From Molecules to Formulations: A Data-Centric Approach to Electrolyte Development for Lithium Metal Batteries

Abstract

The rational design of electrolytes for lithium metal batteries (LMBs) has traditionally relied on trial-and-error experimentation. However, the rapid growth of battery research data and recent advances in computational tools provide an opportunity to transform this paradigm. This thesis aims to integrate data science with fundamental electrolyte science to extract new insights and to accelerate the discovery of new organic liquid electrolytes for LMBs. In the first attempt, the largest experimental Coulombic efficiency (CE) dataset, curated from over 100 scientific articles, is analyzed using advanced data analytics combined with physically informed feature engineering that encodes fluorination degree, carbon chain length, and oxygen content. The resulting insights reveal that higher fluorination degree, shorter carbon chains, and lower oxygen content correlate with improved CE performance. Notably, boiling point emerges as a decisive descriptor when other factors are comparable, with lower-boiling electrolytes associated with higher CE. Guided by these findings, a predictive CE model (R² = 0.85) enables the development of a new electrolyte formulation achieving a CE of 99.6% in Li||Cu cells, ~4.8 V oxidative stability, and 80% capacity retention after 700 cycles in Li||NMC811 cells. The interesting discovery of boiling point has motivated the second work, which challenges the prevailing design philosophy for localized high-concentration electrolytes (LHCEs), mandating that diluents should remain inert toward the Li+ solvation shell, minimizing all cation–diluent interactions. Through Raman spectroscopy, multinuclear NMR, and MD simulations, we show that a moderately coordinating solvent that restructures the Li⁺ solvation environment while preserving the aggregate-rich anion structure can significantly improve transport properties. Therefore, it enables 99.7% CE and stable cycling for over 550 cycles in 4.4 V Li||NMC811 cells with a high-loading cathode (4.2 mAh cm⁻²) at 1C, thereby overcoming the transport limitations inherent to conventional LHCE systems. Together, this thesis demonstrates that the convergence of data science and materials science provides a systematic pathway toward rational electrolyte engineering for next-generation batteries