Using machine learning for prediction of spray coated perovskite solar cells efficiency: From experimental to theoretical models

Low-cost perovskite solar cells (PSCs) have experienced unprecedented gains in power conversion efficiency (PCE) of up to 25% of lab-scale devices. To be realized in the market, however, PSCs are not only required to be efficient but also scalable in production. While spray coating has viability as an industrial manufacturing process for perovskite photovoltaics scaling, optimizing the spray conditions is often seen as a challenging and time-consuming process due to its complex and multidimensional parameters. Herein, we use a machine learning (ML) approach to capture the relationship between spray parameter settings to the resultant photoconversion efficiency (PCE) of PSCs from experimental collected data points. This data-driven approach has the potential to accurately predict PCE values given the manufacturing parameters, enabling optimization and resulting in an increased experimentally recorded PCE. Furthermore, we also used a Convolutional Neural Network (CNN) to predict defect size distributions in the PSC structures to improve the understanding of defect formation mech-anism at given spray parameters. The implications of the results are discussed for optimizing spray manufacturing process of efficient perovskite photovoltaics.

Automated summary

Low-cost perovskite solar cells have achieved a power conversion efficiency of up to 25% in lab-scale devices. However, to be commercially viable, these cells need to be both efficient and scalable in production. In this study, a machine learning approach was used to optimize spray parameter settings and predict photoconversion efficiency values, resulting in an increased experimental PCE. Additionally, a Convolutional Neural Network was utilized to improve understanding of defect formation mechanism in the perovskite structures.