Domestic researchers have developed an artificial intelligence (AI) that creates optimal catalyst combinations for water electrolysis.
The National Research Foundation of Korea (NRF) announced on 14th that a research team led by Professor Kang Jung-koo from the Korea Advanced Institute of Science and Technology (KAIST) and a research team led by Professor William Goddard of the California Institute of Technology (Caltech) jointly designed a machine learning algorithm to predict the performance of multi-element alloy catalysts and developed a water electrolysis catalyst that surpasses existing rare metal-based catalysts.
Water electrolysis is a reaction that decomposes water into hydrogen and oxygen, and water electrolysis catalysts play a role in reducing the amount of energy consumed in this process. However, electric catalysts for water electrolysis face difficulties in terms of economic viability and sustainability due to the large amounts of rare metals such as platinum and iridium contained from the production stage. In particular, in the case of 'multi-element alloy catalysts' that combine three or more metal elements, astronomical time and expense were required to find the optimal combination.
In response, the research team developed a catalyst design method through machine learning to rapidly explore multi-element alloy catalysts and to find the optimal composition. The optimal alloy catalyst formulated through machine learning models exhibited low overpotentials of 24 mV in hydrogen evolution reactions and 204 mV in oxygen evolution reactions. This performance significantly surpasses that of catalysts based on platinum and iridium oxide.
Professor Kang Jung-koo noted, "Through AI-based catalyst design, we were able to discover multi-element alloys with excellent performance in a short time," adding, "The developed alloy catalyst has demonstrated long-term stability for over 100 hours and proves that it can be applied in actual hydrogen production systems."
The results of this study were published in the international journal Proceedings of the National Academy of Sciences (PNAS) on 7th.
References
PNAS (2025), DOI: https://www.pnas.org/doi/10.1073/pnas.2504226122