Machine Learning Assisted Prediction of Dynamics in Current-Driven Nested Skyrmion Bags

Abstract Nested skyrmion bags, characterized by arbitrary integer topological charges (Q), hold considerable potential for application in spintronic devices. Through using micromagnetic simulations, we investigated the spin-orbit torque-driven dynamics of nested skyrmion bags, with calculation results consistent with Thiele equation analysis. For zero-Q bags with varying nesting configurations, our calculations consistently reveal motion with a zero skyrmion Hall angle. Although an increase in the number of internally nested skyrmions induces transverse elongation of the bag due to enhanced internal Magnus forces, we showed that such deformation can be suppressed by strengthening the confinement potential through either increasing the number of outermost domain wall layers or utilizing nanotrack boundaries. Furthermore, we evaluated twelve machine learning models for predicting the skyrmion Hall angle in nested skyrmion bags with arbitrary Q. Gradient boosting models and Neural Network achieved high predictive accuracy across a wide angular range, while the marked failure of linear regression unequivocally confirms the underlying nonlinearity of the physical relationships. Leveraging this predictive capability, we designed a corresponding nested skyrmion bags-based demultiplexer device. Our work enhances the understanding of current-driven dynamics in nested topological structures and offers practical insights for their use in spintronic applications such as racetrack memory and signal-routing devices.