Will Fiber Go to the Desktop?

Declining Optical Component Costs Can Help, But Are Only Part of the Story

The debate over whether copper or optical fiber to the desktop is the best solution has gained momentum as new networking technologies such as high-speed Ethernet, asynchronous transfer mode (ATM) and Fibre Channel have appeared on the horizon, and as the political winds have shifted toward creating a nationwide information superhighway. Until recently, optical fiber has had a clear advantage in only one part of the overall local-area network architecture -- the backbone -- where fiber's tremendous bandwidth capacity, noise immunity and reach have made it the logical choice. Copper has been the obvious choice for so-called horizontal cabling. Within the past year, however, the gap between the cost of bringing optical fiber all the way to the desktop and higher-speed copper-based solutions has narrowed considerably.

Now, tremendous competition is heating up in the final 100 meters from the cabling closet to the desktop as copper and fiber technologies converge from opposite directions to address the next step in networking -- the 100-megabit-per-second LAN. On the one hand, the lure of copper remains strong as vendors extend the capabilities of copper wire to carry these higher data rates, while on the other hand, optical cable, connector and component manufacturers are bringing costs down to the point where they can compete with copper-based products as a result of increased volumes and improved manufacturing.

Several trends are clear. Bandwidth requirements are increasing because of the growing use of client-server applications, the increasing number of nodes connected to LANs and the increasing use of image-based, graphical and multimedia applications. The 10- and 16-megabit-per-second Ethernet and Token Ring LANs of today are starting to strain at the loads being placed on them. New technologies up to the 1000 megabit-per-second range are appearing.

At the same time, businesses are under pressure to be more competitive while holding the line on costs. Those who already have LANs installed want to get the most from their existing investments, while those contemplating new installations must weigh current costs against future needs.

Deciding which way to go for a particular installation is a complex one, however, and involves not only facts but perceptions and uncertainties as well. Perceptions over the relative costs of purchasing, installing and maintaining copper and fiber-based solutions, uncertainties over a business's future growth, its application requirements and other factors affecting bandwidth needs, and uncertainty over not only the role new technology will play in creating new services and capabilities, but its effect on both increased competition and the volume of basic components manufactured, thus driving down costs further.

Reducing Component Costs

Over the past few years, manufacturers of active fiber-optic components have made significant strides in lowering the cost of the basic ingredients that go into fiber-optic hubs, bridges, routers and network interface cards. And with typical original equipment manufacturer (OEM) costs of manufacturing, marketing, distribution and support, every dollar reduction in component cost should result in a three to four dollar savings to the system end user.

For example, the first FDDI transceiver, introduced in 1990, employed a package and pinout specification originally promoted by Hewlett-Packard, AT&T and Siemens. It cost approximately $500 apiece in small quantities. Today, through the evolution of transceiver design and manufacturing, fiber-optic transceivers of similar performance have fallen in price to about $125 in small quantities, a 75-percent reduction in just three years. In volume, these prices have fallen to below $100 each. As a result, FDDI adapter prices have now dropped below $1000 per connection, and are rapidly converging on the price per connection for the different emerging 100-Mb/s Ethernet technologies.

One of the most significant factors in this substantial cost reduction is the market's evolution from the original fiber-optic FDDI transceiver with a 2 x 11 pinout, costly die-cast housing, and expensive MIC/R connector, or the alternative ST simplex transceiver module, to the new FDDI and ATM duplex transceivers with a 1 x 9 pinout, plastic housing and SC connectors.

The new 1 x 9 transceivers are approximately 60 percent of the size of the older 2 x 11 transceivers. As a result, OEMs using these transceivers are able to substantially reduce their cost per FDDI or ATM port because they are able to install more transceivers in a given amount of space. The original design utilized a difficult to manufacture, and therefore expensive, die-cast housing in order to prevent excessive electromagnetic and radio-frequency emissions. The new design uses a plastic injection molded housing capable of offering equal or better shielding while significantly reducing manufacturing costs. At the same time, the internal printed circuit has been made smaller, and has changed from an expensive ceramic substrate to a more cost-effective, multilayer, glass-epoxy substrate.

Better transceiver shielding has another, less obvious effect in reducing network equipment costs. Hubs, routers, switches and other network equipment may have multiples of eight or ten transceivers in a single chassis. The emissions generated by all these transceivers is cumulative, and the total can quickly escalate. In order to meet FCC regulations, designers of such equipment need to spend significant time and effort to prevent these emissions from escaping the box.

HP's 1 x 9 duplex SC transceivers, for example, are specified to meet FCC Class B emission levels with a typical margin of 10 dB when tested in free air. The effect of this extra margin can be significant. By limiting individual transceiver emissions, the overall emission level of any particular system will be reduced. This, in turn, reduces not only the design effort required to attain system-level FCC Class B compliance, but reduces the direct manufacturing costs for shielding, gasketing and other methods used to prevent radiation from escaping. These savings, of course, can be directly reflected in lower end-user costs.

Significant savings in optical component manufacturing costs have also been realized through the reduced cost for the integrated circuits within the fiber-optic transceiver. Through natural product evolution -- finer design rules, improved designs, improved process control -- higher volume manufacturing is possible, dramatically reducing the cost of these integrated circuits.

Besides the cost-cutting gains made by reducing the transceiver size, lowering IC costs and reducing emissions, OEMs have two other options for providing lower cost fiber optic solutions to their customers. One option is to change to a different, higher transmitting frequency (shorter wavelength of light), something that is being reviewed by standards organizations.

Ninety-five percent of all desktop connections have a distance requirement of less than 100 meters. In light of this, it becomes reasonable to ask why a technology capable of transmitting data reliably up to 2000 meters and beyond should be used when something as reliable over a shorter distance would suffice. This is precisely the question prompting recent standards committee activities to add the option of using lower cost, shorter wavelength transceivers to address this need.

By switching to 800-nanometer (nm) light-emitting diodes (LEDs), transceiver costs can be significantly reduced. Traditional FDDI transceivers use 1300-nm LEDs in order to achieve the 2-kilometer specification in the current standard. But the cost of a 1 x 9 FDDI or ATM transceiver with an 820-nm LED is 25 percent less than that of one based on a 1300-nm LED, primarily because the production quantity of 800-nm LEDs is an order of magnitude greater than that of 1300-nm LEDs.

The reliability and performance of 800-nm technology have been well proven over time, as all Ethernet and Token Ring fiber-optic backbone connections currently in use are based on this technology. Although 800-nm technology is limited to hundreds of meters when operating at the 125- to 155 Mb/s FDDI and ATM data rates now becoming popular, as opposed to several kilometers for 1300-nm technology, this is more than adequate for the vast majority of desktop connections.

Network equipment OEMs striving to reduce optical components costs even further have yet another option. Optical transceivers integrate an LED driver IC and quantizer IC into the package along with the LED and optical elements. Savings of up to 25 percent can be realized by using discrete ICs and separate optical "ports" such as HP's HFBR-0300 and HFBR-0400 series of 1300- and 820-nm components instead of fully integrated transceivers.

Designing with discrete ICs and optical ports requires more engineering effort by the OEM, but with readily available printed circuit board layouts from component vendors, that effort is minimized. In addition, as volumes increase, the piece part cost savings can more than offset the additional engineering effort. Today, most Ethernet and Token Ring fiber-optic links use the 820-nm discrete component approach and benefit from substantial savings.

Cabling Conundrum

Although there have been, and will continue to be, dramatic reductions in the cost of fiber-optic components and the systems using them, by some estimates network hardware accounts for only 25 percent of the total cost of a typical network installation, copper or fiber-optic. The remaining 75 percent of the cost is due to labor (35 percent), cabling (20 percent) and connectors (20 percent). Until now, the general perception has been that fiber is more costly to install than copper. But while higher production volumes and advances in connector and optical component technology have greatly reduced the cost of fiber-optic solutions, more stringent EIA/TIA-568 Standard Category 5 requirements addressing the higher emissions inherent in 100-Mb/s data rates have increased the costs associated with copper plant components, installation and testing.

When system-level cost savings are considered over the life of a network, it can be argued that fiber-optic solutions can actually be less expensive than copper. Fiber's superior reliability at higher data rates can reduce operating costs by minimizing network outages, and its higher bandwidth can produce considerable savings by "future proofing," eliminating the need to pull new cable when the network is upgraded to support even higher data rates.

Upgrading existing copper networks to support 100-Mb/s data rates often requires significant cable plant redesign, including not only more expensive cable and electronics, but higher labor, testing and warranty costs. Nearly all 100-Mb/s networks require Category 5 unshielded, twisted-pair (UTP) cable and connectors. The sole exception is the 100Base-VG standard being promoted by HP, IBM and AT&T for Ethernet and Token Ring LANs, which can operate over Category 3 UTP. The bulk of UTP cable installed today is Category 3 or lower. Consequently, new Category 5 cable must be pulled to support the higher data rates used in high-bandwidth LANs like FDDI and ATM.

Meeting FCC emission limits and providing adequate crosstalk isolation at 100 Mb/s can also require the installation of new wall outlets and patch panels. Most conventional wall outlets and patch panels are connected with PVC jacketed copper conductors from several inches to several feet long. The connecting copper wires are highly susceptible to near-end crosstalk because they lack sufficient twists and insulation for 100-Mb/s data rates. Outlets and patch panels meeting the new Category 5 requirements cost significantly more than their predecessors, and have been available only since January of 1993. As a result, any Category 5 cabling installed prior to that date with Category 3 or 4 apparatus will need to be retrofitted before it can support 100-Mb/s data rates.

Extreme care must be exercised when installing copper cable near heavy machinery, heat sources, high-voltage transmission systems and other sources of electromagnetic interference. In addition, both ANSI and EIA/TIA specifications for Category 5 UTP place stringent requirements on the way the cable is pulled and terminated. Because both specifications require that pairs not be untwisted by more than 1/2 inch at termination points, they limit how hard the cable can be pulled during installation to 25 pounds. Fiber-optic cable, in spite of its depiction as being fragile, is able to withstand more than 150 pounds of pull, with some cable designed for horizontal applications able to withstand up to 225 pounds of pull.

Testing and certification requirements also become more stringent when copper is used at high data rates. Tests for crosstalk, impedance mismatch, cable length and near-end crosstalk should be conducted at each cable end, and should include not only the cable, but all the system connectors, including patch panels and jumper cables. And they should be tested at the frequencies specified in the standards -- 31.25 MHz for FDDI TP-PMD and 45 MHz for ATM if MLT-3 coding is approved as the UTP standard.

Threshold of Pain

How far can the installed base of copper technology be pushed? How fast will fiber-optic network equipment prices come down? How quickly will network bandwidth requirements increase? There are no clear answers. As system planners contemplate crossing the 100-Mb/s threshold, where fiber-optic and copper-based technologies converge, they will need to carefully examine near and longer-term costs and tradeoffs in light of current and future applications, and they will have to make complex and often painful decisions. It remains unclear how soon they might make the jump to fiber because in the world of business, solutions that are "good enough" often take precedence over those that are best.

The question isn't whether fiber will reach the desktop, but when. At the 100-Mb/s threshold, cabling, connector and labor costs are already on a par with copper equivalents. In the meantime, fiber-optic component vendors will persist in finding ways to reduce the cost of their products while exploring other ways of making fiber to the desktop more competitive with copper. As is typical in highly competitive environments, OEMs can assume the unprecedented reduction in fiber-optic component prices will likely continue its trend downward for several years to come, enabling them, in turn, to create lower-cost fiber-based network equipment solutions.

# # #