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

Area of Science:

• Physics
• Nonlinear Dynamics
• Computational Science

Background:

• High-dimensional nonlinear systems offer potential for predictive modeling.
• Reservoir computing leverages complex systems for computation.
• Spin-torque nano oscillators (STNOs) are emerging as a novel hardware component.

Purpose of the Study:

• To investigate the efficacy of STNOs as reservoirs in reservoir computing.
• To demonstrate STNO-based prediction for both nontemporal and temporal tasks.
• To assess the robustness of STNO reservoirs against noise and temperature variations.

Main Methods:

• Utilized an asymmetrical seventh-degree polynomial for nontemporal tasks.
• Employed Rössler system time series for temporal tasks.
• Trained the STNO reservoir using alternating current (AC) magnitude and frequency as input parameters.

Main Results:

• STNOs demonstrated strong performance as reservoirs, confirmed by low root-mean-square error values.
• Accurate predictions were achieved even in the presence of significant noise.
• The reservoir's performance remained robust against hyperparameter variations and thermal noise.

Conclusions:

• STNOs are effective and robust candidates for reservoir computing applications.
• The proposed method shows potential for real-world prediction tasks involving complex dynamics.
• The system maintains prediction accuracy across various noise levels and operating temperatures.