Elena Vedmedenko
University of Hamburg
Skyrmions are quasiparticle magnetic textures with long lifetimes due to their non-trivial real space topology [1]. Owing to their metastability, the skyrmions may diffuse without being destroyed in a broad temperature range. This Brownian motion appears particularly interesting in confined geometries that are valuable for different computing concepts like race-tracks or reshuffle devices for stochastic computing. Our micromagnetic simulations of such confined systems show accessibility of several metastable configurations associated with different topological charges starting from randomly generated initial states [2]. Particularly interesting appear the stochastic skyrmions' dynamics leading to a pattern formation of the skyrmion density, involving two distinct time regimes – the internal skyrmion eigenmodes and the diffusive motion of skyrmion ensembles. In time-averaged measurements this skyrmion density often possess integer values just like an ensemble of static skyrmions or looks like a superposition of spin spirals and skyrmions although this is not the case in real time [2]. This makes it impossible to extract the real skyrmion numbers from time-averaged measurements by conventional methods that restricts the possibilities of skyrmion utilization as information bits. Here, we show how machine learning categorically outperforms standard advanced techniques for distinguishing diffusing and spatially stable skyrmions and extracting the total topological charge from time-integrated data of the z direction of the magnetic moments [3].
Fig.1. Representative examples of time-averaged images with metastable skyrmions in irregular islands used in the analysis of [3].
References
[1] J. Hagemeister et al., Nat. Commun. 6, 8455 (2015). [2] A. Schäffer, L. Rozsa, J. Berakdar, E. Y. Vedmedenko, and R. Wiesendanger, Comm. Phys. 2, 72 (2019). [3] T. Matthies, A. Schäffer, T. Posske, R. Wiesendanger, and E. Y. Vedmedenko, Phys. Rev. Appl. 17, 054022 (2022).