Oxygen vacancy management enables uniform and robust TaO x memristor through TiO x buffer

Abstract The emergence of neuromorphic computing has positioned resistive random-access memory (RRAM) synapses as a promising solution to the von Neumann bottleneck. However, oxide-based RRAM devices face persistent challenges including stochastic conductive filament formation, switching variability, endurance limitations, and nonlinear conductance modulation. This study presents a novel interface engineering approach employing a TiOx buffer layer at the TaOx/electrode junction to regulate oxygen ion migration dynamics. The strategic incorporation of TiOx, selected for its superior oxygen-gettering capability and ionic diffusion barrier properties, yields remarkable device improvements: (1) 3.3V reduction in SET/RESET voltage variability, (2) >104 cycle endurance with stable 104 s retention. Synaptic functionality characterization demonstrates linear conductance modulation with high repeatability. In system-level validation, a convolutional neural network utilizing these devices achieves 92.7% MNIST recognition accuracy (20 epochs). This oxygen-ion-migration-managed TaOx memristor represents a significant advancement toward reliable analog RRAM for neuromorphic hardware implementation.