By Judith Meester, Sr. Product Marketing Mgr., Infineon Technologies
The ever-increasing demand for bandwidth is largely attributed to users at the network edge -- office buildings and corporate/educational campuses -- who rely on high-bandwidth-dependent data transfer to succeed in today's swift business environment. But residential customers are also demanding higher speed Internet access separate from their telephone service and at an affordable price.
In fact, it is not beyond the realm of possibility that optimistic predictions about the need for bandwidth could be realized in the next few years. The growing implementation of cable modem and DSL services demanded in residential areas is placing a heavy burden on central offices. This situation is expected to get even worse over the next few years.
The requirements of office buildings, corporate campuses and university campuses prove greater than those of residential customers. Data transfer and email -- with or without attached data files -- is the most frequently used method of communication between workers, regardless of proximity. Combined with video conferencing, voice and fax, such data transfer is placing an ever-increasing load on central offices.
To stay in the game, carriers are forced to examine various ways of meeting customer requirements.
Together, these data-hungry residences and workplaces are propagating a greatly increased need for bandwidth at the central office in metro regions and, therefore, a need for larger communications pipes in long-haul networks. There are a few ways to address these requirements, including dense wave divisional multiplexing (DWDM), which enables data destined for distinct locations to be carried on separate wavelengths on the same fibre; installing more fiber or lighting up dark fibre; and increasing the capacity of existing SONET infrastructures by running greater data rates on existing fibre.
The implementation of DWDM has been the popular choice of many carriers in the metro region. For very short distances, wavelengths can be centered around 850 nm. For distances of 2, 15 and 40 km, wavelengths centered around 1310 nm, spaced at 8 or 10 nm, can be implemented. For greater distances, the C Band of the ITU grid, ranging from around 1530 to 1565 nm, is used. The C Band provides 32 or more wavelengths at 100 GHz, or 0.8 nm spacing, and these wavelengths are compatible with current erbium-doped fiber amplifier (EDFA) technology. In fact, gain-flattened EDFAs can be used to help ensure consistent optical power distribution among wavelengths. It is also possible to use 64 wavelengths at 50 GHz spacing within the C Band. Many vendors are also extending their wavelength capacity into the L Band of the ITU grid (wavelengths greater than 1600 nm), which requires altered amplification technology only recently available.
Using DWDM technology, wavelengths can be added and dropped at desired locations by means of filters, which allow other wavelengths to be expressed through to other locations. Some vendors use banding technology, splitting the C Band into four or more sub-bands. This requires that wavelengths at the edge of each band be used as "guard wavelengths." The main advantage to the use of banding technology is that fewer filters, couplers and splitters are necessary in a system, thus insertion loss and cost are decreased. The disadvantage is that a group of wavelengths within a band must be added and/or dropped at the same location, and those not desired at that specific location are lost.
Whether or not banding technology is used, the disadvantages of a DWDM system are high cost; difficulty in implementing equal optical power distribution among the channels; and the insertion loss suffered by the use of multiple splitters, couplers and filters. In addition, dispersion limits the distance onto which the number of wavelengths can simultaneously be transmitted on one single-mode fiber, and this effect is compounded as the number of wavelengths increases.
For both metro and long haul applications, many carriers have shown foresight by installing "dark" or unused fiber in anticipation of increased traffic. Some have planned to install more fiber to accommodate greater bandwidth, resulting in a cost disadvantage, due to the cable itself, as well as labor associated with its installation.
Installing infrastructures capable of greater data rates is another method of addressing the requirement for increasing bandwidth, especially in long haul networks. In the metro region, some carriers are choosing high data rates over DWDM. Some carriers are planning to install systems capable of OC-192 (9.95 Gbps) and 10 GE (up to 13.2 Gbps, including 25 percent overhead).
In accordance with electrical interface standards for OC-192 -- OIF SFI-4-01.0 and OIF SPI-4-02.0 -- system vendors are currently fabricating modules containing OC-192 chipsets.
Although the demand for OC-768 (39.8 to 43 Gbps) in long haul systems is not expected for another few years, it is important for system designers and manufacturers to begin planning modules with the appropriate mechanical, optical and electrical features and interfaces. Thus, semiconductor manufacturers must be ready, ahead of the curve in the design and development of chipsets capable of processing these speeds. Once such chipsets are available, system vendors can begin evaluation of the chipsets and feedback to the semiconductor vendors, while designing their modules and systems. To this end, work is underway to define OIF SPF-5 and SFI-5 standards, in anticipation of OC-768 requirements.
There are multiple challenges involved in the design and manufacture of such high-speed devices. Complementary metal oxide semiconductor (CMOS) fabrication, the preferred technology due to relatively low cost, is not typically suitable for such high data rates. Semiconductor vendors must therefore use SiGe, InP or GaAs materials systems. The latter two are costly due to low yield, partially resulting from small wafer size. Although slightly prone to jitter, B7HF SiGe is stable. However, B8HF, which includes Cu metallization, has a higher current density than Al metallization, leading to higher frequency capabilities that enable higher speed devices.
Packaging is a second challenge, since bond wire lengths cause detrimental impedances and greatly impact the integrity of signals from high-speed devices. The module design must preserve the features of the high speed chip set, thus some collaborative R&D work between the chip and system vendors is necessary. The design tends to be a delicate balance between optimizing optical, electrical and mechanical features and interfaces, while compromising on module size as little as possible. It is clear that multiple design and wafer fabrication iterations, as well as collaboration with system vendors, are necessary in order to optimize the integrity of both high-speed chipsets and modules.
Additionally, in order to implement OC-768 seamlessly, systems in the existing network should have adequate capacity to accept the new high-speed line cards. In many cases, this may simply be a matter of upgrading backplane capacity and/or inserting the new modules. However, replacing existing systems is unlikely to cost as much as installing additional fibre.
Despite the challenges, high data rates remain an option of choice for many existing and new carriers. For long haul networks, this is the most direct way to address future demands for bandwidth, since it allows carriers to simply upgrade their operational infrastructures. It saves on additional loss of optical power due to insertion loss and dispersion associated with DWDM. Finally, it saves the cost of additional fiber installation.
Since many carriers are currently planning for the future implementation of OC-768 technology, system and semiconductor vendors are involved in the design and optimization of appropriate equipment before the demand for bandwidth increases beyond existing network capacity.