On-Chip Optical Interconnection Networks for Multi/Manycore Architectures

The rapid development of multi/manycore technologies offers the opportunity for highly parallel architectures implemented on a single chip. While the first, low-parallelism multicore products have been based on simple interconnection structures (single bus, very simple crossbar), the emerging highly parallel architectures will require complex, limited-degree interconnection networks. This thesis studies this trend according to the general theory of interconnection structures for parallel machines, and investigates some solutions in terms of performance, cost, fault-tolerance, and run-time support to shared-memory and/or message passing programming mechanisms