PMDs battle for 10-GbE multimode applications
Multimode fiber (MMF) is the most popular optical-fiber medium in premises networks. Lower-cost light sources such as LEDs and VCSELs have contributed to this popularity, particularly for data communications. As the demand for bandwidth in MMF environments increases, standards and transceiver technologies have evolved to support higher data rates. (For more on the properties of MMF and the medium’s role in premises networks, see the sidebar “Multimode fiber and premises networks,” p. 14.)
These standards must take into account the fiber’s modal dispersion, which creates an upper boundary on the bandwidth of the fiber that depends on the wavelength of the light, light profile launched into the fiber, and index profile of the fiber. This bandwidth limit creates a relationship between wavelength, launch condition, transmission distance, and data rate. The IEEE has standardized optical requirements for Fast Ethernet (100 Mbits/sec), Gigabit Ethernet (1 Gbit/sec), and 10-Gigabit Ethernet (10 Gbits/sec) with support for both multimode and singlemode optical fibers.
In particular, the 10-Gigabit Ethernet (10-GbE) standards makers had to wrestle with modal dispersion issues on a wide variety of fiber. That has resulted in several sets of specifications as well as several types of transceivers that network planners must consider when architecting their networks. Several factors will determine which transceiver types will achieve the greatest deployment in MMF applications.
To understand how the market for 10-GbE over MMF will evolve, it is best to look at the history of Gigabit Ethernet. The IEEE P802.3 standards body that governs the Ethernet standard established two standards to address MMF in GbE, 1000Base-SX, and 1000Base-LX. The 1000Base-SX GbE standard has been one of the more successful optical interfaces in communications. Today, there are about 1.5-2 million 1000Base-SX ports shipped per quarter. The 1000Base-SX standard is only specified for various types of MMF and operates at the 850-nm optical wavelength.
The 1000Base-LX standard operates at 1310 nm and is primarily used for singlemode-fiber (SMF) applications. However, it may also be used for some extended MMF applications. Currently, there are about a few hundred thousand 1000Base-LX ports shipped per quarter.
Similarly, the 10-GbE standard currently specifies two different physical media dependents (PMDs) for various MMFs, although a third is under review in the standards committee. The two ratified standards are 10GBase-SR and 10GBase-LX4.
The 10GBase-SR standard uses the same wavelength as the popular 1000Base-SX interface; however, at this higher data rate, the link length is reduced to 33 m over 62.5-µm legacy grade MMF. Like the LX interface, the 10GBase-LX4 standard uses the same wavelength window and supports both MMF and SMF transmission. Unlike the serial LX standard, the 10GBase-LX4 standard uses four wavelengths in the 1.3-µm region at a quarter of the baud rate per lambda so the transmission distance is farther over lower-bandwidth fiber.
The new standard under review is 10GBase-LRM, which is being developed to address concerns that the parallel nature of LX4 reduces the interface’s manufacturability and economical feasibility. The 10GBase-LRM draft uses a single-laser approach for better manufacturability and simpler construction of next generation serial transceivers in form factors such as the XFP.
The Figure on page 13 compares the link distances of the three PMDs over various types of MMF.Implementations of the three 10-GbE PMDs are dominated by XENPAK, X2, XPAK, and XFP optical transceivers-sometimes known broadly as X-modules. To support next generation systems with high-density 10-GbE interface requirements, transceiver manufacturers are working to reduce power consumption, size, and cost within these module families and still support all of the PMDs. Manufacturers are focused on the light sources and interface ICs to realize these goals.
The first cost driver for light sources is the semiconductor material used. There are two basic optoelectronic wafer materials used for 10-GbE transmission: gallium arsenide (GaAs) and indium phosphide (InP). These III-V materials are well-behaved, have superior quality, and are available from leading semiconductor laser-diode suppliers. The 850-nm wavelength standards such as 10GBase-SR and 1000Base-SX use GaAs wafers to manufacture the VCSELs that are the light sources for these transceivers. The 1310-nm standards traditionally have used InP. The rule of thumb has been that GaAs is one-fifth the cost of InP on a material basis, due to a difference in wafer size and their relative volumes.
The second cost difference is the structure of the light source. In the mid-1990s, the market witnessed the commercialization of VCSELs to support 1000Base-SX. VCSELs had three inherent advantages: circular beam for easier coupling, low-threshold currents for reduced power consumption, and wafer testing to enhance manufacturing. Within 10-GbE, 850-nm VCSELs are the dominant light-source structure for the 10GBase-SR standard. For the 1310-nm 10-GbE physical interfaces, the light source has used a DFB structure. However, there are three alternatives to this approach: Fabry-Perot (FP) laser, GaInNAs VCSEL, and short-cavity distributed Bragg reflector (SC-DBR).
The FP laser is the most popular approach for 1310-nm 1000Base-LX links. FPs represent a proven technology with an inherently lower cost than a DFB because of its simpler structure. The disadvantage is that there’s no other market for a high-data-rate FP. The 1000Base-LX market benefited greatly from the SONET/SDH short-reach market.
The SC-DBR is a new type of structure.* The advantages of an SC-DBR are that it yields high output powers at low-threshold currents and high-frequency performance. The disadvantage is that this is a new structure, so quality and reliability must be proven.
The GaInNAs VCSEL is the most attractive structure since it uses GaAs wafers for lower costs. It also has very-low-threshold currents, which result in low power consumption. But there are two key challenges to this approach: output power and wavelength. At 10-GbE data rates, it has proven difficult to reach the necessary output powers. Also, the GaInNAs material system has to be engineered further to reach the nominal 1310-nm wavelength; it naturally emits light at a lower wavelength.
Meanwhile, interface IC technology has harnessed the relentless march of Moore’s Law. Most interface ICs on the market today use 0.13-µm CMOS with next generation products targeting the 90-nm CMOS process. The use of CMOS has greatly reduced power consumption and cost, and as the 90-nm CMOS process is adopted, the market can expect this trend to continue.
Additionally, there have been great strides in electronic dispersion compensation and the use of equalizers to achieve full 10-Gbit/sec line-rate equalization. These innovations are driving the development of the draft 10GBase-LRM standard in the IEEE.
Campus networks that use MMF for optical data transmission have been surveyed many times recently. The intent is to establish reasonable link-length specifications for MMF transmission. Survey results of a broad array of campus networks have shown a good correlation. Specifically, >90% of the links are <300 m. Poor correlation occurs at increments <300 m. Unfortunately, there are no surveys of sufficient scope that evaluate the address link lengths in finer granularity.
Link-length surveys are particularly important for the current LRM standards work since the penalty for transmission distance adds fundamental costs to the complexity of the transmitter and receiver-each 10-m span adds up. These extensive surveys, developed between 1996 and 1999, convinced an array of industry experts that 300-m transmission is required so that an interface can be broadly accepted and successful.
The 10GBase-SR specification is designed for (a) 300-m link length on the higher bandwidth (2,000 MHz-km) MMF and (b) 30-80 m on lower-bandwidth (FDDI-grade) fiber. These links are most frequently found in either the horizontal interconnect to support high-bandwidth applications or the vertical riser links. This reach would also be appropriate for applications such as high-density point-of-presence interconnections.
The 10GBase-LX4 PMD is designed for 300-m link lengths on all MMFs and 10-km singlemode link length. This ability to service longer links enables LX4 to completely service the backbone link in premises networks. Additionally, vertical riser applications using early generation low-bandwidth MMF will often require LX4 to meet the guaranteed 300-m link length.
As currently envisioned, 10GBase-LRM will only be specified for a 220-m link length on all MMF, most notably FDDI-grade fiber. The target of the IEEE committee drafting this standard, P802.3aq, was to achieve a 300-m guaranteed link length. That has recently been reduced to 220 m. This reduction in specified transmission length enables earlier introduction of commercial interfaces but sacrifices transmission distance on low-bandwidth fibers. However, there is an expectation that the market will drive transceiver vendors to be able to guarantee longer distances even over the lowest-bandwidth fibers.
The choice of a PMD optical interface for MMF will always be decided on a cost basis. Certainly, a serial approach such as 10GBase-LRM has always proven to be more cost-effective in the long run. But the 10GBase-LRM draft is targeting an optical specification that is nearly identical to the SMF interface 10GBase-LR while adding functionality in the form of full-line-rate 10-Gbit/sec equalization. The philosophy behind LRM is that the powerful equalization in CMOS interface ICs will enable lower-cost optical interfaces.
However, the focus for transceiver vendors must be to enable longer-reach performance. That will likely force a reduction in optical performance, and the cost of the optics will become a second priority. And that may enable LX4 to continue to dominate the low-bandwidth MMF market longer than initially estimated by industry analysts. Additionally, confusion around the use of LX4 versus LRM will likely increase the adoption rate of 10GBase-SR links.
The 10GBase-LRM approach will succeed since it enables a serial architecture. High-density 10-GbE applications require serial interfaces to achieve port density >16 in a single chassis slot. The most demanding enterprises and data centers will drive the consumption of such interfaces.
Cost, power dissipation, single footprint for all reaches, media, and distance required can contribute to deciding which 10-GbE PMD is the best fit for the application. Each PMD has advantages to support different species of links in the premises network-and since each premises network has unique needs, the network administrator must determine the best solution for the network.
Matt Traverso is senior product marketing manager at Opnext (Fremont, CA).
*K. Shinoda et al., Tenth OptoElectronics and Communications Conference (OECC 2005), July 2005.