Bandwidth Maximization and Allocation for Wavelength-Routed Optical Networks-on-Chip (WRONoC)

• Applicant: Dr.-Ing. Tsun-Ming Tseng
• Subject Area: Computer Architecture, Embedded and Massively Parallel Systems
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 496766278

Wavelength-routed optical networks-on-chip (WRONoC) is a next-generation solution to offer energy-efficient and high-bandwidth communications in multiprocessor system-on-chip (MPSoC). WRONoCs support full bandwidth, i.e., all initiators can communicate with all their targets simultaneously without data collision. However, existing WRONoC topologies assume that only a single bit of data is sent/received at a time between each initiator-target pair, i.e., there is no bit-level parallelism. This conservative setting limits the bandwidth performance of WRONoCs. In this project, we propose to enable bit-level parallel communications for WRONoCs with design automation approaches. Specifically, we aim to optimize the usage of the resonant wavelengths of the on-resonance microring resonators (MRRs) in each signal path, so that the bit-level parallelism of the whole network can be generally improved and the critical signal paths have more parallelism. To this end, we propose to develop a comprehensive WRONoC topology synthesis method to optimize the waveguide structures, signal wavelengths, and the number, locations, and configurations of MRRs of a network for high bandwidths. Besides, for applications that require high communication density in some specific signal paths, we propose to develop a topology revision method that locally changes the waveguides and MRRs in those paths to increase the number of valid signal wavelengths and thus to provide more parallelism. We also propose to accurately model the power of the crosstalk signals to maximally take advantage of the available wavelengths while maintaining good signal quality.

• DFG Programme: Research Grants