Simultaneously transfer data with many light colors
Newswise – New York, NY – June 29, 2023 – The data centers and powerful computers that run artificial intelligence programs, such as large language models, are not limited by the sheer computational power of their individual nodes. It is another issue — the amount of data they can transfer between nodes — that underlies the “bandwidth bottleneck” that currently limits the performance and scalability of these systems.
The nodes in these systems can be more than a kilometer apart. Because metal wires dissipate electrical signals as heat when transferring data at high speeds, these systems transfer data over fiber optic cables. Unfortunately, a lot of energy is wasted converting electrical data to optical data (and back again) when sending signals from one node to another.
In a study published today in Nature Photonics, researchers at Columbia Engineering demonstrate an energy-efficient method for transferring larger amounts of data over the fiber optic cables connecting the nodes. This new technology improves upon previous attempts to transmit multiple signals simultaneously over the same fiber optic cables. Instead of using a different laser to generate each wavelength of light, the new chips require only a single laser to generate hundreds of different wavelengths of light that can simultaneously transfer independent streams of data.
A simpler, more energy-efficient method of data transfer
The millimeter scale system uses a technique called wavelength division multiplexing (WDM) and devices called Kerr frequency combs that take a single color of light at the input and create many new colors of light at the output. The critical Kerr frequency combs, developed by Michal Lipson, Higgins Professor of Electrical Engineering and Professor of Applied Physics, and Alexander Gaeta, David M. Rickey Professor of Applied Physics and Materials Science and Professor of Electrical Engineering, enabled the researchers to obtain clear signals by separate and precise wavelengths of light, with space in between.
“We recognized that these devices are ideal sources for optical communications, where one can encode independent channels of information about each color of light and distribute them over a single optical fiber,” said senior author Keren Bergman, Charles Batchelor Professor of Electrical Engineering at Columbia Engineering, where she is also the faculty director of the Columbia Nano Initiative. This breakthrough allows systems to transfer exponentially more data without consuming proportionally more energy.
The team shrunk all the optical components on chips about a few millimeters on each edge to generate light, coded them with electrical data, and then converted the optical data back into an electrical signal at the target node. They came up with a new photonic circuit architecture that allows each channel to be individually encoded with data while having minimal interference with neighboring channels. This means that the signals sent in any color of light are not confused and difficult for the receiver to interpret and can be converted back into electronic data.
“In this way, our approach is much more compact and energy efficient than comparable approaches,” said the study’s lead author Anthony Rizzo, who performed this work when he was a PhD student in the Bergman lab and is now a research scientist at the US Air Force Research Laboratory Information Directorate. “It is also cheaper and easier to scale because the silicon nitride comb generation chips can be fabricated in standard CMOS foundries used to fabricate microelectronics chips rather than expensive dedicated III-V foundries.”
The compact nature of these chips allows them to communicate directly with computer electronics chips, significantly reducing overall power consumption as the electrical data signals only need to propagate over millimeters rather than tens of centimeters.
Bergman commented, “What this work shows is a viable path to both dramatically reducing system power consumption while simultaneously increasing computational power by orders of magnitude, allowing artificial intelligence applications to continue to grow exponentially with minimal impact. on the environment.”
Exciting results pave the way for real-world implementation
In experiments, the researchers managed to transfer 16 gigabits per second per wavelength for 32 different wavelengths of light for a total single-fiber bandwidth of 512 Gb/s with less than one bit of error on a trillion bits of data transmitted. These are incredibly high levels of speed and efficiency. The silicon chip that relayed the data measured just 4mm x 1mm, while the chip that received the optical signal and converted it into an electrical signal measured just 3mm x 1mm – both smaller than a human fingernail.
“While we used 32 wavelength channels in the proof-of-principle demonstration, our architecture can be scaled to accommodate over 100 channels, which is well within the range of standard Kerr comb designs,” adds Rizzo.
These chips can be fabricated with the same facilities used to make the microelectronics chips found in a standard consumer laptop or cell phone, providing an easy path to volume scaling and real-world deployment.
The next step in this research is the integration of photonics with chip-scale drive and control electronics to further miniaturize the system.
###
About the study
Journal: Nature photonics
The sheet is titled:Massively scalable Kerr comb-driven silicon photonic link.”
The authors are: Anthony Rizzo (formerly Columbia University, currently Air Force Research Laboratory), Asher Novick (Columbia University), Vignesh Gopal (Columbia University), Bok Young Kim (Columbia University), Xingchen Ji (formerly Columbia University, currently Shanghai Jiao Tong University), Stuart Daudlin (Columbia University), Yoshitomo Okawachi (formerly Columbia University, currently Xscape Photonics), Qixiang Cheng (formerly Columbia University, now University of Cambridge), Michal Lipson (Columbia University), Alexander Gaeta (Columbia University), and Keren Bergman (Columbia University)
This work was supported in part by the US Advanced Research Projects Agency—Energy under ENLITENED Grant DE-AR000843 and in part by the US Defense Advanced Research Projects Agency under PIPES Grant HR00111920014. This work was performed in part at the Cornell NanoScale Facility, a member of the National Nanotechnology Coordinated Infrastructure (NNCI), which is supported by the National Science Foundation (Grant NNCI-2025233).
The authors declare no financial or other conflict of interest.
###
LEFT:
Paper: https://www.nature.com/articles/s41566-023-01244-7
DOI: 10.1038/s41566-023-01244-7