Stabilizing Anion-Derived Interphase by Machine-Learning-Accelerated Screening of Out-of-Shell Co-Solvents for Aqueous Zinc Batteries

Abstract

Aqueous zinc batteries (AZBs) lack a stable anion-derived solid electrolyte interphase (SEI) on the Zn anode, resulting in severe competition between Zn deposition and the hydrogen evolution reaction (HER). A conventional in-shell co-solvent coordinates strongly with Zn2+, displacing coordinated water and weakening Zn2+–anion interactions. This introduces a critical trade-off between HER suppression and anion-derived SEI formation. Here, we propose an out-of-shell co-solvent strategy that weakens Zn2+–H2O interactions, thereby enhancing Zn2+–anion interactions. To screen an optimal candidate, machine learning molecular dynamics (MLMD) was employed, achieving a ∼104-fold acceleration over ab initio molecular dynamics (AIMD) without sacrificing accuracy, and identifying N,N-dimethylacetamide (DMAC) from 28 candidates. In situ spectroscopic characterization further reveals that DMAC reconstructs the solvation environment, which facilitates desolvation and mitigates the formation of the inherently anion-lean interface. Consequently, this strategy promotes anion-derived SEI formation, synergistically suppressing HER. The DMAC electrolyte exhibits high Coulombic efficiency in Zn∥Cu cells (99.3% over 950 cycles) and long-term stability in Zn∥I2 full cells (12,000 cycles). Beyond demonstrating a rational electrolyte design, this work illustrates that MD simulations reform the traditional closed loop from material regulation to performance feedback, while ML integration accelerates screening. For bulk-interfacial solvation structure discrepancies, a feedback loop founded on dynamic interfacial processes regulates MLMD parameters, enabling more precise performance regulation.