Data‐Driven Design of Self‐Assembled Monolayers for High‐Efficiency Perovskite Solar Cells

ABSTRACT Self‐assembled monolayers (SAMs) are pivotal for boosting the performance of perovskite solar cells (PSCs). Yet, the intricate link between molecular structure and device efficiency hinders rational SAM design. Here, we introduce a data‐driven strategy that leverages a curated dataset of reported SAMs and their PSC efficiencies, with molecular structures encoded into three distinct segments: anchor group–linker–head group. Based on this fragment‐encoding framework, our strategy focuses on the recombination of fragment units, rather than unconstrained de novo molecular design. Using ensemble learning and SHapley Additive exPlanations (SHAP) interpretability within a cross‐validated framework, we pinpointed the head group as the dominant performance driver. This insight guided the construction of an expanded molecular library by recombining high‐value fragments identified from the curated database. Virtual screening of this library then yielded a synthetically accessible SAM molecule with top‐predicted efficiency, namely S1. Experimental validation revealed that S1 forms a compact, ordered monolayer on NiO x , featuring a well‐aligned HOMO level and a strong interfacial dipole that optimizes electronic coupling. Consequently, S1 enables defect passivation and hole extraction, delivering a champion power conversion efficiency of 26.21%. This study establishes a machine learning paradigm, integrating fragment‐based encoding and explainable AI for data‐driven interface optimization in high‐performance PSCs.