A magneto-ionic synapse for reservoir computing

TL;DR

This paper introduces a lithium-ion magneto-ionic device that mimics synaptic behavior for reservoir computing, demonstrating its application in waveform classification with optimized recognition accuracy.

Contribution

It presents a novel magneto-ionic device controlling magnetic states via ion migration, enabling neuromorphic reservoir computing with non-volatile, analog synaptic functionalities.

Findings

Device successfully emulates biological synapses.

Achieves waveform classification with optimized accuracy.

Demonstrates potential for neuromorphic computing applications.

Abstract

Neuromorphic computing aims to revolutionize large-scale data processing by developing efficient methods and devices inspired by neural networks. Among these, the control of magnetism through ion migration has emerged as a promising approach due to the inherent memory and nonlinearity of ionically conducting and magnetic materials. In this work, we present a lithium-ion-based magneto-ionic device that uses applied voltages to control the magnetic domain state of a perpendicularly magnetized ferromagnetic layer. This behavior emulates the analog and non-volatile properties of biological synapses and enables the creation of a simple reservoir computing system. To illustrate its capabilities, the device is used in a waveform classification task, where the voltage amplitude range and magnetic bias field are tuned to optimize the recognition accuracy.