任何的生產系統都需績效指標做為生產能力的衡量與訂定其生產目標,但績效間會相互影響且具有交互作用,故無法專對於一個績效指標做管理。藉由作業曲線可讓管理者以目前系統狀態,找出一個績效指標與另一個績效指標的數量關係,做為績效間取捨的參考依據。對於半導體產業來說,生產週期時間的管控重點在於生產週期時間的分佈,在實務中,常以生產週期時間95%或98%尾值(即CT95%或CT98%)為目標,而文獻中作業曲線皆以生產週期時間之平均值做描述,本研究以一家半導體封裝廠(A公司)為個案,建構其模擬模型,發展具預估生產週期時間分佈之作業曲線,並探討影響其週期時間分佈的因素。 本研究以模擬之抽樣方式,在變異性不改變的情況下,以機率分配專對於單一樣本值來做預估,也證實模擬方式同樣也可達到作業曲線的效果。由實驗結果發現,變異性對於作業曲線的建構有其限制在,所以由績效關係圖做區間估計,來達到預估的效果。以績效關係圖分析影響CT分佈之因素,說明了不同投料量大小對於CT分佈影響之程度,也證實A公司現行之派工法則即為最適用的法則。
Every production system needs performance indicators to measure its productivity and make its objectives, but there are many effects and interactions among performance indicators. It is difficult to only aim at one performance indicator. Relying on Operating curve allows the manager to quantitatively trade-off one indicator versus another for a given system’s current operating state. Cycle time distribution is the center on the management of cycle time for semiconductor manufacturing. A commonly used measure of cycle time distribution in the practice is 95%-tail or 98%-tail cycle time (namely CT95% or CT98%). All of above cycle time is viewed as mean value in the review of operating curve. A case of semiconductor assembly factory (called factory A) is referred to build its simulation model and developed its operating curve could predict the cycle time distribution. Afterward, there are discussions about the factors, which effect cycle time distribution. This study made use of simulation to reach sampling distribution that estimate value from one sample in the constant variation situation. Later, the simulation is verified the function which operating curve could do. In experiment results, there is the limit of building the operating curve because of the variation. In order to achieve the purpose of prediction, interval estimate is used from the performance relation coordinates. By the way, the coordinates are used to analyze the factors, which effect cycle time distribution. It reveled that different quantity of the lot start effect the level of cycle time distribution and verified the dispatching rule factory A used as the most suitable rule.