Decoupled Spin Hall Voltage Readout for Domain‐Wall Synapses in Embedded Neuromorphic Computing

Domain‐wall (DW) spintronic synapses are attractive for embedded neuromorphic hardware, where DW position encodes nonvolatile multi‐level weights. However, practical DW synapses remain limited by readout schemes that rely on magnetoresistive conductance, restricting dynamic range, or couple sensing and programming paths, weakening voltage output and risking read‐disturb. Here, we introduce a domain‐wall spin Hall (DWSH) synapse with decoupled voltage readout. The device uses current‐induced DW motion for weight programming, while a separated vertical spin Hall path converts DW position into an output voltage through spin‐charge conversion. Differential sensing drives the minimum output voltage toward zero, avoiding the magnetoresistance‐imposed conductance ratio limit, while efficient spin‐charge conversion enables intrinsic voltage gain and may reduce peripheral sensing requirements. Micromagnetic and magnetoelectronic circuit simulations predict 4 ns programming pulses, 1 ns voltage readout, sub‐picojoule read energy, linear and symmetric programming, dynamic range above 50, tunable weight precision, voltage amplification of 1.92, and retention extendable to 19.8 days. A 200‐neuron spiking neural network predicts 83.0% MNIST recognition accuracy with 2.35 nJ inference energy. The results identify decoupled spin Hall voltage readout as a promising architecture for embedded neuromorphic systems requiring robust analog output, low read energy, reduced peripheral sensing burden, and analog weight programmability.