KAIST-KRICT Train Self-Developed AI Model 100,000 Data Points to Be Accumulated by 2028 Minute Process Differences Determine Performance 'AI Modeling' Essential for High Quality and Uniformity Robotic Arms to Be Introduced Later to Enhance Reproducibility
With global competition intensifying to seize leadership in the perovskite solar cell market, the outcome is expected to hinge on the ability to leverage artificial intelligence (AI)-based experimental data. Industry observers say "AI modeling" is essential to improve the quality and uniformity of perovskite solar cells, whose performance is determined by minute differences at each stage, from material development to complex processing procedures.
According to the Korea Advanced Institute of Science and Technology (KAIST) on Wednesday, the research project "Building a Photoenergy Conversion Chemical Materials Hub and Developing AI-Applied Materials-Device Technology (2024-2028)," led by Professor Seo Jang-won, has entered its midpoint this year. The project centers on accumulating tens of thousands of perovskite solar cell experimental data points on an online platform (Chem-DX, Solar Cell) together with the Korea Research Institute of Chemical Technology (KRICT), and training a self-developed AI model on the data.
The AI prediction model, trained on experimental data accumulated in real time, comprehensively considers the characteristics of key factors at each process stage and recommends the material combinations and processing conditions expected to yield the best performance. This enables the development of original materials suitable for the commercialization of future mobility such as electric vehicles and drones, while shortening experimental periods and improving both the efficiency and long-term stability of the cells.
"Compared to the past, when research was driven mainly by individual researchers' intuition, the efficiency of materials research has significantly improved recently," Professor Seo said. "Overseas, there have been cases where AI shortened the material exploration period, which typically took more than three months, to just one week." He added, "We have built a feedback loop in which experiments are conducted using AI-suggested candidate materials and the results are fed back into the system, allowing the AI model's performance to continuously improve."
Since the launch of the AI-based materials hub project, researchers have accumulated a total of 79,070 experimental data points, with the figure expected to exceed 100,000 by 2028. The medium- to long-term goal is to automate not only data analysis but also the collection process itself through physical AI. For example, robotic arms would be used to reproduce thousands of protective coating applications without error, thereby improving the accuracy of experimental data.
Such research is particularly important for narrowing the technology gap with China, which is backed by massive capital. Over the past one to two years, major Chinese power producers including Huaneng Group have successively built perovskite demonstration plants and have been securing real-world usage data. "While Korea is predicting outdoor adaptability by applying 'accelerated conditions' inside laboratories, China is already accumulating 'outdoor data,'" Professor Seo said. "If the gap in demonstration data becomes prolonged, it will eventually translate into a gap in solar market share."
AI-translated from Korean. Quotes from foreign sources are based on Korean-language reports and may not reflect exact original wording.
Seoul Economic Daily (Commentary)
Choi Hyung-wook (Commentary)
Sekyung IN (Commentary)
SedailyIN (Commentary)
Sekyung IN (Commentary)
Copyright ⓒ Sedaily, All right reserved

source