MICHAEL SLUYSKI, AMCC
Over the past few years, the pace of change in optical networking has steadily accelerated to accommodate escalating requirements for higher bandwidths and the drive toward convergence of disparate services such as voice, video, and data across common networks. In addition, the need for more economical deployment options across both wide-area and metropolitan-area environments is driving requirements for longer fiber-optic span distances, lower-cost equipment, and higher port densities.
The evolution of dense wavelength-division multiplexing (DWDM) is also dramatically increasing the capacity of backbone networks by multiplexing from 16 to 80 channels onto a single fiber, with tighter channel spacing offering the potential to drive the number of wavelengths even higher. However, the increase in channel counts has simultaneously raised the bar for the underlying switching capacities required to manage the additional connections. In addition, the migration of DWDM into metropolitan-area-network (MAN) applications poses significant additional challenges because the channels must be transparently managed to efficiently carry a variety of traffic types to meet the disparate requirements of many different users.
Traditional fiber-networking technology requires repeated conversions from the optical domain to the electrical domain and back again at virtually every amplification or switching point. Because every conversion adds significantly to the overall expense of the system, network architects are constantly looking for ways to avoid any unnecessary conversions, especially within long-haul optical transport networks. But because conversions to the electrical domain will continue to be required at network termination points and for periodic signal monitoring along the transport links, these conversions from the optical domain will also need to be handled as efficiently as possible.
Taken together, these requirements for effective and efficient optical-layer management have now become critical factors that will determine the ultimate success and economic viability of tomorrow's wide-area-network (WAN) and MAN infrastructures.
Within current WAN and MAN optical networks, which are primarily based upon Synchronous Optical Network (SONET) and Synchronous Digital Hierarchy (SDH) topologies, the cost of regenerating optical signals can be very expensive, especially when it requires full SONET termination equipment at every regeneration point. Advanced optical amplifiers and repeaters have provided significant improvements in transport distance before requiring a full conversion of signals to the electrical domain, although they are most useful in core backbone links where high volumes of undifferentiated traffic have been pre-groomed for long-distance transport within an exclusively SONET environment. Yet, even in these relatively homogeneous all-SONET environments, optical-layer management can be a key factor in maintaining system integrity.Regardless of whether the transport-link distances can be achieved without requiring a full conversion from the optical to electrical domain, partial conversions at key points can be vitally important for monitoring and assuring adequate signal quality. For example, any link that includes an optical amplifier or regenerator should also incorporate active signal-monitoring capabilities, which will require optical signal splitting and an optical-to-electrical conversion of some portion of the signal.
With the ultimate goal behind DWDM deployment being the provision of more capacity, real-world MAN and WAN environments need to optimize the use of each wavelength's bandwidth. In WAN links traversing the network core, this goal of optimized utilization has typically been accomplished by pre-grooming all traffic such that uniformly grouped signals can be efficiently trans port ed long distances with a minimum of intervening decision points along the way. But for links that are nearer the edge of the transport networks, new-generation equipment needs to provide a higher level of traffic monitoring and grooming cap abil ities within the optical domain to achieve a balance of flexibility, performance, and band width utilization.
For instance, in most cases it would make little economic or practical sense to invest in DWDM and then to map individual GbE connections across individual wavelengths on a one-to-one basis. Therefore, the push for aggregation of multiple connections can very quickly lead to a mix of heterogeneous nonconcatenated traffic traveling within a shared wavelength, with a multitude of different end destination points.
Because such nonconcatenated, channelized signals cannot simply be switched as a single optical wavelength, next-generation crossconnect switches must have the intelligence to quickly recognize in dividual signals and handle them ap propriately, while imposing only a min imal degree of over head for converting between optical and electrical domains.
In essence, new-generation managed optical networks are migrating toward a "distributed-switching" model in which lambda switches and intelligent Layer 1 crossconnect capability is distributed at various points along the optical rings (see Figure 1). This architecture provides seamless and efficient Layer 1 management of heterogeneous traffic types throughout the network, without sacrificing performance or flexibility in either the core or edge environment. Such a global distributed-switching architecture is equally adaptable to using dedicated wavelengths packed with homogeneous traffic for long-haul point-to-point transport or for flexibly managing heterogeneous traffic on dynamically allocated short-haul wavelengths.
From a crossconnect perspective, the emerging need for supporting a managed optical layer with a distributed switching environment presents significant opportunities and challenges for both semiconductor-level and module-level manufacturers. To achieve the required performance levels, next-generation crossconnects need to be closer to the network by providing Layer 1 switching as opposed to traditional Layer 2 switching.
Both asynchronous and synchronous crosspoint designs will play key roles, depending on the environment. For example, higher-speed asynchronous crosspoints will enable heterogeneous MAN implementations to efficiently support different types of native-mode traffic within the same ring. But in longer-haul networks, the major leap forward will come with the innovative use of synchronous-switching crosspoints to provide ultra-high performance. In these instances, the crossconnect will be more of a "time-space-time switch" rather than a simple "space switch" as used in traditional asynchronous designs. These new-generation synchronous crosspoints will incorporate configurable Layer 1 grooming capabilities that can selectively switch SONET or time-division-multiplexed signals between any combination of inputs and outputs, thereby flexibly provisioning high-speed connections to serve changing network requirements.
To the extent that these signal grooming and management capabilities can be implemented at the optical layer using intelligent high-speed synchronous ICs, system manufacturers can more efficiently build high-density, high-performance switching equipment that makes optimal use of Layer 1 optical switching. For example, next-generation synchronous crosspoint-switch ICs will offer the ability to selectively groom out and switch any STS-1 from within an STS-48 or STS-192 stream. Such devices will allow complete flexibility for provisioning of IC-level managed optical connections from any STS-1 input to any STS-1 output. Non-SONET traffic mapped to STS-N-equivalent containers and protocol-independent wrapped traffic can be switched within the same crossconnects that manage SONET traffic.
Deployed at the edge of long-haul SONET transport environments, such high-speed, high-density synchronous grooming switches can optimize bandwidth utilization while efficiently making Layer 1 access decisions to partition out traffic to outlying Internet protocol, GbE, Asynchronous Transfer Mode, Fibre Channel, or other Layer 2 switches. Localized Layer 2 functions such as routing and policy management can be appropriately handled by the outlying switches while the Layer 1 access switches provide high-speed interfaces to the core transport levels and high-performance IC-level switching/grooming of DWDM wavelengths.
Layer 1 access devices effectively support the need for aggregation of heterogeneous traffic to fully use the DWDM-enabled MAN/WAN bandwidths (see Figure 2). At the same time, they also deliver high performance by efficiently collapsing and managing the ADM and DSX client interface functions as close as possible to the underlying optical-network layer.
Synchronous Layer 1 devices also offer the inherent flexibility to act as either a SONET/SDH or data-communications switch. The difference in the two roles is that in SONET, the grooming capability is directed at the STS-1 level or below, whereas in a data switch, the device simply functions as a 2.5-Gbit/channel synchronous switch that can be rapidly configured. By leveraging this IC-level compatibility between data-communications and telecommunications implementations, new-generation Layer 1 optical switching devices are helping to move these markets closer together, further paving the way for transparent convergence of different types of traffic (e.g., voice, data, and video).
At the semiconductor level, the primary challenges involve effectively combining the required densities and performance with the programmable flexibility needed for dynamic configuration of the switching parameters. For example, as future DWDM-driven network capabilities scale to pack significantly more channels onto a fiber, the next generation of high-speed switching systems will have to scale dramatically as well. Ultimately, the foundation for achieving such scalability has to begin at the semiconductor level.
With system designers currently looking at building switches that can support from 1,024 to 2,048 ports, IC manufacturers have to provide highly integrated components that combine significantly more channels on the same chip along with the ability to flexibly configure them in virtually any combination. In addition to reliably switching multigigabit data paths within ultra-tight jitter and skew parameters, new crosspoint ICs also have to deliver minimal power dissipation characteristics to accommodate overall board-level and system-level constraints.
A careful blending of compatible process technologies is required for the consistent and reliable production of high volumes of crosspoint ICs that combine the required densities, high-speed performance, and low-power characteristics. For example, the use of proven mainstream CMOS technologies are needed to achieve the high port densities at a reasonable cost, while the selective use of compatible silicon germanium (SiGe) processes can be leveraged to deliver the high-speed, low-jitter signal-processing demands.
SiGe is a proven technology that provides performance in excess of 50 GHz that combines both high-speed performance and ease-of-integration on a single die, while leveraging all of the low-cost, proven processes currently used for silicon-only devices. Essentially, SiGe works by introducing germanium into the base layer of an otherwise all-silicon bipolar transistor. This germanium doping process re-engineers the band gap of ordinary silicon, thereby speeding electron conduction across the base region, producing supercharged heterojunction bipolar transistors (HBTs) that can operate at speeds greater than 50 GHz.
Most importantly for ultra-dense crosspoint IC designs, because germanium has a lattice-like crystalline structure that is very similar to silicon, it allows a high level of on-chip integration with existing silicon circuitry and can be efficiently produced in existing silicon fabrication facilities. The overall cost of SiGe is only about 10% higher than a standard BiCMOS process, at equivalent geometry levels, and is a small fraction of the cost of exotic processes such as gallium arsenide.
As overall network requirements continue to call for more channels, the crucial linchpin in next-generation networks will support increased scalability, flexibility, and performance in high-speed optical-switching systems. At the network level, the use of distributed switching architectures will provide significantly increased flexibility for effectively handling the convergence of diverse traffic types carried on many different optical channels.
However, the foundation for successful implementation of distributed switching environments ultimately lies at the IC level. Here, next-generation synchronous crosspoint-switching devices are providing the ability to efficiently groom and switch individual signals along with the configurability to dynamically provision crossconnect links as required by network traffic demands. Together, these emerging capabilities deliver a new level of managed optical-layer capabilities that will form the foundation for tomorrow's network infrastructures.
Michael Sluyski is a systems architect at AMCC (San Diego).