Stack-Engineered Mode Selection in PtMn/(Co/Pd)_n Multilayers Enables Deterministic Analog Spin-Orbit Torque Synapses

Spin-orbit-torque (SOT) devices that support both binary and analog switching can bridge spintronic memory and neuromorphic computing, provided the switching mode can be deliberately assigned. Here, we demonstrate that in PtMn/(Co/Pd)n multilayers, the Co/Pd repeat number, n_Co/Pd, serves as a material parameter that determines the reversal mechanism and switching mode. For n_Co/Pd ≤ 7, magnetization reversal is governed by nucleation followed by domain-wall propagation, resulting in binary switching that can undergo a transition to analog behavior through electrical conditioning. For n_Co/Pd ≥ 8, increased structural modulation in the as-deposited state suppresses domain-wall propagation and stabilizes nucleation-dominated analog switching without additional processing. Applying controlled current conditioning to these high-n_Co/Pd stacks produces a hybrid state with smoother long-term potentiation and depression, an expanded number of intermediate states, and neuromorphic classification accuracy exceeding 97%. These results establish stack design and hybrid tuning as scalable strategies for energy-efficient analog SOT synapses in neuromorphic hardware.