Neuromorphic computing implemented by Field-Free memristive switching in CoPt films with multiple inversion symmetries broken

Field-free perpendicular magnetization switching driven by bulk spin–orbit torque (SOT) shows great promise as hardware for neuromorphic computing owing to its multistate, stability, durability, and energy efficiency. Through the introduction of oblique sputtering and a vertical composition gradient, supervised and unsupervised learning are both demonstrated by field-free memristive switching actuated by bulk SOT in CoPt films. A simulated-like magnetization switching with numerous intermediate states is obtained through the application of current pulses with varying repetitions or amplitudes, which enables the simulation of synaptic weight in a supervised neural network. In addition to the synaptic plasticity of long-term potentiation and depression (LTP/D), the integrate-and-fire behavior of biological neurons is also demonstrated. With improved linearity and symmetry in LTP/D, a supervised recognition rate of 92% is achieved. In terms of unsupervised learning, a typical spike-timing-dependent plasticity (STDP) is implemented by modulating the time relationship between pre- and post-current pulses, which indicates 90% accuracy in an unsupervised spike neural network.