Shayan Lal
Research Mentor: Jacob Sellers
Mentor Department: Psychology, LSA
Author(s): Not Available
Session: Session 3 (11:00 AM – 11:50 AM)
Presentation Type: Poster 135
Abstract
Semiconductor electrochemistry is potentially useful for using solar energy to drive chemical reactions such as hydrogen fuel production and CO2 conversion. Currently, photoelectrochemical systems are inefficient and not cost effective. Designing more efficient systems requires analytical tools to better understand how changing aspects of these systems change the energetics at the surface of the semiconductor. Accurate estimation of experimental parameters is essential for interpreting charge transfer processes at semiconductor–solution interfaces. In this project, we developed a computational framework that combines physics-based simulation and machine learning to extract quantitative information from cyclic voltammetry measurements taken on semiconductor electrodes with freely diffusing redox molecules. Specifically, the model targets five parameters: reduction potential relative to band edge, surface oxide thickness, reorganization energy, surface state density and surface state energy. A MATLAB simulation tool was used to generate simulated electrochemical responses and train a Python-based machine learning model for inverse parameter prediction, which was compared with experimental measurements. During the first phase of the project, we reformatted and rescaled simulation outputs to improve alignment with experimental data and optimized data formatting for model training. We also streamlined and verified the MATLAB code to improve readability, performance, and the statistical sampling of input parameters. In the current phase, we are generating large training datasets and refining the learning pipeline to improve predictive accuracy, with the goal of increasing the coefficient of determination (R²) for parameter estimation. This work supports the long-term objective of enabling automated, data-driven interpretation of electrochemical measurements and demonstrates the potential of machine learning to accelerate analysis in semiconductor electrochemistry.
