Interface engineering is proved to be the most important strategy to push the device performance of the perovskite solar cell(PSC) to its limit, and numerous works have been conducted to screen efficient materials. He...Interface engineering is proved to be the most important strategy to push the device performance of the perovskite solar cell(PSC) to its limit, and numerous works have been conducted to screen efficient materials. Here, on the basis of the previous studies, we employ machine learning to map the relationship between the interface material and the device performance, leading to intelligently screening interface materials towards minimizing voltage losses in p-i-n type PSCs. To enhance the explainability of the machine learning models, molecular descriptors are used to represent the materials. Furthermore,experimental analysis with different characterization methods and device simulation based on the drift-diffusion physical model are conducted to get physical insights and validate the machine learning models. Accordingly, 3-thiophene ethylamine hydrochloride(Th EACl) is screened as an example, which enables remarkable improvements in VOCand PCE of the PSCs. Our work reveals the critical role of datadriven analysis in the high throughput screening of interface materials, which will significantly accelerate the exploration of new materials for high-efficiency PSCs.展开更多
基金supported by the National Natural Science Foundation of China (62075006)the National Key R&D Program of China (2018YFB1500200)。
文摘Interface engineering is proved to be the most important strategy to push the device performance of the perovskite solar cell(PSC) to its limit, and numerous works have been conducted to screen efficient materials. Here, on the basis of the previous studies, we employ machine learning to map the relationship between the interface material and the device performance, leading to intelligently screening interface materials towards minimizing voltage losses in p-i-n type PSCs. To enhance the explainability of the machine learning models, molecular descriptors are used to represent the materials. Furthermore,experimental analysis with different characterization methods and device simulation based on the drift-diffusion physical model are conducted to get physical insights and validate the machine learning models. Accordingly, 3-thiophene ethylamine hydrochloride(Th EACl) is screened as an example, which enables remarkable improvements in VOCand PCE of the PSCs. Our work reveals the critical role of datadriven analysis in the high throughput screening of interface materials, which will significantly accelerate the exploration of new materials for high-efficiency PSCs.
