Regensburg 2025 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MA: Fachverband Magnetismus

MA 35: PhD Focus Session: Using Artificial Intelligence Tools in Magnetism

MA 35.4: Hauptvortrag

Donnerstag, 20. März 2025, 11:15–11:45, H20

Future method for estimating parameters in magnetic films using machine learning — •Kenji Tanabe — Toyota Technological Institute, Nagoya, Japan

Estimating material parameters in fabricated materials is a crucial experiment in the field of materials science. Some parameters are difficult or time-consuming to measure. In spintronics, parameters such as the Dzyaloshinskii-Moriya exchange constant are good examples of this. If these parameters could be easily and quickly estimated, our research field would grow rapidly. Here, we present a new method for estimating parameters in magnetic films from a magnetic domain image using machine learning. A magnetic structure is well-known to be deeply related to magnetic parameters. Although a complicated magnetic structure, which often appears in an as-grown magnetic film, is considered random by human eyes, it is influenced by several magnetic energies. Thus, the characteristics of such parameters are probably hidden in the random magnetic structure. Such a relationship suggests that parameters can be estimated from a magnetic domain image by using pattern recognition. We collected a huge number of datasets of magnetic parameters and magnetic domain images made by micromagnetic simulation and/or taken by magnetic microscopes. The datasets were used as training and test data for the convolution neural network, which is a famous technique in machine learning for pattern recognition. We succeeded in the estimation of the parameters from the magnetic image using machine learning. This result may relieve future researchers from the difficulty of measuring parameters.

Keywords: Spintronics; Machine learning; magnetic domain