Machine Learning-Aided 3D-AFM for Identification of Spatial Heterogeneity in Interfacial Solvation Structures

Abstract

Interfacial solvation structures are crucial in wide-ranging applications, including lubrication and electrochemical energy storage. As the sole technique capable of in-situ mapping of three-dimensional (3D) solvation structures, 3D atomic force microscopy (3D-AFM) resolves liquid-solid interfaces with angstrom precision. However, conventional analysis via global averaging or manual selection wastes the massive datasets generated by 3D-AFM, rendering the objective resolution of spatial heterogeneities and 3D features of the interfacial solvation structure unattainable. Here, leveraging the high-throughput capacity of unsupervised machine learning to handle massive datasets, we present a general method integrating 3D-AFM for the unbiased interpretation of interfacial force dataset. Applying this method to the graphite-water interface uncovers a previously obscured spatial heterogeneity, achieving a precise spatial partitioning of the interface into two distinct regimes with divergent mechanical responses. Molecular dynamics simulations attribute this heterogeneity to the nonuniform distribution of trace airborne hydrocarbon contaminants. This work not only elucidates the contaminant-mediated modulation of interfacial water structures, but also provides a general paradigm for identifying chemical non-uniformities at diverse solidliquid interfaces.