以動態規劃模式改善半導體黃光區重疊控制程序之研究

Abstract

摘要 黃光製程是IC生產中最關鍵的一門技術，它決定了整個製程的先進程度，而光罩圖案轉移的優劣會直接影響後續製程的進行。在現行的半導體黃光製程中，由於不同機台、不同光罩(Mask)與晶圓本身都會造成不同層(Layer)間的光罩圖案疊合時產生位移和誤差，尤其隨著關鍵尺寸不斷縮小、晶圓尺寸變大以及光罩數目的增加，黃光重疊控制誤差容忍度也越來越嚴苛，當黃光製程的重疊控制誤差超過誤差容忍度則層間設計電路可能因為位移發生斷路、短路而無法通過電性測試影響良率，進而造成產品報廢。因此，重疊控制誤差控制-準確的濾除不確定及隨機因素並估測出精確的最佳化可調整參數來穩定地降低重疊控制誤差，是控制生產良率的重要指標，也因此，重疊控制(Overlay)量測 - 即針對晶圓上先前已有定義過的圖案，量測一個黃光圖案置於晶圓時的精準度，以避免重疊控制錯誤造成的元件故障，在黃光製程中相形重要。 在生產管理及許多工業工程領域的文獻探討中，發展出了許許多多不同的派工法則來滿足此種限制，諸如：同族(Family-based)派工法則、避免飢餓法(Starvation Avoidance, Glassey and Petrakian, SA)、FIFO…etc.但目前的方式都是被動地被單機限制下(Single Tool Constrain)所發展出來的解法，並無法真正解決問題。 本研究藉整合多個製程與製造的專家控制系統，主動地解決單機限制問題，並運用作業研究之動態規劃模式搭配系統化的邏輯思考方式來撰寫輔助軟體，實際建構出一套模式與系統，期真正能協助解決半導體公司所遇到的問題。

Abstract Photo is a critical process and technology in semiconductor manufacturing and it plays a role how intelligent the IC processing, in which the quality of mask pattern will have a great direct impact on downstream processing. During lithographing process, the motion and difference will occur in mask overlapping due to different tools, different masks and a wafer itself. However, the tolerance for mask overlapping has become more critical as the dimension of metal line shrines to be narrower, the wafer size becomes bigger and the mask layers increase. The circuit may be malfunction as the tolerance in overlapping is over. Therefore, error-overlapping control to move out uncertainty and stochastic risk factors is they key for fab yield by accurately estimate parameters. Overlapping measurement that is predefined pattern for wafer accuracy to avoid breakdown in IC device is very important in Photo. In many research for industrial engineering and operation research, there are so many dispatching rules to satisfy such constraints, such as family-based dispatching rule, starvation avoidance-Glassey and Petrakian, SA, Fifo (First-in-First-Out), etc. All of them are developed to such solutions but cannot be really solve under the single tool constraints. In this paper, we propose a model that integrates actively control the single tool problem. By means of dynamic planning model on operation research and logical thinking with systematic software, such a model and system can practically solve the problem in semiconductor manufacturing.