Interface engineered V2O5-based flexible memristors towards high-performance brain-inspired neuromorphic computing

Abstract

Bio-inspired neuromorphic computing offers a revolutionary approach by replicating brain-like functionalities in next-generation electronics. This study presents two flexible resistive memory devices fabricated using DC magnetron sputtering, D₁(Nb/V₂O₅/Ni) and D₂(Nb/NbO_x/V₂O₅/Ni). Device D₁ exhibits abrupt SET and gradual RESET switching, while D₂ demonstrates fully gradual resistive switching (GRS), highly desirable for analog synaptic behavior. Mechanistically, D₁ is primarily governed by oxygen vacancies, whereas D₂ benefits from the synergistic interplay between oxygen vacancies and interfacial NbO_x/NiO layers, confirmed by XPS depth profiling. These interfacial layers significantly enhance D₂'s GRS performance and synaptic fidelity. Both devices exhibit temperature-dependent control of oxygen vacancies, which dynamically increases the memory window, lowering the ON/OFF ratio. Multilevel resistive states are generated in both devices by controlling the compliance current, with D₂ outperforming D₁ by exhibiting a higher memory window (∼552) and exceptional endurance beyond 7000 cycles. Moreover, both devices effectively replicate biological synaptic functions such as LTP and LTD. However, D₂ also mimics complex neural dynamics, including spike time-dependent and rate-dependent plasticity. Simulation of D₂'s artificial neural network demonstrates ∼86.75% excellent accuracy level, attributed to its linear, symmetric analog weight modulation and multiple conductance states. These results highlight the potential of V₂O₅-based devices for high-performance neuromorphic computing.