Low-Frequency Vibrational Energy Harvesting at the Micro and Meso Scale

by

Haluk Akay

Bachelor of Science, Mechanical Engineering Carnegie Mellon University (2016)

Submitted to the Department of Mechanical Engineering in partial fulfillment of the requirements for the degree of Master of Science in Mechanical Engineering at the Massachusetts Institute of Technology June 2018

Massachusetts Institute of Technology 2018. All rights reserved. Signature redacted Author Department o>echanical Engineering May 19, 2018

Certified by Signature redacted Sang-Gook Kim Professor of Mechanical Engineering Thesis Supervisor

Accepted by Signature redacted Rohan Abeyaratne Chair, Department Committee on Graduate Theses MASSACHUSETS INSTITUTE OF TECHNOLOGY

JUN 2 5 2018 LIBRARIES 2 Low-Frequency Vibrational Energy Harvesting at the Micro and Meso Scale

by Haluk Akay

Submitted to the Department of Mechanical Engineering on May 19, 2018 in partial fulfillment of the requirements for the degree of Master of Science in Mechanical Engineering

Abstract

Energy harvesting from environmentally available vibrations is a solution to providing electric power for remote and mobile applications, such as the Internet of Things. A buckled beam-based MEMS device can harvest power from low frequency, low amplitude vibrations which has never been achieved by a micro-scale device. Due to the poor fabricated quality of the active piezoelectric material in the device, however, the generated power has been an order of magnitude less than expected. It is important to harvest more than 1 micro Watt at the device size smaller than a coin in order to implement this technology for real world application. The buckled beam vibrational energy harvesting device has been evaluated for its power generation performance and its specific microfabrication recipe with respect to piezoelectric materials has been analyzed from a process control standpoint to identify areas of improvement. The process has been redesigned for a simpler and streamlined recipe. Complex residual stress control feedback loops have been replaced with a simple post-fabrication assembly to induce controlled buckling in active MEMS beams. Using a custom-designed fixture, the post-fabrication buckling concept has been demonstrated to achieve accurately buckled beams.

With regard to mesoscale energy harvesting, a product has been designed that converts regular human walking motion to electricity. The device harvests electric power using air bulbs, distributed in the sole of a shoe to drive a series of micro-turbines connected to small DC motors. The number and position of air bulbs is optimized to harvest the maximum airflow from each foot-strike. The system is designed to continuously drive the micro- turbines by utilizing both outflow and inflow from the air bulbs. A prototype combat boot was fitted on the right foot of a 75kg test subject, and produced an average continuous power on the order of lOs of mW over a 22Q load during walking at 3.0 mph. This combat boot provides enough electric power to a passive GPS tracker that periodically relays geographical coordinates to a smartphone via satellite.

Thesis Supervisor: Sang-Gook Kim Title: Professor of Mechanical Engineering

3 4 Acknowledgements

I would like to acknowledge and thank my advisor Professor Sang-Gook Kim for his mentorship and encouragement. I also need to recognize and thank Dr. Ruize Xu, my former group-mate who preceded me on this energy harvesting project for the knowledge he lent me before graduating. I would like to thank the MTL staff, especially Dennis Ward, for their help and exceeding any expectations one could have in providing advice and help while I learned the microfabrication process. Finally, I would like to thank everyone in my personal life who are constantly helping to push me ahead.

5 6 Table of Contents

Chapter 1 Introduction...... 12 1.1 Thesis Objective...... 12 1.2 Thesis Organization...... 12 Chapter 2 Low-Frequency Vibration MEMS Energy Harvesting ...... 14 2.1 Background ...... 14 2.2 Device Design and Dynamic Modeling of Low Frequency Energy Harvesting...... 18 2.3 Piezoelectric Performance of Previous Generation Device ...... 21 2.5 Chemical Vapor Deposition Process Control...... 30 2.5.1 CVD Process Outline ...... 31 2.5.2 Experimental Design...... 32 2.5.3 Testing and Experimental Results ...... 34 2.5.4 Data Analysis...... 36 Chapter 3 Post-Fabrication Buckling Design ...... 39 3.1 M otivation...... 39 3.2 Design Goals ...... 41 3.3 M odeling...... 45 3.3.1 Analytical Buckled Beam Model ...... 45 3.3.2 Experimental Buckled Beam Model...... 49 3.4 Design Paths...... 51 3.4.1 Achieving Buckling without Frame Modification...... 52 3.4.2 Achieving Buckling by Frame Modification ...... 55 3.5 Proposed Designs ...... 56 3.5.1 Experimental Frame Bending Fixture...... 56 3.5.2 Snap-Fit Corrective Frame Fixture Design and Process ...... 59 3 .6 T estin g ...... 6 3 3.6.1 Fixture Construction ...... 63 3.6.2 R esults ...... 6 6 Chapter 4 Energy Harvesting Footwear ...... 71 4.1 Introduction to W earable Energy Harvesting...... 71 4.2 M esoscale Energy Harvesting System Design ...... 73 4.2.1 Air Bulb Design and Placement...... 73 4.2.2 Turbine Enclosure Geometry ...... 76 4.3 Theoretical Calculations of Upper Bound Power ...... 77

7 4.4 GPS - Equipped Combat Boot Prototype...... 79 4 .5 Resu lts ...... 8 2 4.6 Discussion...... 85 Chapter 5 Sum m ary...... 87 5.1 Thesis Summary...... 87 5.2 Future work ...... 88 Appendix A 90 A. 1 PECVD Process Capability Study Data ...... 90

A.2 Electric Testing of Experimental Frame Bending Fixture...... 93 Bibliography 94

8 List of Figures

FIGURE 2 - 2 BUCKLED BEAM ENERGY HARVESTER SCHEMATIC [1]...... 17 FIGURE 2 - 3 MASS-SPRING-DAMPER M ODEL SCHEMATIC ...... 19 FIGURE 2 - 4 CLAMPED-CLAMPED BUCKLED BEAM ENERGY HARVESTER...... 20 FIGURE 2 - 5 POLARIZATION - ELECTRIC FIELD CURVE (R. Xu 2018) [1] ...... 23 FIGURE 2 - 6 JEON RECIPE, CRACKED PZT ...... 26 FIGURE 2 - 7 M ITSUBISHI RECIPE, CRACKED PZT ...... 26 FIGURE 2 - 8 XRD OF ANNEALED PZT (R. Xu BASELINE RECIPE) [1]...... 27 FIGURE 2 - 9 POLARIZATION CURVE FROM FABRICATION WITH NEW SOL-GEL ...... 28 FIGURE 2 - 10 POLARIZATION OF VARIOUS PZT FROM OUR RESEARCH GROUP...... 29 FIGURE 2 - 11 CVD PROCESS BLOCK DIAGRAM ...... 31 FIGURE 2 - 12 EXPERIMENTAL DESIGN GEOMETRY ...... 33 FIGURE 2 - 13 CVD MEASUREMENT SAMPLING ...... 35 FIGURE 2 - 14 FILM THICKNESS OVER THREE CVD RUNS WITH IDENTICAL PARAMETERS...... 35 FIGURE 2 - 15 CVD AVERAGE SILICON DIOXIDE FILM THICKNESS ...... 36 FIGURE 2 - 16 AVERAGE THICKNESS AND VARIANCE FOR COLD-START RUNS...... 37

FIGURE 3 - 1 RESIDUAL STRESS CONTROL FEEDBACK LooP [1]...... 40 FIGURE 3 - 2 CROSS-SECTIONAL VIEW OF BUCKLED BEAM DEVICE ...... 42 FIGURE 3 - 3 CROSS-SECTIONAL VIEW OF SIMPLIFIED PIEZOELECTRIC ENERGY HARVESTER ...... 44 FIGURE 3 - 4 MONOLITHIC MEMS ENERGY HARVESTER DEVICE GEOMETRY ...... 45 FIGURE 3 - 5 CLAMPED-CLAMPED BEAM SCHEMATIC ...... 46 FIGURE 3 - 6 BUCKLED CLAMPED-CLAMPED BEAM SCHEMATIC...... 46 FIGURE 3 - 7 BUCKLED CLAMPED-CLAMPED BEAM MODEL ...... 47 FIGURE 3 - 8 1:35 SCALE MODEL OF BUCKLED BEAM DEFLECTION ...... 50 FIGURE 3 - 9 TOPSIDE VIEW OF MEMS ENERGY HARVESTING DEVICE...... 51 FIGURE 3 - 10 FRAME BUCKLING VERSUS FRAME BENDING ...... 52 FIGURE 3 - I 1 INNER FACE OF FRAME UNDER COMPRESSION ...... 54 FIGURE 3 - 12 FRAME BENDING FIXTURE SCHEMATIC (SIDE VIEW) ...... 57 FIGURE 3 - 13 ALUMINUM BEAM - DEVICE FRAME INTERFACE (NOT TO SCALE)...... 57 FIGURE 3 - 14 ISOMETRIC RENDERING OF EXPERIMENTAL FRAME BENDING FIXTURE...... 58 FIGURE 3 - 15 SIDE RENDERING OF EXPERIMENTAL FRAME BENDING FIXTURE ...... 58 FIGURE 3 - 16 SNAP-FIT SCHEMATIC ...... 59 FIGURE 3 - 17 SNAP-FIT M ODEL FREE BODY DIAGRAM...... 60 FIGURE 3 - 18 EXPLODED VIEW OF SNAP-FIT FIXTURE ...... 62 FIGURE 3 - 19 CROSS SECTIONAL VIEW OF SNAP-FIT FIXTURE ...... 62 FIGURE 3 - 20 DETAIL CROSS-SECTION VIEW OF DEVICE / MOLD INTERFACE...... 63 FIGURE 3 - 21 FRAME BENDING FIXTURE...... 64 FIGURE 3 - 22 FRAME BENDING FIXTURE, SIDE VIEW ...... 64 FIGURE 3 - 23 Two POSITIONS OF BUCKLING OBSERVED ON FRAME BENDING FIXTURE ...... 65 FIGURE 3 - 24 PROFILOMETER MEASUREMENT OF BEAM BUCKLING ...... 67 FIGURE 3 - 25 ACCELERATION OF PROOF MASS, FREQUENCY DOMAIN RESPONSE ...... 68 FIGURE 3 - 26 DEVICE ELECTRIC FUNCTION VERIFICATION (OPEN CIRCUIT VOLTAGE)...... 69

FIGURE 4 - 1 SCHEMATIC OF FOOTWEAR ENERGY HARVESTING SYSTEM ...... 74 FIGURE 4 - 2 PEAK POWER RELATIVE TO LENGTH ALONG SHOE...... 75

9 FIGURE 4 - 3 RADIAL SYMMETRY OF DUAL TURBINE ENCLOSURE ALLOWS FOR BI-DIRECTIONAL FLOW TO CONTINUOUSLY DRIVE GENERATORS...... 76 FIGURE 4 - 4 PROTOTYPE COMBAT BOOT FITTED WITH ENERGY HARVESTING SYSTEM...... 80 FIGURE 4 - 5 PEAK POWER OUTPUT VERSUS VALUES OF LOAD RESISTANCE ...... 82 FIGURE 4 - 6 VOLTAGE OUTPUT MEASURED FOR 0.5 Hz FOOTSTEPS ...... 83 FIGURE 4 - 7 VOLTAGE OUTPUT MEASURED FOR 1.0 Hz FOOTSTEPS ...... 83 FIGURE 4 - 8 VOLTAGE OUTPUT MEASURED FOR LUNGES...... 84

10 List of Tables

TABLE 2 - 2 PZT SPIN-COATING STANDARD OPERATING PROCEDURE ...... 24 TABLE 2 - 3 PZT RECIPE COM PARISON ...... 25 TABLE 2 - 4 SUMMARY OF FACTORS AND LEVELS...... 34 TABLE 2 - 5 SIGN EFFECTS FOR 22 DESIGN [15]...... 34 TABLE 2 - 6 ANALYSIS OF VARIANCE FOR FACTORIAL EXPERIMENT...... 38

TABLE 3 - 1 FUNCTIONAL LAYERS OF BUCKLED BEAM ENERGY HARVESTER ...... 43 TABLE 3 - 2 PREDICTED FRAME WIDTH DISPLACEMENT BASED ON EXPERIMENTAL MODEL ...... 50

TABLE 4 - 1 POWER GENERATED FOR VARIOUS MODES OF MOVEMENT ...... 84

11 Chapter 1 Introduction

1.1 Thesis Objective

The objective of this thesis is to harvest energy at the MEMS scale from low-frequency vibrations below 100 Hz. This operating range which has been identified as the key to implementing microelectromechanical systems (MEMS) energy harvesters for commercial application.

The microfabrication process design involved in producing a bi-stable buckled beam-based vibrational MEMS energy harvester is analyzed and simplified by removing nonfunctional design elements included purely for residual stress balancing and moving buckling operations post-process. Fixtures and methods of achieving post-fabrication modification of MEMS devices have been designed, built, and tested for this purpose.

At the meso-scale, an energy harvesting system embeddable in footwear has been designed to convert the impact of footsteps to usable electric power by way of airflow-driven miniaturized turbines connected to small DC generators. A prototype system has been installed in a combat boot and powers a GPS receiver that can provide information of the user's location upon command.

1.2 Thesis Organization

This thesis is divided into two parts, based on separate efforts to harvest energy at low frequencies. The first part addresses energy harvesting at the microscale using a bi-stable buckled beam MEMS device developed by our group which targets vibration energy harvesting at low frequencies and low amplitude. The second part presents a solution to harvesting energy at the mesoscale from human footstep motion utilizing a turbine system embedded in footwear, driven by airflow generated by the compressive force of the user's

12 foot inside the shoe. The first chapter provides a brief outline of energy harvesting and the objective and organization of the thesis.

The second chapter analyzes the microfabrication process used to produce our group's most recent MEMS device and identifies opportunities for improvement through analysis of process capability and detailed analysis of the fabrication methodology. A higher quality device from a piezoelectric standpoint is fabricated with material properties at par with previous generation devices from our research group. The complexities of the stress control feedback process are highlighted.

The third chapter presents a post-fabrication buckling fixture solution to achieve buckled MEMS thin films after microfabrication. An analytical model of the MEMS frame buckling is developed. A meso-scale model of the buckled MEMS membranes is used to predict range of moment needed in the frame to induce buckling. Different methods of inducing buckling post-process are evaluated, and an experimental frame bending fixture is designed, built, and tested and validated for inducing the correct amount of buckling while preserving electrical function. A snap-fit fixture design is proposed as a production appropriate evolution of the frame bending concept.

The fourth chapter presents energy harvesting footwear. The design is optimized by number and placement of air bulbs within the shoe sole. The turbine enclosure is designed to accept airflow from both outlet and inlet such that the turbines are spinning nearly continuously. A GPS receiver powering prototype is built to demonstrate the function of the system.

13 Chapter 2 Low-Frequency Vibration MEMS Energy Harvesting

2.1 Background

Energy harvesting has played a pivotal role in economic development by enabling agricultural and industrial processes to draw upon the natural environment for their energy needs. Thousands of years ago, water wheels were used to grind wheat into flour. More recently, geothermal heat energy has been used to drive generators to produce usable electric power. With the development of microelectromechanical systems (MEMS), the size of cutting-edge energy harvesting systems has been dramatically reduced to the point where a fully packaged device can be about the dimensions of a large coin.

With the miniaturization of energy harvesting, the source of energy upon which these devices draw for power has changed. Precision microfabrication techniques have enabled the design of devices with critical feature dimensions on the order of nanometers. Such devices can be fabricated from novel materials with piezoelectric, electrostatic, ferromagnetic, and other properties which allow for energy harvesting of naturally occurring vibrations.

The aim of vibrational energy harvesting is to scavenge usable electric energy from naturally occurring vibrations in the ambient environment. One method of transforming vibrational kinetic energy into electricity is to exploit the properties of piezoelectric materials for this purpose. Piezoelectric materials not only deform mechanically when a potential difference is applied across them, but also conversely develop surface electric charge when strained [1]. Vibrational energy harvesting devices using piezoelectric materials are designed to resonate in the frequency range of a targeted ambient vibration such that repeated deformation of the piezoelectric material results in electric charge.

14 Piezoelectric energy harvesting devices can be generally separated into two categories. Bulk piezoelectric devices utilize piezoelectric materials on the mesoscale and scavenge energy from large deformations caused by mechanical impact or other substantial impulses [2]. Large mass of the active piezoelectric material translates to greater surface charge and more electric power generated. Such devices do require significant kinetic energy input to work, however.

The latter category of piezoelectric energy harvesting devices falls in to the family of MEMS devices and utilizes thin-film membranes of piezoelectric materials with thicknesses ranging on the orders from 101 to 10' nanometers. Depending on the thickness and other device-specific geometric designs, these membranes resonate on a wide spectrum of ambient vibration frequencies to produce electricity.

Piezoelectricity and the molecular structure of piezoelectric materials is an important topic to understand in order to address the challenges of fabricating MEMS devices with active piezoelectric components. The mechanical deformation to surface charge piezoelectric effect is found in select materials with non-centrosymmetric atomic structures [3]. In a specific temperature range, dipoles in piezoelectric materials align to create a net polarization within smaller regions of the material known as Weiss domains [4]. In order to truly produce a net polarization of the material, all the Weiss domains must be aligned which can be executed by applying a high electric field at a high temperature point in a process termed "poling."

An example of a piezoelectric material with the stated properties described above is lead zirconate titanate, abbreviated as (PZT). PZT has a polycrystalline, ceramic perovskite structure [3]. PZT has high piezoelectric coupling when compared to other piezoelectric materials, and so is favored for energy harvester fabrication where small amounts of strain must be translated to the largest electric polarization possible [6].

15 When considering potential applications for MEMS energy harvesting devices, both the power requirements of the application and the "niche" benefit of using a MEMS device for this purpose must be taken into account. The so-called "low-hanging fruit" among applications when using these criteria are low-power sensors used to monitor variables such as temperature or flow rate in remote locations. Not only are such sensors being designed for increasingly low power consumption needs, but the need for such constant monitoring often arises in remote locations such as oil pipelines or difficult-to-access situations such as long stretches of piping in refineries which can extend for kilometers.

While evaluating the application of MEMS energy harvesters, it is also important to take a survey of the available vibrations occurring in the potential use environment. The frequency range of many vibrational energy harvesting opportunities falls below 100 Hz for environments such as pipe flow, and motors. For this reason, the target operating frequency of a MEMS energy harvesting device should fall in this range.

Our research group has been developing such thin-film piezoelectric resonators since 2005, with the intention of designing a device that harvests energy from low frequency, low amplitude vibrations on a large operating bandwidth. However, these past energy harvesters operated at a frequency on the order of 104 Hz and at a high amplitude of 4 g [7]. Various efforts using magnets integrated into mechanical assemblies [8] and utilizing soft materials [9] have been reported to lower the operating frequency range of energy harvesters. A method of note to lower operational frequency is to design a nonlinear resonance-based energy harvester with two states of stability [10]. This concept was demonstrated experimentally by way of a mesoscale piezoelectric film supported by a steel sheet. None of these efforts to lower the operational frequency range had been applied to MEMS scale design or microfabricated and tested until a buckled beam energy harvesting MEMS device developed by Xu [1,13]. This device applied the concept of a bi-stable buckled beam design at the MEMS scale to harvest energy from low frequency, low amplitude vibrations. A precisely targeted deflection of 200pm due to buckling was

16 induced by balancing residual stresses in multiple microfabricated material layers in the device. This enabled an array of buckled piezoelectric beams to snap between the two states of stability during excitation. The fabricated monolithic device proved this concept and exhibited the intended behavior during dynamic testing on a mechanical shaker test rig. However, the power generated (85 nW) was an order of magnitude less than what was theoretically predicted, indicating opportunity for improvement.

Figure 2 - 1 Buckled Beam Energy Harvester Schematic [1]

A MEMS device such as the one designed by Xu is fabricated in a clean-rom facility using chemical vapor deposition or sol-gel spin-coating to create uniform material coats on a silicon substrate, and pattern each layer iteratively using photolithography to build the device layer by layer. For a sophisticated device such as the energy harvester designed by Xu, there is added complexity in the microfabrication process because of multiple design requirements that need to be achieved.

The key deliverable of the buckled beam energy harvesting device is the quantity of electric power outputted for an input vibration below 100 Hz and at an acceleration of 1 g. The target value for the RMS power output is on the order of micro-Watts, so that the device can be integrated with a low-power sensor and transmit data through a nodal network without an external power supply. In reality, when the initial batch of devices was fabricated, a peak output power of just 85 nW was measured for testing at 70 Hz and 0.5 g.

17 By evaluating the performance of each of piezoelectric layer and dynamic performance of the device, opportunities to increase the power output through better process control during microfabrication can be identified.

2.2 Device Design and Dynamic Modeling of Low Frequency Energy Harvesting

Vibrational energy harvesting is based on the principle of translating a vibration into usable electric power. As discussed previously, materials with piezoelectric properties are able to experience mechanical strain and create an electric potential difference. In order to induce stain in the active piezoelectric material of the energy harvester, the device structure must be coupled as directly as possible to the stated vibration. The resultant vibrational displacement to the energy harvester can be described using the equation (1) below.

z(t) = zosin (ot) (1)

The model of a vibrational energy harvester can be considered as a mass-spring damper system, the frame of which is directly coupled to the oscillations of the source of vibrations, as shown in Figure 2-3 [1].

18 I F-W 4

k (Spring constant) x(t)

m (Mass) -

b (Damping i.....- z(t)t constant)

Figure 2 - 2 Mass-Spring-Damper Model Schematic

For such a simple mass-spring-damper system described, the system can be modeled using equation 2, with the input vibration z(t) described in equation 1.

m2 + mi + bi + kx = 0 (2)

For such a system, the linear resonant damped frequency can be expressed using equation 3.

b2 0D= k -kb (3)

As shown in equation (3), the natural frequency of such an energy harvester is inversely proportional to its oscillating mass. Early iterations of MEMS piezoelectric energy harvesters featured micro-scale cantilevered beams [12], but these devices had operational frequencies on the order of 104 Hz.

19 The buckled beam energy harvester designed by Xu can be modeled as a doubly clamped beam with proof mass mounted midway along its length, as shown in Figure 2-4. This model has both linear and nonlinear stiffness components.

Piezoelectric layer

InterdigiatedElectrodes

Figure 2 - 3 Clamped-Clamped Buckled Beam Energy Harvester

For a nonlinear system, the natural undamped frequency can be expressed as a function of nonlinear stiffness and linear stiffness [14].

1 kL+keq 1 kL+4c2kN (tN 2T m 27r M(4

In equation 4 ke,, is the equivalent linear stiffness, and 6 is the maximum deflection. The nonlinear system can be modeled by the characteristic bi-stable Duffing equation 5 [11], where kL and kN are the linear and nonlinear beam stiffness, cl. and CN are the linear and nonlinear electromechanical coupling.

m2+m +b+ kLX + kNX3 +CNxVN + CLVL = 0 (5)

The expressions for both linear and nonlinear stiffness are expressed below, and are functions of the beam geometry where W, H, and L are the width, height, and beam length, cE is the elastic constant, and T is the residual stress [1,8]. The stiffness is a lumped sum across i number of material layers in the device.

20 kL = i=1CE,(HU 3 - H + [ Z~ t 1 Toii] (6)

kN 8 1=1 CE,iHi (7)

Linear stiffness k, is broken into two components for beam bending and compressive residual stress in the layer, which has a negative sign and at large magnitudes will render the entire k term negative, modeling the bi-stability of the system through equation (5). When linear stiffness changes sign, the frequency response of the system becomes lower. At low frequency responses, the model predicts a large amplitude shift between the two regions of stability. This bi-stable nonlinear oscillation is the key dynamic response to implementing low-frequency energy harvesting at the MEMS scale.

2.3 Piezoelectric Performance of Previous Generation Device

The piezoelectric performance is tied to the thickness and quality of the PZT deposited and patterned on the wafer. Thicker coats of PZT are desirable as a more massive active piezoelectric layer translates to larger charge accumulation under the levels of strain experienced by the device during vibration testing. PZT is deposited on the wafer in the form of a solution-gel, or sol-gel, and spin-coated to create a uniform film of 70 - 80 nm thickness per coating. The film is pyrolyzed on a hot plate and then photo-lithography technique is used to pattern the 2D surface profile desired for the PZT layer. The unwanted areas are then etched away and the wafer is annealed at high temperature to complete the process. The target number of layers is three, aiming for a total of 240 nm of PZT in the device. However, the number of layers is process-limited. Residual compressive internal stresses in the PZT are the highest for any layer in the device, at approximately 650 MPa. In addition, multiple coatings may prevent healthy formation of the PZT in the perovskite crystalline structure, as described in section 2.1, needed to attain the desirable piezoelectric coefficient.

21 One method of quantifying the piezoelectric properties of a fabricated energy harvesting device is observing the remnant polarization and saturation polarization while applying a large electric field at high temperature in a process known as "poling." This aligns the Weiss Domains and primes the PZT for peak piezoelectric performance.

Remnant polarization Pr has units of charge per unit of square area and is a measure of the orientation of Weiss Domain dipoles that are still aligned even after the large electric field applied during poling is removed. Saturation polarization P, has the same units of polarization and indicates the maximum level of polarization reached at the highest electric field.

As stated before, the most recent energy harvester fabricated by R. Xu (2018) with the buckled beam design was not able to generate more than 85 nW of peak power, despite showing the targeted buckling deflection in the piezoelectric beams and desired nonlinear resonance under 100 Hz. Upon further evaluation it was identified that the PZT quality was not optimal. This was evident from the Polarization-Electric Field curve obtained when poling the device, shown in Figure 2-5. The remnant polarization was measured to be just 3.75 pC / cm 2 [1].

22 30

20- C4

10-

0- 0

-10-

-20

-30 -1500 -1000 -500 0 500 1000 1500 Electric Field (kV/cm)

Figure 2 - 4 Polarization - Electric Field Curve (R. Xu 2018) [1]

Previous work with piezoelectric materials, specifically PZT has yielded better performance in the past from our research group. In 2004, Jeon reported remnant polarization of 20 pC / cm 2 [cite] and others have replicated this quality of PZT on the same order of magnitude, as compared in Figure 2-10.

In order to support the work of R. Xu to successfully fabricate the buckled-beam energy harvester design such that it delivered sufficient target power on the order of micro-Watts, the PZT baseline coating recipe parameters were compared with other recipes for validation. The parameters of interest were specifically related to the coating of PZT on the substrate. The generalized procedure for spin-coating PZT is shown below in Table 2-2. Spin-coating and subsequent patterning of the PZT is done in the Photolithography room of the Technology Research Laboratory (TRL) at MIT's Microsystems Technology Laboratories (MTL).

23

Table 2 - 1 PZT Spin-Coating Standard Operating Procedure

Step Machine Name Operation Parameters 1 Spin-Coater Statically dispense PZT Ramp speed as 500 rpm sol-gel on substrate, for 5 sec, 3500 rpm for followed by spin- 30 sec coating 2 Hot Plates Pyrolyze PZT on hot Ramp temperature as plates 180'C for 1 min, 390-C for 5 min, 180'C for 1min 3 HMDS, Coater Coat with photoresist N/A 4 EVJ Expose N/A (Photolithography) 5 Acid-Hood Etch PZT N/A 6 RTA-Annealing Anneal PZT 700'C for 60 seconds

While steps 3-5 are relatively well-defined processes of photolithography, the coating, pyrolyzing, and annealing of PZT is less understood as a process. Therefore, three recipes were compared with the goal of performing X-ray crystallography (XRD) on the three results in order to observe crystalline structure and intensity peaks that correlate to correctly crystallized PZT. The first recipe used the baseline parameters detailed in Table 2-2. The second recipe was that used by Jeon to create PZT which had reported remnant polarization of 20 pC / cm 2 . The third recipe was that recommended by the manufacturer of the PZT sol-gel, Mitsubishi Materials. These three recipes all called for different process parameters and even different material layer stacks following deposition of silicon oxide by chemical vapor deposition. For example, the Jeon Recipe excludes PT from the material stack. Table 2-3, below, details the parameters for each of the three tested recipes.

24 'I

Table 2 - 2 PZT Recipe Comparison

Recipe Step/ Xu (Baseline) [11 Jeon (2004) [121 Mitsubishi Materials Parameter Post - CVD n/a Anneal in furnace at Anneal in furnace at Annealing 750'C, 30 min 7500 C, 30 min ZrO2 Pyrolysis Bake in furnace at Bake on hotplate at Bake on hotplate at 100*C for 1min, on 350'C for 1 min, in 350C for 1 min, in hotplate at 300'C furnace (add oxygen) furnace (add oxygen) for 5 min, on ramp to 700C for 15 ramp to 700C for 15 hotplate at I 000C min, ramp down to min, ramp down to for 1 min 100 0C 1000C PT Coating Coat 2 Layers PT n/a Coat 2 Layers PT, as per baseline PZT Coating 500 RPM for 3 sec 500 RPM for 3 sec 500 RPM for 3 sec (3 Coats) 3500 RPM for 30 1500 RPM for 30 sec 3000 RPM for 30 sec sec PZTPyrolysis 180C for I min 180C for 1 min 180C for 1 min 390C for 5 min 350C for 5 min 350C for 5 min 180'C for I min 180C for 1 min 180C for 1 min PZT Annealing RTA - 700C for 1 Furnace - 700C for RTA - 700*C for I min min 15 min (add oxygen) (add oxygen)

Each recipe was followed for processing a total of three wafers with the intention of performing X-ray crystallography. None of the wafers was patterned using Photolithography because X-ray crystallography involves destructively imaging samples in a way that the wafer could not be used to produce devices later on. This experiment was purely to investigate the PZT crystal structure of each recipe.

The results of the experiment clearly pointed to Xu's baseline recipe as the optimal recipe for depositing this PZT sol-gel. Neither the Jeon recipe nor Mitsubishi Materials recipe were able to produce an imageable wafer. Both experienced failure modes of cracking. Illustrated in Figure 2-6, the Jeon wafer cracked after annealing the second coat of PZT. As illustrated in Figure 2-7, the Mitsubishi wafer cracked after annealing the third coat of PZT into large pieces.

25 -4

Figure 2 - 5 Jeon Recipe, Cracked PZT

Figure 2 - 6 Mitsubishi Recipe, Cracked PZT

The baseline wafer, however, yielded an acceptable surface upon visual observation. As a result, X-ray crystallography was performed on the baseline wafer despite not having other recipes to compare with. Xu performed the XRD imaging, and the resultant plot is shown in Figure 2-8.

26 PZT (110) U ZrO2 150 o A Pb (Zr.52 Ti0.48) O

100 ZrO2 8 PZT (100) 50 0A4:3 PZT (111) PZT (200) A

0A 4

20 30 40 so 60 Position [291 Figure 2 - 7 XRD of Annealed PZT (R. Xu Baseline Recipe) [1]

X-ray crystallography demonstrated a peak at 1-1-0 indicating promising orientation of the PZT crystal structure. Based on this result, the process recipe was not identified as a risk or a direct cause of the poor remnant polarization of previous buckled-beam devices. Upon further evaluation of the process, it was discovered that the PZT sol-gel being used for processing was several months old. Despite being preserved in a refrigerated environment in a sealed container, degradation of the PZT sol-gel quality was identified as a risk factor. As a result, a new container of PZT sol-gel was ordered from Mitsubishi materials, and the process was replicated with the intention of producing better quality PZT. In addition to using a new sol-gel, other process improvements were implemented, as listed below:

* New PZT Sol-gel from Mitsubishi Materials (opened October 2017; previous was opened December 2016) * Pre-baking substrates for 5 minutes at 75'C prior to all sol-gel spin coating operations to dehumidify wafer * Pre-baking substrate for 5 minutes at 1000C prior to all annealing operations to dehumidify wafer

Two wafer's worth of buckled beam energy harvesting devices were fabricated from the new PZT sol-gel. A selected device was poled at 250'C at an electric field of 250 kV/cm

27 for a time of 30 minutes, and then the Polarization-Voltage (P-V) curve was measured using a probe station. The resultant curve is shown in Figure 2-9 below.

4W

Voltage [V]

Figure 2 - 8 Polarization Curve from Fabrication with new Sol-Gel

The remnant polarization of the PZT is Pr is 23.5 pC/CM2, much more significant then when compared to the polarization curve of the previous device which had remnant polarization of just Pr of 3.75 ptC/CM2 in Figure 2-5. In fact, this device's PZT quality is comparable to previous work from other projects involving PZT either for energy harvesting (Jeon, 2004 [7], and Hajati, 2011 [9]) or ultrasonic medical imaging (Smyth, 2016 [2]). A comparison figure of various reported polarization levels from our research group is shown in Figure 2-10.

28 90

80

70

60

50

0 -j 40

$ 30

20

10

0 11 Y.B. Jeon et. al A. Hajati (2011) K. Smyth (2016) R. Xu (2018) Current Iteration* (2004) (2018)

n Remnant Polarization (Pr) U Saturation Polarization (Ps)

Figure 2 - 9 Polarization of Various PZT from our Research Group

Based on the promising results of the remnant polarization Pr in the most recent fabrication effort with new sol-gel and selected process improvements, a higher output power of the device is expected, on the order of micro-watts.

29 2.5 Chemical Vapor Deposition Process Control

Chemical vapor deposition (CVD) is used in the microfabrication process of the buckled beam energy harvester to deposit material for three separate layers. These three layers are the 300 nm thick silicon oxide below the piezoelectric active layers, the 800 nm thick silicon nitride above the piezoelectric active layers, and the 400 nm thick silicon oxide layer which is the topmost on the device.

The key inputs selected when running the CVD process for an intended material thickness are the deposition rate, measured in units of nanometers per minute, and the time of deposition. Other process inputs such as deposition frequency and material are also detailed in the recipe. Theoretically the thickness of the deposited layer is predicted by the following relation in equation 8.

Theoretical thickness [nm] = Depositionrate x Deposition time [mn] (8)

In practice, however, the measured thickness cannot be accurately predicted solely by multiplying these inputs as suggested in equation 8. The specific machine used to run the CVD process is a Surface Technology Systems CVD (STS-CVD) model installed in the Technology Research Laboratory (TRL) of MIT's Microsystems Technology Laboratories (MTL). After repetitive use of this machine, the user becomes familiar with the unpredictability of the deposition process and it is laboratory custom procedure to have to run at least one or two calibration runs with dummy silicon pieces to determine which input parameters will give the desired deposited film thickness. As a part of comprehensively evaluating the microfabrication process for the buckled beam device, the CVD process was examined in detail.

30 2.5.1 CVD Process Outline

In order to thoroughly design a process control study on the STS-CVD machine, a full understanding of its operation is necessary. The STS-CVD operates with a "black-box" model where the user does not interact with the sample at all during processing, which is advantageous in the fact operator-related error is minimized.

For processing a 4" silicon wafer (which is the correct substrate used in buckled beam energy harvesting device fabrication), a step-by-step walk through of the procedure is summarized as follows, and illustrated as a block diagram in Figure 2-11. After logging into the system's digital interface using user credentials, the user vents the loading chamber. Once this chamber reaches atmospheric pressure, it opens and the user manually places the 4" wafer into one of two loading trays. This is one of the few opportunities for operator error, should the wafer top surface contact anything during loading and become scratched or deformed.

Operator Tasks Loading Chamber Reaction Chamber

Wafer automatically Purgedand pumped loaded into Reaction out Chamber

User Loads Wafer into Pumped and mapped Gas stabilization Loading Chamber

Processing

Wafer automatically unloadedinto 4-1 Purgedand pumped Reaction Chamber out

User unloads wafer +- venting

Figure 2 - 10 CVD Process Block Diagram

31 Once the wafer is in the loading chamber, the chamber is pumped and "mapped" so that the machine knows which tray contains the substrate. After the recipe is edited, saved, and selected, the wafer is loaded automatically into the reaction chamber. The reaction chamber is then purged (10 seconds), pumped out (10 seconds), and then experiences 90 seconds of gas stabilization before the process commences according to the loaded recipe. Following completion of the process, the reaction chamber is pumped out (20 seconds), purged (10 seconds), and pumped out again (10 seconds) before the wafer is automatically unloaded into the loading chamber. After venting the loading chamber, the user can remove the processed wafer and the run is complete.

The film thickness can be measured by a nano-spectrometer using the known refractive index of the deposited material. Before processing the actual experimental wafer, calibration runs on dummy cleaved wafer pieces are necessary to determine which input parameters will give the desired thickness of deposition.

2.5.2 Experimental Design

For a comprehensive analysis of the STS-CVD, experimental design is required to identify running which types of tests will give the best understanding of process capability. From a high level, the first element of the process to be investigated is the "cold start" state, by which the machine is used after being idle for an extended duration of time. The aim of investigating this is to determine whether it is necessary to run an empty run purely for the purpose of "warming up" the machine. This was tested by booking the machine on the earliest Monday morning slot available after a weekend without use and running a 50 nm/min x 3 minute process with 10 replicates consecutively three times. Additionally, tests were run for different deposition rates all at a constant 3 minute deposition time, and conversely runs with different deposition times all at a constant 50 nm/min deposition rate were also run.

32 The second element of the process to be investigated is the process sensitivity with respect to changes in deposition rate versus changes in deposition time. For this part of the experimental design, the two factors are chosen to be deposition rate, and deposition time. The response variable for this experiment is the film thickness. The factor levels were selected as 50 (-) and 100 (+) nm / min for deposition rates and 2 (-) and 4 (+) minutes for deposition times, with 5 replicates each. This is a 22 factorial design testing the two factors at two levels each, illustrated in Figure 2-12.

High (+ ------b ab

B - Deposition Time

(1) a Low (-) ------

Low (-) High (+)

A - Deposition Rate

Figure 2 - 11 Experimental Design Geometry

The corresponding factors and levels are summarized in Table 2-4. The recipe was not to be changed between runs to eliminate deterministic effects. The chosen recipe was Low Frequency Silicon Dioxide, which is used for two of the three STS-CVD processes in the buckled beam energy harvester fabrication process.

33 I

Table 2 - 3 Summary of Factors and Levels

Factor Symbol Level Symbol

50 nm / min Deposition A Rate 100 nm min +

2 minutes Deposition B Time 4 minutes +

The factorial design of this 22 experiment can be summarized run-by-run with table 2-5.

Table 2 - 4 Sign Effects for 22 Design [15]

Factorial Effect Run I A B AB

1 (1) + - - + 2 a + + - -

3 b + - + - 4 ab + + + +

2.5.3 Testing and Experimental Results

Results from experiment are expressed in plots below. Full data can be found in Appendix. Each thickness was measured in two different locations on each lcm2 silicon sample. At each location, two measurements were taken, as illustrated in Figure 2-13.

34 3mm 7mm

3mm

7mm 10 mm

10 mm

Figure 2 - 12 CVD Measurement Sampling

The results from the cold start measurement are below in Figure 2-14. After the machine being in disuse for a period of 48 hours, the same runs were performed at a deposition rate of 50 nm/min for a time period of 3 minutes, with ten replicates in each run.

"Cold Start" Experiment 270 Run 1 - Run 2 - Run 3 265

260

255 IL - A -L 250 - - -.-- 245 -e - -L - S- - -

A S 240 - - * A S U - A A ------235 ~ S

230

22M 0 5 10 15 20 25 30

ALocatio 1, Measurement I N Locatio 1, Measurement2 Sample Number Average

* Locatlio 2, Measurement 1 * Locatim 2, Measument 2 - I Standard Deviation

Figure 2 - 13 Film Thickness over Three CVD Runs with Identical Parameters

35 -1.

The results of the factorial experiment are summarized in the figure below, which took the average value from each run and created a ruled surface.

400 350 ni 300

U' 250 200 H 150 100 E 100 50 0 2 -0 50 U' 4 Deposition Time (min)

HO-50 M50-100 0100-150 a 150-200 0200-250 M 250-300 N 300-350 E 350400

Figure 2 - 14 CVD Average Silicon Dioxide Film Thickness

2.5.4 Data Analysis

When analyzing the results of the "cold start" experiment, we can inspect how the average thickness and variance change from run to run.

36 254 35 252 30 250 o248 25

246 20 ~244 15 242 IF ----

- m --- - 240 10 4 238 5 236 234 0 Run I Run 2 Run 3 Experimental Run

-- Average - Variance

Figure 2 - 15 Average Thickness and Variance for Cold-Start Runs

The average thickness was found by averaging all the measurements taken on all samples for each run. The variance displayed in Figure 2-16 is between the average thicknesses for each sample of each run. This figure suggests that running the STS-CVD multiple times after disuse depends on the process requirements of the microfabricated device in particular. Between Run 2 and Run 3, the average thickness increases by more than 10 nm despite. all input parameters being held constant. For the buckled beam energy harvester, this is a nontrivial change (3%) in material thickness of PECVD Silicon Dioxide that could alter the residual stresses in the device.

When analyzing the results of the factorial experiment, an analysis of variance (ANOVA) can be performed to understand the relative significance of deposition rate versus deposition time and observe any two-way interactions between the factors. The ANOVA table was populated using Factorial Regression on Minitab.

37 -1

Table 2 - 5 Analysis of Variance for Factorial Experiment

Source Degrees of Sum of Mean Square F-Value P-Value Freedom Squares Model 3 175813 58604 1870.47 Linear 2 175620 87810 2802.64 Deposition 1 1523 1523 48.59 Rate (A) Deposition 1 174098 174098 5556.68 Time (B) 2-Way 1 192 192 6.13 0.025 Interactions (A*B) Error 16 501 31 0.025 Total 19 176314

Regression can be performed using the data to create a model for the process based on the inputs of deposition rate and time. The adjusted R2 for this model is 99.56%. The model is expressed below.

Deposited Thickness [nm] = 93.3 time [min] + 0.35 depositionrate - 54.0

Based on this model, we can observe that the largest effect on deposition thickness given the data taken is by varying deposition time as opposed to rate. For practical implementation, recipe adjustment can be done most effectively by changes in deposition time according to this analysis.

38 Chapter 3 Post-Fabrication Buckling Design

3.1 Motivation

The microfabrication process involved with creating each batch of buckled beam energy harvester devices has high costs associated with it due to the number of fabrication steps involved. The largest factor in terms of design complexity is accurately balancing residual stresses as the fabrication process progresses. Iterative stress measurements for each additionally deposited layer are fed into an analytical model of lumped residual stresses so that the subsequent process parameters can be adjusted accordingly to fabricate a buckled MEMS film. This may become a key roadblock when addressing the mass production of buckled beam energy harvesters.

The buckled beam energy harvester design's monolithic structure relies on precisely balanced residual stresses in each of the eight layers of measurable thickness. The challenge in achieving the desired net residual stress and subsequent target buckling deflection is two-fold. Significant analytical modeling is required take as inputs material properties and determine the necessary thickness of each layer to induce controlled buckling in the thin film layers. Secondly, precisely fabricating these layers is nearly impossible, such that a stress control feedback loop is necessary. As fabrication stages progress, each additional material layer must be measured for thickness which is then re- submitted to the analytical model in order to adjust the target thickness for subsequent layers deposited.

39 Thin film deposition on dummy samples

Characterize the thin films (deposition rate, residual stress)

Design thicknesses of the rest control . layers

Deposit the next Plug in the control layer on measured data to real samples the lumped model

Measure the thickness and residual stress

End (deposition

Figure 3 - 1 Residual Stress Control Feedback Loop [1]

As outlined previously, there are a total of 56 fabrication steps, each with a dedicated machine, included in the clean room approved process plan for producing these devices. There are a number of secondary challenges resultant from the design of this long, complicated process, summarized as follows:

* Long production lead-time " High fabrication costs (both material and machine usage) * Low throughput yield due to batch-to-batch process variations

Finally, the large number of steps in the fabrication process include several high-risk stages such as spin-coating substrates with sol-gel or extracting individual devices from wafers, which individually have low percentage yield due to poor process control and together

40 compound to a low overall throughput where yield percentages over 40% for working devices in a given batch is typically not achievable given the existing fabrication process.

The need to accurately balance residual stresses significantly complexifies the fabrication process which is already long and costly due to the device design. The motivation for developing an alternative method to buckle the MEMS membranes after microfabrication has concluded is to simply the production process in this regard.

3.2 Design Goals

The most direct method of simplifying the microfabrication process is to remove the unpredictable and cost-intensive residual stress balancing scheme. This not only helps to loosen tolerances on material thickness for each layer, but also helps to eliminate unnecessary layers whose function is purely to counterbalance residual stresses. A cross- sectional view of the device's material stack is shown in Figure 3-2, illustrating the nine layers of material deposited on the silicon substrate. The intricate material stack presents an opportunity to simplify the device recipe.

41 Silicon Oxide (400nm)

Electrodes

PT (1nnm)

Silicon Oxide (300nm) Silicon Nitride (750nmi) Thermal Oxide (1 000nm)

Silic01 SUbstrate (5_30pmn)

Figure 3 - 2 Cross-sectional View of Buckled Beam Device

In the cross-sectional view of the device, the silicon substrate layer (530prm) is only present along the frame and the proof mass mid-beam section. Otherwise, the MEMS membrane only contains the layers from Thermal Oxide to Silicon Oxide. The figure is not to scale.

In order to simply the device design so as to leave the buckling operation to post-fabrication and merely fabricate functional material layers, the device needed to be redesigned with fewer layers and therefore fewer fabrication steps. Table 3-1 shows a functional break- down of the device on the layer level detailing the key design parameters and purpose of each material included in the stack.

42 Table 3 - I Functional Layers of Buckled Beam Energy Harvester

Layer Material Thickness Stress Function 0 Silicon 530 nm N / A Substrate, ProofMass 1 Thermal Oxide 1000 nm -300 MPa Structural 2 LPCVD Nitride 750 nm 170 MPa Structural 3 PECVD Oxide 300 nm -250 MPa Structural 4 Zirconium Dioxide 70 nm 370 MPa Active Layer 5 PT 10 nm 400 MPa Active Layer 6 PZT 240 nm 650 MPa Piezoelectric Active Layer 7 Electrodes N IA N/A Interdigitated Electrodes 8 LPCVD Nitride 800 nm -200 MPa Passivation 9 PECVD Oxide 400 nm -300 MPa Passivation

As illustrated in Table 3-1, the function of each layer can be generally categorized as either supporting the electric function of the device in terms of transmitting electric potential from the active piezoelectric layer through the interdigitated electrodes to externally facing contact pads or contributing to the dynamic perfonnance in terms of producing a buckled beam structure for the device. As shown in Table 3-1, every material layer except for the gold electrodes and silicon proof mass and frame is under some form of internal stress, whether compressive or tensile. The baseline design diligently accounts for the stresses in each layer and balances the effective internal stress in the beams to achieve the desired buckled deflection target of 200ptm of the proof mass from the frame.

There are advantages and drawbacks to the baseline design of incorporating multiple layers of materials under different internal stress to achieve a specific amount of buckling. The key positive aspect of this design is that the microfabrication process results in a monolithic device in a natural state of buckling that is ready to be installed on a ceramic pin grid array package for testing or use. However, this comes at the cost of a lengthy fabrication process that requires progressive feedback control in order to correctly balance tensile and compressive stresses. Following the fabrication of each additional layer on the device, the stresses are measured and compared to target values. Due to process variation mainly with

43 respect to layer thickness, the internal stresses are not necessarily balanced as desired at any given point during the fabrication process. Therefore, recipe parameters must be adjusted to account for the inaccuracies and maintain the stress balance necessary to achieve targeted buckling.

The original design goal was to eliminate non-functional layers and address buckling the membrane post-fabrication. This can be achieved by eliminating the material layers included in the recipe solely for balancing residual stresses. The cross section of such a simplified device design is shown in Figure 3-3.

Silicon Nitride (800nm11) Electrodes

Silicon Nitride (750nm)

SIliCOnI Substrate (530pmn)

Figure 3 - 3 Cross-sectional View of Simplified Piezoelectric Energy Harvester

By simplifying the material stack, the device becomes a piezoelectrically active array of doubly-clamped micro-beams that are not in any sort of buckling by design. This reduces the number of steps in the microfabrication process from 56 to 40, but also necessitates post-fabrication modification of the device to achieve buckling in the beams, which is the aim of this design. More specifically, the goal of this design was to induce a buckled state of the device's beams after microfabrication is completed and achieve a buckled deflection of 200pLm, identical to the baseline value of buckling.

44 3.3 Modeling

3.3.1 Analytical Buckled Beam Model

When designing a post-processing method of inducing a buckled deflection on the micro- scale beam array, it is helpful to develop a simplified structural model of the beams to analytically characterize what modifications will achieve the desired buckling.

As introduced previously, the monolithic MEMS geometry features an 18 mm by 18 mm silicon frame bounding 28 pairs of beams that support a centrally-suspended silicon proof mass, as shown in Figure 3-4.

28x pairs of beams

L

18 mm

Figure 3 - 4 Monolithic MEMS Energy Harvester Device Geometry

The MEMS device structure can be modeled as clamped-clamped beams of length 1/2 with a centrally suspended proof mass m. The frame spacing across which the mass is suspended is length L. A diagram of this model is shown in Figure 3-5.

45 1/2 1/2

.1 Or

L

Figure 3 - 5 Clamped-Clamped Beam Schematic

In the original baseline monolithic structure, each beam is under residual internal stress such that when the devices are released, there is a net compressive stress, which, when greater than the critical buckling load of the beams, induces buckling in the beams without any external loads to the device. In order to achieve the same buckling, but with the simplified device structure that has no designed residual stresses, the beams must be axially compressed such that they buckle accordingly. This effect can be created by offsetting the frame width to decrease length L, as shown in Figure 3-6.

......

zzzzZor4Z

L-d d

Figure 3 - 6 Buckled Clamped-Clamped Beam Schematic

By narrowing the dimension L by an offset x, the beam is placed under a compressive stress and buckling occurs, raising or lowering the proof mass by a distance S out of the neutral plane.

46 W(X) P A--_------

L-d

Figure 3 - 7 Buckled Clamped-Clamped Beam Model

This buckling can be analytically modeled for a clamped-clamped beam loaded with an axial compressive force P shown in Figure 3-7, resulting in an internal moment M inside the beam. The moment can be simply expressed using the force P and beam deflection w(x) which is a function of distance along the beam, in equation 9.

M = -Pw(x) (9)

For the buckling case, the moment can also be related to Euler-Bernoulli beam theory for the second derivative of deflection as a function of distance long the beam with the following equation (10).

M = )( 2W) (10)

Combining equations (9) and (10) gives a differential equation (11)

Pw + EI d 2 = 0 (11)

47 The second-order differential equation can be solved using the Euler solution for buckling, equation (12).

w = eAX (12)

Substituting equation (12) into a re-arranged equation (11) isolating the second-order term gives us equation (13).

-eAX + t2ellx = 0 (13) El

Where the solution for k can be expressed in equation (14).

2 P (14) El

This gives a general solution to equation (11) as the expression below for deflection in the buckled beam model.

w(x) = A sin x + Bcos x + Cx + D (15)

For the clamped - clamped boundary conditions of this model, we can solve the deflection w(x) as the following relation for the deformed shape of the buckled beam. Because the beam is clamped at both ends, we know that the displacement of the beam w(O) and w(L) at either end must both be equal to 0. From the clamped-clamped boundary condition we also similarly know that the slope of the modeled beam must be zero, such that w'(0) and w'L) are also zero. Therefore, the constants A, B, C, and D can be determined to give a model of the deflection of the beam along its length in equation 16.

w(x) = cisin ( L ) + c 2 cos (2L ) (16)

48 Using equation 16, the force P required to compress the device frame such that the beams buckle 200pim out of the plane can be calculated. By modelling the beams' cross-sectional geometry as a rectangular prism, the moment of inertia I can be approximated, and using material properties of the beam layer stack, an effective Young's Modulus can be approximated to predict an appropriate value for the compressive force needed to be applied to theoretically achieve buckling of 200ptm.

3.3.2 Experimental Buckled Beam Model

The analytical buckled beam model derived in the previous section is helpful in understanding the theoretical structural mechanics of the modeled clamped-clamped buckled beam case and predicting the approximate force needed to compress the fabricated device frame to produce the desired buckling, but the model is based on several assumptions including but not limited to a simplified geometry as well as lumped material parameters. While these assumptions accurately reflect the buckling case of the actual device, a higher degree of granularity can be achieved by way of a more precise experimental model.

Due to the high time and monetary cost of fabricating the actual energy harvesting MEMS devices of interest, a mesoscale scale model was designed to mimic the buckling and determine a relationship between horizontal compression distance and change in vertical buckling height of the device out of the neutral frame plane.

The mesoscale model was constructed at a scale of 1:35. The beam was modeled with a sheet of stainless steel with an aluminum proof mass of 30g. The desired buckled deflection for the scale model was 7 mm to be scaled with the actual target of 200 tm for the MEMS device. Deflections from 0 to 14 mm were investigated with the experimental model. The results of the model, seen in Figure 3-8, can be scaled down to predict the amount of displacement required in the actual MEMS device, shown in Table 3-2.

49 Experimental Model Buckled Deflection 16

12 U

S10 0

____00 8 /

S4

______I S2

0 W 0 0.5 I 1.5 2 2.5 3 3.5 Horizontal Displacement of Free End (mm)

Figure 3 - 8 1:35 Scale Model of Buckled Beam Deflection

Table 3 - 2 Predicted Frame Width Displacement based on Experimental Model Desired Buckling Change in Frame Length o Pm 0 pm 100 Pm 19.7 pm 200 pm* 38.9 ptm 300 pm 68.9 pm 400 pm 87.1 pm *targeted buckling deflection

Based on the results of the experimental beam model, in order to achieve buckling of the piezoelectric beams of a targeted 200 pim, the silicon frame length will have to be narrowed approximately 40 pm, or 0.2 % of the frame length.

50 3.4 Design Paths

Based on the analytical and experimental modeling of buckling in the MEMS device, the key functional requirement of any method that would induce such buckling is the ability to accurately adjust the width of the device frame to a fine tolerance. The question resultant from this requirement is how to achieve this given the current device geometry. As discussed previously, the MEMS device in question is bound by an 18 mm by 18 mm continuous silicon frame on all four sides, as illustrated in the topside view of the device in Figure 3-9.

Figure 3 - 9 Topside View of MEMS Energy Harvesting Device

As depicted in Figure 3-9, the MEMS beams are constrained at their base position by the silicon device frame. In terms of developing a method of post-fabrication buckling, there are two paths forward. The first option involves designing a fixture or process to induce buckling by narrowing the frame width without requiring changes to the original MEMS device design, i.e. taking the current simplified baseline device. The advantage to this option is that no tool or mask changes are required for the microfabrication process, so

51 minimal adjustments needed for the process design. The key disadvantage is overcoming challenges that the rigid frame will pose in terms of designing a fixture to narrow its width. The second option in terms of paths forward to develop methods of post-fabrication buckling involves modifying the geometry of the device frame to allow for re-positioning after fabrication. A main advantage of this option is the constraints of the rigid continuous frame are lifted allowing for different forms of positioning of the frame. The disadvantage of modifying the microfabrication process is the costs of verifying this process change and any possible effects it may have on other steps of fabrication. Both design paths were investigated in parallel in order to thoroughly examine their respective feasibility.

3.4.1 Achieving Buckling without Frame Modification

When considering a method to narrow the intact silicon frame width, two options are apparent, illustrated in Figure 3-10. Compressive forces can'be applied to either end of the frame to deform it without reaching the critical buckling load in an attempt to reduce its width. The other option involves bending the frame sides without plastically deforming the frame with the aim of bringing the opposite two sides closer together.

P M

Buckling Bending

ttt P M

Figure 3 - 10 Frame Buckling versus Frame Bending

52 The feasibility of compressing the frame to bring the opposite ends closer together can be assessed analytically. If we model the frame edges as bars with uniform cross-sections, then the force needed to compress the frame to produce an inward deflection of 6 is shown in equation 17, where A is the cross-sectional area of the frame edge, L is the length of the frame, and E is the elastic modulus of silicon.

P AE (17)

Assuming an elastic modulus of approximately 190 GPa for the silicon substrate, a length of 18 mm, and a cross-sectional area of 0.40 mm 2, it would require maintaining a compressive force of approximately 6.0 N (assuming 2 frame edges) to achieve the projected compression required, on the order of 40 tm, as predicted with the experimental model in section 3.3.2.

The value for P estimated using equation 17 should be compared to the critical buckling load defined by equation 18 for a clamped-clamped beam. While the intention of applying force P to the frame is to compress, buckling in the frame itself is not desirable.

4 iT2 EI Pcritical = 4 (18)(8

Assuming the same geometric and material properties stated when calculating the compressive force, the critical buckling load is approximately 220 N from equation 18. Therefore, the compressive force P (6.0 N) is well below the critical buckling load.

Similarly, the feasibility of bending the frame inwards to bring the opposite ends closer together can be assessed analytically. In order to do so, the frame must be considered in detail. Upon bending the silicon frame, the length along the neutral axis of the frame will theoretically neither elongate nor shrink in length. However, the inner face of the frame radially closer to the curvature resultant from bending will be under compression and will

53 decrease in length. Conversely, the outer face of the frame is under tension and will elongate, as illustrated in Figure 3-11.

radius of curvature

M M

Figure 3 - I I Inner Face of Frame under Compression

If the frame edges are modeled as beams subjected to a moment M, as shown in Figure 3- 11, then the curvature of the frame K can be approximated as a function of that moment as well as the elastic modulus of silicon Es; and the moment of inertia of the frame Jf, in equation 19.

M K= (19) Esi Ifr

By taking the inverse of the curvature found using equation 19, the radius of curvature of the bent frame can be estimated. Given the frame thickness is known, the change in length of the inner face can be approximated as well.

Based on the experimental model, a change in length of the inner face required is approximately 40 pm. The radius of curvature needed in the bent frame for such a length

54 change can be calculated by finding the difference between the extended frame in equation (20) and bent frame, as shown in equation (21). The extended length and desired bent frame distance are both known, so these two equations can be solved simultaneously to find the radius of curvature.

Lextended = 0 27T rcurvature (20)

Lbent = 2 rcurvature sin ( (21)

Solving these equations simultaneously gives a radius of curvature of 77.5 mm, or a

curvature of 12.9 m-. Based on this calculation, the moment found using equation 19 is

determined to be approximately 0.025 N-m.

3.4.2 Achieving Buckling by Frame Modification

If the constraint of developing a post-fabrication buckling process for the continuous silicon frame of the baseline device is lifted, then there is considerably more design freedom to work with. When designing a new frame geometry, however, the functional requirements of the frame still need to be kept in mind. As illustrated in Figure 3-9, the current frame not only provides structural support to the beams suspending the proof mass but also houses the electrically active contact pads from which power is transmitted when the device is actively harvesting energy. A new frame design must still house these contact pads suitably and also provide the same structural support for the device. Not only does the frame act as a support to the device during use and testing but also the frame is often times a fixture point which is used to transfer individual devices during the fabrication process without touching the delicate piezoelectric beams. The functional requirements for the redesigned frame can be listed as follows:

55 * Provide ample surface area for electrical contact pads " Structurally support the micro-scale beam array " Facilitate adjustability with respect to frame width

Based on these defined functional requirements, an optional fourth requirement can be added which splits potential designs into two general categories. The frame width adjustment can either have discrete positions or have a continuous range of adjustability.

3.5 Proposed Designs

Based on the idea exploration throughout various design spaces given different constraints, two parallel paths were chosen to pursue. First, a fixture was conceived to bend the device frame to induce buckling in the piezoelectric beams. Second, a novel frame geometry was designed to allow a snap-fit, one-time frame length modification. Both of these designs are elaborated upon in this section.

3.5.1 Experimental Frame Bending Fixture

The following fixture design was developed for concept proof and experimentation. The concept driving the development of the frame bending fixture was that the frame length could be made shorter by applying a moment on either end of the frame to compress the face on the inner side of the resultant curvature. The fixture design presented here utilizes controlled bending of a cantilevered aluminum beam upon which the device is fastened, shown in Figure 3-12. The dotted silhouette of the beam and MEMS device subassembly represents the unbent mode.

56 MEMS Device Aluminum Beam Steel Wedge Clamp Fixture

M ......

--- X - NIN9991"WwwrMp ...... U11_UO'"0,00* 7Z

Figure 3 - 12 Frame Bending Fixture Schematic (Side View)

When the steel wedge is driven under the aluminum beam, the beam lifts with the increasing height of the wedge. This bends the entire cantilevered aluminum beam. Figure 3-13 shows a close-up view of the interface between the MEMS device frame and the aluminum beam. With a properly mated surface interface, the device frame will bend with the same curvature as the aluminum beam, inducing buckling in the piezoelectric beams.

MEMS Device Frame

Aluminum Beam 0000,0 00 40 00 0000 'S.. * ~*uuSamoa**%*5S

Figure 3 - 13 Aluminum Beam - Device Frame Interface (not to scale)

57 Figure 3 - 14 Isometric Rendering of Experimental Frame Bending Fixture

Figure 3 - 15 Side Rendering of Experimental Frame Bending Fixture

58 3.5.2 Snap-Fit Corrective Frame Fixture Design and Process

While the experimental frame bending fixture was effective in proving the concept of frame bending translating to buckling in the MEMS membranes of the energy harvesting device in that it had a large range of buckling made possible by the adjustable wedge, it was geared towards an experimental application. Then subsequent iteration of design focusing on achieving buckling post-fabrication is a scalable process utilizing a monolithic corrective frame modification process.

Based on analytical and experimental modeling results of the device frame bending, a known curvature i can be identified within a certain tolerance as the benchmark curvature needed to induce desired buckling in the MEMS beams. The experimental fixture design utilizes a strong bond between a bent cantilevered beam and the device to couple bending in the fixture to the silicon frame. This can be achieved more reproducibly and precisely by a fixture rigidly cast in the geometric form of the curvature.

A schematic of the proposed "snap-fit" fixture is shown below in Figure 3-16. The illustration is not to scale, and the curvature is exaggerated in order to demonstrate conceptual function.

P (dist.)

Figure 3 - 16 Snap-Fit Schematic

59 -- '-I

Such a design utilizes the same model of curvature presented in 3.4.1. In this case, however, the parameter of interest is the appropriate force applied on the frame against the mold fixture. The frame can be modeled as beam, where the desired deflection due to bending is known and the applied force is of interest. The boundary conditions of the beam can be considered as free supports that just provide vertical resistance to the applied force. A free body diagram detailing the mechanical deformation of the frame is shown below in Figure 3-17. The distributed force P is lumped as a point load mid-way along the beam.

L ...... I P t t F F

Figure 3 - 17 Snap-Fit Model Free Body Diagram

By summing forces in the z direction, we can find that the reaction force at each support is half that of P. Given the desired curvature of K, we can determine the necessary deflection and respective force needed. At the center of the beam, the moment is PL/4. From the equation 22 below, curvature is related to moment through the elastic modulus and beam's moment of inertia.

M= KE I (22)

60 Therefore, the necessary force required lumped at the middle of the frame model can be found by solving for P.

P= 4KEI (23) L

Knowing the frame length is 18 mm, curvature is 12.9 m-, using an elastic modulus of silicon as 190 GPa, the necessary force P can be estimated as 5.51 N. This frame modification process is a kind of dry molding technique to elastically deform the frame. The mold against which the device is deformed and which defines the radius of curvature is in the form of a half-pipe. Along with the snap-fit fixture design, a standard operating procedure (SOP) must been developed to detail the order of operations for how such a MEMS device can be modified at production scale. This SOP will supplement the fixture design as a defined method of executing the post-fabrication beam bending successfully and take into account operator technique and other variables to maximize efficiency of frame bending. Multiple renderings of the fixture design are displayed in Figures 3-18-20.

61 Figure 3 - 18 Exploded View of Snap-Fit Fixture I 1I Clamp

U~ evice k K~ 2

Mold

Figure 3 - 19 Cross Sectional View of Snap-Fit Fixture

62 C

Mold

Figure 3 - 20 Detail Cross-Section View of Device / Mold Interface

3.6 Testing

Of the proposed designs, the frame bending fixture was selected for construction and testing as it did not require major changes to the microfabrication process, so pre-fabricated MEMS devices could be used for testing the effectiveness of inducing buckling. Additionally, the frame bending fixture design allowed for a continuous range of induced buckling deflection which was desirable from a testing standpoint.

3.6.1 Fixture Construction

The frame bending fixture was constructed according to the design proposed in section 3.5.1. An aluminum beam of thickness 1.6 mm was waterjet cut to a width of 22 mm with a square aperture of dimensions 12.75 mm by 12.75 mm in the center such that the beam did not interfere with the motion of the piezoelectric beams and proof mass during dynamic testing. The MEMS device was adhered in position to the aluminum beam with epoxy

63 adhesive. One end of the aluminum beam was clamped to a steel base and the other end remained free, as shown in Figure 3-21 and Figure 3-22 Steel sheets of thickness 0.25 mm were incrementally inserted beneath the free end to raise its vertical position and actuate the bending in the device.

Aluminum Beam MEMS Device - - Steel Base

Figure 3 - 21 Frame Bending Fixture

& Steel Wedge Aluminum Beam MEMS Device Steel Base

Figure 3 - 22 Frame Bending Fixture, Side View

The MEMS energy harvesting device used for testing did not have any measurable buckling in its immediate post-fabrication state. Due to process issues, the residual stresses

64 of the batch from which this device was released were not balanced correctly resulting in a device that was ideal for testing on the fixture presented in this section.

A closer visual of the mounted device on the fixture qualitatively indicated that as a result of the bent frame, the piezoelectric beams were buckled. Two distinct modes of buckling were observed, shown in Figure 3-23, correlating to the two positions of stability in the bi- stable buckled beam device. By imparting an impulse on the fixture with an action such as a tap of tweezers on the aluminum beam, the beams snapped back and forth between positions of stability.

Figure 3 - 23 Two Positions of Buckling Observed on Frame Bending Fixture

65 3.6.2 Results

The frame bending fixture was tested statically and dynamically. Statically, the actual buckling deflection of the proof mass experienced was measured. Dynamically, the entire fixture was mounted on an electromagnetic shaker and the acceleration of the proof mass was measured while varying the excitation frequency, to verify the resonance of the MEMS device was in the desired range.

Buckling deflection was measured using an optical surface profilometer. Profile measurements of the aluminum beam, the device frame, the piezoelectric beams, and the proof mass were taken. The proof mass and piezoelectric beams' profiles were measured in both positions of stability in order to understand the buckling deflection on either side of the neutral plane. As illustrated in Figure 3-24, the proof mass deflected downwards at an average 207 pm, and deflected upwards at an average 246 pm, for a total range of vertical motion of 453 pm, within 13% of the desired 200 pm.

66 - - .1

Profile of Buckled MEMS Device 1400

1200 ProofMass Device Frame

1000

Beam 800 - ----Aluminum

600 ProofMass

400

- I -

200

0 a 0 2 4 6 8 10 12 14 16 18 20 Distance along Frame (mm)

Figure 3 - 24 Profilometer Measurement of Beam Buckling

Due to the nature of the fixture setup, the cantilevered mode of fixing the aluminum beam meant that raising the free end resulted in the profiles appearing exaggeratedly inclined when viewed at a high level of granularity in z - height.

Dynamic testing was executed by mounting the frame bending fixture assembly on an electromagnetic shaker. The acceleration of the shaker was measured using an accelerometer, and a signal generator controlled the shaker's vibration frequency and amplitude. The movement of the proof mass relative to the device frame was measured using a laser vibrometer. Two different types of dynamic testing were conducted. First, the electromagnetic shaker swept through a range of frequencies from 1 Hz to 1000 Hz at 1 G in order to identify the resonant peak of the device by measuring the proof mass acceleration in the frequency domain. The results of this test are shown in Figure 3-25.

67 Based on the frequency sweep, the device resonance has a relatively wide bandwidth in the range of 0 to 100 Hz.

N . - -- F.5w3.5 g e frqunc sweemesureent teerifni the diw

F3 ur 3-25AceertinofPro Mss reuec DmanRepos

stb1t was 1isal apar9 t Figure4 3-26 sh3 w the S64 ag measurement in4he6time

d Figufre a three-secdextation af Pro Hzasd F.5qG.nTheDopein Rsponpsenta difestene ordesrtoinfy ehathe firameibnig obseraie, and ot oaparnts thth ectonact and earecingrslsnre voltage spikvbrtngte hae ate aonat The reuen -f this tests

acceptable electric function of the device while in the frame bending fixture assembly.

68 4

3Hz Excitation - Time Domain Voltage Measurement 0.40

0.35

0.30

0.25

0.20

0.15

0.10

0.05 ------I-

0.00 0 0.5 1 1.5 2 2.5 3 Time (seconds) Figure 3 - 26 Device Electric Function Verification (open circuit voltage)

Based on the promising results of the electric function and the resonance at excitations below 100 Hz, the device mounted in the frame bending fixture assembly was tested at frequencies from 10 Hz to 100 Hz for open circuit voltage and proof mass displacement at intervals of 10 Hz, all at I g. Electric function was verified at these operating frequencies. Data has been plotted and is attached in Appendix A.2.

69 70 Chapter 4 Energy Harvesting Footwear

4.1 Introduction to Wearable Energy Harvesting

The human body maintains a diverse portfolio of physical activity. Certain human processes, such as the blood coursing through veins and arteries, run continuously in the background, while others, such as sprinting, occur as acutely executed functions of motion. Regardless of the nature of each activity, every physical motion of the human body presents an opportunity to harvest energy and power a portable device without relying on a battery in remote areas [16]. Of these activities, the impact of the human foot on the ground during walking and running presents the greatest occasion to harvest energy in terms of energy expenditure [17]. While bipedal movement is not a continuous activity, its occurrence is correlated with the need for electric power. This raises the challenge of converting foot- falls into usable electric power.

Regular walking motion at 1.3m/s (3.0mph) typically results in each foot striking the ground at a frequency of approx. 1 Hz for an adult. According to previous investigations, the dynamical vertical force can be up to 130% of the person's body weight [21]. Assuming each foot is lifted just 1cm from the ground, there is a potential energy of approx. 9.6 J for a 75kg adult dissipated with every footstep considering only the vertical motion of the foot. In order to prevent the discomfort of walking on a sponge or swamp, the maximum power harvested should not be more than 10% of the foot energy. The challenge presented by this opportunity is to harvest the foot strike energy as efficiently as possible up to 100 mW (for 1Hz walking), despite unpredictable nature of human motion [18].

The early and most intuitive method of scavenging energy from footsteps utilized piezoelectric materials embedded in areas of footwear that experience the most strain. Shenck and Paradiso demonstrated a method of harnessing footstrike energy piezoelectrically by inserting a PZT foil into the heel area of a shoe [19]. An average power

71 of 8.4 mW for the PZT device was reported for a foot strike frequency of 0.9 Hz. The benefit of such a design is its unobtrusiveness with respect to the user's ergonomic experience with the product. However, the magnitude of energy that can be harnessed is limited by the use of piezoelectric devices which rely on mechanical strain created by the heel inside the shoe to create a potential difference. Moro et al. also demonstrated an energy harvesting shoe design with a system of piezoelectric vibrating cantilevers, but power generated was restricted by the low frequency of foot impact [20]. Above all, the major limitation to piezoelectric energy harvesting from walking is that the total foot energy cannot be converted to the strain of the piezoelectric material without complex additional mechanisms.

We previously developed an air pump-turbine-based system to solve the difficulties of harvesting energy from foot strikes by absorbing the foot strike energy with deformation of air bulbs and utilizing an airflow speed amplification mechanism to drive a micro- turbine [18]. The turbine was activated by airflow and operated with a high frequency, enabling the foot-strike energy to be harvested effectively. The turbine casing was designed specifically to enable the device to operate continuously with airflow from both directions [21]. A prototype was fabricated and then tested under different situations. A 6 mW peak power output was obtained with a 4.9 Q load for each foot strike. Considering that a maximum of 100 mW could be potentially harvested as described above, there was a room for improvement.

This work presents optimized components and system design to solve these challenges by using footsteps to compress nylon lungs embedded in a shoe sole to pneumatically drive miniature turbines and power small generators for electric power. A combat boot has been developed to provide electric power to a passive GPS tracker that can periodically ping geographical coordinates to a satellite relay this information to a command center.

72 4.2 Mesoscale Energy Harvesting System Design

The system for harvesting energy presented in this chapter converts able-bodied bipedal locomotion to usable electricity by way of unobtrusive hardware installed in footwear to generate airflow from the compressive forces applied by feet inside shoes during walking [24]. A diagram of this energy harvesting system embeddable in different forms of footwear is displayed in Figure 4-1. Air bulbs are positioned in the sole, optimized relative to the length of the foot. Outflow from the bulbs is directed towards a two-stage turbine enclosure which drives two DC motors in series to generate electrical power. This energy is stored in a supercapacitor and used to charge a battery which powers a GPS receiver.

4.2.1 Air Bulb Design and Placement

The key design parameters considered for the air bulbs were their effective stiffness and time period of regaining original shape after being compressed. The bulbs' effective stiffness was bound by the requirements that the bulbs not only compress under the user's weight but also self-reset to their original decompressed form and do so fully before the user takes another step. The time available to reset depends on the frequency of foot-strikes during motion which varies with gait and speed of walking or running. The resultant time allowance for the air bulb to regain its initial shape can be as little as 0.3 seconds [22] when stepping with higher frequency. The actual air bulbs selected for testing were large nylon hand pumps used in sphygmomanometers, or blood pressure meters. The volume of each air bulb is 110 cubic centimeters, with length of 7.5 cm and diameter of 4 cm. When installed in the boot prototype, the bulbs were positioned with a tight tolerance with respect to each other and the supportive structure of the shoe interior, but still retained their shape when sitting neutrally in the shoe. Three bulbs were used for prototyping. This bulb type was only used for the proof-of-concept. Future design iterations will include embedded bulbs in the insole or outsole to minimize the discomfort to the user.

73 GPS Transmitter Power Management Module Turbine Enclosure Air Bulbs (3x) Insole Rubber Tubing Boot

Figure 4 - 1 Schematic of Footwear Energy Harvesting System

Optimal placement of the air bulbs along the shoe was determined by two design requirements. The first design requirement was that chosen array of air bulbs produce the maximum air outflow with each step. The second design requirement was that the air bulbs be positioned such that time period during which air outflow was generated was maximized during each step cycle. Furthermore, each additional air bulb added to the system increases the effective stiffness of the entire bulb array. Therefore, the design decision to add each additional air bulb to the system was based on the justification that the marginal contribution of airflow to the turbine enclosure was greater than the resultant decrease in airflow from the array when compared with the original configuration. Testing, detailed below and summarized in Figure 4-2, found three bulbs to be optimal given the stiffness and volume of the off-the-shelf nylon air bulbs selected for this purpose.

Positioning the air bulbs with respect to maximizing air outflow was maximized by identifying high pressure regions of the foot during walking. Although the morphology of human feet varies greatly among individuals, clinical research has shown that the pattern of weight distribution on the foot of an able-bodied human generally has maximums in the heel, ball, and toe regions of the foot [23]. In order to verify this pattern was generally accurate for the test user of the prototype, the test user stood on a sheet of pressure

74 indicating sensor film which confirmed same three regions as bearing the weight of the body.

In order to quantify how varying air bulb position along the length of the shoe directly affected energy harvesting, a single air bulb-turbine-generator assembly was developed specifically for this setup. The relative change in airflow outputted by the air bulb depending on location in the shoe was measured by observing the peak open-circuit voltage from the generator. The test setup was incrementally shifted along the length of the shoe and compressed by the weight of the human test subject at each location. The results of this test are presented in Figure 4-2. In order to maximize the time period of air outflow during each step cycle, spacing between groups of air bulbs was introduced to create sequential compression of the air bulbs based on the gait of human walking motion. Based on these design requirements, two air bulbs were positioned in the heel area of the sole, and one in the ball area.

Optimal Air Bulb Placement 6K,

500

~400 Heel 0~ i/ Ball 200

100 L IIrbe-t 4i.o

n 0 10 20 30 40 50 60 70 80 90 100 Percentage of Length along Shoe Sole

Figure 4 - 2 Peak Power relative to Length along Shoe

75 4.2.2 Turbine Enclosure Geometry

The turbine enclosure was designed to accept flow from either the inlet (exhalation of the air bulbs during compression), or the outlet (inhalation of the bulbs during decompression), while maintaining revolution of the turbines in the same rotational direction regardless of the flow direction. Figure 4-3 depicts a drawing of the turbine enclosure with the top enclosure translucent to illustrate how the symmetric positioning of the two turbines allows conversion of airflow to rotational kinetic energy regardless of which inlet the air enters. The previous iteration of turbine enclosure design also geometrically enabled flow from opposite directions to rotate a single turbine in the same anti-clockwise direction, but this feature was not used to harvest energy during air bulb inhalation on the off-step. The air bulbs used in the previous iteration were off-the-shelf pumps for inflating the sleeve of a blood-pressure meter and had embedded check valves ensuring that a dedicated outlet only allowed airflow outwards. For the prototype presented in this chapter, the check valves were removed, and the bulb inlet was plugged to force the air bulbs to re-inflate after compression by drawing air back through the turbine enclosure on the off-step.

1.0 cm

2.4 cm

3.5 cm Figure 4 - 3 Radial Symmetry of Dual Turbine Enclosure allows for Bi-directional Flow to continuously drive Generators

76 The turbines themselves were designed for efficiency of airflow harvesting by optimizing blade number and internal diameter using computational fluid dynamics (CFD) [21]. The turbine blades were offset by 8.5mm from the housing to prevent contact during use while maintaining efficiency.

The entire energy harvesting system including air bulbs, tubing, connectors, turbine components, and electronics has a mass of 130g. Each combat boot used for testing has a mass of 770g, so the additional weight experienced by the user is 17% of the original baseline shoe weight. Subsequent design iterations will eliminate the majority of mass from PCBs and other meso-structures when the prototype is developed for mass-application products. Since the bulbs will be embedded in cavities in the insole of the shoe, we target a net mass increase on the order of approximately 20 grams.

4.3 Theoretical Calculations of Upper Bound Power

In our group's previous work [18], the theoretical upper limit power by the air bulb compression and mini-turbine energy harvesting was modeled. We assumed the air bulb displacement to be 10 mm, and an amplification factor resulting from the ratio of the sectional area of the pump to that of the rubber tube inlet to be 6 = 500, then the velocity of air entering the turbine can be estimated as 10 m/s for walking at 3.0mph.

The kinetic energy for the airflow entering the turbine at relatively high velocity can be calculated from equation 24.

E = mav2 (24) 2

The mass of air is represented by ma and the velocity of airflow is v. The time period of each step was measured to be approximately 0.2 sec for walking at 3.0 mph. In Figure 4-7 we observed the voltage signature output with respect to time to show working duration to

77 be approximately 1.0 seconds, which includes the inhalation time after the foot step up. The calculation in equation 25 counts only the down strike of the foot on an individual air bulb.

The energy from each step can be estimated as:

E = ma(6 vO)2 = 1(10-4 X 1.29)(500 x 0.2)2 = 0.645J (25) 2 2

Therefore, the upper limit for the power to be theoretically generated from 1.0 Hz foot strikes is approximately 645 mW.

78 4.4 GPS - Equipped Combat Boot Prototype

A product prototype was built with the intent of applying the energy harvesting concept outlined in this chapter to a real-life opportunity. The prototype presented here uses the harvested energy from walking to power a GPS module that relays geographic coordinates when requested by an SMS message from a cell phone. The motivation for selecting GPS functionality as the main application of this prototype was due to the expected use of such a product in remote environments. Potential users of energy harvesting footwear will likely be in an environment where marginal electric power can be valuable, meaning use of this product would take place far from human infrastructure capable of delivering electric power. Such users may include military personnel, emergency first-responders, and outdoor recreational explorers.

A prototype energy harvesting system was built and installed in a combat boot, shown in figure 4-4. Three air bulbs were sourced from off-the-shelf latex blood pressure measurement pumps and embedded in the shoe sole at the optimized locations. The air bulbs are transparently overlaid in figure 4-4; two bulbs are positioned in the heel while one bulb is positioned in the ball region. Rubber tubing was used to direct air from the bulbs to the turbine enclosure. The turbine enclosure and the turbines themselves were 3D printed. The turbine diameter was 1Oimm, with a height of 5 mm. Each turbine was directly connected to the shaft of a 1.5V DC motor used as generators. The turbine enclosure was mounted externally to the outer side of the combat boot.

79 GPS Transmitter

Power Management Module

Turbine Enclosure

Air Bulbs (3x)

Figure 4 - 4 Prototype Combat Boot Fitted with Energy Harvesting System

For experimentally measuring power output, the motors were connected in series to a resistor, and power was estimated by measuring potential difference across the resistor. For functional demonstration, the motors were connected to a boost converter and supercapacitor made by Advanced Linear Devices (EH4295), and the output power was regulated by a power management circuit (EH300). The supercapacitor was used to charge a Li-Po battery which powered an off-the-shelf GPS receiver. Power measurements were conducted while a 75 kg test subject walked at different controlled speeds on a treadmill while wearing the combat boot.

The off-the-shelf GPS receiver used for this prototype communicated on the 3G network. The architecture of the GPS module used for the prototype can be broken down into two functions: receiving a positioning signal from a satellite (to obtain geographic coordinates), and transmission to the GSM network (to communicate the coordinates to a cellphone). The off-the-shelf location tracker receives a low-powered microwave GPS signal from at least three or four satellites (for 2D and 3D positioning respectively). The module hardware amplifies the signal to increase gain over noise, and mixes, filters, and processes the signal in order to convert to measurements of position and time. The GPS receiver took an average

80 of 12.6 seconds to communicate to the satellite, requiring 5.80 J of energy. Communicating location in the form of an SMS message to a cellphone took 13.2 seconds and required 6.25 J of energy. This measurement was conducted by connecting the GPS module to an oscilloscope and measuring current and voltage consumed during transmissions. The values were averaged over 8 runs.

This high power-consuming 3G GPS is actually not necessary for the basic functions provided by this prototype but was used for the ease of using off-the-shelf components for testing and demonstration in certain parts of Asia where 2G is no longer available. Using a custom-built 2G GPS unit with minimal energy needs in the future could reduce footsteps needed per signal by an order of magnitude.

Potential strategies for reducing the power consumption of the GPS module include transmitting the geographic coordinates for longitude and latitude to fewer decimal places to reduce energy costs during transmission via GSM [25]. Additionally, separating the receiving and transmitting processes into two charging cycles to create independency between the two processes will add to the robustness of the design. An initial stage would charge a power bank and releases enough energy to power the GPS receiver and log the geolocational coordinate string for a microprocessor to store the data. The second stage would allow the stored position coordinates to be transmitted. Furthermore, increasing the efficiency of the module by customizing the digital signal processing algorithms and running the microcontroller at a lower speed would reduce the power-consumption of the hardware architecture, and allow for more frequent transmissions based on the same rate of walking.

81 I

4.5 Results

The device was tested at different walking styles and speeds and voltage was measured over a 22Q resistor. This resistance value was optimized by measuring peak power output for a footstep cycle for various load resistances, shown in Figure 4-5.

Optimizing Resistance of Dual Turbine System 1000

800

E 600 41)

0 0~ 400 a) 0~

200

0 0 5 10 15 20 25 30 Load Resistance ( Q)

Figure 4 - 5 Peak Power Output versus Values of Load Resistance

The human factor involved with harvesting energy from footsteps meant that such a device would naturally be subjected to a variety of types of movement by users with varying weights. Due to experimental constraints, a single user of weight 75 kg tested the prototype. The prototype was tested on a treadmill at two different walking speeds (1.5mph and 3.0mph) which corresponded to foot strikes at frequencies of 0.5Hz and 1.0Hz respectively. Voltage measured across a 22Q resistor for these cases is shown in Figures 4-6 and 4-7. Additionally, power measurements were taken while the test subject lunged, attempting to simulate maximum impact to the shoe. These voltage measurements are shown in Figure 4-8.

82 Walking at 1.5mph 5

4

3

1

0 0 1 2 3 4 Time (s) Figure 4 - 6 Voltage Output measured for 0.5 Hz Footsteps

Walking at 3.0mph 5

4 --- 3

2

1

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 Time (s) Figure 4 - 7 Voltage Output measured for 1.0 Hz Footsteps

83 Lunging 5 -I~~

4 ......

3 0) CM

1

0 ' .1 ).5 1 1.5 2 2.5 3 3.5 4 . 5 Time (s) Figure 4 - 8 Voltage Output measured for Lunges

The power measurements in Figures 4-6-8 are based on measurements for voltage over the known resistor load and estimated using the equation expressed in equation 26 relating electric power P to potential difference V and load resistance R.

P = R (26)

The peak power for these tests can be estimated by relating peak voltage and known load resistance, 22K in these cases. The average continuous power was found by calculating the total electrical energy generated over a six-second interval of use and dividing by time. These values are displayed in Table 4-1.

Table 4 - 1 Power Generated for Various Modes of Movement

Movement Type Average Power (mW) Peak Power (mW) 0.5 Hz Footsteps 29.9 161 1.0 Hz Footsteps 86.4 516 Lunges 106 679

84 While power measurements were conducted while the test subject moved in controlled motions on a treadmill, the functional demonstration of the GPS capabilities of the prototype was conducted in a remote area on uneven terrain, allowing the user to create random motions similar to those experienced during routine use. The off-the-shelf GPS receiver used for testing required 15 minutes of activity in the combat boot before enough energy was produced to send a text message with geographical coordinates.

From equation 25, the upper theoretical limit for power output by the turbine was 645 mW at footstep frequency of 1.0 Hz. Based on the average power measurement in table 4-1, this prototype has achieved capturing 13.4% of the foot strike energy. This value is comparable to the initial goal of capturing 10% of the foot strike energy as a goal to balance maximizing harvested energy and minimizing side effects on the user's experience while wearing the shoe during use.

4.6 Discussion

This chapter presented an energy harvesting system designed to convert footsteps into usable electric power. The device used an array of air bulbs which sequentially compressed and decompressed during the user's walking motion, to continuously drive a series of miniature turbines to generate energy. The turbine enclosure was designed specifically to accept flow from opposite directions while ensuring the turbines themselves always turned in the same rotational direction. As a result, at a moderate walking speed of 3.0 mph (footstep frequency of approximately 1 Hz), there was no measurable down-time when the turbines were not rotating and producing electric power.

Because the energy harvesting device depends on human movement, it is difficult to quantify its performance with respect to a wide range of users. The prototype presented in this chapter was tested on a 75 kg subject and generated an average continuous power between 30 mW and 80 mW depending on the gait and speed of the subject's movement. Approximately 15 minutes of varied forms of movement was required to generate enough

85 energy to send a GPS transmission detailing geographical coordinates, sent as a text message to a cell phone.

If the airflow generation and air bulb compression is further optimized within the sole of the boot, the power output can be improved to the theoretical prediction of the previous section. Considering the GPS receiver used in this prototype required very high power for transmission to satellites, this combat boot can be optimized to provide more power at each step to a lower power consuming GPS in order to deliver transmissions more frequently.

Future work on this device will involve optimizing the circuit design by developing a custom designed power management system and low-power GPS receiver to reduce the number of footsteps required to operate as well as to improve the performance of the air bulb-turbine efficiency. In addition, ergonomic factors not considered in this design such as the effect of prolonged, repetitive use on the user's feet need to be taken into account and incorporated into the final product-oriented design.

86 Chapter 5 Summary

5.1 Thesis Summary

This thesis has contributed to the objective of achieving low frequency vibrational energy harvesting at the microscale and mesoscale. At the microscale, the microfabrication process for producing piezoelectric bi-stable buckled beam-based energy harvesters was analyzed and quality of processed piezoelectrically active material (PZT) was identified as an opportunity for improvement. New batches of devices were fabricated with remnant polarization of the PZT layer an order of magnitude larger than the previously reported value measured for this device. The process capability of specific microfabrication steps was also examined from a statistical process view point. The cumbersome method of balancing residual stresses in each material layer of the device to achieve film buckling was identified as a process of significant complexity. This process was moved to post- fabrication, thereby allowing the microfabricated material stack to include just functional layers. A post-fabrication buckling process by inducing bending in the device frame was modeled and a fixture enabling this procedure was designed. An experimental iteration of the fixture was built and tested with actual buckled beam energy harvesters in order to prove the concept and validate its dynamic performance. Electrical function of the device was preserved.

At the mesoscale, an energy harvesting method to convert energy dissipated through walking to usable electricity for commercial application has been optimized and applied to the purpose of powering a GPS receiver embedded in a combat boot. Air bulbs were placed in the sole of the shoe to generate airflow from walking, which was used to drive a turbine assembly on both the down strike and uplift of the foot's motion during walking and jogging. The power generated was stored in a supercapacitor and used to charge a li-po battery that powered a GPS receiver capable of geographic coordinates upon demand at intervals of 15 minutes.

87 5.2 Future work

Future work on the MEMS vibrational energy harvesting device will focus on achieving power generation on the order of micro-watts for frequencies below 100 Hz at amplitudes below 1 g. Armed with the assets of high PZT material quality and a method of enabling bi-stable buckled beams through post-process adjustment fixtures, this objective is now possible to achieve with the next batch of fabricated devices. With a MEMS device capable of delivering power at this level, commercial application is within reach. The energy harvester can be integrated with a low-power sensor and transceivers to create a nodal sensory network powered by vibrations from the ambient environment. Other Internet of Things applications are also open to exploration with this novel energy harvesting opportunity available. The longevity and durability of this MEMS energy harvester with respect to fatigue over high cycle use must also be investigated further.

Future work on the energy harvesting footwear design involves developing a unified shoe design with cavities and other embedded features to reduce the mass of the added components from the energy harvesting system. Furthermore, studying the ergonomics of the energy harvesting shoe is important in understanding any side effects marginal damping may have on human walking. From an electronics stand point, an application-specific GPS receiver with more efficient power electronics and communication algorithms must be developed. Most electronics components used for the prototype demonstration were off- the-shelf products. Optimization of a customized voltage boosting, charge storing, and GPS receiving system can reduce the number of steps required to relay geographic coordinates. Other applications beyond GPS must also be investigated including device charging and variable shoe stiffness among others.

88 89 Appendix A

A.1 PECVD Process Capability Study Data

PECVD process capability study data

Run 1

Deposition Deposition Sample Location 1, Location 1, Location 2, Location 2, Rate [nm / Time Measurement Measurement Measurement Measurement min] [min] I [nm] 2 [nm] I [nm] 2 [nm] 50 3 1 231 234 246 245 50 3 2 240 243 247 250 50 3 3 232 233 260 260 50 3 4 235 237 232 234 50 3 5 247 251 239 234 50 3 6 236 242 267 263 50 3 7 236 235 232 234 50 3 8 233 230 246 252 50 3 9 240 242 236 233 50 3 10 233 230 237 243

Run 2

Deposition Deposition Sample Location 1, Location 1, Location 2, Location 2, Rate [nm / Time Measurement Measurement Measurement Measurement min] [min] I [nm] 2 [nm] I [nm] 2 [nm] 50 3 1 247 241 249 251 50 3 2 243 240 260 259 50 3 3 239 238 240 237 50 3 4 238 243 239 243 50 35 241 237 243 234 50 3 6 238 238 245 241 50 3 7 236 238 234 240 50 3 8 237 238 251 244 50 3 9 238 234 239 233 50 3 10 243 239 239 246

90 Run 3

Deposition Deposition Sample Location 1, Location 1, Location 2, Location 2, Rate [nm / Time Measurement Measurement Measurement Measurement min] [min] I [nm] 2 [nm] 1 [nm] 2 [nm] 50 3 1 253 253 256 256 50 3 2 253 255 251 249 50 3 3 251 249 255 251 50 3 4 254 251 255 260 50 3 5 263 262 266 247 50 3 6 254 256 261 260 50 3 7 252 251 243 246 50 3 8 242 237 241 243 50 3 9 245 244 248 248 50 3 10 246 238 257 262

Run 4

Deposition Deposition Sample Location 1, Location 1, Location 2, Location 2, Rate [nm / Time Measurement Measurement Measurement Measurement mi] [min] n 2 [nm] I [nm] 2 [nn] 50 2 1 152 155 155 150 50 2 2 154 155 159 156 50 2 3 140 142 141 144 50 2 4 156 151 153 156 50 2 5 163 160 159 163

Run 5

Deposition Deposition Sample Location 1, Location 1, Location 2, Location 2, Rate [nm / Time Measurement Measurement Measurement Measurement min] [min] I [nm] 2 [tim] I [nm] 2 [nm] 50 4 1 333 335 333 333 50 4 2 339 338 342 339 50 4 3 338 336 334 335 50 4 4 333 330 331 331 50 4 5 330 330 327 325

91 Run 6

Deposition Deposition Sample Location 1, Location 1, Location 2, Location 2, Rate [nm / Time Measurement Measurement Measurement Measurement min] [min] I [nm] 2 [nm m] 2 [nm] 100 2 1 165 162 170 172 100 2 2 164 162 169 162 100 2 3 159 163 159 161 100 2 4 159 169 163 167 100 2 5 166 167 162 168

Run 7

Deposition Deposition Sample Location 1, Location 1, Location 2, Location 2, Rate [nm / Time Measurement Measurement Measurement Measurement min] [min] 1 [nm] 2 [nm] 1 n 2 [nm] 100 4 1 351 349 348 344 100 4 2 353 355 357 355 100 4 3 364 363 364 364 100 4 4 354 354 356 355 100 4 5 364 366 364 365

92 A.2 Electric Testing of Experimental Frame Bending Fixture

Open circuit electric testing of buckled beam MEMS device mounted on experimental frame bending fixture.

Voltage Measurement 0.5

0.45

0.4

0.35

0.3 AL b Cu 0.25 'SN 0.2

0.15

0.1 V 0.05

0 10 20 30 40 50 60 70 80 90 100 Frequency (Hz)

Figure A.2-1 Open Circuit Peak Voltage at varied operating Frequency

Proof Mass Displacement 0.6

0.5

E 0.4 E

-. 3 E0 0.3 U CL 0.2 ee-100,

0.1

0 10 20 30 40 50 60 70 80 90 100 Frequency (Hz)

Figure A.2-2 Proof Mass Displacement at varied operating Frequency

93 Bibliography

1. Ray Xu. Low-Frequency, Low Amplitude MEMS Vibration Energy Harvesting. PhD thesis, Massachusetts Institute of Technology, 2010. 2. Katherine Marie Smyth. Piezoelectric Micro-machined Ultrasonic Transducersfor Medical Imaging. PhD thesis, Massachusetts Institute of Technology, 2017. 3. Stephen D. Senturia. MicrosYstem Design. Kluwer Academic Publishers, 2001. 4. Jaffe, Bernard. Piezoelectricceramics. Vol. 3. Elsevier, 2012. 5. Kholkin, A. L., N. A. Pertsev, and A. V. Goltsev. "Piezoelectricity and crystal symmetry." Piezoelectric and Acoustic Materials for Transducer Applications. Springer, Boston, MA, 2008. 17-38. 6. Beeby, S. P., M. J. Tudor, and N. M. White. "Energy harvesting vibration sources for microsystems applications." Measurement science and technology 17.12 (2006): R175. 7. A. Hajati and S.-G. Kim, "Ultra-wide bandwidth piezoelectric energy harvesting," Appl. Phys. Lett., vol. 99, no. 8, p. 83105, 2011. 8. Pillatsch, Pit, Eric M. Yeatman, and Andrew S. Holmes. "A piezoelectric frequency up-converting energy harvester with rotating proof mass for human body applications." Sensors and Actuators A: Physical 206 (2014): 178-185. 9. Edamoto, M., et al. "Low-resonant-frequency micro electret generator for energy harvesting application." Micro Electro Mechanical Systems, 2009. MEMS 2009. IEEE 22nd International Conference on. IEEE, 2009. 10. F. Cottone, et al., "Piezoelectric Buckled Beams for Random Vibration Energy Harvesting," Smart Materials and Structures, 21(3):035021, 2012. 11. R. Xu, and S.-G. Kim, "Modeling and Experimental Validation of Bi-Stable Beam Based Piezoelectric Energy Harvester." Energy Harvesting and Systems 3.4 (2016): 313- 321. 12. Y B Jeon, R Sood, J Jeong, and S Kim. MEMS power generator with transverse mode thin film PZT. Sensors and Actuators A: Physical, 122(1):16-22, jul 2005.

94 13. R. Xu, H. Akay, S-G. Kim, Micro Buckled Beam Based Ultra-Low Frequency Vibration Energy Harvester. Hilton Head Solid-State Sensors, Actuators and Microsystems Workshop. 2018 14. Arman Hajati. Ultra Wide-Bandwidth Micro Energy Harvester. PhD thesis, Massachusetts Institute of Technology, 2010. 15. Montgomery, Douglas C. Statistical qualit control. Vol. 7. New York: Wiley, 2009. 16. Yeatman, E.; Mitcheson, P. Energy Harvesting and Power Delivery. In Body Sensor Networks; Yang, G.-Z., Ed.; Springer: London, UK, 2014; pp. 237-272. 17. Paradiso, J.A.; Starner, T. Energy scavenging for mobile and wireless electronics. IEEE Pervasive Comput. 2005, 4, 18-27. 18. Fu, H.; Cao, K.; Xu, R.; Bhouri, M.; Martinez-Botas, R.; Kim, S.G.; Yeatinan, E. Footstep Energy Harvesting Using Strike-Induced Airflow for Human Activity Sensing. In Proceedings of the 2016 IEEE 13th International Conference on Wearable and Implantable Body Sensor Networks, San Francisco, CA, USA, 14-17 June 2016. 19. Fu, H.; Xu, R.; Seto, K.; Yeatman, E.M.; Kim, S.G. Energy Harvesting from Human Motion Using Footstep-Induced Airflow. 1. Phys. Conf Ser. 2015, 660, 012060. 20. Shenck, N.S.; Paradiso, J.A. Energy scavenging with shoe-mounted piezoelectrics. IEEE Micro 2001, 21, 30-42. 21. Moro, L.; Benasciutti, D. Harvested power and sensitivity analysis of vibrating shoe- mounted piezoelectric cantilevers. Smart Mater. Struct. 2010, 19, 115011. 22. Akay, H.; XLi, R.; Seto, K.; Kim, S.-G. Energy Harvesting Footwear. U.S. Patent No.62/449208, filed October 2017. 23. Grieve, D.W.; Gear, R.J. The Relationships between Length of Stride, Step Frequency, Time of Swing and Speed of Walking for Children and Adults. Ergonomics 2007, 9, 379-399, doi: 10.1080/00140136608964399. 24. Hughes, J.; Clark, P.; Jagoe, R.R.; Gerber, C.; Klenerman, L. The pattern of pressure distribution under the weightbearing forefoot. Foot 1991, 1, 117-124.

95 25. Perrucci, G.P.; Fitzek, F.H.P.; Widmer, J. Survey on Energy Consumption Entities on the Smartphone Platform. In Proceedings of the 2011 IEEE 73rd Vehicular Technology Conference (VTC Spring), Yokohama, Japan, 15-18 May 2011; pp. 1- 6, doi:10.1 109/VETECS.2011.5956528.

96