Abstract
During the process of wafers in semiconductor manufacturing, many kinds of data will be collected automatically or manually to monitor the product. It is important to fix the problem of the manufacture process as soon as possible. However, the engineers sometimes can not find out the root-cause from the mass manufacturing data timely and rapidly. This research focused on constructing a conceptual framework for data mining of the semiconductor manufacturing. In this framework, there are four phases: problem definition and problem structuring, data preparation, modeling, and evaluation/ interpretation. Two empirical studies with different type of manufacturing data were applied to the framework. In the first empirical study, we focused on the wafer bin map data and constructed the clustering process by the neural network and the spatial statistic test. It can help engineers to find out the root-cause of problems through the cluster groups of common fail bin patterns. In the second empirical study, we developed the data mining methods of diagnosing defects of the WAT parameters and the related process information. In the use of decision tree and K-W test, we can extract the valuable information or knowledge for the domain engineers to identify the trouble process step. The final results of study showed the domain engineers could improve the effectiveness of trouble shooting.