Dipolarly coupled nanomagnets as hardware emulators of neurons for brain-like computing systems

Magnetic materials and their heterostructures can host a range of nonlinear dynamics including phase transitions and criticality, self-oscillations, synchronization, stochastic resonance, and chaos, which form the basis of many brain algorithms. Recent theory has demonstrated current-induced spike-like dynamics in (synthetic) antiferromagnets and easy-plane ferromagnets. In this research, we present the theory of non-linear hysteretic dynamics emergent in dipolarly coupled nanomagnets and show that such systems can be used as hardware emulators of neurons. Numerical models are developed to (i) explore the regimes of coherent and incoherent oscillations and (ii) estimate the performance metrics (energy, latency, area) of dipolar neurons as a function of geometry and material properties of the nanomagnets. The effect of thermal noise on the linewidth of the output signal and its spectral purity is also quantified. Finally, we show that local-level connectivity between dipolar neurons could be established at the thermodynamic limits solely using magnetic dipolar coupling, or spin diffusion. Our results will guide experimental studies of energy- and area-efficient magnetic devices for brain-like computing.