Date de l'exposé : 17 octobre 2014, 10h30-11h30, salle Petri/Turing
Random number generation for cryptography: past, present and futureRandom number generators (RNGs) are crucial in cryptography. They are used to generate confidential keys, initialization vectors, nonces in challenge-response protocols, padding values, and even masks in side-channel attack countermeasures. Random number generator design evolved significantly in last decades: the classical approach searching for generators whose output passes general purpose statistical test is not sufficient any more. Instead, modern designs are based on a thorough on-line evaluation of entropy rate at generator s output based on its stochastic model and dedicated statistical tests. Since dedicated tests are perfectly adapted to generator s principle, they are cheaper, faster, and more efficient than general purpose, so-called black box statistical tests. It is clear that the modern approach increases significantly security since on-line statistical tests can rapidly detect deviations in RNG operation that can be caused by changing environment, aging or active attacks. However, this new approach is not straightforward: it necessitates detailed study of underlying physical phenomena, constructing efficient and robust electronic circuitries, physical and stochastic modeling, and last but not least, designing efficient statistical tests. We will show on few examples the challenges of this new approach.