Magnetic vortex dynamics in vortex based nano-oscillators for hardware artificial intelligence applications

Details

Supervisors

Abreu Araujo, Flavio

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil physicien, à finalité spécialisée

Abstract

enAs the energy demands of digital computation continue to rise, driven by the rise of artificial intelligence, there is growing interest in neuromorphic computing, a brain-inspired approach promising greater energy efficiency. This thesis investigates spin-torque vortex oscillators (STVOs), nanoscale magnetic devices capable of emulating neuron-like behavior, as a hardware platform for such computing architectures. In the first part, we focus on improving the Thiele equation, a model used to describe the motion of magnetic vortices in STVOs. Traditionally, this model assumes constant gyrotropic and damping terms, which limits its predictive accuracy when the vortex core moves far from equilibrium, regimes that are typical in neuromorphic applications. To address this, we developed a semi-analytical model based on a deformable vortex core ansatz that captures the dip. This allowed us to compute the gyrotropic and damping terms as functions of the core position. We found that the gyrotropic term remains close to its approximated value across displacement amplitudes, supporting continued use of a fixed gyroconstant. In contrast, the damping factor shows a marked increase with oscillation radius. A second, fully numerical approach, based on micromagnetic simulations, confirmed these trends and highlighted the role of realistic deformations not captured analytically. Together, these results yield a more accurate and physically grounded description of vortex dynamics, improving our ability to model and design STVOs for artificial neuron applications. The second part examines the emergence of multiple vortex core polarity reversals, including triple and quadruple events, observed for the first time in nanodots with low aspect ratios and thin (5 nm) magnetic layers. These reversals exhibit instability and sensitivity to initial conditions, likely due to the chaotic dynamics of polarity reversal. Correlations with internal magnetization changes, particularly a threshold-like behavior in the vortex dip region, suggest new mechanisms behind these phenomena. Understanding these dynamics could help emulate spiking behavior in spintronic neurons. Overall, this work deepens the theoretical and computational understanding of STVOs, refining key modeling tools and revealing complex behaviors relevant to neuromorphic hardware. These studies could help realize scalable and energy-efficient brain-inspired computing platforms.