Topologically Epitaxial Conductive Filaments in Oxide Artificial Synapses Enable Neuroglia-Like Regulatory Dynamics for Embodied Intelligence.

Abstract

Resistance random access memory (RRAM) has emerged as a critical device for neuromorphic computing, offering significant potential for synaptic simulation. Nevertheless, it remains challenging to control the stochastic nature of the conductive filaments (CFs) in oxide-based artificial synapses, leaving a critical gap between biological plasticity and neuromorphic reliability. Here, inspirated from the directional guidance of growth factors and the mechanical traction exerted by glia during axonal outgrowth, and apply these biological principles into a topo-epitaxial self-assembly protocol that steers every step of CF growth. By prescribing both the ionic trajectory and the structural registry of the nascent filament, we suppress intrinsic transport stochasticity and enforce crystallographic coherence. The result is atomic-precision control over ion migration and single-crystalline CF formation-achieved within standard CMOS flows, without extra masks or exotic processing. Finally, by constructing a behavior-level model based on the habituation characteristics of oxide artificial synapses, the application in obstacle avoidance is successfully presented. Our synapses empower embodied AI robots with rich, robust, and self-adaptive behaviors.