As electronic hardware integration technologies develop there is an increasingly strong shift towards implementing complete systems within a single chip, extending the now established paradigm of System on Chip towards high density many-core systems by employing Networks on Chip (NoCs) to connect the processing elements. This brings many new challenges to fault-Tolerant design when applied to embedded applications, but also opportunities for new approaches that can leverage the many-core fabric in ways that traditional system architectures could not exploit. This paper describes such an approach by adopting behavioural aspects of social insects as an inspiration towards autonomous, self-repairing systems. Each router in the NoC is considered as a member of a distributed colony and a simple adaptive controller is responsible for determining the behaviour of each node, relying only on a set of sensory inputs local to each node and small amounts of information shared between neighbours. This provides each node with a small amount of "intelligence" that, for this paper, has been implemented within a many-core hardware system to demonstrate an adaptive routing scheme which provides effective network traffic management through simple and decentralised agent-To-Agent communications. The emergent behaviours of the network are then exploited to demonstrate an example of how fault tolerance could be supported within a many-core system without any pre-defined fault handling strategies. A discussion then follows on how the emergent behaviours of this system can be further inspired by social insect colonies to exhibit other autonomous and adaptive behaviours such as dynamic task allocation.
|Title of host publication||2015 IEEE Symposium Series on Computational Intelligence|
|Number of pages||8|
|Publication status||Published - 2015|
|Event||IEEE Symposium Series on Computational Intelligence, SSCI 2015 - Cape Town, South Africa|
Duration: 8 Dec 2015 → 10 Dec 2015
|Conference||IEEE Symposium Series on Computational Intelligence, SSCI 2015|
|Period||8/12/15 → 10/12/15|