Constructing the composition ratio prediction model using machine learning for BaSi2 thin films deposited by thermal evaporation

Abstract

A composition ratio prediction model for BaSi2 thin films deposited by thermal evaporation was constructed using machine learning. BaSi2 was prepared by thermal evaporation in a vacuum chamber, and the composition ratio was measured by energy-dispersive X-ray spectroscopy. The results show that the composition ratio is affected by various experimental parameters. To consider these parameters, kernel ridge regression was performed with Si/Ba ratio as the objective variable, and with experimental parameters as explanatory variables. A good fitting result was obtained by kernel ridge regression. The next step was to select a kernel function. We evaluated four types of kernel functions, and confirmed that two of them, the polynomial kernel and the sigmoid kernel, have relatively high prediction accuracy. Then we investigated different combinations of explanatory variables and found the best combination with the highest generalization performance. From the above, a composition ratio prediction model with a mean absolute error of less than 0.2 was obtained.