P4 Increasing R&D productivity

The ever-increasing complexity of high-tech equipment as well as the ongoing technological progress require more and more R&D effort and skills. R&D organizations see themselves challenged in managing complexity and knowledge while, at the same time, they need to speed up.  

Evolutions in software development bring new tools like generative AI and more proven methods like modelling techniques. In this session we see two industry examples. The first takes us through complex timing and performance analysis where data, while interpreted, will be validated against a known model. The second case deals with defect analysis and root cause prediction based on R&D data sources using  NLP and supervised learning techniques.

The presentation of the two state-of-the-art topics will be followed by a lively discussion between the chair and the presenters in which the reliability of the two approaches will be discussed and the benefits of one approach will be compared against the other. 

In the Accelerando project, a collaboration between Philips Healthcare and TNO-ESI, we have investigated ways to speed up the bug resolution process, which starts when a problem is detected and ends after the documented issue has been formally closed.

We considered 3 steps from this process:

1. Collecting information and entering the issue in a database

2. Assigning the issue to an engineer or team, so called triaging, and

3. The fault analysis required to identify the cause of the issue.

We zoom in on the use of Machine Learning to support the triaging of issues. During our work, we found out that several factors determine whether Machine Learning can replace human triagers.

In our presentation we will discuss possible pitfalls of applying Machine Learning in the industrial practice.

The ASML TWINSCAN system has tremendously evolved the last two decades, delivering ever higher overlay performance (from <20nm in 2001 to <1 nm in 2023) combined with increasing throughput (from ~90 WPH in 2001 to >330 WPH in 2023).
To manage the extreme performance/timing requirements, the computation platform and the software actions have been advanced towards high parallelism and concurrency, respectively. Hundreds of intertwined parallel tasks at runtime make the diagnosis of a performance anomaly a very challenging task, requiring extreme domain knowledge to mentally connect observations on platform level to specific ASML application artifacts. On top of that, the many configuration differences make a knowledge driven analysis virtually impossible.

This presentation presents the PPS model-based methodology and its supporting (T-iPPS) tooling developed by TNO-ESI and applied by ASML on TWINSCAN for fast and automated root-cause analysis of performance anomalies, showcasing a more efficient way of working.

• P1 Digital transformation, the road ahead

• P2 Digital transition: continuous evolution

• P3 Digital engineering in brown field

• Keynotes