An Ag/Al₂O₃/Ga₂O₃/Pt bilayer memristor with low power consumption for neuromorphic applications

Resistive random-access memory-based in-memory computing offers a route beyond the von Neumann bottleneck, and the ultra-wide-bandgap semiconductor gallium oxide (GaO) is an attractive memristor material owing to its chemical and thermal stability. However, purely electrically driven GaO-based electrochemical metallization (ECM) devices still suffer from stochastic conductive-filament growth and poor resistance-state uniformity. Herein, a vertical Ag/AlO/GaO/Pt memristor was fabricated. The ultrathin AlO(5 nm) barrier layer not only suppresses the rapid vertical migration of Agions but also promotes Agnucleation at multiple sites, thereby transforming stochastic single-filament growth into controllable, parallel multi-filament conduction. On the electrically operated device, conductive atomic force microscopy directly revealed 33 discrete current hotspots across a 25marea. The metallic-like positive temperature coefficient of resistance in the low-resistance state, together with the area-dependent resistance, provides strong evidence for a multi-filament conduction mechanism. COMSOL Multiphysics simulations further clarified the current-homogenization effect afforded by the AlOinterlayer. The device exhibits stable bipolar switching (cycles, on/off ratio), multi-level conductance modulation, and lower operating voltages than comparable devices reported in the literature. It also successfully emulates the transition from short-term memory to long-term memory, spike-timing-dependent plasticity, and learning-forgetting behavior. A multilayer perceptron (MLP) built from the experimentally measured conductance states achieves a classification accuracy of 95.68% on the MNIST benchmark dataset, with an average energy consumption of approximately 2.14 nJ per pulse. Notably, without resorting to heterojunction engineering or intentional doping, a single ultrathin AlOinterlayer fabricated by standard processing is sufficient to enable controllable modulation of ECM conductive filaments. This work thus provides a viable and transferable strategy toward array-compatible GaO-based neuromorphic devices.