Low-cost, singlemode transmission with long-wavelength VCSELs
Technical barriers to realizing 1300-nm VCSELs are being removed. A new approach may yield the first commercial devices.
Dave Welch
Gore Photonics Inc.
Vertical-cavity surface-emitting lasers (VCSELs) emitting at 850 nm are commonly used as cost-effective, highly reliable light sources for gigabit data communications. These short-wavelength VCSELs are limited, however, to transmission distances of a few hundred meters over multimode fiber. For longer-distance applications, long-wavelength lasers compatible with singlemode fiber must be employed. Expensive 1300-nm edge-emitting lasers are typically used in these applications.
There is a clear need for a low-cost, high-data-rate light source capable of transmission over extended distances. Long-wavelength 1300-nm VCSELs address this need. These devices possess cost advantages similar to those offered by short-wavelength 850-nm VCSELs, and enable the increased transmission distance provided by 1300-nm edge-emitting lasers. Long-wavelength VCSELs also offer the potential for both low-cost die and simplifications in packaging. Most notably, the narrow beam divergence offered by these devices allows cost-effective singlemode coupling solutions.
Emergence of VCSELs
There has been a strong investment in VCSEL technology by the government, universities, and industrial labs since the late 1980s. Most of the early work was on GaAs VCSELs emitting light in the near-infrared wavelengths between 780 and 980 nm. As a result of these efforts, the technology has developed rapidly and the first commercial products appeared in the mid-1990s.
VCSELs are viewed as a revolutionary technology because they offer a variety of performance and cost advantages over traditional edge-emitting semiconductor lasers. VCSELs provide the following benefits:
circular output beam with a small divergence angle
fast modulation rates
stable performance over temperature
low threshold currents
wafer-scale fabrication and test
1-D and 2-D arrays
Produced much like low-cost, light-emitting diodes (LEDs), VCSELs are comprised of a light-emitting amplification region sandwiched between two semiconductor mirrors. The structure is grown layer by layer onto a wafer substrate, typically by metal-organic chemical vapor deposition (MOCVD), which allows precise control of layer thickness, composition, and uniformity. Devices are then fabricated using standard III-V processing techniques. Critical features are formed lithographically, leading to high yields and repeatable performance characteristics. Because VCSELs are surface emitters, wafer scale testing can be employed. This results in lower testing costs, known good die, and increased packaging yields.
While VCSEL die can be produced at lower costs, VCSELs have an even more profound effect on packaging costs, which generally account for around 80% of the total packaged laser cost. The low-divergence, circular output beam dramatically simplifies optical coupling. VCSELs have also shown sub-milliamp threshold currents, and can operate at high speeds with relatively low drive currents. The low power consumption eases package design by reducing heat problems. Additionally, small changes in the threshold current and output power over typical operating temperatures allow the laser to be operated under fixed bias and modulation currents, while adhering to the link budget requirements. Therefore, the monitor photodiode feedback loop required by edge-emitting lasers can be eliminated for many applications.
To date, the largest market for short- wavelength VCSELs is in high-speed data communications, where 850-nm VCSELs are used as high-speed, multimode transmitters for applications such as Gigabit Ethernet and Fibre Channel.
Long-wavelength VCSEL advances
Considerable research has focused on developing long-wavelength VCSELs emitting at 1300 and 1550 nm. Long-wavelength VCSELs offer the cost benefits of short-wavelength VCSELs plus the ability to transmit long distances over singlemode fiber.
But the development has progressed more slowly than it did with short- wavelength VCSELs. The primary difficulties stem from an incompatibility in the material systems needed to create both good amplification regions and highly reflective mirrors for light of longer wavelengths. Indium phosphide-based (InP) materials are needed to achieve emission at the 1300- or 1550-nm wavelengths; however the mirrors made from these materials do not provide adequate reflectivity. GaAs-based mirrors offer the needed reflectivity at these wavelengths, but lattice mismatch problems do not allow these mirrors to be grown directly on InP substrates.
While researchers are exploring numerous ways to address this problem, to date wafer-fused structures have yielded the best results. Wafer fusion is a process in which materials with different lattice constants are joined under heat and pressure. The resulting interface is atomically continuous. Wafer fusion is a proven, reliable process demonstrated by Hewlett-Packard`s volume production of high-brightness red LEDs.
This process has been applied to long-wavelength VCSELs to overcome the materials incompatibility problems described here. Researchers have produced long-wavelength VCSELs by sandwiching InP active regions between GaAs mirrors. Using this technology, John Bowers` group at the University of California at Santa Barbara (UCSB) demonstrated the first continuous-wave (CW), room-temperature operation of a long-wavelength VCSEL in 1995, operating near 1550 nm. Since then, the UCSB group has raised the CW operating temperature to 71C, and performed various transmission experiments proving the viability of VCSELs as an optical source for long-distance communications.
UCSB`s pioneering work has sparked a great deal of research activity in long-wavelength VCSELs by a number of groups in Europe, Japan, and the United States. Many researchers have used wafer fusion, while some have focused on combinations of fused, epitaxial, and deposited dielectric mirrors. In addition, there has been a great deal of effort in developing materials that can emit at 1300 nm and be grown directly on GaAs. Although these efforts show promise, they are still in the early stages of development.
Despite the success of the UCSB wafer-fused approach and the subsequent efforts by multiple research groups, a number of technological barriers to commercialization have remained. The challenges have included obtaining useful output power (several hundred microwatts) at temperatures up to 70C, achieving lower operating voltages, and obtaining all the output power in a single transverse mode, to allow efficient coupling to singlemode fiber.
Last July, scientists at Gore led by Vijay Jayaraman announced the first long-wavelength VCSELs to meet commercial performance requirements. The VCSELs combine wafer-fused mirrors with an integrated optical pump to overcome the barriers highlighted here (see Fig. 1). The device makes use of two wafer-fused GaAs/AlGaAs mirrors. The mirrors and active region are undoped. The structure includes an integral 850-nm VCSEL vertically adjacent to the 1300-nm VCSEL. Current is injected into the 850-nm VCSEL, which in turn optically pumps the 1300-nm VCSEL, which then emits the 1300-nm laser output through the 850-nm VCSEL and out the top surface of the device.
This optically pumped structure has several fundamental advantages over direct electrical pumping. The undoped 1300-nm cavity has reduced optical losses. The wafer fusion provides only an optical interface, so it does not introduce an additional device voltage drop. Resistive heating is primarily confined to the 850-nm pump. Finally, the absence of current crowding and reduced thermal lensing promotes higher singlemode output powers.
Such VCSELs have reproducibly demonstrated useful singlemode output power from 0 to 70C, low threshold currents, and low operating voltages. These devices also offer temperature performance advantages over edge-emitting lasers (see Table 1). Wavelength shift with temperature is much lower, leading to improved transmission performance, and easing optics design for low-cost coupling. The minimal shift in threshold current and the low power output variability over the operating temperature may enable fixed bias and modulation currents while maintaining a high extinction ratio, eliminating the need for a monitor photodiode feedback loop in some applications. Edge-emitting lasers require a monitor photodiode, and the complementary drive circuitry to compensate for temperature effects. The 1300-nm VCSELs also provide good high-speed performance, and 4-Gbit/sec operation has been demonstrated (see Fig. 2).
The optically pumped 1300-nm VCSELs emit in a single transverse and single longitudinal mode, with a narrow linewidth. This solution differs from 1300-nm Fabry-Perot edge-emitters that have multiple longitudinal modes (see Fig. 3). The 1300-nm VCSELs provide output much like that of a distributed-feedback laser, and are well suited to long-distance transmission.
Long-wavelength VCSELs also provide a low-divergence, circular output beam. The beam differs greatly from the highly divergent beam of an edge-emitting laser, which is very difficult to efficiently couple into singlemode fiber. The output beam of the 1300-nm VCSEL is well matched to singlemode fiber, and allows for efficient coupling and improved coupling tolerances. Lower-cost alignment components and process technologies can be considered, and passive alignment appears feasible. To date, singlemode butt-coupling efficiencies of 80% have been demonstrated.
Bandwidth and distance at low cost
The 1300-nm VCSELs may provide the lowest-cost singlemode laser solutions for wide area networks, interoffice communications, fiber-to-the-home, and fiber-to-the-curb. One of the initial applications for long-wavelength VCSELs will be in singlemode fiber-optic transceivers for use in data-communications equipment for local-area and wide-area networks.
With the increase in data rates of new systems, network equipment providers must address the issues related to the bandwidth and usable transmission distance in multimode fiber. The primary factor determining the transmission distance in multimode fiber is modal dispersion resulting from varying path lengths of different propagating modes. This creates pulse spreading; adjacent pulses will eventually interfere with each other. Effective transmission distance is reduced with increasing operating speeds.
As data rates have increased to 1 Gbit/sec and beyond, lasers have replaced LEDs as the preferred transmitter source, and the challenges associated with using lasers to transmit light over multimode fiber have become more apparent. A problem called differential mode delay (DMD) surfaced during the development of the Gigabit Ethernet standard for transmission in graded-index multimode fiber. Graded-index fiber is designed to compensate for different modal path lengths by allowing the light at the outside of the fiber to travel faster than the light propagating down the fiber`s center. DMD is caused by inefficient compensation for this effect in the fiber. This lack of compensation, combined with the characteristics of an underfilled launch condition resulting from using a laser source with multimode fiber, magnifies the DMD problem, further reducing transmission distance. The Gigabit Ethernet committee was forced to reduce link distance from 500 to 220 m when using a 1000BASE-SX 850-nm transceiver in 160 MHz/km FDDI-grade 62.5/125-micron multimode fiber. A similar 62.5-micron multimode fiber solution using a 1000BASE-LX 1300-nm transceiver requires the additional expense of an offset patch-cord jumper cable for improved modal launch, to enable link lengths up to 550 m.
Although the Gigabit Ethernet group successfully worked through the various difficulties, the issue of effective bandwidth and transmission distance in multimode fiber-based links concerns many users and systems designers, particularly those who want to upgrade network performance while using their installed base of fiber. Higher data rate standards are already in the works. Fibre Channel bodies have initiated work on 2X- and 4X-speed versions of the existing 1.063-Gbit/sec standards. The development of a 10-Gbit Ethernet standard is expected to commence this year. As data rates increase, effective transmission distance in multimode fiber will be further reduced.
Because of multimode fiber transmission distance limitations, lower-cost singlemode solutions are needed. Andy Bechtolsheim, vice president of engineering in the Gigabit Switching Group at Cisco Systems Inc. (San Jose, CA), summarizes the expected benefits of long-wavelength VCSELs, "1300-nm VCSELs enable scalability to higher data rates and longer distances at much lower cost than currently available laser components."
Use in parallel optics
Long-wavelength VCSELs will also be used in parallel optical data links. Parallel optics is an emerging technology, in which array components are used to create high-density, low-cost transmitters and receivers that are compatible with ribbon fiber cables and connectors. Parallel optical interconnects offer the extremely high data-transfer rates required by scalable computers and broadband switches. Parallel optics also provide a low-cost, high-density alternative to serial transceivers used for data-communications networking applications. VCSELs enable a cost-effective approach to realizing parallel optical interconnects. The initial parallel optical modules are typically 12 channels wide and have per-channel data rates of at least 1.25 Gbits/sec, providing an aggregate bandwidth of 15 Gbits/sec.
One application for parallel optical interconnects is in parallel processor-based computer systems in which a problem is broken down into smaller elements, with each element handled by one or more processors. High-speed data buses link the processors.
"The processor-to-processor interconnects are critical to our systems," states Casimer DeCusatis, advisory engineer in Systems/390 Server Division at IBM (Poughkeepsie, NY). "IBM uses predominantly copper interconnects today; however, increasing data rates make parallel optics the likely solution for future systems. Parallel optical data links are expected to provide the necessary bandwidth, while meeting our cost and density objectives".
The 1300-nm VCSELs offer an economical path to overcoming some significant challenges around designing eye safety into parallel optical data links. Because arrays of lasers are used, the power levels of individual lasers must be reduced (versus single lasers) to meet eye-safety requirements. Most data link manufacturers and systems designers are struggling to develop 850-nm parallel modules that meet all Class 1 eye-safety standards. Because 1300-nm light is much less damaging to the eye than sources at 850 nm, more power can be launched into the fiber and eye-safe operation can be achieved without compromising link performance.
"Our machines are typically used in non-restricted environments, so FDA and IEC Class 1 eye safety is required. 1300-nm VCSELs remove eye safety as a major technical hurdle for parallel module developers," notes DeCusatis.
Cost-effective performance
Long-wavelength VCSELs possess fundamental cost advantages over traditional edge-emitting lasers and are capable of high-data-rate communications over long distances. These devices will be used in a variety of data-communications, telecommunications, and high-end computing applications requiring scalability to higher data rates and increased distance, at low cost. u
Acknowledgements:
The author would like to thank the following people for their assistance: Vijay Jayaraman, Gore Photonics Inc. (Lompoc, CA); Tom Goodwin, Chip Mueller, and Mark Donhowe, W.L. Gore and Associates Inc. (Newark, DE); and Richard Kriese, W.L. Gore & Associates Inc. (Austin, TX).
Dave Welch is the business leader at Gore Photonics Inc. (Lompoc, CA), a subsidiary of W.L. Gore & Associates Inc. headquartered in Newark, DE.