- 學位論文
十二吋晶圓製造知識型懸掛式載具派工之研究
Knowledge-Based OHT Dispatching in 300mm Semiconductor Manufacturing
摘要
現行的自動化物料搬運系統(AMHS, Automatic Material Handling Systems)解決方案中，懸掛式搬運系統(OHT, Overhead Hoist Transport)是一個可以對達到在作業區內傳送自動化，並且達成十二吋晶圓廠內機台對機台直接傳送的可行技術。但由於受限於機構設置及無塵室使用率的限制，一個工作區內的懸掛式搬運系統通常架構成一部或多部搬運車(OHT vehicle)沿著單一迴路的軌道運行。本論文主要在求解懸掛式搬運系統之空搬運車的派車問題。 由於懸褂式搬運車的派車問題內含釵h不同的目標或需求，簡單的啟發式派車法則(heuristic rules)或是線性規劃(linear programming)無法滿足所有搬運車派車問題的需求。啟發式法則對於特定需求的派車問題很有效率，其所得到的結果也通常能符合決策者的期望。然而，這個方法不僅對於不同的派車需求不能全部滿足，也不保證能達到最佳化(Optimality)的目標。線性規劃方法當問題趨於複雜時，需耗費較長的計算時間。因此對較大型的問題而言，並不實際。 在本論文中, 我們提出知識型系統的概念，來解決工作區內懸掛式搬運車的派車問題。首先針對基本搬運工作樣式進行分析, 藉由列舉兩個搬運工作的所有樣式，我們可得到了求解兩個搬運工作的最佳派車法則。根據這些基本搬運工作樣式，我們可以延伸到三個或更多搬運工作的派車問題。經由每一次的問題延伸，可以得到新搬運工作樣式的最佳派車法則，並將此基本搬運工作樣式加入原有的基本搬運工作樣式中。在求解最佳派車法則的過程中, 必須根據先前推導出對較小問題的解法，應用到數學推導並融合是人類專家的經驗判斷。對較大規模問題的所有推導過程皆是離線處理完成。實際應用時，則可藉由電腦比對問題樣式和基本搬運工作樣式，來獲得解答。 我們採用概念圖(conceptual graphs)來作為基本搬運工作樣式的表示方式。將欲求解決之問題分解成較小的子問題，利用基本樣式進行比對，在疊代發展出新樣式推導法則。對於無法經由樣式比對來解決的問題，必須進一步分析或是直接由人類專家來解決。其解答可作為此知識型系統新的基本樣式。 應用PROLOG+CG開發工具，我們建構了一個知識型懸掛式派車系統。我們藉由兩輛車、兩個工作和三輛車、三個工作的派車問題，進行系統的可行性評估。 數值結果顯示所開發的知識型懸掛式派車系統可在數秒內解得懸褂式搬運車派車問題之解答。
並列摘要
Among the proposed automatic material handling systems (AMHS) solutions, OHT (Overhead Hoist Transport) is a promising technology to fulfill the requirements of transportation automation in intrabays and to realize tool-to-tool delivery in a 300mm fab. Limited by the hardware mechanisms, an intrabay OHT system is usually configured as a circular loop where one or more OHT vehicles run around the loop. OHT dispatching deals with the assignment of an empty OHT to a transport job There are usually multiple objectives or requirements to OHT dispatching. It is difficult to solve OHT dispatching problems with simple heuristic rules or with linear programming techniques for small-scaled problems. Heuristic rules are efficient in generating a solution and its results are much closer to those from human decision-makers. However, its optimality is not guaranteed and it is unable to fulfill the requirements of multiple objectives. The linear programming approach is more time-consuming and its applicability is limited by the problem sizes. We propose, in this thesis, a knowledge-based solution methodology to the OHT dispatching problems in an intrabay loop. This methodology starts with the analysis of atomic patterns of transport jobs. The optimal dispatching can be achieved by enumerating all the atomic patterns of two jobs. Based on these patterns, we can extend to the dispatching problems of three or more jobs. Each time when we extend the problem size, solutions to new job patterns are found and incorporated as the atomic patterns. During the derivation of solutions, either mathematic inductions or human judgments are used, based on the previously derived solutions to smaller sizes as well as on the human expertise to the problems. All the derivations of solutions are off-line conducted. During on-line applications, the solutions are generated immediately by matching the given job patterns to the atomic patterns. We adopt the conceptual graphs for knowledge representation of atomic patterns. Induction logics are developed to derive the new atomic patterns by decomposing the problem into smaller ones to which the existing atomic patterns can be applied. For problems that are new or without atomic patterns for their decomposed problems, further analysis or human problem-solvers are needed. The solutions to them render themselves new atomic patterns in our knowledge- based system. We implement a knowledge-based OHT dispatching system with PROLOG+CG. The dispatching problems of (2 vehicles, 2 jobs) and (3 vehicles, 3 jobs) pairs are tested to assess the feasibility of the proposed methodology. Numerical results demonstrate that this knowledge-based system generates a feasible solution in seconds.