Analysis of properties related to spin textures, such as the magnetic vortex state, is mainly based on spin configuration data, which is directly related to magnetic parameters involved in the system's Hamiltonian. Here, we focus on magnetic parameter estimation by implementing the machine learning (ML) approach, especially on magnetic force microscopy (MFM) images of vortex states within nanodots generated by micromagnetic simulation. The exchange constant Aex and saturation magnetization Ms as well as exchange length as a reduced parameter Lex(Aex, Ms) are estimated by different convolutional neural network (CNN) models. We also evaluated the CNN models, trained on simulated MFM images with non-zero temperature, on a reference experimental MFM image and found the performance to a satisfactory level of accuracy. Moreover, the same CNN models, trained for binary classification of vortex states based on helicity from MFM images, successfully identified the vortex helicity from simulated as well as experimental MFM images. These findings show the possible application of ML in magnetic parameter estimation and the analysis of magnetic vortex states simply with images obtained from this commonly used imaging technique that is significant in efficient investigation of material properties based on intrinsic parameters for spintronic device applications.
Read full abstract