A generic simple model of synaptic memristor with local activity for neuromorphic applications

Abstract

A non-volatile locally active memristor is a promising candidate for neuromorphic computing based on artificial synapses and neurons, due to its high-speed switching, strong scalability, high computing, and low power consumption. In this paper, a novel generic model of voltage-controlled memristor with local activity and synaptic behavior is proposed. The circuit design of this memristor is very simple and easy to fabricate. Using small-signal analysis, the behavior of local activity is analyzed for this memristor model. Through the theoretical study, three significant parameters are identified to derive an equivalent circuit (small-signal), which is important for the study on dynamics of this memristor. To check the feasibility of the proposed model, a hardware-based implementation is performed through breadboard analysis. Important fingerprints of this memristor are verified both in theoretically and experimentally. The hardware-based results confirm the non-volatile characteristic and synaptic behavior of this memristor. Several experimental results exhibit a tunable modulation of synaptic weights with pulses, which effectively mimic different bio-synaptic characteristics like potentiation, depression, STDP (Spike-Time-Dependent Plasticity), STP (Short-Term-Plasticity), LTP (Long-Term-Plasticity), learning, forgetting, PPF (Paired-Pulse Facility), and PTP (Post-Tetanic Potentiation).