晶圓製造廠機台刪減機制之設計

Abstract

晶圓製造業者常面臨到刪減機台的決策：例如機台變賣或是出借。然而，不同的機台從生產系統中抽離時，對系統的生產績效將會產生不同程度影響。而且，不同產品組合與產出目標之下，對於工作站亦會造成不同程度的負荷。因此，在給定未來可能發生的數種產品需求組合與產出目標下，本文以維持特定的產出目標以及訂單達交率水準為目標，設定晶圓製造廠機台刪減機制，用以規劃最佳機台組合，來因應刪減機台的決策。 依據限制理論的概念，系統應該要避免瓶頸的漂移。因此，在給定某一產品組合下，本機制首先估算每個工作站的產能負荷來決定其最小機台需求數。再者，考量機台負荷以及工件批物流流動的情形，運用區段基礎式週期時間估算法(BBCT)來評估刪減每個工作站之一台機台後對生產週期時間的影響。接著選取對生產週期時間影響最小的機台加以刪減。重複上述評估與選取之步驟，一直到產品生產週期時間上升到預定達交率下所設定的生產週期時間上界為止。最後，重複上述流程，分別針對每個未來可能發生的產品組合比例與產出目標，決定其機台刪減的組合後，進行彙整以得到最終可刪減的機台組合。 模擬結果顯示，本文的機台刪減機制，在產出目標方面，不但使刪減機台後的系統其產出量可以達到預計產出目標，在達交率方面，亦可控制達交率於可接受的範圍內。

Wafer manufactures usually face a strategy of pruning machine units for the need of selling or leasing machines. However, different machine type and corresponding units are pruned from the shop floor will cause different effects on the production performance. Moreover, various product mix and throughput target will also cause different effects on loading of each individual workstation. Therefore, considering the variation of product mix and throughtput target, this thesis aims to design the machine units pruning mechanism for maintaining system throughput target and delivery rate. Based on the concept of Theory of Constrains (TOC), bottleneck wandering must be prevented. The mechanism first estimates the loading of each workstation under a given product mix and throughtput target to determine the least machine units needed. Second, the increment of the production cycle time as one machine unit of a specific workstation was pruned must be estimated. The block-based cycle time (BBCT) methodology, which considers the waiting time due to machine loading and congestion of production flow, is used. Next, one machine unit of the workstation which has the lowest effect on production cycle time is pruned. Repeat the above steps of evaluation, until the production cycle time has reached the upper limit for achieving the target of delivery rate. At last, repeat the above procedure for each possible product mix and throughtput target respectively. Then, all the results are collected so as to generate the machine set being pruneable. Simulation experiments reveal that with machine units pruning mechanism, not only the system can have its throughput reaching the planned target, but also can control the delivery rate within acceptable range.