Investigation of Sample Sizes and Correlation in Multi-Cluster Feature Distributions for an Efficient Encryption System

This paper investigates some practical aspects of the employment of measurable features derived from characteristics of given integrated electronic circuits for the generation of encryption keys pertaining to the circuits, a technique termed ICmetrics. Specifically the paper addresses difficulties introduced by features exhibiting highly diverse distributions, potentially containing many distinct clusters associated with each of the given circuits. For such feature distributions, it is crucial to detect the precise number of clusters associated with each given circuit and the paper discusses the consequentially crucial importance of selecting the appropriate number of samples in order to correctly detect and identify the number of clusters. Moreover, the phenomenon of correlation in multi-cluster features is analyzed and methods of how to successively detect it are presented.