Machine learning guided efficiency improvement for Sn-based perovskite solar cells with efficiency exceeding 20%

Eco-friendly lead-free tin (Sn)-based perovskites have drawn much attention in the field of photovoltaics, and the highest power conversion efficiency (PCE) of Sn-based perovskite solar cells (PSCs) has been recently approaching 15%. However, the PCE improvement of Sn-based PSCs has reached bottleneck, and an unambiguous guidance beyond traditional trial-and-error process is highly desired for further boosting their PCE. In this work, machine learning (ML) approach based on artificial neural network (ANN) algorithm is adopted to guide the development of Sn-based PSCs by learning from currently available data. Two models are designed to predict the bandgap of newly designed Sn-based perovskites and photovoltaic performance trends of the PSCs, and the practicability of the models are verified by real experimental data. Moreover, by analyzing the physical mechanisms behind the predicted trends, the typical characteristics of Sn-based perovskites can be derived even no relevant inputs are provided, demonstrating the robustness of the developed models. Based on the models, it is predicted that wide bandgap Sn-based PSCs with optimized interfacial energy level alignment could obtain promising PCE breaking 20%. At last, critical suggestions for future development of Sn-based PSCs are provided. This work opens a new avenue for guiding and promoting the development of high-performing Sn-based PSCs.

Graphical Abstract