The advent of VCSEL technology in high-speed data communications
VCSELs promise to provide a cost-effective source for 10-Gbit/sec applications, regardless of whether designers use serial, parallel, or WDM technology.
Dr. Robert Bryan and Brian Gibson
Driven by increasing de mand for low-cost, high-speed telecommunications and data communications, the use of vertical-cavity surface-emitting lasers (VCSELs) is exploding. VCSELs are compound semiconductor micro-laser diodes that emit light vertically perpendicular to their p-n junction from the surface of a fabricated wafer. They are used in a variety of data-communications and telecommunications environments that require efficient, high-speed transmission of data, including Gigabit Ethernet, Fibre Channel, Infiniband, WDM, ATM, very-short-reach OC-192, and optical backplanes.
VCSELs have many advantages, including photolithographically defined geometries, circular output beams, high fiber-optic coupling efficiencies, extremely low power consumption, and ultra-high modulation rates. The result is that the race to power the next generation of high-performance, low-cost optical-fiber data-communications equipment appears to be shaping up as a battle between different implementations of this device, based on time-, space-, or wavelength-division multiplexing methods.
VCSELs have become the technology of choice for a wide range of data-communications products. With a low threshold of between 1 and 6 mA, VCSELs offer extremely efficient power conversion. They can deliver transmission speeds ranging from 1 to 10 Gbits/sec, yet have a modulation swing of only 5 to 10 mA, which keeps power consumption low.
The latest generation of VCSELs does not require hermetic packaging, and typical mean lifetimes for well-manufactured devices are over 100 years. At the same time, the circular, low-divergence output beams provided by VCSELs eliminate the need for corrective optics in most applications. The ability to use standard microelectronic-style manufacturing processing coupled with on-wafer testing keeps costs at the low levels required for use in customer-premises applications. Finally, the fact that arrays can be produced without extra packaging steps promotes advanced systems and miniaturization.
When the Gigabit Ethernet standards committee chose an 850-nm light source, light emitting diodes (LEDs), edge-emitting diodes, and VCSELs were considered for the light source of choice. However, LEDs were never a practical choice, primarily because of their slow speed and difficulties caused by their very wide beam emission and broad spectral emissions. Edge emitters dropped out fairly late in the product cycle due to problems in reliability and the operation of the "CD-style" laser structure at high speeds. VCSELs, however, easily operate at high speeds and are extremely reliable.
The result is that VCSELs are currently used in all new end-user 850-nm data-communications products at gigabit speeds and higher, including Gigabit Ethernet and Fibre Channel. VCSELs are also beginning to penetrate the public network for the telecommunications market in certain short-distance applications such as interswitch and intraswitch communications.
According to nearly all industry observers, VCSELs will continue to play a dominant role, as 10-Gbit/sec standards (such as 10-Gigabit Ethernet) begin to emerge in both datacom and telecom applications. The real question is which type of VCSEL implementation will the market turn to for particular 10-Gbit/sec applications? Three implementations offer a reliable solution, and chances are that all will co-exist for various applications.
It appears that the most logical and ideal solution would be a serial 10-Gbit/sec VCSEL solution. However, this approach does not exist yet in commercial implementations. The advantage of this approach is its simplicity-a single laser transmitting data down a single strand of fiber using time-division multiplexing (TDM). The challenge is that while VCSEL devices have been demonstrated at 10 Gbits/sec in the laboratory, commercial products have only just reached the 2.5-Gbit/sec milestone. Even as 10-Gbit/sec VCSELs emerge over the next year, they will have to overcome the obstacle posed by the fact that the current infrastructure is based on much slower speeds.
For example, the multimode fiber that constitutes the vast majority of current premises equipment is limited to link lengths of tens of meters at 10 Gbits/sec.
Another obstacle is that the drive circuits and high-speed receivers needed to implement 10-Gbit/sec serial solutions are still very expensive, as is the required test equipment. But as 10-Gbit/sec VCSELs begin to reach the market and intersect with improvements in drive circuits, receivers, and optical-fiber cable, this approach should prove very attractive.
On the other hand, the space-division multiplexing (SDM) approach, better known as parallel optics, is here today. It will, for just that reason, almost certainly be used in the first practical low-cost very-short-reach OC-192 10-Gbit/sec solution to reach the market. The basic idea is to produce an array of VCSELs designed to run over separate strands of fiber on a single wafer.
The key advantage of parallel optics is that it makes use of existing technology-not only VCSELs, but also receivers, drive circuits, and optical-fiber cables. The result is faster development cycles, because, for example, it's possible to keep the existing deserializer interface. Parallel optics also delivers higher link lengths, because the transmission speed within each strand is at current levels.
The first parallel-optics solutions consist of 1x12 arrays, with the possible migration to 1x4 as VCSEL speeds increase. It is important to note that existing technology offers considerable headroom for further speed increases. For example, the first commercially available 1x12 arrays used 1.25-Gbit/sec lasers, enabling 15-Gbit/sec parallel optical links. Now that 2.5-Gbit/sec VCSELs have appeared, it should only be a short time before a 1x12 array offering 30-Gbit/sec speeds appears.
Parallel optics will likely be the earliest cost-effective solution at the highest speeds (i.e., the biggest data pipe) to reach the market, because it can take the current state-of-the-art in serial technology and multiply it to yield a greater aggregate data rate. This demand for very high and cost-effective data rates is especially acute for rack-to-rack and equipment-to-equipment links.
WDM may be another contender in emerging high-speed data-communications markets. The basic concept is similar to the parallel-array approach-use of multiple existing components operating at current speeds to achieve a new plateau. The difference is, of course, that parallel optics divides the signals on the basis of space, while WDM divides them on the basis of wavelength. The basic advantages are also similar to parallel optics: use of existing components that can be produced at high yields at a relatively low cost and longer link lengths.
There are, however, some significant obstacles that must be overcome before WDM becomes a significant player. WDM requires manufacturing techniques that are considerably more complex than those involved in producing parallel arrays. Unlike parallel arrays, which can utilize the natural wafer spacing, WDM requires that the wavelength-selected VCSELs be diced and manually placed into the device packaging. Tight manufacturing tolerances as well as microscopic components increases the difficulty of this task.
Another problem is that capital equipment required to assemble WDM devices is not available commercially, so it must be designed and built from scratch. WDM devices also require the design and manufacture of drive circuits that are considerably more complex than either parallel arrays or serial VCSEL devices.
While first Gigabit Ethernet and later the upcoming 10-Gigabit Ethernet standards have been the primary drivers of VCSEL technology, manufacturers of outside-plant equipment have already begun moving in this direction. As their speeds increase to the level of the public network, VCSELs offer an increasingly attractive alternative to equipment suppliers that are accustomed to paying several orders of magnitude higher prices for light sources.
VCSELs can be used today within the central office, for example, as a link between two central-office switches or a high-speed communications link within a switch. With the coming of long-wavelength VCSELs (likely within the next year), VCSELs may be deployed outside the plant as the cost-effective solution for connecting the "last mile."
The fastest commercially available VCSEL is a 2.5-Gbit/sec 850-nm oxide VCSEL. This device meets the requirement for a cost-effective light source capable of operating at the 2.12 Gbits/sec required for new Fibre Channel applications. In addition, 1.25- and 2.5-Gbit/sec 1x12 parallel arrays are available that will deliver 10-Gbit/sec bandwidth with headroom to spare. Both serial and WDM solutions can be expected to emerge in the next several years. Serial solutions depend on increasing the speed of the light source, while WDM requires reducing the cost of the circuitry currently needed to accurately control the wavelength of the device.
In the meantime, the 10-Gbit/sec Ethernet standard proposal is in the hands of various working groups. Several different proposals are under consideration, and the chances are that the final standard will involve multiple implementations, allowing each manufacturer to select the technology that works best for them and their application. Concurrently, the Fibre Channel group is working on a 10-Gbit/sec implementation, and there are several other groups working on high-speed links, such as the personal computer I/O bus called Infiniband. In addition, 1x12 VCSEL-based parallel optics implementation for very-short-reach OC-192 is likely from the Optical Internetworking Forum.
Each of these organizations faces similar challenges and technical barriers. As a result, the standards tend to converge around the base-level components described here.
Among the various methods of producing VCSELs, the commercialization of an oxide process for producing high-speed, highly reliable, and highly consistent VCSEL components has been significant. As in silicon technology, the insulating oxide is able to produce insulating regions at strategic points in a device. In the case of VCSELs, a lateral confinement region is created, which results in highly efficient laser cavities. The tight tolerances of the oxide material result in more efficient use of current than ion-implanted VCSEL devices. And the greater uniformity provided by these methods results in less variability among devices and significantly more uniform large-element-count arrays.
Recently, the industry has seen the first commercial 2.5-Gbit/sec 850-nm oxide VCSEL. This product technology is expected to form the critical high-speed optical link for next-generation serial and parallel optical transceivers for 10-Gbit/sec Ethernet, 10-Gbit/sec Fibre Channel, OC-192, and other markets. The oxide VCSELs provide lower current thresholds and smoother slope efficiencies to meet performance requirements of existing and future Gigabit Ethernet standards. These new VCSEL products are available as a bare die or can be built in to TO-cans or optical subassemblies.
The latest generation of VCSELs is blasting through the performance limits of current devices, exactly what is required for the next generation of high-speed data-communications equipment. VCSELs provide a sensible, cost-effective solution for speeding up the transmission of data in the near term, while delivering the reliability that both customer premises equipment and public-network applications require.
Regardless of which implementation method eventually wins, VCSEL technology is certain to be an integral component and will enjoy a long and bright future.
Dr. Robert Bryan is vice president, MicroOptical Devices Div., at EMCORE (Somerset, NJ). Brian Gibson is director of business development, MicroOptical Devices Div.