CMOS-MEMS High Gee Capacitive Accelerometers Amy Wung A report submitted in partial fulfillment of the requirements of the degree of: MASTER of SCIENCE in ELECTRICAL AND COMPUTER ENGINEERING at CARNEGIE MELLON UNIVERSITY Advisor: Prof. G. K. Fedder Reader: Tamal Mukherjee December 23, 2007
47
Embed
CMOS-MEMS High Gee Capacitive Accelerometers · CMOS-MEMS High Gee Capacitive Accelerometers ... the piezoresistor deposition requires temperatures as high as 965 degrees C, ... Chapter
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
CMOS-MEMS High Gee Capacitive Accelerometers
Amy Wung
A report submitted in partial fulfillment of the requirements of the degree of:
MASTER of SCIENCE
in
ELECTRICAL AND COMPUTER ENGINEERING
at
CARNEGIE MELLON UNIVERSITY
Advisor: Prof. G. K. FedderReader: Tamal Mukherjee
December 23, 2007
i
Dedicated to my parents, Edward and Bea Wung, and my brother Derek Wung for their encouragement and love.
ii
Abstract
This report describes the design and testing of a tri-axial high gee integrated CMOS-MEMS capac-
itive accelerometer. Military and industrial applications, such as vehicle crash and safety testing and
in-flight munitions testing, require inertial sensors capable of measuring accelerations up to 20,000
gee’s, where 1 gee = 9.8 m/s2. Commercial piezoresistive high gee MEMS are not integrated with the
sense electronics on a single chip. The CMOS-MEMS capacitive accelerometer design has the advan-
tage of smaller size, weight, and reduced parasitics due to single-chip integration.
The high gee accelerometer capacitive design is a new derivative of previous work at Carnegie
Mellon University on low gee capacitive accelerometers. The structure of this high gee accelerometer
departs from traditional low gee plate and comb drive capacitive sensing accelerometers, instead tak-
ing the form of an array of cantilever structures. The cantilever’s reduced mass and increased stiffness
increases the sensitivity range to the desired 20,000 gee’s. An array of cantilever structures is electri-
cally connected in parallel to obtain larger capacitive sensitivity.
The design methodology presented in this paper will demonstrate the theory and simulations used
to optimize the cantilever mechanical structures and capacitive sensing method. Preliminary electrical
and shock testing results in the 50-30,000 gee range will be presented to show functionality and linear-
A common surface-machined low gee accelerometer structure is a plate mass, with comb capacitive
transducers on two opposite sides and anchored by springs on the other two sides. Figure 2.1 shows a
simplified picture of such an accelerometer structure, with only a portion of the comb drive pictured.
Figure 2.1 A simplified comb drive accelerometer model, where the movable comb fingers and
anchored comb fingers form a capacitive divider, and an external acceleration changes the capacitance.
Only one set of comb fingers are shown for simplicity.
The capacitive comb fingers that are attached to the plate interlace with the anchored comb fingers,
making up a differential capacitive divider. An external acceleration causes the plate and comb fingers
2 Mechanical Design
y
xdeflection
y
y
xdeflection
y
8
Chapter 2 Mechanical Design
to move with respect to the anchored comb fingers. The change in gap results in a change in capaci-
tance, which can be measured to determine the direction and magnitude of the external acceleration.
This method of capacitive sensing is further discussed in Chapter 3.
Lateral high gee accelerometers
For the low gee accelerometer in Figure 2.1, the upper limit of the sensitivity range is determined
by the maximum free range of movement of the plate mass. Assuming the accelerometer is at rest, the
gap between comb fingers is go. The desired high gee accelerometer specification is to displace a dis-
tance less than go under a 20 kilo gee input acceleration.
The displacement sensitivity of the low gee accelerometer, to first order, is given by the following
equation:
. (2.1)
where Δx is the plate’s displacement due to an input acceleration a. The accelerometer displace-
ment in (2.1) is dependent on the structure’s mass m and the spring constant k. The maximum detect-
able acceleration is therefore:
. (2.2)
As an example, a CMOS-MEMS accelerometer with plate length of 260 µm and a plate width of
220 µm has a resonant frequency of 12.7 kHz. [1] The plate mass therefore moves approximately
1.5 nm per gee. The gap between comb fingers at rest is 1.5 µm, which means that the maximum
detectable acceleration for this sensor is 1,000 gees.
In order to modify this design to function under input accelerations of over 20 kilo-gees, the ratio of
stiffness to mass must be increased. The mass of the device is dominated by the large plate, and the
Δxa
------ mk---- 1
ωr2------= =
amax g0 ωr( )2×=
9
Chapter 2 Mechanical Design
mechanical stiffness is determined by the springs. The simple solution would be to increase the stiff-
ness of the springs, and decrease the size of the plate mass. However, decreasing the size of the plate
mass would decrease the number of comb fingers that can fit on the side of the plate, thus decreasing
the total sense capacitance. This has an adverse effect on the sensitivity, as described further in Chapter
3. Increasing the stiffness of the springs is possible by changing the geometry of the springs. However
the springs should not be designed such that the nonlinearities from extensional stress become signifi-
cant, as the plate movement would no longer be linear with acceleration. Thus there are limits to
changing the plate and spring dimensions without having adverse effects of the sensor performance.
Another idea is to take the goal of a stiff, less massive structure, to the practical limit - which would
be a single cantilever beam. The design concept chosen is to create a sensor from an array of anchored
cantilevers. The effective mass of the cantilever and its stiffness then determines the resonant fre-
quency and displacement sensitivity.
Figure 2.2 Top view finite element simulation of a 60 μm long, 2 μm wide free-end cantilever
bending under a 1000 gee input acceleration, next to two stator structures. The color scale indicates lat-
eral deflection.
stator electrode
stator electrode
Max: 2.0 nm
Min: 0 nm
stator electrode
stator electrode
Max: 2.0 nm
Min: 0 nm
10
Chapter 2 Mechanical Design
A simple free-end cantilever exhibits design problems, however. Under an external acceleration,
the free end will deflect, while the anchored end remains in place. Thus the sidewall of the rotor canti-
lever will no longer be parallel to the stator, as seen in Figure 2.2. It would not be correct to assume
parallel-plate operation between the rotor and stator electrodes in Figure 2.2. In parallel-plate opera-
tion, the capacitance for the capacitance between electrodes is given by:
(2.3)
where t and L are the dimensions of the plate, and g is the distance between the plates. It will be shown
in Chapter 3 that for small changes in g, the parallel-plate capacitance changes nearly linearly. The
capacitance for a tilted-plate capacitor is given by:
(2.4)
where g is the average distance between the plates, and θ is the angle formed by the plates. Even for
small values of θ, the capacitance does not change linearly with θ [9]. Therefore, without parallel plate
operation, the output will not be linearly proportional to the input acceleration.
The free-end cantilever design, then, must be modified to ensure that the sidewall of the rotor will
remain parallel to the stator, even under a large input acceleration. The first generation of modified
cantilever structures for a lateral high gee accelerometer was discussed briefly in a previous paper
from Carnegie Mellon University [1]. The lateral high gee accelerometer design is made up of an array
of 80 rotor fingers, where a single comb finger structure is pictured in Figure 1.2(a). The cantilever
structure’s suspension beam is narrow, while the side electrodes are wider and do not bend signifi-
cantly. The most significant deflection comes from bending in the center beam. The shape of the elec-
trodes are designed with the intent to balance the moments on both ends of the center beam, creating a
C εLtg
--------=
C εtLθ
-------- 2g Lθ+2g Lθ–-------------------⎝ ⎠
⎛ ⎞ln≈
11
Chapter 2 Mechanical Design
guided-end condition on the beam. With the center beam deflecting under guided-end conditions, the
side electrodes will move laterally, parallel to the stators. Thus the design preserves the parallel-plate
operation.
The FEMLAB simulations shown in Figure 2.3 illustrates the mechanical design principle. The dis-
placement plot of the bending beam in Figure 2.3(a) shows the guided-end bending caused by the bal-
ancing of moments. The displacement plot of the electrode in Figure 2.3(b) shows that the side
electrodes have approximately the same displacement along its entire length. The displacement is not
exactly uniform over the length of the electrode, but the difference in displacement is within 5% of the
total deflection. From these simulations, the displacement sensitivity of the accelerometer is approxi-
mately 6.4 pm/gee.
12
Chapter 2 Mechanical Design
Figure 2.3 Finite element results for lateral accelerometer cantilever structure bending under
1000 gees input acceleration. a) Plot of lateral displacement of the side electrode. b) Plot of lateral dis-
placement of the center suspension beam.
Vertical high gee accelerometer
The vertical high gee accelerometer’s design was initiated by Tsai [1]. It operates on much the same
principle as the lateral high gee accelerometer. The accelerometer is an array of 112 rotors, where a
single set of fingers are pictured in Figure 1.2(b). The side electrodes are much thicker in the z-axis
than the suspension beams, which ensures that the electrodes do not bend significantly. The mass of
6.0
5.0
4.0
3.0
2.0
1.0
0
y-di
spla
cem
ent [
nm]
0 10 20 30 40 50 60 x-position [μm]
6.0
5.0
4.0
3.0
2.0
1.0
0
y-di
spla
cem
ent [
nm]
0 10 20 30 40 50 60 x-position [μm]
6.55
6.5
6.45
6.4
6.35
6.3
6.25
y-di
spla
cem
ent [
nm]
0 10 20 30 40 50 60 x-position [μm]
6.55
6.5
6.45
6.4
6.35
6.3
6.25
y-di
spla
cem
ent [
nm]
0 10 20 30 40 50 60 x-position [μm]
(a)
(b)
13
Chapter 2 Mechanical Design
the side electrodes are designed to create balancing moments on the suspension beams, to satisfy the
guided-end condition, as demonstrated in the simulation results pictured in Figure 2.4(b). The elec-
trode displacement shown in Figure 2.4(c) is nearly equal along the length of the electrode, within 2%
of the total displacement. The displacement sensitivity is approximately 18.5 pm/gee from the finite
element simulation.
Figure 2.4 Finite element results for vertical accelerometer cantilever structure bending under
1000 gees input acceleration. a) Plot of out-of-plane displacement of the side electrode. b) Plot of out-
of-plane displacement of the center suspension beam.
y-di
spla
cem
ent [
nm]
0 10 20 30 40 50x-position [μm]
16.0
14.0
12.0
10.0
8.0
6.0
4.0
2.0
0
y-di
spla
cem
ent [
nm]
0 10 20 30 40 50x-position [μm]
16.0
14.0
12.0
10.0
8.0
6.0
4.0
2.0
0
y-di
spla
cem
ent [
nm]
18.6
0 10 20 30 40 50x-position [μm]
18.5
18.5
18.5
18.5
18.5
y-di
spla
cem
ent [
nm]
18.6
0 10 20 30 40 50x-position [μm]
18.5
18.5
18.5
18.5
18.5
(a)
(b)
14
Chapter 2 Mechanical Design
2.2 Lateral high g-force accelerometer mechanical design issuesAn undesirable effect of process variation on the high gee CMOS-MEMS accelerometers is the
presence of lateral curl in the lateral accelerometers. In the first generation design of the lateral acceler-
ometer, the suspension beam has a length of 60 µm. The CMOS-MEMS beam includes metal 1, 2, and
3 layers, where metal 1 is 1.2 µm wide, metal 2 is 1.0 µm wide, and metal 1 is 0.8 µm wide, as illus-
trated in Figure 2.5 (a). Significant lateral curl has been observed for these dimensions, as pictured in
Figure 2.5.
Figure 2.5 (a) Cross-section illustration of suspension beam, showing the stepped metal widths.
(b) SEM photograph of first generation lateral high g-force CMOS-MEMS accelerometer design,
showing significant lateral curl. (c) SEM photograph of the same structure, with greater magnification,
illustrating the unequal gaps between the rotor and stators.
Lateral curl can be the result of several causes, including electrostatic forces from dielectric charg-
ing, and fabrication misalignment of the metal layers. The observed curl is consistent in direction and
deflection for all released devices from the same processing run. This evidence suggests that the cause
of the lateral curl in the lateral high g-force accelerometers is due to variation during processing, possi-
bly due to metal layer misalignment and stress gradients.
M etal3
M etal2
M etal1
0.1 μm
0.1 μm
M etal3
M etal2
M etal1
M etal3
M etal2
M etal1
0.1 μm
0.1 μm
(a) (b) (c)
15
Chapter 2 Mechanical Design
Figure 2.6 (a) SEM photographs of test structures with the same dimension as those pictured in
Figure 2.5, processed at the same foundry but in a later processing run. (b) Greater magnification of
the vernier fingers attached to the pictured structures in (a). The vernier fingers are 1 μm wide.
Figure 2.7 a) Top view SEM photograph of a second generation of test structures. b) Greater mag-
nification of the vernier fingers attached to the pictured structures in (a). The vernier fingers are 1 µm
wide at the base, and taper to 0.5 µm wide at the tip.
There is notable variation between processing runs, which also indicates the problem is in metal
layer misalignment or stress gradients. Figure 2.6 and Figure 2.7 both show SEM photographs of sim-
ilar structures from two different processing runs. The devices in Figure 2.6 show approximately
0.8 μm deflection at the tip. In comparison, the devices in Figure 2.7 show less than 0.1 μm deflection
at the tip, although the structures in Figure 2.6 have equivalent dimensions.
(a) (b)
(a) (b)
16
Chapter 2 Mechanical Design
Metal layer misalignment, material stress gradients, and other processing variation in the CMOS-
MEMS process will always be present. The designer does have options however to mitigate the effects
of process variation. One option, which all the structures above already take advantage of, is to make
each metal layer slightly narrower than the metal layer below it. In the accelerometer designs, the
metal 1 layer is 0.2 µm wider than the metal 2 layer, which is in turn wider than 0.2 µm wider than the
metal 3 layer. This cross-section is pictured in Figure 2.5 (a). Assuming that the metal offset between
layers due to process variation is not greater than 0.1 µm, the edges of each metal layer are uncon-
strained allowing the relaxation of axial stress without significant lateral curl. The first generation
accelerometer design already makes use of this design trick, however there is still significant curl, even
when the cross section in Figure 2.8 shows that the misalignment between metal layers is not greater
than 0.1 µm.
The cross-section shown in Figure 2.8 was taken using a focused ion-beam to slice through the sus-
pension beam of a lateral high g-force accelerometer. Although the cross-section shows that the metal
layer misalignment is less than 0.1 µm, several other process non idealities can be observed from cross
section. There is a visible sidewall on the left side of the structure, of approximately 0.1 µm thickness.
The sidewall is either polymer redeposited during the release process, or is an artifact of the FIB cross
section. Due to charging from the nearby structure, it is unclear from the photo if the sidewall is sym-
metrical on both sides. Additionally, the bottom metal in the stack does not have a clear cut vertical
17
Chapter 2 Mechanical Design
sidewall. The top and bottom Ti/W metal barrier layers are wider than the aluminum metal layer. Fur-
thermore, there is a dark area to the left of the bottom metal layer, which could possibly be oxide.
Figure 2.8 Cross-section of a lateral high gee accelerometer. The shadow on the right side is the
unanchored section of the accelerometer that has been cut off, but did not fall into the silicon etch pit
below.
Because the exact cause of the lateral curl is still unknown, the simplest solution to prevent lateral
curl is to widen the suspension beam of the accelerometer. The design of the second-generation lateral
accelerometer suspension beam has a 2.0 µm-wide metal 1 layer, a 1.8 µm-wide metal 2 layer, and a
1.6 µm-wide metal 3 layer. The new design is extremely conservative, with the suspension beam being
almost twice as wide as its predecessor. Since this design change, the offset of the released high gee
lateral accelerometers has been reduced to less than 0.1 µm. The wider structures have been verified to
release using the standard post-CMOS MEMS processing.
18
Chapter 2 Mechanical Design
2.3 Vertical accelerometer mechanical design issuesIn the case of the vertical accelerometer, the 1.2 μm wide, 60 μm long, suspension beams are
expected to curl significantly out of plane of the chip. The curl is due to differing residual and thermal
expansion stress in the metal and oxide layers that compose the beam. Since those beams only contain
one metal layer and one oxide layer, the stress gradient is sufficient to produce a relatively large
amount of vertical curl. For this reason, there are two sets of bending beams in series to provide curl-
matching. Because the curl-matching beams are all the same dimensions, they will curl equally, and
the side electrodes will remain parallel to the substrate after releasing. This design implementation of
curl matching does come at the sacrifice of extra layout area. A perspective view of the released canti-
lever structure is shown in Figure 2.9.
Figure 2.9 Perspective view of high-gee vertical accelerometer cantilever structures.
These structures have been verified to release using the standard post-CMOS MEMS processing.
Figure 2.10(a) shows the measured topology of a released high-gee vertical accelerometer. The mea-
surement equipment has an x-y resolution of approximately 1 µm, which is why the topology image is
not clearly defined. The structures on the right side of the image are the most clearly in focus, which is
where displacement was measured along the curl-matching beams. The displacement of the curl-
matching beams is shown in Figure 2.10(b) and (c) to be approximately equal, so that the electrodes
remain in the same plane as the stators.
19
Chapter 2 Mechanical Design
Figure 2.10 (a) Measured vertical displacement of a released high-gee vertical accelerometer using
a Veeco NT 3300 white light interferometer. (b) Measured displacement of the curl-matching beams
located at cross-section A-A’. (c) Measured displacement of the curl-matching beams located at cross-
section B-B’.
(a)
(b)
(c)
20
Chapter 3 Capacitive Sensing
3.1 Principle of capacitive sensingCapacitive sensing is commonly used in the design of MEMS sensors as a low-power method of
converting mechanical movement to electronic signals. It is especially useful for inertial sensors
because the resulting voltage change will be, to first-order approximation, proportional to the external
acceleration. The high-gee accelerometers described in this paper use the same electrical connections
to form the capacitive bridge as the low-gee accelerometers designed by Tsai [1].
3.2 Lateral high-gee accelerometer capacitive sensingFor the lateral high-gee accelerometer, the design principle is to sense the change in capacitance
caused by an increase or decrease in the gap between comb fingers. Figure 3.1 shows the electric con-
nections in the comb fingers that create the capacitive bridge in the lateral accelerometer. In the cross
section in Figure 3.1(b), capacitance C1 decreases and capacitance C2 increases as the rotor moves to
the left.
For the purposes of calculating the accelerometer’s sensitivity, it is necessary to know the rest
capacitance of each node in the capacitive bridge, and the parasitic capacitance of the output node, Vo,
to ground. The magnitude of the rest capacitance is difficult to calculate by hand accurately, because of
the alternating layers of conducting metal and insulating silicon dioxide, as well as the different widths
between layers. Finite element simulation is used to determine capacitance between the electrodes.
3 Capacitive Sensing
21
Chapter 3 Capacitive Sensing
Figure 3.1 (a) Top-view of a single cantilever in the lateral accelerometer array, with cross-section
marked. (b) Cross-section of one rotor finger between two stators which form the comb drive capaci-
tive bridge used to sense acceleration along the y-axis. Vo represents the sense output voltage, and Vm
is the modulation voltage which powers the capacitive bridge.
The 2-dimensional electrostatic simulation results, pictured in Figure 3.2, give the effective capaci-
tance between a single stator and rotor sidewall to be 2.4 fF, taking into account the stepped metal
widths. In comparison, the parallel-plate approximation, using the minimum gap of 1.0 μm, estimates
the capacitance to be 2.6 fF. Using the maximum gap of 1.4 μm, the parallel-plate approximation esti-
mates the capacitance to be 1.9 fF. For one side of the differential output, there are 40 total cantilever
structures electrically connected in parallel in a single lateral high gee accelerometer. Each capacitor in
the capacitive bridge in Figure 3.3 represents a total capacitance of 40 cantilever structures with elec-
z
y
z
y
(b)
y
x
y
x
A
A’
A
A’(a)
deflection y
C2C1
+Vm +Vm
-Vm-Vm
Vo Vo VoA A’
22
Chapter 3 Capacitive Sensing
trodes connected in parallel. For the differential output, the two capacitors C1A and C1B are matched
and behave identically under an input acceleration. The two capacitors C2A and C2B are matched as
well. At rest, the capacitors each have capacitance Co = 97 fF.
Figure 3.2 Two-dimensional electrostatics simulation of a single rotor between two stator fingers.
The color scale indicates the electric potential.
23
Chapter 3 Capacitive Sensing
Figure 3.3 Schematic representation of capacitive bridge formed by two sets of 40 parallel cantile-
ver structures having parallel electrodes in the lateral high gee accelerometer.
The accelerometer’s sensitivity also depends on how that capacitance changes with the input accel-
eration. This design achieves first-order linear changes in capacitance in response to acceleration. By
approximating the metal stacks as a parallel-plate capacitor, its capacitance is given by (2.3).
When the electrode displaces by a distance , the capacitance change is,
(3.1)
assuming that . This is a reasonable assumption. For example, the FEA simulations discussed in
Chapter 2 give the displacement sensitivity of the rotor to be 6.4 pm/gee for the lateral accelerometer.
For a 20,000 gee input acceleration, the displacement will be 128 nm, which is much less than the
1.5 μm at rest gap. Substituting (2.1), the change in capacitance is, to first order, proportional to accel-
eration:
. (3.2)
V m
-V m
V o +
V o -
C 1 A
C 1 BC 2 A
C 2 B
4 0 c a n ti lev e rs tru c tu resh a v in gp a ra lle le le c tro d e s
4 0 c a n ti le v e r s tru c tu re sh a v in gp a ra lle le le c tro d es
V m
-V m
V o +
V o -
C 1 A
C 1 BC 2 A
C 2 B
4 0 c a n ti lev e rs tru c tu resh a v in gp a ra lle le le c tro d e s
4 0 c a n ti le v e r s tru c tu re sh a v in gp a ra lle le le c tro d es
Δx
ΔC εAg Δx–--------------- εA
g------– εA Δx⋅
g g Δx–( )----------------------- εA Δx⋅
g2-----------------≈ CoΔxg
------= = =
Δx g«
ΔC εAg2------ a
ωr2------⋅=
24
Chapter 3 Capacitive Sensing
Repeating the 2-d finite element simulation in Figure 3.2 while changing the gap between the metal
stack cross-sections gives the change in sidewall capacitance per length over input accelerations. Mul-
tiplying the capacitance per length by the length of the electrode and by the number of cells in the array
gives the data in Figure 3.4, where the capacitance of C1A and C2A is plotted over input accelerations
of gees. The correlation coefficient of both curves in Figure 3.4 are 0.998.
Figure 3.4 Capacitance of C1A and C2A in the lateral accelerometer capacitive bridge, with respect
to input acceleration.
The output voltage of the capacitive bridge will also change linearly. Due to charge conservation on
the high impedance output node, the voltages on the capacitors are constrained by the following rela-
tion:
. (3.3)
where Vo is the one-sided voltage output of the differential capacitive bridge, Vm is the signal powering
the capacitive bridge, Co is the capacitance of C1 and C2 at rest, and ΔC is the change in capacitance
Figure 4.7 Measured CMOS-MEMS high-gee accelerometer output with respect to maximum
deceleration over several drop tests for (a) lateral and (b) vertical accelerometers, with modulation fre-
quency of 1.7 MHz.
One possible cause is approximations made in the mechanical simulation, such as not including the
Ti/W metal barrier layers and not using the exact metal and oxide layer thicknesses, which were not
immediately available. The mechanical simulations were modified to include the average Ti/W, metal,
y = 3.0228x + 46.052R2 = 0.9377
0
100
200
300
400
500
600
700
0 50 100 150 200 250acceleration (g)
high
-g o
utpu
t (μV
)
y = 9.9141x + 105.38R2 = 0.9917
0
500
1000
1500
2000
2500
0 50 100 150 200acceleration (g)
high
-g o
utpu
t (μV
)
(a)
(b)
39
Chapter 4 Testing
and oxide layer thicknesses measured from past processing runs. The lateral accelerometer displace-
ment sensitivity in the modified simulation was 7.7 pm/gee, compared to 6.4 pm/gee from the simula-
tions in Chapter 2. The vertical accelerometer displacement sensitivity in the modified simulation was
16.5 pm/gee, compared to 18.5 pm/gee from the simulations in Chapter 2. The modified simulations
do not show an order of magnitude increase in displacement sensitivity, so it is unlikely that this is the
cause of the lowered measured sensitivity.
Another possible cause is that the amplifier gain is reduced from circuit simulations. Because the
amplifier inputs are not routed to bondpads, to minimize parasitic capacitance, the gain of the on-chip
amplifier has not been measured currently to verify this theory. To test this theory, additional drop tests
were conducted at a lower modulation frequency of 200 kHz, as opposed to 1.7 MHz. These drop tests
were conducted at the Army Research Laboratory on a full-suite drop tower with higher impact capa-
bilities. The drop test results, shown in Figure 4.8, show a measured sensitivity of 49.8 μV/gee for the
vertical accelerometer. This is still much lower than the simulated sensitivity of 130 μV/gee, but higher
than the sensitivity obtained with modulation frequency of 1.7 MHz. This indicates that the gain in the
electronics did contribute to the lower measured sensitivity in earlier drop tests. However it does not
fully explain the lower measured sensitivity for the vertical accelerometer. The vertical accelerometer
sensitivity remains to be investigated further.
The correlation coefficient for the measurement results in Figure 4.8 is 0.994. These results show
that the vertical accelerometer is functioning under shock accelerations beyond 30,000 gees.
40
Chapter 4 Testing
Figure 4.8 Measured CMOS-MEMS high-gee vertical accelerometer output with respect to maxi-
mum deceleration over several drop tests, with modulation frequency of 200 kHz.
4.3 .Future WorkTo investigate the discrepancy between simulated sensitivity and measured sensitivity, the resonant
frequency of the devices will be measured in the future. Amplifier gain will also be measured and
investigated as a possible source of reduced sensor sensitivity. The drop tower tests will be repeated
with an off-chip differential amplifier with a gain of 1000, to improve signal-to-noise ratio and obtain
more accurate sensitivity measurements. Performance reliability over time and temperature should be
investigated. Noise and cross-axis sensitivity should be characterized.
4.4 .ConclusionsA lateral and vertical high gee accelerometer capable of sensing accelerations has been designed
and shown to be functional at low input accelerations. The accelerometer design uses CMOS-MEMS
(a)
y = 0.0498x - 44.168R2 = 0.9938
0200400600800
10001200140016001800
0 10000 20000 30000 40000acceleration (g)
high
-g o
utpu
t (m
V)
41
Chapter 4 Testing
fabrication to be fully integrated with sense circuitry on a single chip, making the sensor sufficiently
small and lightweight for in-flight munitions testing. The tri-axial integrated sensor and circuitry have
been fabricated on a single 2.4 mm x 2.4 mm chip.
The work described here has been shown through hand calculation and FEA simulations to produce
an output signal whose voltage is linear to the external acceleration, up to 20,000 gees. Mechanical
curl due to processing variations has been reduced. Structural release has been achieved. Preliminary
testing indicates that the sense circuitry is functioning as expected, and that the accelerometer responds
linearly under input accelerations below 30,000 gee’s.
4.5 AcknowledgmentsThe work presented has been funded by the Test Resource Management Center (TRMC) Test and
Evaluation/Science and Technology (T&E/S&T) Program through the Naval Undersea Warfare Cen-
ter, Newport, Rhode Island, under the Army Research Laboratory. It has also been funded by a
National Science Foundation Graduate Fellowship. The author thanks Suresh Santhanam and the staff
of the Carnegie Mellon Nanofabrication Facility for processing assistance, and Rudolph Park and
Johns Hopkins University Applied Physics Laboratory for circuit design.
42
References
43
References[1] J. M. Tsai and G. Fedder, “Mechanical Noise-Limited CMOS-MEMS Accelerometers,” in
Technical Digest of the 18th IEEE International Conference on Microelectromechanical Sys-tems, 2005, pp. 630-633.
[2] C. Lo, F. Chen and G. Fedder, “Integrated HF CMOS-MEMS Square-Frame Resonators with On-Chip Electronics and Electrothermal Narrow Gap Mechanism,” in Technical Digest of the 13th International Conference on Solid-State Sensors, Actuators and Microsystems, 2005, pp. 2074-2077.
[3] S. Bedair and G. Fedder, “CMOS MEMS Oscillator for Gas Chemical Detection,” in Proceed-ings of the 3rd IEEE Sensors Conference, 2004, pp. 955-958.
[4] G. Fedder, S. Santhanam, M. L. Reed, S. C. Eagle, D. F. Guillou, M. S. Lu and R. Carley, “Laminated High-Aspect-Ratio Microstructures in a Conventional CMOS Process,” in Techni-cal Digest of the 9th IEEE International Workshop on Micro Electro Mechanical Systems, 1996, pp. 13-18.
[5] Y. Ning, Y. Loke, G. McKinnon, “Fabrication and characterization of high g-force, silicon piezoresistive accelerometers,” Sensors and Actuators A, vol. 48, pp. 55-61, May 1995.
[7] Endevco Corporation, Piezoresistive Accelerometer Model 7268C Datasheet.
[8] J. Suminto, “A Wide Frequency Range, Rugged Silicon Micro Accelerometer with Overrange Stops,” in Technical Digest of the 9th IEEE International Conference on Microelectromechan-ical Systems,1996, pp.180-185.
[9] "Capacitive Sensing," class notes for ECE 18-614, Department of Electrical and Computer Engineering, Carnegie Mellon University, Fall 2004.
[10] H. Luo, G. Zhang, R. Carley and G. Fedder, “A Post-CMOS Micromachined Lateral Acceler-ometer,” IEEE/ASME Journal of Microelectromechanical Systems, vol. 11, pp. 188-195, June 2002.
[11] H. Xie and G. Fedder, “A CMOS Z-axis capacitive accelerometer with comb-finger sensing,” in Proceedings of the 13th IEEE International Conference on Micro Electro Mechanical Sys-tems, 2000, pp. 496-501.
[12] B. Razavi, Design of Analog CMOS Integrated Circuits, pp. 215-217. Boston: McGraw Hilll, 2001.
[13] S. D. Senturia, Microsystem Design. Boston: Kluwer Academic Publishers, 2001.