A superior class of networks for reconfigurable meshes

To achieve fault tolerance through reconfiguration in mesh-connected arrays many rail networks with vastly different effectiveness and cost have been presented. The authors attempt a unified notation of these networks to allow for their comparative evaluation. They further present a method to improve the effectiveness of fault tolerant networks by combining several small switches into large crossbar switches. This method is applicable to almost all rail networks presented in the literature leading to a significant improvement of effectiveness and often also delay time along the network connections for little hardware cost. Furthermore, the switches used provide fault tolerance of the network itself, which is usually unrealistically assumed to be always fault free.<>