Ramp-up階段新舊製程混製生產時的投料批量決策模式

Abstract

半導體產業必須持續提升製程技術以增加競爭力。經由技術不斷地改良與創新，可提升每片晶圓的利用率與可用率。因線寬越小的製程每片晶圓可切割出更多的晶片，降低每顆晶片所攤提的晶圓成本，半導體晶圓製造廠持續推出縮小線寬的製程以增加其競爭力。然而，新製程導入初期因為機台的差異，造成良率偏低且不穩定。若新製程投料批量太多則會造成過多晶圓成本的浪費，反之，若投料量太少則會讓新製程學習緩慢且拉長ramp-up之時間。本研究假設良率學習受制於新製程的累積投料量，在瓶頸機台產能與滿足客戶需求限制下，以最小化投料晶圓成本及存貨成本為目標，建構晶圓新、舊製程的投料批量決策模式。新製程良率學習模式含自我學習與實驗學習，此兩因素受限於新製程的晶圓投料批量。經由案例的敏感度分析，了解良率學習模式參數、新製程的初始良率與新舊製程晶圓成本的差距，對新舊製程投料批量之影響及其管理意涵。

Continuous production process improvement is an important issue in semiconductor companies. The smaller line width of production process is, the more numbers of chip per wafer are. Therefore semiconductor companies continuously introduce new production processes to reduce line width that cuts the chip unit cost down. However, semiconductor companies face with rapid product lifecycles and competitive pressure. A new production process is frequently introduced into a production line when it is ill understood. The yield of new production process is very low and unstable that causes a lot of wafer waste. For meeting customer demands, current and new production processes are frequently existed simultaneously in production line at production ramp-up stage. In this study we assume yield learning of new production process during ramp-up stage is determined by two factors. One is learning-by-doing, and the other is learning-by-experiment. We suppose the two factors are defined by the number of released lots of new production process. In this study we consider the constraints of capacity and customer demands we formulate a mathematical model to determine the released lots of new and current production for minimize raw wafer costs and holding cost. We also perform sensitivity analysis of parameters of learning function and per raw wafer cost for some management insights.