Modular subsystems will play a key role in future network architecture
The applications driving the growth of the Internet are globally distributed, interactive, and bandwidth intensive. They are diverse in their demands on the network in terms of delay, bandwidth, reliability, and security, and are in demand by both businesses and individuals. As evidenced by the ongoing market shakeout, supporting these applications profitably is no easy task.
Nowhere are these requirements more demanding than in 10-Gbit/s optical networking. While the Internet has cemented TCP/IP as the dominant routing protocol for data, the transport environment remains segmented: Ethernet in the enterprise, packet-over-SONET, ATM in the wide-area network, and SONET rings in the metropolitan area network. This segmentation has traditionally been supported by divergent requirements in data communication and telecommunication networks. However, for the first time with 10-Gbit/s optical networking, the transport protocols share a uniform data rate and optical transmission medium and are being positioned to meet the same market requirements.
With the confluence of the datacom and telecom worlds, new network infrastructure must combine the characteristics of both. Network equipment designers must be prepared to meet the low-cost and time-to-market requirements of the datacom world, along with the stringent reliability and performance requirements of the telecom world.
The problems associated with integrating optics and electronics, designing complex high-speed systems, and providing multiprotocol services at multigigabit rates seem at odds with the market requirement of rapid development and deployment of reliable, cost-effective infrastructure. Equipment designers are addressing this disconnect by adopting a more modular architectural model, leveraging externally developed subsystems to get more functional products to market faster.
This horizontal, modular approach offers multiple benefits. Instead of building an entire system from the ground up, incorporating modules reduces development requirements, enabling network equipment vendors to devote the bulk of their engineering resources to differentiating their product while meeting ever-increasing time-to-market pressures. In addition, a modular approach assists in reducing the overall system footprint—a crucial benefit for system manufacturers under pressure to produce smaller solutions. Modules have the added benefit of promoting higher performance by reducing board-level interconnects, lowering noise, and reducing EMI issues. Last but certainly not least, a higher level of integration drives down power and cost.
In short, a modular approach leads to more cost-effective development of higher-speed, more-intelligent, reliable systems. Advanced 10-Gbit/s optical-networking modules combine and optimize an entire chain of optical, electrical, and software functions from a system perspective, rather than creating subcomponents to support generic functionality (see Fig. 1).
Service providers expect next-generation 10-Gbit/s network equipment to provide a richer feature set and higher port density while reducing cost and power on a per-port basis. To make such systems a reality, network-equipment designers need even more-compact, low-power, and cost-effective subsystems than are available today. Advances in electronics, optics, and packaging promise to deliver the means to develop next-generation modules with a higher level of flexibility, scalability, and performance, as well as reduced cost.
Important advances in optics technology for both short- and long-reach applications are emerging, including long-wavelength VCSELs and integrated tunable lasers. Three directly modulated laser technologies are suitable for short-reach serial 10-Gbit/s applications—850-nm VCSELs, 1310-nm Fabry-Perot lasers, and 1310-nm distributed feedback (DFB) lasers. For long-reach applications, externally modulated DFB lasers with frequencies locked in the 1550-nm C- and L-bands are the only option.
The 850-nm VCSEL technology is suitable only for very short-reach applications (<200 m) because of high chromatic dispersion and multimoded propagation. The 1310-nm Fabry-Perot lasers support operation across longer distances (at least 2 km), and are also limited by dispersion effects. VCSEL and Fabry-Perot lasers enable the design of low-cost, miniaturized optical transmitters because they are easy to align to fibers and do not require the use of bulky optical isolators or peltier cooling devices. Unfortunately, their short reach limits application outside of enterprise and POP/central office applications.
Distributed feedback lasers support operation to at least 12 km with direct modulation and to very long reaches with external modulation because of their high spectral purity. However, they are relatively costly and require a complex transmitter design with optical isolators and temperature stabilization. Uncooled DFBs lift the requirement of temperature stabilization but cannot address the complexity of transmitter integration.
Many efforts are under way to develop longer-wavelength VCSELs. A 1310-nm VCSEL represents the ideal solution for short-reach applications—allowing for low-cost, low-power, miniaturized transmitter assemblies, and also providing the >12 km reach required for metro-area networks and large enterprise applications. VCSEL operation at 1310 nm has been demonstrated by a number of groups in the past year, although challenges remain in the areas of device lifetime, temperature range, speed, and output power. Given this progress, it appears that 1310-nm VCSEL technology will mature in the next few years, leading to dramatic cost, power, and size reductions of short-reach modules.
Long-reach applications demand WDM-compatible operation, which today means complex optics trains comprising laser, modulator, power control, and fixed-frequency wavelength locking components. Recent advances in planar integration allow these complex trains to be combined into one or two components while also making them frequency tunable. These new components will enable the development of long-reach 10-Gbit/s physical-layer modules that deliver flexibility and ease of integration for DWDM links.
Complementing these advances in optics technology, a revolution is under way in high-speed electronics. Serializer/deserializer (SERDES) technology has a strong influence on the performance, power dissipation, and cost of physical-layer modules. The responsibility for meeting the stringent jitter and signal-quality requirements of 10-Gbit/s network standards falls primarily to SERDES circuits.
Early SERDES designs relied on expensive and power-hungry technologies such as silicon germanium (SiGe) and gallium arsenide (GaAs) to achieve the level of performance required for 10-Gbit/s operation. This year, we have seen the commercialization of first-generation designs in 0.18-µm CMOS. These designs offer cost and power advantages over SiGe and GaAs implementations, but current CMOS performance limitations constrain the level of integration and sensitivity that can be achieved. Additional circuitry is required at the module level to work around these issues.
Over the next year or two, smaller CMOS geometries will provide the additional performance needed to bring 10-Gbit/s SERDES technology to the mainstream. In addition to stand-alone SERDES designs, it will be possible to integrate SERDES functionality into complex ASICs that also provide protocol processing and control functionality. This has major implications for 10-Gbit/s modules.
Cost will be reduced as a single standard-technology ASIC replaces several parts implemented in specialty processes. Power consumption will be reduced both through process improvement and through elimination of high-speed interfaces between SERDES and protocol processing circuits. Performance will be enhanced as module vendors take advantage of new integration possibilities. As a result, modules incorporating advanced functionality—such as adaptive data recovery, forward-error correction, and integrated protocol monitoring—will replace simple transponder modules without any penalties in power or cost.
Innovations in packaging will combine with the advances in electronics and optics to deliver modules that surpass the expectations of an integration-hungry market. Vendors are already building high-speed serial electronics into optical subassemblies to improve signal integrity and advance miniaturization.
With reduced component counts and simplified optical trains, this trend will move forward to low-cost design leveraging advanced packaging technologies from both the electronics and optics industries, including optical microbenches and dense organic substrates. Within two years leading-edge short-reach modules will look like compact FLASH cards with fibers, and long-reach DWDM modules will be smaller than a business card.
Another element of growing importance is the software interface to modular subsystems. This interface is embodied in application-program-interface (API) libraries that streamline system integration by providing a software foundation for module configuration, customization, and monitoring. An API must provide the flexibility to create a fully custom implementation while making it simple to configure the module for standards-compliant operation. Module APIs need rich abstraction layers that can control advanced protocol processing, such as SONET signaling and packet filtering, through a straightforward user interface. Software can also provide features beyond those directly implemented in hardware. By encapsulating and extending hardware functionality into software layers that can directly interact with system-level applications, an API dramatically reduces the overall time required to integrate modules into the equipment environment.
Claude Denton is vice president of product architecture at Network Elements, 15425 SW Koll Parkway, Beaverton, OR 97006. He can be reached at email@example.com.