[Other] AI-guided discovery of self-assembled monolayers combining synthetic accessibility and interfacial stability for perovskite solar cells Copy

Navigating the vast chemical space to discover self-assembled monolayers that combine synthesizability, interfacial compatibility and long-term device stability remains a large challenge in perovskite photovoltaics. Here we present an artificial intelligence-guided design framework for self-assembled monolayers that integrates graph neural network-based retrosynthetic predictions with high-throughput screening based on density functional theory and expert chemical knowledge. Over 90,000 structures from 308 fragments were narrowed down to 300 candidates, ranked using a composite score of eight descriptors extending beyond intrinsic electronic properties to include synthesizability, hole reorganization energy λh and structural features (heteroatom content and aromaticity). The top candidate, AI-1, is readily synthesizable and incorporates favourable heteroatom functionalities. AI-1 achieves optimal energetic alignment and forms robust S–Pb coordination bonds at the buried interface, effectively suppressing non-radiative recombination and enhancing interfacial stability. Devices incorporating AI-1 achieved a power-conversion efficiency of 27.2% and retained >95% of their initial efficiency after 2,000 h of maximum-power-point tracking at 65 °C. Mechanistic studies show that the multifunctional methylthio side chain is vital, as it reinforces the interfacial bonding and balances passivation with chemical robustness. These findings demonstrate that artificial intelligence-guided design can systematically deliver self-assembled monolayers uniting high efficiency with long-term operational stability, and they provide general principles for interface engineering.