An artificial visual perception system based on ZnO threshold switching neurons with integrated rate and time-to-first-spike coding

Abstract

The proliferation of wearable electronics and Internet of Things (IoT) has driven the development of energy-efficient sensory processing systems inspired by the spiking mechanisms of the human sensory system. In this study, we present an artificial neuron integrated with an Ag/ZnO/Pt volatile threshold switching (TS) memristor for artificial visual perception and neuromorphic computing. The memristor exhibits electroforming-free operation, stable volatile switching behavior (with a cumulative probability variation of 1.508%), high ON/OFF ratios (∼1.64 × 10⁴), and excellent device uniformity, enabling it to effectively emulate biological neuronal functions such as spike encoding and leaky integrate-and-fire (LIF) dynamics. By integrating the memristor with photoresistors, an artificial visual neuron was developed, capable of spatial integration and letter recognition through distinct oscillation frequencies. Furthermore, an artificial visual perception system incorporating a spiking neural network (SNN) based on ZnO neurons was implemented for Yale facial image classification and MNIST digit recognition, employing the rate coding, the time-to-first-spike (TTFS) coding, and the rate-temporal fusion (RTF) coding strategies. Notably, the artificial visual perception system employing the RTF coding achieved the highest accuracy (94.4% for the Yale facial images and 91.3% for MNIST images) with superior energy efficiency. These results highlight the potential of ZnO-based artificial neurons for energy-efficient neuromorphic computing and intelligent sensory systems.