Machine learning study of surface and bulk compositional features governing Ge-incorporated kesterite solar cell performance
Year: 2025
Type: article
Abstract: Ge-incorporated kesterite (Cu2Zn(Sn1−xGex)Se4, CZTGSe) is considered an alternative to Cu(In,Ga)Se2; however, its conversion efficiency remains limited by significant open-circuit voltage (VOC) losses. One possible reason for this loss is the presence of internal defects in the kesterite, which are closely related to the composition ratio of its thin film. In this study, a machine learning (ML)-based analysis is performed to investigate the effects of elemental composition ratios in kesterite devices. A comprehensive database is compiled by integrating composition analysis results with device parameters obtained from over 6,000 self-fabricated CZTGSe thin films and their devices. Particularly, the composition ratio is distinguished between the surface and bulk to clarify their respective roles in governing device performance. To ensure reliable analysis, various ML models, including Random Forest, Gradient Boosting, LightGBM, and XGBoost, are employed to evaluate feature importance and demonstrated consistent predictive capability. The results reveal that the Ge/IV (IV = Sn+Ge) ratio at the surface is the most influential feature, decisively affecting the overall conversion efficiency, VOC, and short-circuit current. Its dominant role is associated with forming an adaptable conduction-band offset and suppressing Sn-related defect states at the buffer/CZTGSe interface. These results indicate that surface engineering, especially Ge/IV, is a key factor in improving the performance of kesterite solar cells.
Source: Materials Today Communications
Authors: Sanghun Lee, Shinho Kim, Hitoshi Tampo, Takehiko Nagai, Yangdo Kim
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,830
Award: G3434209833
OpenAlex ID: W4416940286