Zero Defect Manufacturing of Microsemiconductors - An Application of Machine Learning and Artificial Intelligence

A real-time quality monitoring of the detection and prediction of a defect in fluid dispensing systems is presented. A case study of adhesive placement and dispensing in a semiconductor production system demonstrates the applicability of a combination of PCA to explain the variations in the amount of dispensed fluid syringe needle placement and event-based learning to express the causal relationship between machine and production state with defect types. The resulting definitions of system state and interrelationship of control parameters build the building blocks of Gene Expression Program (GEP) that predicts the formation of droplets and fail or pass product. The results show 99.93 % of accuracy in prediction of defect which is based on the obtained data from glue dispensing model. This integrated solution provides the genetic signature of the glue dispensing process helping to eliminate defects and the adjustment of system state prior to defect formation.