Principal and Independent Component Analysis for Crypto-systems with Hardware Unmasked Units

Abstract. Principal component analysis (PCA) is a well-known statistical method for reducing the dimension of a data set. This method finds a new representation of the data set conserving only the most important information. It is based on spectral decomposition of the covariance matrix of the data set. In this article, we apply principal component analysis to facilitate simple power analysis. Principal component analysis, however, makes no assumption on the independence of the data vectors. In 1986, Herault and Jutten introduced a learning algorithm based on a version of the Hebb learning rule [1]. This algorithm was able to blindly separate mixtures of independent signals. Independent Component Analysis (ICA) is a powerful tool for signal processing. PCA and ICA permit to improve the signal to noise ratio of signals used for differential side channel analysis, which makes the application of classical methods to recover cryptographic keys more easy. To be a serious threat against present cryptographic implementations, PCA and ICA should be accompanied with other signal processing tools.