Home > News

Neuromorphic computing with nanoscale spintronic oscillators

Nowadays artificial intelligence algorithms have become very efficient for speech or image recognition. But running these programs on conventional computer architectures uses ten thousands time more energy than a brain to complete the same task. Dramatic reduction of energy consumption could be achieved by building brain-inspired computers massively integrating artificial neurons and synapses. However, assembling such huge amount of components (at least a million) in a reasonable space requires nanoscale devices. Until now, nobody had been able to fabricate a nano-neuron stable enough to compute reliably despite its small dimensions.

In this paper [1] an experimental demonstration of stable neuron behavior at the nanoscale is made by performing speech recognition with spin-torque nano-oscillators. The highly stable non-linear magnetization dynamics of spin-torque oscillators was used to mimic a neuron behavior. Each gyration of the magnetization is converted thanks to magneto-resistive effects into an oscillation of the voltage emitted by the oscillator. These oscillations are similar to the spikes emitted by a neuron. Leveraging this response, spoken digits said by different speaker were recognized with a success rate of 99.6%. These results open up a path to compact intelligent chips able to learn and to adapt to the ever changing and ambiguous situations of the real world. Such chips could find numerous applications in every field where recognition is critical, for instance for smart motion control of robots or autonomous vehicles, medical diagnosis assistance or prosthesis design.

This work was made in collaboration between CNRS/Thales laboratory, the National Institute of Advanced Industrial Science and Technology (AIST), the Center for Nanoscience and Nanotechnology (C2N) and the National Institute for Science and Technology (NIST).

[1 J. Torrejon, M. Riou, F. Abreu Araujo, S. Tsunegi, G. Khalsa, D. Querlioz, P. Bortolotti, V. Cros, K. Yakushiji, A. Fukushima, H. Kubota, S. Yuasa, M. D. Stiles, J. Grollier, Neuromorphic computing with nanoscale spintronic oscillators. Nature. 547, 428–431 (2017).