МАГНИТНЫЕ ЭЛЕМЕНТЫ НЕЙРОННЫХ СЕТЕЙ. СВОЙСТВА МАГНИТНЫХ НЕОДНОРОДНОСТЕЙ В СИСТЕМАХ ОГРАНИЧЕННОЙ ГЕОМЕТРИИ

Abstract

This paper discusses the prospects for using magnetic nanostructures as elements of neural networks. At present neural network learning programs are actively used in analyzing and processing large data arrays; however, the development of computer technologies based on the neural network principle still remains open. Possibilities for using magnetic elements as physical carriers of information bits in these systems attract much attention from researchers and technologists due to the presence of several easily controlled parameters (order parameter) in the magnetic system, possibilities for the dimensionality reduction in magnetic elements by using magnetic nanostructures (domain boundaries, vortices, ckyrmions), superquick switching between magnetic states and some other factors. One of the key aspects of research in this regard is to determine basic controlled magnetic parameters in restricted geometries and to identify ways of controlling these parameters through internal and external factors. The paper presents a research on the magnetic ground state in restricted geometries. It deals with the magnetic state rebuilding in the system under changes in both external factors (applied magnetic field, sample dimensions) and internal ones (magnetic anisotropy constant, Dzyaloshinskii-Moriya interaction constant). Calculations were performed within the framework of micromagnetic modelling using the Object Oriented MicroMagnetic Framework ( OOMMF) sogtware. It is shown that the anisotropic exchange interaction (Dzyaloshinskii-Moriya interaction) has a significant effect on the magnetization distribution in restricted geometries. Namely, when changing the value of the Dzyaloshinskii-Moriya constant in the system with uniaxial magnetic anisotropy there is a series of phase transitions observed between magnetic states of different types: transitions from the homogenous magnetic state into the skyrmion-type vortex state (domain structure with the skyrmion-type unidomain state) with subsequent domain structure reversal when changing the value of the Dzyaloshinskii-Moriya constant. In the case of magnetic anisotropy of easy -axis type, chirality and properties of the structures in question do not depend on the constant symbol of the Dzyaloshinskii-Moriya interaction.