High-Performance Resistive Switching Memory-based Sensory-Neuromorphic System for Finger Motion Tracking
발표자
()
초록
내용
In recent, hardware implementation of neuromorphic computing system, which can effectively process numerous unstructured data, has received considerable attention. However, the inferior conductance tunability (non-linearity and asymmetricity) of the conventional synaptic element degrades the recognition accuracy of the hardware neural network (HW-NN). Herein, we demonstrate high-performance resistive switching memory (RSM) with uniform 3D grain boundaries (GBs), which function as ion transport channels, formed via polydimethylsiloxane (PDMS) pre-treatment process. Based on the superior conductance metrics of 3D GB-channel RSM (3D GB-RSM), we implement a sensory-neuromorphic system as a metaverse technology for finger motion tracking. This system is composed of a crossbar-array neural network and a stretchable double-layered photoacoustic strain sensor. The experimental results demonstrate a remarkable recognition rate of 97.9% in the HW-NN finger motion ultrasonic patterns.