7372 IEEE SENSORS JOURNAL, VOL. 17, NO.

22, NOVEMBER 15, 2017

Design and Optimization of Piezoelectric MEMS

Vibration Energy Harvesters Based
on Genetic Algorithm
Seyedfakhreddin Nabavi, Student Member, IEEE, and Lihong Zhang, Member, IEEE

Abstract— Low-power electronic applications are normally in the surroundings such as solar, thermal, etc., energy har-
powered by batteries, which have to deal with stringent lifetime vesting from vibrations by MEMS energy harvesters has been
and size constraints. To enhance operational autonomy, energy gaining increasingly more attention due to high power density,
harvesting from ambient vibration by microelectromechanical
systems (MEMS) has been identified as a vivid solution to ease of energy trapping, and ubiquity in the environment [1].
this universal problem. This paper proposes an automated The vibrations may be harvested from wind flow [2], water
design and optimization methodology with minimum human stream [3], blood pressure [4], or road traffic [5]. To con-
efforts for MEMS-based piezoelectric energy harvesters. The vert vibrations to electricity, electromagnetic-, electrostatic-,
analytic equations for estimating the harvested voltage by the bistable-, and piezoelectric-based techniques have been
unimorph piezoelectric energy harvesters are presented with their
accuracy validated by using the finite element method (FEM) proposed as feasible solutions to MEMS or portable vibration
simulation and prototype measurement. Thanks to their high energy harvesters. Since the functionality of the electromag-
accuracy, we use these analytic equations as fitness functions of netic generators is based on magnetic field strength, a big
genetic algorithm (GA), an evolutionary computation method for permanent magnet, which may unfortunately degrade device
optimization problems by mimicking biological evolution. Our portability degree [6], is normally required to achieve high
experimental results show that the GA is capable of optimizing
multiple physical parameters of piezoelectric energy harvesters energy conversion efficiency. The electrostatic-based technique
to considerably enhance the output voltage. This harvesting effi- can offer a highly compact energy harvester. However, for
ciency improvement is also desirably coupled with physical size polarization of capacitor’s plates, utilizing a high external bias
reduction as preferred for the MEMS design process. To demon- voltage [7] or electret material [8] seems to be necessary,
strate capability of the proposed optimization method, we have which virtually prevents them from forming standalone energy
also included a commercial optimization product (i.e., COMSOL
optimization module) in our comparison study. The experiments harvesters. Moreover, by using the bistable-based technique
show that our proposed GA-based optimization methodology normally for energy scavenging systems with macro dimen-
offers higher effectiveness in the magnitude improvement of sions [9], special packaging has to be sought to leverage both
harvested voltage along with less runtime compared with the piezoelectric effect and electromagnetic field strength simulta-
other optimization approaches. Furthermore, the effects of neously. In contrast, the piezoelectric-based technique, thanks
geometry optimization on mechanical and electrical properties
(e.g., resonant frequency, stiffness, and internal impedance) are to the piezoelectric material deposition compatibility with
also studied and an effective solution to producing maximum micromachining process and the independence from external
power from unimorph piezoelectric harvesters is proposed. voltage supply, has attracted more attention in building up
Index Terms— MEMS piezoelectric, energy harvesters, opti- MEMS-based vibration energy harvesters [10], [11].
mization, genetic algorithm, mechanical and electrical properties. In the recent decade, several approaches to improve energy
conversion efficiency for the MEMS piezoelectric energy
I. I NTRODUCTION harvesters have been presented. Gao et al. [12] proposed a new

E NERGY harvesting from the environment is an elegant

solution to sustainable power supplies for low-power
consumer electronic applications. Among available resources
piezoelectric energy harvester that is attached with an external
cylindrical object made of photo paper to adjust the resonant
frequency and magnitude of the output voltage. This study
clearly demonstrates that different sizes of extension objects
Manuscript received September 1, 2017; accepted September 15, 2017. Date
of publication September 26, 2017; date of current version October 24, 2017. (e.g., the cylindrical one) can impact on the characteristics
This work was supported in part by the Natural Sciences and Engineering of harvester output. Another technique to improve efficiency
Research Council of Canada, in part by the Canada Foundation for Innovation, of the MEMS piezoelectric energy harvester, which was pre-
in part by the Research and Development Corporation of Newfoundland
and Labrador through the Industrial Research and Innovation Fund and Arc- sented by Jung et al. [13], is to modify the location of piezo-
ticTECH R&D Award, in part by the Memorial University of Newfoundland, electric film along the beam surface by selecting an area closer
and in part by the CMC Microsystems. The associate editor coordinating the to the anchor side. In addition, Jia and Seshia [14] proposed to
review of this paper and approving it for publication was Prof. Weileun Fang.
(Corresponding author: Seyedfakhreddin Nabavi.) enhance the efficiency of the MEMS piezoelectric harvesters
The authors are with the Department of Electrical and Computer Engineer- by tuning proof mass size as a practical approach. It was
ing, Faculty of Engineering and Applied Science, Memorial University of found that an end mass occupying about 60%–70% of the total
Newfoundland, St. John’s, NL A1B3X5, Canada (e-mail: [email protected];
[email protected]). cantilever length is optimal for linear response. Furthermore,
Digital Object Identifier 10.1109/JSEN.2017.2756921 Ibrahim et al. [15] presented a new structure with a magnetic

1558-1748 © 2017 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.
NABAVI AND ZHANG: DESIGN AND OPTIMIZATION OF PIEZOELECTRIC MEMS VIBRATION ENERGY HARVESTERS BASED ON GA 7373

oscillator around the piezoelectric cantilever tip in order to The contributions of this work include the following:
enhance the harvested voltage magnitude. The comparison • We validate the accuracy of the analytic model for
between the proposed device and the conventional piezoelec- MEMS-based unimorph piezoelectric energy harvesters
tric energy harvesters shows that the harvested voltage was with FEM simulation and prototype measurement;
enlarged by a factor of 2.8 at the magnetic oscillator speed • We propose a GA-based automated design optimiza-
of 2 m/s (i.e., meters/second). tion methodology with minimum human efforts for
However, all the techniques above are solely based on the MEMS-based unimorph piezoelectric energy harvester
designers’ experience and skills in making a wise decision, design;
which means they have to spend a considerable amount • We also investigate and analyze various effects of phys-
of time to gain optimum design. To address this difficulty, ical parameter changes on the mechanical and electrical
Sunithamani et al. [16] proposed an optimization method by properties of the MEMS energy harvesters.
using variant-mesh analysis, which aims to run numerous sim- The paper is organized as follows. The analytic equations
ulations to investigate the effect of harvester geometry changes for estimating the generated voltage from the unimorph piezo-
on its performance. Furthermore, in a recent study [17], electric energy harvesters is presented in Section II. Section III
analyzing geometric parameters by incrementing the harvester discusses our GA-based optimization methodology with a
aspects to produce a correlation matrix was presented as focus on how to enlarge the harvested voltage by changing the
a solution to MEMS piezoelectric harvester optimization. MEMS piezoelectric dimensions. The experimental results are
However, requiring a large amount of time to complete finite reported and analyzed in Section IV. Eventually the conclusion
element method (FEM) simulations as well as acute human is made in Section V.
observation makes these methods less efficient for optimizing a
number of physical parameters of MEMS piezoelectric energy II. A NALYTIC M ODELLING OF U NIMORPH
harvesters. Thus, a proper design automation technique is P IEZOELECTRIC E NERGY H ARVESTERS
essential for the MEMS piezoelectric harvester designers to The piezoelectric materials, which are used to fabricate
be liberated from the conventional laborious trial-and-error sensors and actuators [23], have crystal structure. By applying
effort. Therefore, in this paper we are motivated to utilize mechanical force on the surface, the structure changes to
some design automation techniques to address this pressing generate electrical voltage. This unique property makes them
challenge during the design process of the MEMS piezoelec- useful for converting mechanical vibration to electrical power,
tric energy harvesters. and vice versa for mechanical deformation if an external
In the literature various approaches for optimization electrical power is applied. Since kinetic energy of vibration
of MEMS structures have been presented [18]. provides periodic deformation on the piezoelectric structure,
Tabatabaei et al. [19] proposed an Artificial Immune System an AC electrical voltage can be collected from its surface.
(AIS) method for piezoelectric energy harvester shape In order to collect electrical voltage from the piezoelectric
optimization. However, this method offers an energy harvester film, two different configurations for electrode deposition
with unusual shape that cannot be readily fabricated (i.e., capacitor-style and interdigitated-style) can be uti-
with regular MEMS fabrication technologies. Moreover, lized [24]. In this study, due to our manufacturing restriction,
the proposed AIS optimization approach is less efficient we can only deposit capacitor–style electrodes in our MEMS
in comparison with the other optimization methods [20]. unimorph piezoelectric energy harvesters. Thus, according to
Kim et al. [21] presented a Computational Experiments (CE) the IEEE standard on piezoelectricity with the special con-
method for design automation, by which a prototyped device sideration of our electrode deposition, the generated voltage
was first tested and then the FEM simulation parameters by piezoelectric materials is proportional to the observed
were corrected upon the obtained experimental results. stress on their structure with the following conversion rela-
After performing this calibration, the objective function for tionship (1) [25]:
maximizing the harvested voltage at low frequency was
utilized in the process of tuning the piezoelectric cantilever V = g31tp σp , (1)
dimensions. Since this proposed method needs prototyping- where V is the generated voltage, g31 is the piezoelectric
based calibration, fabricating energy harvesters two times coefficient with an assumption of dominant stress in the
make it somehow impractical. On the other side, Genetic in-plane direction and zero stress in the out-of-plane direction,
Algorithm (GA), an evolutionary optimization approach, t p is the piezoelectric film thickness, and σ p is the applied
was used to enhance the magnitude of the voltage harvested stress on the piezoelectric material. Since stress can lead to
from traffic roads [22]. However, the proposed objective strain inside the material, their general conversion can be
function in this study is dependent on natural properties of expressed by the following when the object is restricted to
the piezoelectric materials (e.g., piezoelectric coefficient, move only in one direction (i.e., the in-plane direction):
internal capacitance of piezoelectric, etc.), which designers
σp
cannot modify during fabrication process. Thus, considering εp = , (2)
a fitness function with optimizable variables but without a Yp
need of prototyping-calibration is highly demanded. where ε p is the observed strain inside the piezoelectric mate-
To overcome these aforementioned challenges, we propose rial, and Y p denotes the Young’s modulus of the piezoelectric
a new design automation method based on GA in this paper. material. Since the prevalent rectangular-shaped piezoelectric
7374 IEEE SENSORS JOURNAL, VOL. 17, NO. 22, NOVEMBER 15, 2017

Fig. 1. Structural diagram of the unimorph MEMS piezoelectric energy

harvester with deposition of piezoelectric film on the beam surface close to
the anchor side. Fig. 2. Schematic of the Euler-Bernoulli model with an assumption of perfect
bonding between beam and piezoelectric film.

energy harvesters are compatible with the regular MEMS

fabrication process, this paper is focused on such harvesters. piezoelectric film is homogenous, transverse isotropic, elastic,
Typically the piezoelectric vibration energy harvesters can and thin compared to the beam.
be offered in two different configurations: bimorph and uni- By using the method of transformed Section [31], we can
morph. In the bimorph configuration, piezoelectric material transfer the two-layer piezoelectric energy harvester into one
is deposited on two sides of the beam (or called cantilever, unified structure (i.e., beam). The common neutral axis of the
both of which terms are exchangeable throughout this paper) newly formed structure with the reference of the piezoelectric
by bonding approaches (e.g., gluing or hot air pressure) for surface on the top can be computed by (3):
centimeter-scale piezoelectric energy harvesters [26]. Due to n tp Yp tb
Z i f i Ai 2 tp Yb + [tp + 2 ]tb
the limitation of regular MEMS fabrication process in deposi- Zs = i=1 n = Y
, (3)
tion of piezoelectric material on two sides, the unimorph piezo- i=1 fi Ai tp Ypb + tb
electric energy harvesters, which includes the piezoelectric where Z s is the new common neutral axis, Z i is the original
film only on one side, features sound advantages for producing neutral axis for the i t h material, f i is the dimensionless trans-
energy harvesters in the micrometer scale [27]. Therefore, this formation factor based on the ratio of piezoelectric film and
paper is aimed at exhibiting the design automation benefits for beam Young’s moduli, and Ai is the area of the i t h material.
the unimorph piezoelectric energy harvesters. The structural Constant n is the number of the composite materials, which
diagram of the unimorph piezoelectric energy harvester is is equal to 2 in this modeling application. And tb is the beam
displayed in Fig. 1. thickness and Yb is the Young’s modulus of the beam.
In the literature, the analytic models, which can estimate Consequently, the strain on this transformed structure can
the harvested voltage from the piezoelectric energy harvesters, be expressed by (4):
can be categorized into the following three: pin-force model,
M tp
enhanced-pin-force model, and Euler-Bernoulli model. The εp = −   (zs − ), (4)
pin-force model assumes the piezoelectric film is attached to Yp I p + Yb I b 2
the beam as a pin connection. The strain on the beam, which where M is the actuating momentum and the term of (z s − 2p )
t
linearly increases through the beam thickness, is computed indicates of the piezoelectric film. The origin of this coordinate
by the Euler-Bernoulli beam theory, whereas the strain on system in our analysis is assumed to be the neutral axis.
the entire piezoelectric film is considered to be constant. I p and Ib , are the moments of inertia for the piezoelectric
Obviously this model does not consider the bending stiffness film and beam within the coordinate system in Fig. 2. They
of the piezoelectric material. The enhanced-pin-force model can be calculated by (5):
actually improves the pin-force model by taking into account  Zs
the bending stiffness of the piezoelectric material. In this 1   3 
Ip = wz2 dz = w Z3s − Zs − tp .
regard, the strain on the piezoelectric material, which is not Zs −tp 3
constant any longer but linearly increases, is determined by the  Zs −tp
beam deformation. In comparison with the two models above, Ib = wz2 dz
Zs −tb −tp
the Euler-Bernoulli model features the highest modeling accu-
1  3  3 
racy. Since this model assumes the beam and piezoelectric = w tp + tb − Zs + Zs − tp . (5)
film are perfectly bonded together, their deformation should 3
be represented with reference to the new common neutral where w denotes the width of the beam or piezoelectric film.
axis [28]–[30]. Thanks to this advantage, in this paper we By integrating (3) and (5) to (4), we can derive the average
utilize the Euler-Bernoulli model for the design optimization strain on piezoelectric film as (6):
 
purpose. 6Yb tb tb + tp
The Euler-Bernoulli model schematic is depicted in Fig. 2. εp = −    × M,
In this model, the bonding adhesive material has no impact on w Yp2 tp4 + Yb2 tb4 +2Yp tp Yb tb 2tp2 +3tptb +2tb2
the beam stiffness or mass. Moreover, it is assumed that the (6)
NABAVI AND ZHANG: DESIGN AND OPTIMIZATION OF PIEZOELECTRIC MEMS VIBRATION ENERGY HARVESTERS BASED ON GA 7375

Eventually by using (2), we can also compute the average

stress on the piezoelectric material. Then by applying it
into (1), the harvested voltage from the unimorph piezoelectric
harvester is equal to (7):
 
6g31 tp Yp Yb tb tb + tp
V=−    × M.
w Yp2 tp4 +Yb2 tb4 +2Yp tp Yb tb 2tp2 +3tp tb +2tb2
(7)
In order to use (7) for estimating the harvested voltage,
the actuating momentum, which is equal to (8), has to be
computed,
M = (Ib Yb ) k̂, (8)
where k̂ is the average of radius curvature. Based on the
Euler-Bernoulli beam theory, the governing equation of the Fig. 3. The working mechanism flowchart of the GA method.
beam can be expressed by (9):
∂ 4 d(x, t) ∂ 2 d(x, t) reproduction, which strives to find appropriate strings in a
Yb I b + ρA = F(t), (9) population and interpolate them into a mating pool. A number
∂x4 ∂t2 of methods for string selection have been proposed in the
where ρ is the beam density, A is the cross-sectional area of literature, although the main idea is that certain preferable
the beam, d(x,t) is the deflection along the Z-axis as shown in strings from the current population should be chosen, dupli-
Fig. 2, and F(t) represents the external excitation force on the cated and inserted into the pool. The next operator within
beam, whose oscillation frequency is identical to the resonant GA is crossover, where typically two strings are selected
frequency of the beam. Moreover, the radius curvature in terms from the mating pool and a certain quota of these strings
of time t is equal to (10): are exchanged in between. In other words, the recombination
∂ 2 d(x, t) between string pairs produces new strings, called offspring.
k (x, t) = . (10) Finally, mutation operator is performed to change one bit
∂x2
from 1 to 0 or vice versa. This process is also random with
Eventually, the required average of radius curvature for
a very low probability (called mutation rate) on the entire
momentum calculation in (8) can be calculated in (11),
 population. All the three operators are performed on the entire
1 lp population in one GA generation, which is counted by g in
k̂ = k(x, t)dx (11)
lp 0 Fig. 3. Thus, the search and optimization aspect of GA is
mainly provided by the crossover and mutation operators.
where l p denotes the length of the piezoelectric film.
The multi-dimensional search capability offered by GA can
Since the Euler-Bernoulli beam theory is used in the deriva-
effectively prevent it from being trapped by local optima [37].
tion above, the following prerequisites should be observed in
Therefore, a significant feature of GA in comparison with
order to ensure the highest accuracy of the final analytic mod-
the conventional optimization approaches is its advantageous
eling: (1) the ratio between the beam length and its thickness
access to the global optimum.
is greater than 10 [32]; (2) the ratio between the beam length
In this study, our proposed GA-based optimization method
and its width is greater than 2 [33]; (3) the piezoelectric film
is performed to identify optimum physical aspects of the beam
length is much greater than its thickness [34].
and piezoelectric film in the MEMS piezoelectric harvesters
for the efficiency enhancement of energy harvesting. The
III. G ENETIC -A LGORITHM -BASED O PTIMIZATION coverage of the electrodes is defined identical to the size of
The Genetic Algorithm (GA) is an evolutionary comput- the piezoelectric film. In the following section, the capabil-
ing method for searching and optimizing complex problems ity of the GA-based optimization methodology in efficiency
by mimicking biological evolution [35]. The GA starts its improvement and its applicability to the micro-fabrication
operation by generating some random numbers for a group of process will be discussed and demonstrated.
variables to maximize or minimize certain defined fitness func-
tion(s). The GA solutions are normally coded in binary string
IV. E XPERIMENTAL R ESULTS AND D ISCUSSIONS
structure. During the computation, three different operators can
be typically applied to the set of solutions to provide new In order to evaluate accuracy of the presented analytic
evolutionary population. The flowchart of the GA mechanism equations in Section II for harvested voltage of the unimorph
is illustrated in Fig. 3 [36]. piezoelectric energy harvesters, one piezoelectric energy
As shown in Fig. 3, GA starts with initialization, where the harvester with the listed parameters in Table I was imple-
variables are coded in the form of fixed-length binary string. mented in MATLAB (Version 2014) for analytic computation,
Each variable can be randomly selected with equal probability. while the other one with the same properties was simulated
Usually the first operator that performs on a population is with COMSOL Multiphysics (Version 5.2) for finite element
7376 IEEE SENSORS JOURNAL, VOL. 17, NO. 22, NOVEMBER 15, 2017

on the physical geometry optimization of the harvesters, rather

than having to consider any external parameters (e.g., load
impedance) additionally.
The proposed optimization method was implemented in
MATLAB genetic algorithm toolbox (Version 2014) to
enhance the magnitude of harvested voltage besides device
size reduction. The piezoelectric energy harvesters were
excited with a sinusoidal vibration, whose amplitude of accel-
eration is 0.5g (1g = 9.8 m/s2 ) with a frequency identical
to the resonant frequency specification of the corresponding
energy harvester.
The applied fitness function and constraints of our GA
Fig. 4. The estimation of the harvested voltages from the unimorph optimization are defined by (12):
MEMS piezoelectric energy harvester by using analytic computation and FEM
simulation. Maximize : {V}
TABLE I Subject to : design rules of the optimizable parameters
U TILIZED PARAMETERS IN THE A NALYTIC C OMPUTATION AND (12)
FEM S IMULATIONS FOR A U NIMORPH P IEZOELECTRIC
E NERGY H ARVESTER For all the five physical optimizable parameters (i.e., lb ,
tb , l p , t p , and w), upper and lower bounds are defined as
per the design rule constraints as listed in the second row of
Table II. Moreover, the observed results by using a commercial
product (i.e., COMSOL optimization module) for MEMS
structural optimization are listed in the fourth row of this
table (named as Com-OPT). To run the COMSOL optimization
module, the geometry of any piezoelectric energy harvester
in 3D space was first defined, and isotropic materials for both
beam and piezoelectric film were considered. We selected the
Time Dependent Study to measure the harvested voltage for
50 ms with a time interval of 5 ms. The optimization module
is configured as follows: Monte Carlo as the optimization
method, 600 as the maximum number of objective evaluations,
and objective maximization as the optimization type. Further
information regarding the COMSOL optimization module can
be found in [38]. To demonstrate and compare the GA capa-
method (FEM). The comparison between the harvested volt- bility, GA was executed 4 times with the same generation size
ages by using the analytic computation technique and the of 20 but with different population sizes ranging from 5 to 45,
FEM simulation is shown in Fig. 4. One can observe that as shown in the rows of OPT1-OPT4. To make a comprehen-
the proposed analytic computation after reaching the steady sive comparison, we also list the resonant frequency, generated
state (around 1.3 ms) has high accuracy and reliability. Thus, peak voltage (V peak ) at a specific frequency, and runtime
we can ensure that such an analytic model has solid potential in Table II.
for being utilized in the magnitude estimation of harvested According to the listed data in Table II, the un-optimized
voltage from the unimorph piezoelectric energy harvesters. case, as named Un-OPT, had a large occupied area with
It should be noted that such high accuracy between ana- extremely small V peak . The optimized unimorph energy
lytic computation and FEM simulation can be attributed to harvester by COMSOL optimization module, labeled as
the relatively simple structure of the unimorph piezoelectric Com-OPT, could enlarge harvested V peak by a factor of 27%
energy harvesters and the applied Euler-Bernoulli beam theory. in comparison with the un-optimized case, while it took
Moreover, our sufficient satisfaction of the accuracy-related 338 minutes in the computation. After the first GA optimiza-
prerequisites as listed in Section II is highly essential. tion run (labeled as OPT1) with a small population size of 5,
Since the analytic equations presented in Section II are able the generated V peak was highly reduced in comparison with
to accurately estimate the harvested voltage for the given har- Un-OPT although the GA optimization process was done very
vester dimensions, we opt to utilize them in a fitness function quickly. By increasing the population size to 15, the required
of our GA optimization, which includes five physical variables runtime was increased along with the energy harvesting
defined as optimizable parameters, i.e., beam length (lb ), beam efficiency. OPT3 with a sufficiently large population size
thickness (tb ), piezoelectric-film length (l p ), piezoelectric-film of 30 demonstrates that increasing the population size can
thickness (t p ), and beam or piezoelectric width (w). Note that help improve the magnitude performance of the piezoelectric
we define the harvested voltage instead of the harvested power energy harvesters by a factor of 3.96 although more runtime is
as our fitness function of GA. In this way, we can only focus expected. It is worth noting that selecting a proper population
NABAVI AND ZHANG: DESIGN AND OPTIMIZATION OF PIEZOELECTRIC MEMS VIBRATION ENERGY HARVESTERS BASED ON GA 7377

TABLE II
U N -O PTIMIZED AND O PTIMIZED F IVE P HYSICAL A SPECTS OF THE MEMS P IEZOELECTRIC V IBRATION E NERGY H ARVESTERS
BY COMSOL O PTIMIZATION M ODULE AND GA W ITH D IFFERENT P OPULATION S IZES

size is highly critical for the GA optimization. As the data

of OPT4 shows, although an even larger population size is
utilized, the runtime is significantly increased but the perfor-
mance improvement in terms of V peak is very little compared
to OPT3.
On the other hand, the obtained results for OPT3 show
that the magnitude performance improvement was actually
emerging on top of dimension reduction in several aspects,
such as piezoelectric film length, beam thickness, and beam
width. This is indeed highly desirable for MEMS piezoelectric
design optimization. In addition, the comparison between
the obtained data by using GA and COMSOL optimization
module clearly demonstrates that GA has gained less runtime
but with higher capability in energy harvesting efficiency
improvement. Besides the computation results from the ana-
lytic equations, the eighth column in Table II also provides the
FEM numerical simulation results for the harvested voltage Fig. 5. Fabrication process flow for manufacturing our MEMS unimorph
magnitude by using COMSOL Time Dependent Study. It is piezoelectric energy harvesters.
found that both results above are in line with each other very
well. In general, from Table II one can observe that a slim Silicon-on-Insulator (SOI) wafers. The top surface of the
beam (i.e., larger length, smaller width, and thinner layer) may silicon layer is doped by depositing a phosphosilicate
lead to larger magnitude of the harvested voltage. Moreover, glass (PSG) layer and being annealed at 1050° C for one hour
a shorter and thicker piezoelectric film tends to contribute in argon (as depicted in Fig. 5 Step-1). Thereafter, the PSG
considerable improvement to the energy conversion for the layer is removed via wet chemical etching. Then the piezoelec-
MEMS unimorph piezoelectric energy harvesters. tric film, 0.5 μm aluminum nitride (AlN), is deposited over the
As reflected from Table II, the GA-based optimization wafer by reactive sputtering technique (as depicted in Fig. 5
method can successfully optimize a large number of free Step-2). Subsequently, the wafer is coated with photoresist
variables (i.e., five physical parameters) that the designers can and the piezoelectric film is lithographically patterned. After
select during the design process. However, for the most of the completion of this step, the top electrode, which includes
commercially accessible fabrication processes, the thickness 20 nm chrome and 1 μm aluminum, is deposited and patterned
of the layers (e.g., beam layer and piezoelectric layer) is through a liftoff process. In the next step, the silicon layer is
normally fixed. In other words, the designers have no control lithographically patterned, and Deep Reactive Ion Etch (DRIE)
on the thickness amount during the fabrication process. Thus, is performed to etch the silicon layer down to the oxide
to consider such manufacturing limitation, in this study we also layer (as depicted in Fig. 5 Step-3). Then on the front side
performed the GA optimization and associated comparison of the wafer, protection material is deposited (as shown
on three harvester physical aspects (i.e., beam length lb , in Fig. 5 Step-4). After the wafer is reversed, the substrate is
piezoelectric film length l p , beam and piezoelectric width w) lithographically patterned and etched by using RIE and DRIE
as reported in Table III. Moreover, several harvesters with the methods. Eventually the deposited protection material on the
distinct physical aspects were fabricated by a micro-fabrication front side is stripped by using a dry etch process. As shown
process and then verified by our experimental measurement. in Fig. 5 Step-5, this MEMS process can release our clamped
The simplified fabrication process flow, which was utilized unimorph piezoelectric harvester prototypes.
to fabricate our harvester prototypes, is sketched in Fig. 5. In Table III, the coverage ranges for the three optimizable
This process starts with 150 mm n-type double side polished parameters based on the design rule constraints and our
7378 IEEE SENSORS JOURNAL, VOL. 17, NO. 22, NOVEMBER 15, 2017

TABLE III
U N -O PTIMIZED AND O PTIMIZED T HREE P HYSICAL A SPECTS OF THE MEMS P IEZOELECTRIC V IBRATION E NERGY H ARVESTERS
BY COMSOL O PTIMIZATION M ODULE AND GA W ITH D IFFERENT P OPULATION S IZES

5 to 3 can considerably reduce the GA runtime. However,

the efficiency improvement of the COMSOL optimization
module is not significant. Although Un_OPTF was selected
to use the maximum available physical dimensions, its perfor-
mance on the harvested voltage magnitude is not necessarily
superior. Moreover, in comparison with Com_OPTF, OPTF1
can achieve equivalent harvested voltage magnitude but with
much less runtime. Among all the candidates, OPTF2 can
achieve the largest harvested voltage magnitude with reason-
able runtime.
In addition, compared to the measured output voltage,
we can see that the analytic model presented in Section II is
capable of estimating the harvested voltage magnitude with
accuracy of around 80%. Our analyses show that such a
difference between the analytic estimation and experimental
measurement is mainly due to neglect of air damping ratio
(normally 0.01 - 0.05 [28]) and absence of electrode thickness
during the analytic computation. On the one side, if we
Fig. 6. SEM images of the fabricated micro-harvesters (a) un-optimized
consider an air damping ratio of 0.015 in the FEM simulation,
(Un-OPTF), (b) optimized with population size of 15 (OPTF1) and the simulation results would be very close to the measurement
(c) optimized with population size of 30 (OPTF2). results. On the other side, the electrode thickness may take up
to 10% of the beam thickness. Furthermore, it can be seen
budgetary chip plan are listed. It is worth mentioning that that our proposed GA optimization methodology with proper
the size of an un-optimized harvester, labeled as Un_OPTF, is size of population is able to enhance the harvested voltage
selected with the maximum allowable sizes in all the dimen- by a factor of 31% while reducing the harvester physical size
sions. The COMSOL optimization module result, labeled as by 50%. In contrast, the efficiency of the optimized harvester
Com_OPTF, is listed in the fourth row of Table III. On top of with COMSOL optimization module (i.e., Com_OPTF) is
the GA optimization study from Table II, two promising evolu- improved only by 6.4% while it needs a lot more time for
tionary population sizes of 15 and 30 were selected to perform completion of the optimization. Therefore, we conclude that
GA-based optimizations in the new scenario here as listed by our presented analytic model is very helpful for estimating
OPTF1 and OPTF2, respectively. Table III includes the peak harvested voltage of the MEMS unimorph piezoelectric energy
harvested voltage magnitude from the analytic computation harvesters and our proposed GA-based optimization method-
in Column 5 by using our presented analytic equations and ology can facilitate the MEMS design both effectively and
the FEM simulation in Column 6 by using the COMSOL efficiently.
Time Dependent Study at the harvester resonant frequencies. One may observe that the reported peak harvested voltage
Moreover, Un-OPTF, OPTF1, and OPTF2 were fabricated by magnitudes for the studied harvesters in Table II are generally
using the micro-fabrication process above. Com_OPTF was less than those listed in Table III. This is all due to the
not selected into our silicon-area-constrained chip fabrication differences of two applied technologies and excitation inputs.
due to its comparable V peak performance but significantly In Table II the beam and piezoelectric film use aluminum
inferior runtime efficiency with reference to OPTF1. The and PZT respectively, while in Table III the two materials are
measurement results are also included in Table III (Column 7). silicon and AIN. Moreover, we used 0.5g acceleration as the
The SEM images of the prototyped devices are illustrated excitation input for the computation/simulation in Table II,
in Fig. 6. and 2g acceleration in the experiment of Table III for eas-
From the summarized data in Table III, one can observe ier measurement data reading. Therefore, one can conclude
that number reduction of the optimizable parameters from that the performance of our proposed GA-based optimization
NABAVI AND ZHANG: DESIGN AND OPTIMIZATION OF PIEZOELECTRIC MEMS VIBRATION ENERGY HARVESTERS BASED ON GA 7379

TABLE IV frequency of OPTF1 is 9.2% greater than that of Com_OPTF

C OMPUTED M ASSES OF THE U N -O PTIMIZED AND due to their clear difference in stiffness. Therefore, it can be
O PTIMIZED P IEZOELECTRIC E NERGY H ARVESTERS
concluded that the geometry optimization is of high signifi-
cance for the unimorph piezoelectric energy harvesters due to
the sensitivity of stiffness besides device mass.
Another vital parameter to be investigated for the perfor-
mance of one piezoelectric energy harvester is its impedance.
The internal piezoelectric impedance, which is known as a
source impedance, can be modeled by VanDyke equivalent
circuit [39]. According to this model, the internal piezoelec-
tric generator impedance is dependent on several physical
properties of the resonator, such as stiffness, mass, mechani-
cal damping, and piezoelectric capacitance. Consequently the
geometry optimization proposed in this paper can be used to
change the internal impedance of the piezoelectric harvesters.
Fig. 8 illustrates the simulated internal impedance for the
four energy harvesters as listed in Table III, including the
resistance curves (with the maximum amount at the resonant
frequency) and the reactance curves (with zero magnitude at
the resonant frequency). In this regard, 3D geometry of these
devices was modeled by COMSOL Multiphysics and 0.01 was
considered as a damping ratio around the device resonant
frequency. Then 0.01 volt was applied on the top side of the
piezoelectric film as a test voltage to measure the impedance in
Frequency Domain Study. Among the measured impedances,
it is observed that the maximum impedance belongs to the
Fig. 7. Tip displacement versus excitation force for the un-optimized and energy harvester with the minimum resonant frequency and
optimized harvesters with consideration of three optimizable parameters. the minimum piezoelectric film area, which is OPTF2. In con-
trast, Com_OPTF, which has higher resonant frequency than
methodology is independent of technology material properties Un_OPTF, features the minimum internal impedance.
and external excitation inputs. Such intricacies can be understood by the following rea-
As reflected in Table III, geometry modification of energy soning. Since any piezoelectric generator has an internal
harvesters may change the amount of the resonant frequency. impedance, whose amount is a function of device reso-
In theory the resonant frequency of a mechanical resonator nant frequency and piezoelectric film area, the piezoelec-
can be computed by (13): tric energy harvester with higher resonant frequency and
larger piezoelectric film area would offer lower internal
1 k
f= , (13) impedance. By providing an optimum impedance load cor-
π m responding to the piezoelectric internal impedance, a perfect
where k is the stiffness and m is the mass of the harvester. That impedance matching network would be offered so that the
is to say, the resonant frequency is largely dependent on the maximum harvested power can be extracted from the energy
physical aspects of the harvester. Table IV lists the computed harvesters.
mass for the presented energy harvesters (i.e., the un-optimized The von-Mises stress distributions for the four unimorph
and optimized ones) in Table III. There is certain mass piezoelectric energy harvesters from Table III by applying
change that can be observed from the various harvesters above. a 0.01 N constant force at their tip edge under the Load
Among them, Un_OPTF is the heaviest while OPTF2 is the Edge feature in Stationary Study of COMSOL Multiphysics
lightest. are shown in Fig. 9. According to this figure, the minimum
In order to compute the required stiffness for resonant observed actuation stress belongs to Un_OPTF with mag-
frequency calculation by using (13), Fig. 7 exhibits the tip nitude of 2.9 ∗ 109 N/m2 (Fig. 9(a)), while the maximum
displacement versus excitation force for the four presented one happens to OPTF2 with magnitude of 6.9 ∗ 109 N/m2
harvesters from Table III. To measure the tip displacement, (Fig. 9(d)). That is to say, Un_OPTF and OPTF2 gain the
we used COMSOL Multiphysics to model the harvesters in lowest and highest energy conversion efficiency in terms of
3D space and applied variable forces by edge load at tip the harvested voltage, respectively. Fig. 9(b) and (c) demon-
location in Stationary Study with Parametric Sweep function. strate that Com_OPTF and OPTF1 have almost similar peak
The slope of each curve can clearly represent the stiffness von-Mises stress with magnitude of 4.5 ∗ 109 N/m2 and
of the corresponding energy harvester. One can observe that 4.43 ∗ 109, N/m2 , respectively. As a result, it can be con-
Un_OPTF has the maximum stiffness and OPTF2 is the cluded that OPTF2 has the best performance in terms of both
minimum one. Furthermore, even though the Com_OPTF and actuating stress and occupied area in comparison with the other
OPTF1 devices have almost identical masses, the resonant un-optimized and optimized MEMS piezoelectric harvesters.
7380 IEEE SENSORS JOURNAL, VOL. 17, NO. 22, NOVEMBER 15, 2017

Fig. 8. Internal impedance of (a) Un_OPTF, (b) Com_OPTF, (c) OPTF1 and (d) OPTF2 piezoelectric MEMS harvesters.

Fig. 9. 3-D views and von-Mises stress distributions of un-optimized and

optimized piezoelectric MEMS harvesters: (a) Un_OPTF, (b) Com_OPTF, Fig. 10. 3-D views and induced charge densities of un-optimized and
(c) OPTF1 and (d) OPTF2. optimized piezoelectric MEMS harvesters: (a) Un_OPTF, (b) Com_OPTF,
(c) OPTF1 and (d) OPTF2.
To illustrate the direct effect of our proposed design automa- also theoretically reflected from our presented equation set in
tion methodology on performance enhancement, the induced Section II. Thus, a unimorph piezoelectric energy harvesting
charge density by the piezoelectric effect was computed and device with the capability of forming higher stress would
exhibited in Fig. 10. By using COMSOL MEMS module, be able to generate larger harvested voltage. Our proposed
the induced charge density was measured by applying a 0.01 N GA-based optimization methodology can optimize the physi-
force at the tip edge of each harvester. Figure 10 clearly cal geometry of the unimorph piezoelectric energy harvesters
demonstrates that the highest charge density with magnitude in order to form higher stress and in turn generate higher
of 0.025 C/m2 belongs to the GA-optimized harvester with output voltage.
sufficient population size (i.e., OPTF2). In contrast, the min-
imum charge density of 0.011 C/m2 was observed for the V. C ONCLUSION
un-optimized harvester, Un-OPTF. Moreover, the comparison In this paper we proposed a GA-based design and opti-
of the induced charge density between the optimized harvester mization method for MEMS unimorph piezoelectric energy
with COMSOL optimization module (i.e., Com_OPTF) and harvesters. The analytic equations for estimating the generated
our GA-optimized harvester with population size of 15 (i.e., voltage from the MEMS unimorph piezoelectric harvesters
OPTF1) shows that they have very similar capability in pro- were first presented and then validated by using COMSOL
ducing charge with magnitude of 0.019 C/m2 and 0.015 C/m2 , Multiphysics, a commercial FEM tool, as well as experi-
respectively. On the other side, one should not ignore the fact mental measurement of our prototype devices. The proposed
that OPTF1 can run 19.7 times faster than Com_OPTF. GA-based optimization methodology in this study demon-
A comprehensive comparison among the von-Mises stress, strated an enhancement of energy harvesting efficiency by 31%
induced charge density, and measured peak harvested voltage in comparison with un-optimized harvesters. This improve-
shows that there is a direct relationship between actuating ment was gained along with physical size reduction in several
stress, induced charge, and harvested voltage amount. This is aspects such as beam width and piezoelectric film length.
NABAVI AND ZHANG: DESIGN AND OPTIMIZATION OF PIEZOELECTRIC MEMS VIBRATION ENERGY HARVESTERS BASED ON GA 7381

In addition, the comparison between the proposed optimiza- [16] S. Sunithamani, P. Lakshmi, and E. E. Flora, “PZT length optimiza-
tion method in this paper and available commercial product tion of MEMS piezoelectric energy harvester with a non-traditional
cross section: Simulation study,” Microsyst. Technol., vol. 20, no. 12,
(e.g., COMSOL optimization module) shows that our pp. 2165–2171, 2014.
GA-based optimizer has higher optimization efficiency in [17] G. Sordo, E. Serra, U. Schmid, and J. Iannacci, “Optimization method
enlarging the harvested voltage magnitude and reducing the for designing multimodal piezoelectric MEMS energy harvesters,”
Microsyst. Technol., vol. 22, no. 7, pp. 1811–1820, 2016.
computation time. Moreover, some effects of the geometry [18] L. Zhang, V. Masek, and N. N. Sanatdoost, “Structural optimization
optimization on harvester properties (such as resonant fre- of Z-axis tuning-fork MEMS gyroscopes for enhancing reliability and
quency, mass, stiffness, and internal impedance) were studied. resolution,” Microsyst. Technol., vol. 21, no. 6, pp. 1187–1201, 2015.
[19] S. M. K. Tabatabaei, S. Behbahani, and P. Rajaeipour, “Multi-
The developed design and design automation techniques would objective shape design optimization of piezoelectric energy harvester
be essential for the MEMS unimorph piezoelectric energy using artificial immune system,” Microsyst. Technol., vol. 22, no. 10,
harvester designers to be liberated from the conventional pp. 2435–2446, 2016.
[20] S. Nabavi and L. Zhang, “MEMS piezoelectric energy harvester design
laborious trial-and-error effort. and optimization based on genetic algorithm,” in Proc. IEEE Int.
As the future work, we intend to add the optimization Ultrason. Symp. (IUS), Sep. 2016, pp. 1–4.
of resonant frequency and quality factor into our established [21] J. Kim et al., “Design optimization of PZT-based piezoelectric cantilever
beam by using computational experiments,” J. Electron. Mater., vol. 45,
GA-based optimization platform. To reach these objectives, no. 8, pp. 3848–3858, 2016.
we need to investigate the analytic modeling of resonant fre- [22] G. Ye, J. Yan, Z. J. Wong, K. Soga, and A. Seshia, “Optimi-
quency and quality factor for the MEMS unimorph piezoelec- sation of a piezoelectric system for energy harvesting from traffic
vibrations,” in Proc. IEEE Int. Ultrason. Symp. (IUS), Sep. 2009,
tric energy harvesters. Moreover, we should develop advanced pp. 759–762.
evolutionary algorithms for the purpose of multi-objective [23] A. L. Araújo, V. S. Carvalho, C. M. M. Soares, J. Belinha, and
optimization. A. J. M. Ferreira, “Vibration analysis of laminated soft core sand-
wich plates with piezoelectric sensors and actuators,” Compos. Struct.,
vol. 151, pp. 91–98, Sep. 2016.
R EFERENCES [24] N. Chidambaram, A. Mazzalai, and P. Muralt, “Measurement of effective
piezoelectric coefficients of PZT thin films for energy harvesting appli-
[1] S. Nabavi and L. Zhang, “Portable wind energy harvesters for low-power cation with interdigitated electrodes,” IEEE Trans. Ultrason., Ferroelect.,
applications: A Survey,” Sensors, vol. 16, no. 7, p. 1101, 2016. Freq. Control, vol. 59, no. 8, pp. 1624–1631, Aug. 2012.
[2] C. Zhang, X.-F. He, S.-Y. Li, Y.-Q. Cheng, and Y. Rao, “A wind energy [25] A. H. Meitzler, H. F. Tiersten, A. W. Warner, D. Berlincourt,
powered wireless temperature sensor node,” Sensors, vol. 15, no. 3, G. A. Couqin, and F. S. Welsh, III, IEEE Standard on Piezoelectricity,
pp. 5020–5031, 2015. Society. IEEE, 1988.
[3] R. Song, X. Shan, F. Lv, and T. Xie, “A study of vortex-induced energy [26] S. Roundy and P. K. Wright, “A piezoelectric vibration based generator
harvesting from water using PZT piezoelectric cantilever with cylindrical for wireless electronics,” Smart Mater. Struct., vol. 13, no. 5, p. 1131,
extension,” Ceram. Int., vol. 41, pp. S768–S773, Jul. 2015. 2004.
[4] H. Zhang et al., “A flexible and implantable piezoelectric generator [27] P. Janphuang, R. Lockhart, N. Uffer, D. Briand, and N. F. de Rooij,
harvesting energy from the pulsation of ascending aorta: In vitro and “Vibrational piezoelectric energy harvesters based on thinned bulk PZT
in vivo studies,” Nano Energy, vol. 12, pp. 296–304, Mar. 2015. sheets fabricated at the wafer level,” Sens. Actuators A, Phys., vol. 210,
[5] A. Moure et al., “Feasible integration in asphalt of piezoelectric cymbals pp. 1–9, Apr. 2014.
for vibration energy harvesting,” Energy Convers. Manage., vol. 112, [28] J. I. Daniel, Engineering Vibration. Englewood Cliffs, NJ, USA:
pp. 246–253, Mar. 2016. Prentice-Hall, 2001.
[6] Q. Zhang and E. S. Kim, “Microfabricated electromagnetic energy [29] T. Eggborn, “Analytical models to predict power harvesting with piezo-
harvesters with magnet and coil arrays suspended by silicon springs,” electric materials,” M.S. thesis, Dept. Mech. Eng., Viginia Polytech. Inst.
IEEE Sensors J., vol. 16, no. 3, pp. 634–641, Feb. 2016. State Univ., Blacksburg, VA, USA, 2003.
[7] Y. Zhu, S. O. R. Moheimani, and M. R. Yuce, “A 2-DOF MEMS [30] S.-H. Kim, J.-H. Ahn, H.-M. Chung, and H.-W. Kang, “Analysis of
ultrasonic energy harvester,” IEEE Sensors J., vol. 11, no. 1, piezoelectric effects on various loading conditions for energy harvest-
pp. 155–161, Jan. 2011. ing in a bridge system,” Sens. Actuators A, Phys., vol. 167, no. 2,
[8] M. Perez, S. Boisseau, P. Gasnier, J. Willemin, M. Geisler, and pp. 468–483, 2011.
J. L. Reboud, “A cm scale electret-based electrostatic wind turbine for [31] R. C. Hibbeler, Mechanics of Materials. Englewood Cliffs, NJ, USA:
low-speed energy harvesting applications,” Smart Mater. Struct., vol. 25, Prentice-Hall, 1997.
no. 4, p. 45015, 2016. [32] A. L. Herrera-May et al., “Analytical modeling for the bending resonant
[9] Y. J. Gao, Y. G. Leng, S. B. Fan, and Z. H. Lai, “Performance of frequency of sensors based on micro and nanoresonators with complex
bistable piezoelectric cantilever vibration energy harvesters with an structural geometry,” IEEE Sensors J., vol. 11, no. 6, pp. 1361–1374,
elastic support external magnet,” Smart Mater. Struct., vol. 23, no. 9, Jun. 2011.
p. 95003, 2014. [33] L. G. Villanueva, R. B. Karabalin, M. H. Matheny, D. Chi, J. E. Sader,
[10] A. Rivadeneyra et al., “Tunable MEMS piezoelectric energy harvesting and M. L. Roukes, “Nonlinearity in nanomechanical cantilevers,” Phys.
device,” Microsyst. Technol., vol. 22, no. 4, pp. 823–830, 2016. Rev. B, Condens. Matter, vol. 87, no. 2, p. 024304, 2013.
[11] S. Saxena, R. Sharma, and B. D. Pant, “Design and development of [34] S. Kumar, R. Srivastava, and R. K. Srivastava, “Design and analysis
guided four beam cantilever type MEMS based piezoelectric energy of smart piezo cantilever beam for energy harvesting,” Ferroelectrics,
harvester,” Microsyst. Technol., vol. 23, no. 6, pp. 1751–1759, 2016. vol. 505, no. 1, pp. 159–183, 2016.
[12] X. Gao, W.-H. Shih, and W. Y. Shih, “Flow energy harvesting using [35] L. Zhang, R. Raut, Y. Jiang, and U. Kleine, “Two-stage placement
piezoelectric cantilevers with cylindrical extension,” IEEE Trans. Ind. for vlsi analogue layout designs,” IEE Proc.-Circuits, Devices Syst.,
Electron., vol. 60, no. 3, pp. 1116–1118, Mar. 2013. vol. 153, no. 3, pp. 274–280, Jun. 2006.
[13] H. J. Jung et al., “Design and optimization of piezoelectric impact- [36] A. E. Eiben and J. E. Smith, Introduction to Evolutionary Computing,
based micro wind energy harvester for wireless sensor network,” Sens. vol. 53. Berlin, Germany: Springer-Verlag, 2003.
Actuators A, Phys., vol. 222, pp. 314–321, Feb. 2015. [37] S. Nabavi and L. Zhang, “Design and optimization of wideband mul-
[14] Y. Jia and A. A. Seshia, “Power optimization by mass tun- timode piezoelectric MEMS vibration energy harvesters,” in Proc. 31st
ing for MEMS piezoelectric cantilever vibration energy harvesting,” Eurosesnosors, Sep. 2017, vol. 1, no. 4, p. 586.
J. Microelectromech. Syst., vol. 25, no. 1, pp. 108–117, Feb. 2016. [38] COMSOL Optimization Module. Accessed: Apr. 10, 2017. [Online].
[15] D. S. Ibrahim, A. G. A. Muthalif, N. H. D. Nordin, and T. Saleh, “Com- Available: https://www.comsol.com/optimization-module
parative study of conventional and magnetically coupled piezoelectric [39] N. Kong, D. S. Ha, A. Erturk, and D. J. Inman, “Resistive impedance
energy harvester to optimize output voltage and bandwidth,” Microsyst. matching circuit for piezoelectric energy harvesting,” J. Intell. Mater.
Technol., vol. 23, no. 7, pp. 2663–2674, 2016. Syst. Struct., vol. 21, no. 13, pp. 1293–1302, 2010.
7382 IEEE SENSORS JOURNAL, VOL. 17, NO. 22, NOVEMBER 15, 2017

Seyedfakhreddin (Koorosh) Nabavi was born Lihong Zhang (M’06) received the Ph.D. degree
in Hamedan, Iran, in 1991. He received the in electrical engineering from the Otto von
B.Sc. (Hons.) degree in biomedical engineering Guericke University of Magdeburg, Magdeburg,
(bioelectric) from the Sahand University of Technol- Germany, in 2002. He was a Post-Doctoral Research
ogy, Tabriz, Iran, in 2013, and the M.Sc. degree in Associate with Concordia University, Montreal, QC,
electrical and electronic engineering from Ozyegin Canada, Dalhousie University, Halifax, NS, Canada,
University, Istanbul, Turkey, in 2015. He is currently and the University of Washington, Seattle, WA,
pursuing the Ph.D. degree with the Department USA. He is currently an Associate Professor with the
of Electrical and Computer Engineering, Memor- Department of Electrical and Computer Engineer-
ial University of Newfoundland, St. John’s, NL, ing, Faculty of Engineering and Applied Science,
Canada. Memorial University of Newfoundland, St. John’s,
His research interests include design and modeling of piezoelectric MEMS NL, Canada. His current research interests include very large-scale integration
transducers for portable and biomedical applications. computer-aided design, mixed-signal integrated system/circuit design, micro-
electromechanical systems design and design automation, wireless sensor
networks, microfluidics and biosensors, and microprocessor-based instrumen-
tation for ocean and biomedical applications.