ANALYSIS OF CRYPTOGRAPHIC ALGORITHMS AGAINST THEORETICAL AND IMPLEMENTATION ATTACKS

This thesis deals with theoretical and implementation analysis of cryptographic functions. Theoretical attacks exploit weaknesses in the mathematical structure of the cryptographic primitive, while implementation attacks leverage on information obtained by its physical implementation, such as leakage through physically observable parameters (side-channel analysis) or susceptibility to errors (fault analysis). In the area of theoretical cryptanalysis, we analyze the resistance of the Keccak-f permutations to differential cryptanalysis (DC). Keccak-f is used in different cryptographic primitives: Keccak (which defines the NIST standard SHA-3), Ketje and Keyak (which are currently at the third round of the CAESAR competition) and the authenticated encryption function Kravatte. In its basic version, DC makes use of differential trails, i.e. sequences of differences through the rounds of the primitive. The power of trails in attacks can be characterized by their weight. The existence of low-weight trails over all but a few rounds would imply a low resistance with respect to DC. We thus present new techniques to effciently generate all 6-round differential trails in Keccak-f up to a given weight, in order to improve known lower bounds. The limit weight we can reach with these new techniques is very high compared to previous attempts in literature for weakly aligned primitives. This allows us to improve the lower bound on 6 rounds from 74 to 92 for the four largest variants of Keccak-f. This result has been used by the authors of Kravatte to choose the number of rounds in their function. Thanks to their abstraction level, some of our techniques are actually more widely applicable than to Keccak-f. So, we formalize them in a generic way. The presented techniques have been integrated in the KeccakTools and are publicly available. In the area of fault analysis, we present several results on differential fault analysis (DFA) on the block cipher AES. Most DFA attacks exploit faults that modify the intermediate state or round key. Very few examples have been presented, that leverage changes in the sequence of operations by reducing the number of rounds. In this direction, we present four DFA attacks that exploit faults that alter the sequence of operations during the final round. In particular, we show how DFA can be conducted when the main operations that compose the AES round function are corrupted, skipped or repeated during the final round. Another aspect of DFA we analyze is the role of the fault model in attacks. We study it from an information theoretical point of view, showing that the knowledge that the attacker has on the injected fault is fundamental to mount a successful attack. In order to soften the a-priori knowledge on the injection technique needed by the attacker, we present a new approach for DFA based on clustering, called J-DFA. The experimental results show that J-DFA allows to successfully recover the key both in classical DFA scenario and when the model does not perfectly match the faults effect. A peculiar result of this method is that, besides the preferred candidate for the key, it also provides the preferred models for the fault. This is a quite remarkable ability because it furnishes precious information which can be used to analyze, compare and characterize different specific injection techniques on different devices. In the area of side-channel attacks, we improve and extend existing attacks against the RSA algorithm, known as partial key exposure attacks. These attacks on RSA show how it is possible to find the factorization of the modulus from the knowledge of some bits of the private key. We present new partial key exposure attacks when the countermeasure known as exponent blinding is used. We first improve known results for common RSA setting by reducing the number of bits or by simplifying the mathematical analysis. Then we present novel attacks for RSA implemented using the Chinese Remainder Theorem, a scenario that has never been analyzed before in this context.