Machine learning-based quality assurance in the production of X-ray detectors, FHNW School of Computer Science

Early detection of chip defects in HPC modules using machine learning-based methods

DECTRIS is an innovative, world-leading company that manufactures X-ray detectors based on Hybrid Photon Count (HPC) technology. The company is currently undergoing a transformation from single-unit production to small-batch manufacturing.

Against this background, quality control is a key step in the manufacturing process of HPC modules, especially in an international and increasingly competitive environment.

Early identification and elimination of defects along the production chain is of central importance for economic reasons. During product qualification and validation, increasing amounts of data are collected, offering significant potential for predicting device adjustment parameters.

The use of traditional analytical approaches has so far not achieved the desired results in identifying and eliminating defects.

The Institute for Data Science FHNW examined existing data for correlations and modeled dependencies within the data using machine learning (ML) algorithms. Deep learning, in particular, is well-suited for predicting the quality of manufactured end products based on pixel or chip properties.

In selected cases, a successful machine learning-based method has been developed that enables early prediction of the relationship between measurement data and quality characteristics of chips across production steps. Predictions of specific calibration parameters were deemed directly relevant to practical application, as they are suitable for improving the combination of sub-components in production and have the potential to increase production yield. Based on the promising results of the feasibility study, an Innosuisse project has now been submitted.