Reliable Digital Twins in Manufacturing for Robust and Uncertainty-Aware Prediction of Critical-to-Quality Parameters

3:55 pm - 4:20 pm

Data-driven and digital approaches applied to material suppliers’ shop-floor operations deliver substantial value to semiconductor customers while generating significant internal financial returns. Merck Electronics has demonstrated this through scrap elimination, reduced quality control measurements, increased throughput, and better quality for their customers. This success is built on manufacturing expertise combined with interpretable AI to create competitive differentiation and enable scalability without quality compromise. Critical enablers include ontologies, digital twins, robust MLOps, and uncertainty quantification that ensure reliable predictions in production environments. This presentation outlines how ML can be applied reliably on real-world shopfloors to generate value at scale.

In semiconductor materials manufacturing, maintaining high-quality standards is critical for consistent performance, optimal yield, and avoiding excursions. Predictive quality with uncertainty quantification enables dynamic adjustments in chemical production processes to address natural variations in raw materials and environmental conditions. By forecasting key quality parameters before production with confidence intervals, manufacturers can reduce cycle times and decrease scrap rates while creating measurable customer value.

Effective implementation requires reliable machine learning systems that automatically address feature and concept drift caused by manufacturing process changes. These models must be interpretable to ensure manufacturing staff can depend on their predictions in day-to-day operations. After a year, model performance drift and lack of prediction intervals mean outliers are often missed as prediction error grows. Robust MLOps-driven AI combined with explainable AI (XAI) significantly boosts outlier detection while maintaining low false positive rates.

Our approach integrates digital twins with semantic data layers that power predictive maintenance and automated root cause analysis. Predictive quality models and intelligent raw material allocation operate on virtual formulations to forecast critical parameters. By combining MLOps with explainable AI and uncertainty quantification techniques, reliable AI scales these operational improvements without compromising quality commitments to customers, creating lasting competitive advantage in semiconductor materials manufacturing.

Featured Speakers

Steven Eulig

Head of Digital Solutions, Merck

Steven Young Eulig is the Head of Digital Solutions at the Electronics business of Merck KGaA, Darmstadt, Germany. Since joining the company in 2020, he has held responsibilities spanning in-house consulting, business development, and semiconductor customer engagement. In 2024, Steven took over leadership of the global Digital Solutions organization, which drives the business's digital transformation by being deeply embedded in manufacturing, supply chain, and commercial operations. His department also leads data engagements with customers to improve product quality and performance in the fab. Beyond improving the foundations, the Digital Solutions organization is focused on creating tangible profit impact for the Electronics business.

Prior to joining Merck KGaA, Darmstadt, Germany, Steven Eulig was a strategy and data science consultant for 3 years. He received an MSc degree in Physics from Ruhr University and is currently a research scholar in Physics at Harvard University. He also acts as a researcher at the NSF-funded Institute for Artificial Intelligence and Fundamental Interactions (IAIFI) and is part of its industry committee.

Gianni Klesse

Head of Data Science and Digital Business, Merck

Dr. Gianni Klesse is the head of the Data Science & Digital Business team within the Digital Solutions organization at EMD Electronics, the Electronics business of Merck KGaA, Darmstadt, Germany. His team specializes in designing and implementing advanced machine learning and AI solutions across the Manufacturing, Supply Chain, and Commercial domains. In addition, Gianni spearheads Electronics’ data-sharing engagements with leading-edge customers in the semiconductor industry.

He earned his doctorate in Computational Biophysics from the University of Oxford, complemented by a M.Sc. in Computational Science from the University of Amsterdam and a B.Sc. in Physics from the University of Heidelberg.