Development of an Intelligent Scheduling System for a Semiconductor Foundry Fab

由於晶圓廠就如同一個大型的彈性製造系統且晶圓的製程相當複雜，因此派工問題對於生產控制扮演相當重要的角色。對於動態的派工系統而言有兩個關鍵性的議題主宰著彈性製造系統的效能；其中一個就是如何選出適當且關鍵的顯著系統資訊當作判斷系統狀況的指標，另外一個就是派工系統的設計方法-也就是所謂分類機器的設計。相對於單一個派工法則，如果我們能有效的依據一些有用的系統的資訊為當時的系統狀況選擇適合的派工法則並且建立一個動態的派工法則知識庫則會使得彈性製造系統的產出提高。為了提供彈性製造系統一個可以選取最佳系統資訊並且有高歸納力的分類器，本篇論文提出了一個自我組織的派工模型。為了達到上述的目標，本論文所提出的派工系統模型將會包含了模糊理論、基因演算法、還有類神經網路。除此之外，本研究更利用模擬軟體建立了一個彈性製造系統，並且在模擬實驗證明本論文所提出的動態派工模型相較於單一派工法則可以得到較優產出。With the selection of the real-time salient information of machines and parts and then a rule’s dispatching mechanism is built for the scheduling task, the dynamic scheduling rules would outperform static ones in a flexible manufacturing system (FMS). Scheduling plays an important role in the production control in a foundry fab, which can be seen as a huge FMS. For a dynamic scheduling system, two critical issues dominate the performance of a scheduled FMS, one is the selection of system attributes and the other is the design of the dispatching mechanism, namely, the classifier design. This thesis proposes a self-organizing scheduling model (SOSM) aiming to provide a FMS with the optimal attribute selection and high generalization, high-accuracy classifier. The proposed scheduling model combines several intelligent methods to achieve these goals, including fuzzy set theory, genetic algorithms, and neural networks. A typical FMS model is conducted in the experiments for the demonstration. Experimental results show that the SOSM outperforms the static dispatching rules under three different performance criteria.中文摘要 i Abstract ii Contents . iv List of Tables v List of Figures vi Chapter 1 Introduction 1 1.1 Motivation 1 1.2 Literature Survey 2 1.2.1 FMS 2 1.2.2 Dynamic Scheduling Problems of a FMS 3 1.3 Contributions 4 1.4 Thesis Organization 5 Chapter 2 Background Knowledge 6 2.1 Overview of Machine Learning and Intelligent Scheduling 6 2.1.1 Applying Machine Learning to Intelligent Scheduling 6 2.2 Attribute selection approaches 8 2.3 Support Vector Machine 10 2.3.1 Linearly separable SVM 10 2.3.2 Linearly nonseparable SVM 11 2.3.3 Nonlinear SVM 14 2.3.4 Multi-class Support Vector Machines 16 Chapter 3 Selecting Optimal System Attributes with GA-based FFEI 20 3.1 Introduction 20 3.1.1 Supervised Feature Mining (SFM) 21 3.1.2 Relative Importance and Weighting Factors 21 3.2 Fuzzy entropy based feature evaluation index 23 3.2.1 Minimizing the FFEI via GA 28 3.3 Experiment result of feature selection 29 3.3.1 Validity demonstration with classification accuracy 32 Chapter 4 Scheduler Design Using Support Vector Machines 34 4.1 Introduction 35 4.1.1 Scheduler training part 36 4.2 Construct the Scheduling Knowledge Base 38 4.3 Adjusting Dispatching Rule with Voting Strategy 41 4.3.1 One-against-one Multi-classification SVM 41 4.3.2 Voting Strategy 41 4.4 Preparing the Training Data 43 4.5 On-line Scheduling Part 43 Chapter 5 Experiments 45 5.1 Layout and Assumptions of FMS 45 5.1.1 Training Set Preparation 51 5.1.2 Testing Set Preparation 53 5.1.3 Data Scaling 53 5.2 Experimental Results 53 Chapter 6 Conclusion 67 6.1 Future work 6