AI for materials
Everything depends on materials.
From improving new batteries for electric vehicles to studying the atomic structure of metallic glass, ORNL researchers are laser-focused on finding the revolutionary materials that will make the world a better, more sustainable place in the 21st century.
How does one discover a new material, you ask? Today, it’s with an assist from AI.
There are billions of potential element-and-material combinations that could yield revolutionary new materials. Finding the best combination is a monumental challenge, tailor-made for AI. ORNL researchers are tackling it with large language models, AI programs that can recognize and generate text and human language.
“We have trained a large language model on all the experimental data that we have or is in the literature,” said Bobby Sumpter, section head for Theory and Computation at ORNL’s Center for Nanophase Materials Sciences. “For us to digest all of that is really hard. But for the large language model to digest it and pull it together is actually fairly quick, especially on a computer like Frontier.” Located at ORNL, Frontier is among the world’s fastest supercomputers.
ORNL scientists are developing models to predict the properties of new materials in a millisecond. Rather than spend days or weeks sifting through the best material candidates, scientists can have the top options ready for development and testing in a matter of minutes.
Other breakthrough advances include the rapid analysis of imaging data. For example, when scientists use powerful microscopes to see how atoms of a material respond to changing temperature and pressure, extracting the information is extremely laborious and time-consuming.
This is where deep learning provides a solution by enabling rapid image inference and analytics to identify material defects and to follow them in real time.