AlN Thin-Film Piezoelectric-on-Silicon Transducers for In-Sensor Computing

This paper reports the first MEMS physical reservoir computing device based on an aluminum nitride (AlN) thin- film piezoelectric-on-silicon (TPoS) transducer. Being an attractive piezoelectric material due to its high efficiency, high power handling, CMOS compatibility, and environmental friendliness, AlN has attracted much attention in both academic and industrial research. Physical reservoir computing (PRC) is an alternative neuromorphic computing framework that can implement in-sensor edge computing in MEMS sensors. This paper presents the first physical reservoir computing device and system based on an AlN sealed membrane transducer. Various classification and computing tasks were performed, including NARMA tasks, hand-written digit recognition, and arrhythmia detection. Test results showed improved performance compared to the literature.