| Summary: | 碩士 === 中華大學 === 資訊管理學系 === 93 === The complicated manufacturing process and high quality demand are two features of Very Large Scale Integration (VLSI). To monitor the whole manufacturing process and explore the reason of low yield, WAT plays an important role. But too many parameters of Wafer Acceptance Test (WAT) cause difficulties for engineers to explore the problems. This research explores the effective WAT argument combination and reduces the number of WAT parameters to explore the reason of low yields with Related-Beta Base Function Neural Network (R-BBFNN) and Genetic Algorithm (GA). R-BBFNN is able to gather statistics of parameter’s upper-lower bound and network training. GA is able to propagate genes and find the optimal solution.
This research consists of three modules with R-BBFNN and GA, which are R-BBFNN population module, R-BBFNN child module and GA module. R-BBFNN population module uses the population as input to train the network weight. R-BBFNN child module feedbacks the error of network. GA module optimize the population. In the experiment result, the population converges in a variation environment, the best hit rate of problem causing parameters is 80% and above, and the number of the test parameters are greatly decreased.
|
|---|
