AI-Driven Literature Mining Accelerates Discovery of Heat-Stable Lead-Free Dielectrics

Researchers at Seoul National University have demonstrated a transformative approach to materials science by combining artificial intelligence with literature mining to discover new lead-free dielectric materials capable of maintaining high performance at elevated temperatures. Led by Professor Ho Won Jang of the Department of Materials Science and Engineering, with first author Kwanwoo Song, the team published their findings in Nature Communications, outlining a framework that shifts material development from trial-and-error experimentation to a predictive, data-driven methodology. Dielectric materials are essential components in multilayer ceramic capacitors, which are ubiquitous in consumer electronics, electric vehicles, and aerospace systems. While high dielectric constants enable greater energy storage in compact sizes, practical applications require these properties to remain stable across wide temperature ranges. Traditional lead-based materials often fail under these conditions, prompting a push toward lead-free alternatives. However, the vast combinatorial space of potential compositions, coupled with fragmented data across scientific publications, has historically hindered rapid discovery. To overcome these barriers, the research team developed a machine learning pipeline that integrates multimodal literature mining with physics-informed modeling. The system automatically extracted composition, processing parameters, and temperature-dependent dielectric properties from 1,202 records spanning 448 research papers. By incorporating 22 physical descriptors and deploying an ensemble of 30 predictive models, the framework evaluated approximately 150 million virtual compositions. Rigorous screening against performance targets and physicochemical constraints narrowed the candidates to 37. Two lead-free formulations, designated SNBTS1 and SNBTS2, were synthesized for experimental validation. Both materials, doped with 1 mol% and 2 mol% tin respectively, achieved room-temperature dielectric constants of 3,422 and 3,307 while maintaining exceptional thermal stability. Experimental testing confirmed that the tin-substituted compounds significantly outperformed conventional barium titanate, particularly in high-temperature environments. The subtle addition of tin expands the crystal lattice and enhances atomic-scale electrical heterogeneity, creating a favorable balance between energy storage capacity and temperature resilience. Both formulations successfully met stringent international standards for multilayer ceramic capacitors, including X5R, X6R, and X7R classifications. The study establishes a scalable blueprint for accelerating materials discovery across functional oxides and thin-film technologies. By systematically aggregating fragmented scientific data and validating predictions through experimental synthesis, the research team has demonstrated how AI can bridge the gap between theoretical screening and practical engineering. Professor Jang and his colleagues anticipate this methodology will streamline the development of next-generation electronic components, with immediate applications in electric powertrains, high-reliability power electronics, and aerospace systems. The findings underscore a growing paradigm shift in materials science, where computational literature synthesis and physics-guided machine learning collectively redefine the pace of technological innovation.