Identification of the Optimal Set of Informative Features for the Problem of Separating of Mixed Production Batch of Semiconductor Devices for the Space Industry

Abstract

In this paper, we investigate the problem of separation of a mixed production batch of semiconductor devices of space application into homogeneous production batches. The results of the mandatory testing for each item contain a large number of parameters. Many optimization models and algorithms were developed for solving this clustering problem in the most efficient way. However, due to a rather high data dimensionality, such algorithms take significant computational resources. We analyzed methods of reducing the dimensionality of the data set with the use of factor analysis based on Pearson matrix in order to improve the accuracy of the separation. We investigated efficiency of the proposed method for separating a mixed lot of semiconductor devices which consists of two, three, four and seven homogeneous batches, with various methods of selection and rotation of factors. It was shown on real data that with any orthogonal rotation, with an increasing number of homogeneous batches in the sample, the clustering accuracy decreases. Moreover, it was impossible to identify a universal clustering model with a limited number of factors for dividing a mixed lot composed from an arbitrary number of homogeneous batches. Thus, the use of the multidimensional data was shown to be inevitable.