A Spin-Orbit Torque Device Toward Energy-Efficient Neuronal Population Performance

Precise control over the intermediate states in a spin-orbit torque (SOT) device with multi-levels has been demonstrated a unique strategy toward high dense storage for an analog memory, which bridges the solid-state device and the neuromorphic computing into a newly developed research field in recent years. In this study, we demonstrate a SOT switching mechanism in the devices composed of Pd/Co/Ta trilayer featuring robust stabilization of intermediate magnetic states, which leads to precise controllability and superior stability of multi-levels. Magnetization reversal of Pd/Co/Ta devices by SOT takes place through the gradual nucleation of reversed domains with restricted domain wall (DW) motion due to DW precession over Walker breakdown, which guarantees stable magnetic multi-levels. Furthermore, we demonstrate that individual reversed magnetic domain can be regarded as a single neuron, and the whole device becomes an artificial neuron population (NP). Therefore, the period required for the magnetization reversal can represent the response time of NP. The DW velocity or nucleation rate in SOT devices thus can be corresponding to the interaction between neurons; accelerated DW motion (or nucleation), leading to faster reversal, can mimic the case that excited neurons would influence neighboring ones to quickly respond to stimulus together. We demonstrate that the relationship between stimulus strength and duration, commonly observed in biological NPs, can be achieved in the artificial NP, composed of a Pd/Co/Ta SOT device. Furthermore, we reveal the modification of the response time of artificial NP under the lowest energy consumption by tuning the DW velocity. Our work not only explores the fundamental physics of the SOT dynamics via both experimental imaging and theoretical micromagnetic simulation, but also demonstrates the potential neuromorphic application as a new branch of neuronal population. Our findings will open a revolutionary avenue for establishing biomimetic neurons toward the energy and time efficient neuromorphic technology.