Assisting the Design of Chiral Metal Halide Semiconductors with Machine Learning

Chiral metal halide semiconductor (CMHS) can directly resolve circularly polarized light and has high carrier mobility and strong defect tolerance, which is highly promising for applications in quantum communication, asymmetric catalysis, medical diagnosis, and so on. However, the tortuous and costly experimental process of designing high-performance CMHS is detrimental to the application of CMHS. Therefore, there is a need to find efficient methods for the design of CMHS.<br/>Here, we built the map between the structure, composition, and chiral intensity of CMHS efficiently using machine learning. Then, we performed SHAP analysis on the established model and calculate the chirality intensity of 1080 potential chiral CMHSs through high-throughput screening. By combining the result of SHAP analysis and prediction, we infer that 1NEA⁺ in A-site, Cu²⁺ in B-site, and Cl^- in X-site are positively related to the chirality intensity of CMHSs. Furthermore, we successfully synthesized unreported material, such as 1NEA₂PbCl₄ with a g_CD of 0.00805, which further validates the credibility of our predictive model. Our work provides an in-depth understanding of combining machine learning and experimentation and helps accelerate the development of CMHS.