Neural network-assisted indirect probing of magnetic skyrmions in MnSi via magnetic entropy variation

Abstract

Magnetic skyrmions emerge as promising candidates for next-generation magnetic storage technologies. However, their direct detection requires advanced techniques such as Lorentz transmission electron microscopy or small-angle neutron scattering. In this study, we propose an indirect approach to identify skyrmions in MnSi through the analysis of magnetic entropy change (ΔSM). Magnetocaloric measurements reveal both first- and second-order magnetic phase transitions, where subtle entropy variations correspond to the skyrmion phase. To enhance sensitivity and interpretability, we employ artificial intelligence (AI) techniques—convolutional neural networks (CNNs) and long short-term memory (LSTM) networks—to analyze ΔSM data. Fourier-transformed spectral representations enable CNNs to capture spatial correlations, while LSTMs identify dynamic field-dependent patterns. The models reproduce the experimentally reported skyrmion region (170–230 mT) and distinguish between formation and annihilation processes. These results demonstrate that AI-assisted magnetic entropy analysis provides an effective, low-cost, and experimentally accessible approach for probing magnetic skyrmions, offering a generalizable framework for identifying topological spin textures using conventional magnetometry.