A new artificial intelligence framework developed at Cornell University can accurately predict the performance of high-conductivity lithium-ion battery electrolytes while uncovering the underlying chemical principles that drive their behavior. Published on February 19 in Nature Computational Science, the framework—called SCAN (a dynamic routing-guided, NAE engineering platform)—offers a powerful new tool for designing next-generation batteries with improved energy storage and efficiency. The research focuses on nonaqueous electrolytes—liquid or gel-like materials that enable higher energy density in lithium-ion batteries. These electrolytes are essential for advanced battery performance, but their chemistry involves a complex interplay of salts, solvents, and operating conditions, making rational design challenging. Traditional AI models treat electrolyte formulations as static collections of input variables, learning statistical correlations between ingredients and performance metrics like conductivity. While effective at prediction, these “black-box” models lack transparency and fail to explain the chemical mechanisms behind their outputs. To overcome this limitation, Fengqi You, the Roxanne E. and Michael J. Zak Professor in Energy Systems Engineering at Cornell’s Duffield College of Engineering, and Zhilong Wang, a postdoctoral researcher and the study’s first author, created a dynamic, interpretable framework. SCAN treats salts, solvents, and operating conditions as distinct but interacting components. It processes chemically meaningful descriptors for each category separately, then adaptively integrates them using a routing mechanism that mimics how different chemical pathways influence ion transport. When tested on a large experimental dataset of lithium-ion electrolytes, SCAN reduced prediction errors by over 65% compared to state-of-the-art machine learning methods. Crucially, it maintained high accuracy across the full range of conductivity values—including rare, high-performance formulations that are most valuable for next-generation batteries. “Interpretability and integration with physical principles are essential for trustworthy, scalable design tools in energy materials,” said You, who is also a senior faculty fellow at the Cornell Atkinson Center for Sustainability. “We’re not just predicting outcomes—we’re revealing the chemistry behind them.” The development is part of the broader Cornell AI4S Initiative, which brings together researchers from diverse disciplines to apply AI to pressing challenges in energy, materials, and sustainability. This work complements a recent study published in Science Advances by You and Wang, which introduced a coordinated framework for AI-driven solid-state battery research that combines machine learning, simulations, and experimental feedback. “These two studies represent both a strategic vision and a practical implementation path,” You said. “Together, they demonstrate how AI can be used not just to predict better materials, but to understand and accelerate innovation in battery science.”