Structure-Engineered Nanoporous Vanadium Oxide Memristors for Reconfigurable Synapse-Neuron Integration and Synergistic Robotic Motion

Abstract

Neuromorphic sensory-to-motor interfaces require compact devices that can combine nonvolatile synaptic weight storage with volatile neuronal firing, yet these functions typically rely on distinct material and circuit mechanisms. Here, we report a structure-engineered VOy/nanoporous VOx heterostructure that enables electrically selectable nonvolatile and volatile switching within a vanadium oxide memristor platform. Annealing-induced interfacial diffusion and oxidation produce an asymmetric stack comprising a crystalline VOy layer that supports threshold insulator-to-metal transition dynamics and an oxygen-vacancy-rich nanoporous VOx region that promotes filamentary conductance modulation. In a 16 & times; 16 crossbar array, identically fabricated cells are reconfigured either as artificial synapses exhibiting multilevel retention and analog long-term potentiation/depression or as artificial neurons producing relaxation-oscillator spiking and diverse neuronal response features. By pairing two cells as a one-synapse-one-neuron unit, the programmed synaptic conductance modulates the neuronal firing frequency and measured current-spike amplitude, thereby linking analog weight storage with spike-based signal generation. Using measured device characteristics, a hardware-informed spiking neural network recognizes rock-paper-scissors images with high accuracy, and its output commands are coupled to a memristive synergistic motor system that drives a robotic hand to generate counter-gestures. These results suggest that structure-engineered nanoporous vanadium oxide memristors can serve as reconfigurable building blocks for neuromorphic sensory-to-motor interfaces.