Bio-inspired memristive plasticity enhanced by magnetic substrate design in CuO/V2O5 heterostructures for energy-efficient neuromorphic devices

Abstract

This study presents two adaptable resistive memory devices fabricated using DC magnetron sputtering: D1 (Ag/CuO/V2O5/NiMnIn/Ni) and D2 (Ag/V2O5/CuO/V2O5/NiMnIn/Ni), demonstrating strong potential for integration into neuromorphic computing (NC). Device D1 shows abrupt SET and analog RESET, whereas D2 exhibits gradual SET and RESET switching behavior that closely mimics biological synaptic functionality. Devices D1 and D2 are primarily influenced by the electrochemically active electrode (Ag) and oxygen vacancies. The trilayer structure of V2O5/CuO/V2O5 enhances both performance and synaptic functionality. The D2 device achieves multilevel resistive states under a magnetic field and demonstrates remarkable endurance, sustaining over 5000 cycles. Furthermore, both D1 and D2 effectively replicate the key biological synaptic functions, such as long-term potentiation (LTP) and long-term depression (LTD). Notably, D2 displays a more refined linearity in LTP/LTD characteristics compared to device D1. Additionally, D2 mimics complex neural functions, such as pair-pulse facilitation in the presence of a magnetic field and spike-timing-dependent plasticity. Artificial neural network simulations of the D2 device achieve a high recognition accuracy of 82.5%, characterized by its linear, asymmetric, and gradual weight change across multiple conductance states. These outcomes highlight the significant promise of CuO/V2O5-based devices in achieving an outstanding performance NC system.