半導體產業大型廠房之設施規劃

Abstract

半導體產業由於廠房和機台取得的前置時間都很長，為了快速反應市場需求的變動，晶圓廠逐漸採取「預建大型廠房」的產能擴充策略。亦即先建一個大型廠房，然後再根據市場需求，逐步添購機台。此等大型廠房約當是一般晶圓廠的兩倍大。面對如此大的廠房空間，會產生一些設施規劃的決策問題：到底要當一個廠來規劃，還是當成兩個廠來規劃？如果是當成兩個小廠，每個小廠內部應該如何規劃較佳？兩個小廠之間的運輸應該如何規劃？決策問題兼具學術與實務價值，但是過去卻甚少探討。本計畫提出三種設施規劃方案，等候網路模式進行績效評比，期能幫助管理者在各種不同的產能需求情境下，選出最適的廠房空間規劃模式；為了公正進行績效評比，我們逐一探討各方案的最佳運作模式；也探討每一設施規劃方案適用的情境。本研究擬以等候網路模式進行績效評估，擬以數學規劃結合基因演算法求解廠區內和廠區間的最佳營運模式，並擬以資料挖況技術找出判斷各方案是否適用的簡易決策規則。

In semiconductor industry, the lead times for establishing a fab building and the acquisition of equipments are quite long. To quickly respond volatile market demand, semiconductor companies are adopting a 「large-fab-space」 policy to fast expand fab capacity. That is, a large scale fab building is established in advance, and equipments are progressively moved in according to the pace of market demand. Such a large-scale fab may be two times the size of a typical wafer fab. This characteristic raises some new research issues in facility layout. Should such a large-scale fab be designed as one fab or two fabs? What is the best layout among two-fab alternatives? How to make two fabs effectively support each other in capacity? How to effectively design material transportation system for each facility layout alternatives? What are the capacity upper bounds for adopting a particular facility layout alternative? This research proposes three facility layout alternatives and aims to develop methods to answer the above questions. We attempt to use queueing network models and discrete event simulation techniques to evaluate the performance of each facility layout alternative. Mathematical program and meta-heuristic techniques are considered in addressing the issues of inter-fab capacity support. Data mining techniques are also considered in deriving decision rules for selecting facility layout alternatives.