關鍵鏈專案管理在半導體一元化代工服務之研究

Abstract

一元化代工服務（統包）在今日顯得日益重要，主要是因為它可以用來幫助積體電路設計公司所推出的新產品，在短時間內製造完成以便即早來搶奪市場佔有率。世界前三大晶圓製造廠—台灣積體電路公司，聯華電子公司以及特許半導體—全都有替他們的顧客提供此類的服務。雖然一元化代工服務能夠為這些積體電路設計公司提供較佳的服務，但是大量且煩人的作業卻因此移到這些晶圓製造廠上。他們必須以達交來滿足顧客，因此晶圓製造廠需要一個監控一元化代工服務訂單加工時間的機制來保護交期表現。 在本論文中，我們利用「限制理論」[6]與「關鍵鏈專案管理」[7]的觀念來建構半導體供應鏈模型與架構。我們應用「關鍵鏈專案管理」，其主要目的在於確保訂單能夠準時達交。然而「關鍵鏈專案管理」是否能夠套用在半導體一元化代工服務是我們首要研究的問題。「關鍵鏈專案管理」裡有兩項基本的觀念：「專案緩衝」與「緩衝管理」。前者可用來計算交期而後者當成監控的機制。「專案緩衝」對交期表現有很大的影響。所以去驗證「專案緩衝」的計算是否合理也是必須的。同樣地，Dr. Goldratt 隨意地將專案緩衝切成三等分也被許多學者質疑，因以我們也必須要驗證這一點。 我們以新竹科學園區一家知名半導體廠商的資料來作為研究的個案。我們以eM-Plant TM [43]來設計模擬實驗，並且驗證了(1)在我們的個案中，專案緩衝的計算可能是不正確的但是當緩衝管理加上Critical Ratio (CR)作為第二層判斷時，運作效果十分良好(2)我們所提出的方法，CCPM3+CR，跟CR相較之下，在同樣的情境之下有較優異的達交率(3)驗證了Dr. Goldratt隨意地將「專案緩衝」切成三等分是合理且可行的(4)當我們用FIFO所得到的平均加工時間做為交期的基準時。CCPM3+CR比CR得到較高的達交率與產出。

Nowadays, the one-stop shopping turnkey service, which facilitates time to market and helps grab the market share, is more important to the IC design houses. Although it may provide better services to IC design houses, laboring tasks are therefore shifting to those FABs. On time delivery is one of the key performance indices of turnkey services. Therefore, a mechanism to control cycle time of turnkey service order to protect due-date performance is necessary. We present a semiconductor supply chain model constructed under the concept of Theory Of Constraints (TOC) [6] as well as Critical Chain Project Management (CCPM) [7]. The objective of CCPM intends to improve on time delivery [7]. However, whether CCPM is applicable for the semiconductor turnkey services deserves to be studied. There are basically two major concepts in CCPM. They are “project buffer” and “buffer management.” The former is utilized to calculate the due date while the latter serves as the monitoring and control mechanism. The project buffer strongly influences the due-date performance. To verify the feasibility of the estimate of project buffer proposed by Dr. Goldratt is necessary. In the meanwhile, Dr. Goldratt’s arbitrarily divided the project buffer into three equal zones for buffer management [7], which has been criticized by many researchers, needs to be justified, too. A case with real world data from a foundry FAB in Hsinchu Science-Based Industrial Park in Taiwan is employed. We conduct experimental simulation using eM-Plant TM [43] and demonstrate that (1) the estimate of project buffer may be improper in this typical case while buffer management works well with critical ratio (CR) as a tie-break rule, (2) the concept to divide project buffer into 3 equal zones (CCPM3) is reasonable and workable in this typical case, (3) CCPM3+CR, our proposed method, yields superior average on-time-delivery percentage (AOTDP) than the legacy dispatching rule, CR, does, and (4) due date in this study is designated based on the average cycle time obtained with first in first out (FIFO) where 50% of AOTDP will be achieved. CCPM3+CR yields higher AOTDP and higher throughput (TPT) than CR does in this typical case.