| Abstrakt: |
Domain wall (DW) devices have emerged as promising candidates for ultrafast and low power spintronic computing systems. However, the hardware implementation of these systems requires device models that are capable of describing all the underlying physical phenomena more accurately. This article presents a robust compact model that efficiently predicts the motion of DW under the influence of current induced spin Hall effect (SHE) and incorporates the effects of Dzyaloshinskii-Moriya interaction (DMI), longitudinal magnetic field, and thermal fluctuations to determine the position of the DW. The proposed model achieves a velocity of ~400 m/s that matches well with the micromagnetic simulations with an error of less than ±5%. Furthermore, the proposed model is also demonstrated to capture the behavior of leaky-integrate-fire (LIF) neuron and synapse. To enhance the robustness of the model, process variations have been incorporated to enable the analysis and optimization of DW-magnetic tunnel junction (DW-MTJ) based neuron and synapse. Finally, an all-spin neural network (NN) is implemented on MNIST dataset for image classification. The proposed DW-MTJ based NN achieves a training accuracy of 92.4% and it shows better performance in terms of area, energy, and latency when compared to NNs based other technologies such as spin transfer torque (STT)-MTJ, spin orbit torque (SOT)-MTJ, resistive random access memory (RRAM), phase change RAM (PCRAM), and ferroelectric-RAM (FeRAM). |
