Artificial Intelligence Is Used By Researchers To Find Novel Materials For High-Tech Computing

Utilising state-of-the-art artificial intelligence (AI) technologies, a team of researchers led by Trevor David Rhone, assistant professor in the department of physics, applied physics, and astronomy at Rensselaer Polytechnic Institute, has discovered unique van der Waals (vdW) magnets. The team specifically used semi-supervised learning to find transition metal halide vdW compounds with significant magnetic moments that are projected to be chemically stable.

These two-dimensional (2D) vdW magnets may find use in spintronics, quantum computing, and data storage.

Rhone has a speciality in using material informatics to find new materials with surprising features that promote science and technology. At the nexus of AI and materials science, materials informatics is a growing academic discipline. The most recent study conducted by his team was recently highlighted on the cover of Advanced Theory and Simulations.

Due to their unexpected features, 2D materials, which can be as thin as a single atom, were first discovered in 2004 and have sparked a significant amount of scientific interest. Because their long-range magnetic ordering endures even when they are reduced to one or a few layers, 2D magnets are noteworthy. The reason for this is magnetic anisotropy.

Low dimensionality and this magnetic anisotropy may interact to produce novel spin degrees of freedom, such as spin textures, which can be applied to the creation of quantum computing architectures. 2D magnets can be employed in high-performance and energy-efficient devices and span the entire spectrum of electrical properties.

Rhone and his team used AI to implement semi-supervised learning, a type of machine learning, to discover the properties of the vdW materials through high-throughput density functional theory (DFT) calculations. A combination of labelled and unlabeled data is used in semi-supervised learning to find patterns in the data and make predictions. The lack of labelled data, a significant problem in machine learning, is reduced via semi-supervised learning.

AI saves time and money, according to Rhone. “The normal materials discovery procedure necessitates pricey, time-consuming simulations on a supercomputer. Even more time and money may be spent on lab experiments. The process of finding new materials could be expedited with the help of artificial intelligence.

A supercomputer’s initial subset of 700 DFT simulations was used to train an AI model that, when applied to a laptop, could predict the properties of thousands of material candidates in a matter of milliseconds. Next, the group discovered intriguing candidate vdW materials with significant magnetic moments and low formation energy. Chemical stability, a crucial prerequisite for laboratory synthesis and later industrial applications, is indicated by low formation energy.

We can easily examine materials with various crystal structures using our framework, according to Rhone. This framework can also be used to examine mixed crystal structure prototypes, such as a data collection comprising both transition metal halides and transition metal trichalcogenides.

According to Curt Breneman, dean of Rensselaer’s School of Science, “Dr. Rhone’s application of AI to the field of materials science continues to produce exciting results.” He has advanced our understanding of 2D materials with unusual features and is also anticipated to help develop future quantum computing technologies, according to the statement.