ASML
Graduation assignment: Analyze data to diagnose performance issues in ASML machines
Introduction
This is a graduation assignment for a master’s student in Engineering, Mathematics, Computer Science or Data Science with good communication and analytical skills, a hands-on attitude, who deals well with uncertainty, and has experience with data analytics.
Job Mission
ASML has thousands of lithography machines operating 24/7 at chip factories throughout the world. Due to their extreme accuracy, problems with lithography machines occur quite regularly. Downtime in a high-end chip factory is extremely costly: nominal costs for customers can be $20 per second of unscheduled downtime.
Performance problems on these machines are manually diagnosed mainly by constantly measuring (at sample intervals of minutes/hours/days) thousands of signals generated inside the machine. These signals are collected daily for all ASML machines in the field. In the past, this kind of data has mainly been used by human experts.
The ambition is to transition to structural automated monitoring whereby software helps capture the knowledge of the experts to allow generalists to deal with common failure modes.
You will investigate ways to preprocess signals (time series data) collected from a large population of ASML machines in such a way that understood abnormal behavior (“symptoms lead to problem via known mechanism”) are recognized and made explicit. This helps disseminate knowledge, check “known” assumptions and, where possible, the system should makes unexpected behavior (“strangely enough X leads to Y”) explicit. This helps empirically investigate complex system behavior and discover forgotten or unknown dependencies. We would prefer actual working examples of the techniques/ideas/approaches discussed.