Utilizing machine learning algorithm in predicting the power conversion efficiency limit of a monolithically perovskites/silicon tandem structure

Abstract

Tandem structures have been introduced to the photovoltaics (PV) market to boost power conversion efficiency (PCE). Single-junction cells’ PCE, either in a homojunction or heterojunction format, are clipped to a theoretical limit associated with the absorbing material bandgap. Scaling up the single-junction cells to a multi-junction tandem structure penetrates such limits. One of the promising tandem structures is the perovskite over silicon topology. Si junction is utilized as a counter bare cell with perovskites layer above, under applying the bandgap engineering aspects. Herein, we adopt BaTiO 3 /CsPbCl 3 /MAPbBr 3 /CH 3 NH 3 PbI 3 /c-Si tandem structure to be investigated. In tandem PVs, various input parameters can be tuned to maximize PCE, leading to a massive increase in the input combinations. Such a vast dataset directly reflects the computational requirements needed to simulate the wide range of combinations and the computational time. In this study, we seed our random-forest machine learning model with the 3×10 6 points’ dataset with our optoelectronic numerical model in SCAPS. The machine learning could estimate the maximum PCE limit of the proposed tandem structure at around 37.8%, which is more than double the bare Si-cell reported by 18%.