Machine-learning-enabled vectorial opto-magnetization orientation

TL;DR

This paper introduces a machine learning inverse design method to efficiently and accurately generate light beams for controlling arbitrary three-dimensional magnetization orientations, improving flexibility and efficiency over traditional methods.

Contribution

The paper presents a novel machine learning approach for inverse design of light beams to achieve prescribed 3D magnetization orientations, enhancing control capabilities.

Findings

The method accurately produces incident beams for arbitrary 3D magnetization orientations.

Machine learning inverse design is time-efficient and flexible.

Applicable to various magnetization control scenarios.

Abstract

Manipulation of light-induced magnetization has become a fundamentally hot topic with a potentially high impact for atom trapping, confocal and magnetic resonance microscopy, and data storage. The control of the magnetization orientation mainly relies on the direct methods composed of amplitude, phase and polarization modulations of the incident light under the tight focusing condition, leaving the achievement of arbitrary desirable three-dimensional (3D) magnetization orientation complicated, inflexible and inefficient. Here, we propose a facile approach called machine learning inverse design to achieve expected vectorial opto-magnetization orientation. This pathway is time-efficient and accurate to produce the demanded incident beam for arbitrary prescribed 3D magnetization orientation. It is highlighted that the machine learning method is not only applied for magnetization orientations, but also widely used in the control of magnetization structures.