晶圓製造廠光阻噴管配置問題之探討

Abstract

微影製程(黃光製程)是晶圓製造過程中重要的一環，微影製程主要有三大步驟，分 別為光阻塗佈、曝光以及顯影。其中，光阻塗佈除必須塗佈在晶圓上的光阻外，因該作 業特性，另有閒置、保養與沖洗成本發生。閒置成本是光阻噴管在機台上閒置一段時間 後，下次作業前需執行空噴灑的動作，以避免噴管頭產生結晶，造成不良品，多次空噴 灑將造成光阻的浪費，因此使用率較低的光阻噴管盡量不裝在機台上，以免造成光阻的 浪費。保養成本是每季更換機台上的光阻噴管時，則原先的噴管取下後必須浸泡在OK82 液中保養，稱為保養成本。新替換的光阻噴管從保養液取出後，須用欲替換的光阻沖洗 噴管後方可使用，此成本稱為沖洗成本。因此光阻噴管在機台各槽配置的組合對上述三 項成本有極大的影響。本研究將探討晶圓製造廠光阻噴管在機台上的配置問題。已知下 一季產品需求及本季噴管配置的情境下，考量機台、產品需求與光阻上、中及下層的相 關限制下，以總成本最小化為目標，建構光阻噴管配置決策模式。隨之，探討模式求解 時間與機台數及光阻種類的關係，及瞭解模式最佳解的一些特性，由此構建求解法則， 以縮小最佳解收尋空間，達成快速求解及該解有良好精確性的目的。最後由實際案例驗 證該決策模式及求解方法，以利業界實際應用。

In this study we investigate the photo-resist nozzle allocation problem. The costs of photo-resist wasted spraying, OK82 maintenance and to purge are determined by the allocation of photo-resist nozzles on the photo-resist spinners. Therefore photo-resist nozzle allocation is an important issue for semiconductor fab. Photo-resist wasted spraying cost is that a photo-resist nozzle should be sprayed before its next work for its idle time being larger than or equal to a given value. This spray is asked for quality reason and it is a waste of photo-resist. The cost due to wasted spraying as photo-resist wasted spraying cost. When a photo-resist nozzle is taken away from a photo-resist spinner, it should be maintained in the liquid of OK82. This cost is called OK82 cost. Before a photo-resist nozzle is replaced on a photo-resist spinner, it should be purged by the corresponding photo-resist. We call it as purged cost. In this study, given the following period product demands and the current period photo-resist nozzle allocation, and under the constraints of machines, product demands and among photo resist relationships, we formulate an integer programming ( IP) model for minimizing the total cost to determine the optimal photoresist nozzle allocation. The items of photo-resist wasted spraying cost, OK82 cost, and purged cost are considered in the objective function. Then we investigate the properties of optimal solution in the proposed IP model. Based on these properties we construct rules to reduce the solution space for the problem solving computational time. Finally real data are collected to verify the proposed model and the accuracy of the proposed problem solving approach.