Artificial Refractory Neuron Based on Cs₂AgBiBr₆ Nanoionic Memristor for Efficient Motion Information Processing.

Abstract

Biological neurons are highly efficient in encoding motion information, and their physical realization can inspire the development of emerging bionic machine vision technologies for autonomous driving and video monitoring applications. However, due to the difficulty in emulating the complex ion movement dynamics within neurons, the use of hardware to comprehensively emulate neuronal firing dynamics for encoding function replication remains challenging. Herein, we report a bio-inspired artificial neuron based on a Cs₂AgBiBr₆ memristor, which can replicate both neuronal firing and refractory period behaviors for motion information processing. We demonstrate that the bidirectional migration of Ag⁺ and Br^- ions within Cs₂AgBiBr₆ can induce threshold conductance switching along with a strong built-in internal electric potential, mimicking neuronal excitation-resting responses to voltage pulse stimuli. The artificial neurons can dynamically respond to continuous inputs from moving objects, encoding motion features (such as velocity, direction, and acceleration), into compressed feature maps for accurate classification and trajectory prediction. Our biomimetic encoding and prediction framework offers a promising strategy for developing neuromorphic systems with biologically realistic behaviors that may rival the capabilities of the human brain in processing complex dynamic information.