There are three general ways to achieve aggregate bandwidths of 10 Gbit/s. One is to combine multiple channels in parallel. Another is to combine four or more wavelengths along a single fiber using coarse wavelength division multiplexing (CWDM). Also, it is possible to have a serial solution using a single channel at 10 Gbit/s. Recent developments suggest that all of these will prove to be viable solutions. Cielo launches serial transponder At the recent Optical Fiber Communication conference (OFC), Cielo Communications launched a 10 Gbit/s 850 nm serial transponder. The transponder is currently sampling to development partners with volume production commencing in July 2001. Picolight announced a similar product in August 2000, which will have general availability in the third quarter of 2001. The product is designed for OC-192 very short reach (VSR), 10 Gigabit Ethernet and proprietary backplane interconnects up to 300 m in length. Other products targeted at the VSR application use either 1310 nm serial or 850 nm parallel solutions. At present, 1310 nm serial solutions suffer from high power consumption, and high testing and packaging costs associated with the use of edge-emitting lasers, while 850 nm parallel solutions involve more complicated interconnect designs using multiple components. "Cielo s transponder takes the best from both worlds by offering a high-speed serial solution over MMF that is less expensive, requires less than half the power, and is half the size of current alternatives" said Jeff Bisberg, product marketing manager at Cielo. Blaze has a coarse solution Blaze Network Products, a company based in Dublin, California, is one of the leading proponents of coarse WDM technology at 850 nm. Several other companies, including Agilent and Finisar, have announced longer wavelength CWDM products. Blaze s Afterburner XGSX transceiver (see page 43) combines four wavelength channels, each operating at 3.125 Gbit/s, and uses four VCSELs operating at 778, 801, 825 and 850 nm. The transceiver contains an optical sub-assembly (OSA) that contains VCSELs, drivers, detectors, TIAs, CWDM filters, and multiplexers and demultiplexers in a single, low-cost package. Blaze has developed an internal process to assemble the elements of the OSA using passive alignment. The company obtains oxide VCSELs from a number of suppliers, including Emcore. Blaze claims that the Afterburner-XGSX, which measures 1.2 2.57 0.53 inches, is the smallest pluggable 10 Gigabit transceiver in the industry. The initial pricing of the product is well below $800 for modest volumes, and product shipments are due to begin in early June. "Customers have been pushing the transceiver industry for years for higher port densities to support the increased numbers of links on the network," says Kirk Bovill, director of product marketing for Blaze. "While the initial implementations of 10 Gigabit Ethernet will not require a high number of ports per blade, they will in the near future." Alvesta uses a four-channel format At OFC, Alvesta announced the availability of its Model 3100 optical transceiver, which uses four channels and parallel optics to deliver an aggregate bandwidth of 10 Gbit/s. The transceiver (see ) has a footprint of 15.5 34.6 mm (0.61 1.36 inches) and is priced in the $300600 range, depending on volume. Each module contains arrays of four 850 nm VCSELs and four PIN detectors, and transmission takes place over a 12-fiber ribbon (four fibers are used in each direction, and four are dark). This type of product was also the subject of a multi-source agreement between Agilent and Mitel (see Compound Semiconductor May 2001, p17). Alvesta claims that their solution offers the lowest power consumption (less than 1.0 W) and lowest cost of any 10 Gbit/s solution. explains Alvesta s contention that a four-channel parallel solution is more cost-effective than serial (one-channel) or 12-channel parallel solutions. A small channel count requires the use of high-speed components and leading edge technology, which is more expensive - this is represented by the curve labeled "technology cost". The cost of implementing more than one channel increases incrementally since more components are required - this is shown in the "linear aggregation" curve. In reality, the aggregate link cost is reduced in small arrays due to the benefits of integrating the components, while for larger arrays the opposite is true due to lower yield and increased assembly complexity. The "integration and yield" curve indicates the actual aggregate link cost. The "total cost" curve, which has a minimum in the four to six channel range, is the product of the technology cost and the aggregate link cost. Two equipment suppliers, Force10 Networks and BrightLink Networks, have already chosen Alvesta s 10 Gbit/s Model 3100 optical transceivers for VSR applications. Mitel s Smart OSA approach Mitel is developing parallel fiber modules with aggregate bandwidths of up to 30 Gbit/s (for the 12-channel, 2.5 Gbit/s version). The company has developed its Smart OSA high-volume manufacturing technology to accurately position the VCSELs or detectors with respect to the fiber arrays. The Smart OSA unit begins with a single chip containing an array of four devices, either VCSELs or detectors, spaced 250 m apart to match the spacing of fibers in a standard MTP or MPX connector. The chips are fabricated by Mitel. Using a solder reflow technique for very accurate alignment, one or more such chips are attached to a leadframe, as shown in . Because the two end elements are close to the edges of the chips, several chips can be combined to form an array with up to 12 elements. The leadframe also contains two circular alignment holes that correspond with guide pins in the MTP and MTX connectors - this ensures that the fibers in the connector align accurately with the optical devices. This approach results in a compact, simple and robust system, which is compatible with the high temperatures required for the reflow soldering process. The use of chips with four devices allows several combinations of module to be built, namely transmitters and receivers with 4, 8 and 12 channels, or transceivers containing both four VCSEL and four detectors. Importantly, four-element chips have far higher yields than chips with arrays of eight or twelve devices.