A Bio-Inspired Visual System With Single-Spike Coding Based on VO2 Memristors

Abstract

Bio-inspired visual systems, owing to their event-driven operation, rapid response, and low-power processing capabilities, have emerged as promising candidates for next-generation efficient visual processing architectures. However, existing bio-inspired visual systems frequently employ rate coding, which necessitates a large number of spikes, thereby leading to high energy consumption and latency. This work demonstrates a compact single-spiking visual neuron and system based on a VO₂ memristor. With a feedback-controlled leaky integrate-and-fire circuit, the neuron emits only one spike under each input stimulus. By further introducing an optical sensor into the input path, illuminance can be directly encoded into spike timing. In simulations, the resulting single-spike visual system achieves 92.88% accuracy on the Modified National Institute of Standards and Technology (MNIST) classification task, while reducing the average energy consumption and average decision time by <inline-formula> <tex-math notation="LaTeX">$12.2\times $ </tex-math></inline-formula> and <inline-formula> <tex-math notation="LaTeX">$2.96\times $ </tex-math></inline-formula>, respectively, compared with rate coding. These results indicate that the proposed neuron and system provide a new route toward efficient neuromorphic visual systems.