Energy-efficient synthetic antiferromagnetic skyrmion-based artificial neuronal device.

Area of Science:

• Spintronics
• Neuromorphic Computing
• Materials Science

Background:

• Magnetic skyrmions are promising for spintronics due to their unique properties.
• The skyrmion Hall effect (SkHE) hinders their application by causing undesirable motion.
• Synthetic antiferromagnetic (SAF) skyrmions offer a solution by enabling straight-line motion.

Purpose of the Study:

• To propose an integrate-and-fire (IF) artificial neuron model utilizing SAF skyrmions.
• To demonstrate a binarized neural network accelerator for image classification using this model.
• To enhance energy efficiency in neuromorphic computing devices.

Main Methods:

• Development of an IF artificial neuron model based on SAF skyrmions within an asymmetric wedge-shaped nanotrack.
• Leveraging inter-skyrmion repulsion to mimic biological neuron integrate-and-fire mechanisms.
• Implementation of a binarized neural network accelerator using SAF skyrmion-based neurons and synaptic devices.

Main Results:

• The SAF skyrmion model successfully replicates IF neuron behavior with adjustable thresholds.
• The proposed accelerator demonstrates high energy efficiency, outperforming SRAM and STT-MRAM.
• Achieved significant improvements in energy efficiency (2.31x vs. SRAM, 1.36x vs. STT-MRAM) and throughput efficiency per Watt.

Conclusions:

• SAF skyrmions provide a viable path to overcome SkHE for practical spintronic devices.
• The proposed IF neuron model and accelerator offer a highly energy-efficient solution for neuromorphic computing.
• This research paves the way for advanced, low-power AI hardware.