Bayesian Optimization of photonic curing process for flexible perovskite photovoltaic devices

Photonic curing is a thin-film processing technique that can enable high-throughput perovskite solar cell (PSC) manufacturing. However, photonic curing has many variables that can affect the processing outcome, making optimization challenging. Here, we introduce Bayesian Optimization (BO), a machine-learning framework, to optimize the power conversion efficiency (PCE) of photonically cured MAPbI3 PSCs on ITO-coated Willow Glass. We apply BO with four input variables—MAPbI3 concentration, additive CH2I2 volume, pulse voltage, and pulse length. These input variables were dynamically adjusted in response to the new data, an example of a human-machine partnership. With the limited experimental budget of 48 conditions, we achieved a champion PCE of 11.42% and predicted 14 new conditions resulting in >10% PCE. Beyond simple optimization, we examined the relationships between pairs of inputs with two-dimensional contour plots and investigated the relative importance of each input to gain insight into photonic curing. We demonstrate that BO is a powerful tool in process optimization and can be adapted to other PSC manufacturing cases. [Display omitted] •Bayesian optimization optimizes four inputs for flexible perovskite solar cells.•Human-in-the-loop ML: expanding the optimization boundaries in response to new data.•For 48 different sets of inputs, champion 11.47% device efficiency is achieved.•More inputs that produce high-performing devices are found than traditional method.•Sharply Additive explanation interprets the regression model.