Quantifying the Magnetic Interactions Governing Chiral Spin Textures Using Deep Neural Networks.

Area of Science:

• Condensed Matter Physics
• Materials Science
• Computational Science

Background:

• Chiral multilayer films host nanoscale topological spin textures crucial for advanced computing.
• Conventional methods for quantifying magnetic interactions are often specialized, time-consuming, and resource-intensive.
• High-throughput imaging of domain configurations offers a potential solution for rapid interaction analysis.

Purpose of the Study:

• To develop a machine learning (ML)-based approach for simultaneously determining key magnetic interactions (symmetric exchange, chiral exchange, anisotropy) from chiral domain images.
• To validate the ML model's predictive accuracy and its ability to uncover physical interdependencies between magnetic parameters.
• To demonstrate the utility of ML-driven techniques as a high-throughput complement to conventional experimental methods.

Main Methods:

• Development and training of a convolutional neural network (CNN) model using over 10,000 binarized domain images.
• Simultaneous prediction of symmetric exchange, chiral exchange, and anisotropy parameters from single images.
• Validation of model predictions against independent experimental measurements.

Main Results:

• The ML model achieved an R-squared value greater than 0.85 in predicting magnetic interaction parameters.
• The model independently learned and revealed physical interdependencies between the magnetic parameters.
• Model predictions on microscopy data showed consistency with established experimental findings.

Conclusions:

• ML-driven techniques provide a valuable, high-throughput method for determining magnetic interactions in chiral multilayer films.
• This approach can significantly accelerate the development of materials and devices for nanoscale electronics.
• The study highlights the potential of ML to streamline complex materials characterization processes.