Stochastic Switching in a Magnetic-Tunnel-Junction Neuron and a Bias-Dependent Néel-Arrhenius Model

Back hopping or telegraphic switching describes stochastic bistate oscillation in magnetic tunnel junctions (MTJ) at high bias, which significantly increases the write error rate of memory storage. Nev-ertheless, this unfavorable stochastic switching could be utilized to construct extremely compact spiking neuron circuits, where both the spike frequency and duty cycle are proportional to the applied bias voltage. This MTJ neuron is the fundamental building block of future all-spin neural networks. This paper analyzes the mechanism of stochastic switching in the MTJ neuron in detail. The self-heating effect at high bias is identified to induce thermal perturbation through a reduced energy barrier of a weakened perpendicular magnetic anisotropy. The high spike frequency is measured up to 10 MHz, which is mainly limited by the transition time between states and the multistate switching when approaching the Curie temperature of the ferromagnetic phase. Finally, to quantitatively describe the stochastic switching phenomenon, we establish a bias-dependent Neel-Arrhenius compact model, which calibrates well with experimental data. Based on this model, the potential and design guideline for realizing MTJ neurons with a spike frequency toward the gigahertz range are also discussed.