Piezoelectric Micromachined Ultrasonic Transducers

0 downloads 0 Views 19MB Size Report
Over the past two decades, a variety of physical mechanisms have been exploited to ...... [38] D. Blackstock, Fundamentals of Physical Acoustics. : John Wiley ...
Piezoelectric Micromachined Ultrasonic Transducers for Fingerprint Sensing By YIPENG LU B.S. (Jilin University) 2007 M.S. (Shanghai Jiao Tong University) 2010 M.S. (University of California, Davis) 2015 Dissertation Submitted in partial satisfaction of the requirements for the degree of DOCTOR OF PHILOSOPHY in Mechanical and Aerospace Engineering in the OFFICE OF GRADUATE STUDIES of the UNIVERSITY OF CALIFORNIA DAVIS Approved: _______________________________________ Professor David A. Horsley, Chair _______________________________________ Professor Nesrin Sarigul-Klijn _______________________________________ Professor Tingrui Pan Committee in Charge 2015

i

Piezoelectric Micromachined Ultrasonic Transducers for Fingerprint Sensing

Copyright © 2015

by

Yipeng Lu

Yipeng Lu May 2015 Mechanical and Aerospace Engineering

Piezoelectric Micromachined Ultrasonic Transducers for Fingerprint Sensing Abstract Fingerprint identification is the most prevalent biometric technology due to its uniqueness, universality and convenience. Over the past two decades, a variety of physical mechanisms have been exploited to capture an electronic image of a human fingerprint. Among these, capacitive fingerprint sensors are the ones most widely used in consumer electronics because they are fabricated using conventional complementary metal oxide semiconductor (CMOS) integrated circuit technology. However, capacitive fingerprint sensors are extremely sensitive to finger contamination and moisture. This thesis will introduce an ultrasonic fingerprint sensor using a PMUT array, which offers a potential solution to this problem. In addition, it has the potential to increase security, as it allows images to be collected at various depths beneath the epidermis, providing images of the sub-surface dermis layer and blood vessels. Firstly, PMUTs are analyzed using FEM and analytical method to obtain maximized PMUT sensitivity by optimizing the layer stack and electrode design, and the increased coupling coefficient via series transduction. Moreover, a broadband PMUT with 97% fractional bandwidth is achieved by utilizing a thinner structure excited at two adjacent mechanical vibration modes with overlapping bandwidth. In addition, we proposed waveguide PMUTs, which function to direct acoustic waves, confine acoustic energy, and provide mechanical protection for the PMUT array.

ii

Furthermore, PMUT arrays were fabricated with different processes to form the membrane, including front-side etching with a patterned sacrificial layer, front-side etching with additional anchor, cavity SOI wafers and eutectic bonding. Additionally, eutectic bonding allows the PMUT to be integrated with CMOS circuits. PMUTs were characterized in the mechanical, electrical and acoustic domains. Using transmit beamforming, a narrow acoustic beam was achieved, and high-resolution (sub-100 µm) and short-range (~1 mm) pulse-echo ultrasonic imaging was demonstrated using a steel phantom. Finally, a novel ultrasonic fingerprint sensor was demonstrated using a 24×8 array of 22 MHz PMUTs with 100 µm pitch, fully integrated with 180 nm CMOS circuitry through eutectic wafer bonding. Each PMUT is directly bonded to a dedicated CMOS receive amplifier, minimizing electrical parasitics and eliminating the need for through-silicon vias. Pulse-echo imaging of a 1D steel grating is demonstrated using electronic scanning of a 20×8 sub-array, resulting in 300 mV maximum received amplitude and 5:1 contrast ratio. Because the small size of this array limits the maximum image size, mechanical scanning was used to image a 2D PDMS fingerprint phantom (10 mm by 8 mm) at a 1.2 mm distance from the array.

iii

Acknowledgements Many thanks to my research advisor Prof. David A. Horsley for guiding my research, especially, his big picture insight and valuable advices. I would also like to thank Prof. Bernhard E. Boser for his helpful advice and time contribution in our collaboration. Also I would like to thank my other two thesis committee members, Prof. Nesrin Sarigul-Klijn and Prof. Tingrui Pan, for reviewing the manuscript and helpful suggestions. Thanks to all my labmates at UCD MEMSlab. In particular I thank Andre Guedes and Stefon Shelton for help on device fabrication, Ofer Rozen for helpful discussion, and Stephanie Fung and Qi Wang for help on device characterization. Also I would like to thank Dr. Richard Przybyla at UC Berkeley for helpful discussion. Special thanks to my main collaborator HaoYen Tang at UC Berkeley, who designed ASIC chips which allowed me test my devices much more easily. It was a pleasure to collaborate with such a pleasant and efficient partner. Thanks to the Berkeley Sensor and Actuator Center (BSAC). It was a valuable experience to be able to discuss our work with BSAC industry members. Also many thanks for research funding from BSAC industry members, including Invensense, Qualcomm, Murata and Capella. Special thanks to Julius Tsai and Mike Daneman at Invensense, and Dr. Ronald Polcawich and Gabriel Smith at US Army Research Lab, for helpful discussion and device fabrication. Finally, I would like to thank my family for their support and encouragement. To my parents, Yuxia and Fengjin Lu, my brother, Yizhi, and my fiancéé, Lizhi Tao: I would not be where I am today without you.

iv

Table of Contents Abstract.......................................................................................................................................... ii Acknowledgements ...................................................................................................................... iv Table of Contents .......................................................................................................................... v List of Figures............................................................................................................................. viii 1.

Introduction ........................................................................................................................... 1 1.1. Fingerprint sensor ....................................................................................................... 1 1.1.1. Optical fingerprint sensor ........................................................................................... 1 1.1.2. Pressure fingerprint sensor .......................................................................................... 2 1.1.3. Thermal fingerprint sensor .......................................................................................... 3 1.1.4. Capacitive fingerprint sensor ...................................................................................... 4 1.1.5. Ultrasonic fingerprint sensor....................................................................................... 6 1.2. Ultrasonic Pulse-Echo Imaging .................................................................................. 8 1.3. Ultrasonic transducers ................................................................................................. 9 1.3.1. Conventional ultrasonic transducer............................................................................. 9 1.3.2. CMUT ....................................................................................................................... 10 1.3.3. PMUT ....................................................................................................................... 11

2.

PMUT Design and Modeling.............................................................................................. 14 2.1. Vibration and resonant frequency ............................................................................. 14 2.2. Piezoelectric coupling and equivalent circuit ........................................................... 15 2.3. Film Thickness Optimization .................................................................................... 16 2.4. Electrode Area Optimization .................................................................................... 19 2.5. Acoustic pressure directivity..................................................................................... 21

3.

High fill-factor annular PMUT array ............................................................................... 23 3.1. Background ............................................................................................................... 23

v

3.2. PMUT array design ................................................................................................... 24 3.3. Fabrication ................................................................................................................ 27 3.4. Characterization ........................................................................................................ 31 3.5. Summary ................................................................................................................... 36 4.

Ribbon shape PMUT array ................................................................................................ 38 4.1. Background ............................................................................................................... 38 4.2. PMUT Design ........................................................................................................... 39 4.3. Characterization ........................................................................................................ 43 4.3.1. Mechanical domain characterization ........................................................................ 44 4.3.2. Electrical domain characterization ............................................................................ 47 4.3.3. Acoustic domain characterization ............................................................................. 49 4.4. Summary ................................................................................................................... 52

5.

Broadband PMUTs with additional anchor dual resonance modes .............................. 54 5.1. Background ............................................................................................................... 55 5.2. Design and fabrication .............................................................................................. 56 5.3. PMUTs with dual resonance modes ......................................................................... 61 5.3.1. Vibration of rectangular membrane .......................................................................... 61 5.3.2. Thin PMUT membrane for broad bandwidth ........................................................... 64 5.3.3. Measurement results ................................................................................................. 65 5.4. Summary ................................................................................................................... 71

6.

PMUT array based on cavity SOI wafers ......................................................................... 73 6.1. Background ............................................................................................................... 73 6.2. Fabrication ................................................................................................................ 74 6.3. Characterization ........................................................................................................ 79 6.3.1. Mechanical Domain .................................................................................................. 79 6.3.2. Electrical Domain ..................................................................................................... 83 6.3.3. Acoustic Domain ...................................................................................................... 85 6.3.4. Summary ................................................................................................................... 88

7.

Pulse-echo ultrasonic imaging using cavity SOI PMUTs ................................................ 90 vi

7.1. Background ............................................................................................................... 90 7.2. PMUT Design and Modeling .................................................................................... 92 7.2.1. AlN vs. PZT .............................................................................................................. 92 7.2.2. Acoustic beam pattern of a single transducer ........................................................... 93 7.2.3. PMUT array .............................................................................................................. 95 7.2.4. Beam-forming and scanning ..................................................................................... 96 7.3. Short range pulse echo imaging experiment results ................................................. 98 7.4. Long-range fat sensor ............................................................................................. 107 7.5. Summary ................................................................................................................. 108 8.

PMUT with Increased Coupling Coefficient via Series Transduction......................... 110 8.1. Background and PMUT Design .............................................................................. 110 8.2. Results and Discussion ........................................................................................... 112

9.

Wave-guide PMUTs .......................................................................................................... 115 9.1. Background ............................................................................................................. 116 9.2. Device Design ......................................................................................................... 117 9.3. Device characterization ........................................................................................... 121 9.4. Summary ................................................................................................................. 124

10.

Fingerprint sensor......................................................................................................... 125 10.1. Background ............................................................................................................. 125 10.2. Design of Ultrasonic Fingerprint based on PMUTs ............................................... 126 10.3. Characterization ...................................................................................................... 129 10.4. Fingerprint Pulse-echo Imaging .............................................................................. 133 10.5. Package of the fingerprint sensor ............................................................................ 135

11.

Conclusions .................................................................................................................... 137

References .................................................................................................................................. 139

vii

List of Figures Fig.1-1 System diagram of an optical fingerprint sensor. ............................................................... 2 Fig.1-2 Pressure-based fingerprint sensor with T-shaped protrusions. ......................................... 3 Fig.1-3 Sensing principle of a thermal fingerprint sensor: (a) arrayed resistive microheater elements; (b) sensing principle. .............................................................................................. 4 Fig.1-4 The conceptual model of a capacitive fingerprint sensor. .................................................. 5 Fig.1-5 Scanning mechanism of ultrasonic fingerprint sensor. ...................................................... 7 Fig.1-6 Ultrasonic fingerprint sensor based on micromachined ultrasonic transducers (MUTs). .. 8 Fig.1-7 (a) cross section of a single PMUT, and out of plane deflection of the PMUT with (b) positive and (c) negative voltage inputs when functioning as a transmitter. ........................ 12 Fig.2-1. The first (a) and second (b) mode shapes of the PMUT obtained from FEM model (COMSOL Multiphysics). .................................................................................................... 15 Fig.2-2 Equivalent circuit model of a PMUT including electrical, mechanical and acoustic domains. ................................................................................................................................ 16 Fig.2-3 Diagram of axis-symmetric unimorph plate..................................................................... 17 Fig.2-4 Sum of stresses in radial and tangential directions at r=a=12.5µm inside the AlN layer. Good agreement is observed between the analytical and FEM models................................ 18 Fig.2-5 FEM simulated electrical potential in the middle of the piezoelectric layer at the center of the membrane with various thickness of the piezoelectric layer on a fixed 2.5 µm Si layer under a uniform 100 kPa pressure. ....................................................................................... 19

viii

Fig.2-6 FEM simulated stresses in the radial and tangential direction of a 50 µm diameter AlN PMUT with 1 Pa uniform load on its surface. The sum of the stresses is equal to zero at = 2/2, the optimal electrode radius. .......................................................................... 20 Fig.2-7 Piezoelectric coupling versus electrode radius ratio. ....................................................... 21 Fig.2-8 Analytical directivity of transducers with different diameters. ........................................ 22 Fig.3-1 Optical image of the PMUT array and close-up picture of individual 25 µm diameter PMUTs. Left: the 8 channels are connected to individual bond-pads through the topelectrode metal and a single bond-pad is connected to the common bottom electrode. Right: The sacrificial poly-Si is removed via 2µm by 4 µm etch holes between each PMUT. The rings of PMUTs are connected through the top-electrode metal layer (dark gold) and bottom electrode (white). .................................................................................................................. 25 Fig.3-2 Schematic diagram of focal depth control of the PMUT array. The 8 annular channels are shown in cross-section. ................................................................................................... 26 Fig.3-3 Fabrication process flow. ................................................................................................. 28 Fig.3-4 Cross-section diagram and optical images of two etching hole designs: 2 µm x 4 µm etching holes connected to the PMUT through a buried polysilicon channel (a, c) and 2 µm diameter etching holes located in the center of the PMUT (b, d). ........................................ 29 Fig.3-5 Confocal laser microscope measurements showing a 2 µm by 4 µm etch hole (a) before and (b) after Parylene sealing and (c) height profile measurements. .................................... 30 Fig.3-6 SEM images: (a) Cross-section of Parylene sealed PMUT and (b) the close up picture of buried channel and etching hole. .......................................................................................... 31 Fig.3-7 Frequency response measured in air showing 0.8% frequency mismatch across the PMUT array. ......................................................................................................................... 32

ix

Fig.3-8 FEM and measured results of center frequency in air of PMUTs (750 nm AlN and 800 nm SiO2) with various diameter. .......................................................................................... 32 Fig.3-9 Frequency response of a single PMUT measured in air before and after Parylene sealing. ............................................................................................................................................... 33 Fig.3-10 Electrical impedance of a 35 µm diameter PMUT (750 nm AlN /800 nm SiO2) with no Parylene coating. ................................................................................................................... 34 Fig.3-11 Pulse responses of displacement measurements of a 25 µm diameter (750 nm AlN/800 nm SiO2) PMUT, (a) in air, (b) in fluid (Fluorinet-70). ....................................................... 35 Fig.3-12 Measured frequency response in fluid (Fluorinert-70) with velocity converted to pressure on the surface of PMUTs. ....................................................................................... 36 Fig.3-13 Simulated pressure distribution based on the measured PMUT parameters (a) without focus control, (b) focused at 1 mm axial distance, and (c) focused at 1.5 mm axial distance. ............................................................................................................................................... 36 Fig.4-1 3D diagram of PMUT ribbon array, (b) close-up picture. ............................................... 39 Fig.4-2 The first 5-order resonant modes of a 40 µm×200 µm PMUT. ....................................... 40 Fig.4-3 Harmonic FEM analysis with 1 V amplitude applied to the AlN layer showing the predicted frequency response and corresponding mode-shapes. .......................................... 41 Fig.4-4 Resonant frequency (a) and static displacement (b) versus thickness of the SiO2 elastic layer....................................................................................................................................... 43 Fig.4-5Optical images of the 1×8 40 µm×200 µm PMUT array. ................................................. 44 Fig.4-6 Measured displacement amplitude (a) and phase (b) frequency response of a 40 µm×200 µm PMUT (0.75 µm AlN/ 0.8 µm SiO2) in air. ................................................................... 45

x

Fig.4-7 Displacement frequency response of PMUTs from three wafers (W1, W2, W3) having different film thicknesses. ..................................................................................................... 46 Fig.4-8 (a) 100 Vpp pulse displacement response measurement of a 40 µm×200 µm PMUT with 1500 nm AlN/200 nm SiO2; (b) FFT of time response signal in (a); (c) FFT of time response of a PMUT driven with 5 Vpp input. ..................................................................... 47 Fig.4-9 (a) Electrical impedance measurement setup and (b) measurement of a single 40 µm×200 µm PMUT with 750 nm AlN/800 nm SiO2. ........................................................................ 48 Fig.4-10 (a) Time-domain pressure measured using a hydrophone and (b) FFT of the time response................................................................................................................................. 51 Fig.4-11 Pulse-echo measurement set-up (a) and measured pulse response (b). The distance between the array and the fluid-air interface is 2.2 mm. ...................................................... 52 Fig. 5-1: Laser confocal microscope images of a single PMUTs with thick metal additional anchor. ................................................................................................................................... 57 Fig. 5-2: Fabrication process flow. ............................................................................................... 58 Fig. 5-3: SEM image of plated gold air bridges............................................................................ 59 Fig. 5-4: Measured frequency response in air showing the 7% frequency variation that occurs with 60% overetch. ............................................................................................................... 60 Fig. 5-5: Measured frequency variation versus overetch percentage. .......................................... 61 Fig. 5-6: FEM simulation results of the first 5-order resonant mode shapes of a fully clamped 30 µm × 200 µm membrane....................................................................................................... 62 Fig. 5-7: FEM simulation results of resonant frequencies for different modes of a rectangular PMUT with different length (the other side is fixed, 30 µm). .............................................. 64

xi

Fig. 5-8: Image of a single 30µm×200µm PMUT. Inset: close-up showing the top electrode and 3 µm etching holes. ............................................................................................................... 65 Fig. 5-9: Measured displacement frequency response in air shows that the odd harmonics ((1, 1), (1, 3), …) modes are excited. ................................................................................................ 66 Fig. 5-10: (a) Impedance measrurement setup; (b)Measured electrical impedance of the PMUT. ............................................................................................................................................... 67 Fig. 5-11: Frequency response of PMUT displacement sensitivity amplitude (a) and phase (b) in air with various dc bias. ........................................................................................................ 68 Fig. 5-12: PMUT resonant frequency (a) and displacement sensitivity at resonance (b) measured in air versus dc bias. .............................................................................................................. 69 Fig. 5-13: (a) system diagram and (b) optical image of measurement setup for acoustic pressure. ............................................................................................................................................... 70 Fig. 5-14: Measured acoustic pressure in fluid (FC-70) from a single 30µm×200µm PMUT demonstrates a large 97% fractional bandwidth at 3.7 MHz center frequency due to the thin PMUT membrane and the excitation of the (1, 1) and (1, 3) modes, which have overlapping bandwidths. ........................................................................................................................... 71 Fig.6-1 3D schematic diagram of a PMUT array based on cavity SOI wafers; (b) optical images of the fabricated 72×9 PMUT array. ..................................................................................... 75 Fig.6-2 Fabrication process flow of PMUT array based on cavity SOI wafers. ........................... 76 Fig.6-3 SEM cross section images of a 50 µm AlN PMUT showing the vacuum cavity beneath the device layer Si. ................................................................................................................ 78 Fig.6-4 (a) X-ray diffraction (XRD) measurement of the PZT film on Pt bottom electrode; (b) rocking curve measurement showing 1.3° full width at half magnitude (FWHM). ............. 79

xii

Fig.6-5 Frequency response measured in air: PZT PMUTs (1.1 µm PZT/ 2.5 µm Si) vs. AlN PMUTs (0.8 µm AlN/ 2.5 µm Si). ........................................................................................ 80 Fig.6-6 Measured and FEM (a) resonant frequency and (b) static displacement of PZT and AlN PMUTs in air versus radius................................................................................................... 82 Fig.6-7 Displacement frequency response in air of 50 um PZT PMUTs (a) before and (b) after poling. ................................................................................................................................... 83 Fig.6-8 Impedance measurement of a single 50 µm PMUT with (a) PZT and (b) AlN piezoelectric layers. PZT PMUTs show a high coupling constant kt2 = 12.5% and 50 Ω electrical impedance around the resonant frequency, well-matched to circuits. .................. 85 Fig.6-9 The acoustic pressure generated by PMUTs excited with a 4-cycle 10 MHz 25 Vpp input, demonstrating that a 250 µm thick PDMS layer has minimal effect on the pressure amplitude............................................................................................................................... 86 Fig.6-10 Measured pressure map of a 15x9 array driven at 10 MHz with 18 Vpp using phasedarray beam forming. .............................................................................................................. 87 Fig.6-11 Pressure measurements made with and without beam forming. .................................... 88 Fig. 7-1: Vector diagram for beam pattern calculation. ............................................................... 93 Fig. 7-2: (a) Calculated beam patterns at 1.5 mm away from a single 50 um transducer with various working frequencies; (b) calculated beam patterns at 1.5 mm away from an 8 MHz transducer with various diameter. ......................................................................................... 94 Fig. 7-3: Acoustic on-axis pressure pattern of an 8 MHz transducer with various diameter. ..... 95 Fig. 7-4: Optical images of the fabricated 72×9 PMUT array. ..................................................... 96

xiii

Fig. 7-5: Beam-forming PMUT array is used to achieve narrow acoustic beam at lower frequency, 8 MHz, where low acoustic attenuation makes the beam-forming array favorable. ............................................................................................................................................... 97 Fig. 7-6: Beamforming using groups. The beam-forming pitch is shown as twice the PMUT pitch but may be any integer multiple. The focused spot is translated by switching from the 1st group of odd-numbered PMUTs (blue) to the 2nd group of even numbered PMUTs (red). The scan step is equal to the PMUT pitch. ........................................................................... 98 Fig. 7-7: Acoustic pressure measurement setup using a needle hydrophone. .............................. 99 Fig. 7-8: Measured pressure from a 15×9 PMUT array with 70 µm (a) and 140 µm (b) beamforming pitch. ...................................................................................................................... 100 Fig. 7-9: FFT result of the ringing down part of the measured pressure signal. ......................... 101 Fig. 7-10: (a) Measured acoustic beam-pattern compared with simulation results for a 15-element PMUT array with 140 µm and 70 µm beam-forming pitch; (b) 8 MHz transmit pressure measured using a hydrophone; 2-D acoustic pressure pattern in x-z plane (PMUT array in xy plane)................................................................................................................................ 102 Fig. 7-11: (a) System diagram of pulse-echo imaging using AlN cavity SOI PMUT array and 1.8V 180 nm CMOS ASIC interface; (b) optical image of the ASIC. ............................... 103 Fig. 7-12: 1D B-scan pulse-echo imaging of a tilted steel phantom: inset is the used phantom. 104 Fig. 7-13: Pulse-echo time response resulting from (a) steel strips and (b) the gap between steel strips. ................................................................................................................................... 105 Fig. 7-14: Measured pulse-echo ultrasonic C-scan image of a 2-D steel phantom. ................... 106 Fig. 7-15: Pulse-echo time responses with and without beam-forming resulting from the steel phantom at steel and the gap between steel. ....................................................................... 107

xiv

Fig. 7-16: (a) Pulse echo response and reconstructed image from a tissue phantom; (b) measured thickness of fat layer versus phantom [67]. ........................................................................ 108 Fig.8-1 Cross-section of a single PMUT with differential driving. ............................................ 110 Fig.8-2 PMUT operation principle in transmit mode: voltage between electrodes TE-1 and TE-2 generates opposite electric field beneath them, which results in opposite stresses inside the film and bending moment to deflect the membrane. .......................................................... 112 Fig.8-3 Fabrication process flow based on cavity SOI wafers with only 1 more mask to pattern top electrode (TE). .............................................................................................................. 113 Fig.8-4 (a) top-view optical image and (b) cross-section SEM image of the proposed PMUT. 113 Fig.8-5: Measured frequency responses of displacement sensitivity and electrical impedance. 114 Fig.9-1 Cross-section diagram of ultrasound imager. The MEMS wafer and CMOS wafer are bonded together using Al-Ge eutectic wafer bonding. Each column of 8 PMUTs shares a common top electrode (TE) contact while individual bottom electrode (BE) contacts connect each PMUT to a dedicated receive amplifier. Waveguides etched into the 220 µm thick Si handle layer of the MEMS wafer serve to direct the ultrasound towards the imaging target. .................................................................................................................................. 118 Fig.9-2 Cross-section scanning electron microscope (SEM) images of a single waveguide PMUT. ............................................................................................................................................. 119 Fig.9-3 Optical images of the waveguide PMUT array. Top: complete device showing the 8×24 array. The pitch between adjacent PMUTs is 100 µm. Bottom: image of the MEMS die after being debonded from the CMOS die. Each 8-PMUT column shares a common TE contact while individual BE contacts are used for each PMUT. ..................................................... 119

xv

Fig.9-4 2D axisymmetric (the axis is perpendicular to the PMUT membrane) FEM simulation results: the absolute acoustic pressure from a PMUT without (a) and with (b) waveguide. ............................................................................................................................................. 121 Fig.9-5 (a) Measured waveguide PMUT pulse-echo response produced from a steel target immersed in Fluorinert FC-70; (b) 2-D pulse-echo ultrasonic image of a steel phantom obtained using a 4×20 subarray. ......................................................................................... 123 Fig.10-1. Cross-section system diagram. .................................................................................... 128 Fig.10-2. (a) Optical images of the 24×8 PMUT array after de-bonding to remove the CMOS wafer; (b) cross-section SEM images of a single PMUT after partial de-bonding to remove the MEMS wafer. ................................................................................................................ 128 Fig.10-3. System diagram of pulse echo imaging with each PMUT accessible and separate driving and sensing electrodes [81]. ................................................................................... 129 Fig.10-4. (a) Mode-shape of a 5×5 sub-array measured at 28.4 MHz using a scanning LDV; (b) displacement frequency response of the 25 PMUTs. Inset: histogram of the peak displacement sensitivity for the 25 PMUTs tested. ............................................................ 131 Fig. 10-5 Measured acoustic pressure using a needle hydrophone. ............................................ 132 Fig.10-6. 1-D Pulse-echo ultrasonic imaging of a steel phantom using electronic scanning: inset pictures are the time-domain pulse-echo response and an optical image of the steel phantom. ............................................................................................................................................. 134 Fig. 10-7 (a) Fingerprint phantom made from patterned PDMS sealed with a PDMS sheet; (b) 2D pulse-echo ultrasonic image of the PDMS fingerprint phantom..................................... 135 Fig. 10-8 Transmission ratio verses hard layer thickness. .......................................................... 136

xvi

1. Introduction

1.1. Fingerprint sensor Biometric identifiers are the distinctive, measurable characteristics used to distinguish individuals, such as fingerprints, palm prints, faces, DNA, irises, retinas, voice and others. Among them, fingerprints are the most commonly used due to their uniqueness, universality and convenience. Fingerprint ridge details are generally described at two different levels: pattern and minutia. Fingerprint patterns are the macro details of the fingerprint such as arch, loop and whorl; fingerprint minutiae are ridge bifurcations and endings [1]. Inking is the traditional way of fingerprint capturing, which is inconvenient and time consuming for digitization. In contrast, the live scan fingerprint sensor can capture a digital fingerprint image in real time [2]. Over past decades, various mechanisms have been employed to capture an electronic fingerprint image, including optical, capacitive, pressure and acoustic mechanisms [3]. This section will introduce different kinds of fingerprint sensors and their advantages and disadvantages. Especially, we will present an ultrasonic fingerprint sensor based on pulse-echo ultrasonic imaging.

1.1.1. Optical fingerprint sensor Optical fingerprint capture is typically based on the frustrated total internal reflection (FTIR) phenomenon, as illustrated in Fig.1-1 [2, 4]. The prism or platen is designed to have similar refractive index to that of human tissue. Therefore, when a finger touches the platen, the light that passes through the glass upon valleys (air on the glass surface) is totally reflected, while the light that passes upon ridges is only partially reflected. Then the reflected light is focused by a 1

lens onto a camera and the fingerprint image is captured [2]. The optical sensor will not be deceived by presentation of a photograph or printed image of a fingerprint from a camera, since FTIR images a three-dimensional surface. But the sensor itself is complicated and has bulky optics system which makes it very hard to integrate with portable devices. Moreover, it will have poor image quality due to excessively dry fingers, which reduce their refractive index [5].

Fig.1-1 System diagram of an optical fingerprint sensor.

1.1.2. Pressure fingerprint sensor Many research groups have been working on miniature fingerprint sensors which have the potential to be integrated with portable devices, such as laptop or smart phones. One example is a fingerprint sensor based on pressure sensing mechanism, which generally utilizes piezoelectric effect or sensing capacitance variation [6]. As shown in Fig.1-2, each pixel of the sensor is a pressure sensor with a cavity structure. This device is CMOS-process compatible and each pixel is easily accessed with supporting sensing circuits beneath MEMS cavity structures. The pressure sensor consists of top and bottom electrodes and a T-shape protrusion on the top electrode. When a finger is placed on the sensor surface, fingers’ ridges will be contact with the top electrode, and the generated force will bend the membrane, reduce the distance between top

2

and bottom electrodes and therefore increase the capacitance between electrodes, while the capacitance of pixels beneath fingers’ valley keeps constant. The capacitance variation can be detected using the integrated circuit and then converted to a fingerprint image [6]. This kind of sensors are easily integrated with electronics and therefore compatible with portable devices. Moreover, they are not sensitive finger humid condition, dry or wet. However, they are fragile and a protection a layer is required, which will blur the image.

Fig.1-2 Pressure-based fingerprint sensor with T-shaped protrusions.

1.1.3. Thermal fingerprint sensor The sensing principle of a thermal fingerprint sensor is shown in Fig.1-3 [7]. The device has a dense array of heaters and temperature dependent resistors. Thermal insulation layer or cavity is used to provide thermal isolation from the substrate. When a finger is placed on the sensor surface, the heater elements in contact with ridges of the fingerprint will show less of temperature rise than those in contact with the valleys, because of the different thermal conductivities between the bridges and air filled in the valleys. Then the temperature-rise differences will cause variation in electrical resistance, which can be can be easily converted into electrical signals by circuits. However, after a short time, the image will vanish after the chip and 3

finger reaching at thermal equilibrium. Moreover, these sensors tend to be less accurate in hot environments, when the temperature difference is not high between the ridges and the valley [3].

Fig.1-3 Sensing principle of a thermal fingerprint sensor: (a) arrayed resistive microheater elements; (b) sensing principle.

1.1.4. Capacitive fingerprint sensor Among these fingerprint sensors, capacitive fingerprint sensors[8] are the ones most widely used in consumer electronics because they are fabricated using conventional complementary metal oxide semiconductor (CMOS) integrated circuit technology [9]. The conceptual model of a capacitive sensing scheme is shown in Fig.1-4[10].The finger is modeled as the upper electrode of the capacitor, and the metal plate in the sensor cell as the lower electrode. These two electrodes are separated by the passivation layer of the silicon chip and air.

4

Fig.1-4 The conceptual model of a capacitive fingerprint sensor. The sensor consists of an array of 2-D array of sensing electrodes using standard CMOS processing covered by a thin dielectric layer. By measuring capacitive difference between electrodes and a finger placed on the sensor, ridges and valleys of a fingerprint pattern can be resolved, due to relative permittivity difference between human tissue of finger’s bridge and air filled in the finger’s valley. However, capacitive fingerprint sensors are extremely sensitive to contamination and moisture on the finger. In reality, the real conductive layer to be the upper electrode is the live skin cell layer beneath the epidermis layer with dead skin cell. The permittivity of the air and tissue depend on its relative humidity (RH) and temperature, and is very sensitive to the ambient conditions. For example, the permittivity of the air is expressed in the following equation [11]:

=

where

is permittivity of vacuum,

is pressure of air (mmHg), and

[1 +

211

( +

48

)10 ]

is absolute temperature (K),

is relative humidity (%),

is pressure of the saturated water at temperature T.

5

1.1.5. Ultrasonic fingerprint sensor Ultrasonic fingerprint sensor is based upon pulse-echo imaging due to the reflection effect of ultrasound as it propagates through mediums with various acoustic impedance. After a finger is placed on the sensor, a pulses wave with a narrow beam is launched from the sensor. Due to acoustic impedance of human tissue is much larger than that of air, the reflection ratio is much larger at the interface sensor/air than that from sensor/tissue. Therefore, the acoustic echo amplitude for the former is much larger than the latter. Through scanning the narrow acoustic beam within an area, recognizing the echo amplitude at each location, the fingerprint pattern within the area can be obtained. Note that the mechanism of pulse-echo ultrasonic imaging is discussed in the next section in detail. Compared with other fingerprint sensors, ultrasonic fingerprint sensors offer a potential solution to be insensitive to finger contamination or humid conditions, because the fingerprint’s valleys and ridges are easily distinguished due to the great difference between their acoustic impedance (430 Rayl for an air-filled valley and ~1.5 MRayl for a ridge’s human tissue). In addition, ultrasonic pulse-echo imaging has the potential to be more secure as it allows images to be collected at various depths beneath the epidermis, providing images of the sub-surface dermis layer and blood vessels.

6

Fig.1-5 Scanning mechanism of ultrasonic fingerprint sensor. A scanning mechanism for ultrasonic fingerprint sensor was reported in [12, 13], as shown in Fig.1-5. A single fixed focus high frequency ultrasonic transducer is used to launch a narrow acoustic beam. The ultrasonic beam strikes a rotating acoustic mirror which redirects the sound wave toward imaging target, and the return echoes are sent back along the same path. However, it requires a transducer with a long focal length, and therefore large transducer aperture, or high frequency to maintain a small focus size [12]. To minimizes overall transducer focal length and focus size, scanning transducer method in a plane, parallel to the imaging plane, has been proposed [12]. But this results in a more complex, higher power motion control subsystem, and low frame rate. Micromachined ultrasonic transducers (MUTs), originally developed for medical ultrasound applications, solve these problems and a fingerprint sensor based on capacitive MUTs (CMUTs) has been reported.[14] Here, we demonstrate a novel ultrasonic fingerprint sensor based on piezoelectric MUTs (PMUTs) that can be fully-integrated with CMOS circuitry. The system diagram of fingerprint sensor based on MUTs is shown in Fig.1-6. Whereas a CMUT is based on capacitive electrodes separated by a submicron vacuum gap, a PMUT consists of a solid piezoelectric capacitor, greatly simplifying fabrication and improving mechanical robustness. More details can be found in Session 10. 7

~500 µm Contamination

100 - 300 µm 75 - 150 µm Dermis layer Epidermis layer Transmitting wave

Coupling material

Echo time delay

Echo wave

Ultrasonic transducers

Fig.1-6 Ultrasonic fingerprint sensor based on micromachined ultrasonic transducers (MUTs).

1.2. Ultrasonic Pulse-Echo Imaging Ultrasonic Pulse-Echo Imaging have been used in many applications, such as nondestructive testing (NDT), ranging and velocity sensing, industrial automation, object recognition, collision avoidance, and medical imaging [15-18]. Compared with X-ray imaging, there is no ionizing radiation exposure associated with ultrasonic pulse-echo imaging. The ultrasonic image is performed based on the reflection of pulse waves from structures with various acoustic impedance. The reflection and transmission ratios, R and T, can be expressed:

where

=

(1-1)

=

(1-2)

and

are acoustic impedance of two different kinds of medium. Acoustic impedance

is equal to the product of density, , and sound of speed of the medium, , as below: =

(1-3)

8

An ultrasonic transducer can be used for both launching acoustic waves and for detecting acoustic pressure of echo waves. From the time difference between pulses waves transmitted and the echo waves (reflected pulse waves) back to the transducer, we can extract the distance of the target from the transducer given the sound of speed. Combining the above information, we can extract the material property (acoustic impedance) from the amplitude of the echo. There are different kinds of ultrasonic imaging modes, such as A-mode (amplitude mode), B-mode or 2D mode (brightness mode, imaging of a plane vertical to the transducer), C-mode (imaging of a plane parallel with the transducers), M-mode (motion mode), harmonic mode for better contrast or Doppler mode for blood flow test [19].

1.3. Ultrasonic transducers Conventional ultrasonic transducers are largely based on bulk piezoelectric ceramic with poor acoustic coupling to air or water, and are also expensive to machine into two-dimensional (2D) transducer arrays needed for 3D imaging [20, 21]. In contrast, micromachined ultrasonic transducers (MUTs) have a compliant membrane structure with low acoustic impedance for good coupling to air and liquids [22-24]. Furthermore MUTs have several other advantages over conventional ultrasonic transducers, including small element size, low power consumption, improved bandwidth, low cost, easy fabrication of large arrays with compact designs, and easy integration with supporting electronics [25-28].

1.3.1. Conventional ultrasonic transducer Ultrasonic transducers can work as both a transmitter and receiver, using converse and direct effects of piezoelectric materials, respectively. In transmit mode, the vibration would be produced due to a potential difference between piezoelectric materials, and in the receive mode, the voltage would be generated when acoustic pressure loaded on the surface. Conventional 9

ultrasonic transducers are based on thickness mode of piezoelectric block and its resonance frequency depends on the thickness, which indicates it is challenging to make transducers with difference resonant frequencies on the same wafer. Conventional ultrasonic transducers generally have high coupling coefficient due to using e33 piezoelectric effect. The coupling coefficient of conventional transducer can be calculated using is the coefficient of the stiffness tensor, and

=

, where

is piezoelectric constant,

is permittivity. Due to its high piezoelectric

coefficients, bulk piezoelectric ceramic PZT has been used for conventional ultrasonic transducers. Since the PZT itself exhibits much higher acoustic impedance (~30 MRayl) than that of biological tissue or water (~1.5 MRayl), a substantial part of the acoustic energy would be lost at the rear interface and not directed into the forward direction, resulting in poor resolution and sensitivity, if not properly matched acoustically [29]. Therefore, a matching layer is needed between PZT and biological tissue. For a continuous sinusoidal acoustic wave, a 100% transmission can be achieved when the matching layer thickness is equal to wavelength in matching layer, and acoustic impedance of matching layer and

=(

/4, quarter )

/

, where

are acoustic impedance for piezoelectric material and imaging target [30]. For a

broadband transducer, the acoustic impedance of the matching layer should be modified to be =(

)

/

. To further improve impedance matching, two or more matching layers are

needed [29]. PVDF is another common material used for thickness mode ultrasonic transducer, which has relatively smaller piezoelectric constant but better acoustic impedance matching and smaller Young’s modulus.

1.3.2. CMUT Capacitive MUT (CMUT) consists of a thin movable plate suspended over a gap. A metal

10

layer on top of the thin plate or the thin plate itself, if conductive, forms the top electrode of the capacitor and the underlying conductive substrate acts as the bottom electrode [28]. When a dc voltage is applied between the two electrodes, the movable plate is attracted toward the substrate by the electrostatic force, which is equilibrated by a mechanical restoring force due to the stiffness of the plate. In transmitting mode, driving the capacitor with an alternating voltage generates ultrasound. In receive mode, if the movable top plate is subjected to ultrasound pressure, an electrical current is generated due to the capacitance change under constant bias voltage [28]. The generated electrostatic force,

FCMUT = where

, can be calculated as below:

Co (Vdc + Vac ) 2 2 go

is dc bias,

is ac driving signal,

is the gap thickness,

is capacitance. Then, we

can conclude amplitude of this current is a function of the frequency of the incident wave, the bias voltage and the capacitance of the device. To achieve high sensitivity, generally CMUT needs a high dc bias or small gap, which causes circuit or fabrication complexity. In addition, there is pull-in effect when voltage is large enough to make the plate move more than 1/3 of the gap thickness. This could cause reliability issue when gap thickness is not uniform. To solve these problems, collapse-mode CMUTs have been studied [31]. In collapse-mode operation, the dc bias voltage is first increased beyond the collapse voltage (pull-in voltage), then reduced without releasing the collapsed membrane, and the center of the membrane is always in contact with the substrate. In the case of a circular membrane, the maximum ac displacement occurs along the ring formed between the center and the edge of the membrane [31].

1.3.3. PMUT In transmit mode the electric field between the top electrode (TE) and the bottom electrode

11

(BE) creates a transverse stress in the piezoelectric layer due to the inverse piezoelectric effect [32]. The generated stress causes a bending moment which forces the membrane to deflect out of plane, launching an acoustic pressure wave into the surrounding medium, as shown in Fig.1-7. As a receiver, an incident pressure wave deflecting the plate creates transverse stress which results in charge on the electrodes due to the direct piezoelectric effect.

Fig.1-7 (a) cross section of a single PMUT, and out of plane deflection of the PMUT with (b) positive and (c) negative voltage inputs when functioning as a transmitter. Until recently, PMUTs were less well-developed than CMUTs because thin-film piezoelectric materials technology was immature, resulting in low and poor reproducibility of piezoelectric coefficients. Compared with well-developed CMUTs, PMUTs do not require a high polarization voltage (which can exceed 190V for CMUTs [33]), to achieve the required transducer sensitivity. While the required polarization voltage of a CMUT diminishes as the 12

capacitive gap is decreased, small gaps require tight fabrication tolerances and can result in reduced manufacturing yield. PMUTs also have the advantage of higher capacitance, which results in lower electrical impedance, allowing better matching to supporting electronic circuits and less sensitivity to parasitic capacitance.

13

2. PMUT Design and Modeling

2.1. Vibration and resonant frequency The first and second vibration mode shapes were obtained by FEM, as shown in Fig.2-1, where the first mode is the best for a high acoustic coupling efficiency. According to force equilibrium in the z-direction, symmetric vibration of a uniform circular thin plate is described by [30, 34]:

D∇4 (w(r)e jωt ) + ρ

∂2 (w(r)e jωt ) =0 ∂t 2

(2-1)

where w(r) is the deflection amplitude of the plate in the z-direction at radial distance r, ω is the PMUT working frequency, ρ is the mass per unit area, and D is the flexural rigidity of the plate. D can be obtained from:

E( z) z 2 (2-2) dz 2 1-ν ( z ) where E(z) is Young’s Modulus and ( ) is Poisson’s ratio of the material at a distance z from D= 

the neutral axis, which is the point where the sum of stresses is equal to zero. The integration is conducted for z varying from the bottom to the top of the plate. Considering a fully-clamped boundary condition at the circular edge, we can obtain the first mode’s resonant frequency as follow: f1 = (3.2 a)4

D

(2-3)



ρ where a is the radius of the circular membrane. proportional to radius squared as below:

14

is proportional to thickness hand inversely

∝ ℎ⁄

(2-4)

Because of increased damping in fluid compared with that in air, the center frequency of the PMUT will be reduced. The center frequency of a clamped circular plate with a fluid load on one side [27, 35, 36] can be approximately written as: =

,

1+

0.67

(2-5)

where ρfluid is the fluid density.

(a)

(b)

Fig.2-1. The first (a) and second (b) mode shapes of the PMUT obtained from FEM model (COMSOL Multiphysics).

2.2. Piezoelectric coupling and equivalent circuit The equivalent circuit model for a PMUT is shown in Fig.2-2, where Co is PMUT electrical capacitance, Cm motional capacitance, Lm motional inductance, Zac acoustic impedance, η electrical-mechanical coupling coefficient, and Ae effective area, which is equal to 1/3 the surface area. Cm, Lm and Zac can be obtained from the following equations [37, 38]: = =

(2-6)

(2-7)

( )

15

=

(1 −

(

)

(

+

where ( ) is mode shape expression, speed in the coupled media, and =2

+

)

)

(2-8)

is density of the coupled media,

is acoustic

can be obtained: (2-9)

+

Using an energy method [22, 26, 39, 40], η can be obtained: = where

(2-10) is the effective piezoelectric coefficient, Zp is the distance of the middle of the

piezoelectric layer from neutral axis, and Ipiezo is an integral representing the piezoelectric coupling to mode shape (x): =2

∂ ∂

+

1∂ ) ∂

(2-11)

where is radius of the top electrode for the PMUT. Electrode size and layer thickness can be optimized to have maximum , which will be discussed in the following.

Fig.2-2 Equivalent circuit model of a PMUT including electrical, mechanical and acoustic domains.

2.3. Film Thickness Optimization

16

Neutral axis

Axis of Symmetry Fig.2-3 Diagram of axis-symmetric unimorph plate. A cross-section schematic of a unimorph plate is shown in Fig.2-3, where

and

are

thickness of piezoelectric and elastic (passive) layers, respectively. For analyzing the stress and strain of the bending membrane, the neutral axis of deflection must be determined. Since the membrane vibration amplitude is much smaller than its dimensions, we can assume it to be plane strain, i.e. no strain in the direction vertical to the membrane. As a result, the radial and tangential stresses inside the thin membrane can be obtained using polar coordinate: Ez ∂ 2 w(r ) 1 ∂w(r ) ( +ν ) 2 2 1-ν ∂r r ∂r Ez 1 ∂w(r ) ∂ 2 w(r ) σθ = - 2 ( +ν ) 1-ν r ∂r ∂r 2

σr = -

(2-12)

The z-coordinate is measured relative to the neutral axis, which is computed from the following equation:

 F =

top

bottom

top σ r  E( z) z dz =0   dz = bottom 1-ν ( z ) σθ 

(2-13)

17

Fig.2-4 Sum of stresses in radial and tangential directions at r=a=12.5µm inside the AlN layer. Good agreement is observed between the analytical and FEM models. Given the vibration mode-shape of the membrane, Eq. (2-12) and (2-13) allow the stress to be calculated. Using these equations, the layer stack (0.75 µm AlN/ 0.8 µm SiO2) was chosen to maximize the piezoelectric stress at a given deflection, thereby maximizing pressure sensitivity. To verify the analytical model, finite element method (FEM) simulation results for the sum of stresses in the radial and tangential directions at the membrane edge ( = ) are compared in Fig.2-4. Close agreement is observed both for computed stress and for the location of the neutral axis. Furthermore, FEM simulation was used to determine the best piezo/silicon layer stack to optimize receiving performance. Fig.2-5 shows electrical potential in the middle of the piezoelectric layer at the center of the membrane with various thicknesses of the piezoelectric layer on a fixed 2.5 µm Si layer under a uniform 100 kPa pressure. The optimal designs for PZT and AlN PMUTs are 1.1 µm PZT/ 2.5 µm Si and 0.8 µm AlN/ 2.5 µm Si, respectively. AlN PMUTs show ~3× higher optimal receiving sensitivity due to its small dielectric constant, while PZT has a larger piezoelectric coefficient e31 than that of AlN, and therefore higher transmitting sensitivity. 18

Fig.2-5 FEM simulated electrical potential in the middle of the piezoelectric layer at the center of the membrane with various thickness of the piezoelectric layer on a fixed 2.5 µm Si layer under a uniform 100 kPa pressure.

2.4. Electrode Area Optimization Here, we analyze it in both transmitting and receiving modes. For example in receiving mode, for a clamped disk with uniform pressure p loading, we can obtain [41]: ∇ 4 w=

p D

(2-14)

Considering the boundary condition of clamped circular plate, w = dw = 0 @ r = rD , we can dr

obtain the solution for the deformation: (2-15)

( )=

Substituting (2-15) into (2-12), we obtain: ( )= ( )=

(

(

)

)

(1 + )

− (3 + )

(1 + )

− (1 + 3 )

(2-16)

where zp is the distance from the middle of the piezoelectric layer to the neutral axis of deflection. To maximize the electromechanical coupling, the top electrode should cover an area where there

19

is no sign change of the stress. Because stresses in both the radial and tangential directions, and

( ), contribute piezoelectric charge, the location where

( )+

( )

( ) = 0 is used to

determine the edge of the top electrode. Fig.2-6 shows stresses in the radial and tangential directions for a 50 µm diameter AlN PMUT with 1 Pa uniform pressure load on its surface. The radius where the sum of stresses is zero,

= √2 /2 , is the optimal electrode radius in

receiving mode. An analytical study on electrode optimization of PMUTs with different boundary conditions was reported in [42], where clamped and simply-supported plates show maximum transmitting efficiency when the electrode radius is 60% and 100% of the plate radius, respectively.

Fig.2-6 FEM simulated stresses in the radial and tangential direction of a 50 µm diameter AlN PMUT with 1 Pa uniform load on its surface. The sum of the stresses is equal to zero at ⁄ = √2/2, the optimal electrode radius. For optimization in transmit mode, we use energy method and need to assume a mode shape. The static deflection is very close to the mode shape of the first resonant mode, as below: ( )=

( )

(2-17)

20

2

2 where ( ) = (1 − ( ⁄ ) ) . Transformer ratio

is proportional to

, which is related to

electrode area coverage, as shown in Fig.2-7. The optimal electrode radius is

= √2 /2,

agreeing with the result of analysis in receiving mode.

Fig.2-7 Piezoelectric coupling versus electrode radius ratio.

2.5. Acoustic pressure directivity In reality, the PMUT is an edge-clamped membrane instead of a uniform piston. According to Equation (2-15), the average velocity of PMUT membrane, av

u0 = u0 / 3

, can be obtained [38]:

(2-18)

where u0 is the deflection magnitude at the center of the PMUT membrane. This relationship is the reason that the effective area used in the equivalent circuit model is one third the surface area. The pressure created by the motion of a single PMUT can be obtained [38]: P= j

where ,

R 0 P0 av j ( wt − kr ) e Ddir (θ ) r

( ),

(2-19)

and R0 are the angle of incidence, directivity, theoretical surface pressure,

and Rayleigh distance, respectively and are defined as follow: 21

av

P0 = ρ 0 c0u0

av

(2-20)

R 0 = A / λ = ka Ddir (θ ) =

2

/ 2

(2-21)

48 J 3 ( ka sin θ ) ( ka sin θ ) 3

(2-22)

In the far-field, the PMUT is equivalent to a spherical acoustic source with radius equal to the Rayleigh distance R0. The theoretical directivity for PMUTs operating at 10 MHz (λ = 150 µm in tissue) with 30 µm, 100 µm, and 200 µm diameter is shown in Fig.2-8, demonstrating that the 30 µm PMUT produces the most omnidirectional pattern.

Fig.2-8 Analytical directivity of transducers with different diameters.

22

3. High fill-factor annular PMUT array

This section presents a 1.2 mm diameter high fill-factor array of 1,261 piezoelectric micromachined ultrasonic transducers (PMUTs) operating at 18.6 MHz in fluid for intravascular ultrasound (IVUS) imaging. At 1061 transducers/mm2, the PMUT array has a 10-20 times higher density than previous PMUT arrays realized to date. AlN based PMUTs described in this paper are fabricated using a process compatible with the fabrication of inertial sensors, RF resonators and CMOS integrated circuits. The PMUTs are released using a front-side sacrificial etch through etching holes that are subsequently sealed by a thin layer of Parylene. FEM and analytical results including resonant frequency, pressure sensitivity, output acoustic pressure and directivity are given to guide the PMUT design effectively, and are shown to match well with measurement results. Due to the PMUT’s thin membrane (750 nm AlN/ 800 nm SiO2) and small diameter, a single 25 µm PMUT has approximately omnidirectional directivity and no near-field zone with irregular pressure pattern. PMUTs are characterized in both the frequency and time domains. Measurement results show a large displacement response of 2.5 nm/V at resonance and good frequency matching in air, a high center frequency of 18.6 MHz and wide bandwidth of 4.9 MHz when immersed in fluid. Phased array simulations based on measured PMUT parameters show a tightly focused, high output pressure acoustic beam.

3.1. Background Most of the previous work on PMUTs has focused on lead zirconate titanate (PZT) film, because of its high piezoelectric coefficient. While PZT has higher piezoelectric constants than the aluminum nitride (AlN) film used in this work, the lower dielectric constant of AlN allows 23

for comparable performance to be achieved, especially in terms of sensitivity in the receive mode [22]. In addition, unlike PZT films that require high fabrication temperature (around 800 °C) [24, 43], AlN is deposited at a low temperature (10 MHz) arrays requiring halfwavelength (λ/2) element pitch. Previous PMUTs were fabricated by a through-wafer etching approach [45, 46], resulting in low fill-factor, small element count, and therefore poor acoustic efficiency. Here, we present PMUTs with 10-20× higher density (1061 transducers/mm2) than the highest density PMUT arrays realized to date, 56 transducers/mm2 [45] and 123 transducers/mm2 [46]. This result was achieved using an AlN fabrication process originally developed for RF MEMS resonators, filters [47] and inertial sensors [48]. The small 5 µm spacing between PMUTs enables a high fill factor, and therefore a large array gain, which is the ratio of the array’s output pressure to that of a single PMUT. In addition, the small array can be used for high resolution medical imaging, especially intravascular ultrasound (IVUS) imaging, which requires small transducer size (< 2 mm) and high operating frequency (>10 MHz). In IVUS, close proximity to the target allows the transducer to pick up relatively weak ultrasound signals, which provides diagnostic information inaccessible from a noninvasive transducer.

3.2. PMUT array design Optical image of the PMUT array demonstrated here and a close-up picture of individual PMUTs are shown in Fig.3-1. The PMUTs are arranged in annular rings that are electrically connected through the top electrode metal layer. Eight concentric channels are formed by 24

connecting eight groups of adjacent rings. As summarized in Table 1, the number of rings in each channel was selected to yield approximately the same active area (and therefore acoustic pressure output per unit volt) for each channel. As shown in Fig.3-2, the phase delay of the signal applied to each channel can be controlled in order to control the focus point of the array. This allows a narrow acoustic beam without the need for an acoustic lens and enables electronic focus control, eliminating the need for mechanical scanning along the axis of the acoustic beam.

Fig.3-1 Optical image of the PMUT array and close-up picture of individual 25 µm diameter PMUTs. Left: the 8 channels are connected to individual bond-pads through the top-electrode metal and a single bond-pad is connected to the common bottom electrode. Right: The sacrificial poly-Si is removed via 2µm by 4 µm etch holes between each PMUT. The rings of PMUTs are connected through the top-electrode metal layer (dark gold) and bottom electrode (white).

25

Pulses_B

B Focal depth control A

Pulses_A

pMUT Annular Array 8

7

6

5

4

3

2

1

2

3

4

5

6

7

8

Fig.3-2 Schematic diagram of focal depth control of the PMUT array. The 8 annular channels are shown in cross-section. Table 3-1: PMUT annular array parameters. Channel #

1

2

3

4

5

6

7

8

# of Rings

5

3

3

3

2

2

2

1

#of PMUT

61

108

162

216

174

198

222

120

Membrane area (mm2)

0.03

0.05

0.08

0.1

0.09

0.1

0.1

0.06

Capacitance (pF)

1.94

3.44

5.16

6.89

5.50

6.30

7.07

3.82

A PMUT’s resonant frequency is proportional to the membrane thickness, h, and inversely proportional to the radius squared,

[30]:

∝ ℎ⁄ In addition, the pressure sensitivity

(3-1) is proportional to the radius squared and inversely

proportional to the thickness squared,

26

∝ ( ⁄ℎ )

(3-2)

To compare two PMUTs with the same resonant frequency but different diameters

and

using Equation (3-1) and (3-2), the ratio of the pressure sensitivities is: ⁄

=( /

which means

) ⁄(

) =



(3-3)

, the sensitivity for PMUTs with a fixed frequency

proportional to radius squared, ∝ 1/

/

, is inversely

:

(3-4)

For the same area, the PMUT array with a high fill-factor has a large effective area to generate acoustic wave or sense acoustic pressure. The fill-factor of an array with PMUT radius, a, and space between adjacent PMUTs, , is given by: ⁄(2 + ) (3-5)

=

Therefore, we can define a figure of merit (FOM), which is the product of the fill-factor, FF, and PMUT sensitivity =

×

:

= ⁄(2 + ) (3-6)

Eq. (6) shows that for PMUTs with a specific resonant frequency, a smaller PMUT pitch (smaller radius, , or space between adjacent PMUTs, ) will give a higher FOM, indicating higher output sound pressure per unit area.

3.3. Fabrication The process flow used for device fabrication is shown in Fig.3-3, where steps (a-d) were performed in the Sandia National Labs AlN MEMS fabrication process [47, 48] and steps (e-f) were performed in the UC Berkeley Marvell NanoLab. This process incorporates a sacrificial polysilicon release pit that precisely defines the PMUT diameter, thereby enabling a small device 27

size (25 µm and even smaller) with close spacing (5 µm) and eliminating the need for throughwafer etching. The sacrificial polysilicon is etched by vapor phase XeF2, releasing the PMUT membranes as shown in step (e). Etching holes are sealed and the device is insulated via vaporphase deposition of Parylene-C to enable fluid immersion, as shown in step (f). Two kinds of etching holes are designed and fabricated, as shown in Fig.3-4, (a) 2 µm×4 µm etching holes connected to the PMUTs through a buried polysilicon-filled channel and (b) 1 or 2 µm diameter etching holes located in the center of each PMUT. Etching holes are required to be small enough so that they can be sealed without filling the cavity beneath the PMUT membrane. Optical images of fabricated PMUTs with the two different etching holes are shown in Fig.3-4 (c) and (d).

(a) Polysilicon deposition and pattern.

(b) 1st SiO2 deposition, CMP and 2nd SiO2 deposition (0.8 µm).

(c) Oxide etch, CVD W contact and bottom electrodes deposition.

(d) Sputtering of AlN (0.75 μm), vias etch and top electrodes deposition.

(e) Opening of etch holes and XeF2 etch.

(f) Parylene sealing of etch holes.

Silicon Ti/TiN/Al

Polysilicon Al/TiN

SiO2 AlN

Fig.3-3 Fabrication process flow.

28

Tungsten Parylene

(a)

(b) Silicon Polysilicon

Aluminum Nitride Aluminum/TiN

SiO2 Ti/TiN/Al

Gold Tungsten

Fig.3-4 Cross-section diagram and optical images of two etching hole designs: 2 µm x 4 µm etching holes connected to the PMUT through a buried polysilicon channel (a, c) and 2 µm diameter etching holes located in the center of the PMUT (b, d). A high fill-factor array (1061 transducers/mm2) with etch holes as small as 2 µm × 4 µm was successfully released, as shown in Figure 1. A PMUT array with reduced fill factor (155 transducers/mm2) allowed successful release using 1 µm diameter center etch holes, suggesting that XeF2 depletion occurs during the release of dense arrays. 3D confocal laser microscope images of an etch hole before and after Parylene sealing are shown in Fig.3-5, demonstrating that the 3 µm Parylene layer successfully seals the etch hole. SEM pictures of the cross-section of Parylene sealed (750 nm AlN/40 nm SiO2) PMUTs, and a close-up image of a Parylene-sealed etch hole are shown in Fig.3-6.

29

Fig.3-5 Confocal laser microscope measurements showing a 2 µm by 4 µm etch hole (a) before and (b) after Parylene sealing and (c) height profile measurements.

(a)

30

(b) Fig.3-6 SEM images: (a) Cross-section of Parylene sealed PMUT and (b) the close up picture of buried channel and etching hole.

3.4. Characterization A Laser Doppler Vibrometer (LDV) (OFV 512 and OFV 2700, Polytec) is used in conjunction with a network analyzer (E5061B, Agilent Technologies) to measure the displacement frequency response in air, as shown in Fig.3-7. The average peak displacement sensitivity is 2.5 nm/V at a center frequency of 25 MHz. Measurements of 8 PMUTs selected from each annular ring show a small center frequency mismatch of 0.2 MHz (0.8%) across the array, demonstrating good fabrication uniformity. FEM and measured results of center frequency in air of PMUTs (750 nm AlN/800 nm SiO2) with various diameters are shown in Fig.3-8. The measured frequencies are ~10% lower than those predicated by FEM simulation. This result could be due to the AlN and SiO2 layers being thinner than expected, or may indicate that the FEM boundary conditions (perfectly clamped at the PMUT’s radius) do not capture the true boundary conditions of the PMUT. The frequency response results of a single PMUT in air measured before and after Parylene sealing are shown in Fig.3-9. The measurements show that the Parylene layer reduces the dynamic displacement sensitivity, equal to the product of static displacement sensitivity and quality factor (Q), from 2.5 nm/V to 0.36 nm/V. This is mostly caused by the reduction of the Q from 167 to 45 due to the additional Parylene layer. The static displacement sensitivity in air, which is approximately the same as the fluid-immersed dynamic 31

displacement (fluid-immersed Q ~ 1), is only reduced by approximately 50%, as FEM simulation results predict. Reducing the Parylene thickness is expected to improve performance while still sealing the etch holes.

Fig.3-7 Frequency response measured in air showing 0.8% frequency mismatch across the PMUT array.

Fig.3-8 FEM and measured results of center frequency in air of PMUTs (750 nm AlN and 800 nm SiO2) with various diameter.

32

Fig.3-9 Frequency response of a single PMUT measured in air before and after Parylene sealing. Impedance measurements were conducted using a wafer probe station, transimpedance amplifier (TIA), and network analyzer. Due to the relatively large cable capacitance in the measurement setup, it was difficult to observe resonance-antiresonance frequencies for high frequency (25 MHz) and small PMUTs (25 µm diameter) using this setup. Instead, the electrical impedance of a 1×12 linear array of 35 µm diameter PMUTs without Parylene sealing is shown in Fig.3-10. The amplitude and phase peaks show the resonance of the PMUT membrane, and the resonant frequency matches well with that obtained via LDV measurement.

33

Fig.3-10 Electrical impedance of a 35 µm diameter PMUT (750 nm AlN /800 nm SiO2) with no Parylene coating. Fig.3-11 shows pulse responses of displacement measurements of the 25 µm diameter PMUT(750 nm AlN/800 nm SiO2), driven by a 4 cycles high voltage pulse signal in air (a) and fluid (b). Fluid-immersed measurements were conducted in Fluorinert-70, which has acoustic impedance similar to that of human tissue and high electrical resistivity, eliminating the need for full insulation of all electrical connections to the device. The ring-down time in air and fluid shows the PMUT has a much lower quality factor (Q) in fluid than in air, which is caused by greater damping in fluid. These results match well with the frequency response in fluid, as shown Fig.3-12. The fluid-immersed transducer has a high 18.6 MHz center frequency and wide 4.9 MHz bandwidth. The center frequency is shifted from 25 MHz in air to 18.6 MHz in fluid because of the increased damping provided by the fluid. This bandwidth is smaller than that of some CMUTs [49, 50] due to the PMUT’s thick membrane (~ 1.9 µm, compared to 0.88 µm for the CMUT [49]) and therefore large effective mass. Reducing the PMUT membrane thickness is 34

expected to increase the bandwidth. The peak PMUT membrane vibration velocity is 1.5 mm/s/V, which corresponds to a surface pressure of 2 kPa/V. The output pressure is ~10× lower than a CMUT with a 120 nm capacitive gap and 40 V DC bias [49]. However, the PMUT doesn’t require a small gap and operation at a high DC bias. Furthermore, the PMUT’s output pressure can be increased by using a thinner Parylene sealing layer, a thinner PMUT membrane or using PZT instead of AlN film. Phased array simulations were conducted by using Eq. (28) to calculate the pressure field from a single PMUT based on the experimentally measured PMUT velocity. Fig.3-13 (a) is the pressure distribution without focus control and Fig.3-13 (b) and (c) are the pressure distributions with the focus set to 1 mm and 1.5 mm focal depth, respectively. This simulation demonstrates the ability to vary the focal point by controlling the phase of the signal applied to the 8 annular rings. Furthermore, it shows a small focus width of 100 - 150 µm, with acoustic pressures of 9 kPa/V and 6 kPa/V at focus depths of 1 mm and 1.5 mm, respectively. These focus points are ideal for the targeted IVUS imaging application.

(a)

(b) Fig.3-11 Pulse responses of displacement measurements of a 25 µm diameter (750 nm AlN/800 nm SiO2) PMUT, (a) in air, (b) in fluid (Fluorinet-70).

35

Fig.3-12 Measured frequency response in fluid (Fluorinert-70) with velocity converted to pressure on the surface of PMUTs. (a)

kPa/V 1.8

No focus control 2

(b)

kPa/V 9

Focus target @ 1 mm 2

(c)

kPa/V 6

Focus target @ 1.5 mm 2

5.5

1.6

8

1.4

7

1.2 1 0.8

1

0.6

1.5

6 5 4

1

3

0.4

2

0.2

1

4.5 Depth - Axial distance (mm)

1.5

Depth - Axial distance (mm)

Depth - Axial distance (mm)

5

1.5

4 3.5 3 2.5

1

2 1.5 1

0.5

0

0.2 0.4 Lateral Distance (mm)

0.6

0.5

0

0.2 0.4 Lateral Distance (mm)

0.6

0.5 0.5

0

0.2 0.4 Lateral Distance (mm)

0.6

Fig.3-13 Simulated pressure distribution based on the measured PMUT parameters (a) without focus control, (b) focused at 1 mm axial distance, and (c) focused at 1.5 mm axial distance.

3.5. Summary A 1.2 mm diameter, high fill-factor array of 1,261 AlN PMUTs was fabricated and characterized. At 1061 transducers/mm2, the array has a 10-20× higher density than the best

36

PMUT arrays realized to date. FEM and analytical results including resonant frequency, pressure sensitivity, output acoustic pressure and directivity were shown to match well with measurement results. The PMUTs in the array have center frequencies which match within 0.2 MHz (0.8% of the 25 MHz center frequency), demonstrating good fabrication uniformity. The peak PMUT membrane vibration velocity in the fluid is 1.5 mm/s/V, which corresponds to 2 kPa/V pressure sensitivity. Phased array simulations based on measured PMUT parameters show high output pressure (9 kPa/V) of the focused acoustic beam, demonstrating the feasibility of the array for use in IVUS applications.

37

4. Ribbon shape PMUT array

This section presents a fine pitch (50 µm) and high fill-factor (80%) linear array of ribbonshaped Aluminum Nitride (AlN) piezoelectric micromachined ultrasonic transducers (PMUTs). To improve PMUT sensitivity, the optimized piezoelectric/passive layer stack (750 nm AlN/ 800 nm SiO2) was identified via FEM simulation. PMUT harmonic modes were studied, and symmetric electrodes were used to excite odd harmonic modes, which possess superior acoustic coupling in comparison to even harmonics. The PMUT arrays were characterized in the mechanical, electrical and acoustic domains. The optimally-designed layer stack was verified to produce the maximum displacement sensitivity, 19.3 nm/V, in frequency response measurements. In pulse-response measurements, only the first harmonic mode was observed at small input voltage amplitude while both the first and third harmonics were observed at large (100 Vpp) amplitude due to transducer nonlinearity. Fluid-immersed measurements were performed with a needle hydrophone used to measure the 50 kPa peak-to-peak output pressure of a 1x8 PMUT array excited with a 140 Vpp 2 MHz pulse sequence. Finally, pulse-echo measurements were performed using a custom CMOS ASIC.

4.1. Background The lower dielectric constant of AlN (ℇAlN = 10.7 versus ℇPZT = 1300) [22, 51] allows comparable receiver sensitivity to be achieved in large AlN PMUTs (>500 µm diameter and > 10 pF capacitance) operating at 100’s of kHz [22, 52]. However, PMUTs operating at MHz frequencies have 10× to 100× smaller area and charge output than these earlier devices, making

38

pulse-echo measurement more challenging. While single ZnO PMUTs operating at 200 MHz have been demonstrated for high-frequency ultrasound [53], array transducers exploit microfabrication technology to enable features such as phased-array beam-forming. Here, we present a 2 MHz AlN PMUT array, and demonstrate for the first time pulse-echo ultrasound measurement using AlN PMUTs operating at MHz frequencies.

4.2. PMUT Design A PMUT array, shown in Fig.4-1, is used to increase the output pressure and enable pulseecho imaging over greater range. The array has a 50 µm pitch and 80% fill-factor, which are important for minimizing grating lobes of the acoustic beam [38]. The array was fabricated in a process using a front-side sacrificial etch of a buried polysilicon layer [48, 54, 55] to precisely define the released dimensions of the PMUT, eliminating the need for through-wafer etching and leading to a small 40 µm device size with close 10 µm spacing.

Fig.4-1 3D diagram of PMUT ribbon array, (b) close-up picture.

39

Fig.4-2 The first 5-order resonant modes of a 40 µm×200 µm PMUT. The mode-shape and frequency of the first five resonant modes of a 40 µm×200 µm PMUT were obtained using finite element method (FEM) modeling (COMSOL MultiPhysics), as shown in Fig.4-2. The modes are indexed (n, m) where n is the number of nodes along the width of the PMUT and m is the number of nodes along the length. Because the PMUT’s width (40 µm) is much smaller than its length (200 µm), the first five modes have n = 1 and m varies from 1 to 5. The even harmonic modes have low acoustic coupling efficiency because in these modes, equal areas of the membrane have positive and negative deflection, and the net volume flow above the membrane is close to zero. The odd harmonic modes have higher acoustic coupling due to the net difference in deflection, which reduces with increasing mode order. Therefore, the first (1,1) mode has the greatest acoustic coupling. Neglecting damping, the resonant frequency of this mode for a clamped rectangular plate with dimensions 2a×2b is: [56]

= 4 2( where

+

+

)

(1)

, , ℎ are the flexural rigidity, density and thickness, respectively. In a multi-layer

membrane,

and D are the average density and flexural rigidity of the layers weighted by their

thickness [54]. When b>>a, the resonant frequency is dominated by the smaller dimension a. 40

Therefore, the smaller dimension, 2a = 40 µm, is chosen to define the resonant frequency and the large dimension, 2b = 200 µm, is chosen to produce a large active acoustic area. Harmonic simulations were conducted using a piezo-acoustic FEM model to determine the electrode layout that maximized coupling to the odd harmonic modes. the top electrode was centered on the PMUT, and simulations were conducted varying the width of the electrode relative to the 40 µm width of the membrane in order to find the electrode that produced the maximum displacement in the (1,1) mode. The simulated frequency response with the optimum 22 µm width is shown in Fig.4-3, demonstrating that only the odd harmonic modes are excited by the electrode. In this simulation, the layer thicknesses used are 0.75 µm AlN and 0.8 µm SiO2, resulting in 7.2 MHz, 7.95 MHz and 9.55 MHz for the m = 1,3, and 5 modes, in good agreement with (1).

Fig.4-3 Harmonic FEM analysis with 1 V amplitude applied to the AlN layer showing the predicted frequency response and corresponding mode-shapes.

41

The layer thicknesses of the PMUT membrane were optimized by FEM simulation to maximize the displacement sensitivity. For a fixed AlN film thickness (0.75 µm), increasing the thickness of the SiO2 elastic layer from 0.1 µm to 1.5 µm increases the membrane stiffness and mass, resulting in a nearly linear increase in the (1,1) mode resonant frequency from 5 MHz to 10.5 MHz, as shown in Fig.4-4 (a). At the same time, thicker SiO2 increases the distance between the neutral axis and the middle of piezoelectric AlN layer, thereby increasing the piezoelectric bending moment in the transmitting mode and piezoelectric charge output in the receiving mode. Because the increased stiffness and increased bending moment are competing effects, an optimum exists. Fig.4-4 (b) shows the simulated displacement sensitivity versus SiO2 layer thickness: for 750 nm thick AlN, the optimum SiO2 thickness to achieve maximum displacement is 800 nm.

(a)

42

(b) Fig.4-4 Resonant frequency (a) and static displacement (b) versus thickness of the SiO2 elastic layer.

4.3. Characterization An optical image of the 8-PMUT array is shown in Fig.4-5. PMUTs with varying AlN/SiO2 thicknesses (750 nm AlN/40 nm SiO2, 750 nm AlN/800 nm SiO2, 1500 nm AlN/800 nm SiO2) were fabricated on three different wafers and measured to verify the predictions of the FEM model. Devices were released using vapor-phase XeF2 to remove a sacrificial polysilicon layer beneath each PMUT. While earlier devices fabricated in this process [48, 55] use large (10-20 micron) openings at the device edges to allow XeF2 to undercut the device, such large openings are problematic for immersed ultrasound experiments. Here, 2-µm etching holes spaced by 23 microns were used, simplifying the process of sealing the etch holes before immersion experiments. Timed XeF2 etching was conducted using visual observations of the etch front to determine the 15 min etch time required to undercut the 25 micron distance between etch-holes.

43

Fig.4-5Optical images of the 1×8 40 µm×200 µm PMUT array.

4.3.1. Mechanical domain characterization A Laser Doppler Vibrometer (OFV 512 and OFV 2700, Polytec) was used in conjunction with a network analyzer (E5061B, Agilent Technologies) to measure the displacement frequency response of a 40 µm×200 µm PMUT (0.75 µm AlN/ 0.8 µm SiO2) in air, as shown in Fig.4-6. In agreement with the FEM simulation, only odd harmonic modes are observed. The measured resonant frequency of the (1,1) mode was 6.25 MHz, ~10% lower than predicted by FEM, a result which is consistent with the expected film thickness tolerances of the process. The quality factor (Q) of this mode was 110, and a large displacement sensitivity of 19.3 nm/V was observed at resonance.

44

(a)

(b) Fig.4-6 Measured displacement amplitude (a) and phase (b) frequency response of a 40 µm×200 µm PMUT (0.75 µm AlN/ 0.8 µm SiO2) in air. The measured displacement frequency responses of 40 µm×200 µm PMUTs from three different wafers having different film thicknesses (750 nm AlN/800 nm SiO2, 1500 nm AlN/200 nm SiO2, 750 nm AlN/200 nm SiO2) are shown in Fig.4-7. The measurements show that the 750 nm AlN/800 nm SiO2 layer stack produces the maximum displacement sensitivity of 19.3 nm/V. When normalized by the quality factor (Q = 110), this result agrees well with the FEM prediction of 0.16 nm/V static displacement sensitivity shown in Figure 5(b). Compared to the optimal design, the thinner membrane (750 nm AlN/200 nm SiO2) has approximately three times lower displacement sensitivity (6.3 nm/V) at the (1,1) resonance frequency, a result that is also in good agreement with the FEM model shown in Figure 5. We also observe that the quality factor in air 45

is limited by air damping and is approximately proportional to the membrane thickness squared [57], with Q equal to 110, 136, and 52 for PMUTs from wafer 1 (1550 nm thickness), wafer 2 (1700 nm thickness) and wafer 3 (950 nm thickness).

Fig.4-7 Displacement frequency response of PMUTs from three wafers (W1, W2, W3) having different film thicknesses. The time-domain displacement response to pulse inputs was measured on a 40 µm×200 µm PMUT (1500 nm AlN/200 nm SiO2) in air. The PMUT was driven with a 4-cycle 8.3 MHz pulse-train using a high-voltage ultrasound pulser integrated circuit (HV7361, Supertex Inc.). The input pulse amplitude was varied to study the effect of the amplitude on the displacement response. The time-domain measurement at 100 Vpp, Fig.4-8 (a), exhibits beating because two vibration modes are excited in the PMUT. The fast Fourier transform (FFT) of the time-domain signal, Figure 9(b), shows that the displacement signal has frequency content at 8.3 MHz and 9.3 MHz, corresponding to the frequencies of the (1,1) and (1,3) modes (shown in Figure 3 (a) and (c)). At a lower, 5 Vpp, input voltage, the FFT of the response, Figure 9(c), shows that only the (1,1) mode is excited. The difference between these experiments is that the peak-to-peak displacement amplitude at 100Vpp input is over 100 nm, whereas it is an order-of-magnitude smaller (8.75 nm) at 5Vpp input. At large amplitude, mechanical nonlinearities (e.g. those 46

occurring due to in-plane stretching of the membrane) contribute to increased nonlinearity in the time response.

(a) Low voltage driving in air

1

1

0.8

0.8

Normalized Intensity

Normalized Intensity

High voltage driving in air

0.6 0.4 0.2 0

(b)

6

7

8 9 Frequency [MHz]

10

11

0.6 0.4 0.2 0

(c)

6

7

8 9 Frequency [MHz]

10

11

Fig.4-8 (a) 100 Vpp pulse displacement response measurement of a 40 µm×200 µm PMUT with 1500 nm AlN/200 nm SiO2; (b) FFT of time response signal in (a); (c) FFT of time response of a PMUT driven with 5 Vpp input.

4.3.2. Electrical domain characterization The electrical impedance spectrum of a single 40 µm×200 µm PMUT (750 nm AlN/800 nm SiO2) was measured using a Network Analyzer and transimpedance amplifier (TIA), as shown in Fig.4-9. The 6.22 MHz resonant frequency extracted from the impedance measurement agrees

47

well with the 6.25 MHz value extracted from LDV measurements. The effective electromechanical coupling coefficient, keff2, 1.68% was extracted from the impedance measurement using [58]: =

(2)

where fa and fr are the anti-resonant frequency, 6.275MHz, and resonant frequency, 6.222 MHz, respectively.

(a)

(b) Fig.4-9 (a) Electrical impedance measurement setup and (b) measurement of a single 40 µm×200 µm PMUT with 750 nm AlN/800 nm SiO2.

48

4.3.3. Acoustic domain characterization Bond pads were damaged during the XeF2 release step of the PMUT arrays having the optimized 750 nm AlN/200 nm SiO2 layer stack. Therefore, a 1×8 array of 40 µm×200 µm PMUTs with the 750 nm AlN/200 nm SiO2 layer stack was wire-bonded on a printed circuit board (PCB) to perform fluid-immersed ultrasound experiments. Fluorinert FC-70 was used as the immersion fluid because it has acoustic impedance similar to that of human tissue and high electrical resistivity which eliminates the need to insulate the bond wires. To prevent the fluid from entering the etch holes, the array was sealed using a 250 µm thick polydimethylsiloxane (PDMS) sheet, a material that is biocompatible and has similar acoustic impedance with human tissue. A 40 µm diameter needle hydrophone (Precision Acoustics) located 1.1 mm from the array was used to measure the output pressure. Two cycles 140 V peak-to-peak pulses were used to drive the PMUT array, resulting in 50 kPa peak-to-peak output pressure, shown in Fig.4-10 (a). An FFT of vibration ring down signal in the pulse response, Fig.4-10 (b), shows a ~2 MHz center frequency, which is lower than the 4.15 MHz measured in air due to the increased mass loading of PDMS and fluid. In addition, from both time-domain and FFT results, we observe that the third mode is excited. The measured pressure is consistent with the displacement measurements. From the FFT results, we obtain ~0.85 MHz -3dB bandwidth and displacement amplitude in fluid, air,

, to be estimated as 46 nm from the displacement in

, using the following equation: =

×

×

= 2.35. This result allows the

/

(3)

49

where

is the driving voltage and

and

are the quality factors in air and fluid,

respectively. Finally, the acoustic pressure, P, generated by the vibration amplitude

is

estimated to be 41.5 kPa using the following equation: =

×2

×

×(

where f is the frequency (2 MHz),

/(4

))

/

(4)

is the acoustic impedance (~1.4 Mrayl),

is

the effective vibration area of the PMUT (0.04 mm2), and R is the distance of the hydrophone from the PMUT array (1.1 mm). The small difference between the 41.5 kPa estimated pressure and the 50 kPa measured pressure may be due to fact that the model assumes an omnidirectional acoustic pressure pattern while the true pressure pattern may be slightly directional.

(a)

50

FFT of pulses response in the fluid

Normalized Intensity

1

(b)

0.8 0.6 0.4 0.2 0 1

1.5

2 2.5 Frequency [MHz]

3

3.5

Fig.4-10 (a) Time-domain pressure measured using a hydrophone and (b) FFT of the time response. Fluid-immersed pulse-echo measurements were conducted using a custom ASIC [59, 60] to transmit and receive signals from the 1x8 PMUT array. The ASIC produces 32 V unipolar 2 MHz pulses to transmit ultrasound and has an integrated transmit-receive switch and low-noise preamplifier to receive ultrasound. In the pulse-echo measurement setup, shown in Fig.4-11 (a), echoes from the fluid-air interface 2.2 mm away from the array were measured. The received echo, shown in Fig.4-11 (b), is 50 mV peak-to-peak. Transmitting wave Echo wave

Fluid/air interface

Fluid coupling material PMUTs

ASIC (a)

51

Amp Pulser

ADC

30

Pulse echo (mV)

20 10 0 -10 -20 -30

(b)

6

7

8 9 Time (μs)

10

11

Fig.4-11 Pulse-echo measurement set-up (a) and measured pulse response (b). The distance between the array and the fluid-air interface is 2.2 mm.

4.4. Summary A fine pitch (50 µm) and high fill-factor (80%) linear array of ribbon-shaped PMUTs was fabricated and characterized. An optimized layer stack for the PMUT membrane (750 nm AlN / 800 nm SiO2) was determined from FEM simulation results and verified via experimental measurements. A symmetric electrode was chosen to excite the odd harmonic vibration modes as these have higher acoustic coupling efficiency than the even harmonics. The displacement response to a pulse input shows that the (1,1) mode is dominant, with mechanical nonlinearity producing a small (1,3) mode component at large (>100V) pulse amplitudes. Immersed hydrophone testing resulted in a pressure amplitude that is consistent with the amplitude predicted using measured displacement amplitudes. Finally, a pulse-echo ultrasound experiment was demonstrated for the first time using AlN PMUTs at a frequency above 2 MHz. Pulse-echo measurements are challenging due to AlN’s small piezo-coefficient in comparison to PZT and the relatively small size of the 1×8 array of 40 µm×200 µm PMUTs. When the array was driven with a 32V input pulse, a 50 mV peak-to-peak signal was obtained for the echo from the fluid-air 52

interface 2.2 mm away from the array. This experiment demonstrates that CMOS-compatible AlN PMUTs are feasible for future ultrasonic imaging applications such as in portable devices for mobile health care.

53

5. Broadband PMUTs with additional anchor dual resonance modes

This section proposed broadband piezoelectric micromachined ultrasonic transducers (PMUTs) with additional anchor fabricated by a simple process, where front-side etching is used to release PMUT membrane and patterned thick metal layers are used to define the effective boundary of each PMUT. The PMUTs with additional anchors show ~10× lower frequency variation when compared with PMUTs without additional anchors. Furthermore, to obtain broad bandwidth and thereby enable high imaging resolution, we proposed PMUTs working on dual resonance modes with overlapping bandwidth. Firstly, analytical analysis and FEM simulation results demonstrate that we can successfully control frequency difference between two desired vibration modes. Secondly, analytical analysis was conducted to show a thin film is desired for a broad bandwidth of each mode. Based on the analysis, we proposed a 30 µm × 200 µm ribbon PMUT, which was characterized in the mechanical, electrical and acoustic domains. Measurement results demonstrate a large displacement sensitivity of 500 nm/V in air and pressure response of 0.3 kPa/V in fluid measured 1.4 mm away from the PMUT, equivalent to 13.6 kPa/V average pressure on the PMUT surface. Finally, The PMUT achieves a 97% fractional bandwidth by utilizing a thinner structure excited at two adjacent mechanical vibration modes with overlapping bandwidth.

54

5.1. Background Here, we present PMUTs based on chemical solution deposited (CSD) lead zirconium titanate (PZT) thin films [61]. When designing MUT arrays, it is desirable to minimize the spacing between MUTs because this results in high fill-factor, and therefore greater acoustic efficiency per unit area. In addition it enables the MUTs to be spaced with a pitch that is equal to or less than half the acoustic wavelength (λ/2), thereby minimizing grating lobes [30, 38]. Through-wafer Si etching is a simple process to release the membrane for PMUTs. Compared with wet isotropic etching [20], through-wafer DRIE (deep reactive ion etching) improves fillfactor but the PMUT size and pitch are still limited by the wafer thickness and DRIE aspect ratio [24, 62]. While PMUTs based on cavity SOI wafers [15, 63] can achieve fine pitch, cavity SOI wafers have relatively poor thickness control (on the order of ±0.5 µm), necessitating thicker PMUT membranes and resulting in narrow bandwidth. Front-side etching enables thinner PMUT membranes, but makes it more difficult to precisely define the released PMUT’s dimensions, a parameter which has a strong effect on the resonant frequency. While the PMUT boundaries can be precisely defined using a buried sacrificial layer [54], this buried layer increases fabrication complexity and cost. Here, we proposed a novel process, where front-side etching is used to release PMUT membrane and patterned thick metal layers are used to define the effective boundary of each PMUT. Finally, a broad bandwidth is desirable for ultrasonic transducers to increase axial ultrasonic pulse-echo imaging resolution [64]. A broad bandwidth is also needed for harmonic ultrasonic imaging which enables improved imaging contrast [65]. Unlike other MEMS resonator devices, a higher coupling ratio of ultrasonic transducers requires more energy emitted into surrounding environment instead of stored in the transducers. Therefore, a low quality factor

55

(Q) is desired, which will result in a short ringing up and down time for pulse driving, and thereby a shorter time of flight (TOF) indicating better axial imaging resolution. Increasing resonant frequency is also one method to reduce TOF, but it will cause larger acoustic attenuation, which is proportional to frequency in fat and frequency squared in water. Prior PMUTs had low fractional bandwidth due to their thick membranes: 26% in [32], 30% in [66] and 43% in [58]. Here, we demonstrate a PMUT with a thin membrane (0.5µm PZT/ 0.5µm SiO2) that achieves a relative broad fractional bandwidth. Furthermore, we proposed a rectangular shape PMUT, which has controllable resonant frequency difference between two adjacent desired resonance modes. Finally, through excitation the two vibration modes with overlapping bandwidths, and overall bandwidth is increased by nearly a factor of two.

5.2. Design and fabrication Unlike vibration of thickness mode used in conventional ultrasonic transducers, MUTs require a thin released membrane to work in a flexural mode. Instead of using sacrificial material and chemical-mechanical planarization (CMP) to precisely define PMUT dimension, additional anchor on the top around PMUT boundary is proposed in this paper. Fig. 5-1 shows a laser confocal microscope image of a 50 µm diameter PMUT fabricated based on the process flow shown in Fig. 5-2. The circular PMUT is composed of a unimorph membrane (0.05 µm Pt/0.5µm PZT (52/48)/0.1µm Pt/0.036µm TiO2/0.5µm SiO2), actuated via an annular top electrode, and released through a central etching hole.

56

Fig. 5-1: Laser confocal microscope images of a single PMUTs with thick metal additional anchor. Fig. 5-2 shows cross-section fabrication process flow. This process is also compatible with fabrication of other MEMS devices, such as micro-robot, lateral and in-plane torsional actuators, ultrasonic motor and RF MEMS switches [5]. As shown in step (a), bottom electrode Pt/TiO2, piezoelectric PZT (52/48) and top electrode Pt films are deposited and then patterned after vaporized deposition of SiO2 passive layer. Step (b) shows atomic layer deposition (ALD) barrier Al2O3 or HfO2 / Al2O3, which is subsequently patterned, followed by deposition and patterning of film Metal 1 (Au/Pt/Cr) to connect PMUT top and bottom electrodes with bond pads. As shown in step (c), sacrificial photoresist (PR) is deposited and patterned, which is followed by that Metal 2 (Au) deposition and patterning to form metal bridge. Also Metal 2 film is deposited on additional anchor area to increase anchor stiffness. Step (d) shows PR is removed using O2 57

plasma to release the metal bridge. Subsequently, passive SiO2 is etched to form etching holes, which is used for further XeF2 etching Si to release the membrane. However, it is challenging to precisely define the released PMUT’s dimensions via timed XeF2 Si etching. Therefore, the total 4 µm thick gold layer is designed at the boundary of the PMUT membrane to provide additional mechanical support. To inspect etching status, a dummy ruler is made through patterning top electrode metal layer around an etching hole with the same size and next to the real PMUTs.

Fig. 5-2: Fabrication process flow. Furthermore, it is desirable to reduce parasitic capacitance, receiving voltage sensitivity,

=

=

, as below: (5-1)

58

, to increase PMUT

where, capacitance, and

is generated charge on electrodes due to incident pressure,

is total

is PMUT capacitance. Here, the high dielectric constant PZT (ε = 1200)

beneath the bond pads and interconnect was removed in order to improve the PMUT’s receive sensitivity. This also prevents damage to the PZT layer from wire-bonding. The connection between the PMUT’s Pt top electrode and the Au interconnect layer is made via a plated gold air bridge, shown in Fig. 5-3, that prevents shorting of the top and bottom electrode layers. The air bridge is formed using a sacrificial photoresist (PR) layer, Figure 3(c), that is removed at the completion of the process using an oxygen plasma, Figure 3(d). As a result of removing the PZT layer from beneath the bond pads and interconnect, the total capacitance,

, of a 50 µm

diameter PMUT is reduced from 128 pF to 60 pF, which indicates doubling the PMUT’s receive sensitivity.

Fig. 5-3: SEM image of plated gold air bridges.

59

Fig. 5-4: Measured frequency response in air showing the 7% frequency variation that occurs with 60% overetch. The displacement frequency response of PMUTs were measured in air via a laser dropper vibrometer (LDV, OFV 512 and OFV 2700, Polytec) in conjunction with a network analyzer (E5061B, Agilent Technologies). The nominal diameter of the PMUT with the thick metal anchor is 50 µm, and measured displacement frequency response of it is shown in the Fig. 5-4. Then the same PMUT was put back into the chamber for further die-level XeF2 etching, and then displacement frequency response was measured again. Measurement results show that a 60% (80 µm diameter) overetch produced only a 0.5 MHz shift in the 6.5 MHz center frequency. Finite element method (FEM) simulation results (COMSOL Multiphysics) show that this overetch would result in a 4MHz shift for PMUTs without the thick metal anchor. This metal layer defines the effective PMUT anchor, reducing the sensitivity of the resonance frequencies to overetch during the XeF2 release step. Figure 4 shows the measured and simulated resonant frequency in air versus overetch percentage (8% to 65%) for PMUTs with and without the thick metal anchor. PMUTs with thick metal anchors show ~10× lower frequency variation when compared with the simulated results for PMUTs without additional metal anchors.

60

Fig. 5-5: Measured frequency variation versus overetch percentage.

5.3. PMUTs with dual resonance modes 5.3.1. Vibration of rectangular membrane FEM simulation results of the first 5-order resonant mode shapes of a fully clamped 30 µm × 200 µm membrane are shown in Fig. 5-6. Due to length of one side (200 µm) is much larger than the other onw (30 µm), the first five (n, m) modes have n=1. Among them, for even harmonic modes, (1, 2) and (1, 4) modes, one part of the membrane moves up, while the other part moves down, and therefore overall effective area is nearly zero resulting in low coupling between mechanical and acoustic domains. For odd harmonic modes, the effective area reduces with increasing number of the mode order. Here, we proposed PMUTs working with (1, 1) and (1, 3) modes, which have overlapping bandwidth to achieve a broad bandwidth.

61

Fig. 5-6: FEM simulation results of the first 5-order resonant mode shapes of a fully clamped 30 µm × 200 µm membrane. To design PMUTs with overlapping bandwidth for adjacent working modes, we need to control resonant frequency difference between them. Since PMUT membrane thickness, ~1 um, is much smaller than its length 30 um, we can model it as stretched rectangular membrane fixed at the boundary. Free vibration of a rectangular membrane can be expressed [30]: =

( , , )

where

( , , )

(

+∅ )

(

+∅ )

is displacement at x, y and t,

(5-2) and

are wave number in x and y directions

respectively, ∅ and ∅ are phase value, and A is vibration amplitude. For a fully clamped rectangular membrane with length a and b, the boundary condition is: ( , , )

=

( , , )

=

( , , )

Then we can obtain that



=

(5-4)

=

(5-5)

and

=

( , , )

=0

(5-3)

are integral times of

62

, as below:



+

where

=

and

(5-6)

are wave number in x and y directions, respectively, and

is total wave

number, which can be obtained: = where

/

(5-7)

is angular frequency, and c is effective acoustic speed inside the membrane. Therefore

we can resonant frequency for (n, m) mode, =

2 [( ⁄ ) +

]

/

, can be expressed as below [30]: (5-8)

where c is effective acoustic speed for flexural mode vibration, a and b are PMUT length. When one side of PMUT, b, is much larger than the other side, a,

will be dominated by the smaller

one, a. Furthermore, frequency difference between desired (1, 1) and (1, 3) modes can be expressed by: ∆ = where



=

( 1 + 8 (1 +

) − 1)

(5-9)

= / . Then we can expect that for the fixed length of one side of rectangular

membrane, a, increasing the length of the other side, b, can reduce frequency difference between desired (1, 1) and (1, 3) modes. FEM simulation results of resonant frequencies for different modes of a rectangular PMUT with different length are show in Fig. 5-7. The results agree with analytical analysis and show that resonant frequency is almost same when increasing PMUT length from 100 µm to 200 µm. Moreover, the frequency difference between desired (1, 1) and (1, 3) modes is reduced from 1.27 MHz to 0.25 MHz. Material properties used in the FEM simulation are listed as below: SiO2, Young’s modulus, E, is 72 GPa, and density, , is 2200 ⁄

; Pt, E is 120 GPa and

E is 230 GPa and

is 4000

is 21450 ⁄



; PZT, E is 83 GPa and is 7500

.

63



; TiO2,

Fig. 5-7: FEM simulation results of resonant frequencies for different modes of a rectangular PMUT with different length (the other side is fixed, 30 µm).

5.3.2. Thin PMUT membrane for broad bandwidth After controlling frequency difference between desired modes, bandwidth of each mode is desired to be broad enough for enabling overlapping and thereby broad bandwidth. For a second order system, quality factor (Q) can be obtained as below: =(

where

(5-10)

)⁄

is natural frequency, m is effective mass, and b is damping. Then, the bandwidth of a

PMUT, ∆ , can be obtained: ∆

=

⁄ = ⁄

(5-11)

For PMUTs with contact into human tissue or fluid, total damping b is dominated by acoustic damping and proportional to membrane area, A, as below: ∝

(5-12)

And assuming the same density, effective mass is proportional to the product of A and membrane thickness, h, as below: ∝ ℎ

(5-13) 64

Therefore, we can obtain that the bandwidth is inversely proportional to the thickness of the membrane, as below: ∆

∝ 1⁄ ℎ

(5-14)

Taking the advantage that PMUT frequency is defined by both membrane size and thickness, a thinner membrane can be selected to enable a broad bandwidth, while PMUT membrane size can be used to compensate and obtain the desired frequency.

Fig. 5-8: Image of a single 30µm×200µm PMUT. Inset: close-up showing the top electrode and 3 µm etching holes.

5.3.3. Measurement results Laser confocal microscope images of a single 30 µm × 200 µm rectangular ribbon PMUT are shown in Fig. 5-8. The membrane is released through 3 µm etching holes, which will allow them to be sealed by a thin polymer for fluid immersed measurement. To dis-activate the even harmonic modes, symmetric top electrode was selected. The ribbon PMUT is designed to have broad bandwidth by selecting dimensions to create closely spaced resonance frequencies of the (1,1) and (1,3) vibration modes. To obtain large and stable polarization, the 0.5 µm thick PZT

65

film is poled by 10 V DC in room temperature for 2 minutes. Then displacement frequency response of PMUTs were characterized in air using LDV, as shown in Fig. 5-9. Measurement results show that the odd harmonic modes ((1, 1), (1, 3), (1, 5)) are excited and the (1, 1) and (1, 3) modes are separated by approximately 0.4 MHz. The peak displacement sensitivity, 500 nm/V, occurs at the (1, 1) mode’s 5.5 MHz resonance frequency.

Fig. 5-9: Measured displacement frequency response in air shows that the odd harmonics ((1, 1), (1, 3), …) modes are excited. Fig. 5-10 shows measured electrical impedance of the PMUT. The resonant frequencies of (1, 1) and (1, 3) modes extracted from the impedance measurement are 5.49 MHz and 5.89 MHz, respectively, which agree well with the values extracted from LDV measurements. The electromechanical coupling coefficient, kt2, can be extracted from these impedance measurements using [58]: = where C and

=

(5-15)

and C are the motional capacitance and PMUT capacitance, respectively, and

are the anti-resonant frequency and resonant frequency, which are 5.57 MHz and 5.49

MHz, respectively, for (1,1) mode. The bond pads and interconnect was not removed for this rectangular PMUT and the capacitance caused by them is 4.5 times of capacitance of the real 66

PMUT. The electromechanical coupling coefficient is 13.9% after the calibration of real PMUT capacitance.

(a)

(b)

Fig. 5-10: (a) Impedance measrurement setup; (b)Measured electrical impedance of the PMUT. Furthermore, displacement frequency response measurements were conducted at small ac voltage (0 dBm) and various dc bias voltages -5 V, 0V and +5 V. The measured displacement sensitivity amplitude and phase of (1, 1) mode in air are shown in Fig. 5-11 (a) and (b), respectively. Resonant frequency and peak displacement sensitivity amplitude are 5.56 MHz and 477.5 nm/V, 5.59 MHz and 549.5 nm/V, and 5.66 MHz and 480 nm/V for cases of 0 V, -5 V, and 5V dc bias, respectively. Because dc bias will change PZT polarization, peak displacement sensitivity amplitude varies with dc bias. In addition, resonant frequency is also changing with dc bias, because it will cause additional stress inside piezoelectric film due to piezoelectric effect 67

and therefore affect the membrane effective stiffness and resonant frequency. Furthermore, resonant frequency for PMUT with both -5 V and +5V dc bias is higher than that of PMUT with 0 V dc bias. That indicates tensile stress is generated in both cases and therefore, PZT polarization is switched. This is also verified by displacement sensitivity phase measurement, shown in Fig. 5-11 (b), that peak displacement sensitivity phase is -103, 88, and -100 for cases of 0 V, -5 V, and 5V dc bias, respectively.

(a)

(b)

Fig. 5-11: Frequency response of PMUT displacement sensitivity amplitude (a) and phase (b) in air with various dc bias. In addition, we also observed the phenomenon of PZT polarization hysteresis through the measurement of PMUT displacement frequency response. It is indicated by that resonant frequency can be different for PMUT in (1, 1) mode with the same 0 V dc bias, which are 5.5 MHz and 5.56 MHz shown in Fig. 5-9 and Fig. 5-11, respectively. To further study PZT polarization, PMUT was driven with small ac voltage (0 dBm), and dc bias was swept with a small step 0.5 V, from 10 V to -10V and then back to 10 V. Fig. 5-12 (a) and (b) show measurement results of dependence of the resonant frequency and the peak displacement sensitivity amplitude (nm/V) on dc bias, respectively. As shown in Fig. 5-12 (a), polarization switching is indicated by the minima of the measured resonant frequency, where stress generated

68

by dc bias varying the sign. For example, when reducing dc bias from +10 V to -10V, firstly tensile stress will reduce from a specific number to zero (dc bias from 10 V to 0), and then compressive stress will increase from zero to a specific number (before polarization switching), and then tensile stress generated increase from zero to a specific number (after polarization switching). This result is also verified by the sign reversal of the measured displacement sensitivity in Fig. 5-12 (b). Both plots show polarization hysteresis loop and polarization domain switching occuring at +0.5 V and –4 V, corresponding to coercive fields Ec = 1 V/µm and –8 V/µm. The asymmetric hysteresis and butterfly loop may be caused by preferential alignment along the c-axis of the film, due to defect dipoles or residual stress within the film [12]. Finally, instead of using expensive specific instrument to measure PZT polarization switching and hysteresis, here we directly measured displacement response generated by e31 piezoelectric effect.

(a)

(b) Fig. 5-12: PMUT resonant frequency (a) and displacement sensitivity at resonance (b) measured in air versus dc bias. Ultrasound experiments were conducted with a single 30µm×200µm ribbon PMUT and

Fig. 5-13 shows system diagram and optical image of the measurement setup. PMUTs are wirebonded on a printed circuit board (PCB), and a glass tube is glued on the PCB to make a small 69

tank. Here, Fluorinert, (FC-70, 3M), is used for coupling material, because it has acoustic impedance ~1.3 Mrayl similar to that of tissue, 1.5 Mrayl, and high electrical resistivity, 2.3 × 10





, to eliminate epoxy protection for devices. In addition, the PMUT was driven

with a 20V, 4 MHz pulse using a custom 1.8 V interface ASIC [67]. The excitation was unipolar to avoid repoling the PZT, which will generate acoustic pressure with 2f frequency (f is driving frequency). The acoustic pressure was measured using a needle hydrophone (Precision Acoustics, 40 µm effective area) 1.4 mm away from the PMUT.

Fig. 5-13: (a) system diagram and (b) optical image of measurement setup for acoustic pressure. The time-domain acoustic pressure of pulse response is shown in Fig. 5-14. It shows pressure response of 0.3 kPa/V in fluid measured 1.4 mm away from the PMUT, equivalent to 13.6 kPa/V average pressure on the PMUT surface. Furthermore, it demonstrates ~ 0.4 µs ringing-down time, which indicates a broad bandwidth. Frequency-domain response is computed via fast Fourier transform (FFT) for the ringing-down part of measured pulse response. The (1, 1) and (1, 3) modes in fluid are observed and the center frequencies of them are at 3.7 MHz and 5.3 MHz, respectively, lower than those in air (shown in Figure 4) due to the mass loading of the fluid. The reduced ratio of resonant frequency for the (1, 1) mode is 33% (from 5.5 MHz to 3.7 MHz) much larger than that for the (1, 3) mode, 10% (from 5.9 MHz to 5.3 MHz). That is caused by larger effective coupling area of the (1, 1) mode, which is ~3× of that in the (1, 3) mode. 70

Furthermore, each mode has broad enough bandwidth and overlaps with each other. Finally, the ribbon PMUT demonstrates a large 97% fractional bandwidth at 3.7 MHz center frequency due to the thin PMUT membrane and the excitation of the (1, 1) and (1, 3) modes, which have overlapping bandwidth.

Fig. 5-14: Measured acoustic pressure in fluid (FC-70) from a single 30µm×200µm PMUT demonstrates a large 97% fractional bandwidth at 3.7 MHz center frequency due to the thin PMUT membrane and the excitation of the (1, 1) and (1, 3) modes, which have overlapping bandwidths.

5.4. Summary PMUTs were fabricated having a thin membrane composed of 0.5 µm piezoelectric PZT layer and 0.5 µm passive SiO2 layer. To reduce the effect of overetch on the PMUT’s resonant frequencies, a plated thick metal layer at the PMUT boundary was demonstrated, which proved to lower the frequency variation ~10× when compared with FEM simulations of PMUTs without the metal anchor. Instead of studying the PZT hysteresis loop using a specific instrument, we directly measured the displacement frequency response of PMUTs at small (-10 dBm) ac voltage and various dc bias voltages, and the results show that polarization switching occurred at coercive fields of Ec = 1 V/µm and –8 V/µm. Furthermore, analytical analysis and FEM simulation were conducted to successfully control frequency difference between two desired 71

modes of a rectangular PMUT. Finally, due to the PMUT’s thin membrane, each vibration mode has broad enough bandwidth and the two desired vibration modes were designed with overlapping bandwidths. The resulting ribbon PMUT has a broad 97% fractional bandwidth at 3.7 MHz center frequency, indicating its potential for high imaging resolution or harmonic ultrasonic imaging.

72

6. PMUT array based on cavity SOI wafers

This section presents high fill-factor piezoelectric micromachined ultrasonic transducer (PMUT) arrays fabricated via a novel process using cavity SOI wafers. The simple three-mask fabrication process enables smaller diameter PMUTs (25 µm) and finer pitch than previous processes requiring through-wafer etching. PMUTs were fabricated with diameters from 25 µm to 50 µm, resulting in center frequencies from 13 MHz to 55 MHz in air. Two types of devices, having different piezoelectric layers, lead zirconium titanate (PZT) and aluminum nitride (AlN), were fabricated and characterized. Comparing 50 µm diameter devices, the PZT PMUTs show large dynamic displacement sensitivity of 316 nm/V at 11 MHz in air, which is ~20× higher than that of the AlN PMUTs. Electrical impedance measurements of the PZT PMUTs show high electromechanical coupling

= 12.5%, and 50 Ω electrical impedance that is well-matched to

typical interface circuits. Immersion tests were conducted on PZT PMUT arrays. The fluidimmersed acoustic pressure generated by an unfocused 9×9 array of 40 µm diameter, 10 MHz PZT PMUTs, measured with a needle hydrophone 1.2 mm away from the array, was 58 kPa with a 25 Vpp input. Beam-forming based on electronic phase control produced a narrow, 150 µm diameter, focused beam over a depth of focus > 0.2 mm and increased the pressure to 450 kPa with 18 Vpp input.

6.1. Background When designing MUT arrays, it is desirable to minimize the spacing between MUTs because this results in high fill-factor, and therefore greater acoustic efficiency per unit area, and 73

because it enables the MUTs to be spaced with a pitch that is equal to or less than half the acoustic wavelength (λ/2), thereby minimizing grating lobes [30, 38]. Front-side etching using a sacrificial layer and etch holes has been used to make a high fill-factor PMUT array, but required a complicated multi-layer fabrication process and an additional layer to seal the etch holes after the release etch [32]. Through-wafer etching is a simpler process, but it results in a large PMUT dimension and pitch. Compared with wet isotropic etching [20], through-wafer DRIE (deep reactive ion etching) improves fill-factor but the PMUT dimension and pitch are still limited by the wafer thickness and DRIE aspect ratio. For example, PMUTs with 65 µm dimension and 100 µm pitch were fabricated via through-wafer DRIE [24]. A second challenge for small-dimension PMUTs fabricated via through-wafer etching is the PMUT’s center frequency is increasingly sensitive to diameter variations resulting from DRIE tolerances. To solve this issue, a two-step DRIE process was proposed, allowing 50 µm dimension and 80 µm pitch PMUTs to be fabricated [62].

Here, we avoid the need for through-wafer etching, relying on a more

manufacturable process based on cavity SOI wafers with precisely defined cavity diameter to produce PMUTs with a minimum diameter of 25 µm and minimum pitch of 45 µm. The process is both simple, requiring only 3 photomask layers, and high performance, enabling roughly 4x more PMUTs per unit area than previous through-wafer DRIE processes.

6.2. Fabrication A 3D schematic diagram of a PMUT array based on cavity SOI wafers and the cross section of a single PMUT are shown in Fig.6-1. In the 72×9 array, the top electrodes of the 9 PMUTs in each column are connected together to minimize the number of electrical connections to the die. The PMUTs demonstrated here are fabricated using cavity SOI wafers with a 2.5 µm device Si layer over 2 µm deep vacuum-sealed cavities of various diameters [63]. Devices were 74

fabricated using both lead zirconium titanate (PZT) and aluminum nitride (AlN) piezoelectric layers to allow the performance of these two types of PMUTs to be compared.

Fig.6-1 3D schematic diagram of a PMUT array based on cavity SOI wafers; (b) optical images of the fabricated 72×9 PMUT array. PMUT arrays were fabricated via a simple 3-mask process. A cross-section of the fabrication process flow is shown in Fig.6-2. The process is based on custom-fabricated cavity SOI wafers (IceMos Technology, Belfast, Ireland), as shown in step (a). The first mask is used to pattern 2 µm deep cavities in the handle wafer, after which the handle and device wafers are bonded in vacuum, followed by grinding and polishing to produce the desired 2.5 µm thickness of the device layer Si. The cavities define the location of each PMUT and are vacuum-sealed to eliminate the possibility of squeeze-film damping beneath the PMUT membrane. Alignment 75

marks etched into the handle wafer at the same time as the cavities are exposed by selectively etching openings in the Si device layer. The bottom electrode metal and piezoelectric layer are then deposited via sputtering. In AlN PMUTs, sputtering is conducted at 0.2 mm. Pressure measurements with and without beam-forming demonstrate a 3× increase in pressure using beam-forming, as shown in Fig.6-11. kPa

1.3

450

Axial distance (mm)

400 350

1.25

300 250

1.2

200 150

1.15

100 50

1.1

-300 -200 -100 0 100 200 Lateral distance (μm)

300

Fig.6-10 Measured pressure map of a 15x9 array driven at 10 MHz with 18 Vpp using phased-array beam forming.

87

Fig.6-11 Pressure measurements made with and without beam forming.

6.3.4. Summary Here, we presented PMUTs fabricated via a simple three-mask process using cavity SOI wafers. The process enables the fabrication of PMUTs with small diameter (25 - 50 µm) and fine pitch (45 - 70 µm), over a range of frequencies from 13 MHz to 55 MHz. Unlike previous processes that use through-wafer DRIE to release the PMUTs, cavity SOI allows smaller spacing between PMUTs, resulting in higher fill factor, and has the practical advantage that the cavity SOI wafer does not have through-wafer holes or require back-side wafer processing, simplifying wafer handling and increasing device yield. PMUTs were fabricated using both AlN and PZT piezoelectric layers to compare the performance of these two materials. While sputtered AlN has the advantages of lower deposition temperature (400°C versus 600°C for PZT) and much lower dielectric constant, sputtered PZT films were shown to have 16 times higher piezoelectric coefficient, ultimately resulting in 28 times higher displacement sensitivity than AlN PMUTs of the same diameter. Ultrasound experiments conducted using an immersed 15×9 PZT PMUT array operating at 10 MHz resulted in 450 kPa peak-to-peak pressure with 18 V peak-to-peak 88

input, a beam width of 150 µm FWHM, and a depth of focus > 0.2 mm. These results show that cavity SOI PMUTs are promising devices for ultrasound imaging.

89

7. Pulse-echo ultrasonic imaging using cavity SOI PMUTs

Conventional ultrasound imaging uses costly bulk piezoelectric transducers and high voltage electronics. Low-cost and low-voltage ultrasound transducers would enable many new applications in healthcare, biometrics, and personal health-monitoring. This work demonstrates short-range and high-resolution ultrasonic imaging using Aluminum Nitride (AlN) PMUT arrays, which are compatible with CMOS circuitry and wafer-level mass manufacture. Furthermore, because AlN has low dielectric constant, which means that lower input currents are needed to drive the array, here a local custom interface ASIC (requiring only 1.8 V power supply) with onchip charge-pump (1.8V to 32V) provides sufficient power for this purpose. In addition, instead of achieving a narrow acoustic beam using high frequency, causing greater acoustic attenuation, or large transducer size, resulting in large pixel and undesired near-field pattern, here we demonstrated a 90 µm acoustic beam width with low-frequency 8-MHz PMUTs using transmit beam-forming. Pressure map is measured using a 40 µm needle hydrophone and the results agree with simulation results. Furthermore, a scanning method was proposed to achieve a sub-100 µm focus size and meanwhile maintain a 70 µm small scanning step. Finally, 1-D and 2-D pulseecho imaging were conducted, which demonstrated the feasibility of AlN PMUTs for high quality of ultrasonic imaging in a short range.

7.1. Background A high fill-factor PMUT array is desirable to minimize grating lobes and increase acoustic 90

efficiency per unit area. However, reported PMUTs have large dimensions and pitch and therefore low fill-factor; this results from fabrication using through-wafer etching [62]. Frontside etching using a sacrificial layer and etch holes has been used to make a high fill-factor PMUT array, demonstrated in MEMS-2014 [32], but required a complicated multi-layer fabrication process and an additional layer to seal the etch holes after the release etch which reduces fluid-immersed PMUT performance [32]. PMUTs based on cavity SOI wafers have the advantages of a simple fabrication process and a high fill-factor, with device characterization of lead zirconate titanate (PZT) and Aluminum Nitride (AlN) PMUTs first demonstrated in [15]. Here, we demonstrate for the first time short-range (mm) and high-resolution pulse-echo ultrasonic imaging (~100 µm) using CMOS-compatible AlN PMUT array. Relative to PZT, AlN is lead-free and has the advantages of low-temperature (