Predictive Maintenance of Pump and Abatement Equipment in a 300mm Semiconductor Fab

Effective maintenance strategy in semiconductor manufacturing factories (Fabs) must enable sufficient tool availability to meet fab capacity needs while minimizing the all-in costs of that capacity - including the parts and labor for scheduled maintenance and the parts, labor, and product losses associated with unscheduled maintenance. There are also the costs associated with developing and maintaining various sensors and controls algorithms. One part of this equipment maintenance strategy involves the scheduling of maintenance activities for complex sub-systems such as dry vacuum pumps and associated abatement systems. This paper presents a case study on how univariate FDC can be used as a Predictive Maintenance (PdM) technique by utilizing existing sub-system sensor data, enabling improved maintenance scheduling with deeper insights into process effects on the sub-systems performance as it pertains to a focus on equipment availability optimization. By utilizing “direct measurements” instead of probabilistic measurements, such as Machine Learning (ML), such an approach can generate unambiguous signals that predict over ninety percent of the associated sub-system failures.