ENG-SEMICON: Coherent Co-packaged Optics (C2PO) using Offset-QAM Modulation

University Of Washington, Seattle WA

Investigators

Abstract

The advent of computationally intensive applications such as generative artificial intelligence (AI) has created a vital demand for dense and ultra-low power multi-TB/s inter/intra-rack optical communications (with <100m reach) as well as low-latency chip-to-chip interconnects. Similar needs will be soon required for edge-to-cloud connectivity and 5G/6G front/back-haul networks. Silicon photonic (SiPho) Transceivers have shown a great promise to address this challenge by ultimately co-packaging optical transceivers with high performance GPU/FPGA/SoCs in a same package (“co-packaged optics” or CPO). Among various realization approaches, micro-ring modulators (MRM) have significantly improved the energy-efficiency and shoreline (or edge) bandwidth (BW) densities in terms of Tb/s/mm compared with conventional Mach-Zehnder modulators (MZM) and VCSELs. State-of-the-art demonstration could achieve 100Gb/s data-rates per wavelengths with ~5pJ/b energy-efficiencies and less than 0.5Tb/s/mm BW densities using multi-level amplitude modulation. These numbers are still an order of magnitude behind what future AI processing demands. Since there is a large energy penalty in scaling up baud-rates, vast parallelization degrees should be utilized to address multi-TB/s aggregate off-package data-rate demands. So far Wavelength division multiplexing (WDM) have been proposed and demonstrated to do so. While multiplexing can be a near future solutions, it is evident that advanced coherent modulations like QAM can be an ultimate solution to increase spectral-efficiency and overall aggregated bandwidths per fiber for CPO. However, today’s coherent optical transceivers are not yet suitable for CPO applications in AI datacenters. In this proposal, we are introducing Coherent CPO (C2PO) to enable compact and low-power QAM modulation using MRMs. Unlike today’s datacenter connectivity, future of interconnects will rely on massively parallelized multi-channel energy/area efficient coherent optical links that should be co-packaged with processor and accelerators. Key technologies such as generative AI, autonomous vehicles, AR/VR and 6G all require such a critical backbone technology paradigm shift. Our proposed work enhances key metrics by 10x-100x compared with today’s commercial and R&Ds solutions. Proposed method can be extended in future to higher QAM modulations such as QAM-64 to support data-rates of +1Tb/s per wavelength in each fiber. Education, workforce development, and outreach activities within this project focus on the semiconductor industry's future needs. These activities encompass: Developing course materials for the co-design of RF/High-speed circuits with emerging devices and creating analog layout and design automation (in Python) for high-school internship/outreach programs. PI will leverage various mechanisms to achieve broader impacts in diversity, education, and outreach through this project. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.