A Multiphysics Reservoir Computing System with Mass-Spring Metamaterials and Spintronic Readout for Vibration Analysis

Mechanical metamaterials, which can be modeled as chains of mass-spring resonators, are drawing significant interest from researchers across various engineering fields due to their unique properties and simple design. The continuous advances in developing new technologies and algorithms for artificial intelligence are opening a path to employing the mechanical properties of these metamaterials to perform tasks such as acoustic wave recognition, smart sensing and adaptive dynamical vibration absorption. A possible approach in this sense is based on the idea of reservoir computing (RC). RC is an emerging neuromorphic computing paradigm that leverages the intrinsic dynamics of nonlinear systems for efficient temporal data processing. In this work, we propose a multiphysics physical reservoir combining 2-dimensional nonlinear mechanical metamaterials (modelled as a network of massspring resonators) with spintronic technology as the electrical reading scheme of the masses' dynamics. In particular, this hybrid system exploits the stray field interactions between the magnetic tunnel junctions on top of the masses of the metamaterials and the mechanical dynamics of the elastic medium to achieve a highly compact, tunable, and energyefficient reservoir. The proposed design can operate across a broad frequency spectrum, from hertz to gigahertz, and can be scaled to the nanometric regime. We show that the nonlinear dynamics of this metamaterial maps the input to a highdimensional space, achieving over 90% accuracy for a vowel recognition task, comparable to state-of-the-art reservoir computing implementations, with minimal preprocessing. The integration of elastic inputs simplifies signal injection by directly utilizing mechanical vibrations, making this system particularly suited for real-time sensing and computing applications. With its scalability, efficiency, and accuracy, this hybrid spintronicsmechanical metamaterial used as RC system holds significant potential for applications in biosensing, machinery anomaly detection, and edge computing tasks such as voice and acoustic pattern recognition.