VCSELs turn to high-speed transmission

Jeff Hecht, Contributing Editor

The technology for vertical-cavity surface-emitting lasers is growing explosively. First demonstrated only about a decade ago, the VCSEL is a major shift in semiconductor-laser technology. Instead of a laser cavity in the plane of the junction layer, the VCSEL has resonator mirrors above and below the junction layer, so light oscillates in a "vertical cavity" perpendicular to the junction. This gives VCSELs properties quite different from conventional edge-emitters.

So far, the biggest applications of VCSELs have been in high-speed local-area networks such as Gigabit Ethernet, where 850-nm sources are used with multimode fibers. However, fiber dispersion and loss at 850 nm limit transmission rates and distances. Emerging higher-capacity networks like 10 Gigabit Ethernet will require longer-wavelength sources for most interbuilding applications. That poses some tough challenges, but developers are have made considerable progress and look to push VCSELs all the way to the metro network.

A generic view of a VCSEL is shown in Fig. 1. In this example, the mirrors above and below the junction layer are formed by depositing a series of thin layers of semiconductor material, with two alternating compositions that differ in refractive index. These function as distributed Bragg reflectors, with the refractive-index differential causing them to reflect wavelengths selected by the layer spacing back into the laser cavity. One is a total reflector, and the other transmits a small fraction of the incident light—around 1%—which forms the laser beam. The figure shows a common type of device in which the beam emerges through an n-doped substrate.

Other VCSEL structures differ in detail, but retain the idea of a pair of highly reflective mirrors that provide strong reflection back into the laser cavity. In this sense the cavity resembles that of a low-gain helium neon gas laser because VCSELs have low gain per pass. Semiconductor lasers have much higher gain per unit length, but the VCSEL includes only a very short gain region because the beam passes vertically through the junction layer. One consequence is that VCSEL powers generally are in the milliwatt range, lower than edge emitters.

VCSEL cavities also include spacers, typically above and below the junction layer. These spacers control the length of the VCSEL cavity, and thus select the longitudinal modes and the output wavelength.

The internal structure of a VCSEL is more complex than that of an edge emitter, but in the end the design reduces costs. Both VCSELs and edge emitters can be fabricated in large numbers on a wafer, but the VCSEL structure is complete on the wafer, so devices can be tested before the wafer is diced. In contrast, edge emitters are incomplete until the wafer is cleaved to form facets, so they must be tested individually. Testing and packaging costs tip the economic balance toward VCSELs.

Normal fabrication techniques readily yield monolithic two-dimensional arrays of VCSELs. Separate inputs can drive individual devices for optical processing and switching, or for supplying input to separate fibers. Grading the spacer layer across the wafer produces lasers with a range of wavelengths, potentially useful for wavelength-division multiplexing (WDM).

The circular cross section of VCSELs gives better control over beam size and divergence than for edge emitters. VCSELs normally emit from areas 5 to 30 µm across, avoiding the broad divergence of edge emitters perpendicular to the thin junction plane. This aids coupling of VCSEL output into fibers, with the smaller-aperture VCSELs compatible with single-mode fibers and larger ones compatible with the multimode graded-index fibers used for local-area networks.

Although standard VCSELs cannot match the output powers of edge-emitting lasers, they are more powerful and have better beam quality than LEDs. Together with their higher modulation speeds and low cost, this makes VCSELs preferred light sources for local-area networks such as Gigabit Ethernet.

Another advantage of VCSELs is a lower sensitivity to scattering of transmitted signals toward the fiber. The distributed-Bragg-reflection output mirror strongly rejects the laser wavelength, and any noise that reaches the inside of the laser cavity is overwhelmed by the high circulating power. In contrast, signals scattered back into an edge emitter can cause serious noise problems, requiring the use of optical isolators.

The first VCSELs were made in gallium-arsenide (GaAs) materials, and so far VCSELs have been most successful at wavelengths of 750 to 850 nm and 980 nm. Much of this success comes from the comparative ease of making distributed Bragg reflectors. The refractive index of gallium aluminum arsenide (GaAlAs) varies significantly with the mixture, so it's easy to make alternating layers with large enough refractive index for high Bragg reflection. The higher the index contrast, the fewer alternating layers are needed to make a reflector.

Intense development has focused on longer wavelengths of 1300 to 1700 nm, where fibers have lower attenuation and dispersion. However, the standard material used for edge emitters in that range, indium gallium arsenide phosphide (InGaAsP), has proven difficult for VCSELs. The refractive index changes little with composition, so thick stacks of alternating layers are needed for distributed Bragg reflectors. Unfortunately, material losses accumulate with the number of layers, which together with other problems such as low thermal conductivity makes InGaAsP VCSELs with thick mirrors impractical.

Dielectric mirrors offer higher reflectivity, but they don't conduct current, making electrical excitation difficult. Nonetheless, M. C. Amann et al. of the Technical University of Munich (Germany) reported encouraging results on an InP substrate in a postdeadline paper at the European Conference on Optical Communications (ECOC) in September 2000. They excited the dielectric-mirror side of a 1500-nm VCSEL through a buried tunnel junction, and coupled light out through a distributed Bragg mirror of InGaAlAs.

Developers are working on many alternative approaches, with some companies so secretive they decline to provide details. Bonding GaAs reflectors to InGaAsP active layers has worked in the laboratory, but that elaborate process is not attractive for industrial production.¹

An alternative is to grow GaAs materials on InP substrates, called metamorphic growth because of the 3.7% difference in lattice spacing.² At CLEO 2000 in San Francisco, CA, W. Yuen and colleagues at Bandwidth9 (Fremont, CA) reported using this approach to make 1600-nm VCSELs emitting up to 0.45 mW. Starting with an InP substrate, they deposited a lattice-matched stack of InAlGaAs/InAlAs layers, and an active region containing InGaAs quantum wells. Then using metamorphic growth, they deposited a stack of alternating GaAs/AlGaAs layers as the top distributed Bragg reflector.

Larry Coldren's group at the University of California-Santa Barbara (UCSB) is investigating alternative materials for use with InP. In another ECOC 2000 postdeadline paper, the group reported a continuous-wave 1550-nm VCSEL with aluminum gallium arsenide antimonide (AlGaAsSb) distributed Bragg reflectors grown on an InP substrate. Cladding layers of InP separated the reflectors from an InGaAs active layer.

Using a buried tunnel junction, a team at the Technical University of Munich (Munich, Germany) has made a 1550-mm VCSEL that can be modulated at 2.5 Gbit/s. In a postdeadline paper at ECOC 2001, they reported output power to 0.5 mW at 80°C, important because lasers in actual systems may need to operate at high temperatures.

Another promising approach is developing long-wavelength materials compatible with GaAs. A team at Sandia National Laboratories turned to InGaAsN, a nitride compound compatible with GaAs that also has been tested for solar cells. The researchers made 1300-nm VCSELs on a GaAs substrate, with distributed Bragg reflectors of GaAs/GaAlAs, and used InGaAsN only for quantum wells in the active layer. They licensed the technology to Cielo Communications (Broomfield, CO).

These and other long-wavelength VCSEL technologies are in intense commercial development.

Both dispersion and attenuation can limit 850-nm transmission. The proposed 10 Gigabit Ethernet standard (see table) envisions a 65-m limit in conventional 50/125-µm graded-index multimode fiber, but developers have gone well beyond that. In a postdeadline paper at CLEO 2000, Martin C. Nuss' group at Bell Labs (Holmdel, NJ) reported transmitting a single 10-Gbit/s signal through 1.6 km of a high-bandwidth 50/125 graded-index fiber developed by the parent Lucent Technologies. At ECOC 2000, the same group reported extending its experiments to coarse wavelength-division multiplexing (CWDM), with 10-Gbit/s channels at 815, 822, 828, and 835 nm, squeezing 40-Gbit/s through 310 m of the same fiber. In a CLEO 2000 postdeadline paper, a team from Ulm and Bell Labs reported sending 10 Gbit/s through 2.8 km of the high-speed multimode fiber, limited by attenuation. The Ulm group is even testing single-mode fibers at short wavelengths, a technology largely abandoned 20 years ago with the opening of the 1300-nm window.

The next step is tunable VCSELs, which are already starting to emerge. Most interest is in the long wavelengths used for telecommunications. The leading approach is to fabricate an external mirror suspended above the substrate (see Fig. 2). Moving that mirror up and down changes the cavity length and thus the output wavelength. These approaches usually are based on MEMS technology, with the entire structure grown monolithically, then selected layers etched away to leave the suspended mirror. Applied voltages then move the external mirror to tune the laser. Tuning range potentially can exceed 100 nm, T. Amano and colleagues from the Tokyo Institute of Technology said in a postdeadline paper at ECOC 2001.

So far the highest powers have come from optically pumping the VCSEL with a high-power fiber laser. electrically driven tunable VCSELs have lower power, often less than 1 mW. Expect to see continued progress in tuning range, output power, and modulation speed.

1. Y. Qian et al., Appl. Phys. Lett. 71, 25 (July 1997).
2. J. Boucart et al., IEEE J. of Sel. Areas in Quant. Elec. 5, 520 (May/June 1999).

This article was originally published in Laser Focus World, February 2001.