标题/Title：

Dipole-Moment-Knowledge-Guided Molecular Design for Perovskite Surface Passivation: A Gemma-Language-Model and DFT-Driven Framework

影响因子/Impact Factor：2.9

链接/Link： https://doi.org/10.1002/adts.202501318

摘要/Abstract：

One of the key challenges in the large-scale application of perovskite solar cells is stability. Researchers have found that passivation molecules play a crucial role in mitigating interface defects, thereby enhancing stability. Traditionally, the design of passivation molecules has relied on the expertise of chemists and materials scientists. In this study, we introduce a novel approach driven by a language model and dipole-moment-knowledge-based strategy for passivation molecule design. Specifically, we employ the open-source Gemma model, which is pre-trained and fine-tuned on the PubChem and QM9 datasets. This fine-tuning enables Gemma to generate passivation molecules with higher dipole moments. Further density functional theory (DFT) validation reveals that molecules designed by Gemma improve the stability of perovskite structures with surface defects by approximately 27.75%. Additionally, electronic density of states and charge distribution analysis further support these findings. This study highlights the potential of language models in the design of next-generation photovoltaic device materials, particularly in passivation molecule development.