Machine learning-guided electrolyte–interface co-design for Li-based anode-free batteries: Toward a predictive framework for stable metal deposition

Lithium-based anode-free batteries (AFBs) can increase cell-level energy density by eliminating the pre-existing negative electrode and plating cathode-derived lithium onto a bare current collector during charge. However, their lifetime is severely limited by finite lithium inventory loss caused by unstable solid electrolyte interphase formation, dead lithium, electrolyte decomposition, and interfacial contact failure. Existing AFB reviews mainly summarize cell architectures, electrolyte formulations, current-collector modifications, or solid-state designs, while machine-learning-focused battery reviews usually address broad materials discovery, electrolyte screening, or degradation prediction across Li-ion and Li-metal systems. A clear gap remains in linking electrolyte chemistry, interfacial evolution, practical cell constraints, and machine-learning descriptors into one AFB-specific design logic. This review addresses that gap by treating AFBs as a coupled electrolyte–interface–data problem. It distinguishes AFB-specific evidence from broader Li-metal and Li-ion studies and explains why flooded Li||Cu, excess-Li, or generic electrolyte datasets cannot be directly transferred without considering first-cycle lithium loss, E/C ratio, cathode loading, stack pressure, bare-current-collector nucleation, and pouch-cell validation. Finally, an ML-enabled electrolyte–interface co-design framework is proposed, defining descriptors, model choices, uncertainty treatment, validation metrics, and closed-loop optimization routes for stable zero-excess-lithium batteries.