JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO.
5, OCTOBER 2015
A MEMS Implementation of a Classic
Parametric Resonator
Shai Shmulevich, Inbar (Hotzen) Grinberg, and David Elata, Member, IEEE
Abstract We present a microelectromechanical systems
realization of a classic parametric resonator. This parametric
resonator is ideal in the sense that the electrostatic stiffness,
which may be time modulated, is not affected by motion. We also
present a simple, efficient, and intuitive model of parametric
excitation. This model predicts the minimal modulation
amplitude required to obtain an unbounded response in a
parametric system with linear damping. We show experimental
results in which the system is operated as a Meissner
resonator. [2014-0381]
Index Terms Electrostatic actuators, parametric excitation,
parametric resonators, tapered comb-drive.
I. I NTRODUCTION
M ICRO-ELECTRO-MECHANICAL (MEMS) reson-
ators are prevalent in sensors and filters, because of
their excellent dynamic response (e.g. high quality factor)
and their compatibility with electrical driving and sensing.
Electrostatic resonators were first proposed in the pioneering
work of Nathanson [1]. Since then, much effort has been
invested in achieving MEMS resonators with a linear dynamic
response [2], [3]. Extensive effort has also been invested in
reaching high resonance frequencies and very high quality
factors [4], [5], which are essential for filtering, clocking and
sensing applications.
In recent decades, significant work has been devoted to
utilize nonlinear dynamics in micro-systems, with specific
interest in parametrically actuated resonators [6]. Parametric
resonators are dynamic systems in which at least one of the
physical parameters (i.e. mass, damping, or stiffness) varies
periodically in time, resulting in special properties of their
dynamic response.
When a linear resonator is excited at the resonance
frequency, from rest at the unloaded state, the dynamic
response amplitude grows and eventually saturates at a peak
value. In contrast, when a perfect parametric resonator is
excited at the appropriate frequency, its dynamic response
Manuscript received December 16, 2014; revised February 2, 2015; accepted
February 6, 2015. Date of publication February 25, 2015; date of current
version September 29, 2015. This work was supported in part by the
Russell Berrie Nanotechnology Institute, and in part by the Micro and Nano
Fabrication Unit through the TechnionIsrael Institute of Technology, Haifa,
Israel. Subject Editor A. Seshia.
The authors are with the TechnionIsrael Institute of Technology,
Haifa, 32000, Israel (e-mail: shaishm2@gmail.com; inbar28@campus.
technion.ac.il; elata@technion.ac.il).
This paper has supplementary downloadable multimedia material available
at http://ieeexplore.ieee.org provided by the authors. This includes a PDF file,
providing Appendix A and Appendix B. This material is 71.33 kB in size.
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JMEMS.2015.2402223
1057-7157 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1285
amplitude grows exponentially [7], [8], and it cannot
be bounded by linear damping. Besides the possibility
of achieving large (theoretically unbounded) response
amplitudes, another appealing property of parametric
resonators is the sharp transition between stable and unstable
responses. This sharp transition enables sensitive frequency-
shift detection for sensing applications [9][11]. Furthermore,
in parametric excitation the applied driving frequency may
differ from the response frequency. This allows to use
parametric resonators in filtering applications (e.g. frequency
dividers) [12], [13], and to attenuate feed-through noise in
sensing applications. Hence, parametric actuation offers many
advantages for MEMS transducers.
However, existing MEMS parametric resonators
introduce many additional nonlinear effects besides pure
time-modulation of stiffness. In many of these devices, both
non-modulated and modulated stiffnesses are affected by
motion [14][18]. These nonlinear effects often overshadow
the characteristics of classic parametric excitation. In order
to implement a classic parametric resonator, it is necessary
to find a way to time-modulate stiffness, which is otherwise
not affected by motion.
Nonlinear effects are sometimes considered as blessing in
disguise, since they eventually cap the response amplitude.
After all, in realistic devices an unbounded response is
impossible. However, in the present work we aim to produce
the best parametric excitation: in the presented device, the
time-modulated electrostatic stiffness is unaffected by motion
and the nonlinear effects which cap the motion are only due
to the mechanical spring.
Recently we presented a MEMS parametric resonator that is
compatible with classic parametric systems [19]. In the present
study we present a comprehensive analysis of the dynamic
response of that device, and present a new model which
explains parametric excitation in a simple and intuitive way.
We present a parametric MEMS resonator that allows pure
time-modulation of a linear stiffness (i.e. motion independent
stiffness), without introducing any unwanted nonlinear electro-
static effects. This enables to design and fabricate a parametric
resonator that behaves exactly as a classic Meissner or Mathieu
system, depending on the modulation scheme [20][22].
In the next Section a simple excitation (SE) model of
parametric actuation is introduced. The SE model considers
a specific form of excitation which is tailored to simplify the
modeling of the system response. The SE model provides a
very intuitive explanation of parametric resonance. In addition
this model includes linear damping, and enables an intuitive
analytic prediction of the minimal modulation amplitude that
1286 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO. 5, OCTOBER 2015
is required to reach an unbounded response. In Section 3
a classic parametric resonator is introduced. This parametric
resonator is based on a tapered comb-drive [23] that provides
a motion-insensitive stiffness (i.e. linear stiffness), which can
be tuned by voltage. In the present study this tuning is utilized
to achieve time-modulation of the system stiffness, enabling
parametric excitation.
Experimental validation of the response of the classic para-
metric resonator is presented in Section 4, where the system is
operated as a Meissner resonator. In addition, the theoretically
predicted minimal modulation amplitude which is necessary to
overcome linear damping, is experimentally validated.
In the discussion (Section 5) we address a nonlinear
capping phenomenon which we observed in the experiments.
We explain why this nonlinearity is not associated with the
actuation scheme, but is rather related to inherent imperfection
of the elastic spring.
II. A S IMPLE E XCITATION (SE) M ODEL OF
PARAMETRIC E NERGY P UMPING
A. Background
A basic model of parametric resonance is Hills
equation [22], [24],
x + k(t)x = 0 (1)
where k(t) is a time-periodic function. Depending on the
functional form of k(t), the amplitude of the oscillations that
are solutions of Hills equation, may be either bounded for all
time or may grow unbounded.
One well-known special case of Hills equation is the
Mathieu equation [20], [22],
 
x + 2 + 2 cos(2t) x = 0 (2)
in which stiffness is an additive composition of a constant
component and a single harmonic component. The argument
of the cosine in (2) is sometimes 2t, as appears here, and
sometimes t (e.g. in [9]). The implications of this choice
are elaborated in Appendix A found in the supplementary
downloadable multimedia.
The Mathieu equation has been extensively investigated
since it is associated with classic problems in nonlinear
dynamics (e.g. inverted pendulum). The oscillations which
form the solutions of the Mathieu equation may be either stable
(i.e. bounded for all time), or unstable.
Another special case of Hills equation is the Meissner
equation [21], [25]
 
x + 2 + 2 sgn (cos(2t)) x = 0 (3)
in which stiffness is an additive composition of a constant
component and a square wave-form component. Here as well,
the argument of the cosine may have two representations.
The parameter in equations (2) and (3) corresponds to
the frequency of the system. For example, if the system was
(i.e. = 0) then would have been the same as
linear
n = k/m of a linear system. The square of this parameter
is therefore associated with the average stiffness of the
system (i.e. 2 k in a linear system). The term 2 in
Fig. 1. Stability map of a parametric system governed by: (a) Mathieu
equation, (b) Meissner equation. S marks regions of a stable dynamic
response, and U marks regions of an unstable response.
equations (2) and (3) is the amplitude of stiffness modulation.
As is common, the modulation frequency in (2) and (3) has
been normalized by the natural frequency n of the system
(See Appendix A found in the supplementary downloadable
multimedia).
The dynamic response of the system may be stable or
unstable depending on the modulation amplitude and
frequency. Each system has its own stability map which
describes the relation between the modulation parameters and
the stability of the dynamic response. The stability maps of
these special cases are shown in Fig. 1, where S and U
mark the regions of stable and unstable dynamic responses.
An unstable dynamic response is unchecked by linear
damping, and only a nonlinear mechanism can limit the ampli-
tude of dynamic vibrations. For low values of 2 the windows
of unstable response is very narrow (Fig. 1), though it may be
shown that this sharp window is blunted by linear damping.
To better understand the essence of parametric excitation, it
is constructive to consider the simple excitation (SE) model.
In this model (of which we have not found previous refer-
ence) the excitation scheme is tailored such that the analytic
derivation of the dynamic response is both simple and intuitive.
Moreover, this model explicitly includes damping which is
known to affect the tip of instability windows in practical
systems. In the present study we use the SE model to consider
the first instability window, i.e. when excitation modulation
frequency corresponds to 2n . In the next subsection we
present this model in detail.
B. SE Model
Consider a simple 1-DOF linear resonator with a mass m,
a nominal stiffness ka and a linear damping coefficient c. The
system starts from zero displacement x(0) = x 0 = 0 and a
small arbitrary positive velocity v(0) = v0 > 0, and it is not
subjected to any external force. The natural frequency of the
system at its initial state is n = ka /m. Next, assume that
the system includes a feature which allows to switch (at will)
between two different possible values of stiffness: ka and
a (k > 1) as illustrated in Fig. 2a. The velocity of the
kb = kk
mass will diminish when it reaches the maximal displacement
x 1 = v0 /a at time ta = /2a where a = n = ka /m.
We assume that damping is sufficiently small, c  1, so that
the natural frequency of damped oscillations is very close to
the natural frequency of the undamped system. The motion in
SHMULEVICH et al.: MEMS IMPLEMENTATION OF A CLASSIC PARAMETRIC RESONATOR
Fig. 2. Undamped response of the SE model. (a) Illustration of the stiffness scheme showing stiffness switching during one cycle. The dashed triangular
regions indicate the elastic energy pumped into the system in this cycle. (b) Simulation of parametric resonance with k = 2 and v0 = 1. With every cycle the
maximal displacement and maximal velocity increase by k.
the time interval 0 t ta is therefore harmonic, and the
velocity is va = x 1 a cos(a t).
At time ta , the stiffness is switched from ka to kb (points
and  in Fig. 2a) and hence the stored elastic energy in the
spring is instantaneously increased from 12 ka x 12 to 12 kb x 12 . The
mass will now freely travel back to x = 0,  but this will take
a . This time
a shorter time interval tb = /2b = /2 k
interval is a new harmonic response event, where the mass is
released from rest at x 1 . Accordingly, the velocity in the time
 
range ta t ta + tb is vb = x 1 k a sin a (t ta ) .
k energy stored in the system will increase by a factor of k 2 ,
At time ta + tb the mass
 reaches x=0 with a velocity of
vb max = x 1 b = x 1 k a.
At this point in time, the stiffness is switched back to ka ,
and the stiffness switching scheme is repeated with negative
displacements x < 0 (Fig. 2a). Effectively, the time interval
0 t ta + tb constitutes a half-cycle of the dynamic
response.
The SE model is ideal in the sense that stiffness is switched
between two extreme values: Stiffness is switched to the higher
value when the motion amplitude is maximal and velocity
(i.e. kinetic energy) vanishes. In this way, the increase in
stiffness maximizes the associated increase in energy. Stiffness
is switched to the lower value when the velocity is maximal
and the motion (i.e. elastic energy) vanishes. In this way,
the decrease in stiffness does not decrease the energy
(i.e. no energy is extracted from the system due to the
switching scheme).
It follows that duty-cycle, D = tb / (ta + tb ) is not fixed
(as in the Meissner and Mathieu resonators) but rather depends
  
on modulation ratio k, D = 1/ 1 + k . If operated in this
fashion, the SE model is actually a self-excited parametric
oscillator rather than a parametric resonator.
It can be shown that the total dissipated energy in the first
half-cycle is given by
1  
D = ka x 12 k+1 (4)
4 Q
where Q = ka m/c is the quality factor of the system when
stiffness is at the nominal value ka .
1287
The energy pumped into the system in the first half-cycle,
due to the abrupt increase in stiffness, is given by
1  
Win = ka x 12 k 1 (5)
2
As long as the pumped energy exceeds the dissipated energy,
the half-cycle will end at an increased speed. In this way,
when stiffness is repeatedly switched between ka and kb with
the previously described time intervals, a cycle of duration
 of = 2(ta + tb ) will be effective. With no damping, the
for
and the maximal velocity will increased by a factor of k,
every additional cycle (as presented in Fig. 2b).
For non-zero damping, equating (4) and (5) yields the
minimal stiffness ratio required to overcome the linear
damping, and reach an unbounded dynamic response. This
critical ratio is given by
 
2
kcr = 1 + (6)
2Q
The SE model excitation scheme, with its coordinated duration
of stiffness modulation, pumps net energy into the system in
each half cycle for k > kcr . This results in an unbounded
motion response similar to the Mathieu and Meissner
resonators.
In recent years many different electrostatic actuators were
used to induce parametric excitation [6]. However, these
electrostatic actuators inevitably included additional nonlinear
effects, and did not afford a pure modulation of stiffness. The
additional nonlinear effects meant that parametric excitation
was tainted, and this affected the stability map of the systems.
In this work we use a transducer which enables to tune the
linear stiffness without inducing any other effects [23]. In the
next Section we will show how we use this device to time-
modulate the stiffness of a linear resonator, and implement
a classic parametric resonator.
III. A C LASSIC PARAMETRIC R ESONATOR
Consider a symmetric comb-drive with linearly tapered
finger lengths, as presented in Fig. 3 [23]. This design enables
1288 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO. 5, OCTOBER 2015

Applying this voltage scheme to (7) yields,

 
2
k V Veq
m x + c x + k 1
k


2k V Vac Veq Vac sgn (cos (2t)) x = 0
2 2

(10)
We may rewrite this equation in the non-dimensional form
Fig. 3. Schematic view of the symmetric comb-drive with linearly tapered   
finger lengths [23]. d2x n dx 2 2
+ + n (1 K T ) n 2K T Vac 1 Vac 2
d 2 Q d

to change the linear stiffness of the system by applying a sgn (cos (2 )) x = 0 (11)
dc voltage to the stator fingers. The equation of motion of
this system is given by, where

  k n d x n d x d2x n2 d 2 x
n = , = t = t, = , =
m x + c x + k k V V 2 x = 0 (7) m n d n dt d 2 n2 dt 2
2
km Vac k V Veq
Q= , Vac = , KT = (12)
where k V = n gO0 wL is the stiffness softening coefficient c Veq k
(See Appendix B found in the supplementary downloadable The first term in the square brackets of (11) is the constant
multimedia). stiffness of the system ( 2 ), and the second term is the
Here m is the rotor mass, c is the linear damping coefficient, time-modulated stiffness ( 2 ). The term K T represents the
k is the nominal linear stiffness of the suspension, n is normalized electrostatic stiffness of the system, which may
the number of tapered fingers on each side of the rotor, be used to tune the location of the instability window [26].
is the permittivity of free-space, w is the device thickness, The time-modulated stiffness is affected by the electrostatic
g is the gap between rotor and stator fingers, and OL is the stiffness K T and by the amplitude of the modulated
initial overlap where motion is in the range OL < x < OL. 2 must not exceed V 2 , but this
voltage Vac . It is clear that Vac eq
It is clear from (7) that the stiffness of the system can be 2 > 0.
is trivial since all it really means is that Vdc
tuned down by application of a dc bias, and that this stiffness The form of (11) resembles that of the Meissner equation (3)
is otherwise constant, i.e. it is unaffected by motion. (and more specifically resembles the form (A4 - supplementary
This actuator induces a diverging electrostatic force which downloadable multimedia)), and includes a linear damping
is a perfectly linear function of displacement. This force is term. The fact that the second term in the square brackets
exactly the opposite of the restoring force in a linear spring, is preceded by a minus sign is not important due to the effect
and hence we refer to this actuator as a linear anti-spring. of the sgn function.
Moreover, applying a time-modulated voltage will cause time- Finally, it is clear from the second term of (11), that for
modulation of the system stiffness, resulting in parametric modulations of higher order n > 1 (i.e. lower modulation
excitation. However, the modulated stiffness is proportional frequencies), damping is more effective. In fact, it seems that
to voltage-squared. This means that if the applied voltage has the effective quality factor is inversely proportional to the
a square waveform V (t) = Vdc + Vac sgn (cos(t)), then the modulation order n. This point will be elaborated in future
average voltage squared (and hence the average stiffness) will work.
also be affected by Vac , and not only by Vdc as would be In the next section we present experimental results of the
preferable. new parametric resonator, when it is operated as a Meissner
Therefore, we choose to apply a voltage of the form resonator. In addition, we present experimental results of the
minimal modulation amplitude required to achieve a strongly
diverging response, for several values of effective damping.
V (t) = 2 V 2 + V sgn (cos (2t))
Veq ac ac (8) For low damping, this minimal modulation amplitude (i.e. the
tip of the instability window) is similar for both the Meissner
2 = V 2 +V 2 = const such that the average stiffness and the SE parametric resonators. This is because at small
where Veq dc ac damping (see (6)), a low modulation ratio k = kb /ka 1
is constant.
means that the duty-cycle D = tb /(ta + tb ) converges to 50%.
Consequently, the voltage-squared is an additive compo-
sition of a constant component and a square wave-form
IV. E XPERIMENTAL R ESULTS
component.
A. Test Devices
Test devices were fabricated using SOIMUMPs technology
V (t)2 = Veq
2
+ 2Vac Veq
2 V 2 sgn (cos (2t))
ac (9) (run 43, [27]). The devices were designed (Fig. 4) such that
SHMULEVICH et al.: MEMS IMPLEMENTATION OF A CLASSIC PARAMETRIC RESONATOR
Fig. 4. Schematic view of an electrostatic resonator with two actuators shar-
ing the same shuttle. The lower transducer is a standard double-sided comb-
drive actuator, and the upper transducer is the anti-spring. The configuration
of the wiring dictates the type of actuation. (a) Driving scheme for a standard
linear resonator, enabling the study of stiffness down-tuning as function of
Vtune , and the measurement of the linear damping of the system. (b) Driving
scheme for a parametric resonator.
Fig. 5. Microphoto of a typical test device with partial schematics of
the experimental setup. The insets show the anti-spring with tapered finger
lengths, and the standard double-sided comb-drive actuator.
the shuttle can be excited by a standard double-sided linear
electrostatic actuator (driving scheme (a) in Fig. 4), or be
excited parametrically by an electrostatic anti-spring (driving
scheme (b) in Fig. 4).
Test devices were fabricated in a (100) single crystalline
silicon, and the suspension flexure beams were oriented in the
(110) direction. A typical device is presented in Fig. 5. The
anti-spring transducer (upper), and the standard double-side
comb drive (lower), are connected in series, sharing the
same rotor. Therefore, the same damping mechanism is equally
effective for any mode of operation of the system. The rotor is
1289
Fig. 6. Measured natural frequency-squared as function of tuning voltage-
squared. The excellent linear fit validates that linear modulation of the stiffness
may be achieved. The inset in the figure presents down-tuning of the measured
natural frequency f n as function of Vtune . These results are compatible with
the findings presented in [28].
suspended on four folded-beam springs and its nominal natural
frequency is about 700Hz. The finger width and minimal
trench are both designed to be 5m, and the device layer
thickness is 25m. The overlap length of the anti-spring
fingers was tapered by 0.4m/finger.
B. Experimental Set-Up and Preliminary Characterization
The devices were first driven by the linear double-sided
actuator to measure the suspension stiffness and the
system damping, and to investigate the tuning properties when
an additional fixed bias voltage was applied to the anti-spring.
In this experiment, the frequency response of the system was
measured in atmospheric pressure, using an experimental setup
similar to the one shown in Fig. 5. For this experiment we used
a lock-in amplifier for driving and sensing.
The rotor (i.e. shuttle) was subjected to a constant dc bias
voltage of Vdc =5V. A low Vac signal (typically 0.1VRMS)
was applied to the comb-drive stator, and a dc tuning voltage,
Vdc +Vt une , was applied to the anti-spring stators. A laser
vibrometer was used to measure the vibration velocity.
Reasonable SNR was achieved by the lock-in amplifier which
provided the reference ac signal to the comb-drive actuator
and acquired the signal from the vibrometer with respect to
the same ac frequency.
Figure 6 presents the measured natural frequency-squared,
f n2, as function of the applied electrostatic tuning voltage-
squared, Vt2une . The very high linear correlation of the data
(R 2 =0.99997) validates the predicted down-tuning relation
in (7), and validates that the stiffness may be modulated
without introducing any unwanted nonlinear effects (similar
results have been previously shown in [28]). The inset in
the figure presents down-tuning of the measured natural
frequency f n as function of Vt une .
C. Measured Response of a Meissner Resonator
Next, the test device was operated as a Meissner parametric
resonator, by applying modulated voltage to the anti-spring
(Fig. 5), which in turn modulated the stiffness of the device.
The experiments were performed at ambient vacuum of
1290 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO. 5, OCTOBER 2015
Fig. 7. A strongly diverging response for a modulated square-wave signal
of 6[V ]. The inset shows that the modulation signal has twice the frequency
of the response, as expected. Mechanical bumpers were designed to bound
motion at 10m displacement. The response grows exponentially, and is
eventually capped before the bumpers by nonlinear mechanical effects [29],
[30].
Fig. 8. Stability map of the first instability window of a Meissner resonator.
Theoretical boundaries are indicated by the solid line, and the measured results
are indicated by marks. The inset presents a larger region of the stability
map. Measurements were performed in vacuum of 6mTorr.
6mTorr. Modulation frequency was slowly swept in the
vicinity of double the natural frequency of the system.
This frequency is associated with the first instability
window (Fig. 1b). At the detection of a strongly diverging
motion increase, the system was brought back to rest. Then,
the parametric actuation was repeated in that very same
frequency, to verify that the strongly diverging response
indeed results from parametric excitation. Figure 7 presents a
typical measured response of the device, when it is actuated
from rest. It is evident from the exponential growth of the
motion amplitude envelope (thick line), and from the double
frequency actuation with respect to the response frequency
(inset), that the response is indeed a typical parametric
response. The dynamic response is eventually bounded
by a nonlinear effect [29], [30], which is not associated
with parametric resonance and will be discussed later
in Section 5.
Fig. 9. Minimal stiffness modulation ratio, kcr , at the tip of the instability
window, for different values of Q. The good fit between measured and
predicted values validates that the system can respond as a SE resonator. For
low values of Q (i.e. higher damping) a larger kcr is required, but this also
affects the duty-cycle. Since the system was driven as a Meissner resonator
with a duty-cycle of D=50%, the prediction for Q=20 and Q=35 are not
as good.
To characterize the first instability window of the Meissner
resonator, different voltage modulation amplitudes were
applied. For each modulation amplitude, frequency was swept,
back and forth, to identify the window in which the response
becomes unstable. Figure 8 presents the measured points at
which the response switched from stable to unstable. The
very good correlation between the measured points and the
predicted stability map, confirms that the parametric resonator
presented in this work can perform as a classic parametric
resonator.
D. Measured Response Threshold
The test devices were operated as linear resonators to
measure the quality factor of the system, Q, associated with
linear damping. Specifically, Q values were extracted from the
sharpness of measured frequency sweeps, at different vacuum
levels [31]. At each vacuum level, the device was then operated
as a parametric resonator. The amplitude of the modulated
voltage was gradually increased until a strongly diverging
response was observed. During these experiments, frequency
was swept to ensure that the bottom tip of the instability
window (Fig. 8) was identified. The minimal modulated
voltage associated with the instability window tip was thus
measured. Using the tuning relation (Fig. 6), the minimal
stiffness ratio, kcr necessary for inducing instability was
extracted.
The duty-cycle of the SE model is determined by the
stiffness ratio, k (Fig. 2b). However, in sufficiently high values
of Q, the SE model duty-cycle converges to 50%. This means
that for sufficiently high values of Q, the instability window
tip of the Meissner resonator and that of the SE resonator, are
the same.
Figure 9 compares the predicted (Eq. (6)) and measured
values of the minimal stiffness ratio, kcr . The good fit between
prediction and measurement validates that the effect of linear
damping is properly accounted for in the SE model.
SHMULEVICH et al.: MEMS IMPLEMENTATION OF A CLASSIC PARAMETRIC RESONATOR
V. D ISCUSSION AND C ONCLUSION
In this work we have presented a parametric resonator
based on a symmetric comb-drive anti-spring [23]. This
parametric resonator enables to modulate stiffness without
inducing any additional nonlinear effects. Specifically, we
demonstrated that this resonator may be operated as the classic
Meissner resonator or as an SE resonator. Though the option
was not demonstrated in the present study, a Mathieu resonator
(Eq. (2)) may also be implemented by applying a voltage of
the form V (t) = (Vdc + Vac cos(t))1/2 [9].
However, the response measured in Fig. 7 demonstrates
that the motion is bounded by some unforeseen effect - the
maximal motion was smaller than the maximal designed travel
range of 10m. We attribute this capping of vibration
amplitude to mechanical nonlinear stiffening, which is inherent
to folded-beam suspensions [29], [30]. The source of this
stiffening is inertial effects which induce membrane stiffening
in the flexures of standard folded-beam suspensions. Though
standard folded-beam suspensions are specifically designed to
respond as linear springs, they do so only in static loading.
In harmonic response the standard folded-beam suspension is
linear only in an extremely limited range of motion of the
shuttle [29]. We explained this phenomenon in detail, and
proposed a solution for the problem in a recent work [30].
In any case, this mechanical stiffening is insignificant for
small displacements, and does not affect the threshold of the
instability we investigated.
The presented parametric device in which stiffness modula-
tion is purely time-dependent, allows further investigation and
implementation of classic parametric resonators and oscillators
in MEMS.
In this work we presented the SE model of parametric
excitation which includes linear damping, and we derived a
simple analytic form for the critical modulation associated
with the onset of a strongly diverging response. As discussed
in Section 2.2, the SE model is an effective way to pump
energy into a resonator. Based on the analytic relation between
quality factor and critical modulation (Eq. (6)), operation of
the system as an SE resonator enables an alternative method
to measure damping.
R EFERENCES
[1] H. C. Nathanson, W. E. Newell, R. A. Wickstrom, and J. R. Davis, Jr.,
The resonant gate transistor, IEEE Trans. Electron Devices, vol. 14,
no. 3, pp. 117133, Mar. 1967.
[2] W. C. Tang, T.-C. H. Nguyen, and R. T. Howe, Laterally driven
polysilicon resonant microstructures, Sens. Actuators, vol. 20, nos. 12,
pp. 2532, Nov. 1989.
[3] C. Marxer, O. Manzardo, H. P. Herzig, R. Dandliker, and
N. De-Rooij, An electrostatic actuator with large dynamic range and
linear displacement-voltage behavior for a miniature spectrometer, in
Proc. Transducers, Sendai, Japan, 1999, pp. 786789.
[4] C. T.-C. Nguyen, MEMS technology for timing and frequency control,
IEEE Trans. Ultrason., Ferroelectr., Freq. Control, vol. 54, no. 2,
pp. 251270, Feb. 2007.
[5] D. Weinstein and S. A. Bhave, Internal dielectric transduction in
bulk-mode resonators, J. Microelectromech. Syst., vol. 18, no. 6,
pp. 14011408, Dec. 2009.
[6] K. Moran, C. Burgner, S. Shaw, and K. Turner, A review of parametric
resonance in microelectromechanical systems, IEICE Nonlinear Theory
Appl., vol. 4, no. 3, pp. 198224, 2013.
1291
[7] R. B. Karabalin, X. L. Feng, and M. L. Roukes, Parametric nanome-
chanical amplification at very high frequency, Nano Lett., vol. 9, no. 9,
pp. 31163123, 2009.
[8] C. Guo and G. K. Fedder, A quadratic-shaped-finger comb parametric
resonator, J. Micromech. Microeng., vol. 23, no. 9, p. 095007,
2013.
[9] K. L. Turner, S. A. Miller, P. G. Hartwell, N. C. MacDonald,
S. H. Strogatz, and S. G. Adams, Five parametric resonances in
a microelectromechanical system, Nature, vol. 396, pp. 149152,
Nov. 1998.
[10] C. B. Burgner, W. S. Snyders, and K. L. Turner, Control of MEMS on
the edge of instability, in Proc. IEEE TRANSDUCERS, Beijing, China,
Jun. 2011, pp. 19901993.
[11] Z. Yie, M. A. Zielke, C. B. Burgner, and K. L. Turner, Comparison
of parametric and linear mass detection in the presence of detection
noise, J. Micromech. Microeng., vol. 21, no. 2, p. 025027,
2011.
[12] J. F. Rhoads, S. W. Shaw, K. L. Turner, and R. Baskaran, Tun-
able microelectromechanical filters that exploit parametric resonance,
J. Vibrat. Acoust., vol. 127, no. 5, pp. 423430, Jan. 2005.
[13] B. S. Strachan, S. W. Shaw, and O. Kogan, Subharmonic resonance
cascades in a class of coupled resonators, J. Comput. Nonlinear Dyn.,
vol. 8, no. 4, p. 041015, Jun. 2013.
[14] S. G. Adams, F. M. Bertsch, K. A. Shaw, and N. C. MacDonald, Inde-
pendent tuning of linear and nonlinear stiffness coefficients [actuators],
J. Microelectromech. Syst., vol. 7, no. 2, pp. 172180, Jun. 1998.
[15] W. Zhang, R. Baskaran, and K. Turner, Tuning the dynamic behavior
of parametric resonance in a micromechanical oscillator, Appl. Phys.
Lett., vol. 82, no. 1, pp. 130132, 2003.
[16] J. F. Rhoads, S. W. Shaw, K. L. Turner, J. Moehlis, B. E. DeMartini,
and W. Zhang, Generalized parametric resonance in electrostatically
actuated microelectromechanical oscillators, J. Sound Vibrat., vol. 296,
no. 4, pp. 797829, 2006.
[17] R. Lifshitz and M. C. Cross, Nonlinear dynamics of nanomechanical
and micromechanical resonators, in Review of Nonlinear Dynamics and
Complexity, vol. 1, H. Schuste, Ed. Weinheim, Germany: Wiley, 2008,
pp. 152.
[18] J. F. Rhoads, S. W. Shaw, and K. L. Turner, Nonlinear dynamics and its
applications in micro- and nanoresonators, J. Dyn. Syst., Meas., Control,
vol. 132, no. 3, p. 034001, Apr. 2010.
[19] S. Shmulevich, I. Hotzen, and D. Elata, An ideal MEMS parametric
resonator using a tapered comb-drive, in Proc. Eng., vol. 87. Brescia,
Italy, 2014, pp. 14811484.
[20] . Mathieu, Mmoire sur le mouvement vibratoire dune membrane de
forme elliptique, J. Math. Pures Appl., vol. 13, pp. 137203, 1868.
[21] E. Meissner, Ueber Schttelerscheinungen in systemen mit periodisch
vernderlicher Elastizitt, Schweizer Bauzeitung, vol. 72, no. 11,
pp. 9598, 1918.
[22] A. H. Nayfeh and D. T. Mook, Nonlinear Oscillations. Hoboken, NJ,
USA: Willey, 1979.
[23] K. B. Lee and Y.-H. Cho, A triangular electrostatic comb array for
micromechanical resonant frequency tuning, Sens. Actuators A, Phys.,
vol. 70, nos. 12, pp. 112117, Oct. 1998.
[24] G. W. Hill, On the part of the motion of the lunar perigee which is a
function of the mean motions of the sun and moon, Acta Math., vol. 8,
no. 1, pp. 136, 1886.
[25] H. Hochstadt, A special Hills equation with discontinuous coefficients,
Amer. Math. Monthly, vol. 70, pp. 1826, 1963.
[26] S. Shmulevich, I. Hotzen, and D. Elata, Tuning the first instability
window of a MEMS Meissner parametric resonator using a linear
electrostatic anti-spring, in Proc. IEEE MEMS, Estoril, Portugal, 2015.
[27] [Online]. Available: http://www.memscap.com/products/mumps/
soimumps
[28] D. Scheibner, J. Mehner, D. Reuter, U. Kotarsky, T. Gessner,
and W. Dtzel, Characterization and self-test of electrostatically
tunable resonators for frequency selective vibration measurements,
Sens. Actuators A, Phys., vol. 111, no. 1, pp. 9399, Mar. 2004.
[29] S. Shmulevich, A. Joffe, I. Hotzen, and D. Elata, Are folded-beam
suspensions really linear? in Proc. Eng., vol. 87. Brescia, Italy, 2014,
pp. 624627.
[30] S. Shmulevich, I. Hotzen, and D. Elata, Dynamically-balanced folded-
beam suspensions, in Proc. IEEE MEMS, Estoril, Portugal, 2015.
[31] S. Shmulevich, M. Lerman, and D. Elata, On the quality of quality-
factor in gap-closing electrostatic resonators, J. Micromech. Microeng.,
vol. 23, no. 11, p. 115010, 2013.
1292 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO. 5, OCTOBER 2015
[32] D. Elata, Modeling the electromechanical response of electrostatic
actuators, in MEMS/NEMS: Handbook Techniques and Applications,
vol. 4, C. T. Leondes, Ed. New York, NY, USA: Springer, 2006,
pp. 93119.
Shai Shmulevich received the B.Sc. and
M.Sc. degrees from the Faculty of Mechanical
Engineering, TechnionIsrael Institute of
Technology, in 1999 and 2012, respectively,
where he is currently pursuing the Ph.D. degree.
From 2000 to 2011, he held a research and
development position with the industry, focusing on
mechanical design and electro-optical systems. His
fields of interest include design, simulation, and
characterization of electrostatic MEMS devices.
Inbar (Hotzen) Grinberg received the B.Sc. degree
from the Faculty of Mechanical Engineering,
TechnionIsrael Institute of Technology, in 2008,
where she is currently pursuing the Ph.D. degree.
She held a student research and development posi-
tion with Rafael Advanced Defense Systems Ltd.,
during her undergraduate degree. Her research inter-
ests include design, simulations, and fabrication of
MEMS actuators.
David Elata (M01) received the B.Sc. degree
from Ben-Gurion University, Beer Sheva, Israel,
in 1986, and the M.Sc. and D.Sc. degrees from
the TechnionIsrael Institute of Technology
(Technion), in 1989 and 1993, respectively, all in
mechanical engineering. From 1993 to 1996, he
was a Post-Doctoral Research Staff Member with
the Lawrence Livermore National Laboratory, and a
Visiting Post-Doctoral Fellow with the Geophysics
Department, Stanford University. Since 1996, he has
been with the Faculty of Mechanical Engineering,
Technion. His current research interests are in modeling and design of
MEMS actuators, and the development of novel concepts for MEMS devices.