AI Accelerates Discovery of Ultra-Efficient Energy Conversion Materials, Slashing Development Time from Years to Just Months
- categorization Research
- writing date 2026.05.11
- author Communication Team
- hits 21
AI Accelerates Discovery of Ultra-Efficient Energy Conversion Materials, Slashing Development Time from Years to Just Months
△ AI-based new material discovery system and performance comparison of newly discovered materials.
By leveraging AI, the time required to develop new materials, previously spanning several years, has been reduced to just a few months. A research team in Korea has succeeded in rapidly identifying energy conversion materials with extreme efficiency from 16,000 candidates.
A joint research team, comprising Professor Kim Yong-joo of the Department of Materials Science and Engineering at KU (President Kim Dong-One), Professor Jung Yeon-sik of the Department of Materials Science and Engineering at the Korea Advanced Institute of Science and Technology (KAIST), and Professor Oh Min-wook of the Department of Materials Science and Engineering at Hanbat National University, developed high-entropy chalcogenide thermoelectric materials (HECs) with world-class efficiency using active learning technology.
*Active learning: A method in which AI does not learn from an entire dataset but autonomously selects and learns from the data that is most likely to improve relevant performance.
Thermoelectric materials convert temperature differences into electrical energy and represent a key solution for recycling waste heat, which accounts for up to 70% of global energy consumption. However, in the case of high-entropy materials (HEMs) that contain multiple elements blended to enhance performance, identifying the optimal elemental ratio from among a vast number of possible combinations was virtually impossible using conventional methods.
To address this challenge, the research team constructed a Bayesian optimization-based closed-loop experimental framework. Using the average quality factor, a special physics-informed descriptor, as a training indicator, this AI model autonomously proposes the most promising candidate compositions for the next round of experiments.
As a result, the team successfully discovered three new HEMs with a thermoelectric figure of merit higher than 2.0 by synthesizing and testing only 80 samples, corresponding to just 0.5% of the 16,206 total candidate compositions. This represents a dozen-fold improvement in search efficiency compared to conventional approaches to new material development.
*Thermoelectric figure of merit: An indicator of a material’s thermoelectric conversion efficiency. Higher values indicate superior performance, and a value of 2.0 or above is considered commercially feasible.
Professor Kim Yong-joo said, “Our study demonstrates that AI can go beyond simply learning from data; based on physical principles, it can propose optimal elemental combinations that even expert experimentalists might overlook. We have opened a path through which researchers with limited domain expertise can design complex new materials with the help of AI.”
The results of this study were published online on February 17, 2026, in Advanced Materials, a globally renowned international journal in the field of materials science, and were selected as a back cover article in recognition of the study’s excellence.
*Article title: Active Learning-Guided Accelerated Discovery of Ultra-Efficient High-Entropy Thermoelectrics
*DOI: 10.1002/adma.202515054
*URL: https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202515054
This study was supported by the National Research Foundation of Korea and the Korea Institute of Energy Technology Evaluation and Planning.
[Research team photo]
