In the field of artificial intelligence, the use of artificial neural networks inspired by the operation of our brains has been identified as a promising way of reducing the energy consumed by the algorithms that process information. However, no neural network incorporating compact, energy-efficient spintronic components was yet available.
Researchers at Irig have succeeded in developing a novel magnetic tunnel junction with two free layers capable of emulating the operation of a spiking neuron. This artificial neuron, stimulated by the applied electrical voltage, emits pulsed output signals whose frequency depends on the amplitude of the current flowing through the junction. The device operates without the need a magnetic field. Furthermore, it is not perturbed by stray magnetic fields, making it an ideal candidate for mimicking the behavior of neurons in a dense neural network. The spintronic neuron's compact size (less than 100 nm) and its low power consumption (around 10-11 J/pulse) make it a major asset in the design of embedded electronics applications.
Figure: Schematic of a perpendicular Magnetic Tunnel Junction with free layers that mimic the activity of a spiking neuron. On the right, the curve of its mean frequency response as a function of applied voltage.