The emergence of very high performance processors is necessary to achieve future required application performance in terms of resolution (audio, video and computing) and CPU power / total MIPS (real-time encoding-decoding, data encryption-decryption, artificial intelligence). Distributed manycore architectures are key to achieve such performance levels but require organized high-speed communication between processors, capable of supporting an aggregate data rate exceeding 100TB/s. Metallic interconnect is highly inefficient in this role due to unachievable tradeoffs between design parameters. Silicon photonics, and the concept of on-chip optical data transport, is a potential technological solution to alleviate some of these issues involved in exchanging data between cores in SoC architectures, and in particular that of energy (around 1pJ/bit/mm in metallic interconnect). This research track aims to contribute to the ongoing assessment of the suitability of integrated optical interconnect for on-chip data transport. In this context, we have contributed to the state of the art at world-class level:
Predictive performance assessment of optical interconnect per technology node. We were the first to quantify in detail the performance levels for future technology nodes.
Development of several passive and active optical network on chip topologies exploiting wavelength and targeting real communication requirements.
Modeling and simulation at the system level of optical networks on chip integrated with an industrial communication protocol.
Currently funded work is focusing on the design of energy-efficient broadcast-aware nanophotonic interconnect that goes below the 0.1pJ/bit energy consumption barrier for an aggregated 100TB/s total bandwidth and on the association of stochastic computing with PCM-enhanced silicon photonics to implement a matrix multiplication accelerator for in-network neural networks.
This work is the result of 8 PhD theses (2 ongoing) and 3 postdocs, conducted in the framework of several national (RMNT Heteropt, LabEx CominLabs 3D-Optical-Manycores, ANRAAPG OpticALL2, ANRAAPG Octane), European (FP6-IST PICMOS, FP7-ICT WADIMOS, FP7-ICT HELIOS, Horizon Europe Neuropuls) and international (Horizon 2020 MSCA REDI) projects.
Collaborations: CEA-LETI (FR), INRIA-IRISA (FR), C2N (FR), Mentor Graphics (FR), Kalray (FR), ICB (FR), IMEC (BE), STMicroelectronics (IT), Ecole Polytechnique de Montréal (CA), Concordia University (CA), Hong Kong University of Science and Technology (CN), RMIT (AUS).