High-Throughput Calculation and Machine Learning for Design of Two-Dimensional Hybrid Perovskites

Two-dimensional (2D) hybrid perovskites have garnered significant attention due to their unique structural and electronic properties, making them promising candidates for applications in photovoltaics, light-emitting diodes, and photodetectors. Understanding the structure-property relationships within this vast composition space is crucial for designing novel materials. Given the extensive design space, high-throughput calculations and machine learning are particularly suitable for investigating design rules and discovering new compositions for target applications.<br/>In this study, we employed a fingerprint vector composed of organic descriptors in numerical form to characterize the organic spacers in 2D perovskites. This unique set of fingerprints enables accurate prediction of various properties of 2D perovskites, including their energetics and synthesis feasibility. To derive design rules for the energetics of 2D perovskites, we generated large datasets through high-throughput calculations, capturing structural and electronic property information for a diverse array of 2D perovskite materials. These datasets were then used to train and validate machine learning models, establishing accurate structure-property relationships. To assess the synthesis feasibility of the predicted 2D perovskite structures, we compiled synthesis information from the current literature, including existing experimental structures and simulation studies. We developed a new descriptor for synthesis feasibility based on the organic fingerprint vector, enhancing our predictive capabilities. Leveraging the physical insights obtained from exhaustive enumeration of organic molecules within the defined chemical space, we conducted a targeted search to identify 2D perovskites with desired energetics and high synthesis feasibility.<br/>In conclusion, our work presents a robust computational framework for the energetic design of 2D hybrid perovskites, offering significant insights and tools for the research community. The integration of high-throughput DFT calculations and machine learning not only accelerates the discovery process but also deepens our understanding of the factors influencing the energetics of these materials. We anticipate that our approach will have broad implications for the design of advanced materials across various applications.