Bioinspired Magnetic Vision Synapse for Near-Sensor In-Memory Magnetic Information Encoding

Abstract

Animals use magnetoreception to sense magnetic cues for orientation and navigation.1 Inspired by this capability, artificial magnetic sensory hardware requires not only magnetic-field readout, but also a route for converting magnetic stimuli into internal device states that retain information about the input history. Here, we demonstrate a coupled magnetoelectric (ME)−VO2 magnetic-synapse prototype for magnetic-field-driven conductance-state modulation. A Metglas/PMN-PT/Metglas ME front end converts AC magnetic stimuli into voltage signals, which are processed through a signal-conditioning pathway and applied to the gate of an ionic-gel-gated VO2 transistor. The resulting gate modulation changes the VO2 channel conductance through volatile electrostatic modulation and more persistent proton-mediated electrochemical modulation. Under magnetic-field-pulse inputs, the coupled prototype exhibits transient current response, paired-pulse facilitation, and a transition from short-term to long-term plasticity, indicating that the VO2 conductance state can encode the amplitude, duration, and temporal history of magnetic stimuli. We further show that the ME front end can reconstruct a two-dimensional magnetic-field pattern generated by a coil array, providing a spatial magnetic input source for subsequent conductance-state encoding. This work should therefore be regarded as a prototype demonstration of magnetic-input-driven conductance modulation and conductance-state-based encoding, rather than a fully integrated magnetic-vision or autonomous neuromorphic computing system. The results suggest a possible route toward device-level front-end processing of magnetic information, while monolithic integration, low-power signal conditioning, and array-level implementation remain important directions for future development.