An Electromagnetic Micro Power Generator for Low-Frequency Environmental Vibrations Based on the Frequency Upconversion Technique

14 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 19, NO. 1, FEBRUARY 2010

An Electromagnetic Micro Power Generator forLow-Frequency Environmental Vibrations Based

on the Frequency Upconversion TechniqueIbrahim Sari, Tuna Balkan, and Haluk Külah, Member, IEEE

Abstract—This paper presents a microelectromechanical-system-based electromagnetic vibration-to-electrical power gen-erator that can harvest energy from low-frequency externalvibrations. The efficiency of vibration-based harvesters is pro-portional to excitation frequency, so the proposed generator isdesigned to convert low-frequency environmental vibrations toa higher frequency by employing the frequency upconversion(FupC) technique. It has been shown that the generator can effec-tively harvest energy from environmental vibrations of 70–150 Hzand generates 0.57-mV voltage with 0.25-nW power from a sin-gle cantilever by upconverting the input vibration frequency of95 Hz–2 kHz. The fabricated generator size is 8.5×7×2.5 mm3,and a total of 20 serially connected cantilevers have been usedto multiply the generated voltage and power. The generatordemonstrated in this paper is designed for the proof of concept,and the power and voltage levels can further be increased byincreasing the number of cantilevers or coil turns. The perfor-mance of the generator is also compared with that of a same-sized custom-made traditional magnet–coil-type generator andwith that of a traditional generator from the literature to prove itseffectiveness. [2009-0136]

Index Terms—Array of cantilevers, energy harvesting, energyscavenging, frequency upconversion (FupC), micro powergenerator.

I. INTRODUCTION

W ITH INCREASING global warnings on environmentalissues, clean energy sources have particularly become

important in the past decades. In the meantime, low-cost, low-power, and miniaturized devices can easily be manufacturedusing advanced techniques of microelectromechanical-systemtechnology and the electronics industry. These improvementsenabled many of the electronic devices to be miniaturized andoperated at considerably low power levels, leading to batteries

Manuscript received May 21, 2009; revised August 30, 2009. First publishedDecember 22, 2009; current version published February 3, 2010. This workwas supported by The Scientific and Technological Research Council of Turkey(TÜBITAK) under Grant 104E119. Subject Editor R. T. Howe.

I. Sari was with the Department of Mechanical Engineering, Middle EastTechnical University, 06531 Ankara, Turkey. He is now with the School ofElectronics and Computer Science, University of Southampton, Southampton,SO17 1BJ, U.K. (e-mail: [email protected]).

T. Balkan is with the Department of Mechanical Engineering, Middle EastTechnical University, 06531 Ankara, Turkey.

H. Külah is with the Department of Electrical and Electronics Engineeringand the MEMS Research and Application Center, Middle East TechnicalUniversity (METU), 06531 Ankara, Turkey.

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JMEMS.2009.2037245

and other conventional capacitive sources to be replaced byenvironmental sources.

Researchers have so far worked on harvesting energy fromenvironmental sources like solar, thermal, wind, and vibrationsto power up these devices [1], [2]. Among these techniques,vibrations seem to be more attractive due to their availability,high power density, and easiness of integration to microfabrica-tion and assembly [3]. So far, mainly three types of vibration-based techniques have been utilized to harvest energy:capacitive [4]–[11], piezoelectric [12]–[17], and electromag-netic [17]–[36].

These techniques have been well studied in the literatureso far, and it is possible to find examples of applications topower up various devices. Each of the vibration-based energy-harvesting methods has its own advantages and drawbacks interms of power density, integration, electrical matching, and soon. However, they all intersect at one point where they normallyneed a high-vibration-frequency medium to be effective. Thisarises from the mathematical fact that the maximum generatedpower of these techniques is proportional to the cube of thevibration frequency and drops dramatically at low frequencies(1–100 Hz) [34], [36]. However, it is at these low frequencieswhere most ambient vibration exists. For this reason, vibration-based resonant generators are effective at frequencies of sev-eral kilohertz, but at lower frequencies, they are ineffective[35], [36]. The proposed electromagnetic generator solves thisproblem by mechanically upconverting the low-frequency vi-brations to a higher frequency. This technique was proposedfirst by Külah and Najafi with a milliscale implementation [19]and shown to work in microscale for the first time in the workof Sari et al. [21].

In this paper, the microscale implementation of the techniqueis presented with experimental results. The effectiveness ofthe proposed design has been experimentally verified throughcomparative tests using the following: 1) a same-sized custom-made traditional magnet–coil-type generator and 2) a typicaltraditional generator from the literature by Williams et al. [22],[23]. Finally, the parameters that affect the overall performanceof the generator have been investigated extensively throughsimulations and experimental work.

II. DESIGN AND SIMULATION

Based on the discussion made in the Introduction, thepower output from vibration-based generators is proportional to

1057-7157/$26.00 © 2009 IEEE

SARI et al.: MICRO POWER GENERATOR FOR ENVIRONMENTAL VIBRATIONS BASED ON THE FupC TECHNIQUE 15

Fig. 1. Illustration of the proposed design. (Upper left) isometric, (upper right) side, and (lower) schematic views.

excitation frequency. For this reason, the main objective of theproposed design is to accept low-frequency vibrations as inputand create high-frequency vibrations to act as a linear motiontransformer or, literally, to be a “frequency upconverter.” Thisis achieved using the basic mechanical vibration theory, whichstates that when underdamped structures are excited by aninitial condition such as displacement or velocity, their responsewill be an exponentially decaying out oscillatory motion. Thisvirtually enables a converter mechanism with enough designfreedom to be implemented. In order to realize such a design,the necessary condition is to construct a mechanism that peri-odically excites the generator.

Fig. 1 shows the proposed generator that is simply imple-menting the excitation concept explained earlier. It is composedof two mechanical structures: 1) the upper diaphragm and 2) thearray of cantilevers located right below the diaphragm. The di-aphragm is made of Parylene C and holds a NdFeB magnet forboth frequency upconversion (FupC) and power generation bymeans of electromagnetic induction. The diaphragm–magnetassembly resonates by vibrations in the range of 1–100 Hz.The cantilevers are also made of Parylene C, they have ahigher resonance frequency of 2–3 kHz, and each of themhas a coil for induction. Also, at the tip of each cantilever,nickel is electroplated for interaction with the magnet. As thediaphragm resonates in response to external vibrations, it getscloser to the cantilever array. The distance between them isadjusted such that the magnet catches the cantilevers at a certaininstance of its movement, pulls them up, and releases themat another point. The released cantilevers start resonating attheir damped natural frequency with the given initial condition,realizing the FupC. The motion of the released cantilevers ex-ponentially decays out, and before it completely dies, the cyclestarts again.

The equivalent mechanical model of the system is shown inFig. 2, where m, k, and b denote the equivalent mass, stiffness,

Fig. 2. Equivalent mechanical model of the system.

and damping constants, respectively, and the subscripts m and crepresent the magnets and the cantilevers, respectively. The po-sition coordinates are defined by z and δ with z correspondingto the relative displacements with respect to the base, and δ tothe initial offsets with respect to the base. The whole systemis excited by environmental vibrations of frequency ω anddisplacement amplitude of y. The excitation frequency is closeto the natural frequency of the diaphragm–magnet assemblyand much smaller than the natural frequency of the cantilevers.This causes only the diaphragm–magnet to be excited by envi-ronmental vibrations, and the cantilevers to remain unaffected.Using this model, the magnet–diaphragm and the cantilever dy-namics can be expressed with two separate second-order lineardifferential equations of motion. As a result, the displacement


of the magnet–diaphragm assembly can be represented with thefollowing equation:

zm(t) =

(ω

ωnm

)2

Y√(1−

(ω

ωnm

)2)2

+(2ζ ω

ωnm

)2

sin(ωt + ϕ) (1)

where ϕ is the phase angle expressed by

ϕ = − tan−1

⎛⎜⎝ 2ζm

ωωnm

1 −(

ωωnm

)2

⎞⎟⎠ . (2)

In the aforementioned equations, ωnm is the natural frequencyof the magnet–diaphragm assembly, and ζm is the overalldamping ratio of the magnet–diaphragm assembly.

The relative displacement of the cantilever tip point with re-spect to the base (zc) can be obtained with the application of theinitial conditions zc(0) = zo and zc(0) = 0 and is expressed by

zc(t)=zoe−ζcωnt√

1−ζ2c

(ζc sin(ωdt)+

√1−ζ2

c cos(ωdt))

. (3)

The relative velocity of the cantilever’s tip point is thenobtained by taking the time derivative of the aforesaid equationthat results in

zc(t) = −zoωnce−ζcωnt√

1 − ζ2c

sin(ωdt). (4)

In the equation given earlier, ωnc is the natural frequency, ζc isthe damping ratio, and ωd is the damped natural frequency ofthe cantilevers defined by

ωd = ωn

√1 − ζ2

eq. (5)

As a result, when only the initial conditions are considered, theoutput voltage generated by a single cantilever becomes

ε = −BLP zc. (6)

In the aforementioned equation, B is the magnetic flux densityand LP is the practical coil length. These terms are derived andinvestigated in detail for the cantilevers in the previous work ofSari et al. [34]. This equation can be expanded to give

ε = BLP zoωne−ζeqωnt√

1 − ζ2eq

sin(ωdt). (7)

If it is assumed that the generated voltage is used for driving anelectrical load with an equivalent input resistance of RL, thenthe power term can be expressed with the following:

P =12

ε2

(RL + Rc)2RL (8)

and can be represented in expanded form as

P =12

(BLP

RL + Rc

)2

RL

[zoωnc

e−ζcωnct√1 − ζ2

c

sin(ωdt)

]2

. (9)

Fig. 3. Representation of the contact force.

From (7) and (9), it can be observed that the voltage and powerobtained from the generator will be exponentially decaying outsignals.

Before going into simulation of the system, the equationsderived so far must be modified so that both the cantilever andthe magnet positions are measured from the same referenceposition. The offsets of the magnet (δm) and the cantilever (δc)from the base position are already shown in Fig. 2. If the base ofthe system is considered as the datum, then the dynamic motionequation for the magnet can be rewritten as

mmzm + bmzm + km(zm − δm − δmst) = −mmy − mmg.(10)

In the last equation, δm is the offset of the magnet from the baseto the static equilibrium position, δmst is the static deflection ofthe magnet due to gravity, and g is the gravitational accelerationconstant. The static deflection and gravity terms cancel out eachother to give

mmzm + bmzm + km(zm − δm) = −mmy. (11)

Similarly, for the cantilever, the following equation can be ob-tained considering the same datum that is used for the magnet:

mczc + bczc + kc(zc − δc) = −mcy. (12)

The simulation of the system consists of two modes. The firstmode is the separate motion of the magnet and the cantilevercovered by the last two equations. The second mode is thecombined motion of the magnet and the cantilever, where themagnet and the cantilever move together as a single body dueto the magnetic attraction force between them. The governingequation for the second mode is

(mm + mc)z + (bm + bc)z + km(z − δm)

+ kc(z − δc)= −(mm+ mc)y. (13)


TABLE IOPTIMIZED PARAMETERS OF THE GENERATOR

The condition for transition between these two modes can bedefined using the contact force between the magnet and thecantilever. The contact force must be analytically larger thanthe magnetic attraction force for the bodies to remain intact.Otherwise, they will move independently from each other.Fig. 3 shows the positive directions of these two forces, andwith reference to these directions, the combined and separateoperation conditions can be represented by the following in-equalities:

Combined operation T > −Fm

Separate operation T < −Fm. (14)

The minus sign in the aforementioned inequality is due to thefact that T is the contact force between the bodies, and itis compressive if T > 0. Fm is the magnetic attraction forcebetween the bodies during contact and is defined by

Fm =B2Am

2μo. (15)

In (15), B is the magnetic flux density defined previously, Am

is the magnetic contact area, and μo is the permeability of freespace (μo = 4π · 10−7 T · m/A).

The contact force (T ) can be expressed by either of thefollowing:

mmz + bmz + km(z − δm) = −mmy + T

⇒ T = mm(z + y) + bmz + km(z − δm). (16)

The magnetic attraction force Fm shown in Fig. 3 is not re-flected to dynamic equation (16) as it is a gluing force betweenthe two bodies. The catch and release points are determinedfrom (16), together with the condition defined by (14). Usingthe model described by (11) through (13) and the contactforce defined by (16), the system is optimized and simulatedin the Matlab Simulink environment. The derived power andvoltage terms are optimized using a pattern search algorithmin Matlab to have maximum output from the proposed device.The optimized generator parameters are tabulated in Table I.According to the optimized results, a maximum voltage of0.67 mV and a power of 0.33 nW are estimated to be obtainedfrom a single cantilever of the generator. Fig. 4 shows thesimulated absolute positions of the cantilevers and the magnetobtained using the parameters of Table I.

The system has three different dynamics: 1) cantilever;2) magnet–diaphragm; and 3) combined motion of the can-tilevers and the magnet–diaphragm assembly. During the


Fig. 4. Simulated relative positions of the magnet and the cantilevers.

Fig. 5. Simulated voltage output from a single cantilever.

simulation, the model is switched between these three dynamicmodes by continuously checking the effective conditions ateach sample time step. The shaded area on the left (borderedby the dotted line) shows the combined low-frequency mo-tion of the cantilevers and the magnet, whereas the shadedarea on the right (bordered by the long dashed line) showsthe separate motion of the cantilevers (high-frequency plot)and the magnet (low-frequency plot). The generated voltagefrom a single cantilever is shown in Fig. 5, which has anexponentially decaying out form similar to the motion of thecantilevers.

III. FABRICATION

In this section, the fabrication steps of the FupC designare explained. In order to realize the microfabrication steps ofthe generators, a total number of six masks are designed andfabricated. The cantilever base and diaphragm are fabricated atthe same time and attached to each other later on. Fig. 6 showsa sample layout drawing for a single chip that consists of the

base and the diaphragm. The dimensions of the features are alsoshown on the layouts.

Fig. 7 shows the fabrication process flow of the FupC design.Parylene C is used as the structural material for the cantileversand the diaphragm because it allows much larger deflectionsbefore mechanic failure compared to silicon [37]. First, athermal oxide layer of 200-nm thickness is grown on the siliconsubstrate. This layer provides isolation between the metal con-tact pads and the silicon substrate. Next, a 1-μm-thick paryleneis deposited by chemical vapor deposition and patterned byreactive ion etching (RIE) at the contact pads and cantileverareas [Fig. 7(a)]. Then, coil turns are formed by sputtering andpatterning the first metal layer [Fig. 7(b)]. As the next step,a second 1-μm-thick parylene layer is formed on the metaland patterned to provide electrical isolation between the firstand second metal layers, and vias are opened at the necessarypositions on parylene by RIE to provide contact between thefirst and second metal layers [Fig. 7(c)]. The second metal layeris then sputtered and patterned to complete the metal routes[Fig. 7(d)]. The thickness of the cantilevers is defined mainly


Fig. 6. Schematic drawing of the FupC design.

Fig. 7. Microfabrication steps of the FupC design.


Fig. 8. Photograph of the (left) FupC design base and (right) diaphragm.

by the deposition of the third layer of 13-μm-thick paryleneand patterning [Fig. 7(e)]. Afterward, a 9-μm nickel (Ni) isdeposited by electroplating and patterned by liftoff to form themagnetic actuation areas on the cantilevers [Fig. 7(f)]. Then, afinal layer of 1-μm-thick parylene is deposited and patternedto act as a protection layer for the magnetic areas [Fig. 7(g)].Finally, silicon substrate is through etched from the backsideby deep RIE (DRIE) [Fig. 7(h)]. The exposed oxide layer onthe back is wet etched in a buffered hydrofluoric acid solutionto release the devices [Fig 7(i)]. The devices are then cleaned inacetone and isopropyl alcohol. Then, the magnet is glued to thediaphragm [Fig. 7(j)], and the two chips are combined togetherto form a single device [Fig. 7(k)].

The distance between the magnet and the cantilevers aredefined by using a spacer. Fig. 8 shows the photographs ofthe components of the fabricated FupC design. On the left,the base is placed on a printed circuit board (PCB) for wirebonding, and on the right, the diaphragm is assembled with themagnet stack.

IV. IMPLEMENTATION AND TEST RESULTS

The fabricated micro generators are tested using a vibrationshaker system purchased from Brüel & Kjaer. Fig. 9 showsthe schematic diagram (upper) and corresponding photographs(lower) of the shaker system that is composed of a shaker table,a control unit, an amplifier, an accelerometer, and a softwareinterface. The desired control input values, such as acceleration,velocity and displacement levels, operation frequency range,and other settings, are all entered to the system through a userinterface running on a computer. The software then downloadsthese operation parameters to the main control unit that runs theshaker table at the desired modes and levels of operation. Thecommand signals from the controller are amplified before beingtransmitted to the shaker table by an amplifier unit.

In order to run the shaker table in a closed-loop manner,operation acceleration and frequency levels are measured con-tinuously by an accelerometer and fed back to the controllerunit. The important properties of the shaker system are shownin Table II. Tests can be carried out in a frequency range of10 Hz–20 kHz and up to a bare table acceleration of 75 gwith a maximum force rating of 45 N. Fig. 10 shows thefabricated prototype placed on the shaker table (upper left)and the assembled prototype with its subcomponents preparedfor testing (bottom right). The FupC assembly, shown on the

upper left-hand side, is composed of four major components:a PCB, a FupC base, a diaphragm, and a spacer. The baseis glued permanently to the PCB using an epoxy that canbe applied as a thin film so that minimum roughness can beachieved at the applied surface. Then, the magnet is attachedto the diaphragm manually using the same epoxy under themicroscope for the best alignment of the magnet with respectto the cantilevers. The distance between the FupC base and thediaphragm–magnet assembly is adjusted using silicon spacersof 0.5 mm thickness.

Fig. 11 shows the subcomponents of the FupC assembly.On the lower left-hand side of Fig. 11, the FupC assemblyjust before being attached to the coupling base is shown. Thetapered M5 hole on the coupling base is used to attach the wholeassembly to the shaker table. The two thin wire bonds carryingout the output signals from the device to the PCB can also beseen in the lower right-hand side of the figure.

Fig. 12 shows the photograph of the FupC assembly cou-pled to the shaker table. In order to provide space for theaccelerometer shown in the lower right-hand corner, four9-mm-long spacers with an internal radius of 3.05 mm are used.The M3 bolts pass though these spacers and are fixed by thecorresponding nuts.

Table III lists the test results for the FupC design, togetherwith the simulation results obtained previously. Tests are car-ried out by sweeping the frequency from 70 to 150 Hz and withan input vibration displacement range of 0.44–2 mm. From thetable, it can be seen that a maximum peak power and a voltageoutput of 0.25 nW and 0.57 mV are obtained, respectively,from a single cantilever of the FupC design at an excitationfrequency of 95 Hz and an input displacement of 1.1 mm. Itcan also be seen that the estimated and measured values fora single cantilever are in good agreement. It follows from (7)that the only source of error in the estimation of the voltageoutput arises from the magnetic flux density. This is becausethe experimental setup is made of a complex geometry and themagnet is continuously oscillating up and down. In addition,there is a small amount of stray magnetic field over the shakertable that naturally exists and distorts the actual magnetic fieldlines from the magnet. When added up, all these parametersand uncertainties make it hard to estimate the magnetic fieldover the system precisely, even by finite-element software.

Fig. 13 shows the estimated and measured output voltagesfrom a single cantilever plotted on the same set of axis. In thefigure, the catch and release points of the cantilevers and thepeak voltage output are also indicated. A total of five cycles areshown in the figure, and each cycle lasts for about 10 ms, whichis composed of the combined motion and the individual motionof the cantilevers and the magnet. Voltage generation is realizedright after the release of the cantilevers where they oscillatewith their damped natural frequencies. From the closed-upviews, the catch and release points and the maximum generatedvoltage are clearly seen. It can be deduced that the simulatedand actual voltage waveforms are in close agreement with eachother for a single cantilever of the device.

In order to prove the effectiveness of the proposed generator,a same-sized traditional large mass coil (LMC)-type genera-tor has also been fabricated and tested for performance. The


Fig. 9. Block diagram of the test setup and photographs of the components.

TABLE IIIMPORTANT PROPERTIES OF THE SHAKER SYSTEM

Fig. 10. Fabricated prototype prepared for testing.

LMC-type generator is one of the most popular traditionalelectromagnetic energy scavengers, and it has been extensivelyinvestigated in the literature [22]–[25], [29]–[31]. Fig. 14 shows

Fig. 11. Components of the FupC assembly prepared for testing.

the fabricated traditional generator. It is composed of a basewith planar coils and a diaphragm–magnet assembly. The baseshown on the lower left-hand corner is stationary, whereas themagnet–diaphragm assembly shown on the upper right-handside is movable to induce voltage at the terminals of the coil.


Fig. 12. FupC assembly coupled to the shaker table.

TABLE IIICOMPARISON OF THE SIMULATION AND THE

TEST RESULTS OF THE FupC DESIGN

Table IV compares the performances of the FupC and LMCdesigns.

From the given data, it can be seen that even a single can-tilever of the FupC design is capable of generating two times thevoltage and six times the power of the traditional LMC design.However, considering that the LMC and FupC designs havesinusoidal and decaying-out waveforms, respectively, it wouldbe more meaningful to compare the total energy harvested in acertain cycle of, for example, 1 s. In this case, the LMC designcan collect 23 pJ, whereas the single cantilever of the FupCdesign can collect 20 pJ. There are a total of 20 cantileversused in the FupC design, meaning that the proposed FupCmechanism performs 17 times better than the LMC design.In addition, it should also be noted that the LMC design iseffective only at the resonance frequency of the device, whilethe FupC design has almost the same output performance inany frequency as far as the cantilevers are caught and releasedwithin the physical limits of the generator. In the literature,the LMC design by Williams et al. is the first vibration-based electromagnetic-type micro energy generator proposed

Fig. 13. Simulation and test results plotted on the same set of axis for a singlecantilever of the FupC design.

Fig. 14. Traditional LMC design fabricated for comparison with the proposeddesign.

and has been used as a benchmark model in most of the studies[22], [23]. For this purpose, the performance of the FupC designis also compared with that of this design.

Table V shows the important performance parameters of theLMC design by Williams et al. and the FupC design. The firstrow is the original parameters of the design presented in thework of Williams et al., whereas in the second row, the scaled-down version of this design is shown, simply obtained by recal-culating the performance of the device at 113 Hz. Referring tothe table, a single cantilever of the proposed FupC design hasa better performance in terms of both power and voltage levels.The energy level of the design by Williams et al. is much higherthan that of a single cantilever of the FupC design. However,when the overall performance of 20 cantilevers is considered,the proposed design would still perform much better than the


TABLE IVCOMPARISON OF THE PERFORMANCES OF THE LMC AND FupC DESIGNS

TABLE VCOMPARISON OF THE PERFORMANCES OF THE LMC DESIGN BY WILLIAMS et al. AND THE FupC DESIGN

traditional generator. Another important feature of the FupCdesign is that it can operate at a wide frequency band aroundthe natural frequency of its diaphragm as long as the inputdisplacement amplitude is enough to achieve the catch andrelease of the cantilevers. In order to prove this, Fig. 15 shows

the magnitude of the peak output voltage at sample frequencypoints. It is obtained by sweeping the frequency slowly from50 to 150 Hz at a test displacement of 0.4–3.5 mm and record-ing the peak amplitudes at certain frequencies. In the plot, thedata from two different single cantilevers and the sum of two


Fig. 15. Simulation and test results for the variation of the magnitude of theoutput voltage for different excitation frequencies.

cantilevers, as well as the corresponding simulation results,are shown for comparison. From the given figure, it can beseen that there is a slightly increasing trend of output voltagewith excitation frequency. This is because as the frequencyincreases, the initial release velocity of the cantilevers and, thus,the release heights also increase slightly, resulting in an increasein the output voltage. This can also be verified by the simulationresults shown in the figure. This test shows that continuous andconstant voltage generation in a frequency range is possiblewith the proposed FupC design unlike traditional generators.

The simulation and test results for a single cantilever of theFupC design have been presented so far, and it is shown thatthey are in good agreement. However, when the test results ofmultiple cantilevers are investigated, it has been observed thatthe overall voltage output is less than what it should be. Thefollowing are the two main reasons leading to this.

1) The difference in natural frequency of the cantileverscaused by the dirt remaining from the fabrication ofthe generators. This results in shifting of each cantileveroutput from each other and degradation of the overalloutput.

2) Asynchronous release of the consecutive cantilevers dueto the manual attachment of the diaphragm–magnet as-sembly to the base. This causes the cantilevers to bereleased at different instances, thus shifting the waveformfrom each cantilever and resulting in a degraded overalloutput.

In the following two sections, these two effects will beinvestigated with experimental results.

A. Effect of Difference in Natural Frequencies

One effect that can degrade the overall output is the differ-ence in the natural frequencies of the cantilevers. This causesthe waveforms to shift from each other after sometime andcauses the overall output to degrade. Fig. 16 shows an exampleof voltage outputs from two such cantilevers that have differentnatural frequencies. The blue and red lines are the outputs fromindividual cantilevers, whereas the green line is the overall

Fig. 16. Two cantilevers with nonidentical natural frequencies.

output from these two. From the plot, it can be observed thatalthough the cantilevers start oscillating at the same time, i.e.,released at the same instant, the waveforms deviate from eachother after the second cycle. This causes the overall output tobe less than what it actually should be. The main cause of thedifference in the natural frequencies of the cantilevers is thermal(cool) grease (CGR 7016 from AI Technology, Inc.) residuesremaining after the final step of the fabrication procedure. Thissubstance is used to attach the process wafer to a handle waferto improve thermal conduction during the DRIE step. However,it is quite hard to clean it afterward, which results in residuesremaining on the cantilevers. These remains change the effec-tive mass and, thus, the natural frequency of the cantileversand cause the overall output from the generator to degrade. Atypical cantilever has dimensions of 1000 × 430 × 15 μm3 anda mass on the order of a few micrograms, and it is found outthat a thermal grease residue of dimensions 50 × 50 × 5 μm3

can affect the overall output.

B. Effect of Asynchronous Release of Cantilevers

Another factor affecting the overall output from the can-tilevers is the asynchronous release of the cantilevers, whichresults in time shift or phase-angle difference in the outputvoltage of individual cantilevers. In order to illustrate this,Fig. 17 shows the measured voltage output from two cantileversthat are released at different intervals. The blue and red lines arethe outputs from individual cantilevers, whereas the green lineis the overall output from these two. From the figure, it can beobserved that, due to different release times, the output voltagesare shifted from each other, degrading the overall output. Thetime shift and phase-angle difference are related to each otherwith the damped natural frequency of the cantilevers and aredetermined by the following:

Δt =ΔΦωd

(17)

where Δt, ΔΦ, and ωd are the time- and phase-angle differ-ences and the damped natural frequency of the cantilevers,respectively. The phase-angle difference is a quantity that is


Fig. 17. Two cantilevers released at different instances.

TABLE VISAMPLE VALUES RELATING TIME AND PHASE SHIFT TO

RELEASE-HEIGHT DIFFERENCE FOR A DAMPED NATURAL

FREQUENCY OF ωd = 1.5 kHz AND f = 100 Hz

TABLE VIIEXPERIMENTAL VALUES SHOWING TIME SHIFT (Δt) AND THE

CORRESPONDING RELEASE-HEIGHT DIFFERENCE (Δz)

much easier to interpret than the time difference Δt. Forexample, a phase difference of Δϕ = 180◦ between two can-tilevers directly implies that the overall output from these twocantilevers is zero because the positive and negative intervalsof these two decaying-out sinusoidal signals directly overlapand cancel out each other. Table VI tabulates the sample valuesof time and phase shift and the corresponding release-heightdifference (Δz) for a nominal release height of z = 200 μmand at a damped natural frequency of ωd = 1500 Hz and anexcitation frequency of f = 100 Hz. From the tabulated values,it can be seen that a release-height difference of only 25 μm cancause a phase-angle difference of 180◦ to completely flatten theoverall output.

Table VII shows a number of typical experimental values oftime difference Δt, phase difference Δϕ, and correspondingrelease-height difference Δz. It can be seen that there is an in-accuracy of 3–17 μm in the attachment of magnet to diaphragm

and/or spacers that result in a time shift of 0.03 ms (16◦)to 0.2 ms (109◦), respectively. The test results show that atime shift of more than 0.1 ms corresponding to a heightdifference of 8 μm adversely attenuates the overall output of thecantilevers.

The reason of the release-height difference is the differentrelative distances from each cantilever to the magnet, which arecaused by the manual attachment of the components by epoxy.During the manual application of epoxy, the topology of thelayer can be made smooth up to a certain extent that wouldcreate a tilt on the mating surfaces that are attached. In orderto show whether such a tilt exists, the height of each side ofthe magnet assembly is measured under the probe station thatis shown in Fig. 18. The height of each corner is measuredwith reference to the smooth PCB surface. Then, the heightdifference between the consecutive cantilevers is calculatedgeometrically from similar triangles, using the following:

Δzij = ΔhijΔd

Δr= Δhij

5007000

(18)

where Δhij is the difference in the heights of the corners withrespect to the PCB base, i.e., Δhij = hkl − hij and Δzij isthe corresponding height difference between the height of eachcantilever with respect to the diaphragm. In the final expression,i, j, k, and l represent the corner numbers, and in (18), the termsΔd and Δr correspond to the horizontal distance between eachcantilever and corner in micrometers, respectively. The mea-surements have been made from various devices,and typicalresults are then tabulated and shown in Table VIII.

In the final table given earlier, the maximum height variation[Δzij (max.)] in each data set is shown. The measurementsshow that the vertical height difference between the cantileversand the magnet is compatible with the ones derived from thetime difference values tabulated in Table VII. This proves thatthe time differences between the cantilevers are due to themanual attachment of the device components.

V. CONCLUSION

In this paper, an electromagnetic micro energy generator thatupconverts low-frequency environmental vibrations to a higherfrequency has been presented. The microscale implementationof this design has been shown first here, and the concept hasbeen proven to work in microscale with the actual test resultsfor the first time. It has been shown that a voltage and a poweroutput of 0.57 mV and 0.25 nW can be obtained from a singlecantilever of the microscale design, respectively.

The aim of this paper is to show that the proposed generatorconcept works in microscale and is more efficient comparedto a same-sized traditional micro generator operating under thesame conditions. For this purpose, the device parameters havebeen optimized in a conservative manner for the verificationof the concept. It has been shown that the proposed generatorperforms much better than a custom-made traditional LMCdesign and a typical traditional generator from the literature[22], [23].

The outputs from individual cantilevers are in close agree-ment with simulations; minor deviations are observed due to the


Fig. 18. Illustration of the height-difference measurement under the probe station.

TABLE VIIIRESULTS OF HEIGHT-DIFFERENCE MEASUREMENTS

UNDER THE PROBE STATION

prediction of magnetic flux density, which is quite hard to esti-mate precisely. However, the outputs from multiple cantileversare not compatible with the output from a single cantilever dueto the difference in the natural frequencies of the cantilevers andthe asynchronous release of the cantilevers. These two pointshave been investigated extensively and analyzed with the testresults. It has been shown that the residues coming from thermalgrease that is used at the final step of the fabrication and themanual attachment of the components of the generator causethis deviation. In the future, by employing a different materialthan thermal grease and using an automated attachment tech-nique, these errors can be minimized.

It is also possible to further improve the generated voltageand power by decreasing the coil width to increase the coilturns or by increasing the number of cantilevers. For example,in this design, a coil width of 20 μm has been used, and itcan be decreased to 2 μm to increase the generated outputs.Initial calculations show that this improvement leads to a6.5-fold increase in voltage and power output levels. Withfurther improvements in design parameters, it is possible toimprove the performance of the proposed generator.

REFERENCES

[1] A. D. Joseph, A. Kansal, M. B. Srivastava, J. Randall, T. Buren, G. Troster,T. J. Johnson, W. W. Clark, L. Mateu, and F. Moll, “Energy harvestingprojects,” Pervasive Comput., vol. 4, no. 1, pp. 69–71, Jan.–Mar. 2005.

[2] S. Roundy, E. S. Leland, J. Baker, E. Carleton, E. Reilly, E. Lai, B. Otis,J. M. Rabaey, P. K. Wright, and V. Sundararajan, “Improving power outputfor vibration-based energy scavengers,” Pervasive Comput., vol. 4, no. 1,pp. 28–36, Jan.–Mar. 2005.

[3] J. A. Paradiso and T. Starner, “Energy scavenging for mobile and wirelesselectronics,” Pervasive Comput., vol. 4, no. 1, pp. 18–27, Jan.–Mar. 2005.

[4] F. Peano and T. Tambosso, “Design and optimization of a MEMS electret-based capacitive energy scavenger,” J. Microelectromech. Syst., vol. 14,no. 3, pp. 429–435, Jun. 2005.

[5] S. Roundy, P. K. Wright, and J. Rabaey, “A study of low level vibrationsas a power source for wireless sensor nodes,” Comput. Commun., vol. 26,no. 11, pp. 1131–1144, Jul. 1, 2003.

[6] M. Mizuno and D. G. Chetwynd, “Investigation of a resonance microgen-erator,” J. Micromech. Microeng., vol. 13, no. 2, pp. 209–216, Mar. 2003.

[7] S. Roundy, P. K. Wright, and K. S. J. Pister, “Micro-electrostaticvibration-to-electricity converters,” in Proc. ASME IMECE, New Orleans,LA, 2002.

[8] A. Nounou and H. F. Ragaie, “A lateral comb-drive structure for energyscavenging,” in Proc. ICEEC, 2004, pp. 553–556.

[9] P. D. Mitcheson, P. Miao, B. H. Stark, E. M. Yeatman, A. S. Holmes,and T. C. Green, “MEMS electrostatic micropower generator for lowfrequency operation,” Sensors Actuators A, Phys., vol. 115, no. 2/3,pp. 523–529, Sep. 21, 2004.

[10] T. Sterken, P. Fiorini, K. Baed, R. Puers, and G. Borghs, “An electret-based electrostatic μ-generator,” in Proc. 12th Int. Conf. Solid State Sens.,Actuators Microsyst., Boston, MA, 2003, pp. 1291–1294.

[11] S. Meninger, J. O. Mur-Miranda, R. Amirtharajah, A. P. Chandrakasan,and J. H. Lang, “Vibration-to-electric energy conversion,” IEEE Trans.Very Large Scale Integr. (VLSI) Syst., vol. 9, no. 1, pp. 64–76, Feb. 2001.

[12] S. Roundy, “Energy scavenging for wireless sensor nodes with a focus onvibration to electricity conversion,” Ph.D. thesis, Dept. Mech. Eng., Univ.California, Berkeley, CA, 2003.

[13] P. Glynne-Jones, S. P. Beeby, and N. M. White, “Towards a piezoelectricvibration-powered microgenerator,” Proc. Inst. Elect. Eng.—Sci. Meas.Technol., vol. 148, no. 2, pp. 68–72, Mar. 2001.

[14] G. Poulin, E. Sarraute, and F. Costa, “Generation of electrical energy forportable devices: Comparative study of an electromagnetic and a piezo-electric system,” Sensors Actuators A, Phys., vol. 116, no. 3, pp. 461–471,Oct. 29, 2004.

[15] T. Sterken, K. Baert, C. Van Hoof, R. Puers, G. Borghs, and P. Fiorini,“Comparative modeling for vibration scavengers,” in Proc. IEEE Sensors,2004, vol. 3, pp. 1249–1252.

[16] N. S. Shenck and J. A. Paradiso, “Energy scavenging with shoe-mountedpiezoelectrics,” IEEE Micro, vol. 21, no. 3, pp. 30–42, May/Jun. 2001.

[17] S. Roundy, B. P. Otis, Y. Chee, J. M. Rabaey, and P. Wright, “A1.9 GHz transmit beacon using environmentally scavenged energy,” inProc. ISPLED, Seoul, Korea, Aug. 25–27, 2003.

[18] P. Glynne-Jones, M. J. Tudor, S. P. Beeby, and N. M. White, “Anelectromagnetic, vibration-powered generator for intelligent sensor sys-tems,” Sensors Actuators A, Phys., vol. 110, no. 1–3, pp. 344–349,Feb. 1, 2004.

[19] H. Külah and K. Najafi, “An electromagnetic micro power generator forlow-frequency environmental vibrations,” in Proc. 17th IEEE MEMS,pp. 237–240.

[20] C. Serre, A. Perez-Rodriguez, N. Fondevilla, E. Martincic, S. Martinez,J. R. Morante, J. Montserrat, and J. Esteve, “Design and implementationof mechanical resonators for optimized inertial electromagnetic micro-generators,” Microsyst. Technol.—Micro- Nanosyst. -Inf. Storage Process.Syst., vol. 14, no. 5, pp. 653–658, Apr. 2008.

[21] I. Sari, T. Balkan, and H. Külah, “An electromagnetic micro power gen-erator for low frequency environmental vibrations based on the frequencyup-conversion technique,” in Proc. 22nd IEEE MEMS, Sorrento, Italy,Jan. 25–29, 2009, pp. 1075–1078.

[22] C. B. Williams and R. B. Yates, “Analysis of a micro-electric generatorfor microsystems,” Physica A, vol. 52, no. 1/3, pp. 8–11, Mar.–Apr. 1996.

[23] C. B. Williams, R. C. Woods, and R. B. Yates, “Feasibility study of avibration powered micro-electric generator,” in Proc. IEE Colloq. Conf.Power Sources (Digest No. 96/107), May 8, 1996, pp. 7/1–7/3.

[24] C. Shearwood and R. B. Yates, “Development of an electromag-netic micro-generator,” Electron. Lett., vol. 33, no. 22, pp. 1883–1884,Oct. 23, 1997.

[25] M. El-hami, R. Glynne-Jones, N. M. White, M. Hill, S. Beeby, E. James,A. D. Brown, and J. N. Ross, “Design and fabrication of a new vibration-based electromechanical power generator,” Sensors Actuators A, Phys.,vol. 92, no. 1–3, pp. 335–342, Aug. 1, 2001.

[26] C. T. Pan and T. T. Wu, “Development of a rotary electromagnetic mi-crogenerator,” J. Micromech. Microeng., vol. 17, no. 1, pp. 120–128,Jan. 2007.


[27] T. von Buren, P. D. Mitcheson, T. C. Green, E. M. Yeatman,A. S. Holmes, and G. Troster, “Optimization of inertial micropower gener-ators for human walking motion,” IEEE Sens. J., vol. 6, no. 1, pp. 28–38,Feb. 2006.

[28] P. H. Wang, X. H. Dai, D. M. Fang, and X. L. Zhao, “Design, fabricationand performance of a new vibration-based electromagnetic micro powergenerator,” Microelectron. J., vol. 38, no. 12, pp. 1175–1180, Dec. 2007.

[29] M. Duffy and D. Carroll, “Electromagnetic generators for power har-vesting,” in Proc. 35th IEEE Power Electron. Spec. Conf., 2004, vol. 3,pp. 2075–2081.

[30] W. J. Li, T. C. H. Ho, G. M. H. Chan, P. H. W. Leong, andH. Y. Wong, “Infrared signal transmission by a laser-micromachined,vibration-induced power generator,” in Proc. 43rd IEEE Midwest Symp.Circuits Syst., 2000, vol. 1, pp. 236–239.

[31] N. N. H. Ching, H. Y. Wong, W. J. Li, P. H. W. Leong, andZ. Y. Wen, “A laser-micromachined multi-modal resonating powertransducer for wireless sensing systems,” Sensors Actuators A, Phys.,vol. 97/98, pp. 685–690, Apr. 1, 2002.

[32] R. Amirtharajah and A. P. Chandrakasan, “Self-powered signal processingusing vibration-based power generation,” IEEE J. Solid State Circuits,vol. 33, no. 5, pp. 687–695, May 1998.

[33] I. Sari, T. Balkan, and H. Külah, “A wideband electromagnetic mi-cro power generator for wireless microsystems,” in Proc. Transducers,Jun. 2007, vol. 1, pp. 275–278.

[34] I. Sari, T. Balkan, and H. Külah, “An electromagnetic micro powergenerator for wideband environmental vibrations,” Sens. Actuators A,Phys., vol. 145/146, pp. 405–413, 2008, DOI: 10.1016/j.sna.2007.11.021.

[35] I. Sari, T. Balkan, and H. Külah, “An energy harvesting MEMS frequencydetector,” in Proc. 6th IEEE SENSORS, Atlanta, GA, Oct. 28–31, 2007,pp. 1460–1463.

[36] I. Sari, “Design, fabrication and implementation of a vibration basedMEMS energy scavenger for wireless microsystems,” Ph.D. dissertation,Middle East Tech. Univ., Ankara, Turkey, 2008.

[37] Product Specifications, Parylene Knowledge, Specialty Coating Syst. Inc.,Indianapolis, IN, 2006.

Ibrahim Sari received the B.Sc., M.Sc., and Ph.D.degrees in mechanical engineering from Middle EastTechnical University, Ankara, Turkey, in 1999, 2002,and 2008, respectively.

Since 2008, he has been a Research Fellowin the School of Electronics and Computer Sci-ence, University of Southampton, Southampton,U.K., mainly investigating levitation and propul-sion of microparticles. His research interests includeMEMS, microfabrication, vibration-based energyscavenging, MEMS transducers, and control systems

and applications.

Tuna Balkan received the Ph.D. degree in me-chanical engineering from Middle East TechnicalUniversity (METU), Ankara, Turkey, in 1988.

He is currently a Professor in the Department ofMechanical Engineering, METU, where he is alsothe Assistant Director of the CAD/CAM/RoboticsCenter. His research interests include system dynam-ics, control systems, system modeling, simulationand identification, fluid power control, robotics, andindustrial robotic applications.

Haluk Külah (S’97–M’03) received the B.Sc. andM.Sc. degrees in electrical engineering (with highhonors) from Middle East Technical University(METU), Ankara, Turkey, in 1996 and 1998, respec-tively, and the Ph.D. degree in electrical engineer-ing from the University of Michigan, Ann Arbor,in 2003.

From 2003 to 2004, he was a Research Fellowin the Department of Electrical Engineering andComputer Science, University of Michigan. InAugust 2004, he joined the Department of Electrical

and Electronics Engineering, METU, as a Faculty Member, where he has beenthe Deputy Director of the MEMS Research and Application Center since 2008.His research interests include MEMS sensors, mixed-signal interface electronicdesign for MEMS sensors, bio-MEMS, and MEMS-based energy scavenging.

Dr. Külah was the recipient of several prizes at the 2000 and 2002 DesignAutomation Conferences and the 2002 Student Design Contest, which weresponsored by a number of companies, including Cadence, Mentor Graphics, TI,IBM, Intel, and Compaq. His M.Sc. thesis received the 1999 Thesis of the YearAward given by the Prof. M. N. Parlar Education and Research Foundation,METU.

Documents

An Electromagnetic Micro Power Generator for Low-Frequency Environmental Vibrations Based on the Frequency Upconversion Technique