晶圓製造廠兩階段投料法則之構建

Abstract

由於半導體生產設備的昂貴，因此大多數的晶圓製造廠為了提高生產設備 的利用率，莫不採用機器一有空閒即下料的生產策略。然而這樣的下料策 略卻會造成在製品量過高，生產週期時間延長的情形，以致影響交貨時效 ，喪失市場競爭優勢。瓶頸資源是整個系統產出績效的限制，所以考慮瓶 頸資源是必須的。再者，傳統之產能規劃，均假設長期產能規劃若可行則 保證短期產能規劃可行。然而實際上並非如此，因此在本論文中將結合投 料法則與負荷推估，以確保投料計劃表的可行性。本論文所發展之兩階段 投料法則，即是在考慮瓶頸資源的條件下，分兩階段來進行投料。在第一 階段，經由使用者定義或吾人程式定義的瓶頸，透過避免飢餓法的概念， 並且考慮實際每日瓶頸之產能限制，決定是否下料。而於第二階段，吾人 則採用預視的方法，將通過第一階段可行的下料量，試行投料，觀察於未 來之生產週期時間內是否有超載的現象。若無超載，則將其排入投料計劃 表內；否則，不進行投料。經由實例驗證的結果顯示，本模式的確能構建 出一個產能上確實可行之投料計劃表，並使瓶頸資源上的負荷差異較其他 二法為小，因此對於各產品之生產週期時間有縮短的功效。 Owing to high capital costs of equipment, most wafer fabs tend to release a lot as machines are available to maintain high equipment utilization. However, this has resulted in high WIP and long cycle time. Corresponding with failing to meet the on- time delivery and losing the marketing competitive edge.The system throughput of the wafer fabs is limited by the bottleneck machines. Therefore it is a must to take bottleneck machine into account while constructing the lot-releasing timetable. Moreover, traditional capacity planning assumes that the feasibility of long-term capacity plan guarantees the feasibility of short-term capacity plan. However, it is not so ideal in fact. Therefore it will take both lot-releasing policy and load-traversing into account in this paper to ensure the feasibility of lot-releasing timetable.Under the consideration of bottleneck machines, this paper will develop a two-stage order releasing algorithm by two stage. In the first stage, the bottleneck is defined either by user or the program. It will determine whether releasing lots or not via the concept of Starvation Avoidance and the daily capacity limit of bottleneck. In the second stage, this paper takes a method to provide visibility for future load. It checks whether future load is overloaded or not in the product*s cycle time. If not, the lot will be scheduled in the lot-releasing timetable; otherwise, the lot will not release.The experience studies show that the lot- releasing timetable constructed by the two-stage order releasing algorithm is feasible. It makes the difference of load on bottleneck narrow down compete with the other two methods. Therefore it has the effect of reducing cycle time of each product.