Adjustable artificial neuron based on vortex magnetic tunnel junction

Yuxuan Wu,Yanxiang Luo,Like Zhang,Shige Dai,Baoshun Zhang,Yan Zhou,Bin Fang,Zhongming Zeng
DOI: https://doi.org/10.1063/5.0195602
IF: 4
2024-03-18
Applied Physics Letters
Abstract:In this Letter, we demonstrate an adjustable artificial neuron based on vortex magnetic tunnel junction (MTJ). By applying a bias current to vortex MTJ, the device exhibits splendid characteristics of stochastic switching and nonlinear rectification. The stochastic switching probability induced by spin transfer torque as a function of bias current can simulate sigmoid activation functions. The nonlinear spin-torque microwave rectification through injection locking is similar to a ReLU-like activation function. These two behaviors further are used to perform the recognition of handwritten digits in the Mixed National Institute of Standards and Technology database, with a produced accuracy of up to 93.56% and 93.25%, respectively. Our work provides a potential way for the construction of artificial neuron based on vortex MTJ.
physics, applied
What problem does this paper attempt to address?