Silicon photonics are on the verge of a more serious market presence, and while that's something to take notice of, it's not going to change the world overnight.
That's pretty much what I concluded after OFC/NFOEC a year ago. (See Silicon Photonics Signals Red Alert for 100G.)
But Intel Corp. has rekindled the hype.
Last week at the Open Compute Summit, the company said it's handed engineering samples of 100Gbit/s photonic interfaces to Arista Networks Inc. and other customers. These aren't pluggable modules, but on-board optics for data-center applications -- connecting a server to a top-of-rack switch, for example.
Then there's Cisco Systems Inc., which has been showing (but not really talking about) its own 100Gbit/s optical module, a pluggable format called CPAK. It's a smaller alternative to the CFP client-side 100Gbit/s interface that's shipping today, and it's presumed to be using silicon photonics from the Lightwire acquisition. Two Cisco sources mentioned during the fall that CPAK shipments were "imminent," but reports now say CPAK will be debuting sometime this summer.
Is either item really that big a deal yet? What Intel showed is already possible without silicon photonics. Intel did have its optics mounted directly onto the board, as opposed to putting them in a transceiver module -- and while that offers a density improvement, it's also been done. Finisar Corp. showed it at OFC/NFOEC last year, as Chris Cole reminded me.
(He's Finisar's director of transceiver engineering, but we were talking about his own opinions about the announcement. You can read more in his comments on this EE Times thread.)
Cole also doubted that Intel's 100Gbit/s is that close to commercialization, given that the company doesn't expect 100Gbit/s server interfaces to be a big deal for a few years yet.
CPAK, meanwhile, is a Cisco-only module. Most people think Cisco's doing CPAK mainly as a way to keep its module margins high. Optics have been a pretty good source of Cisco profits for years, and the gray market of off-the-shelf devices might be poised to eat away at some of that profit.
"It makes it easier for them to defend having Cisco transceivers," says Karen Liu, an analyst with Ovum Ltd. (See Gray Market Optics Get a Brand Name and Cisco Modules Pay Off -- for Other Companies.)
CPAK is supposedly the result of Cisco's impatience with industry-standard 100Gbit/s modules, which are moving to the CFP2 and CFP4 multiservice agreements (MSAs) eventually. But as analyst Alex Henderson of Needham & Co. points out, CPAK won't arrive that much earlier than CFP2 (he says both are waiting for development of a particular physical connector) and doesn't blow away CFP2 in power consumption; both should be around 7.5 W, he thinks.
So, in both cases, you can argue that the revolutionary nature of silicon photonics hasn't shown itself yet. The power and density haven't severely thrashed the competition -- and because these are 100Gbit/s devices, I'm not sure silicon will crater the price, either. It's not time to panic.
What would bother me more, if I sold optical components, is the number of competitors coming to transceivers due to silicon photonics.
Analyst James Kisner of Jefferies & Co. Inc. listed several in a recent report.
He thinks Kotura Inc. is going to launch a QSFP+ pluggable transceiver for less than $1,000 late this year. He believes Acacia Communications Inc., the startup developing 100Gbit/s coherent transceivers, is using silicon photonics. And he thinks Luxtera Inc., which recently found a manufacturing partner in STMicroelectronics NV, might create a CFP competitor.
Even if silicon photonics offers no immediate advantage, you're ending up with more competitors in a space that's always fighting for decent margins. I suppose that's nothing new in optical transceiver sales, and it's bound to make for a dynamic OFC/NFOEC.
— Craig Matsumoto, Managing Editor, Light Reading