Stochastic IMT (Insulator-Metal-Transition) Neurons: An Interplay of Thermal and Threshold Noise at Bifurcation

A stochastic neuron, a key hardware kernel for implementing stochastic neural networks, is constructed using an insulator-metal-transition (IMT) device based on electrically induced phase-transition in series with a tunable resistance. We show that such an IMT neuron has dynamics similar to a piecewise linear FitzHugh-Nagumo (FHN) neuron. Spiking statistics of such neurons are demonstrated experimentally using Vanadium Dioxide (VO2) based IMT neurons, and modeled as an Ornstein-Uhlenbeck (OU) process with a fluctuating boundary. The stochastic spiking is explained by thermal noise and threshold fluctuations acting as precursors of bifurcation which result in a sigmoid-like transfer function. Moments of interspike intervals are calculated analytically by extending the first-passage-time (FPT) models for Ornstein-Uhlenbeck (OU) process to include a fluctuating boundary. We find that the coefficient of variation of interspike intervals depend on the relative proportion of thermal and threshold noise. In the current experimental demonstrations where both kinds of noise are present, the coefficient of variation is about an order of magnitude higher compared to the case where only thermal noise were present.