| Summary: | 碩士 === 中華大學 === 科技管理研究所 === 90 === The end point curve of an etching process in semiconductor manufacture can be used to determine the quality of a product . The product line can monitor the change of the end point curve on real time to avoid the abnormal product produced to minimize the defect loss. This research attempts to find a way to replace those current methods which rely on human to check and determine the abnormal pattern during an etching process in semiconductor manufacture. A back-propagation neural network is adopted in this research to recognize the pattern change in etching process because the method has a self-learning characteristics. As long as a better pattern is picked, the system can classify the pattern sample via the pattern recognition training. The ability to judge patterns can be enhanced by learning new patterns to renew the previous result of the classification. This system can improve the pattern classification of an etching process and provide a suggestion to stop running the product line. In this research, back-propagation neural network algorithm is selected and the digitized information from the end point curve of the etch process is used for analysis. Any recipe from a random picking three or more manufacture recipes provides 200 training samples for learning, then the remaining 100 samples are used to test the training result and recognize result to get up to 96%. By this way, this research can develop a system for the end point curve pattern recognition of etching process in semiconductor manufacture.
|
|---|
