Extracting the parameters of two-energy-level defects in silicon wafers using machine learning models

This study introduces a pioneering machine learning (ML)-based methodology to characterise two-level defects in the bulk of silicon wafers. Bulk defects have a critical impact on the efficiency of silicon solar cells. By identifying the specific parameters of these defects, namely, their energy levels and capture cross-sections, researchers can devise strategies to mitigate their effects. It is often assumed that bulk defects are single-level defects following the Shockley-Read-Hall recombination statistics. However, two-level defects or even multi-level defects are common as well. At present, it is challenging to distinguish between single-level defects and two-level defects, and to extract the parameters of a two-level defect. This study proposes an ML-based approach to distinguish between one- and two-level defects based on temperature- and injection-dependent lifetime spectroscopy with an accuracy above 90 %. Furthermore, if the defect is identified as a two-level defect, this study presents another ML method to extract its defect parameters, with a correlation coefficient above 0.9 for the energy levels. • TIDLS reveals critical insights into defects that limit performance in photovoltaic (PV) devices. • Two-level defects are common in silicon (Si) wafers, but traditional methods often treat them as single-level defects • We propose using ML to classify single- and two-level defects and extract two-level defect parameters from TIDLS measurements. • The ML models were validated against both simulations and experimental data.