請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/40625
標題: | 良率分析流程在工程意圖層次之知識建模 Modeling of Yield Analysis Procedure (YAP) Knowledge at Engineering Purpose Level |
作者: | Wei-Chuan Lin 林韋銓 |
指導教授: | 張時中 |
關鍵字: | 知識工程,良率分析流程,知識建模,工程意圖, Knowledge Engineering,Yield Analysis Procedure(YAP),Knowledge Modeling,Engineering Purpose, |
出版年 : | 2008 |
學位: | 碩士 |
摘要: | 如何快速提昇良率一直是半導體產業最關心的議題,根據此議題本研究主要為探討良率分析流程(Yield Analysis Procedure, YAP)在工程意圖層次(Engineering purpose level)之知識建模(Knowledge modeling),以促進良率分析流程知識的分享與重複使用進而提高良率分析流程的績效。由於先前相關的研究在良率分析流程的建模上並未考量工程師執行良率分析的目的(Goal)以及在良率分析流程中所發生的不同情況(Situation),因此本論文的核心著重在於如何建立一個符合工程師思維的良率分析流程知識模型,來支持以條件為基礎的良率分析流程情況判斷(Situation assessment)和意圖路徑控制(Routing controls),進而達成不同的良率分析流程目的。
本研究所提出的良率分析流程建模架構,主要採用知識取得與文件結構化(Knowledge Acquisition and Documentation Structuring, KADS) 的方法並且結合問題解決方法(Problem Solving Methods, PSMs)與本體論工程(Ontology engineering)等方法。提出的模型架構分為三個不同的層次:意圖層(Purpose layer)、意圖連結規則層(Purpose-Link Rule layer)以及知識資源層(Knowledge Resource layer)。 在意圖層的建模上,我們提出了複合意圖(Composite purpose)的觀念與徐仕杰(2007)所提出的原子意圖(Atomic purpose)作區別。複合意圖不僅包括了本身該有的輸入以及輸出的資料描述,還包括啟動該複合意圖的起始條件(Pre-condition)與結束條件(Terminating-condition)。藉由複合意圖的起始條件與良率分析流程的狀態比對,我們可以決定該複合意圖是否被驅動。反之,複合意圖的結束條件則可以用來判斷工程師的目的是否被達成。此外,帶有特定目的的複合意圖也可以更進一步的被拆解成幾個子意圖與相對應的子目的,而本研究進一步提出意圖連結規則(Purpose-link rule)來表達在不同情況下各個子意圖之間的關係。 為了減少一般自然語言所可能帶來的模稜兩可和誤解,我們提出了統一資源模型(Unified resource modeling)來描述所有在良率分析流程中所介紹過的分析意圖、意圖連結規則以及不同的良率分析情況,以利由電腦執行運算。為了希望所建立的統一資源模型可以涵蓋較廣範圍的良率分析流程知識,除了徐仕杰(2007)所提出的半導體領域資料的知識外,我們所建立的統一資源模型還包括了在良率分析流程中所使用的統計工具知識以及一般資料的知識。 我們設計了一個機制來有效地整合與串連所提出的三層結構模型,其中包含五個重要概念:輸出輸入資料集(I/O Dataset)、輸出輸入徵狀(I/O Symptom)、良率分析情況(YAP Situation)、良率分析意圖(YAP Purpose)、良率分析意圖連結規則(YAP Purpose-Link Rule),與四個串連整合步驟:(1)首先先將良率分析工具之輸入輸出資料集轉換成可一般理解之徵狀,(2)然後透過轉換後的徵狀紀錄在所提出的良率分析流程狀態內以敘述工程師目前所掌握的分析狀況,(3)接著藉由所定義之良率分析意圖找出該對應之進入意圖前提條件與結束條件,(4)最後則根據不同良率分析流程狀態的內容靠良率分析意圖連結規則動態規劃出下一個合適的分析意圖。 根據所提出的三層結構模型,我們採用由上而下的(top-down)、反覆的(iterative)以及遞增的(incremental)建模方法來從實務資訊中取得良率分析流程的知識。由上而下的建模方法是為了要確保在意圖層次的良率分析流程知識建模可以符合工程師實際的想法,另一方面也是為了避免良率分析流程的知識建模僅只在系統相依的操作層次。在良率分析流程的知識建模上,我們首先參考半導體廠內良率分析的標準作業程序(Yield analysis standard operating procedure)作為初步的知識來源,而對於後續的知識建模我們則是藉由半導體廠內的實際案例分析,以及來自經驗豐富的工程師寶貴建議,反覆地擴增現有的知識內容,同時也進一步的測試我們所提出的結構模型的彈性與適用性。 Recognizing the importance of rapid yield ramp-up in semiconductor manufacturing industry, this thesis studies the modeling of yield analysis procedure (YAP) knowledge at purpose level to enable knowledge sharing and reuse among engineers for better YAP performance. As compared to former researches without considering engineer’s analysis goals and various situations involved in YAP, this thesis aims at modeling the YAP knowledge to support YAP situation assessments and routing controls for resembling engineer’s behaviors on reaching specific goals. A framework applying the KADS (Knowledge Acquisition and Documentation Structuring) methodology and integrating both PSMs (Problem Solving Methods) and ontology engineering techniques for modeling YAP knowledge is proposed. The proposed framework contains three layers: Purpose, Purpose-Link Rule, and Knowledge Resource. For modeling of analysis purposes, the concept of composite purpose is proposed to differentiate the atomic purpose modeled by former researches. In specific, a composite purpose is not only described by its input/output data but also characterized by its pre-condition and terminating-condition. A pre-condition specifies the YAP status that a composite purpose should be triggered while a terminating-condition specifies the goal status that a composite purpose should be achieved. A composite purpose with specific goal will be further decomposed into sub-purposes and corresponding sub-goals. To facilitate the dynamic control of sub-purpose routings, the associations between current YAP situations and succeeding sub-purposes are represented as purpose-link rules in IF-THEN format. To reduce the ambiguity and misunderstanding in nature languages, all the wordings (resources) involved in YAP purposes, purpose-link rules, and situations (status) are further modeled and integrated as a unified resource structure so that they can be manipulated by computer programs. In the unified resource structure, in addition to the knowledge of domain data names specified by former researches, both the key words about applying statistics tools and common knowledge are formally specified for the comprehensive modeling of YAP knowledge. Besides, a mechanism with four steps is designed to integrate the proposed three layers, which contains five key concepts of YAP knowledge from function level to purpose level: I/O Dataset, I/O Symptom, YAP Situation, YAP Purpose and YAP Purpose-Link Rule. And the four steps are used to link them together, which are: (1) I/O Symptom Transfer used to transfer I/O dataset of analysis tool into comprehensible I/O symptom, (2) YAP Situation Formalization used to describe YAP status that indicates what engineer has confirmed or suspect based on the current finding symptoms, (3) YAP Purpose Definition directs the course of pre-condition toward a specific goal, and (4) YAP Purpose-Link Rule Definition represents the association between the purposes with respect to various situations. Based on the proposed three-layered framework, the acquisition of YAP knowledge adopts top-down, iterative and incremental modeling approach. The top-down modeling is to align of YAP knowledge model with engineer’s thinking at purpose level while avoiding to be mired in modeling at system-dependent operational level. In specific, the top-down yield analysis Standard Operating Procedure (SOP) from a foundry serves as the knowledge source for the first modeling iteration. With real case studies and additional inputs from veteran engineers, follow-up modeling iterations are then incrementally conducted to test the flexibility and applicability of the proposed framework for YAP knowledge modeling. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/40625 |
全文授權: | 有償授權 |
顯示於系所單位: | 工業工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-97-1.pdf 目前未授權公開取用 | 3.91 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。