Security issues play an important role in almost all modern communication and computer networks. As Internet applications continue to grow dramatically, security requirements have to be strengthened. Hyperelliptic curve cryptosystems (HECC) allow for shorter operands at the same level of security than other public-key cryptosystems, such as RSA or Diffie-Hellman. These shorter operands appear promising for many applications. Hyperelliptic curves are a generalization of elliptic curves and they can also be used for building discrete logarithm public-key schemes. A major part of this work is the development of computer architectures for the different algorithms needed for HECC. The architectures are developed for a reconfigurable platform based on Field Programmable Gate Arrays (FPGAs). FPGAs combine the flexibility of software solutions with the security of traditional hardware implementations. In particular, it is possible to easily change all algorithm parameters such as curve coefficients and underlying finite field. In this work we first summarized the theoretical background of hyperelliptic curve cryptosystems. In order to realize the operation addition and doubling on the Jacobian, we developed architectures for the composition and reduction step. These in turn are based on architectures for arithmetic in the underlying field and for arithmetic in the polynomial ring. The architectures are described in VHDL (VHSIC Hardware Description Language) and the code was functionally verified. Some of the arithmetic modules were also synthesized. We provide estimates for the clock cycle count for a group operation in the Jacobian. The system targeted was HECC of genus four over GF(2^41).

La multiplication de polynômes est une opération fondamentale en théorie de la complexité. En effet, pour de nombreux problèmes d’arithmétique, la complexité des algorithmes s’exprime habituellement en fonction de la complexité de la multiplication. Un meilleur algorithme de multiplication permet ainsi d’effectuer les opérations concernées plus rapidement. Un résultat de 2016 a établi une meilleure complexité asymptotique pour la multiplication de polynômes dans des corps finis. Cet article constitue le point de départ de la thèse ; l’objectif est d’étudier les conséquences à la fois théoriques et pratiques de la nouvelle borne de complexité.La première partie s’intéresse à la multiplication de polynômes à une variable. Cette partie présente deux nouveaux algorithmes censés accélérer le calcul en pratique (plutôt que d’un point de vue asymptotique). S’il est difficile dans le cas général d’observer l’amélioration prévue, certains cas précis sont particulièrement favorables. En l’occurrence, le second algorithme proposé, spécifique aux corps finis, conduit à une meilleure implémentation de la multiplication dans F_2[X], environ deux fois plus rapide que les logiciels précédents.La deuxième partie traite l’arithmétique des polynômes à plusieurs variables modulo un idéal, telle qu’utilisée par exemple pour la résolution de systèmespolynomiaux. Ce travail suppose une situation simplifiée, avec seulement deux variables et sous certaines hypothèses de régularité. Dans ce cas particulier, la deuxième partie de la thèse donne des algorithmes de complexité asymptotiquement optimale (à des facteurs logarithmiques près), par rapport à la taille des entrées/sorties. L’implémentation pour ce cas spécifique est alors nettement plus rapide que les logiciels généralistes, le gain étant de plus en plus marqué lorsque la taille de l’entrée augmente. / The multiplication of polynomials is a fundamental operation in complexity theory. Indeed, for many arithmetic problems, the complexity of algorithms is expressed in terms of the complexity of polynomial multiplication. For example, the complexity of euclidean division or of multi-point evaluation/interpolation (and others) is often expressed in terms of the complexity of polynomial multiplication. This shows that a better multiplication algorithm allows to perform the corresponding operations faster. A 2016 result gave an improved asymptotic complexity for the multiplication of polynomials over finite fields. This article is the starting point of the thesis; the present work aims to study the implications of the new complexity bound, from a theoretical and practical point of view.The first part focuses on the multiplication of univariate polynomials. This part presents two new algorithms that should make the computation faster in practice (rather than asymptotically speaking). While it is difficult in general to observe the predicted speed-up, some specific cases are particularly favorable. Actually, the second proposed algorithm, which is specific to finite fields, leads to a better implementation for the multiplication in F 2 [X], about twice as fast as state-of-the-art software.The second part deals with the arithmetic of multivariate polynomials modulo an ideal, as considered typically for polynomial system solving. This work assumes a simplified situation, with only two variables and under certain regularity assumptions. In this particular case, there are algorithms whose complexity is asymptotically optimal (up to logarithmic factors), with respect to input/output size. The implementation for this specific case is then significantly faster than general-purpose software, the speed-up becoming larger and larger when the input size increases.