Ultra-low-energy skyrmion-based learning automata element for adaptive edge intelligence.

Area of Science:

• Spintronics
• Neuromorphic Computing
• Machine Learning Hardware

Background:

• Magnetic skyrmions are stable, nanoscale spin textures ideal for low-power computing.
• Learning automata require efficient hardware for adaptive decision-making.
• Integrating learning automata with memory devices can reduce power consumption.

Purpose of the Study:

• To implement a skyrmion-based learning automaton for energy-efficient, in-memory learning.
• To demonstrate a hardware-efficient logic-learning mechanism using skyrmion motion.
• To establish a scalable building block for edge-oriented machine intelligence.

Main Methods:

• Utilized micromagnetic simulations (MuMax3) with a Co/Pt bilayer.
• Mapped finite-state transitions to skyrmion motion along nanoscale tracks.
• Employed spin-orbit torque for skyrmion state transitions driven by optimized current densities.

Main Results:

• Achieved stable skyrmion nucleation, motion, and detection.
• Demonstrated a skyrmion-based Tsetlin Machine element for interpretable logic learning.
• Reported ultra-low energy consumption of 4.86 aJ per state update.
• Observed a fast 5ns transition time for skyrmion motion.

Conclusions:

• Skyrmion motion provides a viable mechanism for implementing learning automata.
• The proposed device offers significant energy reduction (99%) compared to existing methods.
• This work presents a scalable and reconfigurable platform for future edge AI hardware.