<p> This Master Thesis presents a new method of analyzing and classifying liquid crystal textures, using feed-forward neural networks and different clustering techniques. Liquid crystal phases are generally identified by human experts by polarizing optical microscopy observations of textures, based on typical defects, the smoothness or sharpness of domains and the birefringence colors of the films. The thesis aims to establish a novel algorithmic technique for liquid crystal texture analysis and classification. Using image analyzing software, a characterization vector with 22 parameters is extracted for each texture. Advanced clustering algorithms are used to classify textures based on those characteristic parameters. Furthermore, a ranking of the measurements is proposed to refine the accuracy of classification. The proposed methodology will lead to a reliable and simple technique for the physical investigation of liquid crystal materials. </p>

Mobile appliances and especially smart cards have found more and more applications in the past two decades. A little more than ten years ago, the security of those devices still only relied on mathematical complexity and computational infeasibility to force cryptographic systems. Unfortunately, during the execution of cryptographic algorithms, unintentional leakage may be observed. Indeed, the power consumption or the electromagnetic emanations of the device are correlated to the encryption/decryption process. Those unintended channels are called “sidechannel”. Our work was not targeted at the discovery of new “side-channel” sources but rather at a thorough investigation of two of them: the power consumption and the electromagnetic emanation in the near-field domain. In this respect, we dealt with three different aspects of the problem: 1. We carried out many experiments on small microcontrollers but also on FPGAs in order to provide an explanation on the sources and on the set up of an efficient measurement process. Moreover, we provide the first XY scanning pictures of the electromagnetic field radiated by a small microcontroller. 2. Obtaining several measures of the observed side-channel, how is it possible to statistically analyzed these observations? We detail here the different methods available and we introduce an enhancement in the Template Attack process with Principal Component Analysis. 3. Finally, on the basis of this experience, we tried to answer the following question: “Is it possible to provide a theoretical tool to evaluate secure implementations?” The idea was to follow the notion of “Physical Computer” introduced by Micali and Reyzin. In this respect, we provide here two metrics that we consider necessary to evaluate both the strength of the adversary and the information held in the leakage. Respectively we choose the average success rate and the Shannon's mutual information.