Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Contents lists available at ScienceDirect

Commun Nonlinear Sci Numer Simulat

journal homepage: www.elsevier.com/locate/cnsns

Research paper

Design and modeling of a MEMS accelerometer based on

coupled mode-localized nonlinear resonators under
electrostatic actuation
Ming Lyu a, Jian Zhao a,∗, Najib Kacem b, Bin Tang c, Pengbo Liu a, Jiahao Song a,
Heng Zhong a, Yu Huang d
a
State Key Laboratory of Structural Analysis for Industrial Equipment, Dalian University of Technology, Dalian, China
b
Univ. Bourgogne Franche-Comté, FEMTO-ST Institute, UMR 6174, CNRS/UFC/ENSMM/UTBM, Department of Applied Mechanics, Besançon,
France
c
Institute of Electronic Engineering, China Academy of Engineering Physics, Mianyang, China
d
College of Ocean and Civil Engineering, Dalian Ocean University, Dalian, China

a r t i c l e i n f o a b s t r a c t

Article history: A novel dual proof mass accelerometer is proposed by introducing mode localization in
Received 25 February 2021 two electrostatically coupled resonators. The levering mechanism is utilized to amplify the
Revised 30 May 2021
inertial force applied axially to the two weakly coupled resonators. The dynamic model
Accepted 6 July 2021
considering the electrostatic and mechanical nonlinearities is established and solved by the
Available online 10 July 2021
method of multiple scales, and also is validated by the harmonic balance method (HBM)
Keywords: coupled with the asymptotic numerical method (ANM). The linear and nonlinear sensitiv-
Mode localization ities depicted as the difference of relative shift of amplitude ratio are investigated as well
Accelerometer as the main influencing factors. It has been found that linear sensitivity is up to 4 orders
Electrostatic coupling of magnitude higher than that expressed by the difference of relative frequency shift. Also,
Nonlinear dynamics the nonlinear sensitivity is further increased by 1.47 times comparing to the linear sensi-
Multiple scales method tivity. Moreover, the resolution is also greatly improved when the sensor is driven beyond
its critical amplitude. Finally, the effect of the electrostatic coupling on the sensing perfor-
mances is explored, and the optimal coupling voltage is theoretically identified at the limit
of mode aliasing.
© 2021 Elsevier B.V. All rights reserved.

1. Introduction

MEMS accelerometer is one of the earliest commercialized MEMS products and is widely used in the fields of automotive,
consumer electronics and aerospace [1, 2]. Practically, exploring effective method to improve the sensitivity of such sensors
is in great demand for the field of high precision detection. Common accelerometers can be divided into two types according
to the sensing mechanisms, static and dynamic sensors [3–5]. As a typical dynamic sensor, the resonant accelerometers can
measure the acceleration by evaluating the frequency shift induced by inertial force, which have been widely used due to

Abbreviations: HBM, harmonic balance method; MMS, method of multiple scales; ANM, asymptotic numerical method; RSAR, relative shift of amplitude
ratio; RSF, relative shift of frequency.
∗
Corresponding author.
E-mail address: Jzhao@dlut.edu.cn (J. Zhao).

https://doi.org/10.1016/j.cnsns.2021.105960
1007-5704/© 2021 Elsevier B.V. All rights reserved.
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Nomenclature

M proof masses
d distance of resonator from elastic pin
Lg distance from the center of the proof mass to the anchor
l resonator length
h resonator thickness
b resonator width
E Young’s modulus
c linear viscous damping
gc , ga air gap size
le electrodes length
Q quality factor
Vc coupling voltage
Vdc1 , Vdc2 bias voltage
Vac1 actuate voltage
wi,j jth amplitude of the ith resonator
ωi,j jth frequency of the ith resonator

Greek Symbols
γ external acceleration
ρ resonator density
θ proof masses rotation angle
ε0 dielectric constant

their high sensitivity and quasi-digital signal. Hence, the research topic of achieving high sensitivity has attracted extensive
attentions. One effective way is to introduce the leveraging mechanisms to amplify the external inertial force. Comi et al.
[6] used a single-stage lever to improve the sensitivity, which was verified through experiments that the sensitivity can
be up to 455Hz/g. Ding et al. [7] used a two-stage lever mechanism in the resonant accelerometer for measuring tilt, and
the high sensitivity was verified through close-loop and open-loop experiments. Enhancing the accelerometer performances
cannot be achieved by increasing the number of amplifying levers indefinitely, otherwise the sensitivity will be decreased
[8]. Therefore, the improvement of sensitivity by just using the levering mechanism encounters a bottleneck. Besides, the
performance of resonant accelerometer is seriously restricted due to its frequency drift and temperature sensitivity.
Recently, mode-localized sensors composed of two or more resonators coupled to each other allowing spatial energy
expansion have been widely reported [9–11]. Compared with traditional resonant sensors, such sensors introducing mode
localization mechanisms possess advantages of high sensitivity and good common-mode rejection [12–15]. Another key iden-
tifier of such sensing method is that the readout metric is basically based on amplitude modulation instead of traditional
frequency shift [16, 17]. Spletzer et al. [18] first utilized mode localization in two coupled microcantilevers as a mass sensor.
And, by comparing the eigenvalue output with the frequency shift, the sensitivity is improved by two orders of magni-
tude. Zhang et al. [19] proposed a mode-localized accelerometer composed of two resonators coupled by an elastic beam,
and it is experimentally verified that the variation of the amplitude ratio induced by inertial force is more than 300 times
higher than the traditional frequency shift. Zhao et al. [20–22] have also done a series of work on the application of mode
localization in mass sensors and inertial sensors. Moreover, the three output metrics of the sensitivity commonly used in
mode localization sensors are depicted by eigenstate, amplitude ratio, and amplitude shift, and it was found that the output
sensitivity of amplitude ratio is the highest [22].
To further improve the performance of the mode localized sensors, a series of methodologies have been introduced such
as adjusting the actuation voltage, varying coupling method and changing the number of coupled resonators [23, 24]. The
coupling strength is the most critical factor, which can be modulated by the coupling style involving electrostatic coupling
[25] and mechanical coupling [21]. The electrostatic coupling has the benefit in terms of high tunability, and was verified for
the first time in [25]. After that, Zhao et al. [26] proposed a three-degree-of-freedom mode-localized sensor using an elec-
trostatic coupling, they demonstrated that this coupling method is more sensitive than other methods through experiments.
Lyu et al. [23] proposed a mass sensor design based on a pair of electrostatically coupled clamped-clamped microbeams of
unequal length, the sensitivity output under different coupling voltages has been studied, and its maximum can be achieved
while tuning the electrostatic coupling. In [27], a mode localization accelerometer with four degrees of freedom and two
coupling methods in parallel is proposed, and the open-loop circuit test has been performed to verify that the normalized
sensitivity range can reach 2050%. In addition to the high sensitivity of mode-localized sensors, the anti-interference ability
to external environmental disturbances is another advantage. Thiruvenkatanathan et al. [28] studied the eigenstates and fre-
quency shift output under different external temperatures and pressure, respectively, and the results showed that eigenstate
output has better robustness. In [29], it has been demonstrated that the amplitude ratio output can not only improve the

2
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Fig. 1. Sketch of the dual-mass mode-localized accelerometer.

Table 1
Parameters selected in the simulation.

Parameter Value Parameter Value

Length of resonator (l) 150 μm Length of electrode (le ) 100 μm

Resonator height (h) 1 μm Resonator width (b) 4 μm
Young’s modulus (E) 169 GPa Density (ρ ) 2320 kg/m3
Air gap (ga ) 2 μm Air gap (gc ) 20 μm
Distance (d) 5 μm Distance (Lg ) 260μm

sensitivity but also eliminate the ambient pressure drift. The current research on mode-localized accelerometers is mainly
limited to linear vibrations, while to improve the resolution, the vibration amplitude of the resonator must be increased and
the nonlinear behavior of the device must be considered.
In this paper, a novel dual proof mass accelerometer is proposed by introducing mode localization in two electrostati-
cally coupled resonators. This structure uses a lever mechanism to transmit the motion of the proof mass in term of axial
force applied to the weakly coupled resonators, which causes vibration localization. The Multiphysics model is established
while including electrostatic and mechanical nonlinearity. The coupled equations of motion have been solved using Galerkin
discretization associated with the method of multiple scales (MMS), and the results have been verified by a computational
solving procedure based on the harmonic balance method (HBM) coupled with the asymptotic numerical method (ANM).
The simulations have been performed to investigate the sensitivity changes for the linear and nonlinear cases. Moreover,
the nonlinearity functionalization for acceleration sensing utilizing mode localization has been demonstrated to enhance
the sensitivity and vibration amplitude. When the drive voltage is increased, not only the sensitivity can be improved but
also the resolution can be increased with localized vibrations of the second mode. Finally, the effect of the coupling voltage
on the sensor’s sensitivity is analyzed in detail, and the mode aliasing is clearly shown through the amplitude-frequency
response curve, so that the theoretical optimal coupling voltage can be obtained.

2. Structure of the accelerometer

The proposed structure of the accelerometer is shown in Fig. 1, and the relevant parameters are listed in Table 1. The two
proof masses M are suspended on the substrate by the elastic pin, and two resonators are placed at a distance d from the
elastic pin. The two resonators are clamped-clamped beams driven by electrode 1 and electrode 2 respectively, and initially
balanced. It is worth noting that the two resonators are coupled with an electrostatic force. When the two proof masses are
subjected to the external acceleration γ along the e2 direction, the resonators state becomes unbalanced by the axial forces
transmitted to the resonators via a lever mechanism, thus resulting in a drastic amplitude change and vibration energy lo-
calization. Then, the magnitude of the external acceleration can be estimated by calculating the amplitude ratio variation
of the two resonators. In this paper, an accurate Mmultiphysics model of nonlinear electrostatically coupled accelerometer
is established, which can be used as a quick tool for improving the performances of accelerometers by utilizing mode lo-
calization. Thus, it will useful in some applications that require high sensitivity, such as monitor vital signs and inclination
measurement [2, 30].

2.1. Equations of motion

The studied structure is shown in Fig. 1, the sensitive part of the accelerometer is composed of two coupled clamped-
clamped beams subjected to a viscous damping c˜ and actuated by an electric load. The transverse deflection of the two

3
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

microbeams w˜ 1 (x˜, t˜) andw

˜ 2 (x˜, t˜) are governed by the nonlinear Euler–Bernoulli equation, which is the commonly used ap-
proximate equation of motion for a thin beam.
     
Ebh3 ∂ 4 w˜ 1 x˜, t˜ ∂ 2 w˜ 1 x˜, t˜ ∂ w˜ 1 x˜, t˜
+ ρ bh + c˜
⎡ ∂ x˜4  ∂ t˜2 ∂ t˜
  2 ⎤ 2  
12
l ∂w , ∂
Ebh ˜ x˜ ˜
t w˜ 1 ˜, t˜
x
−⎣N dx˜⎦
1
˜γ + =
2l ∂ x˜ ∂ x˜2
0
  2
ε0 b Vdc1 + Vac1 cos  ˜ t˜ H (x˜) ε0 bVc2
  2 −     2
2 ga − w ˜ 1 x˜, t˜ 2 gc + w ˜ 1 x˜, t˜ − w
˜ 2 x˜, t˜
      (1)
Ebh3 ∂ w ˜ 2 x˜, t˜ ∂ 2 w˜ 2 x˜, t˜ ∂ w˜ 2 x˜, t˜
4
+ ρ bh + c˜
⎡ ∂ x˜ ∂t ∂ t˜
  2 ⎤ 2  
12 4 ˜2

l ∂w ˜ 2 x˜, t˜ ∂ w˜ 2 x˜, t˜
Ebh
−⎣−N ˜γ + dx˜⎦ =
2l ∂x ˜ ∂ x˜2
0
ε0 bVdc2
2
H (x˜) ε0 bVc2
−   2 +     2
2 ga + w
˜ 2 x˜, t˜ 2 gc + w
˜ 1 x˜, t˜ − w
˜ 2 x˜, t˜

And the Heaviside function and boundary conditions are defined by

l − le l + le
H (x˜) = H x˜ − − H x˜ − (2)
2 2

   
∂ w˜ 1  ˜ ∂ w˜ 1  ˜
˜ 1 0, t˜ = w
w ˜ 1 l, t˜ = 0, t = l, t = 0
  ∂∂w˜x2   ∂ x˜
˜
  ∂ w˜ 2  ˜ (3)
˜ 2 0, t˜ = w
w ˜ 2 l, t˜ = 0, t˜ = l, t = 0
∂ x˜ ∂ x˜
where x˜ is position on the resonator, c˜is the linear viscous damping. E, ρ , b, h, l represents Young’s modulus, density,
widths, thickness, and lengths, respectively. gc , ga is the air gap size, w ˜ 1, w
˜ 2 are the two resonators transverse deflections,
le is electrodes length, ε 0 is dielectric constant,N
˜ γ is axial force applied to the resonators, M is the proof mass, Lg is the
distance from the center of the proof mass to the anchor, d is the distance of resonator from elastic pin.
When an acceleration γ is applied along the vertical axis, the proof masses can rotate an angle θ due to the leverage
mechanism, then the amplitude of the axial force exerted on each resonator can be approximated as [31].

M Lg
˜ γ = kR |dθ | ≈
N |γ | (4)
d
In order to facilitate the calculation, the following dimensionless variables are used.

˜1
w x˜ EI ˜2
w gc
w1 = , x = , t = t˜ , w2 = ,R = (5)
ga l ρ Al 4 ga ga

Due to Vdc1 >>Vac1 , (Vdc1 +Vac1 cos(t))2 ≈Vdc1 2 +2Vdc1 Vac1 cos(t), substituting Eq. (5) into Eq. (1), we can obtain
⎧  
⎪  1 2
⎪
⎪ ∂ 4 w1 ∂ 2 w1 ∂ w1 ∂ w1 ∂ 2 w1
⎪
⎪ + +c − Nγ + α1 dx
⎪
⎪ ∂ x 4 ∂ t 2 ∂ t 0 ∂ x ∂ x2
⎪
⎪
⎪
⎪ α H (x )Vdc1 2
α2Vc 2
2V Vac1 cos(t )
⎪
⎨= 2 − + α2 H (x ) dc1
( 1 − w1 ) 2
(R + w1− w2 ) 2
(1 − w 1 )
2
(6)
⎪  1 2
⎪
⎪ ∂ w2 ∂ w2
4 2
∂ w2 ∂ w2 ∂ w2
2
⎪
⎪ + +c − −Nγ + α1 dx
⎪
⎪ ∂ x 4 ∂ t 2 ∂ t 0 ∂ x ∂ x2
⎪
⎪
⎪
⎪ α2Vc 2 α2 H (x )Vdc2 2
⎪
⎩= −
( R + w1 − w2 ) 2
( 1 + w 2 )2

c˜l 4 ˜ γ l2
N
 g 2 6ε0 l 4 ρ Al1 4
, α1 = 6 , α2 = ,τ = ˜τ
a
c= , Nγ = , =  (7)
EIτ EI h E h3 g3a EI

4
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Fig. 2. Evolution of the dimensionless static position resonator 1 versus Vdc 1 ; (a) coupling voltage Vc ; (b) acceleration γ ; (c) gap ga .

The Galerkin method is applied to discretize the nonlinear equations to eliminate spatial dependence, and the reduced
order model has been obtained in Appendix A.1.
⎧
⎪
⎪ 
Nm
⎪
⎨ 1
w ( x, t ) = w s,1 ( x ) + q1,i (t )φ1,i (x )
i=1
(8)
⎪
⎪ 
Nm
⎪
⎩w 2 ( x, t ) = w s, 2 ( x ) + q2,i (t )φ2,i (x )
i=1

2.2. Static deformation

To obtain the safe driving voltage range for the proposed accelerometer, it is necessary to calculate the static pull-in
voltage of the sensing part. We remove the time dependent terms in Eq. (6) leading to the static deformation equations of
the coupled resonators. After discretizing Eq. (9) by the Galerkin method, the ANM is used to determine the pull-in voltages.
⎧  
⎪  1 2
⎪
⎪ d 4
w ∂ w ∂ 2 ws,1
− Nγ + α1
s, 1 s, 1
⎪
⎪ dx
⎪
⎪ dx 4
0 ∂x ∂ x2
⎪
⎪
⎪
⎪ α2Vc2 α2Vdc1 2 H (x )
⎪
⎨= − +
(R + w s,1 − ws,2 )
2
(1 − ws,1 )2  (9)
⎪  2
⎪ d4 ws,2
⎪
1
∂ ws,2 ∂ 2 ws,2
⎪
⎪ − −N γ + α dx
⎪
⎪ d x 4 1
0 ∂ x ∂ x2
⎪
⎪
⎪
⎪ α2Vc 2
α2 H (x )Vdc2 2
⎪
⎩= −
(R + ws,1 − ws,2 ) 2
(1 + ws,2 )2
As shown in Fig. 2, the method of controlling variables is used to explore the influence of different parameters on the
static pull-in voltage. Here, the distances between the resonator and the electrode are equal, and the voltage of Vdc 1 is set
to be equal to Vdc 2 . Fig. 2 (a) shows the effect of different coupling voltages on the static pull-in voltage. It can be seen that
the coupling strength has almost no impact on the static deformation. In the case of a constant coupling voltage, magnitude
of the external acceleration also has little effect on the pull-in position as shown in Fig. 2 (b). When reducing the driving
gap ga while keeping other conditions unchanged, the pull-in voltage is significantly decreased as shown in Fig. 2(c). It can
be seen that the driving gap is the most important factor affecting the pull-in of the resonator. If the driving gap ga is
determined, the pull-in voltage is almost determined. When the selected driving voltage is far from the pull-in voltage, it
can be ensured that the device will not collapse.

2.3. Eigenvalue analysis

Through eigenvalue analysis, the relationship between frequency and electrostatic force can be obtained [32], and it is
convenient to find the operating balance point of the device. To obtain the linear undamped eigenvalue equation, Eq. (8) is
substituted into Eq. (6), then the damping and higher-order nonlinear terms are removed resulting in a system of cou-
pled linear equations describing the undamped free vibrations of the resonators. The detailed process can be found in

5
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Fig. 3. The natural frequency variation with the Vdc 1 under different initial voltage of Vdc 2 .

appendix A.3. The variation of the natural frequencies with Vdc 1 under different Vdc 2 is shown in Fig. 3. We found that
the value of Vdc 1 at the mode veering point is equal to Vdc 2 regardless of the initial value of Vdc 2 . Thus, it is necessary to
make the voltages of Vdc 1 and Vdc 2 equal when the device is in the balanced state. Here, the coupling voltage Vc =215V, the
other parameters are listed in Table 1.

3. Perturbation analysis

In this section, a second-order MMS is adopted to explore the response under different acceleration disturbances. Since
the first mode is the dominant mode, only one mode is considered (Nm =1) and the other modes are ignored in the Galerkin
discretization. For convenience, in Eq. (8), we replace q1,1 , q2,1 by q1 , q2 , respectively. The method of multiple scales is then
used on the resulting equation with an appropriate scaling for the quadratic and cubic nonlinearities. A low bookkeeping
parameter ε is introduced to distinguish the effect of each term on the equations, and each term of the equation is mul-
tiplied by a small parameter ε of different orders according to its importance in the system dynamics. The equations of
motion can be obtained as follow.
⎧
⎨q̈1 +
⎪ ε 2 cq˙ 1 + κ11
2
q1 + ε κ12 q21 + ε 2 κ13 q31 − ε 2 κc q1
+ε κc q2 + ε f1 cos (t ) = 0
2 2
(10)
⎪q̈2 + ε 2 cq˙ 2 + κ21
⎩
2
q2 + ε κ22 q22 + ε 2 κ23 q32 + ε 2 κc q1
−ε 2 κc q2 = 0

Where the parameters are given in Appendix A.1, and different scales times Tn are introduced as

Tn = ε n t (n = 0, 1, 2 ) (11)

To describe the motion near resonance, detuning parameters σ 1 , σ 2 are defined as follows

 = κ11 + ε 2 σ1
(12)
κ11 = κ21 + ε 2 σ2
The solution of Eq. (10) is assumed by a second-order time scales form as follows

q1 = q10 (T0 , T1 , T2 ) + ε q11 (T0 , T1 , T2 ) + ε 2 q12 (T0 , T1 , T2 )

(13)
q2 = q20 (T0 , T1 , T2 ) + ε q21 (T0 , T1 , T2 ) + ε 2 q22 (T0 , T1 , T2 )

Substituting Eq. (13) into Eq. (10), and according to the order of ε , we obtain the following equations:

ε0 : D20 q10 + κ11

2
q10 = 0
(14)
D20 q20 + κ21
2
q20 = 0

ε1 : κ11
2
q11 + D0 (D0 q11 + D1 q10 ) + D1 D0 q10 = −κ12 q210
(15)
κ21
2
q21 + D0 (D0 q21 + D1 q20 ) + D1 D0 q20 = −κ22 q220

6
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

ε2 : D0 (D0 q12 + D1 q11 + D2 q10 ) + D1 (D0 q11 + D1 q10 )

+D2 D0 q10 + κ11
2
q12 = −cD0 q10 − (−κc q10 + κc q20 )
−κ13 q10 − κ12 q10 q11 − f1 cos (κ11 T0 + σ1 T2 )
3
(16)
D0 (D0 q22 + D1 q21 + D2 q20 ) + D1 (D0 q21 + D1 q20 )
+D2 D0 q20 + κ21
2
q22 = −cD0 q20 + (−κc q10 + κc q20 )
−κ23 q20 − κ22 q20 q21
3

The solutions of Eq. (14) is given as follows

q10 = X1 exp (iκ11 T0 ) + X1 exp (−iκ11 T0 )

(17)
q20 = X2 exp (iκ21 T0 ) + X2 exp (−iκ21 T0 )

Where X1 and X2 are complex functions (X̄1 , X̄2 are their complex conjugate), and by utilizing the following polar form to
express

X1 = A1 eiβ1 , X2 = A2 eiβ2 (18)

Where A1 , A2 represent the amplitudes of the two resonators, substituting Eq. (18) into Eq. (15) and Eq. (16), by separating
the imaginary and real parts, the following relationships can be obtained.
⎧
⎪ f1 sin ϕ1 = −2(κ11 cA1 + A2 κc sin (ϕ2 ) )
⎪
⎪  2   2 
⎪
⎪ 3κ11 κc + 6κ11 3
σ1 A1 + 10κ12 − 9κ11
2
κ13 A1 3
⎪
⎪ 2
⎪
⎪ −3κ112
A2 κc cos ϕ2
⎪
⎪ 1
f cos ϕ =
⎨ 1
3κ112

κ21 cA2 (19)

⎪sin ϕ2 =
⎪
⎪ A1 κc
⎪
⎪
⎪
⎪ 10κ22
2
A2 2 − 9κ23 κ21 2
A2 2 + 3κ21
2
κc
⎪
⎪ A2
κ σ κ σ
⎪
⎪cos ϕ2 = +6 3
1 + 6 3
2
⎩ 21 21
3κ212 A κ
1 c

By simplifying the Eq. (19), we can obtain the equations of the amplitude frequency response
 2
A2 2 10A2 2 κ22
2
− 9A2 2 κ23 κ21
2
+ 3κ21
2
κc + 6κ21
3
σ1 + 6κ21
3
σ2 A2 2 κ21
2 2
c
+ =1 (20)
9A1 2
κ21
4 κ2
c A1 2 κc2

   2
A2 2 κ11
2
10A2 2 κ22
2
− 9A2 2 κ21
2
κ23 + 3κ21
2
κc + 6κ21
3
( σ1 + σ2 )
4 10A1 3
κ12
2
− 9A1 κ11
2
κ13 + 3A1 κ11
3 2
κc + 6A1 κ11
3
σ1 −
A1 κ21
2

9 f12 κ11
4

 2
A2 2 κ21 c
4 A1 κ11 c + A1
+ =1 (21)
f12
Using Eq. (20) and Eq. (21), a six-order implicit expression for the unknown amplitude A2 can be derived and transformed
into a third-order equation via the change of variableU = A22 , which enables the computation of A2. Then, A1 can be derived
using Eq. (20),where the amplitude of q1 and q2 are respectively equal to 2A1 and 2A2 .
To verify the results of MMS, HBM coupled with ANM is used to compute a reference solution. ANM+HBM is a very
powerful numerical method that has been widely used to solve nonlinear problems [33]. It is not introduced in detail
herein but can be viewed in [34]. When the quality factor Q=10 0 0 0, the coupling voltage Vc is equal to 215V, the DC
voltages Vdc 1 =Vdc 2 =35V and the Vac 1 =5mV, the amplitude-frequency response curves of the two methods in balanced state
are shown in Fig. 4 (a), and the results of two methods are in good agreement, which proves that the solution obtained
by MMS can accurately describe the dynamic characteristics of the device. When the quality factor Q is increased up to
20 0 0 0 and other parameters remain unchanged, as shown in Fig. 4(b), it can be seen that these two methods are still very
consistent even when the sensor operates in the nonlinear vibration regime.

4. Numerical simulations

The previous section has proved that the method of MMS can accurately analyze the dynamic characteristics of the de-
vice. Thus, it will be used to determine the sensitivity of the accelerometer. Due to the mechanical nonlinearities dominating
the resonator dynamics compared to electrostatic nonlinearities, the frequency responses exhibit a hardening behavior with
strong amplitude dependence in the balanced state. The amplitude frequency response under different actuation Vac 1 is

7
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Fig. 4. Amplitude frequency responses of the two resonators with Vc =215V, Vdc 1 =Vdc 2 =35V, Vac 1 =5mV, (a) Q=10 0 0 0; (b) Q=20 0 0 0.

Fig. 5. The responses of the two resonators under different actuation voltage Vac1 .

shown in Fig. 5, where the coupling voltage Vc =215V, the DC voltage Vdc 1 =Vdc 2 =35V, and the quality factor is Q=20 0 0 0. As
the driving voltage Vac 1 increases, the amplitude frequency response curve gradually shows hardening behavior. When the
voltage Vac 1 is equal to 3mV, the amplitude curve of the two resonators shows an infinite slope without bi-stability. This
amplitude point is called the critical amplitude in balanced state [35, 36], it is the amplitude transition point for the vibra-
tion of the device from linear to nonlinear behavior. Since the device can easily exceed the critical amplitude and reach the
nonlinear region under the disturbance of external acceleration, it is necessary to analyze the sensitivity effect for nonlinear
vibrations.
To investigate the effect of different dynamic behaviors on acceleration sensing, the simulation is carried out to explore
the dynamic behavior of the system under the same acceleration disturbance for the linear and the nonlinear initial balanced
vibration state, respectively. When the voltage of Vac 1 is equal to 2mV, the amplitude frequency curve of the two resonators
in the balanced state is shown in Fig. 6(a), and the DC voltage Vdc 1 =Vdc 2 =35V is much smaller than the pull-in voltage
which is around 185V. The device vibration exhibits linear behavior under this excitation voltage, at this time, the vibration
becomes localized under the acceleration disturbance of 0.5g, as shown in Fig. 6(b). On the first mode, the amplitude ratio of
the two resonators w2 /w1 , which is initially equal to 1, increases up to 1.86. On the second mode, the amplitude ratio of the
two resonators w1 /w2 rises from 1 up to 2.11. These amplitude ratios for the two modes are calculated using the rectangular
and circular points depicted on Fig. 6(b). In this case, the amplitude ratio of mode 2 is higher than mode 1. Subsequently,
we only increase the AC voltage for simulation while ensuring that the other parameters remain constant. For Vac 1 =4mV,
the balanced state under nonlinear behavior is shown in Fig. 6(c). When the same acceleration disturbance is applied to the
accelerometer, the vibration on the second mode is more localized than that in the linear case as shown in Fig. 6(d). The
amplitude ratio of w1 /w2 is equal to 2.76. On the first mode, the amplitude ratio w2 /w1 is equal to 1.79, which is reduced
compared to the linear case. It is worth noting that one of the resonators shows nonlinear vibration behavior. Since the two
resonators are electrostatically coupled, there is a bifurcation topology transfer between the two coupled beams when the
sensor is driven beyond its critical amplitude [37]. Consequently, the other resonator displays a hardening behavior even if
its vibration is below the critical amplitude in the balanced state.

8
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Fig. 6. Amplitude frequency response of the device before and after a 0.5g acceleration disturbance; (a) and (b) Vac1 =2mV, Vdc1 =Vdc2 =35V, Vc =215V; (c)
and (d) Vac1 =4mV, Vdc1 =Vdc2 =35V, Vc =215V.

Fig. 7. Acceleration disturbance under linear behavior; (a) the response of resonator 1, (b) difference of RSAR between two modes, (c) the response of
resonator 2, (d) difference of RSF between two modes.

4.1. Linear case

When the driving voltage is small, the vibration amplitude of the two resonators is less than the critical amplitude,
and they will operate in the linear dynamic range. Fig. 7 shows the amplitude-frequency response curve and the change of
the scale factor under the disturbance of external acceleration, where the DC voltage Vdc 1 =Vdc 2 =35V, the coupling voltage
Vc =215V and the driving voltage Vac 1 =1mV. From Fig. 7(a) and (c) show that the amplitude of the two resonators is lower
than the critical amplitude in the balanced state. When the acceleration disturbance varies from -1g to 1g, the vibration
energy restrained in the first mode is gradually transferred to the second mode. In order to analyze the accelerometer
scale factor for two different approaches under a uniform scale, the sensitivity output methods of the difference of relative
shift of amplitude ratio (RSAR) and frequency between two modes are compared in Fig. 7(b) and (d). When the device is
operating in linear behavior, the difference of RSAR between two modes is linear in the range of -1 to 1. The difference
of RSAR between two modes is defined as (w1,2 /w2,2 -w1,2 0 /w2,2 0 )/ (w1,2 0 /w2,2 0 )-(w1,1 /w2,1 -w1,1 0 /w2,1 0 )/ (w1,1 0 /w2,1 0 ), and the
difference of relative shift of frequency (RSF) between two modes is defined as (ω1,1 -ω1,1 0 )/ (ω1,1 0 )- (ω1,2 -ω1,2 0 )/ (ω1,2 0 ),
where wi,j represents the jth amplitude of the ith resonator, ωi,j represents the jth mode frequency of the ith resonator, the

9
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Fig. 8. Acceleration disturbance under nonlinear behaviors while Vc =215V; (a) the response of resonator 1, (b) difference of RSAR between two modes, (c)
the response of resonator 2, (d) difference of RSF output between two modes.

superscript zero represents the amplitude and frequency of two modes before acceleration disturbance. It can be found that
using the difference of RSAR between two modes as the output can increase the sensitivity by four orders of magnitude.

4.2. Nonlinear case

When the actuation voltage Vac 1 is increased to 4mV while maintaining the same other parameters as the linear case,
the vibration amplitude of the device exceeds the critical amplitude in the balanced state. When the device is operating
in nonlinear behavior, sensitivity output of the two modes will no longer be equal, as shown in Fig. 8 (b) and (d). The
difference in RSAR between two modes is improved by 1.3 times compared with the linear behavior. The variation of the
frequency responses of the two resonators driven in the nonlinear regime is shown in Fig. 8(a) and (c), for acceleration
disturbance ranging from -1g to 1g. It is worth noting that the vibration amplitude of the nonlinear case is nearly 4 times
higher compared to the linear case. Since mode localization sensitivity detection is based on the amplitude measurement,
the smallest detection amplitude directly determines the smallest detectable acceleration. Therefore, utilizing nonlinear high
amplitude can improve the acceleration resolution of the device.

4.3. Sensitivity analysis

To investigate the device sensitivity, we carry out simulations for an acceleration ranging from -1g up to 1g for several
actuation voltages Vac 1 , where Vdc 1 =Vdc 2 =35V, Vc =215V, and Q=20 0 0 0. Fig. 9 shows the RSAR of mode 1 and mode 2 and

Fig. 9. RSAR and difference of RSAR between two modes as a function of the acceleration disturbance.
10
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Fig. 10. The simulations of the responses of the device with different coupling voltage; (a) and (d) are the amplitude responses of two resonators when
Vc =150V; (b) and (e) are the amplitude responses of two resonators when Vc =200V; (c) and (f) are the amplitude responses of two resonators when
Vc =250V; (g), (h), (l) is the difference of RSAR between two modes when Vc =150V, 200V, 250V, respectively.

the difference of RSAR between two modes as a function of the acceleration disturbance under different driving voltages.
Even though results of mode 1 and mode 2 are completely different for the change of the excitation voltage Vac 1 , both modes
are sensitive to acceleration. When the acceleration in the range of -1g to 0g, the amplitude of resonator 1 is greater than the
amplitude of resonator 2 in mode 1 and contrary to the aforementioned result in mode 2. Thus, below the resonator critical
amplitude, the difference of RSAR between the two modes is linear with respect to the acceleration as shown in Fig. 9.
As the driving voltage Vac 1 increases, the RSAR gradually decreases for mode 1. However, when the applied acceleration
is positive, the RSAR is enhanced as the driving voltage Vac 1 increases in the second mode. Since the RSAR of mode 2 is
higher and increases with respect to the driving voltage, not only the sensitivity can be improved but also the resolution
can be significantly enhanced when the resonators vibrate in the nonlinear regime. Fig. 9 shows that the difference of RSAR

Fig. 11. Mode-aliasing effect of different coupling voltage; (a) frequency loci veering with acceleration perturbations; (b), (c), (d) are the amplitude fre-
quency response curves when Vc =100V, 150V, 200V respectively.

11
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

between two modes can be used as the sensitivity output to replace the relative amplitude ratio, thus only one output can
be used in the entire acceleration dynamic range.
In addition to the driving voltage, the coupling voltage is the other factor affecting the sensitivity output. The frequency
responses and the difference of RSAR between two modes under different coupling voltages are compared in Fig. 10, where
the actuation voltage Vac 1 =5mV, Vdc 1 =Vdc 2 =35V, and quality factor Q=10 0 0 0. We also notice that for the same driving
voltage, the amplitude under different coupling voltages is nearly unaffected as shown in Fig. 10 (a)-(l). However, it should
be noted that when the coupling voltage is reduced, the frequency difference between the two modes is also reduced,
especially for resonator 2. Moreover, the results in Fig. 10 (g), (h) and (l) show that using the difference of RSAR between
two modes as the output, while reducing the coupling voltage can significantly increase the sensitivity. Thus, we noticed
that compared with a mechanical coupling, the electrostatic coupling has the advantage of adjustable sensitivity.
The preceding results demonstrate that decreasing the coupling voltage can improve the sensitivity, however, when the
coupling voltage decreases, the frequency difference between the two modes decreases and may cause mode aliasing. By
varying the coupling voltage, we obtain the frequency loci veering in Fig. 11 (a). With the decrease of the coupling voltage,
the natural frequencies of the two modes gradually approach, especially the frequency of mode 1 increases significantly.
From the amplitude-frequency response curve in Fig. 11 (b)-(d), when Vc is equal to 100V in the balanced state, the two
modes are completely indistinguishable and resonator 2 generates mode aliasing. Thus, when Vc decreases to 100V or less,
the effective sensitivity range of the accelerometer does not continue to increase with decreasing Vc . Unlike mechanical
coupling for which the strength is fixed for a given fabricated device, the electrostatic coupling can be adjusted to achieve
the theoretical optimal value of sensitivity below mode aliasing.

5. Conclusions

In this paper, a dual-mass MEMS accelerometer based on the mode localization phenomenon has been designed and
modeled. Two separate leveraging mechanisms were introduced to amplify the inertial forces applied axially to two electro-
statically coupled identical resonators, thus generating mode localization. The main conclusions are listed as follows: (i) A
mathematical model considering the electrostatic and mechanical nonlinearities was established, and the dynamic response
was calculated by using MMS. The solving accuracy was validated by the results consistency between the MMS and the
ANM+HBM method. (ii) The accelerometer specifications operating in the linear and nonlinear ranges were investigated in-
volving sensitivity, resonant frequency, and response amplitude. The results showed that the linear sensitivity in terms of
the difference of the relative shift of amplitude ratio (RSAR) is significantly enhanced with up to 4 orders of magnitude
compared with the difference of the relative shift of frequency (RSF). (iii) When the driving voltage increases, the resonators
vibrate in the nonlinear regime. The nonlinear sensitivity and resolution are improved for mode 2. (iv) The effect of the
coupling voltages on the device performances was explored, and the mode aliasing is clearly shown through the amplitude-
frequency response curve enabling the theoretical identification of the optimal coupling voltage.
Future work will be proceeded including the manufacturing of the proposed sensor, in which the elastic pin design
is considered to be the critical component since it has a significant impact on the leverage mechanism and the device
performances.

Credit author statement

Ming Lyu: Wrote the original draft, Conceptualization, Methodology, Jian Zhao: Conceptualization, Methodology, Writing
-review & editing, Najib Kacem: Conceptualization, Writing - review & editing, Bin Tang: Validation, Pengbo Liu: Validation,
Jiahao Song: Validation, Formal analysis, Heng Zhong: Validation, formal analysis, Yu Huang: Writing-review and editing. All
authors have read and agreed to the published version of the manuscript.

Declaration of Competing Interest

No conflict of interest exists in the submission of this manuscript, and the manuscript is approved by all authors for
publication. I would like to declare that the work was original research that has not been published previously, and not
under considered for publication elsewhere, in whole or in part. All the authors have approved the manuscript that is
enclosed.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (Grant No. U1930206), Natural Science
Foundation Project of Liaoning Province (2019-KF-02-03) and project of State Key Laboratory of Structural Analysis for In-
dustrial Equipment, China (Grant No. GZ19202). This project was also performed in cooperation with EUR EIPHI program,
Europe (Contract No. ANR 17-EURE-0 0 02).

12
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

Appendix A.

A.1. Reduced order model

In order to simplify the calculation without affecting the accuracy, we expand the coupling and driving electrodynamic
terms up to first and third orders, respectively
 

Nm
1 
Nm 
Nm
1
q̈i, j + ci, j q˙ i, j + λ
4
i, j qi, j − 2α1 qi,m ∫ φ 
i,m w s,i dx + α1 qi,m qi,p ∫ φ i,m φ i,p dx
0 0
 m=1
 m=1 p=1

1 
Nm
1
∫ w s,i φi, j dx + qi, j ∫ φ  i, j φi, j dx
0 0
j=1
1 
Nm
1 
Nm
1
+ −(−1 )i+1 Nγ − α1 ∫ ws,i dx qi, j ∫ φ  i, j φi, j dx = 2α2Vdc,i
2 2
qi,k ∫ H φi,k φi, j dx
0 0 0
j=1 k=1

Nm 
Nm
1 
Nm 
Nm 
Nm
1
−3(−1 )i α2Vdc,i
2
q2,k q2,l ∫ H φi, j φi,k φi,l dx + 4α2Vdc,i
2
qi,k qi,l qi,h ∫ H φi, j φi,k φi,l φi,h dx (22)
0 0
k=1 l=1 k=1 l=1 h=1

Nm
1 
Nm 
Nm
1
−6(−1 )i α2Vdc,i
2
qi,k ∫ H φi, j φi,k ws,i dx + 12α2Vdc,i
2
qi,k qi,l ∫ H φi, j φi,k φi,l ws,i dx
0 0
k=1 k=1 l=1

Nm
1 
Nm
1 φi, j φ1,k 
Nm
1 φi, j φ2,k
+12α2Vdc,i
2
qi,k ∫ H φi, j φi,k w2s,i dx − 2(−1 )i α2Vc2 q1,k ∫ dx + 2(−1 )i α2Vc2 q2,k ∫ dx
0 0 R3 0 R3
k=1 k=1 k=1
1  
+δ i
α
1 2 2Vdc,1Vac,1 H 1 + 2ws,1 + 3w2s,1 + 4w3s,1 φi, j dx cos (t )
0

where i is the ith microbeam, Kronecker symbol δ 1 1 =1, δ 1 2 =0.

A.2. The integration parameters

κi1 2 = λi, j 4 − (−1 )i+1 Nγ ∫ 1 φ  i,1 φi,1 dx − 2α2Vdc,i 2 ∫ 1 H φi,1 2 dx + 6(−1 )i α2Vdc,i 2 ∫ 1 H φi,1 2 ws,i dx
0 0 0
−12α2Vdc,i 2 ∫ 1 H φi,1 2 ws,i 2 dx − α1 ∫ 1 ws,i 2 dx ∫ 1 φ  i,1 φi,1 dx − 2α1 ∫ 1 w s,i φi,1 dx ∫ 1 φ  i,1 w s,i dx
0 0 0   0 0
κi2 = 3(−1 )i α2Vdc,i 2 ∫ 1 H φi,1 3 dx − α1 ∫ 1 w s,i φi,1 dx ∫ 1 φ  i,1 2 dx
0 0 0
−2α1 ∫ 1 φ  i,1 φi,1 dx ∫ 1 φ  i,1 w s,i dx − 12α2Vdc,i 2 ∫ 1 H φi,1 3 ws,i dx
0 0 0  (23)
κi3 = −4α2Vdc,i 2 ∫ 1 H φi,1 4 dx − α1 ∫ 1 φ  i,1 φi,1 dx ∫ 1 φ  i,1 2 dx, i = 1, 2
0 0 0
2α2Vc 2
κc =
R3   
f1 = −2α2Vdc1Vac1 1
H 1 + 2ws,1 + 3ws,1 2 + 4ws,1 3 φ1,1 dx
0

where i is the ith microbeam.

A.3. linear undamped eigenvalue problem

The coupled linear equations describing the undamped free vibrations of the resonators are given as follows

 2α2Vc 2 q1,1 − q2,1 1
φ1,1 (x )φ2,1 (x )dx
 
∫0 1 w s,1 (x )2 dx q1,1 φ1,1 (x )φ  1,1 (x )dx − 0
q̈1,1 + λ1,1 q1,1 − α1
4 1
(R + ws,1 − ws,2 )3
0

2α2Vdc1 2 q1,1 1
φ1,1 2 (x )H (x )dx
0
− =0 (24)
(1 − ws,1 )3


 2α2Vc 2 q1,1 1
φ2,1 (x )φ1,1 (x )dx − q2,1
 
w φ2,1 (x )φ  2,1 (x )dx + 0
q̈2,1 + λ2,1 q2,1 − α1 ∫0
4 1
s,2 (x ) dx q2,1
2 1
(R + ws,1 − ws,2 )3
0

13
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960


2α2Vdc2 2 q2,1 1
φ2,1 2 (x )H (x )dx
0
− =0 (25)
(1 + ws,2 )3
Eqs. (24)–(25) can be rewritten in a matrix format as
   
q̈1,1 q
+  1,1 = 0 (26)
q̈2,1 q2,1

Where
⎡ 4   ⎤
λ1,1 − α1 ∫0 1 w s,1 (x )2 dx 1 φ1,1 (x )φ  1,1 (x )dx 2α2 Vc 2
 1
φ1,1 (x )φ2,1 (x )dx
⎢  0 0 ⎥
⎢ 2α2 Vdc1 2 1 φ1,1 2 (x )H (x )dx (R+ws,1 −ws,2 )3 ⎥
⎢− (R+w2α2−wVc 2
3 −
0
⎥
=⎢
⎢
s,1 s,2 ) (1−ws,1 )3    ⎥ (27)
⎢ 2α2 Vc

2 1
φ2,1 (x )φ1,1 (x )dx
1 
λ2,1 − α1 ∫0 w s,2 (x ) dx
4 2 1
φ2,1 (x )φ 2,1 (x )dx ⎥

⎥
⎣ 0  0
2α V 2 1 φ 2 (x )H (x )dx
⎦
(R+ws,1 −ws,2 )
3 2 dc2 2,1
− (R+w2α2−w
2
Vc
3 −
0
s,1 s,2 ) (1+ws,2 )3

For a given value of voltage Vdc1 , the static deflections of the two microbeams ws,1 and ws,2 are evaluated. Finally, the
eigenvalues of the matrix  of Eq. (27) are calculated and their square roots correspond to the sensor eigenfrequencies.

References

[1] Effa D, Nieva P, Andrew Z, Lancaster S. Design and modeling of a MEMS accelerometer for a novel Virtual Button user interface. In: 2009 IEEE Toronto
International Conference Science and Technology for Humanity (TIC-STH). IEEE2009; 2009. p. 597–602.
[2] Babatain W, Bhattacharjee S, Hussain AM, Hussain MM. Acceleration sensors: sensing mechanisms, emerging fabrication strategies, materials, and
applications. ACS Appl Electron Mater 2021;3:504–31.
[3] Fang Z, Yin Y, He X, Han F, Liu Y. Temperature-drift characterization of a micromachined resonant accelerometer with a low-noise frequency readout.
Sens Actuat, A 2019;300:111665.
[4] Daeichin M, Miles RN, Towfighian S. Large-stroke capacitive MEMS accelerometer without pull-in. IEEE Sens J 2020:1–10.
[5] Partridge A, Reynolds JK, Chui BW, Chow EM, Fitzgerald AM, Zhang L, et al. High-performance planar piezoresistive accelerometer. J Microelectromech
Syst 20 0 0;9:58–66.
[6] Comi C, Corigliano A, Langfelder G, Longoni A, Tocchio A, Simoni B. A resonant microaccelerometer with high sensitivity operating in an oscillating
circuit. J Microelectromech Syst 2010;19:1140–52.
[7] Ding H, Ma Y, Guan Y, Ju B-F, Xie J. Duplex mode tilt measurements based on a MEMS biaxial resonant accelerometer. Sens Actuators, A
2019;296:222–34.
[8] Su X-PS, Yang HS. Two-stage compliant microleverage mechanism optimization in a resonant accelerometer. Struct Multidisc Optim 2001;22:328–34.
[9] Chen Z, Cai G, Li Y, Chen Q, Wu W. Performances and improvement of coupled dual-microcantilevers in sensitivity. Sens Actuators, A 2019;288:117–24.
[10] Li L, Han J, Zhang Q, Liu C, Guo Z. Nonlinear dynamics and parameter identification of electrostatically coupled resonators. Int J Non Linear Mech
2019;110:104–14.
[11] Marquez S, Alvarez M, Plaza JA, Villanueva LG, Dominguez C, Lechuga LM. Asymmetrically coupled resonators for mass sensing. Appl Phys Lett
2017;111:113101.
[12] Ilyas S, Younis MI. Theoretical and experimental investigation of mode localization in electrostatically and mechanically coupled microbeam resonators.
Int J Non Linear Mech 2020;125:103516.
[13] Wang DF, Chatani K, Ikehara T, Maeda R. Mode localization analysis and characterization in a 5-beam array of coupled nearly identical micromechan-
ical resonators for ultra-sensitive mass detection and analyte identification. Microsyst Technol 2012;18:1923–9.
[14] Ramakrishnan S, Balachandran B. Energy localization and white noise-induced enhancement of response in a micro-scale oscillator array. Nonlinear
Dyn 2010;62:1–16.
[15] Ouakad H, Ilyas S, Younis M. Investigating mode localization at lower and higher order modes in mechanically coupled MEMS resonators. J Comput
Nonlinear Dyn 2019;15:031001.
[16] Rabenimanana T, Walter V, Kacem N, Le Moal P, Bourbon G, Lardiès J. Functionalization of electrostatic nonlinearities to overcome mode aliasing
limitations in the sensitivity of mass microsensors based on energy localization. Appl Phys Lett 2020;117:033502.
[17] Rabenimanana T, Walter V, Kacem N, Le Moal P, Bourbon G, Lardiès J. Mass sensor using mode localization in two weakly coupled MEMS cantilevers
with different lengths: Design and experimental model validation. Sens Actuators, A 2019;295:643–52.
[18] Spletzer M, Raman A, Wu AQ, Xu X, Reifenberger R. Ultrasensitive mass sensing using mode localization in coupled microcantilevers. Appl Phys Lett
2006;88:254102.
[19] Zhang HM, Yuan WZ, Li BY, Hao YC, Kraft M, Chang HL. A novel resonant accelerometer based on mode localization of weakly coupled resonators. In:
2015 Transducers-2015 18th International Conference on Solid-State Sensors, Actuators and Microsystems (TRANSDUCERS). IEEE; 2015. p. 1073–6.
[20] Wang Y, Zhao C, Wang C, Cerica D, Baijot M, Xiao Q, et al. A mass sensor based on 3-DOF mode localized coupled resonator under atmospheric
pressure. Sens Actuators, A 2018;279:254–62.
[21] Zhao C, Zhou X, Pandit M, Sobreviela G, Du S, Zou X, et al. Toward high-resolution inertial sensors employing parametric modulation in coupled
micromechanical resonators. Phys Rev Appl 2019;12:044005.
[22] Zhao C, Wood GS, Xie J, Chang H, Pu SH, Kraft M. Comparative study of different output metrics for a three weakly coupled resonator sensor. In:
Transducers- International Conference on Solid-state Sensors. IEEE; 2015. p. 2196–9.
[23] Lyu M, Zhao J, Kacem N, Liu P, Tang B, Xiong Z, et al. Exploiting nonlinearity to enhance the sensitivity of mode-localized mass sensor based on
electrostatically coupled MEMS resonators. Int J Non Linear Mech 2020;121:103455.
[24] Hajjaj AZ, Jaber N, Ilyas S, Alfosail FK, Younis MI. Linear and nonlinear dynamics of micro and nano-resonators: Review of recent advances. Int J Non
Linear Mech 2019;119:103328.
[25] JY Thiruvenkatanathan P, Lee JE-Y, Seshia AA. Enhancing parametric sensitivity using mode localization in electrically coupled MEMS resonator. In:
15th International Conference on Solid-State Sensors, Actuators and Microsystems Transducers2009; 2009. p. 2350–3.
[26] Zhao C, Wood GS, Pu SH, Kraft M. A mode-localized MEMS electrical potential sensor based on three electrically coupled resonators. J Sens Sens Syst
2017;6:1–8.
[27] Peng B, Hu K-M, Shao L, Yan H, Li L, Wei X, et al. A sensitivity tunable accelerometer based on series-parallel electromechanically coupled resonators
using mode localization. J Microelectromech Syst 2019;29:3–13.

14
M. Lyu, J. Zhao, N. Kacem et al. Commun Nonlinear Sci Numer Simulat 103 (2021) 105960

[28] Thiruvenkatanathan P, Yan J, Seshia AA. Common mode rejection in electrically coupled MEMS resonators utilizing mode localization for sensor ap-
plications. In: 2009 Joint conference of the 22nd European Frequency and Time Forum and the IEEE International Frequency Control Symposium,
IEEE20 09; 20 09. p. 358–63.
[29] Zhang HM, Zhong JM, Yuan WZ, Yang J, Chang HL. Ambient pressure drift rejection of mode-localized resonant sensors. In: 2017 IEEE 30th International
Conference on Micro Electro Mechanical Systems (MEMS). IEEE2017; 2017. p. 1095–8.
[30] Zhang Y, Kim WS. Highly sensitive flexible printed accelerometer system for monitoring vital signs. Soft Rob 2014;1:132–5.
[31] Kacem N. Nonlinear dynamics of M&NEMS resonant sensors: design strategies for performance enhancement [Ph.D]. Grenoble: INSA-Lyon; 2010.
[32] Abdel-Rahman EM, Younis MI, Nayfeh AH. Characterization of the mechanical behavior of an electrically actuated microbeam. J Micromech Microeng
2002;12:759–66.
[33] Kacem N, Baguet S, Hentz S, Dufour R. Computational and quasi-analytical models for non-linear vibrations of resonant MEMS and NEMS sensors. Int
J Non Linear Mech 2011;46:532–42.
[34] Cochelin B, Vergez C. A high order purely frequency-based harmonic balance formulation for continuation of periodic solutions. J Sound Vib
2009;324:243–62.
[35] Kacem N, Hentz S, Pinto D, Reig B, Nguyen V. Nonlinear dynamics of nanomechanical beam resonators: improving the performance of NEMS-based
sensors. Nanotechnology 2009;20:275501.
[36] Kacem N, Baguet S, Hentz S, Dufour R. Nonlinear phenomena in nanomechanical resonators: mechanical behaviors and physical limitations. Mec Ind
2011;11:521–9.
[37] Bitar D, Kacem N, Bouhaddi N. Investigation of modal interactions and their effects on the nonlinear dynamics of a periodic coupled pendulums chain.
Int J Mech Sci 2017;127:130–41.

15