Ultrafast stochastic spiking neurons based on field-free magnetic tunnel junction

In this study, we showcase the hardware implementation of stochastic spiking neurons (SSNs) utilizing the back hopping phenomenon in a field-free magnetic tunnel junction with a well-established fabrication process and excellent scalability. The neural network constructed using this SSN array demonstrates an ultrafast response speed on the order of tens of nanoseconds, a high accuracy of 97.6% in handwritten digit classification, and superior adaptability across various tasks, surpassing the performance of the majority of reported SSNs. This research carries significant implications, positioning our SSNs as promising candidates in the realm of efficient and robust neuromorphic computing applications.