Tignis - DMEA244-P01

Deposition of thick SiC epilayers presents unique process development and control challenges due to the long deposition time and high dimensionality. We aim to assess the feasibility of reducing process development costs and suppressing process output variance by employing machine learning enhanced technologies. We will employ simulation studies that replicate real-world process development and process control tasks. Our experiments will compare the process development costs and process output variance that can be achieved with various machine learning solutions and compare results with classic design of experiments or advanced process control baseline strategies. Specifically, for process development tasks we will explore the benefit of employing Bayesian optimization techniques with informative priors leveraging insight from similar processes or simulations to more efficiently sample a recipe space. For process control, we will assess the output variance suppression that can be attained using our proprietary machine learning run-to-run control algorithm. Collectively, this work will assess the potential value in employing machine learning process development and control for thick SiC deposition and give recommendations on the most promising machine learning approaches. Finally, we will design a software solution that would allow process engineers to employ these machine learning approaches after further development in subsequent work.