In the last decade, the field of microprocessor architecture has seen the rise of multicore processors, which consist of the interconnection of a set of independent processing units or cores in the same chip. As the number of cores per multiprocessor increases, the bandwidth and energy requirements for their interconnection networks grow exponentially and it is expected that conventional on-chip wires will not be able to meet such demands. Alternatively, nanophotonics has been regarded as a strong candidate for chip communication since it could provide high bandwidth with low area and energy footprints. However, issues such as the unavailability of efficient on-chip light sources or the difficulty of implementing all-optical buffering or header processing hinder the development of scalable photonic on-chip networks. In this paper, the area and laser power of several photonic on-chip network proposals is analytically modeled and its scalability is evaluated. Also, a graphene-based hybrid wireless/optical-wired approach is presented, aiming at enabling end-to-end photonic on-chip networks to scale beyond thousands of cores

Peer Reviewed