Orientation-Selective Memory Switching in Quasi-1D NbSe₃ Neuromorphic Device for Omnibearing Motion Detection.

Abstract

Intelligent neuromorphic hardware holds considerable promise in addressing the growing demand for massive real-time data processing in edge computing. Resistive switching materials with intrinsic anisotropy and a compact design of non-volatile memory devices with the capability of handling spatiotemporally reconstructed data is crucial to perform sophisticated tasks in complex application scenarios. In this study, an anisotropic resistive switching cell with a planar configuration based on lithiated NbSe₃ nanosheets is demonstrated. Benefitting from the highly aligned diffusive channel associated with a quasi-1D van der Waals structure, the memristor patterned along NbSe₃ atomic chains presents robust memory switching behavior with superior stability, particularly the low set/reset voltages (0.4 V/-0.36 V) and extremely small standard deviation (0.041 V/0.051 V), among the best compared to state-of-the-art devices. More importantly, unlike traditional resistive switching materials, anisotropic ion migration in NbSe₃ crystals leads to a high orientation selectivity in the conductance update. Custom-designed neuromorphic hardware contributes to the implementation of omnibearing motion recognition for automatic pilot applications, yielding a high accuracy of 95.9% considering variations. This article presents a new strategy based on NbSe₃ crystals to develop a neuromorphic computing system with intelligent application scenarios.