| Summary: | 碩士 === 華梵大學 === 工業管理學系碩士班 === 92 === Under consideration of reducing operating cost, some TFT-LCD panel manufacturers substitute traditional break first-join later method with newer join first-break later method in their cell process. The problem with the latter approach is the unavoidable yield loss due to the joining of defective panels to the good ones. The quality loss will further cut down the yield and have tremendous impact on profitability. To reduce such loss, the research explored the possibility of using various improved “Yield Mapping” heuristics to solve the problem more efficiently. The divide and conquer concept of these heuristics is to first divide the into subgroups and then transfers the defects on TFT and CF substrates of each subgroup into an O-X matrix to form a series of smaller assignment problems and then solve them in less time.
The research first explored the possibility of adjusting arrangement of substrate sequence to improve the process yield. Next, five mapping strategies were developed. For each strategy, three rules were used for further mapping those unmatched substrates. The effectiveness of the heuristics was evaluated with simulation data based on three different performance indices. The results indicate that determining the weight by summing up the defect locations on each substrate is a more consistent and better performing strategy.
Keywords: Yield mapping, Mapping strategy, Neural Network, Weight mask
|
|---|
