The 400-Gig Vision
The subject came up at last week's Ethernet technology summit, put on by The Ethernet Alliance , where an informal audience poll showed a lot more interest in 400-Gbit/s Ethernet than in Terabit Ethernet. (See Facebook: Yes, We Need 100-GigE.) Technological barriers probably stacked the vote, but Terabit might not be as far-fetched as Bob Metcalfe made it sound last year. (See Bob Metcalfe on the Terabit Ethernet.)
That's the impression I got after Finisar Corp. (Nasdaq: FNSR) engineer Chris Cole gave a talk on the possibilities for 400 Gbit/s. This wasn't a product introduction or a standards pitch. The goal was to create a feasible 400-Gbit/s plan from today's building blocks.
I'll give you his punchline first: "25 Gbit/s is where you're going to make the money for the foreseeable future," he said. (Half-jokingly. Don't run off and launch a startup based on this.)
OK, 25-Gbit/s components aren't really today's building blocks -- did I mention that was my phrase, not his? -- but they're clearly on the way. The next step in 100-Gbit/s links will be to develop 25-Gbit/s directly modulated lasers, something that's shown up in enough research papers to be a likely bet, Cole said.
After that would come 25-Gbit/s electrical interfaces. At that point, the industry could halve the Kansas-like width of the 100-Gbit/s CFP module, Cole said.
All that 25-Gbit/s work could be the foundation for a 400-Gbit/s standard. Cole, who I think said he was drawing from a 2007 Telcordia Technologies Inc. proposal, described a scheme of 16 wavelengths running at 25 Gbit/s each, grouped into four clusters of four wavelengths.
The resulting block diagram looks a lot like the 25-Gbit/s-based 100-Gbit/s module. But it's wider -- probably wider than a CFP.
The alternative is to use a slower bit rate and fancy modulation, as companies are doing at 100 Gbit/s already, with dual polarization quadrature phase shift keying (DPQPSK! Yesss!) and the like. Whether that's the answer, as opposed to going the 25-Gbit/s route, "will mostly depend on the state of integration of the photonic components," Cole said.
That was the bulk of his talk, but he briefly noted that "to do Terabit, we'll have to do all the tricks," meaning faster optics, faster electronics, and multibit modulation. A standard involving 40 lanes of 25 Gbit/s apiece would at least multiply to the right figure without resorting to any sci-fi advances. It would take a lot of work, though, particularly on the electrical side.
— Craig Matsumoto, West Coast Editor, Light Reading