The purpose of neuromorphic computing is to take inspiration from the brain to build hardware neural networks that can learn to perform useful tasks with low energy consumption [1]. In this talk, I will show that spintronic nano-oscillators based on magnetic tunnel junctions can act as artificial neurons [2]. I will present our first results of pattern recognition with small networks of coupled oscillators [3]. I will then show that these microwave nano-neurons open the path to wireless deep learning.
[1] J. Grollier, D. Querlioz, et M. D. Stiles, « Spintronic Nanodevices for Bioinspired Computing », Proc. IEEE, vol. 104, no 10, p. 2024 2039, oct. 2016.
[2] J. Torrejon et al., « Neuromorphic computing with nanoscale spintronic oscillators », Nature, vol. 547, no 7664, p. 428 431, juill. 2017.
[3] M. Romera et al., « Vowel recognition with four coupled spin-torque nano-oscillators », Nature, vol. 563, no 7730, p. 230, nov. 2018 .
CNRS/Thales lab, Palaiseau, France