Machine Learning Drives a Path to Defect Engineering for Suppressing Nonradiative Recombination Losses in Cu2ZnSn(S,Se)4 Solar Cells

Abstract

Recent kesterite developments encouraged researchers to use Cu₂ZnSn(S,Se)₄ (CZTSSe)-based photoabsorber materials in diverse optoelectronic applications. However, the detrimental bulk and interface defects induced high carrier recombination at corresponding regions, stagnating further improvement in the performance of kesterite solar cells. In this work, a machine learning (ML)-guided strategy is employed to optimize the amount of germanium (Ge) incorporation to enhance the baseline performance of CZTSSe. The ML model indicates that incorporation of <5% Ge concentration is optimal for achieving higher device performance and reducing open-circuit voltage (V_OC) loss. It also revealed that engineering defects controlling carrier density plays a crucial role in achieving high-quality devices. Building upon the optimized Ge incorporation, silver (Ag) is subsequently introduced to further passivate shallow-level copper (Cu)-related defects. Experimental validation confirms that Ge incorporation effectively improves the device performance by suppressing deep-level defects in both the space-charge region and the quasi-neutral region. As a result, an improved carrier separation process, minority carrier lifetime, and reduced nonradiative carrier recombination losses increased device performance by more than 20%. Finally, the champion device with double-cation incorporation of Ag and Ge in the vacuum-proceeded CZTSSe absorber layer delivers enhanced device performance from 9.11 to 11.32%.

Keywords: CZTSSe; incorporation; kesterite; machine learning; solar cell.