Contribution of communications to dependability in massively-defective general-purpose nanoarchitectures

Many previous studies have analyzed the issues of dependability in the physical layers of future nanoarchitectures. Here we consider the upper layers and especially the contribution of communications to reinforcing the dependability of general-purpose nanoarchitectures (GPN). We adhere to the idea that a GPN should be viewed as a single-chip massive multiprocessor system, consisting in immersing a few high-performance processors in a grid of medium-performance processors to enable efficient execution of both parallelizable and not parallelizable applications. We especially discuss the operation of the communication layer and introduce the mechanism of "blind" task allocation through a contract network protocol. In the framework of this GPN sketch, we show that the communication layer could dramatically contribute to improving the dependability of massively-faulty GPNs by tolerating the increasing number of defective nodes expected with the reduction of dimensions, so that chips with as much as 30% of defective nodes could be validated for deployment