AI Screens 150 Million Compositions to Find New Lead-Free Electronics Materials
Seoul National University researchers used AI to search roughly 150 million possible chemical compositions and identify two new lead-free dielectric materials that stay stable at high temperatures.
Step by step
- 1
Mine text, tables, graphs from 448 papers
- 2
Build a 1,202-record dataset
- 3
Screen 150 million compositions with AI
- 4
Narrow to 37, synthesize and test two
Researchers at Seoul National University have used artificial intelligence to search through roughly 150 million possible chemical compositions and identify new lead-free materials for electronics that remain stable at high temperatures. The team, from the university's College of Engineering, was led by Professor Ho Won Jang of the Department of Materials Science and Engineering, with integrated M.S./Ph.D. student Kwanwoo Song as first author. The findings were published in Nature Communications.
The researchers were searching for dielectrics β insulating materials that block the direct flow of electricity while storing electrical charge. Dielectrics are essential to multilayer ceramic capacitors (MLCCs), components found in smartphones, electric vehicles and other electronics. A higher constant lets a component of the same size store more electrical energy, but a useful material must also keep that performance as temperatures rise.
To find such materials, the team combined multimodal literature mining β extracting information from text, tables and graphs in published papers β with physics-informed machine learning. Large language models pulled compositions and processing conditions from 448 scientific papers, while graphs were converted into numerical data to recover how each material's dielectric properties changed with temperature. Altogether, the process produced 1,202 records, to which the researchers added 22 physical descriptors covering elemental composition and microstructure so that data from different papers could be compared directly.
The team then combined 30 independently trained machine learning models to predict three key measures of dielectric performance at once, using how closely the models agreed with each other to prioritize compositions with greater predictive confidence. That process narrowed the roughly 150 million possible compositions down to 37 candidates. Two of these were synthesized and tested in the laboratory, and both showed high dielectric constants together with strong stability at elevated temperatures.
The two new materials, both variants of a lead-free β a class of material whose electrical response changes gradually with temperature β containing tin, were compared against barium titanate, a widely used dielectric whose dielectric constant changes sharply around 125 degrees Celsius. The new materials maintained relatively stable dielectric constants across a broad temperature range.
Terms explained
The story so far
- Open ASR Leaderboard Adds Its First Indian-Language Speech Recognition Test
- New Algorithm Designs Collapsible Lattices Beyond Origami and Kirigami
- Scientists Turn Hard-to-Recycle PVC Plastic Into High-Performance Engine Lubricant
- Scientists Solve Decades-Old Mystery of Why Plastic Pipe Welds Are So Strong
- University of Tokyo Turns Food Scraps Into a Concrete-Beating Building Material
- Diamond Can Generate Electricity When Bent, Hong Kong Study Finds
- AI Screens 150 Million Compositions to Find New Lead-Free Electronics Materials
