Efficiency and Stability Analysis of 2D/3D Perovskite Solar Cells Using Machine Learning

Abstract

A dataset containing 599 data points from 146 publications on 2D/3D perovskite solar cells is analyzed using machine learning. The predictive models are developed for power conversion efficiency (PCE) using eXtreme Gradient Boosting regression, random forest regression and artificial neural networks while association rule mining is used to analyze the stability data to identify the descriptors leading to high stability 2D/3D cells. A predictive model is also developed for the bandgap to predict the missing values in the dataset for the use in PCE predictions. Models for both bandgap and PCE predictions are quite successful. The thickness of inorganic layer (n), radius of anion (R x), and 2D cation (R m) are found to be the most important descriptors for bandgap predictions; n and R m, together with the bandgap, are found to be deterministic for PCE in regular cells while the bandgap, n, and conduction band energy of hole transport layer are the most influential descriptors in inverted structures. Association rule mining analysis for the stability indicates that the cells with layered perovskite structures are more stable while the 2D and 3D cations leading to the most stable cells are found to be butylammonium and formamidinium-Cs mixed cation respectively.