Accelerating Bonding Overlay Error Optimization in Fine-Pitch Wafer-to-Wafer Hybrid Bonding Through Machine Learning

Fine-pitch wafer-to-wafer hybrid bonding is pivotal for advancing semiconductor integration and requires high overlay accuracy to ensure device performance and yield. Overlay error optimization by trial-and-error experimentation is often resource-intensive and time-consuming. This research investigates the application of machine learning (ML) to expedite the optimization process of overlay errors in fine-pitch wafer-to-wafer hybrid bonding. We develop data-driven predictive ML models that can predict overlay errors for a given set of overlay correction values to assist the process engineers during their design of experiments (DOE). This will significantly accelerate hybrid bonding process learning cycles while enhancing process maintenance, resulting in improved product quality and increased manufacturing efficiency. When fully integrated, these ML driven overlay error optimization techniques for hybrid bonding can be streamlined to automate predictions, optimize corrections, and offer real-time insights. Our results underscore the efficacy of machine learning in enhancing the efficiency and precision of semiconductor manufacturing, offering a promising pathway for improved production reliability and cost-efficiency.