Synthetic antiferromagnetic domain wall device as artificial neuron for neuromorphic computing

Abstract

Spin-based neuromorphic computing (NC) is investigated as a power-efficient enabler for artificial intelligence [1-3]. Hence, magnetic tunnel junction (MTJ) based domain wall devices are investigated as synthetic synapses and neurons to mimic their biological counterparts [2-3]. While a biological neuron resets itself after firing, such a function has not been demonstrated experimentally in spintronic systems [3]. The self-reset feature of neurons reduces the power consumption and the complexity of NC architecture. In this work, we present a domain wall (DW) based artificial neuron that automatically resets without applying a reset pulse.At first, we simulated a synthetic antiferromagnetic (SAF) DW device where the free layer (CoFeB) is driven by spin-transfer-torque (STT) to achieve integrate function. Leakage function was achieved when the antiferromagnetic coupling (AFC) between the free and pinned layers pushes the DW towards the starting position automatically (Fig. 1). The left quarter region of the free layer is ferromagnetically coupled (FC) to the pinned layer so that as DW returns (during reset), it is stabilized at the initial position (Fig. 2(a) and Fig. 2(b)). Experimentally, the integrate and leakage functions are also achieved as illustrated in Fig. 2(d) and Fig. 2(e), suggesting that an SAF-based neuron is viable for NC.  Fig.1 Proposed design of the artificial neuron. Domain wall is moved forward by STT and (after firing), reset backward by SAF coupling.  Fig. 2 (a and b) DW position (free layer) vs time graphs for different AFC strength in the simulations. (c, d and e) Kerr images of the fabricated SAF neuron device with magnetic field in the -z direction. (c) No current applied and device is in antiparallel configuration, (d) current is applied, pushing the DW and achieving the integrate function, and (e) current is removed and the device automatically resets, indicating the leakage function.