Machine learning catches elusive magnetic phenomenon

A magnetic insulator can induce magnetization in an adjoining non-magnetic insulator through a phenomenon known as the magnetic proximity effect. While this effect could be harnessed to develop new low-power electronic and spintronic devices, distinguishing it in real materials can be challenging. Andrejevic et al. developed a machine learning algorithm that automatically determines whether this effect is measured in reflectometry data.

The authors found their machine learning framework cut the detection limit by a factor of two, enabling them to discern the magnetic proximity effect more conclusively near their instrument’s resolution limit.

“The method allows us to better resolve the magnetic proximity effect in reflectometry measurements and can potentially inform the design of analysis frameworks for other experimental techniques,” author Nina Andrejevic said. “The workflows and code developed in this work are already being tested by researchers at the National Institute of Standards and Technology to support data analysis in the broader scientific community.”

To develop their framework, the researchers used a machine learning building block called a convolutional neural network to construct an autoencoder, which is a model that can discover structure within unlabeled data. This drew out patterns and relationships in the data, including subtle spectral signatures that would otherwise be ignored.

“Original spectral data are high dimensional, and machine learning facilitates analysis in a learned, lower-dimensional latent space,” author Zhantao Chen said. “Nuanced spectral features reflecting the proximity effect can be amplified and thus better identified by machine learning in such a space.”

Andrejevic added that the work could also be applied to the elusive superconducting proximity effect, which has applications in quantum computing.

Source: “Elucidating proximity magnetism through polarized neutron reflectometry and machine learning,” by Nina Andrejevic, Zhantao Chen, Thanh Nguyen, Leon Fan, Henry Heiberger, Ling-Jie Zhou, Yi-Fan Zhao, Cui-Zu Chang, Alexander Grutter, and Mingda Li, Applied Physics Reviews (2022). The article can be accessed at https://doi.org/10.1063/5.0078814 .