A Data Mining Approach for Optimizing Manufacturing Parameters of Wire Bonding Process in IC Packaging Industry and Empirical Study

In the wire bonding process, different setting value of manufacturing parameters affect the quality of the solder joints and the speed of wire bonding significantly. And when a new product is introduced, it is difficult to quickly obtain effective parameters simply by relying on the experience of the engineers. The study aims to propose a data mining framework to help the IC packaging factories to increase production capacity by reducing the number of downtimes and speeding up the wire bonding time. The proposed data mining framework employed random forest and XGBoost method to classify the occurrence of defect, MARS algorithm to predict the wire bonding time, and genetic algorithm to search for the manufacturing parameters. By exploring the production data, the proposed framework finally provides a set of recommended manufacturing parameters of the second bonding process. This study cooperates with a leading assembly company in Taiwan to validate the proposed framework. The empirical result reveals that it improves the packaging yield by 0.03%. The domain engineers can also find parameters more systematically and quickly by the proposed framework.