Machine Learning-Aided Band Gap Engineering of BaZrS Chalcogenide Perovskite.

The non-toxic and stable chalcogenide perovskite BaZrS fulfills many key optoelectronic properties for a high-efficiency photovoltaic material. It has been shown to possess a direct band gap with a large absorption coefficient and good carrier mobility values. With a reported band gap of 1.7-1.8 eV, BaZrS is a good candidate for tandem solar cell materials; however, its band gap is significantly larger than the optimal value for a high-efficiency single-junction solar cell (∼1.3 eV, Shockley-Queisser limit)─thus doping is required to lower the band gap. By combining first-principles calculations and machine learning algorithms, we are able to identify and predict the best dopants for the BaZrS perovskites for potential future photovoltaic devices with a band gap within the Shockley-Queisser limit. It is found that the Ca dopant at the Ba site or Ti dopant at the Zr site is the best candidate dopant. Based on this information, we report for the first time partial doping at the Ba site in BaZrS with Ca (i.e., BaCaZrS) and compare its photoluminescence with Ti-doped perovskites [i.e., Ba(ZrTi)S]. Synthesized (Ba,Ca)ZrS perovskites show a reduction in the band gap from ∼1.75 to ∼1.26 eV with <2 atom % Ca doping. Our results indicate that for the purpose of band gap tuning for photovoltaic applications, Ca-doping at the Ba-site is superior to Ti-doping at the Zr-site reported previously.