多維貝氏切割法在半導體製程改善之應用

Multiple Bayesian Segmentation for Process Improvement in Semiconductor Manufacturing

黃銘勳

指導教授 ： 盧鴻興

國立交通大學/理學院/統計學研究所/碩士(2009年)

https://doi.org/10.6842/NCTU.2009.00241

摘要

因為半導體產業是一種高投資產業，每間公司在製程上所投資的成本都很高，因此，對每間公司而言，如何使製程穩定便是一個很重要的目標。因為想要使製程穩定，所以當製程發生問題時，工程師便需要去找到製程發生問題的地方，但由於半導體產業的製程相當複雜，與製程相關的參數非常的多，因此當製程發生問題時，如何利用這麼多的製程參數去找到製程發生變動的所在位置，對工程師而言實為一大挑戰。
在現有文獻中，對於檢驗製程是否發生變異並找出其正確位置，往往都著重在針對單一參數去找尋製程變動的位置，因此開發多變量偵測製程變異之系統極為重要。在本篇文章中，將利用數學方式建立模型，提供多變量偵測製程之方式，幫助工程師更有效率解決製程中有問題的地方。
本篇論文所提供的方式，主要想法來自於賴政言於2008 年所探討的 SBS (Single variable Bayesian Segmentation)，並對其做改良，進而將此方法推廣至多個維度上。由於半導體產業的製造過程中，常常會出現離群值，而對數學上建立模型產生困擾，因此，本文中對於離群值之出現，也提供一個方式來解決離群值對建立模型的影響。

關鍵字

貝式 ； 多變量 ； 分群

並列摘要

Because the semiconductor industry is a high-invest industry, all companies invest a lot of money in the process. So, it is an important target to make the process stable for every company. Because the target is to make the process stable, when some changes occur to the process, engineers need to find the position where a change occurs. Because the semiconductor industry’s process is quite complex, there are many different parameters that are relating to the process. When there are some changes in the process, how do engineers use these parameters’ data to find the position where the change occurs is an enormous challenge for engineers.
In current literature, the most methods are using one parameter’s data to find a position where changes occur to the process. Therefore, it is very important to develop a system that can use all parameters’ data to find a position where the change of the process occurs. In this
paper, we will establish a model using mathematics, provide a multivariate method to detect the process, and provide engineers a more effective solution to find the position where change occurs.
The main ideal of this paper comes from SBS (Single variable Bayesian Segmentation) that was investigated in Zheng-Yan Lai (2008). This paper improves on it and provides a new method that can use many parameters’ data to find the position where change occurs. In the manufacturing process in the semiconductor industry, outliers often occur and affect the model that is established in mathematics. Therefore, this paper also provides a method to solve it.

並列關鍵字

Bayesian ； Segmentation ； Multivariate

參考文獻

Good Rate”, National Chiao Tung University, Master, 2008.

[2] A.R. Taylor and S.R. Duncan, “Case studies in Bayesian segmentation applied to CD

control”, INTERNATIONAL JOURNAL OF ADAPTIVE CONTROL AND SIGNAL

[3] Marc Lavelle, “Optimal Segmentation of Random Processes”, IEEE TRANSACTIONS

ON SIGNAL PROCESSING, Vol.46, No.5, MAY 1998.