SANTA CLARA, Calif.--It's decision time for a handful of companies aiming to bring silicon photonics to the market to reduce the costs of 100 Gbit/second links. A key standards group will create a draft spec by May, but so far none of the conflicting proposals from silicon photonics vendors appear to have adequate backing.
The IEEE 802.3bm group is setting a range of standards for low cost 40 and 100 Gbit/second Ethernet. They include a 4 x 25G module interface that silicon photonics vendors hope to use in 2014 to provide low cost, high speed links inside and between racks in data centers.
"Our biggest challenge is the different proposals from silicon photonic vendors are fairly evenly separated in votes," said Dan Dove, a director of technology at Applied Micro Circuits Corp., who chairs the group. "We have a lot of decision making in the next couple months," he said, noting proposals need a yes vote from three-quarters of the engineers attending.
"It's not uncommon at this stage in the standards process to have disparity, but…to get 75 percent of the room to agree may be difficult," said Dove, a veteran of IEEE standard efforts.
The sticking point seems to be modulation schemes. In straw polls to date the group of 100 or so attending the meetings has been fairly evenly divided between four options:
Fujitsu supports discrete multi-tone;
Broadcom, Cisco and Luxtera back pulse amplitude modulation;
Avago and Luxtera support parallel single mode;
Huawei, IBM and Kotura want wavelength-division multiplexing.
"They all have their strengths and weaknesses, said Dove. "I would say the main sticking point is just sort of political logistics."
Click on image to enlarge.
May will be a key month in the effort to reach a consensus on low-cost 100G Ethernet for big data center and carrier switches.
in terms of WDM, it nicely lends itself to PCI Express protocol- where lanes are bonded together (from x2 to x16)to increase port speed. WDM makes repurposing simple for transceiver makers. And since PCIe is inherently a short reach technology, would only complement the Ethernet space.
Would seem like a 'win-win-win' to me! :-)
Nice article Rick. I should mention that the bullet items separating proposals out by company were not provided by me, but probably gleaned from the IEEE P802.3bm website.
Within the IEEE, participation is by individual and not by company. I did not, nor would I characterize proposals by company name.
Standards are good for commerce and the consumer, at a high level. However, if creating standards limits the R&D options for the relatively few major players spending billions on Si Photonics, then this could be counter-productive in the near term. Is it too early to be locking things down in 100 Gb/s standards?
This viz was created before one of the GOP debates to see which of the potential candidates had a better job creation record.-"Peter F. Spittler":http://presentationsolutions.org/2012/11/01/phil-melugin/
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.