Abstract
Because of the great physical complexity of the precision equipment used in semiconductor manufacturing, it is very difficult to mathematically model the machine behaviors for equipment parameter setting. Trial - and - errors and/or several runs of DOE (Design of Experiment) are often needed in equipment parameter settings for new product production. This research attempts to use Neural Network to model the machine behaviors in order to provide an approach for machine parameter setting thus eliminating or reducing the wasted materials, time, and other resources while conducting the search for proper parameter setting given a set of target machine performance. The research used a two-layer back propagation network to model wire bonder machines used in the packaging industry. Real data from two I.C. packaging plants are used. The result shows that the method is feasible. Benefits of the approach include: reducing trial experiments, in machine setup for new products, thus reducing waste of materials, labors and times, and finding a intelligent method for parameter setting.