Machine Learning-Based Prediction of Antiferromagnetic Skyrmion Formation

The antiferromagnetic (AFM) skyrmions are topologically protected nanoscale spin configurations that have the potential for applications in data processing, logic, and neuromorphic computing. Owing to their resistivity to stray fields, negligible demagnetization energy, and zero net topological charge, they are the prominent spin textures for spintronic devices. However, the analysis of creation, stabilization, and manipulation of AFM skyrmions for various applications is a complex task that involves solving complex equations and simulations. To resolve this issue, a novel approach is proposed based on training the machine learning (ML) neural network model. The training dataset is extracted through micromagnetic simulations by varying perpendicular magnetic anisotropy (PMA), Dzyaloshinskii–Moriya interactions (DMIs), saturation magnetization, and exchange constant at different temperatures on a shape configured nanotrack. The ML approach has shown promising results and holds the potential to significantly accelerate the development of skyrmion-based devices. It assists with identifying the optimal parameters for AFM skyrmion formation under different conditions. This neural network model achieves a 97.91% accuracy and completes inference in just 39.15 milliseconds (ms), in contrast to micromagnetic simulations that require over 200 hours to process more than 16 000 samples.