Dual functionality of NbOx memristors for synaptic and neuronal emulations in advanced neuromorphic systems

Abstract

In this study, we reveal a novel NbO_x memristor structure that significantly advances neuromorphic computing by modulating compliance current (CC). This structure emulates the dynamic functionalities of artificial synapses and neurons, addressing the challenge of accurately imitating biological counterparts. Our memristors uniquely exhibit both memory switching (MS) and threshold switching (TS) properties through oxygen content modulation, enabling a single device to undertake diverse synaptic plasticity and neuronal functions. Our findings demonstrate the dual functionality of NbO_x-based memristors: acting as TS and neuron-like devices at high CCs, and as MS and synapse-like elements at low CCs. At a higher CC, the device exhibits effective TS behaviors, including favorable threshold and holding voltages, along with efficient wait and recovery times. This has enabled the creation of a restricted Boltzmann machine (RBM) model for reproducing Modified National Institute of Standards and Technology (MNIST) database images and the implementation of the leaky integrate-and-fire (LIF) model, which in turn opens possibilities for utilizing spike neural network (SNN) frameworks. At lower CCs, the memristor displays synaptic characteristics such as short-term and long-term memory, facilitated by retention loss for offline MNIST simulation. The NbO_x memristor represents a critical component for the future of neuromorphic devices and capable of performing both neuronal and synaptic functions. It paves the way for sophisticated, integrated computational models, thereby significantly contributing to the neuromorphic engineering field. Through comprehensive structural and functional analyses, this study underscores the potential of NbO_x memristors in neuromorphic computing and lays the foundation for future high-performance neuromorphic device advancements.