Neural network analysis of the magnetization reversal in magnetic dot arrays

TL;DR

This paper uses neural networks and simulations to analyze the complex magnetization reversal processes in ultra-dense magnetic dot arrays, revealing nonuniform configurations and soliton-antisoliton pairs.

Contribution

It introduces a combined simulation and neural network approach to study magnetization dynamics in magnetic dot arrays, including a novel classification method for magnetic configurations.

Findings

Detection of soliton-antisoliton pairs at multiple scales

Identification of nonuniform magnetic configurations

Effective classification of intra-dot states

Abstract

We simulated the remagnetization dynamics of the ultra-dense and ultra-thin magnetic dot array system with dipole-dipole and exchange coupling interactions. Within the proposed 2D XY superlattice model, the square dots are modeled by the spatially modulated exchange-couplings. The dipole-dipole interactions were approximated by the hierarchical sums and dynamics was reduced to damping term of the Landau-Lifshitz-Gilbert equation. The simulation of 40 000 spin system leads to nonequilibrium nonuniform configurations with soliton-antisoliton pairs detected at intra-dot and inter-dot scales. The classification of intra-dot magnetic configurations was performed using the self-adaptive neural networks with varying number of neurons.