2017 IEEE 67th Electronic Components and Technology Conference Wafer-Level Vacuum-Packaged Piezoelectric Energy Harvesters Utilizing Two-Step Three-Wafer Bonding Nan Wang, Li Yan Siow, Lionel You Liang Wong, Chengliang Sun, Hongmiao Ji, Darmayuda I Made, Peter Chang, Qingxin Zhang, Yuandong Gu Institute of Microelectronics, A*STAR (Agency for Science, Technology and Research), 2 Fusionopolis Way, #08-02 Innovis Tower, Singapore 138634, sunc@ime.a-star.edu.sg. Abstract This paper reports on the successful implementation of a wafer-level vacuum-packaged, CMOS-compatible aluminum nitride (AlN) based microelectromechanical system (MEMS) energy harvester (EH). The reported EH features high Q-factor (709.3) and high-g survivability (harmonic at 20g), achieved through wafer-level vacuum-package scheme which reduces the air damping effect and increases the Q-factor, overcoming the tradeoff between vibration amplitude and output power density for EHs operated in air. A power of 468.77μW, bandwidth of 71Hz (3.66%) is delivered by a ~0.119cm 3 footprint (1 0.7 0.1705cm 3 ) EH at 20g sinusoidal input vibration, equates to a record power density of ~3.93mW/cm 3. This novel packaging and design scheme, which utilizes two-step three-wafer wafer level eutectic bonding, allows size reduction and shock-resilience improvement of future EH. The ability to harvest broad spectrum mechanical vibration energy with small footprint and high-g survivability, makes the reported EH one step closer towards powering the next generation battery-less Smart Tire Pressure Monitoring System (TPMS). Furthermore, the whole integration process, including the wafer-level vacuum-packaging process, is CMOS compatible, making the reported EH viable for mass production with low fabrication cost. Keywords-Piezoelectric; Energy Harvester; Waferlevel; Vacuum Packaging; Eutectic bonding; I. INTRODUCTION Over the recent years, a great deal of research attention and development effort in microelectromechanical systems (MEMS) and wireless sensors has been paid to micro-sensors, microactuators, as well as micro-systems which are formed by integrating microelectronics with micromechanical devices for various types of applications [1-3]. In quite a number of application scenarios, these microsystems could not be powered by external power source because either they are standalone or embedded in certain environments, where it is very difficult to have physical connection to the outside world. Energy harvesters, especially the piezoelectric ones, are widely used in the vibration systems due to their ability to convert mechanical energy from vibration into electrical energy, in order to power those wireless sensors [4-7]. There are several factors which affect the capability of the energy harvesters to generate power, including the electromechanical properties of the piezoelectric material that is used in the energy harvester, the nature of the vibration source from which the mechanical energy is being harvested, as well as the structural properties of the energy harvester designed which the energy conversion from mechanical energy to electrical energy takes place. In order to maximize the energy harvested from the ambient vibration source, several methods were carried out, of which the most popular method is to change the structural configuration of the piezoelectric energy harvesters. Cantilever plates or beams, due to their relatively easy construction and effectiveness in harvesting energy from ambient vibration source by mechanical amplification mechanism, are the most commonly used structures for piezoelectric energy harvesters [8-9]. Previous studies on piezoelectric energy harvesters usually focus on improving the performance of single element device through more optimized design of the geometry, device structure, and mechanical boundary. The geometric configuration and dimension of an energy-harvesting device usually remains unchanged once the device is implemented. Such a harvesting structure may become less effective in power generation when it operates in a varying-frequency vibrating system. As the demand for powering remote wireless sensors nodes (WSN) increases significantly over recent years, a great deal of research emphasis has been on piezoelectric energy harvesters (EHs) due to their ability to power micro sensors by converting ambient vibrational mechanical energy to electrical energy. Cantilever beams, being the most commonly used structures for piezoelectric EHs due to easy implementation and effective energy harvesting mechanism, are facing two main challenges: 2377-5726/17 $31.00 2017 IEEE DOI 10.1109/ECTC.2017.108 733
1. In order to have high output power density, the vibration amplitude of the cantilever has to be large. However, large vibration amplitude degrades the Q factor due to increased air damping effect, and subsequently reduces the output power density. Therefore, there is a tradeoff between the vibration amplitude and output power density for EHs operated in air. 2. In order to be operated in harsh environments with high acceleration such as Tire Pressure Monitoring System (TPMS), the fabricated EHs have to be shock resilient and mechanically reliable. In order to overcome the aforementioned two challenges, a novel two-step three-wafer wafer-level vacuum packaging scheme is proposed. In the proposed wafer-level vacuum packaging scheme, the MEMS structural wafer is eutectic bonded to two capping wafers, with one capping wafer bonded to the top surface of the MEMS structural wafer and the other capping wafer bonded to the bottom surface of the MEMS structural wafer. As such, the cantilevers which consist the energy harvesters will be vibrating in a vacuum environment, minimizing the air damping effect and enhancing the Q factor. Consequently, the output power can also be improved. Meanwhile, cavities are pre-defined on both the capping wafers, with the depth of the cavities carefully calculated, in order to allow sufficient vibration amplitude of the cantilevers for output power, yet serve as mechanical stoppers to protect the cantilevers, making the fabricated energy harvesters shock resilient and mechanically reliable. II. STURCTURE DESIGN Figure 1 shows the 3D illustration and the resonant model of the designed wafer-level vacuum-packaged EH. In short, the vacuum-packaged EH consists of three parts: top and bottom cap Si with cavities; a middle MEMS device which is sandwiched by the top and bottom cap Si wafers. The cantilever beams of the EH could vibrate within the vacuum sealed room. The vacuum cavity reduces the air damping effect and enhances the Q-factor of the EH, overcoming the aforementioned tradeoff and increases the output power. The MEMS part of the proposed energy harvester consists of 6 cantilevers with a connected proof mass attached to the tip of each of the constituting cantilevers, as shown in Figure 1 (b). The proof mass is designed to connect to all six constituting cantilevers, in order to limit the torsional modes of the cantilevers. Also, the reliability of the fabricated energy harvester can also be improved. Figure 1: 3D illustration (a) and resonant model (b) of the designed vacuum-packaged energy harvester. The proof mass is designed to connect to all six constituting cantilevers, in order to limit the torsional modes of the cantilevers. Also, the reliability of the fabricated energy harvester can also be improved. III. FABRICATION PROCESS The fabrication process of the MEMS structural wafer has been reported in our previous work [10]. Hence, it will not be repeated here for brevity. Figure 2 shows the simulated 3D integration process. In summary, (a) the functional Mo/AlN/Mo stack was deposited on a SOI wafer, followed by top and bottom Al electrodes with Al bonding ring and AlN thin film patterning to form the MEMS wafer; (b) top cap wafer was processed with three-layer processes, i.e., backside alignment marks, stand-off with Ge bonding ring and top Si cavity etch; (c)-(d) the first bonding was done between the MEMS wafer and top cap wafer, followed by MEMS wafer backside release etch ((c)for top side view and (d) for bottom side view); (e) bottom cap wafer was processed with two-layer for the standoff and bottom Si cavity; (f) after the second bonding 734
between the MEMS-top cap wafer and bottom cap wafer, the vacuum-package EH was obtained. Cavities with depth of 270μm are etched on both top and bottom cap wafers for the fabricated energy harvester to vibrate with sufficiently high amplitude. Eutectic Al-Ge wafer-level bonding is employed twice during the fabrication process to bond three wafers together, i.e., bonding of the top cap wafer and the bottom cap wafer to the MEMS wafer, creating the vacuum cavity for the cantilever to vibrate within. As compared with conventional chip-level vacuum packaging, packaging cost has been significantly reduced. Mechanical stoppers are employed to limit the vibration amplitude, increasing high g survivability and reliability of the fabricated EH. Figure 2: 3D integration process illustration of the vacuum-packaged energy harvester. (a) functional Mo/AlN/Mo stack was deposited on a SOI wafer, followed by top and bottom Al electrodes with Al bonding ring and AlN thin film patterning to form the MEMS wafer; (b) top cap wafer was processed with three-layer processes, i.e., back-side alignment marks, stand-off with Ge bonding ring and top Si cavity etch; (c)-(d) the first bonding was done between the MEMS wafer and top cap wafer, followed by MEMS wafer backside release etch ((c)for top side view and (d) for bottom side view); (e) bottom cap wafer was processed with two-layer for the stand-off and bottom Si cavity; (f) after the second bonding between the MEMS-top cap wafer and bottom cap wafer, the vacuum-package EH was obtained. Figure 3: (a) Photograph and (b) the SEM cross section view of the integrated vacuum-packaged energy harvester. Eutectic Al-Ge wafer-level bonding is employed twice during the fabrication process to bond three wafers together, i.e., bonding of the top cap wafer and the bottom cap wafer to the MEMS wafer, creating the vacuum cavity for the cantilever to vibrate within. Figure 3 shows the optical photograph of the top view, and SEM image of the cross-section view of the integrated wafer-level vacuum-packaged EH with a footprint of 1.0cm 0.7cm. Both top and bottom Si cavities have a depth of 270 m. The functional Mo/AlN/Mo stack has thickness of 0.2 m/1.2 m/0.2 m which was deposited on a SOI wafer with 30 m device Si layer. Total thickness of the device is 1705 m. The entire in-house microfabrication process, including the AlN based MEMS process and Al-Ge eutectic wafer-level bonding process, is CMOS compatible. This makes the reported EH very suitable for mass production. IV. CHARACTERIZATION RESULTS AND DISCUSSION Figure 4 shows the frequency response of the generated voltage from both the simulation and measurement results. From the figure we can see that the natural frequency of the device from the measurement results is 1.939 khz which is in excellent agreement with the simulation results of 1.911 khz. Also, the simulated peak open-circuit output voltage is 2.283V, which is very close to the measurement value of 1.8811V. The discrepancy is due to the deviation of the values of material properties that are used in FEM simulation from the value in the fabricated devices. The calculated Q-factor from the measurement results is 709.3, which is essentially the same as the 735
measurement results of 709. Appling this value to the simulation, the simulated voltage sensitivity of 2.283V/g is also close to the measurement result of 1.88V/g. Figure 5 shows the generated open-circuit voltage frequency response at the different input accelerations. Intuitively, as the input acceleration increases from 1g to 10g, the peak open-circuit output voltage increases from 1.88V to 6.92V. However, upon further increment of input acceleration from 10g to 20g, the peak opencircuit output voltage remains at 6.92V and does not increase further. This is beause when the acceleration reaches to a certain value, the deflection of the device is limited by the hard stoppers and therefore, the maximum generated open-circuit voltage is limited at 6.92V. Figure 4: Voltage sensitivity frequency response of both simulation and measurement results. Both the simulated resonant frequency and the simulated Q factor are very close to the measured value. The simulated voltage sensitivity of 2.283V/g is also close to the measurement result of 1.88V/g. Figure 6: The measured output power spectrum of the vacuum-packaged energy harvester with input acceleration from 1g to 20g. A record output power of 468.77 W is shown from the input acceleration of 10g onwards. Upon further increment of the input acceleration from 10g to 20g, the peak close-circuit output power does not increase due to the limitation in vibration amplitude by the mechanical stoppers (cavities on both top and bottom capping wafer), but the bandwidth of the output power increases. The bandwidth of the energy harvester at 20g input acceleration is 71Hz or 3.66%. Figure 5: The measured frequency response of the output open-circuit voltage with input acceleration from 1g to 20g. A record output open-circuit voltage of 6.92V for the vacuum-packaged energy harvester is shown with the input acceleration from 10g onwards. The peak open-circuit output voltage does not increase further after the input acceleration of 10g because the vibration amplitude of the cantilevers are limited by the hard stoppers formed by the cavities on the top and bottom capping wafers. Figure 6 presents a typical output power spectrum frequency response at the different input accelerations. To maximize the output power, an optimal 40k resistor load is used for extracting the output power from the vacuum packaged energy harvester. When the tip of the proof mass touches the bottom of the Si cavities, the device outputs a record power of 468.77 W. Again, upon further increment of the input acceleration from 10g to 20g, the peak close-circuit 736
output power does not increase and remains at 468.77 W. However, the output bandwidth of the fabricated energy harvester increases with the increment of the input acceleration from 10g to 20g, with the measured 3dB bandwidth of the EH at 20g acceleration being 71Hz or 3.66%. photograph of the power management circuit depicted in the inset. A 100 F capacitor is used to store the energy a supply DC output voltage. From 0V to around 1.7V, the circuit works in cold startup stage and the boost circuit draws power from the input. Beyond 1.7V, the circuit works in normal operation stage and the main boost circuit takes over the charging function and draws power from the storage capacitor. Using MPPT (maximum power point tracking) algorithm, the main boost circuit modulates the input impedance and improves the power transfer efficiency. Starting from 0V, after 67.23s, the output voltage of the storage capacitor reaches 3.3V, and then the boost circuits are disabled to protect the storage element. Figure 7: The measurement results of the energy harvester with power management circuit: (a) the harmonic open-circuit output voltage with a peak value of 6.14V with 6g sinusoidal input vibration and (b) DC output from the power management circuit. A 100 F capacitor is used to store the energy could supply 3.3V DC output. Figure 7 shows the measurement results of the energy harvester with power management circuit with 6g sinusoidal input vibration. Figure 7 (a) shows the harmonic open-circuit output voltage with a peak value of 6.14V which will load to the circuit and Figure 7 (b) shows charging process using the fabricated EH with 6g sinusoidal input vibration, with the optical Figure 8: The measured frequency response of the output open-circuit voltage upon undergoing various number of thermal cycles for the first tested device. The resonant frequency of the vacuum packaged EH before undergoing thermal cycling test is 926.030501 Hz, indicated as ref in the figure. After 160 cycles of thermal cycling test, the measured resonant frequency of the vacuum packaged EH is 926.576847 Hz, rendering 590ppm of frequency shift upon 160 thermal cycles. The resonant frequency of the vacuum packaged EH is measured to be 926.741131 Hz after 295 cycles of thermal cycling test, indicating the frequency shift of 767ppm upon 295 thermal cycles. It is to be noted that from 160 cycles to 295 cycles, the frequency shift is only 177ppm, indicating that the device performance is almost fixed after initial thermal cycling. Therefore, thermal annealing after the vacuum packaging of the fabricated EH would help with the frequency stability and long-term reliability. 737
Figure 9: The measured frequency response of the output open-circuit voltage upon undergoing various number of thermal cycles for the second tested device. The resonant frequency of the vacuum packaged EH before undergoing thermal cycling test is 929.181478 Hz, as indicated by ref in the figure. After 160 cycles of thermal cycling test, the measured resonant frequency of the vacuum packaged EH is 929.72339 Hz, rendering 583 ppm of frequency shift upon 160 thermal cycles. The resonant frequency of the second vacuum packaged EH is measured to be 929.73065 Hz after 295 cycles of thermal cycling test, indicating the frequency shift of 591 ppm upon 295 thermal cycles. It is to be noted that from 160 cycles to 295 cycles, the frequency shift is only 7.8 ppm, indicating that the device performance is extremely stable after initial thermal cycling. Therefore, thermal annealing after the vacuum packaging of the fabricated EH would help with the frequency stability and long-term reliability. Thermal cycling was carried out to determine the reliability and the long-term stability of the in-house fabricated and vacuum packaged EH under various conditions. The EH was characterized prior to placement in an oven with ramping temperature control for thermal cycling test. A single cycle of thermal cycling test includes the following temperature condition: firstly, the device is held at 125 C for 15 minutes, after that the temperature of the oven is ramped down -40 C in 11 minutes. Then, the temperature of the oven is dwelled at -40 C for another 15 minutes, before finally ramped up to 125 C to start a new cycle. Two vacuumed packaged devices were characterized by obtaining the open circuit voltage when the EH were excited with a 1g acceleration from 850 Hz to 1000 Hz, and the experimentally measured results are summarized in Figure 8 and Figure 9. Figure 8 and Figure 9 show the measured frequency response of the output open-circuit voltage upon undergoing various number of cycles of thermal cycling testing for the two vacuum packaged devices, respectively. For the first vacuum packaged device, the resonant frequency before undergoing thermal cycling test is 926.030501 Hz. After 160 cycles of thermal cycling test, the measured resonant frequency of the vacuum packaged EH is 926.576847 Hz, rendering 590ppm of frequency shift upon 160 thermal cycles. The resonant frequency of the vacuum packaged EH is measured to be 926.741131 Hz after 295 cycles of thermal cycling test, indicating the frequency shift of 767ppm upon 295 thermal cycles. It is to be noted that from 160 cycles to 295 cycles, the frequency shift is only 177ppm, indicating that the device performance is almost fixed after initial thermal cycling. For the second vacuum packaged device, the resonant frequency before undergoing thermal cycling test is 929.181478 Hz. After 160 cycles of thermal cycling test, the measured resonant frequency of the vacuum packaged EH is 929.72339 Hz, rendering 583 ppm of frequency shift upon 160 thermal cycles. The resonant frequency of the second vacuum packaged EH is measured to be 929.73065 Hz after 295 cycles of thermal cycling test, indicating the frequency shift of 591 ppm upon 295 thermal cycles. It is to be noted that from 160 cycles to 295 cycles, the frequency shift is only 7.8 ppm, indicating that the device performance is extremely stable after initial thermal cycling. Therefore, for both vacuum packaged devices, thermal annealing after the vacuum packaging of the fabricated EH would help with the frequency stability and longterm reliability. V. CONCLUSIONS In conclusion, a wafer-level vacuum-packaged AlN MEMS EH with power management circuit has successfully been demonstrated with high Q-factor and high-g survivability, and superior reliability and longterm stability, thanks to the design and novel packaging and design scheme, which utilizes two-step three-wafer wafer level eutectic bonding. The vacuum packaged EH, which achieves 468.77 W power output and 71Hz 3dB bandwidth with a small footprint of ~0.119cm 3 (1 0.7 0.1705cm 3 ) and a power density of 3.93 738
mw/cm 3, is a promising candidate for next-generation battery-less smart tire TPMS applications. Furthermore, the whole integration process, including the wafer-level vacuum-packaging process, is CMOS compatible, making the reported EH viable for mass production with low fabrication cost. ACKNOWLEDGMENT This work was funded by Mubadala Development Company (Abu Dhabi), Economic Development Board (Singapore) and GLOBALFOUNDRIES Singapore under the framework of 'Twinlab' project with participation of A*STAR Institute of Microelectronics (Singapore), Masdar Institute of Science and Technology (Abu Dhabi) and GLOBALFOUNDRIES- Singapore. REFERENCES [1] Julian W. Gardner, Microsensors: Principles and Applications, John Wiley & Sons Ltd., 1994. [2] Massood Tabib-Azar, Microactuators: Electrical, Magnetic, Thermal, Optical, Mechanical, Chemical and Smart Structures (Electronic Materials: Science & Technology), Springer. (1997). [3] Anna Hac, Wireless Sensor Network Designs, John Wiley & Sons. (2003) [4] Wang Q M, Yang Z C, Li F and Smolinski P, Analysis of thin film piezoelectric microaccelerometer using analytical and finite element modeling, Sens. Actuators, A, 113, 1-11 (2004) [5] Shu Y C and Lien I C, Analysis of power output for piezoelectric energy harvesting systems, Smart Mater. Struct. 15, 1499 (2006). [6] Ajitsaria J, Choe S Y, Shen D and Kim D J, Modeling and analysis of a bimorph piezoelectric cantilever beam for voltage generation, Smart Mater. Struct. 16, 447 (2007). [7] Anton S R and Sodano H A, A review of power harvesting using piezoelectric materials (2003 2006), Smart Mater. Struct. 16, R1- R21 (2007) [8] Ng T H and Liao W H, J. Sensitivity analysis and energy harvesting for a self-powered piezoelectric sensor, Intell.Mater. Syst. Struct. 16, 785 97 (2005) [9] Gurav S P, Kasyap A, Sheplak M, Cattafesta L, Haftka R T, Goosen J F L and Van Keulen F, Uncertainty-based design optimization of a micro piezoelectric composite energy reclamation device, Proc. 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conf. 3559 70 (2004) [10] N. Wang, L. Y. Siow, H. Ji, P. Chang, Q. Zhang, C. Sun, and Y. Gu, AlN Wideband Energy Harvesters with Wafer-Level Vacuum Packaging Utilizing Three-Wafer Bonding, 30th International Conference on Micro Electro Mechanical Systems (IEEE MEMS 2017), Las Vegas, USA, 22-26 Jan. 2017. 739