Optimization of a Spin-Orbit Torque Switching Scheme Based on Micromagnetic Simulations and Reinforcement Learning

Area of Science:

• Materials Science
• Computer Science
• Electrical Engineering

Background:

• Spin-orbit torque (SOT) memory is a promising technology for next-generation nonvolatile magnetoresistive random access memory (MRAM).
• Its high-speed operation and excellent endurance make it ideal for cache applications.
• Optimizing switching dynamics is crucial for realizing its full potential.

Purpose of the Study:

• To combine a two-current pulse magnetic field-free SOT switching scheme with reinforcement learning (RL).
• To determine optimal current pulse parameters for the fastest magnetization switching.
• To investigate the impact of material parameter variations on switching determinism.

Main Methods:

• Micromagnetic simulations were employed to analyze switching probability.
• A reinforcement learning setup was implemented to discover optimal current pulse configurations.
• The study analyzed the influence of material parameter variations.

Main Results:

• Switching probability is highly dependent on current pulse configuration for sub-nanosecond timing.
• The RL setup achieved a magnetization switching time of approximately 150 ps, a 50% improvement over non-optimized parameters.
• Deterministic switching is achievable across material variations by adjusting current densities.

Conclusions:

• Reinforcement learning is an effective tool for automating and optimizing SOT switching characteristics.
• The developed two-pulse scheme with RL demonstrates potential for significantly faster MRAM operation.
• Careful adjustment of current densities ensures reliable switching performance despite material variations.