Material informatics for functional magnetic material discovery

R. M. Rowan-Robinson,Z. Leong,S. Carpio,C. Oh,N. A. Morley

DOI: https://doi.org/10.1063/9.0000657

IF: 1.697

2024-01-01

AIP Advances

Abstract:Functional magnetic materials are used in a wide range of “green” applications, from wind turbines to magnetic refrigeration. Often the magnetic materials used contain expensive and/or scarce elements, making them unsuitable for long term solutions. Further, traditional material discovery is a slow and costly process, which can take over 10 years. Material informatics is a growing field, which combines informatics, machine learning (ML) and high-throughput experiments to rapidly discover new materials. To prove this concept, we have devised a material informatics workflow and demonstrated the core components of natural language processing (NLP) to extract data from research papers to create a functional magnetic material database, machine learning with semi-heuristic models to predict compositions of soft magnetic materials, and high-throughput experimental evaluation using combinatorial sputtering and high-throughput magneto-optic Kerr effect (MOKE) magnetometry. This material informatics workflow provides a quicker, cheaper route to functional magnetic materials discovery.

materials science, multidisciplinary,physics, applied,nanoscience & nanotechnology

What problem does this paper attempt to address?

The problem this paper attempts to address is the inefficiency and high cost of traditional material discovery methods in the development of functional magnetic materials. Specifically, traditional material discovery methods are not only time-consuming (potentially taking over 10 years) but also often rely on minor adjustments to the compositions of existing materials, which cannot achieve the rapid breakthroughs needed for material discovery. Additionally, many existing magnetic materials contain expensive or scarce elements, making them unsuitable for long-term solutions.

To tackle these issues, the authors propose a materials informatics-based workflow that accelerates the discovery of functional magnetic materials by combining natural language processing (NLP), machine learning (ML), and high-throughput experiments. This approach aims to:

1. **Database Construction**: Utilize NLP to extract data from research papers and create a database of functional magnetic materials.
2. **Material Prediction**: Use ML models to predict the compositions of soft magnetic materials.
3. **Experimental Validation**: Validate the predictions through high-throughput experiments (such as combinatorial sputtering and high-throughput magneto-optical Kerr effect magnetometry) and feed the experimental data back into the database to continuously optimize the ML models.

Through this workflow, the authors hope to provide a faster and more economical method for discovering new functional magnetic materials, particularly those for green technology applications.