晶圓代工廠之機台當機影響即時估算法

Abstract

不預期的機台當機會使晶圓代工廠現場運作遲緩，導致產品產出減少及生產週期時間拉長。當機台發生當機事件時，除了單一機台本身產能受損外，對晶圓廠的產出績效亦有不良的影響，如何以系統觀來衡量其不良的影響，以決定機台修復順序，乃是本研究探討之主題所在。 本研究的目的是發展一快速估算法，在機台發生當機事件時，能夠即時估算未來每個時間區段瓶頸加工站的產出量變化；若有多部機台同時當機，利用本估算法分別求得各種修復順序下的瓶頸產出量，產出量較高的即為較佳的修復順序決策。 經實例驗證顯示，當有5部機台當機時，共有120（5！）種修復順序，這120種修復順序的瓶頸產出量用ASAP模擬器模擬需時3小時，用本研究所提之演算法所需計算時間不到2分鐘，由此演算法可以快速決定當機機台的修復順序。而且，提供的最佳機台修復順序決策與ASAP模擬器所得之最佳修復順序決策之瓶頸產出量僅差距0.06%。

This research aims to measure the impact of machine down in an IC foundry and develop a quick method to determine the priority of repair when several machines are down and waiting for repair. An algorithm is proposed and implemented, which can quickly evaluate the throughput changes of the bottleneck machine when several machines are down and repaired in different sequences. For a case of 5 machines down, there are 120 (5!) alternatives of machine repair sequence. The proposed algorithm implemented on a PC requires only two minutes to select the best repair sequence, while it requires 3 hours by a proprietary simulation software, AutoSimulation-AP Controller. The bottleneck throughput calculated by the algorithm is only 0.06 % difference comparing to that of the ASAP simulation result.