Метод нейронных сетей для моделирования скирмионной решетки в композитном мультиферроике

Abstract

Magnetoelectric nanofilms are of great interest as functional elements of ultra-dense memory cells. In the ground state they may contain various topological magnetic vortex structures of several nanometers in size. The qualitative and quantitative properties of such structures strongly depend on a set of physical parameters. To calculate the ground state configuration with given parameters, we use the steepest descent method; to study a large parametric space, however, significant computational resources are required. To solve this problem, we propose the use of artificial neural networks (ANN), which can help us uncover the relationship between combinations of parameters and the corresponding ground state configurations, using a relatively small number of pre-computed configurations as training data. The application of the ANN allows one to avoid excessive computational costs in the study of the parametric space and narrow down the parametric area in which the existence of stable non-trivial ground state configurations in the form of a stable skyrmion crystal is possible.