Achieving long-term and short-term synaptic plasticity in adaptive ANNs: A memristor circuit design with switchable dual-mode

Abstract

Memristors hold significant potential as innovative components for building neural networks, offering extensive application possibilities. This work presents a memristor emulator model implemented with a nonlinear module based on operational amplifiers and a one-transistor-one-capacitor (1T1C) module. This implementation enables both volatile and non-volatile modes, thereby expanding the application scope of memristors. By switching modes, the proposed memristor can emulate both long-term and short-term synaptic plasticity. Based on this model, an adaptive artificial neural network is designed, with the synapses composed of a memristor array. This design enables flexible reuse of the neural network between reservoir computing and single-layer perceptrons, effectively reducing the number of memristors used and improving utilization efficiency. It provides a novel solution for constructing high-efficiency, low-power neural networks.