Machine-Learning-Aided Spin-Splitting Analysis and Prediction in 2D Perovskites

Metal halide perovskites are a type of promising optoelectronic materials and are potential hosts of Rashba/Dresselhaus spin-splitting for spin-selective transport and spin-orbitronics. Through a broad search and simulation of chiral and achiral two-dimensional hybrid organic-inorganic perovskites (2D HOIPs), our previous first-principles study addressed the key structural descriptor controlling the spin-splitting in perovskite systems. However, the potential common features within the existing 2D HOIPs hinder our further investigation towards a more universal conclusion. In this work, we conduct high-throughput simulations for randomly generated 2D HOIP structures, utilize principal component analysis (PCA) to systematically quantify the influence of different structural features, and construct a hierarchical neural-network-based model to account for the Rashba/Dresselhaus spin-splitting magnitude in 2D perovskite systems at different accuracy levels.