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

Neuromorphic Computing (NC) is a promising candidate for Artificial Intelligence (AI) applications. To realize NC, electronic analogues of brain components, such as synapses and neurons, must be designed. In spintronics, domain wall (DW) based magnetic tunnel junctions, which offer both synaptic and neuronal functionalities-are one of the promising candidates. An electronic neuron should exhibit leaky-integrate-fire functions, like its biological counterparts. However, most experimental studies focused only on the integrate and fire functions, overlooking the leaky function. Here, we report on a DW neuron device that achieves integration using Spin-Orbit Torque (SOT)-induced DW motion and a leaky function via synthetic antiferromagnetic coupling. By fabricating Hall bar devices in a special geometry, we could accomplish these two functionalities. During the leaky process, the maximum DW velocity exceeded 2500 µm/s. Additionally, we investigated the applicability of our neuron devices using a four-layer Leaky-Integrate-and-Fire (LIF) activated spiking neural network (SNN), achieving 92.57 % accuracy on MNIST and 84.62 % on Fashion-MNIST (F-MNIST) using the PyTorch framework. These results further validate the hardware compatibility of spintronic neurons and highlight their strong potential for enabling next-generation intelligent devices and energy-efficient neuromorphic computing. The proposed design utilizes materials used in SOT-MRAM fabrication and is compatible with CMOS fabrication. Therefore, this neuron can be readily integrated into neuromorphic computing.