As part of the appointment project "Smart Systems & Advanced Data Analytics", a demonstrator for a digital twin in high-mix low-volume semiconductor manufacturing is being created using the example of the chemical mechanical polishing (CMP) process, which has been poorly understood to date.
For this purpose, data from process equipment of different manufacturers and the associated measurement systems are fed into a specially developed data infrastructure based on open-source software, and raw data are analyzed automatically. Resource-efficient, smart sensors allow not only the acquisition of additional data but also the connection of plants without data interface, as they are often used in research operations and small and medium-sized enterprises.
User-friendly dashboards allow process engineers to interact with collected data and models created from it, for targeted prediction, monitoring and optimization of process results. The project focuses on optimization in terms of high, uniform material removal across the wafer while reducing defects.
From a modeling perspective, the project addresses two fundamental challenges of digital twins: While for mass production the scaling problem, i.e. the transfer of models to similar equipment (domain adaptation) is in the foreground, the challenge in research and in high-mix/low-volume manufacturing is to make do with little data. By integrating prior knowledge of our project partners, in the form of expert knowledge, physical wafer-scale models and historical data sets, fast learning, robust and transferable hybrid models are created.