應用微影覆蓋誤差的錯誤診斷分析改善微影製程總體設備效能

Abstract

半導體晶圓尺寸從8吋邁向12吋，線寬從微米縮小至奈米的新世代，晶圓越大線寬越小相對於製程方面也越來越複雜，晶圓製程中微影製程是決定線寬是否能更小的主要因素。 微影製程檢視又可分為三大類，層對層覆蓋誤差檢查（Overlay error）及線與孔的寬度檢查（Critical Dimensions, CD）以及顯影後檢查（After Develop Inspection, ADI），微影製程良率就是取決於這三項參數。 本論文所研究的方向是覆蓋誤差。應用類神經網路（Neural Network）及錯誤診斷分析（Fault Detection and Classification）的觀念，利用晶圓實際生產數據建立類神經網路的輸入層與輸出層數據，運用輸入及輸出數據來訓練與調整隱藏層神經單元數目及權重，找出類神經網路輸入及輸出數據相對應的非線性關係，再利用晶圓實際生產新數據加以驗證類神經網路輸入及輸出相對應的非線性關係。利用監控設備覆蓋誤差的數據作為輸入層與監控產品覆蓋誤差的數據作為輸出層，就能預測產品的覆蓋誤差，然後從黃光微影術生產驗證流程切入，進一步改善並提升微影製程機台總體設備效能（Overall Equipment Effectiveness簡稱OEE）

Semiconductor wafer size has been increased from 8 inch to 12 inch, line width in the IC wafer has been reduced from micrometer to nanometer range. The bigger the wafer size and the smaller the line width, the more complicated the wafer manufacturing process. Photolithography process is the key process in reducing line width in IC wafer. There are three kinds of photolithography process inspection, i.e. overlay error, critical dimension （CD） and after develop Inspection （ADI）, the yield of photolithography process depends on these three parameters.

This research focuses on overlay error. Applying the concept of Neural Network and Fault Detection and Classification, we establish input layer data and output layer data of Neural Network with data collected from actual production run of IC wafer. Using input layer data and output layer data to train and adjust the number of neuron members and the associated weighting factors in the hidden layer, we can establish the nonlinear relationship between input and output data of Neural Network. We then verify the Neural Network model with new sets of data from IC wafer production run. Using the monitor wafer overlay error data from photolithography equipment as input data and the overlay error data from product wafer as output data, we can predict the overlay error data of future product wafer. Through simplification of photolithography inspection process of product wafer, we can improve the Overall Equipment Effectiveness （OEE） of photolithography equipments.