A VCSEL (pronounced "vixel") is smaller and less power hungry than an edge-emitting laser. But the VCSEL's biggest advantage is its price. VCSELs can be tested on-wafer, which means manufacturers can spot bad components prior to packaging, an advantage that traditional edge-emitting lasers don't have. That doesn't ring the death knell for so-called edge-emitting lasers, however. VCSELs are still taking baby steps toward supporting the 1,310- and 1,550-nm wavelengths required for wide-area networking. And it appears unlikely that they will ever replace the lasers used for long-haul telecom transmissions.

Still, the devices are finding their way into short-reach connections, including the box-to-box links of less than 300 m, also known as very short reach (VSR). Here, they've become an integral part of standards, for 10 Gbit/s VSR connections.

VCSELs were developed beginning in the late 1980s, and it's the "vertical" aspect that makes them different. Conventional lasers are fabricated in semiconductor form, but they have to be diced before they can be tested, because the light emits horizontally, parallel to the wafer's surface. If all the conventional lasers on an undiced wafer were simultaneously activated, the light from most of them would be blocked by their neighbors.

VCSELs tilt the laser cavity by 90 degrees, so that the light shines up from the wafer. That means that all the lasers can be activated prior to dicing, and the bad ones thrown out. Yield is further augmented by VCSELs' small size, which allows more components to be built on each wafer. Moreover, VCSELs are inherently smaller than edge emitters and consume less power.

VCSELs also save money in esoteric ways. They create a more circular beam of light, making them more efficient than edge emitters, whose elliptical beams lose some output power entering a glass fiber. "You can use really cheap alignment techniques using multimode fiber [a less expensive type of fiber] because of the circular nature of the beam," said Jitesh Vadhia, vice president of Zarlink Semiconductor Inc.'s optical division.

VCSELs are so cheap, in fact, that many companies don't bother selling them individually, instead concentrating on laser-module subsystems.

"We've had numerous companies approach us wanting to buy at the wafer level, at the die level, at the TO-can level," said Jake Weise, vice president of international business for Bandwidth9 Inc. "But the higher up the food chain you go, the more revenue you're talking about."

Those cost and power savings made VCSELs affordable enough to move outside the rarified telco network. In particular, the VCSEL helped create an affordable, subwatt laser module — which in turn opened the door to fiber-connected enterprise equipment.

"Fiber optics was formerly a technology provided through telecom companies," said Stan Swirhun, chief executive of Picolight Inc. "The VCSEL put fiber-optics in the hands of the IT manager."

Storage was among the first inroads for VCSELs, as companies realized that high-speed fiber optics could provide faster storage interfaces than traditional copper connections. Commercial VCSELs were sold as early as 1992, and volume shipments began in 1996. "The Fibre Channel storage market really helped commoditize VCSELs," said Tom Fawcett, product marketing manager at Agilent Technologies Inc.

Inexpensive connections
VCSELs spread to more general use in short-reach links, those less than 2 km, and already dominate that market. Again, what made VCSELs attractive was that they were cheap enough for these connections, which typically linked boxes within a large data center or enterprise network. Some fiber optics were already in use here, but they were powered by light-emitting diodes (LEDs), whose speed topped out at around 600 Mbits/s. VCSELs, on the other hand, were first commercialized at speeds of 1 Gbit/s.

"Given the same footprint [as LEDs], you're talking about substantially more data," Vadhia said.

Moreover, short links can use 850-nm lasers. Longer-wavelength lasers produce light that doesn't disperse as quickly, but they're also more difficult and more expensive to build. Shorter connections are able to use the cheaper 850-nm light sources, however, and a VCSEL for that wavelength proved relatively easy to build.

Another benefit to VCSELs is that they're easy to put into arrays — groups of lasers that transmit in parallel to provide a fast, aggregate interconnect. All lasers can be assembled into arrays, but because VCSELs can be tested on-wafer, they can also be diced into arrays rather than assembled.

A VCSEL array is connected to a ribbon cable, usually consisting of 12 fiber-optic cables side by side. The resulting parallel optical interconnect allows for very fast transmission of data from one piece of equipment to another, creating an optical extension to the equipment backplane. Copper cabling is slower, less compac, and can't extend to distances of 300 m.

Faster is less expensive
"You're giving IT managers the ability, in a matchbox-sized [module], to accept 30 Gbit/s of data and transmit for 300 m. Five years ago, this was just unheard of," Vadhia said.

Moreover, the modulation — which determines the data throughput of a laser — doesn't have to be so fast, because data throughput is split among the parallel channels. That's an important point because a faster laser is expensive; VCSELs have only begun hitting the 10 Gbits/s mark, let alone 30 Gbits/s. "If you're looking at a pretty short reach where you can directly modulate the laser, you can make a pretty inexpensive interconnect," said Jeff Livas, chief technology officer of the transport division of systems vendor Ciena Corp.

The next step for VCSELs is into the wide-area network, but vendors are quick to caution that VCSELs aren't likely to replace all lasers in networking. VCSELs are praised for their low power consumption, but their output power is likewise low, meaning they aren't strong enough to carry the long-reach connections in telecommunications networks.

That's true of most lasers, but in the case of a VCSEL, the extra amplification cancels some of the devices' inherent benefits. "The VCSEL would become a light bulb inside a more complicated system," said Jack Jewell, chief technology officer of Picolight.

Still, VCSELs might find homes in the metropolitan and access networks, where distances extend to the tens of kilometers. These links are beyond the reach of 850-nm lasers, so VCSELs are undergoing some changes to extend their wavelengths. Wavelengths around 1,310-nm, in particular, would fit well into campus interconnects or even fiber-to-the-home applications, and the volumes could be as high as they are for 850-nm devices, Picolight's Swirhun said.

Getting there is a materials issue. Indium gallium arsenide (InGaAs) provides the right optical gain for 850-nm VCSELs but not for 1,310- and 1,550-nm parts. One obvious alternative is to use indium phosphide (InP), but the brittle, expensive material isn't ideal for what is supposed to be a low-cost, mass-produced part. Also, InP has worse thermal properties than GaAs, so that the devices would heat up, diminishing their optical gain.

Around 1996, Hitachi researchers developed a 1,310-nm VCSEL based on indium gallium arsenide nitride (InGaAsN). The material can be grown on a GaAs substrate, "and by really stretching it, you can make the gain wavelength around 1,310 nm," said Wupen Yuen, vice president of R&D for Bandwidth9.

That's not to say the material is easy to work with. "The nitrogen doesn't want to stay at the right crystal site. It wants to diffuse around," Yuen said.

As a result, most 1,310-nm VCSELs actually reach only 1,260 nm in wavelength. "The longer the wavelength, the more nitrogen or indium is required, and material quality suffers," Jewell said. Note that 1,260 nm is perfectly OK; standard wavelengths in the 1,310-nm range run from 1,260-nm to 1,360 nm.

Power and durability
Even after selecting the right material, there remains the question of getting the devices' output power and durability up to snuff. Most vendors don't expect the 1,310-nm VCSEL market to take off until 2003. "To me, the challenges are still very significant for at least a year or two," Fawcett said. Still, 1,310-nm VCSELs are making progress. Picolight demonstrated its 1,310-nm modules earlier this year, running OC-48 (2.5 Gbit/s) connections over 10- to-20-km links.

For 1,550-nm VCSELs, researchers face some of the same challenges. Here, InGaAsN gives out, leaving InP as the only option, "so again, everything plays against you," Yuen said.

Still, a handful of companies has tackled the 1,550-nm challenge including CoreTek (acquired by Nortel Networks), Bandwidth9 and Applied Optoelectronics Inc.

Bandwidth9 is alleviating InP problems by using the tried-and-true gallium arsenide (GaAs) for one of the two mirrors inside the VCSEL. The innards of a laser include two reflective surfaces; light bounces between them until it "lases" at a particular wavelength. For a 1,550-nm VCSEL, the gain material on one of the mirrors needs to be InP, which in turn forces the substrate to be InP, because no other material will match InP's lattice. But the second mirror could be made of anything — hence, Bandwidth9's choice of GaAs.

Jewell noted that 1,310-nm and 1,550-nm VCSELs are "almost equivalent in technical difficulty," but that doesn't mean every vendor is racing to develop the parts. Many don't believe the 1,550-nm market will ever be substantial.

Zarlink's Vadhia said. "The majority of the action is going to be in the short and medium haul."

In addition to new wavelengths, VCSELs are advancing on the switching-speed front, and here, another problem develops. VCSELs that operate at 10 Gbits/s are going to need new packaging. The TO can that houses VCSELs today develops "terrible" parasitics at 10 Gbits/s, said Jim Tatum, VCSEL technical manager for Honeywell.

"At that point, you have to design a completely different mechanical package that's not the same form factor as a TO can," Tatum said. In turn, the new package changes the mechanics of the laser, which will trigger a need for new laser drivers. "At 10 Gbits/s, the entire infrastructure changes."

"The TO can runs out of steam at 2.7 Gbits/s," Zarlink's Vadhia agreed. "Ten Gbits/s will probably be either in some form of chip-level packaging or some high-speed RF packaging."

Honeywell's answer was to develop a hybrid microwave-ceramic package that mounts onto the edge of a pc board. The new material required a new epoxy, as TO can epoxies weren't sufficient, and welding wasn't possible, Tatum said.

But as with 1,550-nm devices, some vendors don't have immediate plans for a 10-Gbit/s VCSEL.

"We're not big believers" in 10-Gbit/s VCSELs, said Agilent's Fawcett. Agilent's work has shown the devices would be limited to less than 100 m in range, compared with the 10-km links that many 10-Gbit/s lasers serve. And the devices likely would suffer from reliability problems, he said.