EXTENDED ABSTRACT: Perovskite solar cells, with their exceptional optoelectronic properties and potential for lowcost manufacturing, have become a hot topic in photovoltaics. Currently, two device structures of perovskite solar cells, i.e., n-i-p (regular) and p-i-n (inverted) types, are still investigated at the same time. In the past, it was found that the power conversion efficiency (PCE) of p-i-n perovskite devices has lagged significantly behind that of n-i-p devices, largely due to interfacial defect issues between the transparent metal oxide electrode/charge transport layer and the perovskite absorber layer on glass substrates. To address these interfacial defects, organic molecular layers are considered a processable and costeffective passivation material that can synergistically enhance the open-circuit voltage and fill factor of photovoltaic devices, thereby substantially boosting the efficiency of p-i-n perovskite cells. Machine-learning-based generative models present an innovative, intelligent approach for the design and screening of passivating molecules. This presentation focuses on the design and screening of passivation molecules, exploring the integration of molecular generative models and screening workflows guided by theoretical calculations. Three approaches for the design and screening of perovskite passivating molecules will be introduced: (1) molecule database screening based on structural and property features, (2) molecule generation model design based on structural similarity, and (3) generative large language model design constrained by structural characteristics. By employing these machine learning techniques, the presentation will discuss how to streamline the material design and screening process effectively, ultimately enhancing the power conversion efficiency of perovskite solar cells and clarifying the impact of molecular structure on properties and device efficiency.
Keywords: perovskite solar cells; functional molecule design;generative large language model