With the advent of the future Internet of Things (IoT), conformal and lightweight flexible electronics have been in the spotlight for portable and wearable applications. Several researches have been reported to fabricate IoT components such as sensors and generators using excellent flexibility, which enables to respond to low frequent mechanical vibrations. In particular, acoustic sensors are an essential component for realizing an intuitive voice user interface (VUI) by converting the analog signal of the sound wave into a digital electrical signal. Recently, intensive studies for resonant acoustic sensors with high sensitivity have been demonstrated by mimicking the basilar membrane in the human cochlea. The basilar membrane converts the resonant sound into an electrical impulse to transmits acoustic information to the brain, exhibiting a similar mechanism with piezoelectricity. In this thesis, a miniaturized acoustic sensor for mobile applications was reported by controlling the resonant frequency band via ultrathin substrate and maintaining high sensitivity even in downsized dimensions.
In addition, generators have been required to power various electronic components. Among various energy harvesting principles, such as piezoelectric, thermoelectric, and electromagnetic, the triboelectric genarators have recently attracted because of advantages including high power output, facile device fabrication, and versatile configuration. In order to improve the performance of triboelectric generators, approaches for increasing electrostatic charge have been applied by modifying the surface to the nano-scale and micro-scale. In this thesis, fabrication of triboelectric generator was demonstrated by establishing a uniformly aligned 50 nm grating paterrn with a large-area wafer scale and replicating pattern to a flexible substrate.
Voice user interaction (VUI) has been attracting significant interest due to outstanding intuitiveness of human-machine interrelationship (HMI) in hyperconnected society. The conventional condenser microphones based on measuring capacitance between two vibrating conductive membranes stimulated by sound wave are most widely commercialized in smart-devices. However, critical drawbacks such as low sensitive recognition, power consumption, unstable amplifier, and limited single signal have not been yet improved. A flexible piezoelectric acoustic sensor was recently reported, showing highly sensitive multi-tunable frequency bands. However, this resonant based flexible piezoelectric acoustic sensor is still restricted for mobile applications and small IoT systems due to the large device size. In this thesis, a highly sensitive miniaturized piezoelectric acoustic sensor (MPAS) is demonstrated for machine learning applications. The MPAS on the ultrathin substrate was designed with geometrical modification to distribute resonance frequency inside 100 Hz to 4000 Hz. Lateral dipole formation matching with geometric structure of multi-channel electrode was induced by residual stress control in piezoelectric membrane. The outstanding sensitivity of MPAS was represented compared to the conventional condenser microphone and previous resonant piezoelectric sensors.
In order to realize a practical application through an acoustic sensor, it must be combined with machine learning algorithm. A flexible resonant acoustic sensor with multi-channels has advantages of tunable resonant band by selecting the most sensitive two channels at each frequency. In this thesis, we report biometric authentication and secure application using resonant acoustic sensors. The signal output from MPAS is analyzed for machine learning signal processing through Fast Fourier Transform (FFT) and Gaussian Mixture Model (GMM). For machine learning, MPAS is combined with a Microcontroller unit (MCU) that enables wireless communication. In addition, the voice information transmitted to the machine learning server is stored in a database and applied to testing procedure by comparing new input signal. Two averaging techniques are used to select the two most sensitive channels at each frequency for superior signal processing of multi-channel resonant signals. The error rate of speaker identification analyzed through machine learning algorithm is decreased to 56 % compared to a commercial MEMS microphone by utilizing resonant flexible acoustic sensor. In addition, for real-time practical application, the machine learning algorithm is improved to have a high speaker identification rate even with a small amount of data and the training time is reduced to less than 10 seconds. Real-time speaker registration, identification of registered speakers, and permission prohibition of unregistered speakers are functionalized in the smartphone application to improve practicality. Finally, biometric authentication and secure application are implemented using a customized smartphone app.
Information of sound sources is the most important data in the Internet of Things (IoT) system for autonomous vehicles, voice control systems, and smart home appliances. Detections of the direction and distance are conducted through analysis and calculation between the sound source and microphone arrays. The recently developed flexible miniaturized piezoelectric acoustic sensor (MPAS) shows a resonant based multi-channel properties with highly sensitive downsized dimension. However, analysis with 360-degree directionality of MPAS has not been conducted. Directionality is an essential characteristic of a microphone to achieve generation of similar sensor outputs in response to any sound source. In this thesis, we report a acoustic sensor array with excellent directionality using circular-type flexible piezoelectric acoustic sensor. The PMN-PT thin film with a seed layer was used by lowering crystallization temperature and a thermodynamic energy barrier. The PMN-PT piezoelectric membrane with a $PbTiO_3$ seed layer presents a low pore and dense characteristics. The circular-type flexible piezoelectric acoustic sensor was bio-inspired with the resonance propagation mechanism of the resonant based basilar membrane. The circular-type flexible piezoelectric acoustic sensor represented high sensitivity of 58mV and –33dB at first resonance and suggested the possibility of application for direction and distance detection through angle calculation and machine learning processing.
Triboelectric generator (TEG) developed with superior output power and facile fabrication has received intensive attractions for power generating from ambient mechanical sources such as low frequent human motion, machinery vibration, and fluid/wind flow. In triboelectrification for accumulation of electrostatic charge on TEG surface, it is essential to modify the micrometer or nanometer scale morphologies to induce the improved electric potential in identical area, however, the researches addressing perfectly-aligned pattern and controllable defectlessness in generation of surface charge have not been reported due to the difficulty of uniform pattern fabrication. In this work, we demonstrate the perfectly-aligned and seamless grating with 50 nm using novel modified spacer lithography. Utilizing alternatively deposited spacers (silicon oxide and silicon nitride) and polysilicon layers, multistep pattern downscaling up to 100 nm pitch is implemented in large-area mother substrate. Subsequently, 50 nm grating pattern was transferred onto flexible 8-inches substrate by replication methods for flexible vertical-type TEG fabrication. The output power was represented as 1 mW scale in generator, showing approximate value for powering the operational amplifier (OP-AMP) as circuit element in acoustic sensor systems. This outperformed method is expected to suggest the new fabrication technology in TEG for future sensor node platform and voice usr interface systems.