Large Language Model Guided Discovery of Hole Transport Layer Dopants for Efficient and Stable Perovskite...

Achieving high efficiency and long-term stability in n-i-p perovskite solar cells (PSCs) remains constrained by the hole transport layer (HTL) dopant chemistry. The most commonly used dopant for 2,2',7,7'-tetrakis(N,N-di-4-methoxyphenylamino)-9,9'-spirobifluorene (Spiro-OMeTAD), typically based on lithium bis(trifluoromethanesulfonyl)imide, enables state-of-the-art power conversion efficiency (PCE) but often sacrifices thermal and environmental robustness due to hygroscopicity, ionic migration, and reduced glass-transition temperature. Here, a HTL-dopant-focused large language model (LLM) framework is constructed to mine the literature at scale. Using a corpus of over 70 000 publications for retrieval-guided learning, the model identifies trityl tetrakis(pentafluorophenyl) borate (TrTPFB) as an effective p-dopant that improves hole transport in Spiro-OMeTAD, while also improving the morphology and hydrophobicity of the HTL film. With optimized TrTPFB doping concentration, the champion lithium-free Spiro-OMeTAD based device reaches a PCE of 24.13%, and retains 92.67% and 85.82% of its initial PCE after 900 h thermal aging at 65°C with 30% RH and at 85°C in N₂, respectively. This study shows how LLM can turn scattered literature into useful experimental guidance for exploring efficient, stable perovskite photovoltaics.