Insight into Charge Transport Behaviors in Hematite Photoanodes via Potential-Dependent Impedance Spectroscopy and Machine Learning

We employed machine learning (ML) techniques combined with potential-dependent photoelectrochemical impedance spectroscopy (pot-PEIS) to gain deeper insights into the charge transport mechanisms of hematite (α-Fe₂O₃) photoanodes. By the Shapley Additive exPlanations (SHAP) analysis from the ML model constructed from a small data set (dozens of samples) of electrical parameters obtained from pot-PEIS and the PEC performance, we identified the dominant factors influencing the electron transport to the back contact in the bulk and hole transfer to a solution at the hematite/electrolyte interface. The results revealed that shallow defect states significantly enhance electron transport, while deep defect states impede it, and also one of the surface states enhances the hole transfer to the electrolyte solution. The incorporation of a titanium dioxide underlayer had an effect to increase shallow defect states to promote electron transport and induce a new surface state to promote hole transfer. Cobalt phosphate cocatalyst improved charge separation, leading to enhance electron transport in the bulk, and the cocatalyst surface state promotes the hole transfer at the interface. The understanding of the band diagram based on the selected dominant descriptors provided critical insights into the electronic structure, elucidating the role of defect states and surface states in influencing the overall photoanode performance. This study showcases the potential of combining ML and pot-PEIS for material optimization.