Optimizing Energy Efficiency of Lithography Machines

The semiconductor industry is characterized by its continuous pursuit of miniaturization and performance improvement. In such a rapidly evolving environment, achieving better performances is crucial for sustaining growth, innovation, and profitability. ASML in one such company, as it is currently holding the position of industry leader of the lithography machine production business. The complexity of the tools employed in their production processes yields an abundance of data, presenting a challenge in terms of intricate interpretations and analysis. Machine Learning approaches emerged as one of the key roles in managing and infer conclusions from such vast and complex data setting. Their capacity to enhance operational efficiency by facilitating precise control over variables and settings has the potential to significantly minimize expenses, as the nature of these multi-million Euros machines would amplify any small gains in profitability. Lithography machines demand substantial amounts of energy, contributing to both rising operational costs and environmental concerns. This thesis addresses the pressing need to optimize the energy efficiency of lithography machines to mitigate these challenges. Through the application of various machine learning models, majority of which belongs to the ensemble learning family, this research aims to identify the most important parameters for minimizing the energy leakage during the plasma creation phase in EUV lithography machines and to define a regression model able capable of predicting these signals using alternative inputs. The importance of such a regression model lies in the current way to track the overflowing energy, which are taken from repurposed sensors not built for this specific function. As such, defining a robust model that can avoid the need for data collected from such sensors becomes is of great importance for the company.