Predictive Maintenance Practices for Cryogenic Pumps in Semiconductor Manufacturing

Many critical steps in semi manufacturing need high vacuum or controlled ambient conditions. This need is met through a very extensive network of vacuum and abatement systems. In a typical fab this network consists of many thousands of pumps, abatement, and ancillary equipment. This provides and maintains vacuum levels and quality, two key process parameters. Applying Smart Manufacturing and Predictive Maintenance is key to Operational Excellence and reducing risk and uncertainty associated with unplanned vacuum and abatement downs. In this paper we focus on cryogenic pumps and discuss using both rule-based and statistical models (ML models) to provide maintenance guidance and maintenance prioritization. Rule-based case studies include Helium circuit contamination detection and Cryo regeneration fault detection. Our ML models, were developed, trained, and verified on extensive HVM pump data. They were applied to 3 different pump types and scored high on key binary classifier rate metrics, as high as 93% accuracy and 87% recall depending on pump type. As we collect more data our models will continue to learn and improve and further reduce risk and uncertainty.