Nondestructive reading technique and refreshment circuit for symmetric and asymmetric stochastic memristors

Abstract

Eventually, Neuromorphic computing structures, which are bio-inspired alternatives to more conventional computing techniques, have been more notable. The researchers have attempted to harness the inherent disparity in electronic design to invent neuromorphic systems with intrinsically stochastic behavior. Theoretically, Networks incorporating stochastic neural networks (NNs) component can develop complicated statistical models of their environments. The memristors’ disparity in neuromorphic structures is mimicked in abstract models of noisy and unreliable brain parts. The stochastic memristor is an intrinsic source of disparity that permits neurons to generate spikes stochastically. The stochastic memristors are mimicked in bi-stable stochastic synapses. This paper studies the stochastic behavior of various memristor models. The configuration of the two-transistor-one-memristor (2T1M) synapse is very efficient in the neuromorphic synapse for its capability to adjust the reading and upgrade the weight on-chip by signals and applied with a nondestructive reading mechanism for asymmetric and symmetric stochastic memristors. A refreshment circuit is applied to recover the right weight when any destructive reading operations occur.