Skyr-Net: Deep learning approach for image classification of magnetic structures

Abstract

We investigate magnetic Skyrmions in Co/Pt ferromagnet-heavy metal bilayers via micromagnetic simulations and deep learning algorithms. Skyrmions can be described as small, knot-like magnetic structures that can provide high data density and offer rapid data processing with low energy consumption. Spin textures at the Co/Pt bilayer interface were simulated by solving Landau-Lifshitz-Gilbert equation using MuMax3 software and pre-trained deep learning methods are used for classifying simulated magnetic structure images. Here we introduce Skyr-Net, a novel deep learning model, which was created for classifying magnetic structure images, for the first time. The Skyr-Net achieves best result among deep learning methods for magnetic spin texture classifying in terms of the combination of computing time and accuracy (99.9%).