Data Analytics & Machine Learning

Wafer maps of electrical measurements after frontend production are used to evaluate the quality of the produces devices, therefore patterns need to be identified and classified.

The amount of data generated during the semiconductor manufacturing process and the subsequent verification and reliability analyses is large due to the high complexity of the products, but also diverse because the data sources include electrical and physical measurements, output from different type of sensors and images from physical failure inspection. Combining these data sources and extracting necessary information for decision making processes are important step towards the digitalization of semiconductor production and reliable prediction methods. Data analytics and Machine Learning methods are powerful tools for these challenges, because they are able to consider large amounts of data and at the same time they can handle complex dependencies between data.

Statistical analysis and classification of signal patterns in electrical wafer test data support the automatic identification of test equipment issues.

KAI Competences

KAI develops and applies data mining concepts, advanced statistical screening methods and machine learning algorithms for the semiconductor industry since 2013. In this time KAI established key competences in the following areas:

Methods for advanced statistical device screening to identify risk devices in the electrical measurements after the semiconductor frontend production (e.g. Independent Component Analysis, multivariate screening techniques)
Pattern recognition methods for electrical measurements (e.g. Markov Random Fields, Stochastic Processes)
Image processing techniques for automatic extraction and quantification of visual information
Application of Machine Learning algorithms (e.g. Support Vector Machines, decision trees, Random Forests, Neural Networks)

Combining image processing with Machine Learning (ML) techniques enables automatic classification of degradation features in metal layers.

Research Goals

Combine pattern recognition and machine learning methods to develop an algorithm to judge the quality of the produced devices after semiconductor frontend production (health factor)
Use advanced data analytics and machine learning techniques to relate data from different sources of semiconductor frontend production in order to identify root causes for process variations
Combine classical image processing analysis with machine learning approaches in order to enable efficient and objective extraction of quantitative information out of images from physical failure inspection of semiconductor devices (microstructure of power metallization, degradation process)