Data centers and high-performance computers running artificial intelligence programs are limited less by the computing power of their individual nodes than by the amount of data they can transfer between nodes. It is the latter that currently causes the most problems in the performance of these systems.
The nodes of these systems can be separated by more than a kilometer of cable. Since metal cables get very hot when transferring data at high speeds, these systems transfer the data using fiber optic cables. Unfortunately, a lot of energy is wasted in the process of converting electrical data to optical data (and vice versa) when sending signals from one node to another.
A team consisting of, among others, Anthony Rizzo and Keren Bergman of Columbia University in New York, United States, has devised and demonstrated an energy efficient method for transferring large amounts of data through the fiber optic cables that connect the nodes.
This new technology improves on 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 a single laser to generate hundreds of different wavelengths of light that can simultaneously transfer independent data streams.
The new chip, despite the work it does, is small, as can be seen in this photograph where it appears on top of a coin. (Photo: Lightwave Research Laboratory/Columbia Engineering)
Rizzo and his colleagues discuss the technical details of the new chip in the academic journal Nature Photonics, under the title “Massively scalable Kerr comb-driven silicon photonic link.” (Fountain: NCYT by Amazings)