VCSEL-based optical interconnects are already established in short-reach networks. Chris Simoneaux of Picolight looks at their prospects in other areas of the network.

Optical interconnect technology is rapidly permeating all levels of the modern network. This includes optical switch interconnects for very-short-reach (VSR), rack-to-rack or switch-to-aggregation-device interconnects, short-reach storage-area network (SAN) and local-area network (LAN) links for enterprise networks, as well as longer-reach links for optical metro/access interconnects. The dominant laser communications for these optical interconnections are transceivers based on vertical-cavity surface-emitting lasers (VCSELs). They are solidly entrenched in the network for SAN and LAN backbones. With the appearance of 10 Gbit/s VCSEL-based transceivers, VCSEL array transceivers, and 1310 nm operation, the technology is rapidly migrating "upward" into ultra-high-bandwidth enterprise switching and SAN applications, "inward" into backplanes of large optical switches, and "outward" into optical metro and access applications (see ). The current position Optical interconnection technology has evolved from primitive forms of Fiber Distributed Data Interface (FDDI) to the 10 Gigabit Fibre Channel standard currently being developed. Today, Fibre Channel supports speeds of up to 2.125 Gbit/s and is moving forward to meet requirements for 4 Gbit/s and 10 Gbit/s data rates. Other standards organizations, such as IEEE 802.3ae for 10 Gigabit Ethernet, Optical Internetworking Forum (OIF) and InfiniBand are developing their respective optical link technologies, utilizing VCSEL-based optics and further illustrating the need for mass deployment of high-speed networking components. Many networking OEM suppliers are initiating physical-layer designs even before standards have been adopted, hoping for future harmonization. Meanwhile, VCSEL technology has spread into the VSR optical switch interconnect space via parallel optics, and is poised to infiltrate the metro/access space via longwave (1310 nm) VCSELs (see ). Here, VCSEL technology will help resolve the problem of insufficient bandwidth in some of the optical switch and metro/access segments that have limited access to the high-end metro core and long-haul segments of the network. Until now, the only option for this segment was high-end, very expensive transport-style optics, making it difficult to justify these ports on cost grounds. Early implementations VCSELs were first implemented in the Gigabit Interface Converter (GBIC), 1 9, and Gigabit Link Module (GLM) form factors. This was followed by Small Form Factor (SFF) solder-in design, backed by a multi-source agreement (MSA) and later by the Small Form Factor Pluggable (SFP) MSA module. There is a trend toward lower power, more compact, higher-density modules with system-management functions that will also be essential to the evolutionary process. VCSELs and the new standards The demand for high bandwidth and density, combined with low power consumption requirements, is leading to the deployment of parallel optical modules using arrayed VCSELs and PIN diodes. Current efforts are focused on modules using four- and 12-channel VCSEL array technology with passive coupling optics (for "piping" the light to the outside world), and multichannel alignment processes. Parallel optics are ideal for the VSR links that are required in backplanes for rack-to-rack, switch-to-switch, and switch-to-aggregation device interconnects. These interconnects are earmarked for use in OC-192 VSRC, multichannel Gigabit Ethernet, proprietary 1.6 and 2.5 Gbit/channel links, multichannel XAUI extension, and T11.1 HIPPI (see ). They are expected to migrate into short-reach enterprises as four- and 12-channel InfiniBand modules in the next few years. Today s parallel solutions at 13.125 Gbit/s/channel will move to 5 and 10 Gbit/s/channel solutions with an effective aggregate bandwidth of up to 120 Gbit/s. At this time 1 Gbit/s solder-in and pluggable modules are the transceivers of choice, in the form of a GLM, GBIC, 1 X9, SFF or SFP solution. The NCITS X3T11 is developing a 4 Gbit/s Fibre Channel specification, which may result in a 4 Gbit/s optical transceiver implementation, although there is not yet a working group to define physical implementations. If 4 Gbit/s optical modules are implemented, they will probably take on the same form factor as the current SFP products. Speeding ahead Some believe that a jump to 10 Gbit/s for transceivers is more appropriate. The NCITS, IEEE, and OIF are converging on data rates around 10 Gbit/s. The NCITS X3 body has approved a T11.2 optical working group to develop 10 Gigabit Fibre Channel physical implementation standards. IEEE has a task group called 802.3ae to develop standards for 10 Gigabit Ethernet. The 10 Gbit/s Fibre Channel transceivers will most probably leverage the design of 10 Gbit/s Ethernet as both share plans for a four-lane electrical interface and similar data rates. Another emerging standard is InfiniBand which, if successful, would be integrated into the Fibre Channel link structure during 2002 or 2003. This switch/bus architecture reduces the types of buses in the system to just CPU and InfiniBand. The attached InfiniBand network interface cards will provide the appropriate translation protocol and the electro-optic functions. InfiniBand s native data rate is 2.5 Gbit/s, with plans for serial, and four- and 12-channel parallel versions, resulting in 2.5, 10, and 30 Gbit/s aggregate bandwidth implementations, respectively. Most physical-layer developments will focus on faster data rates and improved system/link management capabilities. An initiative essential for future optical modules is on-board management services. Management implementations are currently being developed with both OEM and optical module vendors, and the end goal is a common electrical interface and data structure to spur widespread adoption by the industry. Equipping for the last mile As mentioned earlier, there is a critical bandwidth deficiency in the metro/access portion of the network comprising the final few kilometers leading to businesses and residences. One solution is the extension of optical fiber through the access network across the last mile. This not only requires a critical mass of deployed fiber, but also the readiness of service providers, and the availability of affordable, easily managed, multiplexing equipment. It is this last point where economical, efficient and high-performance VCSEL-based interconnects are being prominently featured. VCSEL transceiver technology is uniquely qualified to play at all the key aggregation points in the access network. New gigabit data rate VCSEL-based transceivers consume less than 500 mW and require just a fraction of the typically required board space. They offer improved cost-effectiveness and flexibility, while handling not only single-wavelength point-to-point transmission, but also multiple protocols, shared passive-optical network interfaces, and in the near future, coarse WDM or multiple protocols on separate wavelengths. If there are alternatives to the access network, they probably will not be derived from today s generation of DWDM technology. This International Telecommunications Union (ITU) grid technology is two orders of magnitude more costly than VCSEL optics, and optimized for distances over 100 km rather than the 10 km range of metro/access networks. 1310 nm VCSELs The "sweet spot" of the metro/access market will be served by 1310 nm VCSEL transceivers at the 10 Gbit/s data rate, achieving the necessary reach up to 10 km (see ). Picolight has successfully developed 1310 nm VCSEL-based metro/access-specific transceivers, and plans to sample these interconnect solutions later this year. Picolight s approach - and the approach most leading technologists support - is to "stand on the shoulders of proven 850 nm VCSEL technology". 1310 nm VCSEL technology is poised to replicate the successful 850 nm oxide VCSEL transceiver model. Materials, packaging and fabrication approaches that have been proven in the high-volume short-reach VCSEL transceivers are key to developing the volume metro/access market, which dwarfs the network core in terms of unit shipments. In the past few years systems architects and technology planners have begun writing 1310 nm VCSEL-compatible interface specifications into standards such as 10 Gigabit Ethernet, and are debating codification of this technology into Ethernet in the First Mile (EFM). The IEEE 802 LAN/MAN Standards Committee (LMSC) last year formed a new study group to develop a project proposal for this standard, which will apply the proven and widely used Ethernet networking protocol to the customer initial access market. Picolight is working closely with this group to help shape standards and promote the use of future-proofed VCSEL solutions that will provide the critical interconnect technology. The new generation of 1310 nm VCSEL transceivers also promises to be capable of supporting multiple wavelengths, far enough apart (four, 10 or 16 wavelengths, fairly widely spread) to maintain direct modulation, and therefore cost-effectiveness. This future-proof feature can allow an order of magnitude increase in optical bandwidth across the access or metro fiber, without the severe cost penalty which is imposed by DWDM architectures. Packaging advances Optical component packaging becomes critical as high-speed and high-volume requirements converge. Today s current high-speed, high-volume, cost-effective "Ethernet" or "enterprise" VCSEL transceiver packaging model is expected to extend further into the link layers to include the serializer/deserializer (or SERDES), Transmission Protocol, and eventually, the system protocol function. As InfiniBand is implemented, we may see the same trend, as the integration will expand up to the switch/bus interface. OEMs want pluggability, simple board implementation, and more EMI-friendly and robust integrated packaging. Design trends include EMI-enhanced all-metal housings, panel/bezel mounting and hot-pluggability. Long-wavelength 1310 nm transceivers are expected to use much the same packaging technology as 850 nm solutions. The most important access transceiver packaging approach will be hot-pluggable, compact transceiver configurations such as those used in the enterprise and SAN space to enable optical service provisioning on demand. Pluggable transceivers allow customers and service providers to purchase a 32- or 64-socket concentrator or switch and then populate ports as needed without bringing down the network. This concept has revolutionized the way enterprise LAN and SAN networks are designed and deployed, and will have a similar impact on access equipment design as this market evolves. Manufacturing advances In collaboration with industry leaders, such as Newport, Picolight has worked to establish a high-volume automated test and assembly capability for 850 nm solutions that will map well to the coming generation of 1310 nm products. With process advances and new alignment algorithms, significant improvements in alignment capabilities have been achieved over the past several years. Precision computer-controlled motion stages and sophisticated alignment algorithms ensure maximum coupling efficiency. These algorithms are also now being built into the firmware on the manufacturing equipment controller board to reduce intra-instrument communication latency and drive down the cycle time for the alignment process. VCSEL transceiver technology is already firmly entrenched in the optical enterprise for SAN and LAN backbones, and is rapidly penetrating the market for high-speed optical switching equipment. Soon, as many as 250 million desktops and homes worldwide will become targets for the gigabit bandwidth necessary to deliver high-speed Internet access for a host of new broadband entertainment and information applications. The foundation has been established for VCSEL-based interconnects to help deliver on that promise.