Cross-modal neuromorphic computing in GdFeCo devices by spin–orbit torque induced field-free magnetization switching

Abstract

Field-free spin–orbit torque (SOT)-driven perpendicular magnetization switching presents a highly promising pathway for neuromorphic computing applications. In this study, field-free SOT switching is achieved in a GdFeCo device driven by out-of-plane spin polarization. The anomalous Hall effect loop shifts combined with energy-dispersive spectroscopy analysis confirm that the switching mechanism originates from an out-of-plane spin polarization component (σz), generated by a compositional gradient of Gd. Subsequently, a fully spin-based generative adversarial network architecture utilizing GdFeCo SOT devices is built. By exploiting the tunable neuronal and synaptic characteristics of GdFeCo-SOT systems, we achieve cross-modal learning between image and speech domains. A remarkable recognition accuracy rate of above 80% is attained when the generated speech is assessed. Our work introduces a field-free SOT switching strategy based on out-of-plane spin-polarized components, offering significant potential for implementation in neuromorphic computing.