Magnetization Vector Rotation Reservoir Computing Operated by Redox Mechanism

Area of Science:

• Artificial Intelligence
• Materials Science
• Physics

Background:

• Physical reservoir computing leverages nonlinear dynamics in physical systems for efficient AI.
• Magnetic materials offer miniaturization but suffer from high power consumption and complex structures due to external magnetic field and current requirements.

Purpose of the Study:

• To propose a novel redox-based physical reservoir that overcomes the limitations of magnetic materials.
• To utilize the planar Hall effect and anisotropic magnetoresistance for a magnetic-field-free reservoir.
• To demonstrate enhanced expressive power and efficiency in artificial intelligence tasks.

Main Methods:

• Development of a compact, all-solid-state redox transistor-based physical reservoir.
• Exploitation of the planar Hall effect and anisotropic magnetoresistance, which depend on magnetization vector nonlinearities.
• Evaluation of the reservoir's performance on a second-order nonlinear equation task.

Main Results:

• The proposed redox-based reservoir exhibits higher expressive power compared to previous physical reservoirs.
• Achieved a normalized mean square error of 1.69 × 10^-3 on a nonlinear equation task.
• Outperformed a memristor array reservoir (3.13 × 10^-3 error) with less than half the number of reservoir nodes.

Conclusions:

• The redox-based physical reservoir offers a more efficient and powerful approach to AI hardware.
• This design eliminates the need for external magnetic fields, reducing power consumption and structural complexity.
• The results highlight the potential of redox transistors and specific physical phenomena for next-generation AI computing.