Piezoelectric bimorph MEMS speakers

As an electroacoustic element, speakers are now widely used in various devices, such as in-ear applications. With the development of mobile devices and wearable electronic devices, there are increasing requirements for the miniaturization and integration of electronic components, including a need for smaller and more energy-efficient speakers. Thanks to the high integration level, low power consumption, and low manufacturing cost of MEMS technology, MEMS speakers could meet the demanding requirements for miniaturization and high-quality sound reproduction. In recent years, MEMS speakers have received more and more attention.

MEMS speakers use several driving methods: electrostatic driving,^1–3 electromagnetic driving,^4–6 or piezoelectric driving.^7–10 Electrostatic speakers require high driving voltages. Compared with electrostatic speakers, electromagnetic speakers can generate a high sound pressure without a high voltage. However, the need to use a magnet leads to complicated micro-fabrication and packaging processes.¹¹ In contrast, piezoelectric MEMS speakers can provide satisfactory sound pressure levels (SPLs) at relatively low voltages. These speakers are fabricated by depositing piezoelectric thin films. This is a relatively simple fabrication process and makes mass production feasible.

PZT is widely used as a piezoelectric material for MEMS speakers because the high piezoelectric coefficient of PZT can increase the SPL of the speaker. As well as PZT, commonly used piezoelectric thin films include AlN and ZnO. Cho et al.¹² developed a piezoelectric MEMS speaker with a 0.7 μm PZT film on a polyimide film that can produce large displacements. The SPL reached 79 dB at 1 kHz when driven by a voltage of 13 V. Kim et al.⁹ designed and fabricated a MEMS speaker based on a PMN-PT single-crystal membrane. The electrodes were specially designed to form an interdigitated shape to increase the SPL. In recent years, researchers have been optimizing the performance of piezoelectric speakers. In addition to using arrays,¹³ performance can be improved by changing the shape of the diaphragm.¹⁴ Stoppel et al.¹⁵ developed an innovative piezoelectric speaker with a non-closed membrane of size 4 × 4 mm² made of PZT. It had a high SPL of about 110 dB and a satisfactory energy efficiency of more than 105 dB/mW. Wang et al.¹¹ produced a closed circular membrane speaker made from ceramic PZT with a chip size of 6.7 × 6.7 mm². The speaker produced a maximum SPL of 119 dB at the resonant frequency of 9 kHz driven by a voltage of 10 V_pp. However, due to the ferroelectric properties of PZT, actuators based on PZT are intrinsically nonlinear. In contrast, AlN has much better linearity. Moreover, PZT contains lead, which is not environmentally friendly. In addition, the processing of PZT thin films is not as mature as that for other MEMS thin films. The manufacturing process for AlN thin films is more mature and is compatible with CMOS fabrication. Moreover, AlN is environmentally friendly. Therefore, AlN is used as the piezoelectric material in this work.

Although smaller speakers have greater advantages for integration,¹⁶ the smaller the area of the speaker diaphragm is, the smaller the volume of air pushed by it, thus the speaker generates a lower sound pressure. The relation between SPL, the displacement of the diaphragm tip δ, and the area S is given by^11,16

where k is a constant related to the modal function of the bending cantilever beam, ΔV is the volume of air pushed by the diaphragm, V₀ is the initial volume of a sealed space such as the ear canal, p₀ is the ambient pressure, and p_ref is a constant of 20 μPa.

Therefore, new designs are needed to reduce the diaphragm area while achieving a satisfactory SPL. The MEMS speakers referenced above are all based on a unimorph design. Most researchers focus on making innovative designs in the in-plane direction. Few investigations have been done in the out-of-plane direction, i.e., the thickness direction. The novelty and importance of this work lie in the enhancements in the thickness direction. Generally, a bimorph design improves the performance of sensors¹⁷ and actuators^18,19 compared with a unimorph design. According to Ref. 20, the maximum tip displacement of a unimorph actuator is half of that of a bimorph actuator with the same applied voltage. In this work, we developed a small MEMS speaker with a bimorph structure. To the authors’ knowledge, this is the first bimorph piezoelectric MEMS speaker to be fabricated. To assess our idea for making changes in the thickness direction, we adopted a widely used in-plane structural design.^15,21 It comprises four identical triangular piezoelectric cantilevers distributed in a regular square. The bimorph structures in the cantilever actuators are parallel. For the same tip displacement, the total area for the gaps between the triangular cantilevers is smaller than that for rectangular cantilevers, thereby reducing the amount of sound leakage.¹⁴ This work proves the feasibility and superiority of piezoelectric bimorph MEMS speakers through simulation, fabrication, and characterization.

A cantilever unimorph actuator consists of electrode layers, a piezoelectric layer, and a support layer. Its structure is shown in Fig. 1(a). The support layer can be made of a material such as silicon, and its function is to change the position of the neutral plane of the cantilever beam. When an electric field is applied to the piezoelectric layer, the layer stretches or shrinks. This forces the support layers to move together, causing the cantilever beam actuator to bend. In contrast, a bimorph structure comprises two separate piezoelectric layers. The structure is shown in Fig. 1(b). When an electric field is applied to the piezoelectric layers, one layer stretches along its length and the other shrinks. Therefore, even if the neutral plane remains at the mid-plane, the piezoelectric layer bends and deforms when a voltage is applied to the electrodes.

Previous research has shown that with the same geometry and the same applied electric field, a bimorph actuator can achieve twice the maximum tip displacement compared with a unimorph actuator.¹⁹ Moreover, the larger the volume pushed by the bimorph actuator is, the greater the sound pressure it generates.

Bimorph actuators can be connected in two ways, as shown in Fig. 1(b). First, the two piezoelectric layers can be connected in series. They then have opposite polarization directions, and the actuator is driven by an electric field applied between the top and bottom electrodes. Second, they can be electrically connected in parallel, such that the polarization directions of the piezoelectric layers are the same and the electric field directions applied to them are opposite.²⁰ In the latter case, the distance between the actuation electrodes is about half the thickness of the actuator, so that half of the driving voltage can be used to achieve the same electric field as when they are connected in series.²⁰ Therefore, in this work, we connect the bimorph speakers in parallel.

The aim is to reduce the area of the speaker while ensuring a higher SPL than that of a unimorph speaker. The area of the speaker diaphragm was designed to be 1.4 × 1.4 mm². The resonant frequency was set to around 10 kHz. A resonant frequency as high as 10 kHz endows a speaker with wide bandwidth and better high-frequency performance. The thickness of the AlN layer was set to 1 μm to guarantee the film quality such as the piezoelectric coefficient. The diaphragm size depends on the bimorph thickness and the resonant frequency. Compared with a closed diaphragm, a non-closed diaphragm can achieve mechanical decoupling, higher electromechanical efficiency, and linearity. Moreover, the influence of stress on the diaphragm is reduced. However, the gap between the diaphragms should be as small as possible, otherwise an acoustic short can occur.¹⁴ To reduce the thermal viscosity loss around the sidewall of the gap, the gap between the diaphragms was designed to be 8 μm. Due to the gaps, no additional vent hole was required. In addition, the diaphragm displacement and high-frequency SPL were not sensitive to the residual stress of the AlN films because the gaps release the stress. The total die size of the bimorph speaker was 4 × 4 mm², including the active area, pads, and silicon support. The size of a pad was 0.6 × 0.35 mm². The depth of the back cavity was 400 μm.

This work uses the finite-element method (FEM), as modelled in the simulation tool COMSOL, in designing the proposed bimorph speaker. The properties of the materials (AlN, Si, and Mo) used in the FEM model are given in Table I.

In addition, to characterize the performance of the speaker for in-ear applications, a universal 711 ear simulator model was coupled to the speaker model. In the simulation, solid mechanics and piezoelectric equations were applied in the speaker model. The ear simulator and the back cavity were modelled with a perfectly matched layer using the pressure acoustic equation. The thermoviscous acoustic equation was used in the gap between the actuators.¹⁰ The hypotenuse of the triangular cantilever beam was set as a fixed boundary, and the other two sides were free ends. The simulated displacement field for a bimorph speaker is shown in Fig. 2.

The FEM simulation was also used to compare the acoustic performance of bimorph and unimorph speakers. In the simulation, the driving voltage of both speakers was 10 V_rms. Figure 3 is a comparison of the SPLs. The piezoelectric layer thickness, electrode thickness, and total thickness were the same for both structures. The SPLs of both speakers have a first resonance peak around 10 kHz. At this frequency, the four cantilever beams vibrated together so that the SPL reached a maximum. The second peak at 14 kHz was caused by the acoustic impedance characteristics of the 711 ear simulator. The bump around 1 kHz was due to the Helmholtz resonance of the ear simulator. The low-frequency roll-off of the SPL curve was caused by the gap. The low-frequency SPL of the speaker would be lower for a wider gap.¹⁴ The SPL curve for the unimorph speaker is like that of the bimorph speaker, but the overall amplitude is lower by about 6 dB. Thus, the acoustic performance of the bimorph was better than that of the unimorph and the SPL was higher.

Although the structure looks like a commercial MEMS microphone,²¹ the electrodes are fundamentally different. For a microphone, the top and bottom electrodes are patterned so that they partially cover the piezoelectric layers to improve the sensitivity. Moreover, the electrode elements are connected in series to adjust the output capacitance. In contrast, the piezoelectric layers of the bimorph MEMS speakers in this work are wholly covered by the top and bottom electrodes to optimize the SPL. The different intrinsic requirements of sensing and actuating, i.e., sensitivity and SPL respectively, result in these different electrode designs.

The simulation results verify the feasibility of the bimorph speaker and the improved SPL. The designed speaker was fabricated on 4 in. silicon wafers instead of SOI wafers. The fabrication process is illustrated in Fig. 4.

First, 1 μm of thermal oxide was grown on the wafer, and then a 30 nm AlN seed layer was deposited onto the front side. The first 0.1 μm Mo layer was deposited onto the seed AlN layer as the bottom electrode. Then, the first Mo layer was patterned to form an 8 μm gap. After etching the first Mo layer, a 1 μm AlN was deposited onto the bottom electrode. The second Mo layer was deposited on the first AlN layer and patterned in a slightly different way from the first layer as the middle electrode. Then, another 1 μm of AlN was deposited. A third Mo layer was deposited and patterned as the top electrode. All Mo electrodes were patterned by RIE after deposition. The AlN in the middle and the edges was etched to open the gap and expose the electrodes. First, the AlN at the electrode connection and the middle gap was etched to the middle electrode. Second, the AlN at the top and bottom electrode connections and the middle gap was etched to the bottom electrode. All AlN layers were etched by a KOH solution. After AlN etching, the top metal layer was deposited and patterned for the electrode connection. Here, the top and bottom electrodes were connected as one feed of the input signal, and the middle electrode was used as the other feed of the input signal. The pad layer was 50/500 nm Cr/Au. Cr/Au was chosen because of its high conductivity. Since these materials are only used for the contact pads in this work, they can easily be changed to make the entire process CMOS compatible. After the front device layer was completed, deep reactive-ion etching was used to form the back cavity of the speaker. Finally, the wafer was placed in a hydrofluoric acid (HF) solution to release the cantilever beam from the thermal oxide. The fabricated bimorph speaker before it was released is shown in Fig. 5. The SEM images in Figs. 6(a) and 6(b) show the layer stack of the bimorph actuator beam and the fabricated speaker, respectively. The beams are warped upward due to the stress gradient in the AlN layers after they are released.

The fabricated speakers were tested in a ear simulator. The fabricated bimorph speaker chip was wire-bonded onto a printed circuit board (PCB) adapted to the size of the speaker. The PCB used in this experiment has a through hole on the back. The back cavity of the speaker is open to the air through the through hole in the PCB and is not sealed, which reduces the impact of the sealed back cavity on the acoustic performance of the speaker. The ear simulator used in the test conforms to the IEC60711 standard, which simulates the coupling between the speaker under test and the human ear. The ear simulator includes a pressure field measurement microphone, which can be used with a preamplifier to connect to an electroacoustic analyzer. During the measurement process, the electroacoustic analyzer generated an excitation signal to drive the speaker. The sound pressure was received by the ear simulator and processed by the electroacoustic analyzer. To ensure the accuracy of the experimental results, the entire experiment was carried out in an anechoic chamber, which can greatly reduce the influence of noise. The test system is shown in Fig. 7.

The SPL and the total harmonic distortion (THD) are important performance indicators for evaluating speakers. Figure 8 shows the measured SPL obtained under a voltage of 10 V_rms. The SPL was 73 dB at 1 kHz and relatively flat between 100 Hz and 1 kHz. There are two resonance peaks. The peak near 10 kHz is the resonance point of the first mode, which is consistent with the simulation. The SPL was 102 dB at the resonance frequency. The second resonance peak near 14 kHz is a typical response of the 711 ear simulator.

The position of the highest resonant peak is the same as in the simulation results. However, in the measurement results, the extremely low-frequency SPL was degraded. This may be because the speaker actuators were warped upward due to the stress gradient in the AlN layers so that the gaps between the actuators were wider, as shown in Fig. 6(b). Thus, the gaps between the beams were wider than the design value, which resulted in acoustic loss at extremely low frequencies.

Linearity means that the sound pressure produced by the bimorph speaker is linear with the applied voltage. The linearity of the speaker was characterized as follows. The displacement of the tip of the cantilever beam should be proportional to the applied voltage. The cantilever beam vibrates and pushes the air to produce sound. The sound pressure should also be proportional to the volume of air being pushed. Because the area of the cantilever beam is constant, the sound pressure generated by the speaker should be proportional to the displacement of the tip and also proportional to the applied voltage. We measured the sound pressure generated by the speaker for different applied voltages. The test results are shown in Fig. 9. As the applied voltage increased from 1 to 10 V_rms, the output sound pressure of the designed speaker increased linearly, indicating that the designed speaker has good linearity.

The THD is calculated as follows:

where A_out is the amplitude of the SPL. The THD is plotted in Fig. 10. The second harmonic was used. Note that Eq. (4) is cut off at 2f. In the range from 1 to 10 kHz, the THD curve has two main peaks at 4.9 and 7 kHz, which correspond to the second harmonic distortion of the resonance peaks at 9.8 and 14 kHz in the SPL curve, respectively, which is consistent with the relation reflected in Eq. (4). The THD of the bimorph speaker was less than 3% in the range from 1 to 10 kHz and was mainly less than 2%.

In addition to the SPL and linearity, the power consumption of the speaker was calculated from the impedance of the speaker: $P=U2Z$⁠. The electrical impedance of the speaker was measured by an impedance analyzer at each frequency. The magnitude of the measured impedance of the speaker was inversely proportional to the frequency. When the applied voltage was constant, the total power (or apparent power) of the speaker was proportional to the frequency. Note that the total power consumption is calculated here instead of the active power consumption. The measured power curve is shown in Fig. 11. The measured values are consistent with the theoretical values.

To improve the SPL of MEMS speakers, we proposed piezoelectric bimorph MEMS speakers. The SPLs of a bimorph speaker and a unimorph speaker were compared through an FEM simulation, which found that the bimorph speaker produced a higher SPL. When driven by a voltage of 10 V_rms, the speaker resonated at 10 kHz. At that frequency, the SPL was 102 dB. In the range from 100 Hz to 1 kHz, the frequency response of the speaker was relatively flat and the SPL was above 70 dB. The sound pressure produced by the bimorph speakers has good linearity with the applied voltage. In the range from 100 Hz to 10 kHz, the THD of the speaker was less than 4%. These results indicate that the nonlinear distortion of the bimorph speaker was small and the linearity was good. Although the absolute SPL of the speaker was not the highest among piezoelectric MEMS speakers, this work demonstrates that the piezoelectric bimorph design improved the SPL. Our future work will focus on optimizing the shape and thickness of the diaphragm and the fabrication process to attain a higher and flatter SPL curve.

This work was supported by the funding from Natural Science Foundation of China (Grant No. 62001322), the National Key Research and Development Program (Grant No. 2020YFB2008800), and the Nanchang Institute for Microtechnology of Tianjin University.

The authors have no conflicts to disclose.

Y.L. and C.L. contributed equally to this work.

All data included in this study are available upon request by contact with the corresponding author.

Yiming Lang received his B.E. degree in Agricultural Mechanization and Automation from Jilin University in Changchun, China, in 2018. He is currently studying toward a master’s degree at State Key Laboratory of Precision Measuring Technology and Instruments, Tianjin University, Tianjin, China. His research interests include microelectromechanical system (MEMS), Piezoelectric MEMS speaker.

Chengze Liu received his B.E. degree in engineering from Shandong University, Jinan, China, in 2020. He is currently pursuing the master’s degree at State Key Laboratory of Precision Measuring Technology and Instruments, Tianjin University, Tianjin, China. His current research interests include piezoelectric MEMS speaker.

Dr. Ahmed Fawzy currently works at the Nanotechnology Central Lab, Electronics Research Institute. He does research in Electronic Engineering, Optical Engineering and Experimental Physics. He joined the MEMS/NEMS Lab, Tianjin University as a Postdoc and later as a Research Scientist on the fabrication of piezoelectric MEMS devices. Their current projects are “super capacitor based on nanomaterials” and “Analysis and design of piezoelectric MEMS devices.”

Chen Sun received his B.S. degree in optical information science and technology from Nanjing University of Science and Technology, in 2006, and his M.S. degree in optics from Tianjin University in 2015. He is currently an engineer at School of Precision Instruments and Opto-Electronics Engineering, Tianjin University. His research interests include wet processing of silicon-based materials and structures, photo mechanics and digital image processing.

Shaobo Gong received his M.S. and Ph.D. degrees from the Institute of Semiconductors, CAS in 2018 and 2020, respectively. Now he is a postdoctoral fellow at the School of Precision Instruments and Optoelectronics Engineering, Tianjin University. His research interests include piezoelectric functional material and their applications, flexible electronic skin and piezoelectric MEMS transducers.

Menglun Zhang received his B.S. and Ph.D. degrees from Tianjin University, China. Since 2016, he has been an Assistant Professor at School of Precision Instruments and Optoelectronics Engineering, and State Key Laboratory of Precision Measurement Technology and Instruments, Tianjin University. He was a Visiting Scholar to Bioelectronics Group at University of Cambridge. His research field focuses on piezoelectric MEMS sensors and actuators, e.g., microspeakers, microphones and ultrasonic transducers.