Capacitor-Less Memristive Integrate-and-Fire Neuron with Stochastic Behavior

Abstract

Stochasticity in networks of artificial spiking neurons has been theorized to unlock some of the adaptive learning capabilities of biology. To realize this behavior, systems must be designed to operate probabilistically, looking for patterns in a distribution rather than performing a deterministic mathematical function. In order to design these systems efficiently, stochasticity must be natively built into the system at its most fundamental level. Here, we propose an area-efficient non-leaky integrate-and-fire neuron which models membrane potential accumulation with a memristive device instead of a capacitor. The probabilistic behavior of the memristor ultimately generates stochastic firing patterns in the neuron output.