Results
The project reduced the time needed to identify performance improvements at customer sites from six weeks to less than one day. This frees highly skilled experts to focus on other important work and supports the new business model based on productivity-driven service contracts. The estimated additional annual revenue from this contract model is €150 million, demonstrating a major financial impact. At the same time, optimizing scanner performance supports technological progress and innovation. This project aligns with our mission “Data for a Better Future” by improving efficiency and productivity in semiconductor manufacturing, an industry that is essential for technological and societal progress.
The Process
Analyze
Together with the domain expert, we defined what really determines machine performance. We looked at how uptime and productivity are related and which data sources describe them. We also studied how the expert performed the analysis: what information was needed, how much time it took, and how many analyses could be done each year. This gave us a clear understanding of the complexity and the business value before starting the implementation.
Strategy
We designed how different data sources could be combined into one coherent system. A single machine can generate hundreds of gigabytes of data each year, so it is essential to find the important signals in a very large amount of data. We developed modular and scalable data products that transform raw event logs into useful datasets. These datasets make it possible to analyze and improve scanner performance.
Execute
In the first version, a proof of concept, we already showed clear impact: analysis time was reduced from several weeks to a few days. We then expanded this into a minimum viable product (MVP). For the first time, ASML could see performance across the entire fleet of machines. This was presented several times to senior leadership (VP and EVP level). We also added the ability to explain why performance decreases, enabling root cause analysis and performance optimization. This helped engineering teams and leaders quickly identify and solve productivity issues. We called the tool “Where Are My Wafers”, because it shows exactly that: how well the machines perform and where improvements are possible. It also made fleet-wide comparisons possible, for example, understanding why some customers achieve better performance than others.
Maintain & Scale
The system was built using Azure Databricks and Apache Spark. We created trusted datasets, an ASML concept for shared and reliable data across the organization. This supports strong data governance and security. The system runs automatically and is designed to be scalable and flexible, with clear knowledge transfer and governance to support long-term use.
The Technology
What the client says
"I really liked working with Floris. He is a really good listener and can translate my requirements into the right implementation, giving it an extra edge. This means that the final result was better then I ever envisioned."
"[...] smart, highly motivated, innovative and energetic. He thinks critically about why, always focuses on the big picture and aims for a structural solution. He understands the data landscape on the high level and also the nitty-gritty details."
"Floris surprised me by possessing the combination of the following traits: smart, collaborative, enthusiastic, very eager to learn and passion for the technical side and people side of his work."