Input-driven bifurcations and information processing capacity in spintronics reservoirs

Abstract

Spintronics devices generate diverse nonlinear dynamics and have been studied as promising candidates for physical reservoir computing systems. However, the dynamic properties of spintronics reservoirs driven by input streams are largely yet to be uncovered. This study reveals that two types of bifurcation, from order to chaos and from chaos to order, can be induced by increasing the strength of input signals to the spintronics reservoir, and the information processing capacity of the reservoir changes drastically according to these bifurcations. The significant contributions of input-induced diversity in magnetization dynamics are demonstrated through numerical experiments, which include a real-world sensor emulation task. Our results suggest that modulating input settings can generate a diverse repertoire of magnetization dynamics without tuning the physical platform itself, providing valuable insights into neuromorphic applications.