張時中臺灣大學：電機工程學研究所徐仕杰Hsu, Shih-JieShih-JieHsu2007-11-262018-07-062007-11-262018-07-062007http://ntur.lib.ntu.edu.tw//handle/246246/53129整合與重複利用現有資源以達到快速良率提升是製造廠營運一個很重要的議題。針對這個議題，我們選擇良率分析流程(Yield Analysis Procedure, YAP)與其在工程資料分析(Engineering Data Analysis, EDA)系統上的實現，並以半導體製造為例，為我們的研究範疇。本論文透過對良率分析流程的現狀分析，定義關鍵問題，並討論如何藉由知識的整合與重複利用，來達到有效良率分析流程的規劃與實現。 為了促進有效良率分析流程，本論文從工程知識鏈的角度，針對其中需要的知識服務(Knowledge Service)與流程服務(Process Service)做深入的設計與分析。知識服務將良率分析流程知識與工程資料分析系統知識儲存在良率分析流程導引系統上，並提供適當的良率分析知識給有需要的工程師。知識服務設計包含了兩項機制。第一，設計一個統一良率分析資源模型，提供一致性的良率分析流程知識與工程資料分析系統知識表達方法。我們應用語義網路(semantic networks)，將重要的良率分析概念與概念之間的關係用節點(Node)與連結(Link)儲存在統一良率分析資源模型裡。並用這些良率分析概念描述所有的良率分析流程知識與工程資料分析系統知識。一致性的知識表達方法對於知識整合與重複利用非常重要，可以促進整合良率分析流程規劃與其在工程資料分析系統上的實現，。第二，利用工程資料分析系統的操作記錄，分析、萃取出良率分析概念間新的關連性。第一步必須先藉由資料探勘找出工程資料分析程式之間的關連性。第二步，因為我們已經預先用統一良率分析資源模型裡的良率分析概念來描述所有的工程資料分析程式，因此可以藉由工程資料分析程式之間的關連性進一步的萃取出良率分析概念之間的關連性，促使統一良率分析資源模型裡的良率分析知識更加完整。 流程服務提供多步驟流程規劃，幫助工程師達到快速有效的良率分析決策。流程服務設計包含了三項機制。第一，將工程師良率分析的想法對應（map）到適用的工程資料分析程式，此機制幫助工程師快速找到可用的工程資料分析程式，提高工程資料分析程式的使用率，並促進有效率的將良率分析流程實現在工程資料分析系統上。第二，根據統一資源模型內的良率分析流程知識與規劃出所有的良率分析流程，此機制可產生一個符合工程師想法的良率分析流程規劃樹，提供多步驟的良率分析規劃，促進工程師對良率分析流程做全面性的考量與決策。第三，利用資訊綜合報表(Composite Report)萃取良率分析流程中，工程師的狀況評估知識。因為狀況評估知識隱藏在工程師頭腦裡，很難被自動萃取，因此我們設計了資訊綜合報表幫助工程師整理在分析流程中發現的資訊，在良率分析流程規劃樹上找出最佳路徑，利用此報表帶來的好處，鼓勵工程師每次執行完工程資料分析程式時，在系統上留下發現的資訊。這些工程師留下的資訊可以促進萃取、分析狀況評估知識，向有效良率分析流程跨進一大步。Resource integration and reuse for the rapid yield ramp-up is a critical concern to the business strategies in manufacturing industries. Aiming at this issue, we select the yield analysis procedure (YAP) and its realization on engineering data analysis (EDA) system in a semiconductor fab as the problem conveyer. Through the study of current practices, three YAP problems are identified: low reusability of the YAP related knowledge, low quality of the one-step YAP decision, and low efficiency to realize the YAP plan on EDA system. The focus of this thesis is then on the knowledge integration and reuse for effective YAP planning and realization. In order to enabling the effective YAP, this thesis focuses on the detail design and analysis of knowledge and process service from the perspective of knowledge chain integration. Knowledge service is to model both YAP and EDA knowledge at YAP guiding system and provide on-demand knowledge to engineers. The design of knowledge service consists of two tasks. First, design unified resource modeling for unified representation of YAP and EDA system knowledge. To achieve flexibility and scalability, semantic network is selected as the knowledge representation method. The critical yield analysis concepts and their relationships are modeled as various types of nodes and links in unified resource model. These yield analysis concepts can describe most YAP and EDA knowledge. Unified knowledge representation is critical for knowledge integration and reuse. The unified resource modeling enables the integration of YAP purpose planning and realization on EDA system. Second, extract new relationships among the yield analysis concepts in unified resource model via analyzing EDA system usage log. Two steps in this mechanism: 1. Extract EDA function connectivity through usage log mining. 2. According to the EDA function connectivity, the relationship of the yield analysis concept can be extracted because all EDA functions are described by the yield analysis concepts. Process service is to support the multi-step-ahead process planning for proactive YAP decision makings. The design of process service consists of three successive mechanisms. The first mechanism, the easiest task, maps the purpose of a single YAP activity to corresponding EDA functions. Engineers can find required EDA function quickly and the EDA function usage rate can be improved by this mechanism. The EDA function enables the efficient YAP realization on EDA system. The second mechanism generates the full range of YAP planning tree based on the YAP knowledge in unified knowledge model. The YAP planning tree supports the multi-step-ahead planning and achieves seamless integration of engineer’s yield analysis thinking and EDA system execution. This mechanism enable the global YAP decision makings. The final mechanism, the most difficult task, is to extract situation assessment knowledge by composite report. In current practices, the situation assessment knowledge is implicit in engineer’s mind and hard to be extracted. To cope with the problem, we further design the composite report to help engineers summarize the findings in YAP and analyze the critical path on YAP planning tree. Utilize the benefits bring from composite report to encourage engineers keep their finding in system immediately after each EDA function execution. These engineer’s findings collected by composite report enable the effective extraction of situation assessment knowledge to remove the biggest obstacle to the effective YAP generation. To achieve effective YAP planning and realization, this thesis designs five enabling mechanisms from the knowledge and process service perspective. The goal of this thesis is to enable the evolutions of YAP optimization based on these mechanisms.口試委員會審定書 i 誌謝 ii 中文摘要 iii 英文摘要 v Chapter 1 Introduction 1 1.1 Motivation: Resource Integration and Reuse for Rapid Yield Ramp-Up 1 1.2 Problem Conveyer: Effective Yield Analysis Procedure (YAP) 3 1.3 Literature Survey 6 1.4 Research Scope and Methodology 12 Chapter 2 Study of the Mechanism for Enabling Effective YAP 17 2.1 Different Perspectives on YAP 17 2.2 What the Engineers Do in YAP 24 2.3 Problems and Direct Requirements of YAP in Current Practices 28 2.4 Enabling Mechanism for Effective Yield Analysis Procedure 35 Chapter 3 Knowledge Service- YAP Resource Modeling 39 3.1 Knowledge representation of Semantic Networks 39 3.2 YAP Unified Resource Modeling via Semantic Networks 42 3.2.1 Node Classification Modeling 42 3.2.2 Node Relationship Modeling 45 3.3 YAP Activity Purpose & EDA Function Modeling 50 3.3.1 YAP activity Purpose Modeling 50 3.3.2 EDA Function Purpose Modeling 55 3.4 Knowledge Learning Mechanism Enabled by YAP Resource Modeling 56 3.4.1 Critical Path Learning via Engineer Behavior Mining 56 3.4.2 Situation Knowledge Extraction via Composite Report 61 Chapter 4 Process Service- YAP Planning & Realization 66 4.1 YAP Planning Tree 67 4.2 On-Demand EDA Function via Purpose Mapping 69 4.2.1 Algorithm Design 70 4.4.2 An Example 71 4.3 YAP Planning Tree via Multi-Step-Ahead Planning 73 4.3.1 Basic Idea of Multi-Dimensional Planning Action 73 4.3.2 Idea of 1st Generate & 2nd Generation Planning Algorithm 76 4.3.3 1st G Planning Algorithm Design 79 4.3.4 2nd G Planning Algorithm Design 86 4.3.5 Planning Algorithm Evolution 90 Chapter 5 Conclusions and Future Research 94 Reference 972226761 bytesapplication/pdfen-US知識工程知識鏈知識整合與再利用良率分析流程knowledge engineeringknowledge chainresource integration and reuseyield analysis procedurethesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53129/1/ntu-96-R94921066-1.pdf