Adaptive Memristor Synapse with a Biomimetic Astrocyte for Neuromorphic Computing and Artificial Visual Memory

Abstract The autonomy and flexibility of living organisms in resource allocation in fact touch upon the fundamental difference between biological and artificial systems. Although memristors offer a transformative technological pathway for the development of neuromorphic intelligent computing, most memristor-based resource allocation schemes lack intrinsic regulatory mechanisms and remain constrained by fixed trap mechanism and external control. Through both theoretical derivation of a memristor model and experimental implementation, this work proposed a biomimetic neuron−astrocyte tripartite synapse (BNATS) fabricated via a metal sacrificial-layer technique, which exhibited homeostatic plasticity analogous to that of biological astrocytes. With limited cognitive and memory resources, BNATS demonstrated improved computational accuracy and could help to stabilize memory, owing to the proposed strategy more closely aligned with natural biological processes and better fulfilled the requirements of biological plausibility.