Chaos and Relaxation Oscillations in Spin-Torque Windmill Spiking Oscillators

Spintronic neurons that emit sharp voltage spikes are needed for neural-network hardware, for fast data processing with low power consumption. Conventional spintronic neurons, $i.e.$ spin-torque nano-oscillators (STNOs), emit sinusoidal waveforms, though. The authors suggest how to imitate neuron spiking in STNOs in which both magnetic layers are free and thin enough to be switched by current. Simulations show that the chaotic behavior of these frequency-tunable devices can be adjusted by the magnetic stack and current. The proposed spintronic relaxation oscillator is a promising building block for hardware neuromorphic chips, leveraging nonlinear dynamics for computing.