Stochastic spiking in magnetic droplet solitons

Abstract

The nonlinear dynamics of droplet solitons make them highly promising candidates for brain-inspired computing, including neuromorphic systems, deep physical neural networks, physical spiking neural networks, and transformer-based physical architectures. Integrating droplet solitons into neuromorphic device architectures enables low energy consumption and ensures CMOS compatibility, positioning them for efficient hardware implementations of cognitive and perception-related tasks. However, previous studies have primarily relied on the application of external magnetic fields, which poses significant challenges for on-chip integration in practical computing applications. In this paper, we investigate in detail the nucleation, stability, and dynamics of droplet solitons in nanoconstriction-based spin Hall nano-oscillators (NC-SHNOs) by incorporating unconventional spin–orbit torque under bias-field-free conditions using micromagnetic simulations. We demonstrated stochastic spiking-like behavior in droplet solitons due to drift instability. The intrinsic stochastic spiking-like behavior can be implemented to generate true random numbers, which successfully pass the National Institute of Standards and Technology test suite. Our results reveal the nonlinear dynamics of droplet solitons in NC-SHNOs and deepen our understanding of their practical potential as hardware-based true random number generators, with promising implications for secure communication, neuromorphic computing, and reservoir computing architectures.