Hewlett Packard Labs Research Brings Photonics to Content-Addressable Memories

[Labs_Editorial]

In recent years Hewlett Packard Labs’ Large Scale Integrated Photonics Lab (LSIP) has done extensive research in the field of silicon photonics. This research, driven by the increased demand for high-bandwidth low-latency applications, is aimed at increasing the speed as well as reducing the energy consumption of moving data within data centers and HPC (high performance computing) systems. These areas of focus converged in one of the latest projects conducted by LSIP.

In a paper published in the Nanophotonics journal in late October, LSIP’s researchers describe how content-addressable memories (CAMs) can be fabricated using Labs’ exascale silicon photonics PDK (process design kit). CAMs are a class of high-speed memory that searches its entire content within a single cycle, a property that is particularly handy for routers, switches, and SmartNICs in data centers and HPCs, as it allows high-rate packet routing and security analysis.

LSIP’s study shows that the proposed optical CAM designs can operate ten times faster than electrical CAMs. The energy consumption of these optical CAM designs, however, still requires some work. While CAMs consumed more energy using optical technology than electrical, Yanir London, the lead researcher, said the team did manage to identify a set of hardware and architecture modifications that  could narrow the gap between the two environments. London described this outcome as “a step of great significance” toward embedded CAMs circuits in the optical link that operate at signal data rate, therefore resolving current performance and latency bottlenecks.

Yanir London“We were able to experimentally demonstrate for the first time two different flavors of optical CAMs, each leveraging distinctive multiplexing resource,” London said. “That was not trivial to execute, but we were able to showcase small-sized optical CAM architectures. These experiments confirm the potential of our silicon photonics technology in fabricating large-scale CAMs operating at much higher speed compared to electrical CAMs. Moreover, our optical CAM designs eliminate the overhead of optoelectronic conversion as well as data rate down-conversion, thus reducing the total energy consumption.”

Now that multiple architectures have been developed, London said the team will embark on a new set of research.

“We still have work to do,” London said. “The next steps are addressing energy and scaling aspects and extending the applicability of the optical CAMs, through in-lab efforts as well as academic collaborations.”

The paper, called “Multiplexing in Photonics as a Resource for Optical Ternary Content-Addressable Memory,” was developed by Yanir London, Thomas Van Vaerenbergh, Luca Ramini, Antoine Descos, Luca Buonanno, Jinsung Youn, Can Li, Catherine E. Graves, Marco Fiorentino, and Raymond G. Beausoleil.