Laser Blazers

VCSELs take over the world * What are they? * VCSELs vs the rest * Breakthrough technologies

November 27, 2000

28 Min Read
Laser Blazers

As carriers roll out optical technologies around the world, the demand for lasers is bound to rocket. There's no doubt that they'll be needed in vast numbers as optics starts to penetrate metropolitan environments and the "last mile" to the home and business. And there's a good chance that a lot of these will be VCSELs -- vertical cavity surface emitting lasers.

VCSELs (pronounced "vixels") have already proved to be the laser of choice for short, 850 nanometer wavelength transmissions. In fact, they've blown away other types of laser in terms of the quality of the light they produce and the cost of manufacturing.

Now, vendors are hoping to repeat their success at longer wavelengths -- 1310 and 1550 nm. If the track record of short-wavelength VCSELs is anything to go by, companies making long-wavelength versions could pose a serious threat to the traditional laser makers.

In this report, Light Reading sets out to identify the breakthrough technologies. First, there's a technology backgrounder, describing what a VCSEL is. Next, we outline the advantages of VCSELs over other transmitters. After a few words and numbers on applications, we get to the real meat of the article -- who are the players and what technological advantages do they have?

Intriguingly, there's a company called Vixel Corp. that doesn't actually make VCSELs. To find out why, read this report.

Here's a hyperlinked summary of what's on each page:

What Is a VCSEL?
VCSELs vs the Rest
New Applications
Short-Wavelength VCSELs
1310nm VCSELs
1550nm VCSELs

Simply knowing what the acronym stands for is a good start at understanding what a VCSEL is. VCSEL is shorthand for "vertical-cavity surface-emitting laser" and yes, it's got a vertical cavity, and yes, it's a surface emitter, which means that it spews light out of its top surface. This sets VCSELs apart from all other lasers, classed as edge-emitters, which have a cavity that lies in the plane of the wafer (usually thought of as horizontal) and spit light out of the side.

The cavity, by the way, is a crucial part of all lasers. Light is pumped into it and bounces to and fro between mirrors, at the top and bottom in the case of VCSELs. A so-called "active region" between the mirrors amplifies the light on every cycle so that it builds in intensity until… (fanfare) lasing occurs.

Above the threshold for lasing, light adopts a kind of group mentality. One specific narrow band of wavelengths that's in resonance with the cavity is affected. It's now emitted "in phase" with the rest of the light in the cavity. The best way to understand what "in phase" means is with an analogy: It's like soliders marching in step. And like soldiers marching in step, the group is much more powerful than the individual. Soldiers marching over a bridge are told to break step so that they don't cause damaging resonances in the bridge structure, for instance.

So, the basic physical operation of a VCSEL is pretty much the same as a standard laser. But rotating the geometry though 90 degrees makes manufacturing an entirely different proposition.

In an edge-emitter, mirrors are formed by cleaving the semiconductor crystal (breaking it along an atomically-smooth plane). Doesn't work in a VCSEL. Devices that have a simple, single interface to the air, are crazy about electric current -- they need a lot of it to operate. That's because the volume of active material in a VCSEL tends to be much smaller than in a standard edge-emitting laser. To make a device that operates with a more moderate amount of current, much higher reflectivity mirrors are needed (more than 99.9 per cent).

Incidentally, the small volume of active material is the main reason that, even today, most VCSEL vendors have trouble getting really high powers out of their devices.

It wasn't until Jack Jewell and his associates at Bell Labs, the research arm of Lucent Technologies Inc., came up with a way of making very high reflectivity mirrors, that VCSELs really became a practical proposition. The process they pioneered in 1989 was the first time anyone had made a high reflectivity mirror that was all semiconductor and was made in a single process, says Jewell, who is now chief technology officer at PicoLight Inc., a VCSEL manufacturer.

Today, most VCSELs use mirror designs based on the structure proposed by Jewell and his colleagues. Called "Bragg mirrors" or "distributed Bragg reflectors," they work in the same way as fiber Bragg gratings. The mirrors comprise alternating layers of semiconductors, one with a high refractive index, the other with a low refractive index. The thickness of the layers has to be controlled precisely (technically, each layer must have the thickness of a quarter wavelength at the lasing wavelength in that particular material). When these conditions are satisfied, the reflections at all the layer interfaces add up constructively, to create a strong reflected beam. The more layers in the structure, the higher the reflectivity.

The process of fabricating the 40 or more layers of a VCSEL is termed "epitaxial growth". It's often refered to in the industry as "growing" a laser. There are two main techniques: molecular beam epitaxy (MBE) and vapour phase expitaxy (VPE). Both of these techniques build up the laser structure one layer at a time on top of a substrate wafer. When growth is complete, more processing is needed to add electrical contacts and other refinements, but at this stage the wafer can already lase, simply by pumping it with light from another laser.

The fact that the lasers don't need cleaving into individual devices in order to function, offers the advantage of testing them on-chip. More on that next in VCSELs vs the Rest.

The advantages of VCSELs operating at 850 nanometers were so compelling that VCSELs replaced all standard transmitters in local area networks within two years of first commercial availability. At 850nm, VCSELs deliver better performance at a lower cost than other types of transmitter. Here are the details:

  • VCSELs emit a uniform, narrow, circular beam, which simplifies optical system design considerably. It's easy to couple into a fiber -- coupling efficiencies into a fiber are upwards of 80 per cent. In contrast, edge-emitters have an elliptical beam, which is strongly divergent.

  • A VCSEL laser cavity is very short, compared to that of an edge-emitter. So, when the laser is turned off, it takes less time for the remaining light to exit the cavity. As a result it is possible to modulate the laser (turn the beam on and off) directly, which avoids the need for an external modulator. Devices with direct modulation speeds of 2.5 Gbit/s are commonplace -- faster devices could be available soon.

  • Thanks to surface rather than edge emission, VCSELs can be tested on-chip, before the wafer is diced up and packaged into individual components. This means that dud devices can be weaned out at an early stage in the processing, which saves a lot of cash.

  • These devices are small, so it's possible to squeeze huge numbers of them onto a wafer. Cielo Communications Inc. says it can cram 20,000 individual devices onto a three-inch wafer.

  • VCSELs are very efficient at turning electricity into photons. The record is 57 per cent, but typical efficiencies are in the region of 6 to 25 per cent. As a result they consume less electrical power than an equivalent DFB (distributed feedback) laser. For example, a DFB laser might draw 60 milliamps, while a VCSEL would only require 15 mA to produce the same optical output.

  • Since they dissipate less electrical power as heat, VCSELs don't require temperature control. That saves money by eliminating the need for a Peltier cooler or other cooling gadget and the associated circuitry. Not only that, but VCSELs tend to be more reliable, because the chip doesn't heat up its surroundings.

  • Finally, there are two key opportunities for integration. First, VCSELs can be manufactured in arrays of two, four, or even sixteen devices on the same chip. The devices are made to comply with a grid, so that they can be lined up with the fibers in a fiber ribbon, for example. The array is then placed inside a single package, resulting in a cheap, high-data-rate transmitter. VCSELs can also be integrated side-by-side with detectors, to make transceivers.

  • A second form of integration is to drop the VCSEL on top of its driver IC using one of several techniques. Flip chip bonding, for example, uses solder bumps on the IC to connect to the chip above it. This makes it possible to shrink the packaging, which translates into cost savings.

    Sounds great. What's the catch?

    Unfortunately, the materials used to make 850nm VCSELs don't translate well to longer wavelengths. As yet there are no commercial devices that operate at the telecom wavelengths of 1310 and 1550 nm. For more on why long-wavelength VCSELs are so darn tricky to make, see Challenges, coming next.

    Short-wavelength VCSELs have already won the contest and replaced all the other types of transmitter for short-reach applications. This page isn't about them. It's about their long wavelength cousins, which have yet to reach the market.

    Long-wavelength VCSELs could potentially benefit from all the advantages listed on the previous page under VCSELs vs the Rest. But to date, most attempts at making these devices have needed complicated manufacturing processes, which cost time and money and ultimately eliminate the cost advantage altogether. Performance in terms of operating temperatures and output powers has not been up to scratch either.

    The problems all start with materials. Short-wavelength devices are based on a material called gallium arsenide. (It's a compound of gallium and arsenic, so these chips aren’t very tasty.) However, telecom lasers operating at 1310 and 1550 nm are all made from a different material: indium phosphide.

    As noted on page 2, a VCSEL consists of two key sections: an active region and two mirrors. Ideally, manufacturers want the active region and the mirrors to be grown from the same material so they can produce the entire structure in a single process. The desired wavelength dictates the choice of material for the active region. But it's difficult to make mirrors out of indium phosphide.

    Mirror, mirror…

    To understand why it's so tricky to make mirrors out of indium phosphide, it is necessary to understand a concept called lattice matching. (Technophobes can skip the following section…)

    Lattice matching is an essential part of the epitaxial growth process. All semiconductors are composed of atoms arranged in a regular crystal, also called a lattice. To grow one material on top of another, their lattices must have nearly the same dimensions; otherwise, defects form in the crystal. Defects are bad because they kill light emission.

    The requirement for lattice matching, as it is called, restricts the choice of materials for making the mirrors. Usually gallium and arsenic are added to indium phosphide, to create a compound with a different refractive index, denoted as InGaAsP (In = indium, Ga = gallium, As = arsenic, P = phosphorus). But only certain combinations are allowed. And of the combinations that are allowed, the resulting refractive index is not that different from indium phosphide. This means that the Bragg mirrors don't work very efficiently. Lots of layers, a few hundred per mirror, are needed to create high reflectivity. Though it is possible to grow this many layers, it consumes a lot of raw material, takes a long time, and requires painstaking control of the process parameters. In short, it's not very cost effective.

    In the past couple of years, researchers have come up with ways of circumventing the indium phosphide mirror problem. Right now the first long-wavelength VCSELs are sampling, and several products are due for volume production next summer.

    Despite all of these challenges, the market for VCSELs is set to explode in the next few years, according to a recent study. For the details, seeNew Applications, coming next.

    Coming from nowhere a few years ago, the market for VCSEL transceivers is set to explode as new applications appear on the scene, according to a report released earlier this month by Electronicast Corp., an independent market research firm.

    Electronicast's study includes VCSELs at 850 and 1310 nm. It does not include devices at 1550nm, which will take longer to appear and, in any case, will slot into the WDM market, which operates under different rules. In WDM (wavelength-division multiplexing), the price of the laser isn't a significant part of overall equipment costs, so the price advantage of a VCSEL doesn't have much impact.

    1999 VCSEL Consumption ForecastRight now, the main applications for VCSEL transceivers are Fibre Channel (a protocol for storage area networks) and Ethernet, with intrasystem links coming in third place. These markets were worth $262 million in 1999, three times more than in 1998, says Electronicast.

    2004 VCSEL Consumption ForecastBy 2004, the overall market for VCSEL transceivers is predicted to grow to $3.4 billion.

    There are two points worthy of note in these figures. One, the fastest growing application is intrasystem links, which is expected to increase in value by a huge 77 per cent per annum on average over the next five years. Today, connections among routers, switches, and hubs in the central office are often implemented with more expensive DFB lasers at 1310nm. When 1310nm VCSELs become available sometime in 2001, they will start to take over this area, according to Bob Mayer, VP of business development at Cielo Communications Inc., a company developing a 1310nm VCSEL.

    Two, a brand new application will materialize: low-cost solutions for Very Short Range (VSR) Sonet. It's designed for transporting high data rates over Sonet, and aims to make short Sonet links as cheap as Ethernet (see OIF Specs Component Standard ). Electronicast reckons that VSR Sonet will be worth $900 million in 2004, increasing to $3.7 billion in 2009.

    2009 VCSEL Consumption ForecastThe Optical Internetworking Forum is working on a VSR Sonet physical layer specification for moving 10 Gbit/s of data from one place to another over distances of up to 300 meters. Four options have been proposed, all of which are underpinned by VCSELs:

    Single fiber solutions:

  • a single VCSEL operating at 10 Gbit/s (to go the distance at this speed requires the use of a long-wavelength device)

  • four VCSELs, each sending 2.5 Gbit/s over a different wavelength

    Multifiber solutions

  • four VCSELs at 2.5 Gbit/s over four fibers

  • twelve VCSELs at 1.25 Gbit/s over twelve fibers (the advantage here is that using slower modulation speeds allows signals to go further, plus one channel can be used to protect against failure of a single VCSEL in the array)
    For more details of Electronicast's report, see more details of different categories of VCSEL and who's making what, follow these hyperlinks:

  • 1550nm VCSELs

    Short wavelength means around 850 nanometers. It's no coincidence that this is the wavelength at which silicon detectors are most sensitive.

    However, this wavelength is not the optimum for transmitting signals over optical fiber. At 850nm, attenuation is pretty high, so signals can only go up to a kilometer over singlemode fiber, and less than this over multimode.

    Still, there are plenty of applications where a long range is not required, such as Fibre Channel, Ethernet, and intrasystem links (among switches, routers and hubs inside the central office).

    VCSEL arrays, comprising multiple lasers in the same package, make it possible to transmit high data rates cheaply. These days, pretty much every manufacturer supplying single VCSELs at 850nm also offers VCSEL arrays at the same wavelength.


    Technologically, one vendor's 850nm VCSEL is pretty much the same as any other, says Mike Hartmann, director of marketing at E2O Communications Inc. E2O's strategy is simple -- manufacture in volume to achieve low prices.

    Other 850nm VCSEL manufacturers agree. "We're as good as anyone else, but there's no reason to say our VCSELs are better," says Geoff Callow, a technical marketing specialist at Mitel Corp. (NYSE/Toronto: MLT). "It's not what you have, it's what you do with it that counts," he adds.

    In the end, choice of an 850nm single VCSEL boils down to a few basic variables, such as price, optical output power, price, modulation speed, and the color of the salesman's socks.

    When it comes to making VCSEL arrays, there are a few more variables. The choice also depends on the equipment for which the transmitter is intended. For instance, one vendor might sell a 1x12 array with an aggregate capacity of 10 Gbit/s. (Each individual device operates at 1.25 Gbit/s; the rest is overhead). Another vendor can offer the same aggregate capacity from a 1x4 array by using devices that each operate at 3.125 Gbit/s. Yet another product is available that uses a combination of 2.5 Gbit/s lasers in a 1x8 array. Standards are still evolving in this space.

    A couple of vendors are working on methods for increasing output power. They aren't likely to compete with the 850nm vendors because longer distances can be reached more effectively by going to longer wavelengths rather than by increasing output power. High-power applications would include 980nm lasers for pumping fiber amplifiers (still classed as short-wavelength VCSELs because they use the same materials as the 850nm devices).


    The Established Players

    Emcore Corp.

  • Gained VCSEL expertise through its acquisition of Mode, a spinoff from Sandia National Laboratories in 1997

  • In August 2000, announced the availability of a 1x4 VCSEL array capable of up to 10 Gbit/s and a 1x12 array with a data rate of 30 Gbit/s

  • Max data rate 3.125 per channel

  • Has a joint development and marketing agreement with JDS Uniphase Corp. (NYSE: JDSU)

  • Customers include Agilent Technologies Inc. (NYSE: A), which began sampling parallel optical transceivers in November 2000 (see Agilent Unveils 30-Gig Optic Modules). The duo have a three-year supply agreement

    Honeywell Inc. (NYSE: HON)

  • Honeywell Sensing and Control was the first company to commercialize VCSEL technology

  • It's the world's largest VCSEL component supplier, it claims. Electronicast's study backs this up

  • Sells a lot of VCSELs for sensing and other non-telecom applications

  • Plans to offer VCSEL arrays in the future (no specs available yet)

    Gore Photonics Inc., a subsidiary of W.L. Gore and Associates Inc.

  • Gore got into VCSELs when it acquired a company called Optical Concept, which was started by, and just up the road from, the University of California, Santa Barbara (UCSB)

  • Offers single VCSELs and 1x12 arraysInfineon Technologies AG (NYSE: ISX) (formerly Siemens Semiconductor)

  • Makes VCSEL chips for internal consumption

  • Sells a transceiver called PAROLI containing a 1x12 VCSEL array with an aggregate data rate of up to 15 Gbit/s

  • Could be having trouble with its VCSEL arrays. Brightlink Networks Inc. says it ordered some PAROLI modules in 1999 and still hasn't received them. Infineon did not respond to requests for an interview.

    Mitel Corp. (NYSE/Toronto: MLT)

  • Mitel is the second largest manufacturer of VCSEL chips. Electronicast reckons that between them, Honeywell and Mitel produce 75 percent of all VCSEL chips.

  • In 1996, Mitel acquired the foundry of ABB Hafo in Sweden, where it now produces VCSELs

  • Offers single VCSELs and arrays up to 1x12

  • Mitel contends it's got the edge when it comes to making VCSEL arrays. It has a technique for aligning a 1x4 array end-on with a second 1x4 array, to create a 1x8 array. (Add a third chip to create a 1x12 array and so on). That's a major strength because it improves the yield of the manufacturing process considerably, says Mitel's Geoff Callow, a technical marketing specialist. "If you make a chip that's twice as big, you get considerably less than half the yield. So using four rather than twelve gives us a distinct advantage," he says.

    The Newcomers

    Alvesta Inc.

  • Incorporated in April 1999 by Dubravko Babic, John Bowers, and Russell Bik. It's an impressive lineup. Babic is credited with fabricating the first continuous-wave operating long-wavelength VCSEL in 1995, while at UCSB.Dr. Bowers is currently professor of Electrical and Computer Engineering at UCSB and heads up the university's Multidisciplinary Optical Switching Technology Center. Past achievements include leading the team that demonstrated the first 8- and 16-Gbit/s transmission systems at AT&T Bell Laboratories (now Lucent Technologies Bell Laboratories). In 1996, he founded Terabit Technologies and as CEO led the company through product development to successful acquisition by Ciena Corp. (NYSE: CIEN). Bik is ex-Intel Corp., Sun Microsystems Inc., and Kleiner Perkins Caufield & Byers.

  • March 2000: announced its first product, a 10-Gbit/s small form factor transceiver. Inside is a 1x4 VCSEL array plus detectors, each element operating at 3.125 Gbit/s.

    Avalon Photonics

  • Spun out of the Centre Suisse d'Electronique et de Microtechnique (CSEM), Zürich, in 1999 (formerly the Paul Scherrer Institüt).

  • Though the company is new, it leverages the experience from CSEM, which started VCSEL development in 1994 and had sales in 1997.

  • CSEM has also developed a technology for making embossed microlenses, which narrows the transmission angle from 12 degrees to less than half a degree.

  • Offers a range of devices from 850 to 960 nanometers in wavelength and in arrays up to 1x10 and 8x8.

    E2O Communications Inc.

  • Founded in 1998

  • October 2000: Introduced 1x4 array with an aggregate capacity of 10 Gbit/s

  • Also in October, won $38.5 million cash injection. Lead investor was New Enterprise Associates (see E2O Closes $38.5M Third Round).

    Peregrine Semiconductor Corp.

  • Founded in 1990, Peregrine specializes in silicon-on-insulator electronics

  • Dabbles in VCSELs. Has produced prototype 1x4 arrays flip-chip bonded to CMOS-on-sapphire drive circuitry. Each channel in the array operates at 3.125 Gbit/s

    Novalux Inc.

  • Claims to have a technique for getting high powers out of VCSELs, which it calls the "Novalux extended cavity surface-emitting laser" (NECSEL). Novalux won't explain how this works, but does say that it's increased the area that emits light inside the laser (see Novalux Promises Cheaper Lasers and Laser Startup Bags $109 Million)

  • First product is a pump laser at 980nm (for pumping fiber amplifiers)

    PicoLight Inc.

  • PicoLight was started by Jack Jewell and Stan Swirhun, both founders of Vixel Corp. Vixel was the first company committed to commercializing VCSELs, they claim. Neither party will explain what happened in the transition from Vixel to PicoLight. (Incidentally, since it was acquired by the Fibre Channel division of Western Digital Corp., Vixel has moved into systems and no longer makes VCSELs.)

  • According to Jewell, PicoLight pioneered a more advanced type of VCSEL using a technique called oxide-confinement, which is now widely used in the industry. He likens the process to a steam bath: The material is subject to a water-saturated environment at 400 degrees Centigrade. As a result, a layer of aluminium oxide forms on the outside of the device, giving it better optical and electrical properties.

  • Has two products shipping in volume: a 1.25-Gbit/s single VCSEL, and a 2.5-Gbit/s version. Arrays will follow soon.

    Siros Technologies Inc. (formerly Opitek)

  • Siros is developing VCSELs primarily for optical data storage applications

  • It uses technology licensed from Lucent Technologies Inc., which makes it possible to achieve very high power densities, it claims (see Lucent, Siros Extend Laser Pact and Siros Claims VCSEL Breakthrough). The technique is called the Very Small Aperture Laser (VSAL), and it basically concentrates the emission through a very small hole in the top mirror. Robert Thornton, Siros's director of laser development says, "The architecture underlying this device could be used to solve key problems associated with maintaining single wavelength operation when VCSELs are operated at high power levels in excess of 5 milliwatts."TrueLight Corp.

  • Founded in 1997 in Hsinchu, Taiwan, by a team headed by Dr. Kai-Feng Huang and Dr. Kuochou Tai, both from National Chiao-Tung University

  • Offers single VCSELs at 780, 850, and 980 nm. Arrays under development

  • Claims its devices are high-power: up to 3 milliwatts at 12 milliamp current, up to 8 milliwatts at 30 milliamp current

    The key reason for wanting a VCSEL at 1310 nanometers is because this wavelength falls in the low-loss window of optical fiber. VCSELs operating at 1310nm could extend local area networks out to 40 kilometers.

    In addition, light at 1310nm is much less damaging to the eye than light at 850nm, so it's much easier to design links that are eye safe -- much more of a problem when arrays of VCSELs are used, as the combined power of the array must be considered.

    But before they become a viable commercial proposition, 1310nm VCSELs must equal the performance of DFB lasers. That means they need to operate uncooled over a wide temperature range -- many telecom applications require –40 to +85 degrees C. (Adding a cooler to a VCSEL would wipe out any advantages a VCSEL has in terms of cost and power consumption.)

    Recently, several vendors have made technology announcements about 1310nm VCSELs. Unlike short-wavelength devices, there are several distinct options for making 1310nm VCSELs. So, this report will compare vendors according to the materials technology they have adopted. To understand some of the technical issues involved, it is necessary to read the earlier section on Challenges.

    Wafer bonded

    Gore Photonics Inc., a subsidiary of W.L. Gore and Associates Inc., was first with a 1310nm VCSEL announcement in 1998. Its device comprises a 1310nm laser with an integrated optical pump (a second laser at 850nm). The active region of the 1310nm device is based on indium phosphate, while the mirrors and the 850nm laser are based on gallium arsenide. The regions are joined by wafer bonding, an industry-proven process of fusing two semiconductors with dissimilar lattice constants, using heat and pressure.


  • Improved thermal conductivity, since most of the device is GaAs-based. This results in good temperature performance

  • Highest singlemode output power to date


  • Requires optical pump, because wafer-bonded joints have high electrical resistance. Optical pump takes up space and bumps up costs.

    E2O Communications Inc. is also working on an optically pumped 1310nm VCSEL, though it hasn't published any results yet.

    Single epitaxy

    Single epitaxy means that the device can be fabricated in one process. This is advantageous in term of manufacturing -- it's simpler and cheaper than growing several layers and then bonding them. There is no need for an optical pump, which saves on space and materials and helps bring costs down, too. All of the single epitaxy devices are compatible with GaAs mirrors, and benefit from its good thermal conductivity and established manufacturing techniques. However, in general, the active regions do not work as efficiently as those based on InP. There are several approaches:

    1. GaInNAs active region
    The main drawback is that nitrides have low gain, which means that the laser has to be pushed quite hard to give out a decent amount of power.

    Cielo Communications Inc. and Sandia National Laboratories are collaborating on 1310nm VCSEL development (see Cielo Shows Laser with Sandia Labs). Ultimately -- summer 2001 -- Cielo will manufacture the devices. Best results so far are 0.4 milliwatts of singlemode output power and 0.8 mW of multimode power, according to Bob Mayer, Cielo's VP of product development. He also claims that Cielo's devices are stable at 80 degrees C.

    E2O Communications Inc. is also working on this type of VCSEL. It claims 1 mW of output power and 1 milliamp threshold current. "It's a little bit shorter than 1310 nm -- actually it's at 1220 nm," says Wenbin Jiang, E2O's VP of advanced technology. But it's better than PicoLight, which has only achieved 1170 nm, he adds.

    PicoLight Inc. is working on a single epitaxy device. Stan Swirhun, PicoLight's CEO declined to confirm that the device is based on nitrides, but Light Reading's sources say this seems quite likely. Like Cielo, PicoLight is aiming to introduce 1310nm VCSELs in summer 2001.

    2. InGaAs active region
    This is the material from which the active regions of 850nm lasers are made. Normally, there's too much stress in the material to grow it on top of a GaAs substrate, without many defects (atomic-level cracks) in the material. Light Reading believes that one startup, Nova Crystals Inc., has found a way around this. Nova has developed a "compliant universal substrate," which is a stretchy layer that can accommodate the mismatch between the crystal lattices. This would make it possible to grow InGaAs at 1310 nm (see Nova Crystals Demos High-Power VCSEL and Nova Crystal's Little Secret). Nova declined to comment.

    3. Quantum dots
    Quantum dots are particles of semiconductor (in this case, InAs) typically just 100 atoms wide, which are embedded in another material. The wavelength of light they throw out depends on the size of the particle, not just the material, so there's more scope in the design of the laser structure. What's more, the particles are so small they can stretch to fit in with the crystal lattice of GaAs. Researchers at the Air Force Institute of Technology and Imperial College in London are exploring this approach. These are very early stage devices.

    We're working on it…

    Several vendors have said that long-wavelength VCSELs are on the agenda, but are vague about technology and time to market. They include:

  • TrueLight Corp.Others, like Mitel Corp., are exploring the options by supporting university R&D projects.

    1550 nanometers lies in the second low-loss window of optical fiber, which is mostly used for long-haul applications including DWDM. In DWDM, the price of the laser isn't a significant part of overall equipment costs, so the potentially low cost of a VCSEL doesn't have as much impact. To offer a competitive advantage, VCSELs will have to accommodate an additional requirement -- tunability (see Tune In!).

    This means that 1550nm VCSEL makers have to achieve a double whammy. First they have to make a VCSEL that works, then they have to incorporate tunability. It's no wonder that 1550nm devices are still at a pretty early stage of development.

    Only two companies look like they've got both sets of skills:

    Bandwidth9 Inc.

  • Bandwidth9 announced its tunable VCSEL in September 2000 (see Bandwidth9 Claims Laser Breakthrough). The startup has ignored the problem of incompatible mirrors by simply growing GaAs on top of InP anyway. The result is a so-called "pseudomorphic" interface. The epitaxial layer sticks to the substrate "just like amorphous silicon sticks to glass in solar cells," according to Wupen Yuen, Bandwidth9's director of precision epitaxy.

  • Experts in the field question the reliability of the laser, saying that the dodgy interface might generate lots of defects in the structure that would kill light emission. Yuen counters by saying that though the structure does have defects, they don't matter because they don't occur in the active region.

  • Bandwidth9's laser is not particularly high power. The best reported to date is 0.45 milliamps.

  • Maximum reported lasing temperature is only 55 degrees C

  • Tuning is achieved with a micro-electro-mechanical system (MEMS) cantilever. The startup has experience in making these for short-wavelength VCSELs, and it should be relatively straightforward to translate this to a long-wavelength device.

    Coretek Inc.

  • Uses a separate optical pump to generate light. (Note, the device is not wafer bonded). The reason for optical pumping is it makes it possible to generate light from a larger volume of active material than is possible using electrical pumping. As a result, the VCSEL is high power -- at least 2 milliwatts.

  • Because there is more active material, the mirrors don't have to be as highly reflective. This eases the manufacturing requirements.

  • Fairly high pump powers (50 milliwatts at 980nm) are needed. The pump is expensive.

  • Tuning range of 50nm using MEMS

  • Nortel bought Coretek (see Nortel Gambles $1.43 Billion On Tunable Lasers).

    It's also worth mentioning that Novalux Inc. plans to make a tunable VCSEL. It says the device will be electrically tuned (not MEMS) but doesn't give any details. This device is probably still on the drawing board. Novalux says the development could take a couple more years (see Laser Startup Bags $109 Million).

    Although these are the only companies with concrete plans for tunable VCSELs, there is plenty of activity in universities and research institutes. Different methods for making 1550nm VCSELs are emerging, and commercial development is sure to follow. Some research groups worthy of note include:

    University of California, Santa Barbara

  • Larry Coldren's group is working a single-epitaxy approach for making long-wavelength VCSELs (see VCSEL Breakthrough at UCal). (Incidentally, Coldren is one of the founders of Agility Communications Inc., a tunable laser manufacturer).

  • John Bowers's group is using a wafer-bonding approach. However, unlike other wafer-bonded VCSELs, Bowers has managed to reduce the resistance of the bonded joint so that the laser can be electrically pumped. Adil Karim, who works in the group, says these devices have the best high temperature performance -- twice as much as the single-epitaxy device made in Coldren's group, he claims. Another innovation is to make arrays emitting at multiple wavelengths for use in DWDM systems. (Bowers is one of the founders of VCSEL manufacturer Alvesta Inc., which may provide a route to market for this technology.)

    Walter Schottky Institut, Technical University of Munich

  • Markus Amann's group has achieved the lowest threshold current and voltage for a 1550nm VCSEL to date. The VCSEL structure is made in a two-step epitaxial process. One mirror is InP-based, the other is made of low index dielectric. Maximum lasing temperature is around 50 degrees C.

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like