Microring-Based DWDM for CPO Systems: Motivation, Challenges, and Design Approaches

SystemX Affiliates: login to view related content.

Microring-Based DWDM for CPO Systems: Motivation, Challenges, and Design Approaches
Thursday, May 11, 2023 - 4:30pm to 5:30pm
Y2E2 111
Angad Rekhi - NVIDIA
Abstract / Description: 

*To receive email announcements and live stream information for upcoming seminars, please subscribe to the SystemX Seminar/EE310 Mailing list here.

Motivated by the ever-increasing demand for compute and data by datacenter, scientific computing, and AI applications, accelerated computing systems continue to grow in both computational power and interconnectedness. As these systems continue to scale, they place increasingly stringent demands on the bandwidth, bandwidth density, and energy efficiency of interconnects. Existing electrical and optical interconnect solutions are projected to fall short of the requirements imposed by next-generation systems in these key metrics. In this talk, we first review how co-packaged optics (CPO) technology, and in particular, 2.5D integrated optics, presents a compelling solution for the required scaling of interconnect bandwidth. We then motivate the use of dense wavelength-division multiplexing (DWDM) as an architecture choice that can meet the required bandwidth density and energy efficiency targets within the context of 2.5D integrated CPO systems. Finally, we spend the remainder of the talk outlining the implementation of a microring resonator-based DWDM link architecture, touching on electrical, optical, and thermal design considerations.


Angad S. Rekhi received a B.S. in electrical engineering from Caltech in 2014 and an M.S. and Ph.D. in electrical engineering from Stanford University in 2016 and 2020, respectively. Since 2021, he has been a research scientist in the Circuits Research Group (CRG) at NVIDIA Research, where he is working primarily on electronics and photonics for co-packaged optics-based links; he also interned with CRG in 2018, during which he designed analog/mixed-signal hardware for machine learning. From 2014 to 2020, his PhD work with Professor Amin Arbabian at Stanford involved the implementation of ultrasound-assisted wireless connectivity for next-generation IoT devices and the investigation of a novel form of neurostimulation. From 2012 to 2014, he worked with Professor Ali Hajimiri at Caltech on the design and test of integrated electronic-photonic systems. Angad was a recipient of the NSF GRFP in 2014, the NDSEG Fellowship in 2016, the ADI Outstanding Student Designer Award in 2014 and 2018, and the IEEE SSCS Predoctoral Achievement Award in 2019.