the damage towards the bottom of the
sidewall. Combining the SEM images of the distribution of this damage with the optical
profilometer data shows that the onset of damage occurs as the reentrant surface
emerges from under the top surface of the wafer. Low-angle SEM imaging shows these
holes are circular as viewed vertically downwards (Figure 4.21). The probable cause of
Figure 4.21 SEM picture showing the damage observed at narrow etch-gaps towards the trench
bottom caused by ion bombardment (Inset: zoomed, rotated view at a tilt angle of 85º) (courtesy:
Werner Karl).
144
Suspension: Fabrication
At larger trench widths, passivation dominates. Figure 4.22 shows in detail the buildup
of polymer at the top of the sidewall of the through-etched wafer. Energy dispersive X-
ray micro-analysis confirmed the fluorine content of this layer, Figure 4.23. The
scalloping is evident under the polymer, as is the onset of sidewall breakdown 100 µm
from the top. A portion of polymer has detached from the sidewall, showing that the
thicker passivation itself evolves into a non-uniform film. Sidewalls below passivation
Figure 4.22 Passivation build-up observed at wide etch-gaps leading to striations and therefore to
rough sidewalls.
145
Suspension: Fabrication
Figure 4.23 EDX analysis of the beam shows fluorine presence in the darker areas which were
charging up in the SEM, proving the hypothesis of CF2 deposition. (a) SEM image of the beam
section analysed for Fluorine content showing a higher concentration of carbon and fluorine at the
top darker region of SEM image than at the ligher part towards the bottom of the sidewall, (b)
Elemental map of the sidewall highlighting (red) the carbon and fluorine dominated regions, and
(c) Linescan from the top edge of the beam to the bottom edge showing the concentration of
Charging in the SEM, evident as pale borders at the polymer edges, is suggestive of a
potential cause for polymer-film breakdown in the DRIE: it is known that charging of
the polymer film disrupts the ion flux during the etch step [4.34]. Polymer buildup
lower-quality sidewall.
146
Suspension: Fabrication
The passivation layer consists of single-chain carbon difluoride polymer. As the carbon-
carbon and carbon-fluorine bonds are two of the strongest bonds in organic chemistry,
considerable energy is required to either grow or shorten the polymer chains. For
atom from the chain. CF2 radicals, essentially PTFE monomers, can then add to the
chain, with the final addition of a terminating fluorine radical [4.30]. For reduction of
the chain, the process is reversed, and the polymer is etched. High-energy vertically-
directed ions are present during the etching steps, while during passivation, when the
accelerating voltage is zero, lower-energy ions are scattered isotropically from the
At the floor of the trench, there will be an incoming flux of CF2 radicals during the
passivation step. Before any silicon etching can take place during the etching step all the
polymer deposited on the floor during the passivation step must be removed, again as
CF2 radicals. Redeposition of the polymer from the trench bottom to the sidewalls has
[4.35-39]. Analyses and experimental data have indicated that the incoming flux of CF2
radicals as etching products vary at different heights of the trench sidewall [4.39] and
there will be a reduced, lower-incident-angle flux of CF2 radicals towards the top of the
This provides an explanation for etching of the passivation towards the top of the
sidewalls at narrower trench widths. The flux of redeposited CF2 is reduced as the
147
Suspension: Fabrication
trench is etched, causing the polymer thickness deposited on the sidewall over an entire
cycle to be less than the thickness etched, inevitably leading to eventual overetching of
the sidewall. Overpassivation of the sidewalls at wider channel widths can be explained
by the increased flux of CF2 radicals from the broader floor of the trench as more
material is etched from the floors of the wider trench and wider range of deposition
angle is available for the sputtered CF2 radicals. The passivation in this case grows more
than it is etched during each cycle, causing build-up of passivation beyond a critical
observations, other effects could contribute. During the passivation step, the flux of
activating ions at sidewalls might be expected to drop with trench depth. However, the
flux at the top of the sidewalls would be the same for all trench widths, whereas a
ion-flux variations during the DRIE cycles are unlikely to provide a complete
explanation.
The sidewall quality of through-wafer DRIE structures is very dependent on the aspect
ratio of the etched channels. Below about 100 µm etch width, sidewalls tend to be
smooth, albeit overetched, with some bowing leading to ion damage towards the bottom
breakdown in polymer coverage, causing vertical striations and a very rough surface.
width, and a transition to overetching with etch depth for the narrower channels.
148
Suspension: Fabrication
Redeposition of passivation from the trench floor during the etch step is suggested as an
As the aspect ratio of the trenches changes during etching, no single set of process
parameters can be expected to optimize the balance between sidewall passivation and
etch for a through-wafer etch. Ramping of parameters may be one route to improved
sidewall quality.
In DRIE, the profile of the etched structure is controlled by successively varying the
etching and passivation to achieve highly anisotropic features. As shown in Figure 4.24
the etch trench can be either sloping inwards (narrower at bottom than at top), which is
bow (the middle section is etched more laterally than the top or bottom).
An ability to understand, measure and tailor profile shapes is extremely important from
a mechanical standpoint as well as when the structure is used for devices with a strict
profile shape requirement. For example, a deviation from the ideal vertical profile leads
to a variation in the natural frequency of the lateral suspension (see section 5.3).
149
Suspension: Fabrication
The profile evolving with time is dependent on the charge accumulation, ion energy
distribution and etchant/passivant ratio in various parts of the trench, which are
Channel profile and bowing has been discussed in Chen et al. [4.26] and Ayon et al.
[4.40] and ways to control anisotropy and profile have been discussed. Singh et al.
[4.41] presents a model for profile simulation in reactive ion etching. Previous works
[4.9, 42] have mostly focused on how to achieve high anisotropy and control profile and
reduce bowing, but as we present here, bowing can be a boon for some applications, and
the process which causes it, will help designers use various profile shapes and bowing
to their advantage. The effect of process conditions on DRIE etch shape is discussed.
The relative density of etching and passivating species in each cycle decides the profile
of the trench sidewalls. Etching and passivating time ratio, process cycle time, chamber
pressure, SF6 and C4F8 gas flow rate, coil power, and platen power affects the density of
etching and passivating species in each cycle. The cause for the various profile shapes
are:
results in non-complete removal of passivation from the bottom of the trench resulting
in narrowing of the trench, Figure 4.24(a). This leads to a thicker beam cross section
with depth of the etch. Chen et al. [4.26] mentions a shorter etch cycle leading to
150
Suspension: Fabrication
Positive profile: When the etch cycle is longer than the passivation cycle, the
passivation is removed from the sidewall of the trench (towards the bottom of the
trench) leading to a wider trench bottom as the etch progresses, Figure 4.24(b).
Bowing: Removal of passivation coating from the trench sidewall faster than it is
formed causes and aggravates bowing, Figure 4.24(c). Ishihara [4.9], Sun [4.42] show
bowing to be caused by charging of SiO2 mask or a buried layer. In our HAR etching,
we have seen bowing when using photoresist as a mask and the bowing is quite
symmetric with a thin neck and wider top and bottom. Having a SiO2 mask helps
preserve the pattern size as photoresist is prone to edge erosion which changes the
effective masked area to the etching species. Lii [4.37] investigated the bowing effect in
a silicon trench under plasma etching and proposes two reasons for the undercutting of
the sidewall: (a) etching by divergent ions scattered by ion-molecule collisions in the
sheath, and (b) another mechanism involving the scattering of ions from the edge of the
mask. Lii also shows that redeposition of passivation polymer during etching cycle
SF6 plasma consists of positive and negatively charged ions, radicals and atoms
(equation 4.1). The etching mechanisms are considered to be independent of each other
but Lii [4.37] demonstrates an amplification in reaction between ions and reactive
neutral radicals as the trench depth increases. The concentration of radicals is at least
two orders of magnitude higher than ions and the radical flux (~1019 cm-2s-1) is also
151
Suspension: Fabrication
much higher than the ion flux (~1015 cm-2s-1)[4.5]. This could have a very prominent
effect on the bowing of the sidewall as we notice an increase in the isotropic etch with
increased etch depth. At about the middle of the wafer thickness the etching starts
lagging behind the passivation and the profile starts turning negative or reentrant.
Chang et al. [4.43] discuss surface potential buildup due to flux imbalance of isotropic
electrons and anisotropic ions leading to lateral etching of the sidewall (Figure 4.25).
Figure 4.25 Surface potential build-up due to flux imbalance of isotropic electrons and anisotropic
Due to the isotropic nature of the electron and negative ions in the plasma, the top
surface of the sidewall has more negative charge which deflects the positive ions
Most of the analysis of bowing in the literature has been for shallow trenches and SOI,
so there is not a coherent theory that provides a complete explanation for the top edge
152
Suspension: Fabrication
becoming positive, then a maximum undercut causing necking and then the trench
APC angle or Chamber pressure: Chamber pressure has a strong effect on the etch
profile. With higher APC angle, the trench gets narrower with depth; decreasing the
APC angle improves the profile which converges towards vertical sidewalls, and
reducing the pressure further leads to a wider trench bottom. At the same APC angle,
larger areas etch to a wider floor than top, which could be due to an effective increase in
etchant density at the floor of a large trench than a smaller trench at the same pressure.
Etch/Passivate ratio: Increasing the etch cycle compared to the passivation cycle leads
to an increase in the isotropic nature of the etch which leads to a higher bowing of the
profile. It also results in a wider trench bottom than a similar etch with lower
etch/passivation ratio.
Platen Power: Platen power defines the energy of the ions, higher platen power
accelerates the ions faster vertically leading to better profile control as a significant
For tall structures such as features created by through-wafer etching, the process
parameters were selected to avoid grass formation in the etch trench, which means the
process is etch dominant. During the run the process is etch dominant as is evident from
etching of larger trenches which tend to widen out towards the bottom.
153
Suspension: Fabrication
Towards the start of the process, the trenches tend to have a wider bottom than top due
to the process being etch dominant but after some depth the passivation polymer
removed from the trench bottom is not completely removed from the trench which leads
to it depositing back on the sidewall. As the trench etch depth increases more
redeposition occurs on the sidewall thus shadowing the corner of the trench bottom
As the trench etch depth increases the etchant density starts reducing, and ultimately the
overall process for that trench may turn from etch dominant to passivation dominated,
which will also cause narrowing of the trench as the etch depth becomes large.
A bowed profile can be advantageous where a large surface area is needed on the top of
the feature for traces to run but the effect of bowing needs to be compensated to achieve
the mechanical performance target (see section 5.3). Due to bowing the effective
thickness of the beam will be much lower than for the ideal rectangular profile case. In
the case of lateral suspension, the profile bowing is incorporated in the design to create
springs with lower spring stiffness but larger top surface width. The larger top surface
width provides more leeway for the metal traces to be thicker and further apart thus
4.5.9 Notching
plasma, a notching (or footing) effect occurs towards the end of the etch at the bottom
silicon-insulator interface. Such an interface will be present as the etched silicon wafer
154
Suspension: Fabrication
will have either an oxide layer to avoid break through of the plasma or the wafer will be
bonded to a handle wafer using either photoresist or other adhesives. Notching damage
always a concern towards the end of a through-wafer etch at the bottom of the
surface charging can produce a final lateral undercut at this interface, Figure 4.26.
Figure 4.26 Notching damage at the bottom of a through-wafer etched feature. The wafer was
Due to the directional nature of the positive ions, a substantial proportion will travel to
the foot of the trench where they positively charge the insulator layer. In contrast,
electrons, due to their isotropic directionality, impinge on the top surface of the wafer
but cannot get to the bottom of a high aspect ratio trench to neutralize any positive
155
Suspension: Fabrication
charge build-up in the insulator layer. The positively charged insulator surface then
deflects the ions towards the sidewall causing a lateral etch (Figure 4.27).
Figure 4.27 Charge distribution in deep trench and positive charge accumulation at Silicon-
Plasma tools and process conditions have been modified in the past to alleviate and
remove notching. Previous works have focused on ways to engineer the ion distribution
in order to reduce the charge accumulation or increase the ion directionality. Some of
the mechanisms proposed are low temperature etching [4.2], increasing bias voltage to
reduce ion deflection, decreasing chamber pressure [4.5] to improve ion directionality,
increasing the electron irradiance angle to decrease the potential buildup [4.46] and
pulsed plasma etching [4.47]. Most of these mechanisms have been demonstrated to be
effective for shallow etching. Except for pulsed plasma etching, these techniques are not
156
Suspension: Fabrication
Surface Technology Systems (STS) has adapted the pulsed plasma etching for its time-
Running the plasma at low frequency generates more low energy low directionality ions
which neutralize the charge build-up at the top of the trench. Reducing the negative
potential at the top of the trench facilitates more electrons into the trench alleviating the
positive charge build-up at the bottom of the trench. This technique has been
successfully demonstrated to reduce notching in SOI etches [4.44]. The major limitation
is the need for an endpoint detection to stop the normal etch mode and switch to the
Changing the ion angular distribution also affects the etch rate, anisotropy, etch profile
evolution and mask selectivity. Methods employing ion distribution engineering are
dependent on the aspect ratio (RIE lag), pattern factor (silicon exposure), and plasma
Hence an alternative that has been explored is the inclusion of a conductive layer which
can form a path to neutralize the positive ions at the end of the etch. Fransilla et al
[4.48], Rasmussen et al [4.49] have used a metal coating on the backside to work both
as an etch stop and to eliminate notching. Chabloz et al. [4.50] and Kim and Kim [4.51]
demonstrated using a metal layer on a glass wafer anodically bonded to the backside of
157
Suspension: Fabrication
We examine the use of metal coating on wafer backside as a method to avoid notching
in through-wafer silicon etching [4.52]. This is achieved by changing the process flow
the wafer backside. The etch rate and etch characteristics are not affected, as the process
parameters remain unchanged and the etching is done while operating the plasma at
silicon wafers. The process flow schematic is shown in Figure 4.28. The silicon wafers
are 525 μm thick. The silicon wafer was coated with a thin backside metal layer using
either an evaporator or sputterer. The front side of the wafers was coated with a 9 μm
thick AZ 9260 photoresist mask. The wafer was etched for 3 hours and then attached to
a silicon handle wafer using thin Shipley S-1813 resist. Subsequently it was etched until
the trenches were visibly clear. The wafer was then overetched for an additional 15
To investigate the mode of charge dissipation which eliminates notching at the bottom
158
Suspension: Fabrication
Figure 4.28 Process flow schematic for through-wafer etching using a metal backside coating to
eliminate notching.
Figure 4.29 Schematic of etch setup with metal coating directly on the wafer backside and then
Notching was compared for two silicon wafers etched using identical DRIE etch
parameters. One wafer had no backside metal layer; the other wafer had an Aluminum
(Al) backside metal layer as shown in Figure 4.29. The silicon wafers were then bonded
to a handle wafer using photoresist. The wafer without any metal coating had extensive
notching as seen in Figure 4.30a whereas the wafer with Aluminium coating on the
backside showed no signs of notching (Figure 4.30b) after the etch. This experiment
159
Suspension: Fabrication
demonstrated that by using the metal directly on device wafer backside one can
Figure 4.30 (a) Notching damage observed at the foot of a through-wafer etched structure; the
wafer was bonded to a handle wafer using thin photoresist. (b) The same structure on a wafer with
a thin metal coating on the backside; no damage at the foot of the sidewall is visible.
Similar notching damage was encountered for silicon wafers without backside metal
layers bonded to the handle wafer using thermally conductive pastes such as cool-
grease. Both photoresist and cool-grease are electrically insulating. It was observed that
for a thermally insulating adhesive (like photoresist), the damage was more extensive.
Trenches with similar dimensions were etched in a silicon wafer with and without
backside metal coating. The wafer without metal coating showed extensive lateral
etching at the etch foot, Figure 4.31a. The same trench shows a much cleaner surface
towards the bottom of the wafer, if a thin aluminium backside layer is used, Figure
4.31b.
160
Suspension: Fabrication
Figure 4.31 Backside of an etched wafer. (a) Etched trenches with no backside metal coating,
showing extensive notching. (b) Etched trenches with backside metal coating, showing no visible
notching.
The absence of an insulator interface eliminates any charge accumulation at the foot of
the trench. During overetching of the structures which is necessary to compensate for
etch lag, the positive charges of the reactive ions impinging on the metal layer are
neutralized. This is evident from no observable silicon damage due to a deviation of the
Figure 4.32 Charge dissipation at metal surface in a silicon trench with a metal coating on the
wafer backside.
161
Suspension: Fabrication
Figure 4.33 (a) Schematic of test setup with metal coated handle wafer glued to the device wafer at
the outside edges, (b) Schematic of the test setup with silicon handle wafer bonded to the device
To investigate whether the metal layer needs to be connected to the device wafer to
avoid notching, the handle wafer was coated with metal and the device wafer bonded to
it using photoresist patterned in a periphery-ring (Figure 4.33). Once the etch proceeds
through the wafer thickness the handle wafer’s metal layer is exposed.
We observed no notching at the bottom of the structures. This leads us to believe that
charges from any impinging ions dissipated through the metal layer into the handle
wafer. In another setup the device wafer was attached to a silicon handle wafer with
This method would be suitable for cases where metal exposure to plasma is prohibited.
Figure 4.34 Schematic of experimental setup, (a) metal layer connected to the plasma at edges, (b)
162
Suspension: Fabrication
To investigate whether a metal layer not connected to either the device or handle silicon
wafer is sufficient to eliminate notching, a metal coated oxidized silicon handle wafer is
When the metal layer was extending all the way to the wafer edge and exposed to some
plasma at the edges, we did not see any significant notching but there was some
we patterned the metal layer on the handle in a circle smaller that the wafer so it is
completely inscribed inside the handle wafer. In this case we saw notching damage of
the structures (Figure 4.35). This leads us to believe that connectivity of metal layer to
Figure 4.35 (a) Damage due to partially floating metal layer, (b) Notching damage due to
163
Suspension: Fabrication
Figure 4.36 Schematic of test setup with electroplated metal on device wafer backside.
To eliminate the need for a handle wafer, we electroplated nickel on a Cr/Au seed layer
to 5µm thickness (Figure 4.36). This was sufficient to act as a mechanically strong layer
to constraint any of the structures released on completion of etch. The method also
adhesive. The process is cleaner and more economical as it avoids the wastage of
Additionally we can dry release the structures by peeling off the metal layer as shown in
Figure 4.37.
Figure 4.37 Optical image of dry release process by peeling the metal layer from the wafer
backside.
164
Suspension: Fabrication
Various backside metals like Cr [4.50], Ti/Au [4.51], Cr/Cu, Cr/Ni, Cr/Au, Al have
been tested, with identical results. Aluminium has the advantage of being easy to
remove using any alkaline solution such as photoresist developer. Unless used as a
mechanical layer to constraint island geometries falling out upon completion of etch, it
is not critical for the metal layer to adhere well to the backside of the device wafer.
Pulsed plasma mode was investigated to quantify the effect of low-frequency (LF)
Figure 4.38 SEM image of trench profile etched for 30 minutes using (a) LF plasma mode showing
thinning down of the structures towards the bottom as compared to, (b) HF plasma mode which
maintains profile anisotropy better and (c) scallop size for LF mode is 260 nm × 818 nm compared
165
Suspension: Fabrication
We used the process parameters optimized for through-wafer etching in HF mode and
characteristics. Even though the etch rate was identical for both LF and HF etching at 3
µm/min, the profile shape of the etched trench at LF is markedly diverging compared to
HF plasma (Figure 4.38a,b). This is due to an increase in isotropic etching in each etch
cycle as seen by the increase in the size of the scalloping during LF compared to HF
(Figure 4.38c,d). So LF mode etching is not a quick solution to reducing notching and
Figure 4.39 Schematic of experimental setup to investigate effect of good thermal contact between
the device and handle wafer, (a) Wafer bonded on the periphery such that when the structures are
etched through-wafer, the structures are thermally isolated from the handle wafer, and (b) the
device wafer is completely bonded to the handle wafer allowing the isolated structures thermal
To achieve repeatable etch conditions a stable control of the wafer temperature is most
important. Photoresist and wax have low thermal conductance, so other temporary
bonding agents such as cool-grease [4.53] with high thermal conductivity were
investigated for attaching the handle wafer. In another experiment, the backside metal
coating itself was electroplated. The plated metal was chosen to be a few micrometres
thick in order to create a mechanically stable layer, reducing the need for a separate
166
Suspension: Fabrication
in Table 4.2.
In cases where the complete wafer is bonded to the handle wafer using a thin layer of
photoresist, the thermal sinking seems to be sufficient to achieve repeatable etching. For
cases where the device wafer was bonded to the handle by a thin photoresist periphery-
ring, we saw photoresist burn (Figure 4.40) which is indicative of bad thermal path
between the device and chuck. Improper thermal sinking of the wafer causes the
temperature of the wafer to rise to almost 150 C within 5 minutes causing the resist to
be less resilient to the plasma leading to fast degradation appearing as a burnt (grey)
layer. Once the resist is burnt, the plasma attacks the silicon.
Figure 4.40 (a) Optical image of burnt photoresist due to thermal sinking problem – the reflective
surface shows normal photoresist, (b) photoresist is burnt in this case on the proof-mass which has
bad thermal connectivity through the long springs – the grey non-reflective surface is burnt resist.
167
Suspension: Fabrication
The effect of notching in through-wafer silicon DRIE has been discussed. Charging of
an insulator layer is believed to be the root cause for a lateral undercut observed at the
silicon-insulator interface. Coating the backside of the silicon wafer with a thin metal
layer completely eliminates this notching effect. In the presented case an evaporated
aluminium layer was successfully used. Ultimately the choice of metal to prevent
Compared to other previously presented techniques used to reduce notching, which are
mainly based on tweaking the actual DRIE etch parameters, the method presented here
allows the uncompromised optimization of the DRIE process avoiding the danger of
introducing notching. With the backside metal layer the notching is completely
Table 4.3 presents the trends in silicon etch rate, sidewall smoothness, trench profile
control, across wafer uniformity, grassing and etch lag for the various parameters of the
DRIE process. The effect of increase or decrease in the level of the process parameter is
the effects of some of the parameters would be very different. A case in point is the
SF6:C4F8 flow rates, for etchant dominant conditions, the sidewall smoothness
deteriorates if the etchant concentrations are increased by increasing the gas flow ratio,
168
Suspension: Fabrication
lessen the chances of a passivation deposition and breakdown on the sidewall thus
Etch:Passivate ↑ ↑↑ ↓ ↑ ↓ ↑ ↑
SF6:C4F8 flow ↑ ↑↑ ↓ ↑ ↑↑ ↑ ↓↓
Cycle time ↑ ↑↑ ↓ ↑ ↓ ↑ ↑
APC ↓ ↑ ↑↑ ↑↑ ↑ ↑ ↓
Coil power ↑ ↑ ↑ ↑ ↑ ↑ ↓
Platen power ↑ ↑ ↑ ↑↑ ↑ ↑↑ ↓
The arrow ↑ shows the increase and ↓ shows decrease. Double arrow ↑↑ show the
dominant process parameter for that specific etch characteristics. Profile shape refers to
the profile anisotropy and an increase in it means an increase in the verticality of the
sidewall.
For a through-wafer etch using a halo mask, etch lag is not a concern as the trench width
is constant across the features. Similarly, etch uniformity across the wafer is not a
critical effect as the trenches can be over-etched at the edges of the wafers, the side-
effect of the over-etch is increased notching of structures at the wafer edge, but this has
been completely eliminated through the application of a metal layer at the backside of
the wafer. Due to the availability of thicker photoresist it is not critical to tune the
169
Suspension: Fabrication
This effectively reduces the process engineering to finding the right combination of
process parameters which affect the silicon etch rate, sidewall quality, and profile shape
in the desired way for a given silicon exposure area. It is an iterative process and
generally there is sufficient process window to change one parameter to affect a certain
etch characteristic without worsening any other etch characteristics. For etchant
dominant conditions, both sidewall quality and verticality of the profile shape are
strongly dependent on the chamber pressure. Reducing the chamber pressure improves
both the sidewall quality and makes the profile shape vertical; this is due to the
After considering the various issues with DRIE and multiple iterations of parameters to
achieve the desired etch characteristics, we have developed a process recipe for etching
170
Suspension: Fabrication
A process recipe (Table 4.4) has been developed to achieve through-wafer etching
without parameter ramping and with an overall silicon etch rate of 2 μm/minute. The
process flow is as shown in Figure 4.28. The complete etch is divided into two parts, the
first run etches the wafer without handle wafer then the device wafer is mounted on a
handle wafer. The second run etches the wafer through the full thickness.
Parameter Value
171
Suspension: Fabrication
The mask was designed using Nemetschek Vectorworks [4.54] and laser printed by
Delta Mask [4.55] with a resolution of 1.5 μm linewidth and 0.2 μm address grid. The
mask design (Figure 4.41) included suspension designs for horizontal (Figure 4.42) and
Figure 4.41 Full wafer mask design included 20×20 mm dies as well as 10×10 mm and 5×5 mm dies
The {100} single crystal silicon wafer was aligned at 45 to the mask plate to enable
(100) plane along the suspension compliant direction. (100) plane of silicon crystal has
the lowest Young’s modulus [4.56] which reduces the stiffness of the spring in the
172
Suspension: Fabrication
Figure 4.42 Mask layout of the horizontal lateral suspension, the inset shows the spring connector
and the linkages with the filleting along with the packing pieces.
Figure 4.43 Mask layout of the Galperin configuration lateral suspension, the insets show the
173
Suspension: Fabrication
A n-type {100} silicon wafer is used as substrate wafer, 200 nm thick oxide was grown
in a dry thermal furnace. The oxide on one side of the wafer is reactive ion etched using
CHF3 plasma, this side is henceforth referred as the backside of the wafer. A 200 nm
AZ400K developer. The underlying 200 nm oxide on the frontside of the wafer is then
etched using CHF3 plasma in RIE. The wafer is loaded into the STS ICP DRIE etcher to
etch for 3 hours using the process parameters mentioned in Table 4.4. The process flow
is shown in Figure 4.28. The complete recipes for the through-wafer etch is presented in
Appendix C. After the first etch the wafer was unloaded and bonded onto a silicon
handle wafer using a thin photoresist. The wafer was then etched for another 1 hour 20
minutes in the STS ICP DRIE etcher. Upon visual confirmation of the completion of the
etch – by noticing the bright Al layer at the bottom of the trench, the wafers were
unloaded and the handle wafer was released using acetone. The backside Al coating was
stripped in MF219 developer. The dies were separated by breaking the tabs.
Figure 4.44 shows a fully functional lateral suspension, various configurations were
designed and fabricated. The suspension shown in Figure 4.44 has a set of six springs
and three intermediate frames. Tabs were used to keep the devices together while
etching the wafer; the remnants of the tab are visible alongside the die. A silicon lateral
suspension with reduced mass frame and springs with cross-bars is shown in Figure
4.46, the close-up shows the sidewall of the spring end connector. All etch gaps were 40
174
Suspension: Fabrication
μm wide including the rectangular cavities etched in the intermediate frames to reduce
mass.
Figure 4.44 Silicon lateral suspension with six sets of springs and three intermediate frames
fabricated by DRIE.
Figure 4.45 Close-up of silicon lateral suspension with reduced mass frame and spring with cross-
bar.
175
Suspension: Fabrication
Figure 4.46 shows a SEM micrograph of the spring beam sidewall. The sidewall
smoothness varies from 8 nm rms roughness at the top edge to nearly 1 μm rms
roughness at the bottom edge. The rough portion of the sidewall starts at approximately
a quarter way through the etch which is around the time when the lateral etch starts
The profile SEM shows bowing of the spring beam (Figure 4.47a), intermediate frame
(Figure 4.47b), and the trench cross-section (Figure 4.47c). We will investigate the
effect of bowing on the dynamics of the suspension in next chapter. An advantage of the
shape of the bowed profile is the higher rigidity in the z direction while reducing the
rigidity in the x direction. The bowed shape also leads to a lower mass of the beam
176
Suspension: Fabrication
Figure 4.47 Etch profile shape, (a) spring beam cross-section showing an etch profile which is
slightly bowed towards the centre, and (b) the profile shape of a cross-section of an intermediate
177
Suspension: Fabrication
lateral suspension die. The gaps between the springs are 40 μm, thus eliminating the
4.7 Summary
DRIE based on Bosch time-multiplexed etching and passivation cycles was selected as
The etch trenches are designed with uniform width across the wafer to enable etch
uniformity, so the features on the wafer are enclosed within a “halo” geometry.
safeguard the substrate holder from exposure to plasma and also to restraint the packing
pieces of “halo” geometries from falling in the process chamber. Various bonding
techniques are tried and the suggested technique is to coat the backside of the device
wafer with metal and then bonded to a handle wafer using thin photoresist.
Plasma processes suffer from a range of deleterious artifacts like grassing, top edge
which were discussed in detail. Various process and design modifications are
incorporated to reduce or eliminate the deleterious side effects of DRIE. Grassing and
micro-masking are completely removed by increasing the platen power and decreasing
178
Suspension: Fabrication
the processing pressure and introducing O2 plasma de-scum prior to silicon etch.
increased roughness with increasing channel width is proposed and validated through
experimental results. Profile shape is shown to be an effect of sidewall charging and ion
angular distribution; decreasing the chamber pressure increased the anisotropy of the
etch leading to improved verticality of the trench sidewalls. Notching caused due to
A process recipe is developed to attain the desired sidewall quality and etch profile
shape. A variety of suspensions were fabricated with varying spring and frame designs.
The sidewalls vary in roughness from 8 nm RMS to 1 μm RMS. The profile shape of
the spring beams is nearly vertical with some bowing midway through the thickness.
All work covered in this chapter was carried out by the author alone with the exception
4.8 Conclusion
The ability to micromachine thick structures out of single crystal silicon using DRIE is
achieve the desired etch characteristics while reducing or eliminating the deleterious
179
Suspension: Fabrication
For through-wafer etching the important characteristics to optimize the process were the
sidewall quality and profile shape. By engineering a set of process conditions which
related the sidewall quality to the etch trench width, we are able to find a suitable etch
width for reduced sidewall roughness and silicon etch rates exceeding 2 μm/min.
Notching, which can significantly ill-affect the mechanical characteristics of the device,
is completely eliminated by a proposed scheme to coat the wafer backside with a thin
metal layer; this also improved the thermal uniformity across the wafer while etching,
silicon suspensions where the suspension is a vertical extrusion of the 2D design. The
developed process can be used to fabricate other high-aspect ratio through-wafer silicon
micro-mechanical structures.
The next chapter looks at the experimental determination of the suspension dynamics
180
Suspension: Fabrication
4.9 Bibliography
181
Suspension: Fabrication
182
Suspension: Fabrication
183
Suspension: Fabrication
184
Suspension: Fabrication
185
Suspension: Fabrication
[4.51] C.-H. Kim and Y.-K. Kim, "Prevention method of a notching caused by surface
charging in silicon reactive ion etching," Journal of Micromechanics and
Microengineering, vol. 15, pp. 358-361, 2005.
[4.52] S. Kumar and W. T. Pike, "Technique for eliminating notching in through-wafer
etching," presented at 16th MME Micromechanics Europe Workshop,
Gothenburg, Sweden, 2005.
[4.53] Coolgrease CGR7016, http://www.aitechnology.com/thermal.html.
[4.54] Vectorworks Designer, Nemetschek AG, Konrad-Zuse-Platz 1, 81829 Munich,
Germany http://www.nemetschek.net/.
[4.55] Delta Mask V.O.F, Nijmansbos 56, 7543 GJ Enschede, The Netherlands
http://www.deltamask.nl.
[4.56] N. Maluf, An Introduction to Microelectromechanical Systems Engineering,
Artech House, Inc., 2000.
186
five
5 Suspension: Characterzation
fabricated suspension. Any deviation of the suspension resonant frequency and rejection
ratio from the analytical and numerical estimates of chapter 3 are analysed and the
causes for them are discussed. The analytical and numerical models are then updated to
incorporate the effect of fabrication process variation [5.1]. A compensation design for
187
Suspension: Analysis
The normal modes of the suspension along the major axes are determined from slow-
scan imaging in a variable pressure scanning electron microscope [5.2]. For measuring
the resonant frequency the suspension is mounted on a SEM stub such that it is free to
The raster time is set to a few tens of seconds for measuring the resonant frequency. The
same system is used to measure the damping of the suspension with rastering times of
few minutes. The suspension is excited by applying a force in the appropriate direction
by either moving the positioning stage or tapping the SEM chamber from outside. It is
expected that for future measurements an internal piezoactuator will be mounted on the
The scan time is adjusted so that several time periods of oscillation correspond to the
frame acquisition time. Figure 5.1 is a typical slow-scan of the oscillation. The electron
beam (e-beam) in the SEM is rastered parallel to the direction of the movement of the
spring and proof mass. The scan allows us to take a snapshot of the suspension while it
is oscillating. The time taken for scanning the complete frame is known from the SEM
scan software. We count the number of oscillations (noscillations) in a fixed length of the
frame (frameoscillations). Since we know the time taken for the complete frame
(timeframe_full), we can calculate the time taken for the oscillations counted. Dividing the
number of oscillations counted by the time taken for the e-beam to raster over them
gives the number of oscillations per unit time. In Figure 5.1, total time for the full frame
acquisition (timeframe_full) is 5.1 s, the length of the whole frame (framefull) is 5.505 mm,
188
Suspension: Analysis
the length of frame (frameoscillations) containing 40 oscillations is 3.911 mm. The resonant
Figure 5.1 Slow-scan image of suspension under excitation, the edges of the spring and proof mass
appear as a sinusoid due to the rastering of the electron beam parallel to the direction of motion of
the proof mass. The sinusoidal edges show the fundamental mode of the suspension.
On closer inspection we can see the higher-order modes overlapped on-top of the
normal modes. Figure 5.2 shows the overlap of the x1 mode (the first in-plane spurious
mode along X-axis) on the normal x mode, the x1 mode is higher frequency than the x
mode and is visible in the SEM as a higher-frequency oscillation on the springs whereas
the proof-mass and the overall springs have a low-frequency oscillation shape from the
189
Suspension: Analysis
x mode. Similarly both the rotational modes and other translational modes can be
visualized by carrying out a slow scan along the various axes. The modes are excited by
orienting the stage in a way that an externally applied tap on the chamber walls gets
Figure 5.2 Overlap of the first spurious mode along the x-axis (ωx1) on top of the normal x-axis
mode (ωx). The x1 mode is visible as the higher frequency mode on the springs. The x mode is
visible as the low-frequency sinusoid of the proof mass and the springs.
To visualize the out-of-plane z mode, the suspension is tilted so that the beam is parallel
to the Y-axis of the suspension. The constant oscillation of the centre of the proof mass
(Figure 5.3 inset 2) shows the out-of-plane z mode (ωz) whereas the see-saw oscillation
of the proof mass at the corners (Figure 5.3 inset 1) gives the rotation mode about Y-axis
(ωβ).
190
Suspension: Analysis
Figure 5.3 SEM of the suspension with the electron beam scanning along the Y-axis of the
suspension, the insets show the ωz and ωβ modes. Inset 1 shows the see-saw motion of the proof
mass due to the rotational mode (ωβ) about Y-axis of the suspension. The out-of-plane mode (ωz) is
clearly visible at the center of the proof mass in inset 2 as a constant amplitude oscillation.
We calculate the normal mode frequencies from the SEM measurement in a similar way
as the x mode by calculating the time taken for a specific number of oscillations
The SEM based measurement of resonant frequencies is suitable for measuring only
low-frequency systems due to the loss of resolution for images with scan acquisition
191
Suspension: Analysis
time of less than 0.1 second as the SEM’s e-beam rastering is slow. This restricts the
technique to the measurement of suspensions with modes below few hundred hertz.
The quality factor (Q) of the suspension is calculated by carrying out a ringdown test in
a SEM. The Q is determined by the time taken for the amplitude to reduce by 1/e which
is the decay time constant, τ of the suspension. The expression for calculating the Q for
Q = πω0τ (5.2)
Figure 5.4 Ringdown of suspension in SEM slow-scan. Measuring the time taken for the amplitude
of oscillations to drop to 1/e of its initial value gives the decay time constant (τ) of the system.
192
Suspension: Analysis
Figure 5.4 shows the ringdown test for the suspension in UHV where the rastering time
For the ringdown in Figure 5.4 the initial amplitude of oscillation is 68 μm, so decay
time constant (τ) is given by the scan time taken for the amplitude to decay to 1/e of
initial amplitude which is 25 μm. The total time for the scan (timeframe_full) is 24 minutes,
the length of the frame (framefull) is 250.1 μm and the length of frame (frameoscillation)
between the initial oscillation and when it decays to 1/e of its initial value is 114.6 μm.
With such high Q values it was important to stabilize the suspension before imparting
the ringdown impulse. This is achieved by raising the chamber pressure to few hundred
pascals for a few minutes prior to the impulse to dampen out the background
oscillations. The chamber pressure is then reduced to the required level just before the
scan.
The resonant frequency and mode visualization is carried out in the SEM using the
method described in section 5.1. The quality factor (Q) of the suspensions is determined
by carrying out a ringdown test and calculating the decay time constant.
193
Suspension: Analysis
The resonant frequency of the fabricated suspension is measured and compared to the
analytical model developed in chapter 3. Table 6.1 presents the analytically and
determined values. The tolerance on the measured values is due to measurement error
when using a SEM image; some modes are harder to visualize leading to higher errors
Table 6.1 Vibration modes of silicon suspension with 24 μm wide springs. The suspension has two
frequencies compared to the simulated device. Even though the suspension is designed
to have a resonant frequency of 12.2 Hz, the first mode is only 7.9 Hz for the fabricated
suspension, a drop of 35%. The analytical model is more accurate than the FEA model
for calculating the translational resonant frequencies of the suspension. This can be due
to the fact that the FEA model is much simplified and the spring end connector rigidity
194
Suspension: Analysis
is much higher for the actual device which is accounted for in the analytical model. FEA
model is overall quite accurate and predicts all the modes within 10% of their measured
values and is much more accurate than the analytical model for rotational modes. This is
due to the FEA taking care of the bending and torsional components of the stiffness
The drop in the fundamental in-plane mode along X-axis (ωx) is due to the thinning of
the spring during fabrication since a thin spring will reduce the second area moment Ix
4.5.8 the etched features do not have vertical walls and this could cause a significant
change in the stiffness of the springs. An analysis of the spring shape thus becomes
The first spurious mode along X-axis (ωx1) is nearly 9% higher than the model estimate;
this is due to the decrease in mass of spring and frames due to reduction in spring and
The slight variation in the lateral in-plane mode (ωy) and out-of-plane mode (ωz) could
be due to variation in Ix and Iz for the spring and frame due to bowing of the profile
shape.
We are unable to visualize the rotational mode about X-axis (ωα) as the best place to
visualize it is by looking down the X-axis of the suspension and seeing the oscillation of
the edges of the proof mass. This is hard to achieve due to the distance and the almost
195
Suspension: Analysis
latched condition of the proof-mass under maximum deflection due to gravity causing
The rotational modes agree within 10% of the FEA values, which is a good agreement
given that there is higher measurement error in these modes compared to the
translational modes.
Even though the fundamental resonant mode is significantly different from the predicted
value from the analytical and numerical model, the rejection ratio of the higher order
modes to the fundamental is nearly similar to the model. This is due to the dependence
of the modes predominantly on Ix which defines ωx. Modes which are dependent on J
also are independent of the profile shape as J is proportional to Ix for thin springs
(section 5.3.3) and so the rejection ratio is independent of Ix. The slight difference in
some modes can be attributed to their dependence on Iz which is not proportional to Ix.
To analyze the performance of a fabricated device and correctly predict the mechanical
analytical and numerical models. We use an analytical model in section 5.3 for
designing lateral suspensions taking into account the non-ideal etch profile. Numerical
models become complex when the profile shape needs to be taken into consideration.
To avoid complexity associated with meshing bowed beam cross section of high aspect
ratio beams, we use the analytical models to calculate an effective thickness of the
spring and used it in the FEA model with rectangular spring profile so we can obtain
some information on the modes and their shapes. Silicon suspensions are then fabricated
196
Suspension: Analysis
The SEM allows controlling the pressure of the test chamber from 1 to 400 Pa, which is
suitable for seeing the effect of viscosity on the Q, though for measuring the Q at
shown in Figure 5.5 for a suspension. The Q at standard atmospheric pressure (1 atm =
1 bar = 105 Pa) is 1000, decreasing the pressure below 1 atm does not result in
significant change in Q until the pressure is lower than 600Pa. The Q-factor at UHV
100000
10000
Q
1000
Kn=1
100
0.001 0.1 10 1000 100000
Pressure (Pa)
Figure 5.5 Plot of measured quality factor as a function of pressure for 10 Hz silicon suspension.
The variation in quality factor at various pressures can be classified into three regimes
of gas and material damping of structures. In the first two regimes: high pressure and
rarefied regime, gas damping dominates. When the pressure is very low gas damping is
smaller than material damping of the silicon suspension. At high pressure the dominant
197
Suspension: Analysis
gas damping is due to the closing and opening of the springs with oscillation leading to
which does not vary with pressure in the high-pressure regime so the damping is quite
function of pressure and damping reduces with reducing pressure. The transition
between the two gaseous damping regimes is determined by the Knudsen number, Kn ,
which is given by the ratio of the mean free path of the gas molecules to the size of the
damping gap – in this case the separation between the springs. The Knudsen number is
given by
λ k BT (5.5)
Kn = =
L 2πσ 2 PL
where,
λ is the mean free path (m), L is the representative physical length scale (m), kB is the
Boltzmann’s constant (1.38 × 10-23 J/K), T is the temperature (K), σ is the particle
The rarefaction of flow based on Knudsen number is generally divided into four flow
regimes: Continuum regime (<0.001), slip flow (0.001 – 0.1), transition flow (0.1 – 10)
and free molecular (>10). For the silicon suspension with spring gaps of 410 μm, Kn is
unity at a pressure of 330 Pa, and the transition is expected for pressure range of 3.3 Pa
to 3.3 kPa. Below 3.3 Pa the material damping will dominate and above 3.3 kPa the Q
198
Suspension: Analysis
In the environmental SEM, the quality factor for low pressure between 1 Pa and 0.001
Pa is extrapolated from the rarefied regime as the chamber is not capable of varying
The suspension used in this case was etched silicon with a 200 nm oxide on the top
surface and a thin coating of polymer (passivation layer) on its sidewall from DRIE.
Cleaning the passivation layer increase the Q even further at UHV to 250,000. Even
though it is possible to attain such high Q for a bare silicon suspension, it is not realistic
to attain similar Q for the complete device as the presence of other metal and polymer
insulator layers dictate the Q of the device (as metal and polymer have much higher
material loss leading to a low Q) and preempt any aggressive chemical etch required to
In the SEM, the suspensions are mounted such that the regions above and below the die
is unconstrained so at high pressure (>400 Pa until 1 bar) the only damping that occurs
is due to the air flow between the springs. The analytical formula for calculating the
squeeze flow damping coefficient in the high pressure regime (non-varying viscosity) is
given by
14ηlt 3 (5.6)
α squeeze =
Nd 3
where l is the length of the spring, t is the thickness of the spring, N is the number of
spring sets, d is the gap between the parallel sides of a spring and η is the viscosity of
199
Suspension: Analysis
For a suspension with two set of springs and one intermediate frame N = 4, t = 525 μm,
η = 1.98×10-5 kg m-3 s-1, l = 8032 μm, d = 410 μm. The coefficient of squeeze flow
mω 0 (5.7)
Qsqueeze =
α squeeze
Using equation 5.6, we calculate the quality factor for the suspension due to squeeze
flow damping as, Qsqueeze ≈ 14000, which is much higher than the experimentally
determined Q of 1000 at high pressure. This shows other effects besides squeeze film
damping affect the Q-factor at high pressure. Viscosity is directly proportional to the
ambient pressure in the rarefied regime so with decreasing pressure viscosity also
Dynamics
The experimentally measured value of resonant frequency was much lower than the
model estimated value as we discussed in section 5.2.1. The DRIE etch causes a
As discussed in section 4.5.8 the profile shape of an etched channel can be either
tapering (towards or away from the centre line), vertical, bowing or a combination of
these effects. It is generally difficult to maintain a vertical profile shape particularly for
high aspect ratio structures as in this case. For these conditions it is important to
200
Suspension: Analysis
quantify and model the influence of the profile shape on the mechanical performance of
Previous works have analyzed the bending of folded cantilevers in the compliant
direction if the deviation from the rectangular profile is small [5.3, 4] or trapezoidal
[5.5], but the approximations made are not applicable to the more pronounced etch
We quantify the profile shape for through-wafer etched beams and present an analytical
model to evaluate the effect of profile shape on silicon lateral suspensions. A design
various etching and plasma conditions. The time multiplexing of etching and
passivation cycles in DRIE can be tuned to tailor the overall profile shape by varying
various process parameters such as gas flow rate, chamber pressure, RF coil power and
platen power. It becomes increasingly difficult to control the overall profile shape of the
etched structures for deep etches, as various effects relating to changes in reactivity and
charging come into play and change the etching characteristics with etch depth. With
increasing etch depth, the sidewalls accumulate charge distorting the electric field lines
causing the ions to deflect laterally towards the sidewall resulting in lateral etching [5.7,
8]. Also, with increasing etch depth, the reaction between ions and reactive neutral
radicals increases causing accelerated lateral etching by the isotropic radicals [5.9].
201
Suspension: Analysis
Even though it is possible to tailor the overall profile shape of a HAR structure, the
cross-section (CS) is almost never completely linear. This is not much of a concern for
larger mechanical structures but can significantly affect the mechanical properties of
slender structures where the undercut depth becomes a significant proportion of the
structure width.
resonant frequencies compared to the simulated device (Table 6.1). Detailed discussion
Spring cross-sections from various etch runs are analyzed to quantify the undercut. The
beam cross-section is quantified using a scanning electron microscope (SEM) and white
light interferometer. Figure 5.6 shows the SEM of the cross-section of the spring beam.
The mask width of the beam was 24 μm which is reduced to 22 μm in the actual
Figure 5.6 SEM image of beam cross-section. The mask design was for a spring width of 24 μm
which is reduced to 22 μm at top edge of the spring during the etch and the middle section of the
The bowing is further characterised using interferometer as shown in Figure 5.7. The
202
Suspension: Analysis
Figure 5.7 Undercut measurement using Zygo white light interferometer of spring cross-section.
A set of suspensions with 24 μm and 31 μm wide springs are etched to verify the
consistency of the etch profile. Figure 5.8 shows that the undercut is similar for both the
spring thickness, which means the undercut is independent of the feature size and is
The advantage of a feature size independent undercut is that one needs to measure the
undercut for any one feature on the wafer and then the effect of undercut on the
mechanical performance of all feature widths can be compensated using the data from
the initial structure. This assumes that the etch trenches are same across the wafer, it is
expected that for varying trench widths the undercut will also vary.
203
Suspension: Analysis
The non-rectangular beam cross-section (Figure 5.9) has an effect on the bending and
torsion constants Ix, Iz and J. Ix and Iz are area moment of inertia along X and Z axes
respectively. J is the torsional rigidity about the Y-axis. The profile can be quantified by
a depth-dependent beam width, w(t) which can be assumed symmetric about the yz-
plane of the beam. The bending constant for a beam with thickness t is then given by
1 t (5.8)
Ix =
12
∫ w (t ′)dt ′
3
Figure 5.9 Beam cross section of an anisotropically etched beam, (a) rectangular cross-section, (b)
bowed cross-section.
As the beam profile will normally be measured at a series of discrete points, for
example from a SEM cross section, it is more useful to consider the numerical
approximation for beam-width data (wn, tn; n = {1, N }), taken at a series of depths which
204
Suspension: Analysis
Figure 5.10 Discretization of the spring cross-section into trapezoidal elements to calculate the
Approximating the profile as a series of trapezoids, the X-axis bending constant is given
by
1 N −1 2 (5.9)
Ix ≈ ∑ wn (tn +1 − tn )(3wn +1 − wn )
24 n= 0
N −1
1 (5.10)
Iz ≈ ∑ wn (tn +1 − tn ){3(tn + tn +1 − t)2 + (tn +1 − tn )}
12 n= 0
The neutral axis of the beam is assumed to be at half the wafer thickness (t). The torsion
constant can be derived with the help of membrane analogy [5.10]: the torsional rigidity
membrane stretched over an opening the shape of the cross section. From this analogy
the solution for a long thin beam, J = w t /3 can be decomposed for a long thin beam of
3
varying width into contributions from thin slices of width dt΄, giving
1 t (5.11)
J=
3
∫ w (t ′)dt ′
3
205
Suspension: Analysis
coincidental, applying only to thin beams, but has the important effect that the ratio of
deflections due to torsion and bending in the compliant direction is unchanged by cross-
sectional shape of the beam as we observed earlier in section 5.2.1. Hence, although the
etch profile can have a large effect on the values of the normal modes of the suspension,
ratios between modes will tend to be little affected, as the modes of the suspension are
mostly dependent on Ix and J of the beams. For the analysis, an effective beam width,
weff is given by
1 t (5.12)
w 3
eff =
3t
∫w 3
(t ′)dt ′
0
geometry to give the correct values of Ix and J, although the value of Iz will not be valid.
So the ratio of the modified natural frequency to the frequency for a rectangular spring
(5.13)
ωx Ix
=
ω x0 I x0
where ωx0, Ix0 refer to the resonant frequency and second area moment for the nominal
profile.
The various resonant modes for the suspension can be derived by using expressions
from Table 3.5 with updated I and J values for the springs.
206
Suspension: Analysis
Figure 5.11 Etch profile of the spring - total spring thickness vs. etch depth.
numerically calculated the new Ix = 263821 μm4, Iz = 236×106 μm4 and J = 105×104
μm4. The effective width (weff) is calculated as 18.16 μm. Ix and J are just 43% of their
nominal values assuming a rectangular profile with the spring width of 24 μm. Iz is less
sensitive to the etch profile being only 80% of nominal, as the bending constant along
the z-axis is mostly determined by the thickness of the spring which is constant and is
Table 6.2 compares the model estimated values incorporating the profile shape to the
initial model estimates based on ideal rectangular profile and experimentally measured
values. The FEA model was developed using an effective spring width of 18 μm with a
rectangular profile which gives the same Ix and J as the analytical values. However, it is
207
Suspension: Analysis
not possible to simultaneously match Iz, which causes an error in the FEA calculation of
Table 6.2 Resonant modes for a 24 μm wide spring suspension. Comparison between analytical,
The agreement between analytical and FEA model incorporating profile shape with
observed values provide confidence in the lateral suspension models. Analyses of other
suspensions show excellent agreement of FEA and analytical results for a broad range
of geometrical designs.
To compensate for the reduction in frequency, the width of the spring has to be
increased so that the second area moments Ix, Iy and torsional rigidity, J for the spring
with the bowed profile is similar to the values required for achieving the correct
resonant frequencies. The compensation strategy for achieving the performance target is
as shown in Figure 5.12. The addition of a rectangular area to the existing spring profile
208
Suspension: Analysis
to achieve the effective thickness for the required resonant frequency is possible due to
the profile shape being independent of the feature width as demonstrated in Figure 5.8.
Figure 5.12 Compensation for spring etch profile. (a) Ideal, (b) Actual after fabrication, and (c)
increase the second moment of area Ix for the fabricated spring beam to specification.
The additional width of the spring (Δw) is calculated by solving the equation for the
moment of area Ix for spring with nominal thickness required for achieving the specified
2 (5.14)
⎛ Δw ⎞ Aspring
I nominal = I fill + I actual + 2⎜ ⎟
⎝ 2 ⎠ 2
Δw 3t Δw Aspring
2
I nominal = + + I actual
12 4
where Aspring is the cross-sectional area of the spring (Figure 5.12b), Inominal is second
area moment for spring with effective width required for achieving the specified
resonant frequency, Iactual is the second area moment for the spring which is fabricated
and is the basis for the next set of designs, Ifill is the second area moment for the added
rectangular area to increase the I and J components of the spring , and t is the thickness
of the spring (same as thickness of the wafer). To calculate Δw we need to solve a cubic
209
Suspension: Analysis
(5.15)
Δw 3t Δw Aspring
2
+ − (I nominal − I actual ) = 0
12 4
1
I nominal = × 24 3 × 500 = 576000μm 4
12
1 1
I actual = weff t = × 183 × 500 = 243000μm 4
12 12
spring and measured its resonant frequencies, which is presented in Table 6.3 along
with the original 24 μm wide spring suspension. For the analytical model we
numerically calculated the values of Ix, Iz and J from the SEM of the cross-section of the
210
Suspension: Analysis
Figure 5.13 Spring profile for 30 μm wide beam, (a) total spring thickness vs. etch depth, (b) SEM
As we can observe from Table 6.3, the suspension with 30 μm wide spring is close to
our initial specification of 10 Hz. The cross-axis sensitivity also improves as the wider
spring in addition to being compliant along X-axis also increases the stiffness along Z-
axis.
frequencies and comparing it against measured values for a one spring set suspension
with spring width of 24 μm. The analytical, FEA and measured values of the resonant
211
Suspension: Analysis
Table 6.3 Modal frequency for suspension with 24 μm spring which is increased to 30 μm to
compensate for the reduction in resonant frequency due to etch profile shape.
Table 6.4 Resonant modes for 1 spring set suspension with 24 μm wide spring.
The analytical, FEA and observed values are in good agreement for the one spring
suspension. The largest error is in the analytical result for the first on-axis spurious
mode (ωx1), where the observed value is double the model predicted value. This error is
212
Suspension: Analysis
not surprising as the assumption for the analytical calculation that the suspension has a
constant mass density is least valid for a single spring. For the two spring suspension
the agreement improves as the spatial periodicity of the suspension becomes closer to
The design compensation scheme developed in this section allows for the estimation of
resonant modes for a wide variety of lateral suspension designs prior to fabrication. The
good agreement between the analytical and numerical model provides confidence in
5.4 Summary
The resonant frequency and quality factor of the suspension is measured using a
pressure and about 40,000 at UHV. The resonant frequency measured for the
suspensions are much below the model estimates. The rejection ratios however are
The discrepancy in the measured and model estimates is attributed to the non-ideal
profile of the spring beams cross-section. The analytical and FEA models used an ideal
rectangular geometry for the spring beam cross-section whereas SEM and
difference in resonant frequency of the suspension is therefore due to the reduced area
213
Suspension: Analysis
The analytical model is updated to incorporate the effect of the spring beam profile by
using a numerically calculated area moment for the spring beam. The analytical model
is then used to determine an effective spring thickness for the FEA analysis. Both
analytical and FEA model are closer to the measured values after incorporating the etch
profile shape.
A compensation scheme is devised to calculate the spring beam width to attain the
specified resonant frequency. This enabled the design and fabrication of a new set of
suspensions which are able to reach the desired resonant frequency of 10 Hz. The cross-
The author’s major contributions covered in this chapter are the characterization of
microseismometer team for suspensions that were fabricated by the author, since such
measurements are part of complete data sets presented in this chapter, they have not
the author in collaboration with Tom Pike. The spring beam profile characterization and
5.5 Conclusion
The lateral suspension demonstrated a Q of 1000 in air and 40,000 at UHV and
removing the polymer coating from the spring sidewalls increased the Q to 250,000 –
higher than anything reported for bulk fabricated single crystal suspension. The
suspension satisfies the criteria set in chapter 2 to achieve Q ≥ 330 to satisfy the NEA
214
Suspension: Analysis
After incorporating the effect of the etch profile shape, the analytical and numerical
estimates are in excellent agreement with the measured resonant frequencies providing a
high level of confidence on the lateral suspension models and their applicability to other
designs.
As far as the dynamics of the suspension are concerned, a non-ideal etch profile, if
properly incorporated into the design, causes few problems. Bowing reduces Ix and J so
to increase it back to nominal value for the required resonant frequency, the spring is
increased in width which increases Iz thus increasing the cross-axis separation of modes.
Bowing also reduces the mass of the spring which improves the rejection ratio for the
For fabricating metal traces on the springs, the bowed profile of the spring provides a
215
Suspension: Analysis
5.6 Bibliography
216
Six
6 Microseismometer: Prototype
This chapter presents a brief description of the microseismometer prototype and its
various components. Fabrication processes required to build each of the modules are
also detailed along with some overall experimental results for the microseismometer.
The work in this chapter has been a team effort with contributions from the author on
each of the aspects covered. The overall design of the microseismometer is covered in
217
Microseismometer: Prototype
6.1 Seismometer
Seismometers measure the ground motion arising from a seismic event by determining
the motion of the ground (reference frame) relative to the suspended proof-mass
Differential capacitive position transducer is the most widely used precision sensing
changing system (Figure 6.1). Gap changing systems have a non-linear transfer function
whereas for systems where the overlap area of the plates is varied with position the
Figure 6.1 Two implementation of capacitive sensing, (a) distance (gap) changing: the distance
between the plates is varied and (b) overlap area changing: the overlap area between the plates is
open-loop the voltage output of the capacitive transducer is the final output whereas in
218
Microseismometer: Prototype
closed-loop the output of the feedback controller is the final output. Operating a system
with feedback provides the ability to shape the frequency response of the
dynamic range, and sensitivity of the system. Feedback control is further discussed in
section 6.3.
spectrum and so the output allows the greatest dynamic range of the seismic signal over
the band of interest. For an electromagnetic seismometer the proof-mass has a coil
attached to it which moves in a magnetic field. The current induced in the coil is
proportional to the velocity thus allowing a velocity output from the sensor.
magnetic field. The magnetic actuator provides much lower noise than the electrostatic
counterpart as it does not require high voltages. Figure 6.2 shows the various
Figure 6.2 Block diagram of the microseismometer. The four major functional units are the
mechanical suspension, capacitive transducer which converts the mechanical motion of the proof
mass to electrical signal, feedback controller which drives the electro-magnetic actuator to keep the
proof-mass at equilibrium.
219
Microseismometer: Prototype
Section 6.4 provides details of the electro-magnetic actuator designed for the
microseismometer. In section 6.5 we will look at the fabrication aspects of the various
introduction to the assembly of the complete device; section 6.7 looks at damping of the
packaged suspension and experimental test results are presented in section 6.8.
sense the lateral motion of the proof mass (see section 6.2). This transducer determines
the displacement between the fixed plate (pickup) and the parallel moving plate (drive)
attached to the suspended proof mass by measuring the change in capacitance between
the periodic electrode array on the drive and pickup plates. The sensitivity of differential
capacitive transducer increases as the nominal gap between the fixed and moving plate
is reduced. Increasing the sensitivity requires smaller electronic gain leading to smaller
contribution of electronic noise to the overall noise of the system. However, decreasing
the gap between the fixed and moving plates increases the coquette-flow damping of the
Figure 6.3 shows the cross-sectional view of the microseismometer assembly for the
capacitive transducer. The suspension die is bonded to the fixed (capping) die at the
edges while the central portions are separated by a micro-machined cavity. The
suspension and capping die contain an array of drive and output electrodes, the direction
of periodicity of the arrays aligned to the allowed motion of the proof mass. The drive
220
Microseismometer: Prototype
and output electrodes have the same periodicity (Figure 6.4a). The signal from the
Figure 6.3 Cross-sectional view of the microseismometer showing the Capacitive array transducer
with the drive electrodes on the proof-mass and a set of overlapping output electrodes on the top
capping die.
Displacement of the proof mass under acceleration changes the relative position of the
proof mass electrodes and the capping die electrodes. A small change in the relative
position of these electrodes change the overlap between the proof mass electrodes and
the fixed capping die electrodes and so can be transduced as a change in capacitance
between the fixed and moving plates through suitable drive and pick-up electronics.
The LCAT’s output electrodes are deposited on the silicon proof mass and the drive
electrodes are deposited on the fixed glass layer aligned directly above the silicon
substrate. The drive electrodes (Figure 6.4a) are driven with out-of-phase voltages, and
two sets of output electrodes provide a differential output signal. Figure 6.4b shows the
drive and output signals, the frequency of the drive signal is high enough to minimize
low-frequency (so called 1/f) noise common to analog electronics. There are 36
electrodes on each of the +/- DRIVE and +/- OUTPUT transducers. The gap between
221
Microseismometer: Prototype
Figure 6.4 (a) Configuration of the drive and output electrodes, and (b) alternate drive electrodes
are driven out-of-phase by a +DRIVE and –DRIVE square-wave pulse train. The output electrodes
To reduce the parasitic capacitive coupling between the electrodes on the proof mass
and the underlying silicon, a metal ground plane is used under the electrodes on the
proof mass.
“DC” signal due to the Earth’s gravity and a varying, or “AC” signal due to seismic
activity. Only the “AC” signal is of interest with regards to a seismometer. The “AC”
seismic signals are 160dB smaller than the “DC” gravitational signal. When a feedback
is used in such circumstances, the actuator is unable to exert a large enough force to
counteract the DC force and move the suspended proof mass to approximately a null
position without severely degrading the sensor self-noise as the noise generated by the
222
Microseismometer: Prototype
To enable the use of a low-force actuator thus maintaining the low-noise of the system,
limit control. The limit control electronics allows the electrode arrays to transition
between null points. When the suspension is under a steady force, such as a
gravitational field, the proof mass is allowed to be displaced by many array periods of
electrodes to a new null position under open-loop operation. The feedback is then
reapplied.
The feedback drives the proof mass to the nearest null point. The preset limit is only
slightly more than the signal range required to drive the proof-mass between the
adjacent null points. Hence the actuator needs to be only strong enough to drive the
proof mass over a little more than one period of the electrodes.
The feedback force is proportional to the current through the feedback coil. By passing
the feedback signal through a capacitive differentiating circuit, the feedback force is
proportional to the derivative of the sensor output voltage thus generating ground-
equivalents. A current can be driven through a coil with less noise than a high voltage
can be put on the electrostatic plates. Hence an electromagnetic actuator has less noise
223
Microseismometer: Prototype
The magnetic circuit is composed of two sets of external magnets of flattened horseshoe
geometry on each side of the assembled dies. A set of rectangular spiral coils metallized
on the proof mass are connected to the external electronics. As current flows through
the coil, the proof mass experiences a lateral force (Ffeedback) from both arms of the
where B is the magnetic flux density, I is the current in ampere and L is the length of the
The magnets are so designed that the two arms of the spiral coil are within the linear
portion of the magnetic field for most of their movement. The design of the magnet
yokes have a significant effect on the magnetic flux density and numerical simulation is
Figure 6.5 Perspective view of the magnetic circuit. The planar coil on the proof mass is enclosed by
magnetic field from the pair of magnets on either arm of the coil. The feedback force is generated
224
Microseismometer: Prototype
The dies are sealed at the edges using solder, so the traces are deposited under an
mass die, the capping dies and the magnetic units. These three mechanical sub-units are
The proof-mass wafer contains the proof-mass dies which comprises of the suspension
and the circuit fabricated on the suspension. Figure 6.6 shows the complete proof-mass
die with the mechanical suspension and the integrated circuitry for control and
measurement of the suspension. The close-ups show the electrical traces running on the
Figure 6.6 Proof-mass die showing the mechanical suspension with the circuit fabricated on top of
The proof-mass die consists of the mechanical resonating unit, a coil to drive the
225
Microseismometer: Prototype
displacement of the proof-mass and a metal layer around the edge of the die (edge seal
layer) to connect and hermetically seal the proof-mass die to the capping die.
The key requirements for fabricating the features on the proof-mass wafer are:
1. First Insulator layer: to electrically isolate the metal traces on top of the proof-
2. First Metal layer: to deposit a ground-plane under the drive electrodes on proof-
mass to eliminate capacitive coupling with the silicon substrate and for metal
links for coil and under edge seal (proof-mass capping wafer seal).
3. Second Insulator layer: to form a multiple loop coil for the electro-magnetic
actuator and forming a bridge over the metal link going under the edge seal.
4. Second Metal layer: to form the coil, traces along the springs and pads for solder
reflow.
226
Microseismometer: Prototype
Figure 6.7 Process schematic for the complete suspension fabrication with metal and insulator
layers.
Process flow for the full suspension fabrication is shown schematically in Figure 6.7.
The processing of the proof-mass wafer starts with a double sided polished single
crystal silicon wafer. 200nm of oxide is grown thermally in a dry furnace (step 1, Figure
6.7). A 1 μm thick layer of SPR1813 is spun on one side of the wafer and it is patterned
using the ohmic contact mask (step 2, Figure 6.7). The contact pads and the backside
oxide is etched in 49% buffered HF for 6 minutes. To form the ohmic contact, metal
a 25 nm NiCr(80/20 weight percent Ni/Cr) layer and then a 200 nm Au cap layer (step
3, Figure 6.7). The wafer is then taken out of the evaporator and the resist is stripped
before annealing it at 395 C for 1 minute. The contacts are then checked to confirm
227
Microseismometer: Prototype
whether they are ohmic, Figure 6.8 shows a typical I-V curve for ohmic contact pre-
anneal and post-anneal. The resistance of the ohmic contact is 200 ohms. An issue with
the ohmic contact pads is the formation of granular islands on the contact pad as seen in
Figure 6.9, even though the pads appears non-shiny the contact resistance was
Figure 6.8 (a) Contact resistance of pads prior to anneal, and (b) Post-anneal I-V curve for contact
Figure 6.9 Contact pad post-anneal showing a rough surface, the brighter islands is the Au layer
228
Microseismometer: Prototype
Next a 30nm Cr adhesion layer and 200nm Au Metal-1 layer is sputter deposited on the
(Shipley SPR 1813) is spun and patterned lithographically (step 5, Figure 6.7). The
metal layer is wet etched (step 6, Figure 6.7) to form the ground-plane for capacitive
shielding of the capacitive electrodes for pickup, interconnect for making multiple loop
coils and interconnects for connecting the traces to pads which are outside the edge seal
layer. The patterned metal-1 is shown in Figure 6.10. Figure 6.11 shows metal-1
interconnect for making the connection across the edge seal layer underneath the
insulator layer.
Figure 6.10 Mask layout of the metal-1 layer showing the ground-plane, interconnects for the coil
and traces. Metal 1 is 30nm Cr and 200nm Au and is patterned by wet chemical etching.
229
Microseismometer: Prototype
Figure 6.11 Metal-1 interconnect underneath the insulator layer for making the connection between
The next step in the process is the deposition of the insulator layer which in conjunction
with metal-1 makes bridging under metal-2 layers possible. The insulator layer is photo-
the wafer, cured temporarily and then lithographically patterned (step 7, Figure 6.7).
mechanically strong layer with sloping sidewalls suitable for running thin traces across
The photoneece insulator layer is used to cover the ground-plane for subsequent
interconnects to bridge the metal-2 layer across the edge seal layer as seen in Figure
6.11.
230
Microseismometer: Prototype
Figure 6.12 Photoneece, a photo-definable positive polyimide is spin coated on the wafer and
lithographically patterned. The developed and cured photoneece layer has sloping sidewalls
suitable for running thin metal traces across the edge of the insulator pads (courtesy: Werner Karl).
The topmost layer is the metal-2 layer which forms the coil, traces, edge seal layer,
pickup electrodes and the pads for interconnectivity between the proof-mass die and
capping die (Figure 6.13). To fabricate the metal-2 layer, a thin layer of Cr/Cu (30
nm/200 nm) is first sputtered on the full wafer. The Cr/Cu layer is then spin coated with
electroplating bath for plating Cu to a thickness of 5 μm (step 9, Figure 6.7) and then a
6.14a). The photoresist is then stripped and the initial seed layer is preferentially etched
231
Microseismometer: Prototype
Figure 6.13 Metal-2 mask layout showing the coil, traces, edge seal layer, pickup electrodes and the
interconnect pads for connectivity between the proof-mass die and capping die.
Figure 6.14 (a) Electroplated Cu in the photoresist mold, and (b) Electroplated Cu/Au after
removal of the photoresist mold and etching of the initial Cr/Cu seed layer.
Figure 6.15 shows the complete metal-1, insulator and metal-2 layers fabricated on the
proof-mass wafer prior to DRIE. After all the functional metal and insulator layers are
processed, the backside of the wafer is coated with 500 nm thick Al layer in a thermal
evaporator. The frontside of the wafer is then spun coated with a 14 μm thick layer of
photoresist (AZ9260) which is lithographically patterned and developed for etching the
232
Microseismometer: Prototype
suspension (step 11, Figure 6.7). The wafer is first etched in CHF3 plasma in RIE
(Reactive Ion Etching) system to etch the oxide layer (step 12, Figure 6.7). After oxide
etch the wafer is transferred to STS ICP Etcher for DRIE. The DRIE process is carried
out in two steps. The first step etches the wafer partially to a depth of approximately
350-400 μm (step 13, Figure 6.7). The wafer is then removed from the etching chamber
and mounted on a handle wafer using photoresist (step 14, Figure 6.7). The sandwich is
then etched further in the DRIE etcher for the full thickness of the wafer (step 15,
Figure 6.7). Due to the etch lag between the edge and center of the wafer, the complete
wafer is overetched to etch all the trenches on the wafer to full wafer thickness.
On completion of the etch, the wafer is released from the handle wafer by dissolving the
photoresist in a suitable solvent. The backside aluminum layer is wet etched and the
packing pieces fall off the wafer. The dies are separated by breaking the tabs which hold
them together.
Figure 6.15 Optical micrograph of the circuit on top of the proof-mass die prior to DRIE.
233
Microseismometer: Prototype
Figure 6.16a shows the proof-mass die with all the layers. The traces run on the top
Figure 6.16 (a) A complete proof-mass die, and (b) Closeup of the metal traces on a spring
Assembly of the final device is undertaken by reflowing solder to create a seal around
the device die and to form an interconnect between the electrodes on capping wafer and
The suspension die is enclosed between two capping dies (Figure 6.17) made of
borosilicate glass, the top one contains the pickup electrodes and the bottom die has a
cavity to reduce damping. The suspension die is bonded to the pickup die using solder
reflow, the bottom die is bonded to the suspension die using glass frit.
234
Microseismometer: Prototype
Figure 6.17 Exploded view of the capping and proof mass dies. The DT (Differential transducer) die
contains the drive LCAT electrodes. The cavity die is underneath the suspension die and is required
Process steps for the fabrication of the capping wafer is shown in Figure 6.18. The steps
are similar for both top and bottom capping wafer except the presence of metal
electrodes, and traces on the top capping wafer whereas the bottom capping wafer has
no metallization. Both the top and bottom capping wafers are then sand powder blasted
235
Microseismometer: Prototype
Figure 6.18 Schematic steps for metallization of the glass wafer and the subsequent sand powder
The top capping die contains the capacitive pickup electrode array for measuring the
displacement of the proof-mass. In addition it contains a seal ring on its edge. The seal
ring has solder deposited on it by either electroplating, jetting or manual solder ball
placement. The solder is then reflown to form the seal and electrical contact between the
To form the capacitive electrodes, traces and the seal ring on the top capping die, a layer
of 30nm Cr adhesion layer followed by a 200nm Ni and then 100nm Au are sputter
deposited on the glass wafer. Subsequently a 500nm thick photoresist layer (Shipley
SPR 1813) is spun coated on the wafer and lithographically patterned. The metal layers
are then wet etched to pattern the features. Figure 6.19a shows the mask layout for the
236
Microseismometer: Prototype
metallization on the top capping die and Figure 6.19b shows the glass die with full
metallization.
Figure 6.19 (a) Mask layout for the top capping wafer metal layer, and (b) Top capping die with the
metal layer.
Subsequent to the metallization the wafer is powder sand blasted to form the cavity. The
powder sand blasting is done by Crystal Mark [6.6]. Powder sand blasting is a fast low-
cost, anisotropic glass etching process with aspect ratios of 1:6. A nozzle shoots a
stream of fine sand onto the substrate, the etching is done by the physical removal of
glass by the fine sand particles. A laser-cut pre-patterned polymer decal is stuck onto
the wafers to protect the regions of glass wafer which are not etched. The decal also
protects the underlying metal layers from exposure to the sand blasting.
Powder sand blasting creates sloping sidewalls which are quite suitable for creating
cavities to reduce gas damping. 0 shows a powder sand blasted top capping wafer, the
cavities are etched to the specified depth (800 μm) on the metal side of the capping
wafer and then the wafer is turned over and the opposite side is sandblasted with an
appropriate decal to create dicing lines for separating the capping dies. Figure 6.20
237
Microseismometer: Prototype
shows the etched cavity on one side of the wafer and the dicing etch from the other side.
Powder sand blasting is used to create the cavities in the top capping wafer for reducing
Figure 6.20 On the top capping wafer, the side containing the circuit is powder sand blasted first to
etch the cavities, then the wafer is turned over and with a different decal dicing lines are etched
through the glass wafer. This allows for easy release of the dies (courtesy: Trevor Semple).
The bottom capping die is primarily to decrease damping by having a large cavity. The
bottom capping wafer is sand powder blasted to create the cavity to the desired depth on
one side and then sand powder blasted from the backside to create the dicing channels.
Glass frit is then screen printed on the wafer. The seal between the bottom capping
wafer and proof mass die is formed by glass frit bonding which is discussed in section
6.6.2.
The magnetic actuator comprised a yoke piece, on each yoke two Samarium Cobalt
(SmCo) rare earth magnets are mounted and on each magnet a pole piece is used to
focus the field. Figure 6.21 shows the schematic of the magnetic actuator components.
238
Microseismometer: Prototype
The coil on the proof mass lies within the magnetic flux lines and a current running in
Figure 6.21 Magnetic circuit for the microseismometer comprising of magnets, yoke and coil. The
yoke closes the circuit between the set of magnets on either side of the die sandwich. The yoke
under the magnet smoothes the magnetic flux lines and creates a larger linear region for the coil to
move within.
Figure 6.22 Location of the coil within the magnetic field. For maximum deflection force the coil
239
Microseismometer: Prototype
Ansys electromagnetic simulation is used to calculate the magnetic flux density at the
coil on the proof-mass. The magnetic flux density if better concentrated when using
pole pieces on the magnets. The bevelling of the pole piece also affects the strength and
Figure 6.23 Magnetic flux density at the coil for varying level of bevelling on the yoke. As expected
a sharper bevelling leads to a higher magnetic flux density but the linear range is reduced.
The magnetic components are glued to the device assembly. The soft iron yoke is
coated to prevent corrosion. The magnets used are SmCo rare earth magnets.
240
Microseismometer: Prototype
die, the proof mass and capping die is sandwiched between a set of magnets. The
Figure 6.24 Cross-sectional view of the microseismometer assembly showing the yoke, magnets, top
and bottom capping dies, proof mass die and electrode pads for signal readout and feedback.
The top capping die is attached to the proof-mass die using solder as discussed in
section 6.6.1. The bottom capping die is attached to the proof mass die using glass frit,
details of which are presented in section 6.6.2. The magnets are glued onto the yoke and
the pole pieces are held to the magnets by the magnetic field. The magnetic actuator
assembly is then glued on the outside surface of the top capping and bottom capping
die.
Solder reflow is used to form a contact between the pads on proof mass die and top
capping die as the external pads are located on only proof mass die. In addition a solder
ring all around the proof mass die and top capping wafer die is reflowed to form a
hermetic seal. Figure 6.25 shows the schematic of the contact and solder seal between
241
Microseismometer: Prototype
Figure 6.25 Schematic of the contact and seal between the top capping die and the proof mass die.
Solder reflow for bonding needs two wetting surfaces, one each on the two dies being
bonded together. Solder wetting is the ability of solder to stick to a specific metal
otherwise it will ball up to reduce its surface energy. Metals such as gold, copper and
nickel have good wetting property and are used for making the pads where solder is
initially deposited and also the pads which are ultimately bonded during solder reflow.
For the microseismometer, the top capping wafer contains a layer of Cr/Ni/Au on top of
which the solder is deposited. On the proof-mass wafer, the bonding pads are made of
Cr/Cu/Au.
Various methods have been investigated for solder reflow like solder printing, solder
paste screen printing, solder electroplating and solder ball placement either manually or
Figure 6.26 (a) Solder electroplating, (b) Solder jetting, and (c) Solder ball placement (courtesy:
Trevor Semple).
242
Microseismometer: Prototype
Solder ball placement using grooves etched in silicon by DRIE is one of the promising
methods where holes in a silicon wafer are etched which hold the solder balls. The balls
are transferred onto the main wafer (glass top capping wafer with reflow pads) by
application of flux to the reflow pads which are then reflown in-place. Figure 6.27
shows the schematic of the transfer of solder balls using a wafer with grooves for
Figure 6.27 Mechanism of solder ball transfer using grooves on wafers, the grooves are etched
using DRIE. The solder balls are transferred to the top capping die. The top capping and proof-
243
Microseismometer: Prototype
Glass frit is screen printed on the bottom capping wafer, the bottom capping die and
proof mass die are then aligned and heated to 350 C at which temperature the frit melts
Figure 6.28 Glass frit bonding to form seal between the bottom capping die and the proof mass die.
The glass frit bonding is done at a higher temperature than solder reflow temperature so
glass frit bonding step is performed prior to the solder reflow step.
To encapsulate the mechanical resonator in vacuum the solder and glass frit bonding
need to be hermetic. The vacuum encapsulation is performed during the solder reflow
which is subsequent to the glass frit bonding. The top capping die and the proof-mass
die are aligned and placed in the reflow chamber. A reducing gas is then introduced
inside the chamber to remove oxide from the surface of the solder. The chamber is then
evacuated and the temperature of the assembly is raised so that the solder reflows and
wets the pads on the proof-mass die. Figure 6.29 shows the schematic of the vacuum
encapsulation. The region marked by the red square represents the hermetically sealed
244
Microseismometer: Prototype
Figure 6.29 Schematic of the hermetically sealed cavity enclosing the mechanical resonator.
The thermal noise of the suspension increases with the damping of the suspension.
Damping in these suspensions is a sum of materials damping in the spring itself and gas
independent of pressure. The material damping will consist of damping in the silicon
itself, expected to be very small, and losses in the surface layers resulting from the
increased from 40,000 to 250,000 a 6 fold increase by removing the polymer coating
The overall quality factor of the suspension depends on the sum total of the loss
associated with the mechanical anchors, squeeze film, Couette flow, and Poiseuille flow
damping as well as bulk losses associated with material defects and thermo-elastic
damping (TED). The quality factor of the device due to gas damping is given by
1 1 1 1 (6.2)
= + +
Qgas Qsqueeze Qcouette Q poiseuille
Analytical model to calculate squeeze film, Couette flow, and Poiseuille flow damping
are discussed in [6.7-9]. Table 6.5 contains the dimensional parameters used to calculate
245
Microseismometer: Prototype
Table 6.5 Suspension dimensional parameters for damping calculation for microseismometer.
Parameters Value
Number of spring, N 4
Temperature, T 25 C
Using equation 6.3 Qsqueeze ≈ 14000 for a suspension with two set of springs.
246
Microseismometer: Prototype
d gap (6.4)
Qcouette = 2πfm proof −mass
ηAproofmass
where dgap is the distance between the proof mass and the capping wafers and is divided
into two components, one for the couette damping over the electrode array which is
much closer to the proof-mass and the other between the proof-mass and the top and
h2 (6.5)
Q poiseuille = km
12ηAproofmass
The Q for the assembled device is then calculated from equation 6.2, Qgas ≈ 290.
Calcualting the noise equivalent acceleration (NEA) for the device in air using equation
1.4, NEA = 3.65×10-9 ms-2/√Hz. The NEA of the device is lower than the targeted NEA
for the suspension. Couette flow damping is the dominant term in determining the
overall Q of the suspension, so reducing the damping by increasing the gap between the
proof-mass and the capping dies will lead to a higher overall Q enabling the suspension
The fully assembled device is then mounted on a single axis tilt holder. The output of
the coil on the proof-mass is measured using a network analyzer. Figure 6.30 shows a
247
Microseismometer: Prototype
Figure 6.30 A fully assembled microseismometer, for size reference a British penny is placed next to
The microseismometer evaluated here is a single-axis Galperin design. The output from
the coil is measured using a network analyzer. White noise is fed into one of the coils
on the suspension and the response from the other coil is measured by the network
spectrum analyzer. Figure 6.31 shows the measurement setup and schematic of the
54.74 as shown in a close-up of the conditioning unit in Figure 6.32; the conditioning
248
Microseismometer: Prototype
Figure 6.31 Experimental measurement setup for measuring the resonant frequency and Q-factor
Figure 6.32 Close-up of the electronic conditioning unit containing the microseismometer.
Measurement of the resonant frequency and quality factor is carried out for an
assembled microseismometer. Under the external magnetic field the coil on the
249
Microseismometer: Prototype
suspension.
The resonant frequency for the suspension is calculated to be 12.35 Hz from the
response plot (Figure 6.33) which shows the current output of one of the coils under an
Figure 6.33 Response plot for the suspension. The coil generates a current when driven under the
external magnetic field. The peak in the response curve corresponds to the normal resonant modes.
The Q-factor of the suspension is measured by doing a ringdown test on the suspension
(Figure 6.34). The current from the coil reduces as the vibration dies down. We
calculate the decay time for the amplitude and calculate the Q-factor using equation 2.5.
250
Microseismometer: Prototype
Figure 6.34 Ringdown test of the suspension. The current generated in the coil reduces as the
vibration decays.
The Q-factor for the microseismometer is calculated to be Q = 250 from a time decay
constant τ = 6.5 seconds. The NEA for the device is calculated using equation 1, NEA =
3.89×10-9 ms-2/√Hz.
and shock. As the silicon suspension is part of the microseismometer sensor assembly,
1. Transport of the microseismometer sensor assembly using ground and air transport
251
Microseismometer: Prototype
When the sensor assembly is part of a full fledged microseismometer (system), the
suspension can be actively damped to avoid shock and vibration, in which case only
vibration and shock test on the complete assembly is of concern. Testing of the
microseismometer) and is not considered here. Far more important for this work is the
shock and vibration response of a bare die or a packaged die (with capping dies and
magnets) to transportation vibrations and shock level. The bare die is not transported
out of the MEMS facility without the capping dies and the magnetic assembly so the
transportation of packaged die for final assembly is the point of highest vulnerability.
composite includes the worst vibration environments for truck, rail, and air
The composite transportation vibration envelop starts at 6 Hz (the lowest level the
testing equipment can attain) and is flat at 0.015 g2/Hz to 65 Hz at which point it
increases at approximately 6dB/Octave till it gets to 0.025 g2/Hz at 100 Hz. The
vibration envelop is flat at 0.025 g2/Hz till 500 Hz and then drops at -6dB/Octave to
0.0016 g2/Hz at 2000 Hz. The total RMS g level for the transportation composite is
around 5g.
252
Microseismometer: Prototype
0.1
0.01
Ariane acceptance
Transportation
Resonance
Ariane super
Ariane qualification
0.001
1 10 100 1000 10000
Frequency (Hz)
Figure 6.35 Minimum intensity exposure for transportation of packaged die using a combination of
The suspension die was clamped between the DT and cavity die for the vibration and
shock measurements. The packaged suspension was then placed on a Ling Dynamic
Shaker [6.12] (Figure 6.36) at AOPP [6.13], University of Oxford. The vibrator has a
maximum force rating of 8 kN (up to 16 kN with additional modules), the vibration and
shock profiles can be customized. To test the response of packaged sensor, it was placed
in the Galperin configuration on the test bench and the vibrator was programmed using
the Transportation composite vibration profile. The suspension failed around the spring
linkages and also on the frame –listing and picture of the failures are:
253
Microseismometer: Prototype
2. Damage at the corners of the intermediate frames and proof-mass due to the corners
The two classes of failures can be classified as complete fracture and surface spallation.
The packaged suspensions were able to sustain shock up to 75g laterally and 35g
vertically.
Figure 6.36 Ling Dynamic Shaker at AOPP, University of Oxford used for doing shock and
254
Microseismometer: Prototype
Figure 6.38 Frame and proof-mass damage due to collision at the corners.
To avoid failures at the linkages, the geometry was modified and the filleting was
improved. Figure 6.39 shows the new designs for the linkage. The designs were based
To avoid the frames and proof-mass colliding, a variety of damping structures were
designed as shown in Figure 6.40. The design philosophy behind cavity and comb
finger dampers is an increase in viscous damping as the protrusion enters the etched
cavity thus increasing the damping to frame motion. In the spring constraining tabs, the
255
Microseismometer: Prototype
frames are restricted in motion by being inside an envelop which stops the springs from
colliding with the frames. The frame spring damper places a spring at the end of the
Figure 6.40 Damper designs to avoid frames and proof-mass collision, (a) Cavity damper, (b) Comb
finger dampers, (c) Spring constraining tabs, and (d) Frame spring damper.
geometry and the various damping structures. The packaged suspension was then
exposed to transportation composite vibration levels without any failure. The vibration
level was then increased to Ariane acceptance levels (Figure 6.35) which is the
minimum vibration characteristics for a launch using ESA’s Ariane rockets. The
packaged suspensions survived the vibration levels without any failure. The vibration
level was then increased to Ariane qualification and subsequently to Ariane super which
is higher vibration levels than is required for a space launch mission. The linkages did
256
Microseismometer: Prototype
not undergo any failure. The damping structures had some failures as shown in Figure
6.41 which is mostly fractured pieces of silicon on the proof-mass and frames. The
Figure 6.41 Damage to the frames and proof-mass at Ariane super vibration levels for packaged
suspension.
Figure 6.42 Friction damage on proof-mass due to cavity die support fingers.
257
Microseismometer: Prototype
There was additionally some frictional damage to the proof-mass from the cavity die
(Figure 6.42). This will not affect the device performance and can be fixed by changing
A seismometer is most affected by tilt, temperature and pressure. Tilt is a major issue
for a horizontal seismometer but does not pose significant problems for the Galperin
configuration. The equilibrium between gravity and spring force is disturbed when the
temperature changes so the seismometer needs to be compensated for it. Pressure causes
variable buoyancy on the proof mass when it is not sealed and even when sealed it can
saw in chapter 5, the Q for the proof mass changes with pressure. For low self-noise it is
desirable to have a very high Q which is attained at low-pressure, this requires the
6.10 Summary
A complete seismometer sensor unit comprises of the motion sensing and feedback
differential capacitive array transducer (LCAT) is used to measure the motion of the
suspension. The feedback circuit keeps the proof-mass centred, thus increasing the
258
Microseismometer: Prototype
The full fabrication steps for the proof-mass die are presented. Besides the silicon lateral
suspension fabricated using DRIE, the suspension contains metallization and insulator
layers to form the LCAT pickup electrodes and the coil for the electromagnetic
Fabrication steps for the DT capping wafer and Cavity capping wafer are summarized.
Sand blasting is used to etch glass to create the cavities and dice the dies.
The magnetic unit contains a pair of magnets either side of the PM die. The magnets are
Samarium Cobalt (SmCo) rare earth magnets. The yoke pieces are soft iron and the pole
pieces are bevelled to focus the magnetic field. FEA is used to determine the bevelling
angle for the pole pieces to improve the linearity of the magnetic flux density over the
coil.
glass frit reflow. The DT die is bonded to the PM-cavity sandwich by solder reflow.
Solder balls or electroplated solder is used to form the pads used to bond the dies
together.
The dynamics of the microseismometer prototype are then determined by measuring the
current off one of the coils while the other coil is used to actuate the suspension using
white noise electrical signal. The suspension is measured to have a resonant frequency
259
Microseismometer: Prototype
The packaged dies also undergo vibration and shock testing. The linkages and frame
dampers are modified to build suspensions that can withstand rocket launch and
multiple modes of ground and air transportation. The suspension also withstood 75g of
The silicon suspension fabrication was done by the author. The metal layers and
insulator layers were processed by Werner Karl with contributions from the author,
Trevor Semple and Sanjay Vijendran. The solder and glass frit assembly was performed
by Trevor Semple with contributions from Tom Pike, Werner Karl and the author. The
characterization was done by the author with contributions from Tom Pike and Werner
Karl.
6.11 Conclusion
suspension has a Q of 250 in air with a NEA of 0.4 ng/√Hz which is higher than the
target NEA of 0.35 ng/√Hz. The couette flow damping dominates and would need to be
reduced to increase the Q for the suspension in air. This is achieved by increasing the
gap between the pickup electrodes but at the expense of reducing the sensitivity of the
completely eliminate viscous damping and allow the suspension to have lower noise
Shock and vibration tests demonstrate suspension that is suitable for terrestrial and
suspensions looks even more lucrative given the robustness of the suspensions.
260
Microseismometer: Prototype
6.12 Bibliography
261
Seven
7 Conclusions and Future Work
This thesis describes the design, analysis and fabrication of single crystal silicon lateral
suspensions. The suspensions are designed to have a low resonant frequency and a large
proof-mass. The suspension is made from single crystal silicon (SCS) which is a
popular semiconductor material with extremely high Young’s modulus, yield strength
and a very low material defect and a wide repertoire of standard fabrication tools. The
availability of high aspect ratio micromachining tools like STS DRIE etcher enables us
Seismometers are able to detect extremely low level vibration signals. The fundamental
limit to a sensors ability to resolve a vibration signal is set by the overall noise of the
sensor which comprises the electronic and mechanical noise of the sensor. In this thesis
our primary concern is with reducing the mechanical noise of the sensor. The
mechanical noise floor is determined by the Brownian noise of the suspension given as
noise equivalent acceleration (NEA) which is inversely proportional to the mass, time
period and quality factor of the suspension. This is the reason why the suspension
designed here has low resonant frequency (long period), relatively large mass
(compared to MEMS vibration sensors) and high quality factor through use of SCS.
262
Conclusions and Future Work
The lateral suspensions are fabricated by through-wafer DRIE of SCS wafers. The
grams and a quality factor that varies from 1000 at atmospheric pressure to 40,000 in
UHV and as high as 250,000 with proper post processing to remove any lossy materials
To understand the suspension dynamics and explore the design space to maximize
cross-axis rejection ratio while keeping the fundamental mode low, an analytical and
FEA model is used. The models were extended to include the effect of DRIE on profile
shape of the spring beams. The model estimates of the suspension modes were in
excellent agreement with the observed values thus providing an invaluable tool for
0.4 ng/√Hz at atmospheric pressure. The packaged suspensions are also robust to
vibration exposure for transportation and space launch and to shock levels of 75g
laterally and 35g vertically. Vacuum packaging the suspension should lower the noise
floor by an order of magnitude. Some of the other topics that need to be addressed in
263
Conclusions and Future Work
The fully integrated microseismometer is still under fabrication and the complete device
the Q for the suspension is extremely high in vacuum and can lead to substantial
In the current design, magnetic actuation was used for feedback control of the proof-
mass, and capacitative pickup was used to read the displacement of the proof mass.
Alternate system designs should be investigated that have fully electrostatic actuation
and pickup or with optical pickup to see if they have any benefits over the current
design.
Geophones are low-cost devices and so the assembly required for the current device is
not suitable for such a systems. An electronically actuated system is proposed which
can be further developed into a commercial system. Figure 7.1 shows the design layout
264
Conclusions and Future Work
Figure 7.1 (a) 2D model of a fully electrostatic inertial sensor, (b) a fabricated geophone.
The fabrication process uses the same processing used for fabricating the PM die but
The analytical and numerical models used for the suspension in the thesis show good
correlation with experimental results. They can be further expanded and applied to
The numerical model can be developed further to include fluid flow for damping
estimation. This will allow better understanding of the damping mechanism in the
lateral suspension. Further work can also be carried out on electrostatic simulation both
265
Conclusions and Future Work
Damping structures like fingers and springs (Figure 7.2) to engage and increase
damping when the suspension is undergoing its maximum throw were tried though there
Figure 7.2 Damping structures to increase damping when the suspension is at its maximum
deflection.
Another method to reduce the incidences of failure is by metal plating near the contact
points and self-caging of the resonant structure. These methods will work by increasing
energy dissipation using the metal structures and reducing the impact energies.
The shock tests reported in this thesis were all done at atmospheric pressure thus the
packaged suspension which will have extremely high Q of the order of 10,000 – 40,000
or even higher. Feedback based damping will most likely be required to passively damp
the suspension by shorting the coils on the suspension through a large resistor.
Resonating mass systems are popular with inertial sensor designers and the model, and
fabrication technique can be easily expanded into making other inertial sensors like tilt
266
Conclusions and Future Work
Another area where lateral suspensions can be expanded is actuators; the design is
suitable for making long range motion actuators which have extensive use in optical
developed as part of the work can be used for making a wide range of devices.
Figure 7.3 shows some other through-wafer devices fabricated using the process
developed for the microseismometer. This shows the universality of the processing
mechanism. Through-wafer etching is not only suitable for creating very thick single
crystal silicon inertial sensors but in many cases can replace devices created on SOI.
The cost advantage of a single wafer and no release can outweigh the advantages of
SOI.
267
Conclusions and Future Work
Figure 7.3 Some prototype devices fabricated using the MIMD (Metal-Insulator-Metal-DRIE)
process, (a) geophone sensor, 5mm × 5mm, (b) μSEM components, (c) a tuning-fork gyroscope, (e)
This work is done by the author in collaboration with Michael Larsson. We have
developed a new method for electroforming tall metal structures using high aspect ratio
(HAR) silicon molds. Mold fabrication is realized by through-wafer deep reactive ion
etching (DRIE). DRIE has been used for forming molds for making polymer [7.1, 2],
polysilicon [7.3, 4] and electroplating metal structures [7.5] The advantage of the
current technique is the ease of mold separation and an extremely high life of the mold.
268
Conclusions and Future Work
The method is further extended to create metal structures suspended from the substrate.
tall have been electroformed exhibiting extremely smooth sidewalls. The latter is
comparable with typical sidewall quality achievable by LIGA and significantly better
The method described here employs through-wafer silicon etching to realize HAR
molds. Tall metallic structures can be electroplated within through-wafer etched inserts,
passivated by a conformal polymer release layer. The layer smoothes mold sidewall
Figure 7.4 Fabrication sequence for the silicon mold, substrate preparation, electroplating, and
release.
269
Conclusions and Future Work
Mold fabrication and electroforming steps are shown in Figure 7.4. The silicon mold is
formed via through-wafer DRIE, with mask design and process parameters optimized
completely enclosed within other structures, bridges are designed, allowing the
fabrication of shapes such as rotors with hubs without the need for assembly (Step 3 in
Figure 7.4). The silicon mold is subsequently sputtered with a thin coating of Cr/Cu,
electrodeposition from a heated solution of Shipley Eagle ED2100 (Figure 7.5). The
with the underlying seed layer. Low temperature baking reduces reflow along feature
edges and promotes leak-free sealing with the substrate. After electroplating, the mold is
removed by dissolving the release layer in an organic solvent. Metal structures as tall as
230 μm are electroplated (Figure 7.6) and the sidewall rms surface roughness is
measured to be between 15-25 nm. Suspended structures are created in a similar manner
by electroplating on a substrate with the sacrificial layer defining a gap beneath the
structures. Figure 7.7 shows 50 μm thick suspended structures separated from the
The technique of electroplating metal structures using a reusable silicon mold enables
the presence of a release layer, which additionally smoothes mold sidewalls. The use of
bridges and the possibility of fabricating suspended microstructures reduce the need for
manual assembly. Even taller structures can be created by stacking multiple anodically
bonded molds. The process will be developed to create reliable tall micro-structures.
270
Conclusions and Future Work
Electrodeposited
release layer
Silicon Mold
Figure 7.5 SEM image of a silicon mold after DRIE (inset shows the mold with a 10 μm thick
mm
500
236 mm
Figure 7.6 SEM image of a molded metal test structure electroplated to 236 μm height. The sidewall
is vertical except where the structure is anchored to the substrate. Reflow of the release layer forms
271
Conclusions and Future Work
50 mm
15 mm
100 mm
Figure 7.7 Suspended metal structures molded using the fabrication steps depicted earlier. The
The technique can be developed into a reliable process for making tall metal structures.
Through-wafer etching can be used to make single mask devices where the mechanical
element is fabricated by through-wafer etching and the metal is deposited on the device
The process steps are shown in Figure 7.8. The mechanical structure is fabricated by
through-wafer etching the device wafer using a single mask. The corresponding shadow
mask for depositing the metal layers is also prepared by through-wafer etching another
wafer (can be thinned for better resolution). The shadow wafer is placed on top of the
device wafer and then placed in a sputter coater where the metal layers are deposited
through the holes in the shadow mask onto the device wafer. The sputtered layer due to
its isotropic nature deposits on all the surfaces exposed through the shadow mask. This
272
Conclusions and Future Work
is especially useful in devices like lateral comb drives where a metallic layer on the
Thermal isolation is many times very desirable in small-scale devices to reduce the
effect of temperature variation. Zhang and Najafi [7.6] demonstrated a silicon oxide
isolation trench formed by DRIE etching of thin trenches and oxidation to form a
method can also be expanded to form structures that are electrically isolated from the
substrate by channels which are anchored mechanically but are electrically insulating
273
Conclusions and Future Work
This scheme can be used for creating devices which has several different modules
which need to be electrically isolated. The ability to electrically isolate them which still
plane actuator composed of several moving arms all actuated by its own set of comb
drives.
274
Conclusions and Future Work
7.4 Bibliography
275
Appendix A
(The derivations were originally done by William T. Pike and are presented here after
being re-derived by the author.)
The suspension spring unit can be decomposed into a set of beams with rigid end
connectors as shown in figure a1.
The maximum deflection of the beam is given for fixed-fixed beam by Roark [A.1]
FL3 (a.1)
δx =
192 EI x
where E is the Young’s modulus of rigidity, and Ix is the area moment of inertia of the
beam about X-axis.
F 192 EI x (a.2)
k= =
δx L3
The two beams in each spring are in parallel, so the effective spring stiffness for the
spring unit is half the stiffness of one of the beams.
k (a.3)
k spring =
2
The two spring units on either side of the proof-mass is in series, doubling up the spring
stiffness for the complete suspension.
276
Appendix A
k 192 EI x (a.4)
kx = 2 × =k=
2 L3
L = 2l, length of a beam from the linkage to the end connector, so the spring stiffness is
24 EI x (a.5)
kx =
l3
For suspension with n spring sets, the spring constant along X-axis will be kx/n.
A force acting along Y-axis is taken to derive the deflection of a folded cantilever beam
as shown in figure a2.
Figure a2: Free body diagram of the spring elements for calculating spring constant
along Y-axis.
ML Fls (a.6)
θ= =
EI x 2 EI x
where E is the Young’s modulus of rigidity, and Ix is the area moment of inertia of the
beam about X-axis and M=Fs/2 is the moment acting on the beam.
The lateral deflection along the Y-axis for the folded cantilever beam is then calculated
as
Fls 2 (a.7)
δy = sθ =
2 EI x
The spring stiffness for the folded cantilever beam is then calculated as
277
Appendix A
F 2 EI x (a.8)
k= = 2
δy ls
For the mirrored pair of folded cantilever beam, the springs are in series so the spring
stiffness for a single spring unit is 2k. The suspension has two units of spring on either
side of the proof-mass, the spring units are in series leading to a total spring stiffness in
the y direction for the suspension to be
8EI x (a.9)
k y = 4k =
ls 2
For suspension with n spring sets, the spring stiffness will be ky/n.
The spring stiffness for a folded cantilever spring is made of two components: twisting
of the beam due to the moment generated in the elbow and bending of the beam under
the vertical force loading. The boundary conditions are shown in the free body diagram
in figure a3.
Figure a3: Free body diagram for vertical deflection of the folded cantilever beam.
Under vertical loading the beams will bend depending on the end boundary conditions,
table 3.1 in Judy’s Ph.D. thesis [A.2] gives the deflection as:
Fl 3 (a.10)
Δz =
3EI z
278
Appendix A
Fl 3 (a.11)
Δz =
12 EI z
The deflection is proportional to the rigidity of the end connector, given by a factor c
which is numerically determined. After incorporating c the end deflection then is given
as
cFl 3 (a.12)
δz bending =
12 EI z
The elbow undergoes an angular deflection under the torque caused by the opposing
directions of the vertical loading on the elbow given by
Tl Fs l Fls (a.13)
θ elbow = = =
GJ 2 GJ 2GJ
where G is the shear modulus of rigidity, and J is the area moment of inertia about Z-
axis.
Fls 2 (a.14)
δztwist = sθ elbow =
2GJ
The total vertical deflection of the beam under vertical loading is thus
⎛ cl 3 ⎞ (a.16)
F ls 2
k= = 1 / ⎜⎜ + ⎟⎟
δztotal ⎝ 12 EI x 2GJ ⎠
Spring constant for a spring unit is twice the spring constant of the folded cantilever
beams as they add in series. Spring on either side of the proof-mass add in series giving
a spring constant for the suspension as
279
Appendix A
⎛ cl 3 ⎞ (a.17)
ls 2
k z = 4k = 1 / ⎜⎜ + ⎟⎟
⎝ 48 EI x 8GJ ⎠
Figure a4: Free body diagram of spring beams under torsion about X-axis.
The angular deflection about X-axis consists of a contribution from the twisting
occurring about the X-axis of the linkages and a more complex contribution from a
combination of twisting and bending of the beams. Taking the twisting of the linkages
first, as there are two linkages for each spring unit of the suspensions on each side of the
suspended mass, each of length (s-w)/2, the torsional spring constant kα ,link will be
2GJ s (a.18)
kα ,link =
s−w
280
Appendix A
The torsional compliance of the linkages can be made very low by increasing the width
of the linkage. The compliance due to the distortion of the beams will then dominate.
The forces and torques producing this distortion are analysed in the free-body diagram
of figure a4. For a total external torque of τ, each beam will experience a torque T = τ/4
on its connecting face to the linkage to the proof mass. This torque will be opposed by a
combination of a torque U and moment Fl provided by the elbow. Considering the
balance of forces and torques on the elbow, the forces will produce an overall moment
Fs/2 which must be opposed by torques Fs/2. The resulting axial torques will cause a
twist of each beams. Hence the overall deflection of each beam will be due to a
combination of a torque U and force F producing a bend in the z direction and a twist
about the y axis due to the torque Fs/2.
The external torque can therefore be regarded as being distributed between the bending
torque U and force F so that the resulting deflection and twist of the beams cause a pure
rotation of the suspension unit about X-axis. The force F will be opposed by the other
half of the suspension unit but will slightly reduce the effect of the external torque by an
negligible amount Fws due to the forces acting on the linkage, it is neglected for the
derivation. Equating the torques acting on the beam, we get
T = Fl + U (a.19)
The angular deflection for the overall beam is 2αbeam as shown in fig a4(c), the two
beams comprising the folded cantilever pair undergo double the deflection due to forces
acting at the ends and an angular deflection of the elbow. αbeam can thus be calculated as:
Δ elbow (a.20)
Δ beam + = lα beam
2
where the deflection of the beam, Δ beam , is given by Roark [A.1] table 3, expressions 1a
and 3a as:
Fl 3 Ul 2 (a.21)
Δ beam = +
3EIz 2EIz
Fs l Fls 2 (a.22)
Δ elbow = φelbow s = s=
2 GJ 2GJ
The deflection angle of the beam ends is given by Roark table 3, expression 1 as:
Fl 2 Ul (a.23)
α beam = +
2 EI z EI z
Substituting equation (a.21) and (a.22) in equation (a.20) and solving it with (a.23)
gives
281
Appendix A
Fl 2 Ul Fl 2 Ul Fs 2 (a.24)
α beam = + = + +
2 EI z EI z 3EI z 2 EI z 4GJ
reducing to:
⎛ s2 ⎞ 2 EI z (a.25)
l2
U = F ⎜⎜ − ⎟⎟ ⋅
⎝ 4GJ 6 EI z ⎠ l
⎛ ⎞ (a.26)
⎜ ⎟
1 l ⎜ 3 ⎟
= 1−
kα ,beam 4 EI z ⎜ EI z s 2 ⎟
⎜ 4 + 3 ⎟
⎝ GJl 2 ⎠
The total compliance will be sum of the compliances due the torsion of the linkages
derived from (a.18) and (a.26):
1 1 1 (a.27)
= +
kα kα ,link kα ,beam
1 (a.28)
Ixx = mg 2
3
The torque about the Y-axis can be considered as two opposing forces acting along Z-
axis on either side of the proof-mass (figure a5). This means the spring constant for
rotation about Y-axis is proportional to the spring constant for motion along Z-axis (kz).
282
Appendix A
Figure a5: Rotational torque about Y-axis can be decomposed into an out-of-plane
force acting on the springs.
The torque acting about Y-axis is then given in terms of the force acting along Z-axis on
the springs as:
T y = Fz × h (a.30)
where h is the distance from the center of the proof-mass to the spring linkages.
The angular deflection can be calculated from the vertical deflection as:
Δz (a.31)
θβ =
h
where Δz is the vertical deflection of the springs under a vertical force Fz.
Spring constant for rotation about Y-axis can therefore be calculated using a.29 and a.30
as:
Ty Fz h 2 (a.32)
kβ = = = kzh2
θβ Δz
As we can see from the FEA derived value for mode ratio ωβ/ωx and ωz/ωx in figure a6,
the rejection ratio for the rotational mode about Y-axis follows the rejection ratio for the
out-of-plane mode. The analytical expression neglects the torsional effects which will
tend to increase the rejection ratio for the rotational mode about Y-axis. The analytical
expression thus gives the lower limit for the rejection ratio for rotation about the Y-axis
which is useful when trying to increase the rejection ratios.
kβ (a.33)
kzh2 3k z
ωβ = =
1 mh 2
= = 3ω z
I yy 3 m
283
Appendix A
The error between the analytical and FEA calculated rejection ratio for β and z mode are
exactly the same, showing a complete dependence of β mode on z mode.
The torque about Z-axis is comprised of two opposing lateral force acting along Y-axis
on either side of the proof-mass (fig. a7).
The torque acting about Z-axis is then given in terms of the force acting along Y-axis on
the springs as:
Tz = F y × h (a.34)
where h is the distance from the center of the proof-mass to the spring linkages.
The angular deflection can be calculated from the vertical deflection as:
284
Appendix A
Δy (a.35)
θγ =
h
where Δy is the total deflection of the springs under a lateral force Fy.
Spring constant for rotation about Z-axis can therefore be calculated using a.33 and a.34
as:
Fy h 2 (a.36)
Tz
kγ = = = k yh 2
θγ Δy
where Izz = ⅓m(h2 +g2) is the mass moment of inertia about Z-axis..
This simplified approach to calculating the rejection ratio neglects the effect of torsion
on the beams and gives the lower estimate of the rejection ratio consistently so is useful
for quickly exploring the design space.
A3. Bibliography
[A.1] R. J. Roark, Roark's formulas for stress and strain, 6 ed: McGraw-Hill, 1989.
[A.2] J. W. Judy, Batch Fabricated Ferromagnetic Microactuators with Silicon
Flexures, Ph.D. Thesis, Electrical Engineering, University of California,
Berkeley, 1996
285
Appendix B
A numerical method and a computer can be used to evaluate the mathematical model
and estimate the characteristics of the process. Two common numerical methods are the
finite difference method and variational method like Rayleigh-Ritz and Galerkin.
In finite difference method [B.1] the derivatives are replaced by difference quotients (or
Taylor series expressions), the resulting algebraic equation is solved by imposing the
boundary conditions for the values of the solution at mesh points.
The finite element method can be broken down into three basic schemes. First a
geometrically complex domain of the problem is decomposed into simpler subdomains,
the finite elements. Next for each finite element the approximation function is derived
using continuous functions which are a combination of linear algebraic polynomials.
Finally the algebraic relation between the undetermined coefficients are obtained by
satisfying the weighted-integral governing equations for each element. Thus, the finite
element method can be thought of as element-wise application of Rayleigh-Ritz or
weighted-residual methods. The approximation function is normally an algebraic
polynomial and the coefficients are the value of solution at some pre-selected points on
the boundary and within the elements, called nodes.
Various commercial and academic numerical analysis tools are available. For
mechanical static, dynamic and transient analysis FEM tools give detailed result on
stress gradients within the geometry.
286
Appendix B
Silicon is an anisotropic model, the analytical models used in chapter 3 consider silicon
an isotropic material with a modulus of rigidity of 129 GPa. To develop an analytical
model which takes into account anisotropic nature of silicon is non-trivial. It is much
easier to incorporate the anisotropic nature of silicon in numerical model.
Brantley [B.3] gives the numerical values for the elastic constants along the various
crystal planes of silicon as E11 = E22 = E33 = 165.7 × 109 N/m2, E12 = E13 = E23 = 63.9 ×
109 N/m2, and E44 = E55 = E66 = 79.56 × 109 N/m2. Hooke’s law is expressed for silicon
as:
The analytical calculations for the resonant modes are compared against numerical
simulations with isotropic and anisotropic silicon property. The isotropic model is much
more similar to the analytical model than the anisotropic model. The deviation of the
resonant modes between the isotropic analytical model and anisotropic numerical model
is pronounced in modes which are a function of the out-of-plane stiffness of the spring.
&& + [C]u& + [K ]u = 0
[M ]u b.1
[M] is the structure mass matrix, [K] is the structure stiffness matrix and u is the nodal
displacement vector and ü is the nodal acceleration vector.
ith natural circular frequency (radians per unit time) and t is time.
(− ω [M ] + [K ]){φ} = {0}
i
2
i b.3
287
Appendix B
ANSYS uses an eigensolver like Black Lanczos to solve the eigenvalue and
engenvector matrices for the model. Grimes et al [B.4] gives the theoretical basis of the
Block Lanczos solver.
The 3D element used for meshing the model is SOLID45 which has six degrees of
freedom at each of its eight nodes. The element is shown in figure b1.
Figure b1: SOLID45 element used for meshing the suspension model for modal and
stress analysis (source: Ansys manual).
288
Appendix B
289
Appendix B
290
Appendix B
! tb,ANEL,1,1 ! Stiffness Matrix for ! Solve the model for resonant modes
Silicon /SOLU
! tbdata,1, C11,C12,C12,0,0,0 antype, modal, new
! tbdata,7, C11,C12,0,0,0,C11 modopt, lanb, 10, 0, 1000,,off,,2
! tbdata,13, 0,0,0,C44,0,0 !select Block Lanczos solver
! tbdata,19, C44,0,C44 mxpand,10,,,0
!expand 10 modes
! Define Elements solve
ET,1,PLANE42 ! 2D Element !solve
ET,2,SOLID64 ! 3D Anisotropic
Solid ! Use postprocessor (/POST1) to get the
frequencies and mode shapes
! Area Mesh ! The layout is as seen in figure b2.
! Settings for map mesh
ESIZE, ,5
TYPE, 1 !assigns element
number to elements defined
MSHAPE, 0, 2D !use quads for
area
MSHKEY, 1 !mapped meshing
291
Appendix B
! Parameter nomenclature
! linkage - connector between spring & ml = 17760/2 ! proof-mass
frame - [l] length
! spring [s] fl = ml + (nFrame-1)*(fg + ft) -ll
! frame [f] mw = 11632 ! proof-mass
! proof-mass [m] width
292
Appendix B
293
Appendix B
294
Appendix B
295
Appendix B
296
Appendix B
! Map mesh
ESIZE, 100
TYPE, 1 !assigns element
number to elements defined
MSHAPE, 0, 2D !use quads for
area
MSHKEY, 1 !mapped meshing
AMESH, ALL
TYPE, 2
EXTOPT, ACLEAR, 1 !remove Figure b3: Suspension with three
area meshes spring sets and two intermediate frames.
VDRAG, ALL, , , , , , exline
III. Suspension with intermediate
! Specify Boundary conditions frames and rectangular rigid end
csys, 0 connector (Used for verifying the
alls accuracy of analytical models)
! Select the starting and ending springs ! Sunil Kumar, 3 June 2005
ends ! Multiple spring with intermediate
asel, s, loc, x, 0 frames suspension model to verify
asel, r, loc, y, 0 analytical formulae from Chapter 3.
asel, a, loc, y, ! Rectangular end connector – made
(nframe*(lt+st+sg+st+lt+ft)-ft)*2+mw rigid to correspond better to analytical
model
! Anchor the springs
da, all, ux, 0 /PREP7
da, all, uy, 0 /TITLE, MODAL ANALYSIS OF
da, all, uz, 0 LATERAL SUSPENSION
297
Appendix B
298
Appendix B
299
Appendix B
300
Appendix B
301
Appendix B
302
Appendix B
!alls
!nummrg, all
/prep7
303
Appendix B
k,1,-yoke_w/2,gap_h/2+mag_h+y2_h
/prep7 k,2,-
!define parameters yoke_w/2,gap_h/2+mag_h+yoke_h-
solder_h = 48e-6 !gap between PM bevel_y+y2_h
die and DT die k,3,-
frit_h = 1.5e-4 !gap between PM yoke_w/2+bevel_y,gap_h/2+mag_h+yo
die and Cavity die ke_h+y2_h
glass_h = 1.1e-3 !thickness of k,4, yoke_w/2-
glass wafer used for making DT and bevel_y,gap_h/2+mag_h+yoke_h+y2_h
Cavity die k,5,
yoke_w/2,gap_h/2+mag_h+yoke_h-
mass_w=10e-3 !proof mass bevel_y+y2_h
mass_h=525e-6 !wafer thickness k,6, yoke_w/2,gap_h/2+mag_h+y2_h
a,1,2,3,4,5,6
! Coil parameters xx=500e-6
coil_I=1e-3 !coil current yy=500e-6
coil_w=1e-3 !Width cut_y=1.0e-3
coil_h=10e-6 !Height
coil_disp_x=0 cyl4,0,gap_h/2+y2_h+mag_h,gap_w/2
gap_h=mass_h+2*glass_h+frit_h+solde asba,1,2
r_h !vertical gap between magnet arsym,y,all
pole pieces incl. yoke2
gap_w=2e-3 ! Draw Magnets
!gap between pole pieces blc4,-
yoke_w/2,gap_h/2+y2_h,mag_w,mag_h
yoke_h=4e-3 !yoke 1
yoke_w=16e-3 ! Draw Pole pieces
k,123,-yoke_w/2,gap_h/2+bevel2
mag_h=1.5e-3 !magnet k,124,-yoke_w/2+bevel1,gap_h/2
mag_w=(yoke_w-gap_w)/2 k,125,-gap_w/2-bevel1,gap_h/2
k,126,-gap_w/2,gap_h/2+bevel2
y2_h=0.50e-3 !yoke2 - soft iron k,127,-gap_w/2,gap_h/2+y2_h
yoke over magnet - thin one k,128,-yoke_w/2,gap_h/2+y2_h
y2_w1=mag_w
a,123,124,125,126,127,128
air_w=yoke_w*5
air_h=(gap_h+mag_h*2+yoke_h*2)*5 asel,s,loc,x,-yoke_w/2,-gap_w/2
arsym,y,all
! Vary bevel1 and bevel2 to vary asel,s,loc,x,-yoke_w/2,-gap_w/2
parameters arsym,x,all
bevel_y=yoke_h*.5 allsel
bevel1 = y2_h*7/5
!700microns ! Air area
bevel2 = y2_h*(1-.002) !1 blc4,-air_w/2,-air_h/2,air_w,air_h
micron
inf_w=air_w ! Draw coil
inf_h=air_h blc4,3.5e-
3+coil_disp_x,mass_h/2,coil_w,coil_h
! Draw Yokes
304
Appendix B
blc4,-3.5e-3+coil_disp_x,mass_h/2,- asel,r,loc,y,-air_h/2,air_h/2
coil_w,coil_h asel,inve
blc4,3.5e-3+coil_disp_x,- !asel,s,area,,1,4
mass_h/2,coil_w,-coil_h aatt,1,,3
blc4,-3.5e-3+coil_disp_x,-mass_h/2,- allsel
coil_w,-coil_h
! Select air only
allsel asel,s,loc,x,0,0,0
aovlap,all asel,r,loc,y,0,0,0
cm,air,area
!infinite element aatt,1,,1
k,,-inf_w,inf_h,0 allsel
k,,inf_w,inf_h,0
k,,inf_w,-inf_h,0 ! Select yokes
k,,-inf_w,-inf_h,0 asel,s,loc,y,gap_h/2+mag_h+y2_h,gap_
h/2+mag_h+yoke_h+y2_h
! Infinite areas asel,a,loc,y,-gap_h/2-mag_h-y2_h,-
a,kp(- gap_h/2-mag_h-yoke_h-y2_h
air_w/2,air_h/2,0),kp(air_w/2,air_h/2,0) cm,yokes,area
,kp(inf_w,inf_h,0),kp(-inf_w,inf_h,0) aatt,2,,1
a,kp(-air_w/2,air_h/2,0),kp(- allsel
inf_w,inf_h,0),kp(-inf_w,-inf_h,0),kp(-
air_w/2,-air_h/2,0) ! Select bevel
a,kp(-air_w/2,-air_h/2,0),kp(air_w/2,- asel,s,loc,y,gap_h/2,gap_h/2+y2_h
air_h/2,0),kp(inf_w,-inf_h,0),kp(- asel,a,loc,y,-gap_h/2,-gap_h/2-y2_h
inf_w,-inf_h,0) cm,y2,area
a,kp(air_w/2,air_h/2,0),kp(inf_w,inf_h, aatt,2,,1!change 2 to 1 to make it look
0),kp(inf_w,-inf_h,0),kp(air_w/2,- like air
air_h/2,0) allsel
305
Appendix B
306
Appendix B
Simulation layout is shown in figure b6 along with the magnetic vector lines.
Figure b6: Layout of the magnetic assembly and the magnetic flux density vector lines.
B5. Bibliography
307
Appendix C
308
Appendix C
2 Etch Backside dicing lines Sand blasting High velocity SiO2 200 micron
3 Etch Frontside cavity Sand blasting High velocity SiO2 800 micron
4 Release decals Release Acetone
5 Screen print glass frit Screen print Glass frit 60 micron finally
Assembly
309
Appendix C
17) Place wafer in System 80 for de-scum for 1 min with 60 sccm O2 flow and 100W
power to remove any leftover AZ9260 in developed regions.
ETCH 1
18) Etch wafer in STS ICP DRIE system using recipe DARK3S (Parameters same as )
for 4 hours.
19) Unload wafer – check channel depth and calculate time required for etching through
the wafer (nominal thickenss = 525μm).
ETCH 2 – WITH HANDLE WAFER
20) Prepare handle wafer – Dehydrate bake SSP (single side polished) wafer at 150C for
15’.
21) Spin SPR 1813: spread @ 500/500/10 with open lid, then spin @ 1000/1000/20 with
open lid.
22) Align the flats of device and handle wafer and press on the featureless portions to
bond the wafers together.
23) Place the sandwich on a cold hotplate with max temp set to 116C.
24) Bake on the hotplate at 110 C for 1’.
25) After cooldown, place wafer sandwich in STS ICP DRIE for etching.
26) Set etch time to 30’ and watch the wafer to see the dies turning bright when the Al
on backside starts getting exposed to plasma – etch for a further 15’ after the center
die appears bright.
RELEASE
27) Place wafer sandwich in a beaker with 300ml acetone, cover with a Al foil and heat
to 80C, leave for 30-60’ for the handle wafer to detach on its own.
28) Remove device wafer and place it in a clean acetone bath possibly at 80C to
minimize temperature gradient.
29) Clean the device wafer using acetone, IPA and DI.
30) Break the wafer into dies.
31) Indivisually clean each die into acetone, IPA and DI.
32) Fill a small 200ml beaker with 50-100 ml of MF319
33) Place each die into MF319 – it takes about 7-10’ for the Al on wafer backside to
etch off
34) Rinse the die in DI
35) Place each die on a wipe and blow dry the top portion
36) Lift gently (the die could be stuck to the wipe strongly due to the soaked DI so be
carefull) and turn it on its back – blow dry the other side.
37) Inspect in the microscope to check for any resist residue or dirt particles.
310
Page left blank intentionally
311