داده کاوی برای بهبود عملکرد در ساخت نیمه رسانا و یک مطالعه تجربی

During wafer fabrication, process data, equipment data, and lot history will be automatically or semi-automatically recorded and accumulated in database for monitoring the process, diagnosing faults, and managing manufacturing. However, in high-tech industry such as semiconductor manufacturing, many factors that are interrelated affect the yield of fabricated wafers. Engineers who rely on personal domain knowledge cannot find possible root causes of defects rapidly and effectively. This study aims to develop a framework for data mining and knowledge discovery from database that consists of a Kruskal–Wallis test, K-means clustering, and the variance reduction splitting criterion to investigate the huge amount of semiconductor manufacturing data and infer possible causes of faults and manufacturing process variations. The extracted information and knowledge is helpful to engineers as a basis for trouble shooting and defect diagnosis. We validated this approach with an empirical study in a semiconductor foundry company in Taiwan and the results demonstrated the practical viability of this approach

The competition of the semiconductor wafer fabrication depends on cost, quality, and the delivery time, especially quality that is a key factor for enterprises to attain long-term competition. In the age of digital information, owing the rise of e-commerce and information technology, a large amount of data has been automatically or semi-automatically collected in modern industry. Decision makers may potentially use the information buried in the raw data to assist their decisions through data mining for possibly identifying the specific patterns of the data. The semiconductor fabrication processes are complex and lengthy. During the fabrication processes, voluminous data were generated and collected. As the process issue happens, engineers have to identify the root causes of problem as soon as possible to reduce the lost caused by excursion. Most of the engineers rely on their own domain knowledge and experience to identify the specific characteristics of abnormal products. However, such judgments are ineffective and limited by their own domain knowledge. Therefore, it has become an important topic to effectively transfer plethora and complex engineering data into valuable information and knowledge for process improvements and yield enhancement in semiconductor manufacturing. The extracted information and knowledge can assist the engineers as their reference and basis for advanced investigation of the root causes of the defects. This research aims to propose a framework for mining production data to extract knowledge for manufacturing process monitoring and defect diagnosis in order to remove assignable causes and thus improve the yield. In particular, Kruskal–Wallis test (i.e., K–W test) (Kruskal & Wallis, 1952) and decision tree methodology are applied to analyze and classify abnormal process stages in semiconductor manufacturing. An empirical study was conducted by using real data from a fab to validate the proposed approach. The results showed practical viability of this approach that can efficiently limit the scope for defect diagnosis and derive specific decision rules effectively. This paper is organized as follows. Section 1 describes research background, significance, and research aims of this study. Section 2 describes the fundamental of this research and reviews related literatures. Section 3 proposes a research framework with detailed procedures for semiconductor data mining and knowledge discovery from database. Section 4 validates the framework with an empirical study. Section 5 concludes with discussion and further research directions.

Data preparation is an important work in the data mining process and knowledge discovery in database in practice. Although the data selection and data preprocessing are very time-consuming, it cannot be ignored and needs much patience. The inappropriate data may lead to departure of mining results. Therefore, it is necessary to check the engineering data with domain experts iteratively. The proposed framework combines traditional statistical methods and data mining techniques to explore the huge semiconductor manufacturing data. Engineering data are fully utilized and developed effectively. Based on the empirical results, we validate that the proposed approach has practical viability. It helps domain engineers find out root cause when issue happens and provides information for decision makers to understand how to overcome the problem by the analysis framework. Indeed, data mining and knowledge discovery from database may range from its use as input for a decision process to its full integration into an end-user application. The results also can be used in an IT-enabled knowledge-based system for supporting manufacturing and business decisions in high-tech industry. The target variable used in this study is the yield rate that is like a synthetic index of the performance of hundreds of processes. Thus, it may be inconvenient for diagnosing defects because the fault causes may be obscure. Therefore, further studies should be done for fault detection and classification.