Abstract
In the semiconductor assembly and test (A/T) industry, using physics equations to deduce the parameter settings of machines is difficult due to complicated machine movements. Therefore, the purpose of this research is to use neural network to modelize the complex machine behaviors and tool parameter settings and then determine the practical key parameters for test programs. After modelizing the machine attributes with neural network, these attributes were trained and then applied to assist machine settings by using the given machine attributes. The same procedures were implemented in several machines and afterwards the tool matching machine with the performance that is nearest to the expected result and specification was selected to collect the production data. In all A/T factories, test machines are relatively expensive and the test programs are mainly owned by customers; therefore, the techniques of writing test programs are barely existed domestically. However, there is an enormous amount of experience in executing tests and partial modifications. This paper aims to develop the domestic ability of writing test programs through transforming these experiences into self values. Furthermore, there are thousands of procedures in test programs, so how to refine parameters and identify the essential ones in the process of accumulating the ability of writing test programs is extraordinary important. In this study, using neural network back-propagation algorithm, the most crucial parameters were determined and the purpose of developing more efficient test processes was achieved as well. Keywords : Neural network, Back propagation, IC test