Méthodes et architectures basées sur la redondance modulaire pour circuits combinatoires tolérants aux fautes

In this thesis, we mainly take into account the representative technique Triple Module Redundancy (TMR) as the reliability improvement technique. A voter is an necessary element in this kind of fault-tolerant architectures. The importance of reliability in majority voter is due to its application in both conventional fault-tolerant design and novel nanoelectronic systems. The property of a voter is therefore a bottleneck since it directly determines the whole performance of a redundant fault-tolerant digital IP (such as a TMR configuration). Obviously, the efficacy of TMR is to increase the reliability of digital IP. However, TMR sometimes could result in worse reliability than a simplex function module could. A better understanding of functional and signal reliability characteristics of a 3-input majority voter (majority voting in TMR) is studied. We analyze them by utilizing signal probability and boolean difference. It is well known that the acquisition of output signal probabilities is much easier compared with the obtention of output reliability. The results derived in this thesis proclaim the signal probability requirements for inputs of majority voter, and thereby reveal the conditions that TMR technique requires. This study shows the critical importance of error characteristics of majority voter, as used in fault-tolerant designs. As the flawlessness of majority voter in TMR is not true, we also proposed a fault-tolerant and simple 2-level majority voter structure for TMR. This alternative architecture for majority voter is useful in TMR schemes. The proposed solution is robust to single fault and exceeds those previous ones in terms of reliability.Dans cette thèse, nous nous intéressons à la recherche d’architectures fiables pour les circuits logiques. Par ”fiable”, nous entendons des architectures permettant le masquage des fautes et les rendant de ce fait ”tolérantes" à ces fautes. Les solutions pour la tolérance aux fautes sont basées sur la redondance, d’où le surcoût qui y est associé. La redondance peut être mise en oeuvre de différentes manières : statique ou dynamique, spatiale ou temporelle. Nous menons cette recherche en essayant de minimiser tant que possible le surcoût matériel engendré par le mécanisme de tolérance aux fautes. Le travail porte principalement sur les solutions de redondance modulaire, mais certaines études développées sont beaucoup plus générales