Towards Compelling Cases for the Viability of Silicon-Nanophotonic Technology in Future Many-core Systems

Many crossbenchmarking results reported in the open literature raise optimistic expectations on the use of optical networks-on-chip (ONoCs) for high-performance and low-power on-chip communications in future Manycore Systems. However, these works ultimately fail to make a compelling case for the viability of silicon-nanophotonic technology for two fundamental reasons: (1)Lack of aggressive electrical baselines (ENoCs). (2) Inaccuracy in physical- and architecture-layer analysis of the ONoC. This thesis aims at providing the guidelines and minimum requirements so that nanophotonic emerging technology may become of practical relevance. The key enabler for this study is a cross-layer design methodology of the optical transport medium, ranging from the consideration of the predictability gap between ONoC logic schemes and their physical implementations, up to architecture-level design issues such as the network interface and its co-design requirements with the memory hierarchy. In order to increase the practical relevance of the study, we consider a consolidated electrical NoC counterpart with an optimized architecture from a performance and power viewpoint. The quality metrics of this latter are derived from synthesis and place&route on an industrial 40nm low-power technology library. Building on this methodology, we are able to provide a realistic energy efficiency comparison between ONoC and ENoC both at the level of the system interconnect and of the system as a whole, pointing out the sensitivity of the results to the maturity of the underlying silicon nanophotonic technology, and at the same time paving the way towards compelling cases for the viability of such technology in next generation many-cores systems.