Improved Long-Term Prediction of Chaos Using Reservoir Computing Based on Stochastic Spin-Orbit Torque Devices

Abstract

Predicting chaotic systems is crucial for understanding complex behaviors, yet challenging due to their sensitivity to initial conditions and inherent unpredictability. Probabilistic Reservoir Computing (RC) is well-suited for long-term chaotic predictions by handling complex dynamic systems. Spin-Orbit Torque (SOT) devices in spintronics, with their nonlinear and probabilistic operations, can enhance performance in these tasks. This study proposes an RC system utilizing SOT devices for predicting chaotic dynamics. By simulating the reservoir in an RC network with SOT devices that achieve nonlinear resistance changes with random distribution, we enhance the robustness for the predictive capability of the model. The RC network predicted the behaviors of the Mackey-Glass and Lorenz chaotic systems, demonstrating that stochastic SOT devices significantly improve long-term prediction accuracy.