Machine learning based model for plasma prediction in high power pulsed magnetron sputtering processes

The coating development using physical vapor deposition (PVD) is time intensive and expensive. The coating processes are usually developed and improved based on the operator's experience. In order to improve the understanding of the processes during coating deposition, methods of plasma diagnostics can be used. However, this is time intensive and requires the installation of special diagnostics. The aim of the current study is to build a machine learning based model of PVD processes to predict the ionized to excited intensity ratio of the plasma species during the coating process. This enables a knowledge-based, cost-reduced process improvement. Based on measured data of different coating processes, the models were trained and tested in the current study. Therefore, a database was established by measuring process and plasma properties in hybrid processes of direct current magnetron sputtering and high power pulsed magnetron sputtering using an industrial coating unit. Different processes with variations of cathode powers and gas flows of the reactive gases oxygen and nitrogen were processed. The intensities of the ionized and excited species were measured using optical emission spectroscopy with six substrate side positioned collimators for spatial resolution to calculate the intensity ratios. The data was measured time resolved during the coating process. Compared to the measured data, the predictions show similar trends and values for the species within the coating chamber. The model was used to predict the influence of oxygen gas flow for CrAlON coating development. Furthermore, the models can be used for a more targeted process development. This can contribute to the application oriented adjustment of the coating processes and the coating properties.