2:00 PM - DS04.05.03
WITHDRAWN 5/8/22 DS04.05.03 Identifying Materials Selection Criteria for 2D Capping Layer in Perovskite Solar Cells via Machine Learning
Zhe Liu1,Suo Wang1,Chongyang Zhi1,Zhen Li1
Northwestern Polytechnical University1
Two-dimensional capping layers have shown a great passivation effect on perovskite absorbers, which could improve both power conversion efficiency and humidity stability of solar cells. Although many organic halides have been tried by experimental trial and error approach, the best selection criteria of the organic halides are still not very clear to achieve the good passivating effect. In addition, there are still hundreds of unexplored organic halides existing in the databases. The current trial-and-error is not efficient enough to find the best materials. Machine learning model has been shown as an effective tool to identify the most important descriptors of 2D perovskite materials  using Shapley Additive Explanation (SHAP) . But the previous work  did not investigate the impact of the passivation improvement on the power conversion efficiency (PCE). In this work, we adopted the data-driven machine-learning strategy of analysis and aim to establish the most important chemical properties that affect the device efficiency. Particularly, we demonstrate the application of these derived selection criteria to find suitable organize iodides to passivate MA1-x-yFAyCsxPbI3 interface. First, we utilize an initial selection of 17 organic halides – 12 historical data and 5 Latin hypercube sampling (LHS) data – to build a machine learning regression model. The five LHS data are newly added to ensure the balance of the dataset, and therefore ensure the generalizability of the predictive regression model. This regression model correlates the 13 specific chemical properties to device PCEs, which allows us to identify the most important chemical properties by SHAP analysis. The SHAP analysis gives us the specific value range for the most important chemical properties, and we then used these specific criteria to select 3 new materials for the remaining 20 organic halides that have been tried before in literature. Furthermore, we use the chemical property criteria to down-select 50 suitable candidates out of 500 available organic iodides in PubChem databases. Finally, we validated this result experimentally with 3 materials that could achieve device PCE of > 21%.
 Hartono, N.T.P., Thapa, J., Tiihonen, A., Oviedo, F., Batali, C., Yoo, J.J., Liu, Z., Li, R., Marrón, D.F., Bawendi, M.G., et al. (2020). How machine learning can help select capping layers to suppress perovskite degradation. Nature Communications, 2020, 11, 4172.
 Lundberg, S.M., Erion, G., Chen, H., DeGrave, A., Prutkin, J.M., Nair, B., Katz, R., Himmelfarb, J., Bansal, N., and Lee, S.-I. (2020). From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2020, 2, 56–67.