Hardware Model for Stochastic Neuron Based on Magnetic Tunnel Junction in the Subcritical Current Switching Regime.

The stochastic neuron has great importance in neural networks and is one of the most important subjects in machine learning algorithms. Hardware implementation of neural networks has always been of interest to researchers and can significantly increase the performance and applications of neural networks. Hence hardware implementation of the stochastic neuron is also important. In this paper, utilizing stochastic behavior of MTJs in subcritical current regime a hardware model for stochastic neurons is proposed. Using HSpice tool, the proposed model was simulated and simulation results show that the proposed model functionality is similar to the mathematical description of the stochastic neuron and the error of this model is always less than 4.8٥/٠ compared to the mathematical description of the stochastic neuron. Also using corner simulation, it was shown that this model performs properly even in the presence of process variation and its error rate is less than 15.46٥/o and 17.43٥/٠ compared to the mathematical model and ideal simulation, respectively. [ABSTRACT FROM AUTHOR]