Recently, artificial intelligence (AI) has received widespread attention, and artificial neural networks (ANN) are the foundation of AI, while artificial neurons are the basic information processing units of ANN. Spintronic devices mimicking the function of biological neurons are considered as one of the very promising candidates for the implementation of ANN in the future. In this work, we propose to realize the leaky integrate-and-fire (LIF) properties of artificial neurons through the voltage-controlled strain gradients and current induced spin-transfer torque co-driven the motion of domain wall (DW). In addition, the proposed device can be threshold-tuned in various ways, such as strain gradient, current density, and pulse width, etc. In particular, this structure can also break the limitation of ANN learning and recognition ability by the inherent activation function. By modulating the strain gradient applied on the nanowire, the relevant nonlinear activation function can be configured, thereby improving the learning performance of ANN. As shown in Fig. 1(a), the proposed device model consists of ferromagnetic/heavy metal (FM/HM) nanowire deposited on a piezoelectric substrate. By modulating the magnitude of the electrode voltage below the piezoelectric substrate, the voltage-controlled strain gradient applied to the nanowire will be changed, and finally affects the motion of DW. The corresponding magnetoconductance (TMG) can be read by the magnetic tunnel junction (MTJ). Fig. 1(b) shows the artificial neuron with the LIF function. The DW motion dynamics depend on the competition between the gradient magnetoelastic energy of the nanowire and the current driving force. The DW undergoes a total of five cycle current pulses, which finally reaches the threshold TMG of 0.63 and is detected by the MTJ. Next, the DW "fires" a signal output, and resets to the initial state. The above simulation results reveal the great potential of magnetic DW-based electric field modulation of artificial neurons for applications of next-generation artificial intelligence devices.
Fig. 1 (a) Schematic diagram of voltage-controlled DW-based nanodevices. (b) LIF behaviors.