Integrated Field‐Free SOT Domain‐Wall Synapses and MTJ Stochastic Neurons for Hardware Boltzmann Machines

ABSTRACT Spintronics offers a promising approach to energy‐efficient neuromorphic computing by integrating the functionalities of synapses and neurons within a single platform. A key challenge is achieving field‐free spin–orbit torque (SOT) control in both synaptic and neuronal devices within an industry‐adopted spintronic stack. In this study, we demonstrate field‐free SOT spintronic synapses utilizing a CoFeB ferromagnetic thin‐film system, where asymmetrical device design and specifically engineered lateral notches in the CoFeB thin film facilitate effective domain wall (DW) nucleation, movement, and pinning/depinning. This method yields multiple analog, non‐volatile resistance states with enhanced linearity and symmetry, enabling programmable, stable synaptic weights. We provide a systematic measurement approach to improve the linearity and symmetry of the synapses. Additionally, we demonstrate nanoscale magnetic tunnel junctions (MTJs) that function as SOT‐driven stochastic neurons, exhibiting current‐tunable, Boltzmann‐like probabilistic switching behavior, which provides an intrinsic in‐hardware Gibbs sampling capability. By integrating these synapses and neurons into a Boltzmann machine complemented by a classifier layer, we achieve recognition accuracies greater than 98% on the MNIST dataset and 86% on Fashion‐MNIST. This work establishes a framework for field‐free synaptic and neuronal devices, setting the stage for practical, materials‐compatible, and all‐spintronic neuromorphic computing hardware.