Accelerating the Design and Manufacturing of Perovskite Solar Cells Using a One-Shot Automated Machine Learning Framework

Abstract

Researcher-led approaches to perovskite solar cells (PSCs) design and optimization are time-consuming and costly, as the multi-scale nature and complex process requirements pose significant challenges for numerical simulation and process optimization. This study introduces a one-shot automated machine learning (AutoML) framework that encompasses expanding the design space, accelerating the simulation, and optimizing the manufacturing process parameters of PSCs. By integrating support vector regression (SVR), AutoML, multi-objective immune algorithms (MOIA), and reverse engineering methods, the design space for PSCs was expanded 100-fold, reducing the time required by approximately two orders of magnitude, and successfully increasing the simulated photovoltaic conversion efficiency (PCE) of PCSs from 21.83% to 31.29%. 5 optimal combinations were quickly identified from 166 sets of manufacturing parameters for the manufacturing of perovskites using the rapid spray plasma processing (RSPP) technique, consistent with the results of 30 experiments. This optimization strategy not only refines the process but also embeds a continuous improvement mechanism into the system. This work significantly speeds up the full circle of design, simulation, and manufacturing of PSCs, elucidating the early components of the digital twin framework, and providing an important reference for a comprehensive digital twin system for PSCs.