The research team have created an AI model called PROFiT-Net (Predicting and Optimizing Functional Inorganic Materials Through Neural Networks). (Image courtesy of KAIST)
DAEJEON, Oct. 10 (Korea Bizwire) – Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have developed a groundbreaking artificial intelligence-based material design technology that significantly improves prediction accuracy for material properties.
The research team, led by Lee Eok-Kyun and Kim Hyungjun from KAIST’s Department of Chemistry, announced on October 9 that they have created an AI model called PROFiT-Net (Predicting and Optimizing Functional Inorganic Materials Through Neural Networks).
In collaboration with Kim Won June from Changwon National University’s School of Biological and Chemical Convergence and Kim Changho from the University of California, Merced’s Department of Applied Mathematics, the team successfully enhanced AI performance by incorporating fundamental chemical concepts into the learning process.
The new model goes beyond just analyzing crystal structures, taking into account outer electron configurations, ionization energies, and electronegativity. This comprehensive approach has resulted in a dramatic improvement in prediction accuracy for various material properties.
According to the researchers, PROFiT-Net can reduce errors in predicting properties such as dielectric constants and band gaps by 10% to 40% compared to existing deep learning models. These properties are crucial in determining how materials respond to electric fields and the minimum energy required for electrical conductivity, respectively.
“This study demonstrates the potential for AI technology to evolve further by incorporating basic chemical concepts,” said Kim Hyungjun. He added, “We anticipate that this technology can be applied to various fields, including the development of semiconductor materials and functional materials.”
A visual comparison provided by KAIST illustrates the significant reduction in prediction errors achieved by PROFiT-Net compared to existing models across various material properties.
The research findings were published in the Journal of the American Chemical Society on September 25.
Kevin Lee (kevinlee@koreabizwire.com)
