Spin-torque nano-oscillators for reservoir computing applied to time-series data.

Abstract

Recent investigations have discovered the potential of high-dimensional nonlinear systems for predicting nontemporal and temporal tasks. The prediction is demonstrated here for an asymmetrical seventh-degree polynomial and time series of variables in Rössler system by using spin-torque nano oscillator (STNO) as a reservoir corresponding to the nontemporal and temporal tasks, respectively. For training, the inputs are supplied through the bifurcation parameters, (i) magnitude of the alternating current (AC) passed through the STNO and (ii) the frequency with which the AC is applied. The computed values of root-mean-square error affirms that the STNO is a good candidate for the role of reservoir in reservoir computing. Further the output is predicted close to the target even in the presence of noise. The performance of the reservoir is investigated against the hyperparameters for both the nontemporal and temporal tasks. The performance of the reservoir is quite robust in the presence of thermal noise and the accuracy is maintained even at higher temperatures.