- Professor Lee Tae-hoon's team selected for Nvidia's global academic grant program

- High-performance GPUs to power autonomous AI search of catalyst structures

Lee Tae-hoon, a professor in the Department of Energy Engineering at DGIST. [DGIST]
Lee Tae-hoon, a professor in the Department of Energy Engineering at DGIST. [DGIST]

The Daegu Gyeongbuk Institute of Science and Technology announced Wednesday that a research team led by Energy Engineering Professor Lee Tae-hoon has been selected for Nvidia's Academic Grant Program, a global initiative supporting innovative research at universities and institutions worldwide.

The program provides high-performance GPUs and computing resources to selected research projects at universities and research institutions around the world.

Through the grant, Lee's team will receive Nvidia's RTX PRO 6000 Server Edition GPUs to develop an AI-driven autonomous search system for catalyst structures.

Catalysts are key materials that accelerate chemical reactions. Thermal catalysts in particular undergo constant changes in surface atomic arrangement and composition depending on reaction conditions such as temperature and pressure. Improving catalyst performance requires a precise understanding of the atomic structures that emerge during actual reaction processes.

The sheer number of possible atomic arrangements and combinations, however, has made analysis through conventional quantum mechanics-based calculations enormously time-consuming and computationally expensive.

To address this, the team is developing what it calls "agentic AI" technology that combines machine learning with a large language model.

At the core of the system is an autonomous search framework in which AI independently proposes, calculates and evaluates catalyst structures. The team will use a machine learning interatomic potential trained on quantum mechanics calculation data to accelerate computation, while also accounting for the adsorption and desorption of molecules on catalyst surfaces to identify optimal atomic structures and compositions.

An LLM-based AI agent will be integrated into this framework. Once a researcher inputs the catalyst composition and experimental parameters, the AI generates candidate structures, calculates their performance and autonomously determines the direction of the next search based on the results.

The high-performance GPUs provided by Nvidia will be used to train machine learning models and conduct large-scale catalyst structure searches. The team plans to use the system to predict catalyst structures most likely to appear under real chemical reaction conditions, and to apply those predictions to experimental validation and new material design.

The team expects the technology to serve as a foundation for shifting material development away from its traditional reliance on researchers' experience and repetitive calculations, toward an AI-centered autonomous research framework.

"Understanding how a catalyst operates structurally under actual reaction conditions is the starting point for catalyst design," Lee said. "We will build an autonomous search system in which AI independently proposes and validates catalyst structures, and work with experimental researchers to advance it into a tool that can be applied to real catalyst design."

Meanwhile, Lee has built a body of research using quantum mechanics-based precision calculations and machine learning molecular dynamics to investigate defects and surface and interface phenomena in semiconductor and energy materials at the atomic level. More recently, Lee has focused on next-generation material discovery and design research using generative AI and LLM-based AI agents.


nbgkoo@heraldcorp.com