Uwe Krautz, L6S Master Black Belt, GlobalFoundries
Big data and Machine Learning are a powerful combination that can help the semiconductor industry improve product quality and automation. However, the data from production tools or measurement systems (e.g., FDC/SPC signals, raw sensor data, process measurements, etc.) often need to be cleaned and preprocessed before using Machine Learning approaches. Data can exhibit outliers, multicollinearity, or strange distributions that make it hard to use simple modeling techniques like linear regression or more advanced tools like deep neural networks that learn to distinguish artifacts but are not the root cause of a problem. Thousands of data sets need to be filtered and reduced to a limited number of key features. This whole process of data cleansing and feature engineering can be accelerated using JMP data visualization and analysis tools so that data scientists can focus on building ML models without spending too much time fighting with data preparation. The findings from JMP® can be translated to Python-based ML, which can be automated and scaled via cloud-based computing.
