Abstract
The advancement of ultrahigh-voltage lithium metal batteries (LMBs) is hindered by the lack of efficient guidelines for electrolyte material development, with current approaches still largely relying on inefficient trial-and-error methods. Herein, we report a high-performance additive design and screening strategy. A library of phosphorus-centered candidates (containing 15,200 molecules) is constructed by simultaneously introducing electron-withdrawing and electron-donating groups to induce intramolecular charge redistribution and tune the electrochemical properties of the molecules. Through unsupervised machine learning–assisted screening, lithium diperfluoropinacolatophosphate (LiDPFP) is identified as the optimal candidate. LiDPFP features localized electron-deficient and electron-rich regions, which tune interfacial reactivity and enhance lithium ion (Li
+
) coordination to promote contact ion pairs. This unique structure leads to the formation of a phosphorus-oxygen polymer-rich cathode electrolyte interphase and a Li
2
O (lithium oxide)–rich solid electrolyte interphase, effectively suppressing cathode degradation and lithium dendrite growth. Consequently, the as-assembled NCM811||Li cells exhibit stable cycling at 4.8 volts over a wide temperature range from −40° to 80°C. A practical 4.7-volt pouch cell with the energy density of 628 Wh kg
−1
(watt-hours per kilogram) exhibits an 80% capacity retention rate after 112 cycles. This work combines machine learning with molecular design, exhibiting a paradigm for practical ultrahigh-energy-density LMBs.