Infinera Declares WDM War
Will the startup change the world? Probably not. But its long-haul WDM system, starting its first carrier trials in the third quarter of this year, just might uproot the current thinking surrounding optical networks.
Infinera's system, called the DTN, can do both Wavelength Division Multiplexing (WDM) and add/drop multiplexing (ADM). The radical part is that Infinera does this by forcing every wavelength through an optical-electrical-optical (OEO) conversion.
That's a break from mainstream WDM thinking, where OEO is considered cumbersome. "The fundamental assumption of this industry for the last seven years has been that OEO was expensive," says Infinera CEO Jagdeep Singh. "The whole industry has been organized around trying to get rid of OEO."
To cheapen OEO -- in a good way -- Infinera designed two densely packed indium phosphide (InP) chips, one for transmission and one for receiving. Each crams 10 channels of 10-Gbit/s OEO processing into a space less than 5 mm2 -- in the case of the transmitting chip, that's 10 lasers, 10 modulators, waveguides, and an optical multiplexer.
The integration saves money by throwing out the packaging of each element. A laser chip, for example, costs $20 to $50. But the packaging and assembly to complete a discrete laser can raise the price to $1,000, Singh says.
"The reason OEO is expensive is that you have lots of discrete components, each one of which has a significant packaging cost," says Singh. "If you take all those elements and integrate them onto a single monolithic piece of indium phosphide, it dramatically reduces the cost."
Wielding these chips on its linecards, the DTN can pack 40 channels of 10-Gbit/s traffic into a half-rack chassis, Infinera officials say. That means 80 channels per rack. By contrast, the basic version of the CoreStream Agility from Ciena Corp. (Nasdaq: CIEN) takes two racks to handle 80 channels (see Ciena Launches CoreStream Agility).
The compactness is nice, but why obsess about rescuing OEO? Because the alternative, purely optical transport, is a headache.
Digital electronic signals don't have to hit precise levels; all that matters is that the voltage is in the right ballpark for a "0" or a "1." Optical signals aren't like that. They're processed in analog form, meaning metrics such as the power level must be controlled to precision. And a wavelength that gets added or dropped at one node can affect power levels on channels throughout the network.
That means a WDM network has to be planned meticulously, and it takes torturous fine-tuning to install. Moreover, the network has to include components to control the light: dispersion compensators, gain-flattening filters, and the like. OEO avoids the tweaking and the extra components, making for a simpler network.
But what really makes the DTN interesting is one side effect of its OEO nature. Carriers using the DTN can change add/drop assignments at will, simply by swapping out linecards, without having to do the planning or tweaking associated with all-optical WDM. That could be Infinera's trump card, because it means DTN is not only cheap, but gives carriers a level of flexibility they've never had.
Infinera's mystique comes from keeping all this a secret for so long. The 200-employee company raised $150 million and a priceless amount of buzz since launching as Zepton in 2000. (See Infinera's Amp-less Ambition, Infinera Shoots for the Moon, and More on Infinera (née Zepton).)
Infinera has the credentials to make a change in WDM. Singh was formerly head of Lightera Networks, the startup whose acquisition turned Ciena into a WDM powerhouse. He's surrounded himself with bigwigs from the components world, such as Dave Welch, the CTO of SDL Inc., which was sold to JDS Uniphase Corp. (Nasdaq: JDSU; Toronto: JDU); and Fred Kish, who headed the InP semiconductor efforts at Agilent Technologies Inc. (NYSE: A).
It's hard to fully predict the DTN's impact, because Infinera won't discuss prices, and no carrier has tested the system outside of Infinera's labs. But it doesn't take Columbo to figure out who stands to be hurt from all of this.
"Ciena's got to be afraid of it," says Scott Clavenna, chief analyst for Heavy Reading, Light Reading's paid research arm. "If [the DTN] works, the economics are so radical that it would be hard to come up with a reason why someone would buy a traditional WDM system."
That's trouble for anybody in the WDM camp, but it gets worse for Ciena. Because the DTN's chips include crossbars, the system can do wavelength switching, bringing it into partial overlap with Ciena's CoreDirector. CoreDirector does STS1 (51 Mbit/s) switching as well, but it's being used for wavelength switching about 70 percent of the time, Clavenna says, "so it sort of hits Ciena in two directions."
Certain components could be affected, too, because the DTN doesn't need dispersion compensators and similar devices. And its OEO nature means the signal can be regenerated at every node, which could spell trouble for amplifiers and ultra-long-haul equipment.
Singh downplays all these effects and lays no claim to usurping anyone. "It's not a battle of technologies. What we're doing is a new tool that, when coupled with the network that's already available, will allow carriers to build more cost-effective networks."
It should play in Peoria
In fact, the DTN will probably be used as an add-on at first, exploiting its ability to make add/drop decisions easy. That's going to make it useful for delivering services to second-tier cities, the kind that weren't otherwise economical for traditional WDM drops, Infinera officials say. Carriers can use a DTN to drop just one wavelength to a place like Albuquerque, adding more if demand for services grows.
Those kinds of incremental changes might be Infinera's best business for a while. The company's product arrives at a time when long-haul sales are relatively slow and not likely to explode. Moreover, long-haul isn't price elastic; a carrier won't stock up on WDM nodes just because someone holds a Memorial Day sale.
Then there's the question of competition. Singh might be correct in claiming no one else has achieved Infinera's level of chip integration, but some companies are trying. Clavenna says he's heard of projects put on hold during the recession that are thawing now.
"There's a lot of work that's been going on, integration that is sort of lying fallow for some catalyst to get it going. This could be that catalyst," Clavenna says.
Companies would be hard-pressed to match Infinera's achievement, particularly since InP is a difficult material to work with. But if Infinera is that radical, then a close alternative should be enough to nibble away some market share at the low end.
"I've heard there are some other [attempts] out there to build integrated circuits as cheap and as functional. Maybe not quite as compact. So you could still build a system that has hundreds of Gbit/s of capacity," Clavenna says.
— Craig Matsumoto, Senior Editor, Light Reading