Machine Learning Guided Dopant Selection for Metal Oxide-Based Photoelectrochemical Water Splitting: The Case Study of Fe2O3 and CuO

Doping is an effective strategy for tuning metal oxide-based semiconductors for solar-driven photoelectrochemical (PEC) water splitting. Despite decades of extensive research effort, the dopant selection is still largely dependent on a trial-and-error approach. Machine learning (ML) is promising in providing predictable insights on the dopant selection for high-performing PEC systems because it can uncover correlations from the seemingly ambiguous linkages between vast features of dopants and the PEC performance of doped photoelectrodes. Herein, the authors successfully build ML model to predict the doping effect of 17 metal dopants into hematite (Fe2O3), a prototype photoelectrode material. Their findings disclose the critical parameters from the 10 intrinsic features of each dopant. The model is further experimentally validated by the coherent prediction on Y and La dopants' behaviors. Further interpretation of the ML model suggests that the chemical state is the most significant selection criteria, meanwhile, dopants with higher metal-oxygen bond formation enthalpy and larger ionic radius are favored in improving the charge separation and transfer (CST) in the Fe2O3 photoanodes. The generic feature of this ML guided selection criteria has been further extended to CuO-based photoelectrodes showing improved CST by alkaline metal ions doping.

Automated summary

By using a machine learning model, the authors successfully predicted the doping effect of metal dopants into hematite and identified critical parameters. The model was further validated experimentally and suggested that the chemical state, metal-oxygen bond formation enthalpy, and ionic radius are important factors for improving charge separation and transfer. The ML-guided selection criteria were also extended to copper oxide-based photoelectrodes, resulting in improved charge separation and transfer.