Understanding the Magnetic Microstructure through Experiments and Machine Learning Algorithms

Area of Science:

• Materials Science
• Condensed Matter Physics
• Machine Learning Applications

Background:

• Machine learning techniques are increasingly applied across scientific disciplines.
• Understanding magnetic properties from microstructural images is crucial for materials development.
• Co/Pd multilayers exhibit complex magnetic domain structures sensitive to external perturbations.

Purpose of the Study:

• To investigate the modification of magnetic and structural properties in Co/Pd multilayers under ion irradiation.
• To develop and validate a machine learning model for predicting magnetic properties from microstructural images.
• To explore the application of image regression using convolutional neural networks (CNNs) for materials characterization.

Main Methods:

• Co/Pd multilayers were irradiated with Ar+ ions at varying energies (50 and 100 keV) and fluences.
• Magnetic Force Microscopy (MFM) was used to observe changes in magnetic domain structures.
• Micromagnetic simulations were performed to complement experimental findings.
• A CNN was trained and validated on 960 simulated domain images for property prediction.

Main Results:

• Ion irradiation induced domain fragmentation and weaker magnetic contrast up to 10^14 ions/cm^2.
• Higher fluences led to feather-like domains and altered local magnetization distribution.
• Simulations showed that changes in anisotropy and exchange constants qualitatively matched experimental observations.
• The trained CNN accurately predicted magnetic properties from domain images with up to 93.9% accuracy.

Conclusions:

• Advanced neural networks, particularly CNNs, are effective tools for image regression in materials science.
• This approach enables direct prediction of integral magnetic properties from microscopic features.
• The findings highlight the potential of ML-driven characterization for materials under external perturbations.