A neural network classifier with rough set-based feature selection to classify multiclass IC package products

The choice of packaging type is important to the process of researching and developing an integrated circuit (IC). Indeed, for an IC chip designer, the importance can be compared to an architect’s choice of construction design. Since there are considerable variations in characteristics and in the types of products available, collecting information about packaging technologies and products can be difficult and time-consuming. Therefore, finding the means to provide packaging information to designers quickly and efficiently is necessary and important, as this will not only help designers accurately decide on design methods for an IC, but also significantly reduce processing risks. In this study, existing product information, such as the dimensions, characteristics and design and application criteria, of a product was analyzed. One of the biggest issues when data from multi-dimensional measurements are represented as a feature vector is that the feature space of the raw data often has very large dimensions. This study explores the use of rough set attribute reduction (RSAR) to reduce attributes of the IC package family dataset, and artificial neural networks, to construct an efficient IC package type classifier model. The experimental results show that the features produced by RSAR improve on generalization accuracy: the training and testing set classification accuracy rates were 96.9% and 98.2%, respectively.