晶圓代工廠生產週期時間之預測─以M公司為例

Abstract

晶圓代工製造廠由於是一種接單式生產(make to order)，因此十分重視客戶的滿意度，而生產週期時間(cycle time, CT)是該產業影響客戶滿意度的重要績效指標之一。但是晶圓代工產業有明顯的淡旺季之別，產品數量與產品組合會隨著淡旺季的交替而有劇烈的變動，因此如何提供客戶一個準確的CT值來滿足客戶的需求與增加客戶的滿意度，一直是晶圓代工廠的一項重要課題。然而CT值的預估與管理往往只是依賴員工的經驗法則來訂定，缺乏有系統的方法或數學模型來預測CT值，以作為管理CT值之依據。過去許多研究指出生產周轉率(turn rate, TR)為CT的先行指標，TR指標表示工廠內晶圓進行之速度，此指標值越高表示生產速度就越快，CT就越短。在生產完成後才發現CT值表現不佳時，一般情況下，管理者能做的只是亡羊補牢。因此若能藉由每天所能收集到的TR值來評估CT值，便能即時得知目前在製品(work in process, WIP)的CT績效與預測未來的CT值表現。本研究首先找出影響TR與CT的重要因子，再應用逐步迴歸分析法先預測TR值，接著以此TR預測值依照其與CT的關係式來預測CT值。這樣除了可以準確地預測CT值外，也能讓管理者藉由每天工廠的TR值表現，迅速得知WIP之CT值的變化，以利安排各項生產活動，進而確保達到生產目標。本研究最後以一個晶圓代工製造廠之實例來說明本研究方法確實有效。

The manufacturing process of IC foundries is make-to-order, so it is important to have customer’s satisfaction. The customer’s satisfaction is influenced by cycle time which is one of the important performance indexes in IC industry. However, there is an obvious business cycle in IC industry, consequently the orders and product mix always have violent changes during low or peak seasons. Hence, accurately predicting cycle time can improve customer’s satisfaction. It becomes one of the most important challenges of IC foundries. Predicting cycle time is usually dependent on the experienced engineers. There is no mathematical method for forecasting the cycle time. Many studies have pointed out that turn rate is a leading indicator of cycle time and it is defined as the number of operations that have been done for the wafer per unit time. The higher turn rate indicates production speed and the shorter cycle time will be. In general, when the cycle time was deviated from the target after finishing producing products, the customer will be informed that the order will be delayed. Thus, the turn rate can be utilized to forecast and monitor the cycle time of work-in-process(WIP) simultaneously. This study first finds out which factors will influence turn rate and cycle time significantly and the stepwise regression is then employed to forecast the turn rate. The forecasted value of turn rate is utilized to forecast the cycle time. Managers can use the turn rate data to know the cycle time of WIP and arrange all activities of production. A real case from an IC foundry is used to illustrate the effectiveness of the proposed method.