Near-Sensor Analog Computing via Monolithic 3D Piezoelectric Sensor-FeFET for Tactile Sensing System

Abstract

Artificial tactile systems that replicate human tactile perception have garnered attention for use in human-machine interfaces such as prosthetic devices. However, achieving energy-efficient tactile processing with human-level capabilities remains challenging due to high power consumption and latency in conventional front-end circuits. In particular, although piezoelectric sensors offer self-powered and fast response, they are intrinsically limited in detecting static pressure stimuli. To address this limitation, a monolithic 3D-integrated near-sensor analog computing system is presented that co-designs the sensing and computing architecture. The piezoelectric sensor and metal-ferroelectric-metal-insulator-semiconductor (MFMIS) ferroelectric field-effect transistor (FeFET) unit cells are integrated into a 3 x 3 array, enabling local analog processing of static and dynamic tactile signals at the sensor node. Leveraging an MFMIS memory window >2.5 V, the unit cell resolves >= 6 static force levels within a force range of 1 N and exhibits 18.3 Pa-1 sensitivity. Furthermore, this array achieves three distinct weights for the kernel under identical input pressure conditions by tuning the capacitance ratio between the dielectric and ferroelectric layers, enabling real-time noise reduction with a static power consumption of approximately 10 nW without external interface circuits. Ultimately, these findings demonstrate the potential of piezoelectric- and ferroelectric-based near-sensor analog computing for next-generation energy-efficient tactile processing platforms.