Using process experienced correlation table to improve the accuracy and reliability of data mining for yield improvement

Abstract

The rapid innovation of new process technologies in the semiconductor industry, especially 12 inches Fab, along with continuously growing amounts of data, it is difficult to find root cause when problems occur in some process steps. It causes large amount of wafer scrapping. The analysis methods of traditional EDA system rely on experience of senior engineers. They need to define the suspected process step by their experience and then perform analysis. The analysis methods consume large amounts of human resources in order to determine the root cause of process and yield excursions. Hence, it is important that a knowledge retention method be incorporated to improve the efficiency of root cause analysis. Data mining, a new data analysis method that combines information science and technology of statistical analysis, is developed recently. The new generation data analysis method includes statistical, information science and mathematical calculation to find correlation between the target parameter, for example yield and other parameters. It will provide important clue to the analyzer. In addition, it also provides a direction to find root cause rapidly. It is difficult to find the correlation between the target parameter and other parameters by traditional statistical analysis method, and data mining can solve the blind point of the traditional method. This article discusses the design of how to define the relation between all data sources of semiconductor industry based on the experience of senior engineers. And it installs the relation to data mining analysis, it performs the analysis to identify relationship among all data sources. So, engineers can find the root cause of process issue in a short period of time