Taking performance analysis to the system level

Within the Maestro project, ASML and TNO-ESI are working on a model-based methodology to diagnose and predict system performance.

Building on a tool developed in an earlier collaboration, they’re looking to reactively identify and analyze productivity issues in the field as they’re gradually moving on to proactively tackling potential problems.

With every new generation of lithography machines, better focus and overlay are required to print smaller chips. Accordingly, the number of software tasks and their complexity increases tremendously, impacting the performance of the system.
“As the chip details get smaller, we need to do more corrections and more calculations,” explains Jos Vaassen, software product architect at ASML. “At the same time, we’re allowed fewer timing errors because with even the slightest time difference, there will be a temperature variation causing material expansion or shrink, as a result of which we lose focus and overlay.”

To ensure that performance targets are met, ASML and TNO-ESI, in collaboration with Eindhoven University of Technology (TUE), have been looking into an approach to diagnose the system behavior and quickly find and analyze the root cause of anomalies. In the Concerto project (2016-2019), the partners developed a model-based methodology to scrutinize the software execution and keep computational tasks out of the critical path as much as possible. “The aim of Concerto was to make problem-solving in our machines way easier by automatically detecting software hiccups, gathering data and then reconstructing from that data what exactly happened when the issue occurred,” summarizes Vaassen.

As a follow-up, the Maestro project (2020-2023) takes the methodology to the system level. “Most of the productivity issues in the field are directly dealt with at the customer. Only a fraction is passed back to Veldhoven. Unfortunately, performance anomalies related to the execution of the software don’t immediately end up in the right place there, as the support team operates on a system level and has no intimate software knowledge. Consequently, those issues take longer to resolve,” notes Kostas Triantafyllidis, research fellow at ESI. “With Maestro, the aim was to streamline troubleshooting by connecting observed anomalous system behavior to software root causes, so that software-related issues can be easily delegated to the appropriate team.”`