Leaky-Integrate-Fire Neuron via Synthetic Antiferromagnetic Coupling and Spin-Orbit Torque.

Area of Science:

• Spintronics
• Neuromorphic Computing
• Artificial Intelligence

Background:

• Neuromorphic computing (NC) requires electronic analogues of biological neurons.
• Spintronic domain wall (DW) devices offer potential for synaptic and neuronal functions.
• Existing electronic neurons often lack the crucial 'leaky' function.

Purpose of the Study:

• To design and demonstrate a DW neuron device with both integrate-and-fire and leaky functionalities.
• To investigate the performance of these spintronic neurons in a spiking neural network (SNN).
• To assess the compatibility of the proposed neuron design with existing fabrication processes.

Main Methods:

• Fabrication of Hall bar devices utilizing Spin-Orbit Torque (SOT)-induced DW motion for integration.
• Implementation of synthetic antiferromagnetic coupling to achieve the leaky neuron function.
• Testing the neuron device within a four-layer Leaky-Integrate-and-Fire (LIF) activated SNN using PyTorch.

Main Results:

• The fabricated DW neuron successfully demonstrated both integration and leaky functions.
• The leaky process achieved a maximum DW velocity exceeding 2500 µm/s.
• The SNN achieved 92.57% accuracy on MNIST and 84.62% on Fashion-MNIST.

Conclusions:

• The developed spintronic neuron device effectively replicates biological neuron functions, including the leaky mechanism.
• The device shows strong potential for energy-efficient neuromorphic computing and next-generation intelligent devices.
• The design is compatible with SOT-MRAM and CMOS fabrication, enabling straightforward integration into existing technologies.