Neural-like population coding based on spin-torque diode

Neuroscience studies have shown that population coding in biological systems can carry out resilient information processing with ensemble of neurons. Such strategy is valuable for the future development of electronics, particularly as the downscaling of transistors is reaching atomic limits and causing problems of large device-to-device variability and even device failure. In this work, we propose that nanoscale spin-torque diode (STD) based on a magnetic tunnel junction can be used to implement population coding. We also demonstrate that a basis set obtained from a single STD by time multiplexing can realize the generation of cursive letters. Furthermore, different activation functions of an artificial neural network have been acquired. In addition, high recognition rates of the Mix National Institute of Standards and Technology handwritten digits up to 94.88% are achieved using an output function constructed from the experimental data. Our work may provide inspiration for designing neuromorphic computing systems.