Artificial intelligence can be run faster and with less energy using a photonic chip
Data centers and high performance computers running artificial intelligence programs such as large language model are not limited by their own computational power. The \”bandwidth\” bottleneck is caused by another problem, namely the amount of data that can be transferred between the nodes.
These systems allow nodes to be separated by up to one kilometer. These systems use fiber-optic cable transmissions because metal wires can dissipate heat while transferring data. As signals are sent between nodes, there is a lot wasted energy in the conversion of electrical data to optical data and back.
Researchers at Columbia Engineering have published a study in Nature Photonics that demonstrates an energy-efficient way to transfer larger amounts of data via the fiber-optic cable connecting the nodes. This new technology is an improvement over previous attempts to simultaneously transmit multiple signals using fiber-optic cable. The new chips do not require a separate laser for each wavelength, but instead use a single laser that can generate hundreds of different wavelengths.
Source:
https://phys.org/news/2023-06-photonic-chip-enables-faster-energy-efficient.html
