JOURNAL OF APPLIED PHYSICS 109, 014904 共2011兲

Characterization of ferroelectric material properties of multifunctional lead zirconate titanate for energy harvesting sensor nodes Bozidar Marinkovic,1 Tolga Kaya,2 and Hur Koser1,a兲 1

Department of Electrical Engineering, Yale University, P.O. Box 208284, New Haven, Connecticut 06520, USA 2 School of Engineering and Technology, Central Michigan University, ET 100, Mount Pleasant, Michigan 48859, USA

共Received 26 July 2010; accepted 30 October 2010; published online 5 January 2011兲 We propose a microsystem integration technique that is ideal for low-cost fabrication of vibration energy harvesting sensor nodes. Our approach exploits diverse uses of sol-gel deposited lead zirconate titanate, effectively combining fabrication of several microsystem components into a single process and significantly reducing manufacturing cost and time. Here, we measure and characterize thin film parameters—such as the piezoelectric coefficient e31 共−4.0 C / m2兲, the dielectric constant ␧r-eff 共219 at 3.3 V兲, and the total switching polarization 共2Pr ; 52 ␮C / cm2兲—in order to verify this material’s potential for energy harvesting, energy storage, and nonvolatile memory applications simultaneously on the same device. © 2011 American Institute of Physics. 关doi:10.1063/1.3524271兴 I. INTRODUCTION

II. FABRICATION AND TESTING SETUP

Energy harvesting microsensor platforms 共e.g., Smart Sand1兲 could eventually enable wireless sensor networks with ubiquitous deployment, continuous monitoring, long lifetime, added redundancy, and large area coverage.2,3 In this context, the cheaper the sensor node is, the more pervasive the sensor network could be. Unfortunately, efforts to trim manufacturing costs are confounded by the need to have numerous functionalities—such as energy harvesting, information storage, sensing, computation, and communication— simultaneously present within a microsensor.4,5 Each subsystem typically requires a different set of materials and fabrication procedures, which scales up system complexity and increases overall manufacturing costs, regardless of the manufacturing specifics.6 One way to collectively reduce fabrication costs is to utilize a material deposited in a single unit process to achieve multiple functionalities within the same device. Such an approach would dramatically simplify node fabrication and ultimately save on production time and cost. For vibration energy harvesting microsensors utilizing piezoelectric transduction, sol-gel deposited lead zirconate titanate 共PZT兲 offers one such opportunity. Although ferroelectric materials have been previously studied for various applications—such as energy harvesting,1,4,7–10 energy storage,10–13 and nonvolatile information storage14–16—a microsystem incorporating these different functionalities of the same material has not yet been developed. In this paper, we propose a microsystem fabrication method that applies the same metal-PZT-metal structure to achieve at least three different functionalities simultaneously: energy transduction, electrical energy storage, and information storage. We present material characterization of sol-gel deposited PZT and experimental verification of its potential for these three applications.

PZT thin films were fabricated by first spinning a lead titanate precursor on oxidized silicon substrates with previously patterned Pt/Ti bottom electrodes. PZT sol-gel 共Mitsubishi Materials Co., Sanda, Japan兲 was then repeatedly spun and pyrolized in layers 共at 400 ° C兲 until the desired thickness was achieved. This multispin process, combined with standard photolithography, could enable fabrication and arbitrary configuration of different thickness PZT patches on the same substrate. All PZT films were patterned using diluted hydrochloric and hydrofluoric acid. Prior to top Au/Ti electrode deposition, the substrate was annealed at 600 ° C. The thickness of fabricated films was measured with a 5 nm resolution in step height using a Tencor Alpha-Step IQ surface profilometer 共Milpitas, CA, USA兲. X-ray diffraction 共XRD兲 patterns, obtained with Shimadzu XRD-6000 共Kyoto, Japan兲, were used to study the phase constitution of deposited film. Open circuit voltages from stressed PZT films were measured using a custom-built, high input impedance amplifier; these measurements helped determine the material’s piezoelectric coefficient. Other ferroelectric properties such as polarization hysteresis, total switching polarization, switching fatigue, retention, and leakage, as well as chargedischarge properties of metal-PZT-metal capacitors, were measured using a Radiant PrecisionLC Workstation 共Albuquerque, NM, USA兲. Dielectric measurements were conducted using Agilent E4980A 共Santa Clara, CA, USA兲.

a兲

Electronic mail: [email protected].

0021-8979/2011/109共1兲/014904/5/$30.00

III. RESULTS AND DISCUSSION A. Structural and ferroelectric characterization

The multilayer deposition technique described above has been known to yield nanoscale pores,17 also observed in the scanning electron microscopy 共SEM兲 image shown in Fig. 1共a兲. As expected, the pores are mostly concentrated between spin-on layers due to vacancy trapping of sequential sol-gel

109, 014904-1

© 2011 American Institute of Physics

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

014904-2

J. Appl. Phys. 109, 014904 共2011兲

Marinkovic, Kaya, and Koser

FIG. 1. 共Color online兲 Structural characterization of sol-gel PZT. 共a兲 SEM of metal 共0.2 ␮m兲-PZT 共1.5 ␮m兲-metal 共0.2 ␮m兲 on oxidized Si 共100兲 substrate shows nanoscale pores concentrated between spin-on layers. 共b兲 XRD spectrum of such a multilayer film exhibits only perovskite phase formation, as desired for good piezoelectric performance 共Ref. 18兲.

deposition and pyrolysis. XRD results shown in Fig. 1共b兲, however, indicate the presence of only ferroelectric perovskite phase with dominant 共110兲 and 共100兲 crystal orientations.18 The lattice constant of a cubic unit cell was measured to be 4.07 Å. Controlling the spin rate of sol-gel deposition and total number of deposited layers, the thickness of a particular PZT patch can be engineered anywhere between 60 nm to 1.5 ␮m. To study the evolution of material’s basic piezoelectric properties with film thickness, we have fabricated several PZT stacks of various thicknesses. Figure 2共a兲 shows the corresponding hysteresis loops when fabricated metalPZT-metal structures are subjected to a maximum electric field of 700 kV/cm. From these data, we make two observations. First, an approximately inverse relationship exists between ferroelectric thickness and its coercive field 关Fig. 2共b兲兴. Second, 2Pr remains little changed with increasing film thickness, even though the maximum polarization reached at the highest applied electric field gets larger 关Fig. 2共c兲兴. The observed changes in piezoelectric properties are attributable to the larger grain size of thicker PZT films.13 In addition, heat treatments during the fabrication of the thinnest films reduce Pb content, leading to sporadic formations of Pb-deficient pyrochlore PZT, which in turn dilutes the film’s relative permittivity19 共␧r-eff兲 as shown in Fig. 2共d兲. Several conclusions can be made from these experimental results in the context of PZT thin film applications. First,

FIG. 3. 共Color online兲 Experimental setup for measuring the piezoelectric coefficient e31. 共a兲 Silicon cantilever beam with dimensions 17⫻ 1 ⫻ 0.27 mm3 is clamped and displaced to measure the open circuit voltage 共Vout兲 across two PZT patches. The silicon beam is over two orders of magnitude thicker than the PZT films, ensuring ideal strain transfer. 共b兲 Voltage output from 1.5 ␮m thick sol-gel PZT shows excellent linear dependence on mechanical strain and the resulting open circuit voltages add up when the two patches are connected in series.

thicker films are desired in energy harvesting applications, where the output voltage of stressed PZT patch is directly proportional to film’s thickness and accumulated charge to its dielectric constant. Second, for a given area, capacitance is proportional to the ratio of ␧r-eff to PZT thickness. Hence, energy storage applications will most benefit from the thinnest ferroelectric layer that can be practically fabricated, as long as the internal field does not exceed the breakdown value. Third, even though thinner films exhibit larger coercive fields, they are still better suited for nonvolatile memory applications, as they require lower operating voltages to achieve the necessary electric field to switch the polarization. In fact, the metal-PZT-metal capacitors that we have tested are thin enough to be switched directly 共i.e., without any voltage multiplication兲 with the 0.35 ␮m, 3.3 V complementary metal-oxide semiconductor 共CMOS兲 circuitry. This is in contrast with the conventional floating gate flash memory, which typically requires much larger voltages to write and erase each bit.20

B. Energy harvesting potential

In the context of energy harvesting, a piezoelectric material can be characterized by its e31 coefficient, which defines its capability to convert lateral mechanical strain into out-of-plane electric field. Using a cantilever test structure as depicted in Fig. 3共a兲, e31 can be easily deduced for the case of thick sol-gel PZT films. Below, we seek to relate this coefficient to directly measurable quantities 共such as voltage and film thickness兲, starting with the following constitutive equations of piezoelectricity:21

␴i = 兺 Cijs j − 兺 eijE j , j

j

Di = 兺 eijs j + 兺 ␧ijE j . j

FIG. 2. 共Color online兲 Thickness effects on piezoelectric properties of solgel deposited PZT. 共a兲 Hysteresis loops measured with the same applied electric field. 共b兲 Coercive field 共Ec兲 as measured from hysteresis. 共c兲 2Pr as measured from hysteresis. Inset: maximum polarization 共Pmax兲 attained at 700 kV/cm. 共d兲 Relative permittivity of PZT films goes up with thickness, both for single and multiple layers.

共1兲

j

Here, ␴ and D are the stress and dielectric field tensors, respectively. C is the stiffness matrix, s is the strain tensor, E is the electric field tensor, e is the piezoelectric coefficient matrix, and ␧ permittivity matrix of the PZT material. In general, the summations are across all the tensor components; however, given the crystal structure of perovskite PZT, Eq. 共1兲 reduces to the following matrix formulation:21

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

014904-3

J. Appl. Phys. 109, 014904 共2011兲

Marinkovic, Kaya, and Koser

冢 冣冢 冣冢 冣 冢 冣 冢 冣 冢冣冢 冣冢 冣 ␴11 ␴22 ␴33 ␴23 ␴13 ␴12

=

C11 C12 C13

0

0

0

C12 C12 C13

0

0

0

C13 C12 C13

0

0

0

0

0

0

C44

0

0

0

0

0

0

C55

0

0

0

0

0

0

C66

0

−

0

冢

s13 s12

e31

0

0

e31

0

0

e33

0

e15

0

e15 0

0

0

0

0

E1 E2 , E3

0 0 0 0 e15 0 D1 D2 = 0 0 0 e15 0 0 D3 e31 e31 e33 0 0 0

+

s11 s22 s33 s23

␧11

0

0

0

␧22

0

0

0

␧33

冣冢 冣

E1 E2 . E3

s11 s22 s33 s23 s13 s12

共2兲

Following the experimental conditions at the base of the cantilever, we adopt a single degree of freedom model, where the PZT film is subjected to stress along one direction 共x-axis兲 within its film plane. We further assume that no shear stresses or strains exist 共no torsion applied on the beam兲, and that the stresses and strains developed along the orthogonal directions 共y- and z-axes兲 are negligible compared to that along the x-axis 共due to low Poisson ratio of silicon22兲. Under these assumptions, Eq. 共2兲 further reduces to

␴11 = C11s11 − e31E3 , 0 = C12s11 − e31E3 , 0 = C13s11 − e33E3 , D3 = e31s11 − ␧33E3 .

共3兲

The first three equations in 共3兲 require a priori knowledge of PZT stiffness coefficients 共C11, C12, and C13兲; for sol-gel PZT, these are strongly process-dependent and are not convenient for deriving e31 or e33 coefficients. However, if no free charge is allowed to accumulate on the metal electrodes of the PZT stack 共i.e., open circuit conditions on a previously discharged stack兲, the displacement field D3 is zero, and the last equation in 共3兲 yields an expression for e31

e31 = −

Vout␧r−eff␧0 . hPZTs11

共4兲

Here, Vout is the voltage generated across PZT film thickness 共hPZT兲, ␧r-eff is the effective dielectric constant, ␧0 is the permittivity of free space, and s11 is the lateral strain. In deducing the e31 coefficient, the lateral strain is taken to be the same as that calculated for the silicon cantilever alone. Since the PZT film is at least two orders of magnitude thinner than the underlying silicon and is well over an order of magnitude more compliant, this is an excellent approximation. Using the simple cantilever setup in Fig. 3共a兲, we have measured Vout as a function of s11 关Fig. 3共b兲兴. Based on Eq. 共4兲 and a measured ␧r-eff of 834, the corresponding value for e31 is −4.0 C / m2. It is possible that this piezoelectric performance may be further improved by incorporating an alignment layer below the PZT to ensure a more dominant 共110兲 crystal orientation.23 Nevertheless, this simple and costeffective PZT process yields patches that generate approximately 800 mV at one millistrain. What is more, voltage outputs of two PZT patches strained at the same time and connected in series add up without a loss in average energy density 共140 ␮J / cm3 of PZT volume at one millistrain兲. This observation indicates that sufficient voltage to power integrated circuits could still be generated in low vibration environments by connecting even more PZT patches in series. Combining this material with Smart Sand1 could potentially generate up to 6 V at one millistrain 共with two PZT patches on each tether and all four tethers connected in series兲. C. High-value integrated capacitor

Sol-gel PZT can be patterned as a single layer and be used for electrical energy storage on the same sensor node, thanks to its low leakage current 关Fig. 4共a兲兴 and large dielectric constant 关Fig. 4共b兲兴. Note that, at a given bias, the nonlinear capacitance of the metal-PZT-metal structure will quickly decay to its eventual, lowest value 共corresponding to fully switched polarization at long soak times兲. Therefore, in evaluating PZT’s potential as an energy storage medium, the lowest capacitance density values in Fig. 4共b兲 should be used. Furthermore, changing polarization direction within PZT is an inherently dissipative process,24 resulting in energy loss. Figure 4共c兲 shows the efficiency of electrical energy recovery from a storage capacitor as characterized at various dc levels with minimum leakage contributions 共see Ref. 25 for measurement details兲. To put this measurement into perspective, let’s consider a circuitry fabricated with standard, 0.35 ␮m CMOS technology26 and powered by the storage capacitor as it discharges from 3.6 down to 3 V 共⫾10% of nominal 3.3 V rail voltage兲. A 1 mm2 capacitor charged to 3.6 V 关corresponding to 31 nF; Fig. 4共b兲兴 will yield 42 nJ of adjusted energy 关Fig. 4共d兲兴 when discharged to 3 V 共35 nF兲. Considering the measured leakage values in this voltage range, it would take over ten seconds for the capacitor to discharge on its own. With this energy budget, it is possible to provide the CMOS circuit with either high power for a short duration 共e.g., 4 mW for 10 ␮s兲 or lower power for much longer 共e.g., 4 ␮W for 10 ms兲. This “single-shot”

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

014904-4

J. Appl. Phys. 109, 014904 共2011兲

Marinkovic, Kaya, and Koser

FIG. 4. 共Color online兲 Characterization of operational parameters for a 60 nm thick PZT storage capacitor as a function of dc bias voltage. 共a兲 Leakage current density 共measured quasistatically following long “soak” times to ensure elimination of displacement current兲. 共b兲 Nonlinear capacitance at 50 Hz, measured at different polarization states ranging from not-switched to fully switched as obtained with various soak times. 共c兲 Efficiency of energy recovery from the PZT storage capacitor with different soak times. At 3.3 V operating conditions, the efficiency value starts at as low as 50% 共for an initial remnant polarization of +Pr for positive biases applied and vice versa兲 and quickly reaches up to 96% 共once the polarization settles at that bias兲. The final state corresponds to the lowest capacitance shown in 共b兲 but to the highest efficiency of energy recovery. 共d兲 Total stored energy on the storage capacitor. Given a range of operating voltages 共e.g., 3.6–3.0 V兲, the extractable energy corresponds to the difference in stored energy at those voltages 共i.e., ⌬E兲, multiplied by the average efficiency within that range 共e.g., 96%兲 at steady state.

共or “energy-limited”兲 operation mode is ideal when harvested power levels are stochastically varying, and sufficient energy must be stored to ensure correct circuit operation.27,28 Continuous operation is “power-limited,” and is possible only when harvested power exceeds the sum of power consumed by the circuit and that lost due to leakage.29 D. Nonvolatile ferroelectric memory

Single-shot operation may require a means of recording sensed variables in nonvolatile memory until a receiver is ready to read the data.30,31 In this context, the hysteretic nature of polarization in these thin capacitors 关see Fig. 5共a兲兴 renders them conveniently suited for information storage, as well. Each unit cell can be switched between its two remnant polarization states 共+Pr or −Pr兲 by applying sufficiently large voltage across its electrodes. At 5 V, the maximum measured polarization reversal 共2Pr兲 is 52 ␮C / cm2. 2Pr is also a function of pulse amplitude and its duration, as depicted in Fig. 5共b兲. We have found that the size of the unit cell has negligible effect on these measurements25 and to achieve fastest operation and lowest energy consumption, it is always desirable to work with smallest possible unit cells. Assuming the same 3.3 V CMOS technology, a 3 V, 1 ms write pulse on a 25⫻ 25 ␮m2 unit cell yields a 2Pr value of 22 ␮C / cm2. This polarization reversal consumes a total of 206 pJ 共33 ␮J / cm2兲, corresponding to less than 0.5% per bit of the total energy extracted from the 1 mm2 storage capacitor discussed above. The memory cell and its write energy could be much smaller—limited mainly by higher resolution fabrication costs. Once polarization is set, information can be retained for at least 15 days with no more than 25% 2Pr loss 关Fig. 5共c兲兴. This retention time facilitates monitoring appli-

FIG. 5. 共Color online兲 Characterization of 60 nm thick, 25⫻ 25 ␮m2 nonvolatile ferroelectric memory cell. 共a兲 Hysteresis loops of these thin PZT films 共as opposed to the much thicker one depicted in Fig. 2兲 exhibit both larger polarization and higher coercive field values. Inset: 2Pr reaches 52 ␮C / cm2 when 5 V is applied across the memory cell. 共b兲 Polarization change 共⌬P兲 as a function of drive pulse amplitude and width. 共c兲 Retention in “written” polarization over time 共normalized to its starting value, P0兲. Less than 25% of polarization is lost after 15 days. 共d兲 Relative change in 2Pr as a function of number of switching cycles. In this fatigue test, pulses with alternating polarity, 4 V amplitude and 1 ␮s duration were applied. Device can sustain over 108 cycles.

cations in which data retrieval occurs infrequently. The material can sustain over a 100 million switching cycles throughout its lifetime 关Fig. 5共d兲兴, significantly more than conventional flash memory.32 To put this number into context, we note that even if the same memory location were overwritten every 10 s, it could remain functional for over 30 years. IV. CONCLUSION

In summary, we report material fabrication and characterization of sol-gel deposited PZT films in the context of three different applications. We have demonstrated that the same metal-PZT-metal structure can be used as an effective energy transducer, storage capacitor, and nonvolatile memory and have shown that one unit process can be applied to realize all three. The fabrication is simple enough to be implemented directly by the manufacturer of the energy harvester platform itself, without the need for expensive, highresolution processes in external foundries. Combining this approach with low-power circuitry for sensors and wireless communication, a vibration energy harvesting platform could result in a truly integrated and cost-effective node for ubiquitous wireless sensor networks. As an example, an integrated energy harvesting microsensor platform such as Smart Sand could use multifunctional metal-PZT-metal structures to generate more than enough energy at 2.5 g 共100 ␮m at 80 Hz兲 to charge up a 1 mm2 storage capacitor to 3.6 V, and only a small fraction 共4%兲 of the total available energy 共42 nJ if the capacitor is allowed to discharge to only 3 V兲 goes to storing the sensed information at 8 bit resolution 共0.5% of available energy per bit兲. The system requires no voltage multiplication since the thick PZT patches generate more than sufficient output voltage to directly power up both CMOS circuitry and the thin ferroelectric memory elements. This is a significant advan-

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

014904-5

tage over traditional energy harvesting microsystems that incorporate charge pumps or multiplier circuitry to provide the wide range of voltages that their nonvolatile memory and other components need. State-of-the-art charge pumps,33 for instance, can achieve up to 79% efficiency, but only for small load currents on the order of 10 ␮A. In the milliampere range, desired for single-shot operation, the efficiency of such circuits could drop below 20%. Therefore, the elimination of a voltage multiplier or a charge pump circuitry dramatically increases the efficiency of any harvesting system that incorporates our proposed approach. In general, we note that the concept of multifunctional thin films may be extended to other materials, as well. For instance, it may be possible to utilize paramagnetic thin films as both nonvolatile memory elements and as components of integrated inductors in the same device. ACKNOWLEDGMENTS

This work was supported by the National Science Foundation Grant No. ECCS-0601630 and the Office of Naval Research Grant No. N00014-09-01-0197. We also thank Jason Hoffman for technical help in XRD data acquisition. B. Marinkovic and H. Koser, Appl. Phys. Lett. 94, 103505 共2009兲. J. P. Lynch and K. J. Loh, Shock Vib. Dig. 38, 91 共2006兲. G. Müller, T. Rittenschober, and A. Springer, Elektrotechnik und Informationstechnik 127, 39 共2010兲. 4 L. M. Miller, P. K. Wright, C. C. Ho, J. W. Evans, P. C. Shafer, and R. Ramesh, “Integration of a low frequency, tunable MEMS piezoelectric energy harvester and a thick film micro capacitor as a power supply system for wireless sensor nodes,” IEEE Energy Conversion Congress and Exposition, San Jose, CA, 20–24 September 2009, pp. 2627–2634. 5 A. Witvrouw, Scr. Mater. 69, 945 共2008兲. 6 R. A. Lawes, Microsyst. Technol. 13, 85 共2007兲. 7 A. Erturk, J. Hoffmann, and D. J. Inman, Appl. Phys. Lett. 94, 254102 共2009兲. 8 J. L.-C. Blystad, E. Halvorsen, and S. Husa, IEEE Trans. Ultrason. Ferroelectr. Freq. Control 57, 908 共2010兲. 9 D. Benasciutti, L. Moro, S. Zelenika, and E. Brusa, Microsyst. Technol. 16, 657 共2010兲. 10 Y. Lin and H. A. Sodano, J. Appl. Phys. 106, 114108 共2009兲. 11 H.-B. Kang, S.-K. Hong, H.-C. Park, H.-Y. Chang, K.-W. Park, J.-H. Ahn, 1 2 3

J. Appl. Phys. 109, 014904 共2011兲

Marinkovic, Kaya, and Koser

J.-S. Kih, M. Y. Sung, and Y.-K. Sung, Integr. Ferroelectr. 89, 94 共2007兲. J. Sigman, G. L. Brennecka, P. G. Clem, and B. A. Tuttle, J. Am. Ceram. Soc. 91, 1851 共2008兲. 13 G. L. Brennecka and B. A. Tuttle, J. Mater. Res. 22, 2868 共2007兲. 14 T. P.-C. Juan, C.-L. Lin, W.-C. Shih, C.-C. Yang, J.-Y. M. Lee, D.-C. Shye, and J.-H. Lu, J. Appl. Phys. 105, 061625 共2009兲. 15 J. Carrano, C. Sudhama, V. Chikarmane, J. Lee, A. Tasch, W. Shepherd, and N. Abt, IEEE Trans. Ultrason. Ferroelectr. Freq. Control 38, 690 共1991兲. 16 K.-H. Kim, J.-P. Han, S.-W. Jung, and T.-P. Ma, IEEE Electron Device Lett. 23, 82 共2002兲. 17 G.-T. Park, J.-J. Choi, C.-S. Park, J.-W. Lee, and H.-E. Kim, Appl. Phys. Lett. 85, 2322 共2004兲. 18 B. A. Tuttle, T. J. Headley, B. C. Bunker, R. W. Schwartz, T. J. Zender, C. L. Hernandez, D. C. Goodnow, R. J. Tissot, J. Michael, and A. H. Carim, J. Mater. Res. 7, 1876 共1992兲. 19 C. A. Randall, N. Kim, J.-P. Kucera, W. Cao, and T. R. Shrout, J. Am. Ceram. Soc. 81, 677 共1998兲. 20 P. Pavan, R. Bez, P. Olivo, and E. Zanoni, Proc. IEEE 85, 1248 共1997兲. 21 IEEE Standard on Piezoelectricity, ANSI/IEEE Standards, 176-1987 共1987兲. 22 S. D. Senturia, Microsystem Design 共Kluwer Academic, Dordrecht, 2001兲. 23 D. Ambika, V. Kumar, H. Imai, and I. Kanno, Appl. Phys. Lett. 96, 031909 共2010兲. 24 S. C. Hwang, C. S. Lynch, and R. M. McMeeking, Acta Metall. Mater. 43, 2073 共1995兲. 25 See supplementary material at http://dx.doi.org/10.1063/1.3524271 for information regarding the calculation of the energy recovery efficiency and size effects on PZT memory performance. 26 AMS 0.35 Micron CMOS Process Parameters, Austria Mikro Systeme International AG, 1998. 27 L. Chao, C.-Y. Tsui, and W.-H. Ki, “A Batteryless vibration-based energy harvesting system for ultra low power ubiquitous applications,” Proceedings of the IEEE International Symposium on Circuits and Systems, New Orleans, LA, 27–30 May 2007, pp. 1349–1352. 28 T. Kaya, H. Koser, and E. Culurciello, Electron. Lett. 42, 526 共2006兲. 29 A. Kansal, J. Hsu, S. Zahedi, and M. B. Srivastava, ACM Transactions on Embedded Computing Systems 6, 32 共2007兲. 30 D. Jung, T. Teixeira, and A. Savvides, ACM Transactions on Sensor Networks 5, 1 共2009兲. 31 T. Kaya and H. Koser, “A New Batteryless Active RFID System: Smart RFID,” 1st Annual RFID Eurasia, Istanbul, Turkey, 5–6 September 2007, pp. 1–4. 32 R. Micheloni, M. Picca, S. Amato, H. Schwalm, M. Scheppler, and S. Commodaro, Proc. IEEE 97, 148 共2009兲. 33 M. R. Hoque, T. Ahmad, T. R. McNutt, H. A. Mantooth, and M. M. Mojarradi, IEEE Trans. Circuits Syst., II: Express Briefs 53, 364 共2006兲. 12

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp