Accelerated Multi‐Property Screening of Lead‐Free Halide Double Perovskite via Transfer Learning

Abstract

As a promising third‐generation photovoltaic technology, perovskite solar cells have attracted much attention due to their high photoelectric conversion efficiency and low manufacturing cost. However, perovskite solar cells still face problems such as poor stability and lead toxicity, and it is both time‐consuming and expensive to find new materials with properties that meet the demand through traditional trial‐and‐error methods. To address this problem, a multi‐property screening method is proposed for lead‐free halide double perovskite based on a transfer learning technique. First, a source domain model with the formation energy of halide double perovskites as the target property is established, and high‐precision predictive models of Ehull, band gap, bulk modulus, and shear modulus are constructed by the transfer learning technique. In particular, the “continuous transfer” method is proposed. The bulk modulus model, after transfer learning, is used as the source domain model to transfer the shear modulus model again. Finally, the high‐throughput screening of multi‐properties of halide double perovskites are successfully realized, and computationally verified that the Cs2CuIrF6 material has good stability, a suitable bandgap (1.06 eV), and ductility (G/B = 0.27). This proposed transfer learning strategy provides an effective method for screening stable perovskite materials with potential for multiple optoelectronic applications.