Abstract
Semiconductor manufacturing is an extremely complex undertaking. With the characteristic of long processing time and shorter product life cycle, it is important to build up the capability of quickly improving the yield of test vehicles to shorten the time lag between test runs and mass production. To investigate the performance of the test vehicles, wafer acceptance test（WAT）needs to be conducted to see whether there exists any special pattern or not. This kind of diagnosis is usually inefficient since it is performed through visual judgment. For this reason, this research use the concept of data mining, together with the Self-Organize Feature Map（SOM）of Neuron Network method, to automatically identify the existence of special WAT patterns. The possibly faulty causes of the test process in the fabrication can then be discovered. Real WAT data of test vehicles obtained from a famous semiconductor manufacturing company were experimented through the developed diagnostic system. The results show that the system can recognize the existence of clustering phenomenon and patterns on wafers. Possible causes of the patterns can also be inferred from the established WAT knowledge base. By the assistance of this system, engineers can more effectively find out process problems on test vehicle process problem to achieve the goal of yield enhancement.