Bio-Inspired Spintronic Neurons for Smarter AI Hardware

What is it about?

Spintronic devices are promising building blocks for brain-like computing. Most current designs mimic only simple neuron models, limiting their biological realism. This work presents a new type of spintronic neuron device that mimics how real neurons behave more accurately than traditional artificial ones. Using domain wall motion in a magnetic tunnel junction (DW-MTJ), we replicate the FitzHugh-Nagumo neuron model, a more biologically realistic alternative to common simplified models. The device produces oscillatory spikes controlled by current and voltage, and operates with ultra-low energy. We model this device and integrate it into a spiking neural network that performs handwriting recognition with over 98% accuracy, showing its potential for efficient and brain-like AI hardware.

Featured Image

Photo by BoliviaInteligente on Unsplash

Photo by BoliviaInteligente on Unsplash

Why is it important?

Current AI hardware often relies on simplified neuron models that limit how closely machines can mimic the brain. Our work introduces a more biologically accurate spintronic neuron based on the FitzHugh-Nagumo model, enabling more realistic brain-like behavior in hardware. This neuron is built using magnetic domain wall motion, which allows it to operate efficiently with very low energy consumption (as low as 9 femtojoules per spike). We demonstrate its practical use by integrating it into a spiking neural network that achieves over 98% accuracy on the MNIST digit recognition task. This shows its potential to power faster, more energy-efficient, and brain-inspired AI systems, advancing the future of neuromorphic computing

Perspectives