Magneto-mechanical reservoir computing combining a two-dimensional network of nonlinear mass-spring resonators with magnetic tunnel junctions

TL;DR

This paper presents a novel magneto-mechanical reservoir computing system using a 2D network of nonlinear mass-spring resonators with magnetic tunnel junctions for spintronic read-out, demonstrating high accuracy in vowel recognition and robustness for real-time sensing.

Contribution

It introduces a multiphysics reservoir computing platform combining mechanical and spintronic components, with a focus on nonlinear dynamics and practical edge computing applications.

Findings

Achieved over 95% accuracy in vowel recognition.

Demonstrated simplified signal injection suitable for real-time sensing.

Studied the influence of nonlinearity and robustness under parameter variations.

Abstract

Coupled networks of mass-spring resonators have attracted growing attention across multiple fundamental and applied research directions, including reservoir computing for artificial intelligence. This has led to the exploration of platforms capable of tasks such as acoustic-wave classification, smart sensing, predictive maintenance, and adaptive vibration control. This work introduces a multiphysics reservoir based on a two-dimensional network of coupled nonlinear mass-spring resonators. Each mass has a magnetic tunnel junction on top of it, working as spin diode, used as a spintronic read-out. As a proof-of-concept, we have benchmarked this reservoir with the task of vowel-recognition reaching accuracy above 95$\%$. Because the device accepts elastic excitations directly, signal injections are simplified, making it well-suited for realtime sensing and edge computation. We also studied the effect of nonlinearity, demonstrating how it influences the reservoir dynamics, and assessed its robustness under node-to-node variation of the elastic constants.