Bayesian Optimization of Spray Parameters for the Deposition of Ga(2)O(3)-Cu(2)O Heterojunctions.

The accelerated discovery and optimization of materials relies on the integration of advanced experimental techniques with data-driven methodologies. In this work, Bayesian optimization (BO) is applied to optimize the ultrasonic spray pyrolysis (USP) process for the deposition of copper oxides, targeting high-quality Ga(2)O(3)-Cu(2)O heterojunctions for optoelectronic applications. By employing BO with an initial data set of 12 samples and conducting 4 USP parameter optimization cycles, significant improvements in device performance are achieved, with the open-circuit voltage increasing from 288 to 804 mV. During the optimization process, the performance of the model declines, necessitating the identification of a reliable subset of samples from the full data set. Through the application of BO, the cross-validation error of the model is minimized based on the sample selection, whereby accuracy is restored and generalizability is achieved. The subsequent model evaluation reveals two distinct deposition regimes, each characterized by unique process conditions, leading to specific material properties and device performances. These findings not only demonstrate the application of a data-driven experimental workflow in the context of thin film deposition but also highlight the importance of robust data validation and model evaluation.