Neuromorphic behaviors of VO2 memristor-based neurons

Abstract

Neuromorphic computing has the potential to overcome the limitations of the von Neumann Bottleneck and Moore's Law. Memristors, characterized by nanoscale, adjustable resistance, low power consumption, and non-volatility, are considered as one of the best candidates for neuromorphic computing. This paper utilizes an accurate model of VO2 locally active memristor fabricated by HRL Labs to construct second-order and third-order neuronal circuits, which can exhibit 21 different types of neuromorphic behaviors. These behaviors include all or nothing firing, complex spiking, bursting, refractory period, accommodation, chaos, and others. Based on the theories of local activity and the Edge of Chaos (EoC), this paper analyzes the dynamics of the neuronal circuits, demonstrating the biological neurons operate at the EoC, and the neuromorphic behaviors emerge on or near the EoC, which reveals the generation mechanism of the action potential.