Machine Learning for manufacturing anomaly detection
GLOBALFOUNDRIES is the world’s leading specialty foundry. They deliver differentiated feature-rich solutions that enable our clients to develop innovative products for high-growth market segments. GF provides a broad range of platforms and features with a unique mix of design, development and fabrication services. With an at-scale manufacturing footprint spanning three continents, GF has the flexibility and agility to meet the dynamic needs of clients across the globe.
GlobalFoundries wanted to reduce unplanned downtime of the CMP tool by predicting degradation in the process as wafers run through CMP tool. Chipmakers like GF use a process called chemical-mechanical planarization, or CMP, for short. They wanted to leverage artificial intelligence (AI) to build a model that predicts failures.
Google Cloud Implementation
GF collaborated with Google Cloud Platform (GCP) and SpringML to build such a model. The approach to developing the model involved three high-level steps.
- Data ingestion – gathering data and exploring it to understand patterns
- Model development – usually entails applying more than one model to determine the best fit
- Model testing and tuning – ensuring the model are tuned so that it provides the best possible accuracy for predictive maintenance.
Time series data from machines and sensors were ingested into the Google Cloud Platform. Various time series and classification models were built and deployed to the GCP CloudML platform for training, hyperparameter tuning and for inference. SpringML team also ensured that the GF team was familiar with the model and were in a position to take ownership of the solution.