應用田口方法於動態存取記憶體深溝製程技術之改善

Abstract

半導體產業是一個資金密集的產業，以ㄧ般動態存取記憶（dynamic random access memory, DRAM）晶圓製造廠為例，若能減少晶圓製造過程中因缺陷 (defect) 而造成的良率損失，每提升1% 的良率將可能帶來近千萬的收益，因此良率高低已是各晶圓製造廠的關鍵指標。 然而DRAM晶圓的製造過程複雜且繁瑣，在近五百道製程中更以電容深溝 (deep trench) 製造為技術核心，但目前中、外文獻尚無探討電容深溝製造過程中，因硼矽玻璃不足造成缺陷的相關成因及解決方案。本研究探討影響硼矽玻璃不足的關鍵可控因子，並利用田口方法 (Taguchi Method) 分別找出各個品質特性之最佳參數水準組合，再利用標準化各個品質特性之信號雜音比 (Signal to Noise ratio，S/N 比 ) 與權衡取捨分析法（trade-off study）找出最佳的因子水準組合，以達到多重品質特性 (multiple quality characteristics) 最佳化的目的。本研究針對某晶圓製造廠之實際問題進行探討並實際操作，研究結果顯示依本研究建議之方法確可藉由降低電容深溝製作過程中所產生的晶圓缺陷。

The semiconductor industry is a fund concentrated industry. Take the DRAM (Dynamic Random Access Memory) factory as an example, reducing defects in the manufacturing process to gain 1% yield increase could result in at least ten million NT dollars in profit. Therefore, yield is a key index for any semiconductor factory. The manufacture process of DRAM is very complicated. Among those close to five hundreds process steps, deep trench process is a technical core. There has found no related literature in discussing the causes and solutions to defects due to the insufficient BSG remainder in the deep trench process. This research aims to find out the key control factors related to the insufficient BSG remainder in deep trench process. Taguchi Method is then applied to investigate and to find out a good recipe to each quality characteristic for the process. A trade-off method based on standardized signal to noise ratio (S/N ratio) is used to get an acceptable recipe when considering optimizing multiple quality characteristics simultaneously, A real case from one DRAM manufacture was studied. It is shown that defects were reduced in the deep trench process after applying the suggested recipe by this research. .