Light, as a medium for communication, has the unique ability to transmit volumes of data with minimal energy loss. This capability not only sparked the revolution of internet-based communication over fiber optic networks, but also holds the potential to expand computing beyond our current capabilities. At present, data is stored densely in computer chips, but is sent out of the chip through centimeter-long electrical wires in a slow and energy-intensive process, before finally interfacing with optical transmitters.
To bypass this bottleneck, electrical channels can be condensed and converted into light over a compact area using integrated photonic chips. In particular, the silicon photonics technology platform offers the potential for extremely dense data communications due to its high confinement waveguides and compact micro-resonators. However, three major obstacles stand in the way of realizing a low-energy and bandwidth-dense implementation of this technology: the integration of photonics with electronics, optical coupling from the photonic chip to fiber, and scaling up link architectures to multiplex data streams onto many wavelengths.
The work in this thesis aims to confront these three challenges and advance integrated photonics technology to unprecedented bandwidth densities and energy efficiencies, with a focus on the first challenge of photonic-electronic integration. It begins with an overview of the escalating demand for inter-chip bandwidths and the potential solution offered by integrated photonics. Next, this thesis builds a theoretical framework for the performance parameters and sources of energy consumption that are addressed in the subsequent sections. After this introductory context, the thesis describes the achievement of the highest density and largest scale photonic-electronic integration to date, using a dense, 25 um pitch 3D bonding process. An 80-channel array fabricated in this integration records the lowest data link energies to date, at 120 fJ/bit, and transfers data at 10 Gbit/s/channel for a record 5.3 Tbit/s/mm2 bandwidth density.
The discussion then shifts to the issue of chip-to-fiber coupling efficiency, traditionally the greatest source of loss in photonic links. A substrate-removed edge coupler design reduces this loss to a mere 1.1 dB, and an inverse-designed edge coupler taper shows a fourfold length reduction compared to linear tapers. Lastly, the thesis presents designs for wavelength scaling that increase the number of energy efficient channels on a single fiber. Specifically, it demonstrates a multi-channel, polarization diverse micro-comb receiver and a 3D-integrated transceiver with wavelength interleaving to waveguide buses of cascaded resonators.
This thesis builds on photonic device developments to introduce photonic systems with the lowest energy and densest data communications to date. Together, these results unlock the tremendous potential of light as a fast and energy-efficient communication medium between chips, paving a sustainable path towards scaling artificial intelligence and disaggregating computation and memory resources.
Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/96j0-mp97 |
Date | January 2023 |
Creators | Daudlin, Stuart |
Source Sets | Columbia University |
Language | English |
Detected Language | English |
Type | Theses |
Page generated in 0.0015 seconds