決策樹應用於薄膜玻璃濺鍍製程良率分析之研究

機構典藏 > College of Engineering > Graduate Institute & Department of Computer Science and Information Engineering > Thesis > Item 987654321/34970

Please use this identifier to cite or link to this item: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/34970

Title:	決策樹應用於薄膜玻璃濺鍍製程良率分析之研究
Other Titles:	An analysis of the applying decision trees to the process yield of thin layer glass sputtering
Authors:	王茂年;Wang, Maw-nian
Contributors:	淡江大學資訊工程學系碩士在職專班林丕靜;Lin, Nancy Pei-ching
Keywords:	決策樹;資料採礦;迴歸樹;良率;Decision tree;data mining;Regression Tree;Yield
Date:	2007
Issue Date:	2010-01-11 05:50:13 (UTC+8)
Abstract:	近年來，光電產業已成為我國高科技產業的重點工業。為提升獲利能力，從建廠之初，廠商無不希望快速提升製程技術、大幅縮短試產時程及早進入量產；除此之外，工廠亦須在「大量少樣」或「少量多樣」的生產模式中抉擇，以建立最佳生產模式。為建立成功的獲利模型，部分工廠採取「少量多樣」的生產策略，俾將有限資源集中投入工廠量產。處於「少量多樣」的生產環境下，很難在短時間找出良率決策規則，並形成良率決策規則資料庫。本研究在提供廠商一個簡明之良率改善架構的製程模型，以有效提升良率並控制良率的變異。傳統應用決策樹以改善製程良率之研究，大多利用批次(Lot)資料輸入到迴歸樹(Regression Trees)進行分析，但此方法較難在資料有限且良率變異較大的情況下，較難找出製程參數的最佳區間。本研究乃改採生產管理系統(Manufacturing Execution System，MES)與統計製程管制(Statistical Process Control，SPC)中所收集之資料，將原始批號資料轉換成實際進入每一製程的玻璃片資料數量，再分別將每一個製程參數資料各自利用決策樹進行獨立分析，並決定良率的製程參數範圍，以作為製程工程師解決問題的參考依據，進而提升工廠製程良率。本研究係以台南科學園區內某光電廠之薄膜玻璃濺鍍製程案例為實證，檢驗本研究架構之效度，從研究結果顯示：利用本研究所採行之分析方法，除可分別定義較佳良率之製程參數其正向與反向條件外，並可有效協助製程工程師提高製程良率。實作顯示：該光電廠之製程良率已大幅改善，良率較以往增加約20~30%。 For the past few years, Photonics has become the key technology among many Hi-Tec industries in our country. In order to increase profits, Photonics manufacturers all hope to fast improve their manufacture technique, which can save the trial manufacturing process on a great scale and reach the mass production stage earlier. They also need to determine production strategies such as mass quantity but few varieties or great varieties but less quantity to see which the best production mode is. A few manufacturers would choose great varieties but less quantity production strategy to set up a successful profit model, pouring all the limited resources into mass production line. The purpose of the research is to provide manufacturers a simplified production model of yield improvement under the great varieties but less quantity production circumstances and to effectively improve and control yield variation. Many conventional study of applying classification trees to improve process yield would conduct the analysis by inputting batch data to Regression Trees. However, if the data is scarce and yield variation is too big, this method can not effectively distinguish which batch and its quantity. It is also hard to find out which the best range is for process parameter. Therefore, this research chooses to adopt the data collected by MES(Manufacturing Execution System) and SPC(Statistical Process Control) and transform the original batch data into actual data. This research is based on a Photonics manufacturer located in Southern Taiwan Science Park. The research takes its thin layer glass sputtering for example. Moreover, it analyzed every process parameter by decision tree and decided the range of acceptable process parameters of yield in order to provide insights for yield enhancement and lights of problem-solving for process engineers. This case also examined the validity of this manufacturing process model. This research shows that the process yield of the Photonics manufacturer has improved a lot by 20 ~ 30 % compared with its past record.
Appears in Collections:	[Graduate Institute & Department of Computer Science and Information Engineering] Thesis

Files in This Item:

File	Size	Format
	0Kb	Unknown	323	View/Open

Loading...