建構萃取半導體製造智慧之架構以提升營運效率及其實證研究

Empirical Studies to Structure a Manufacturing Intelligence Framework to Improve Operation Efficiency of Semiconductor Manufacturing

徐紹鐘(Shao-Chung Hsu)

指導教授 ： 簡禎富

國立清華大學/工學院/工業工程與工程管理學系/博士(2009年)

摘要

所有企業都面臨到如何兼顧持續獲利與成長，尤其是高科技業，由於製程的複雜與快速的產品生命週期及製程快速的更新，造成許多企業面臨到比過去傳統製造業更嚴峻的挑戰。為了維持競爭力，不斷的追求成本降低，品質提升及加快交期是每個製造業不可避免的使命。 然而一流的企業與其他的差別，除了在技術與規模外，他們累積了許多的製造智慧，可惜的這些智慧往往存在於個人。 另一方面，資訊技術的演進，使得資料儲存不在是問題，取代而來的反而是如何從大量資料萃取出有價值的資訊協助決策。
過去10年，商業智慧的工具蓬勃發展，但都只侷限於從資訊技術觀點出發，而忽略了分析是其中的核心，更不用說為製造量身打造的智慧萃取工具。本研究依據資料採礦流程，各類分析應用及系統分析架構，歸納提出一套針對不同的製造策略與目標發展萃取製造智慧的架構的方法。依造此方法所制定的分析流程及應用程式，可以在企業中改變成以事實為出發的決策模式。
為了驗證架構的成效，我們挑選晶圓圖良率分析問題，透過實證研究證實相關架構的可行性。而進一步透過資訊系統的建置，實際帶來企業的功效，相關的系統已經建置在一個台灣的八吋晶圓廠及一個測試 Bumping 廠並實際反應在良率分析的時效提升。

關鍵字

製造智慧 ； 半導體製造 ； 良率分析 ； 晶圓圖 ； 分群

並列摘要

The whole industry is facing the challenges of profitability and growth year by year. In particular, high-tech industry, which features complicated manufacturing processes, rapid product and process migration, and limited delivery time to customers, is encountering a tougher challenge than traditional manufacturing industry before. This inevitably induces companies to compete with each other by continuously employing new technologies, increasing yield, and reducing costs. What makes a first-rate company stand out from others is, aside from its scale and technology, its competence in analytics, which is based on manufacturing intelligence. However, the manufacturing intelligence exists in humans, processes, system, and data. These can be used to help manufacturing managers and their staffs with ill knowledge discovery, knowledge inference, and knowledge explanation in operation management. The issue of how to extract manufacturing intelligence (MI) and integrate with enterprise system is increasingly important since advanced fabrication technologies are complicated and interrelated. Thus, most existing studies focus on the functionalities, which are built from information technology (IT) viewpoints. They are subjective, non-objective analyses, and they are also restricted by human experiences. Most IT tool vendors proposed packages but brought only a limited benefit to their clients.
This research describes the development of a process, with a systematic approach, to structure an MI extraction framework in semiconductor manufacturing. The applications based on this framework would lead to changes and improvement in the semiconductor industry. The process to generate the framework is linked with specific manufacturing objectives in developing an analytical system with data mining tools. These frequently-talked objectives during manufacturing include cost reduction, yield enhancement, and productivity improvement. The focus of this research is quality improvement.
The values of proposed approach in constructing MI framework are: (1) integration with business process and objectives; (2) decomposition of the analysis task into 4 level and 3 analytical objectives which can be enabled by IT system; and (3) effectively extract value information for decision. To validate the viability, an empirical study of yield improvement via wafer bin map clustering and classification from an 8” wafer fab is conducted for the proposed framework. The results show a good performance in efficiency improvement and effectiveness as to standardizie the analytical process to engineers.

並列關鍵字

Manufacturing Intelligence ； Semiconductor MAnufacturing ； Yield Analysis ； Wafer Bin Map ； Clustering

參考文獻

Chien, C. (2005). Decision analysis and management- a Unison framework for total decision quality enhancement. Yeh-Yeh book gallery, Taipei.

Chien, C., Lee, P. and Peng, C. (2003). Semiconductor manufacturing data mining for clustering and feature extraction. Journal of Information Management, 10(1), 63-84.

Peng, J., Chang, S., Chien, C. and Yang, J. (2005). Constructing a data mining framework of association rule and an empirical study for fault location. Journal of Information Management, 12(4), 121-141.

Agresti, A. (1990). Categorical data analysis. Wiley, New York.

Bergeret, F. and Gall, C.L. (2003). Yield improvement using statistical analysis of process dates. IEEE Transactions on Semiconductor Manufacturing 16(3), 535-542.