Opto-electronic device costs influence integration into local area networks
Opto-electronic device costs influence integration into local area networks
Semiconductor processing techniques are generating affordable optical products for use in data communications networks
Roger Ady
honeywell Inc.
Although discrete gigabit transceiver modules are available, future communications interconnections will be made using arrays of vertical cavity surface-emitting laser sources and gallium arsenide detectors, and fiber ribbon cables. These types of links are beginning to appear in the marketplace, and with them, effective data rates are limited only by the upper limit of the opto-electronics (currently about 2 Gbits/sec) and the number of fibers.
For several years, market forecasts have predicted that high-volume, high-speed data communications needs were just around the corner--two years away, at most. But, because the data communications market perceives fiber-optic products as too expensive, this market has not materialized.
For example, approximately 5% to 8% of existing local area network ports are fiber-optic, but are used almost exclusively in backbone networks. This percentage is actually a reduction from the 10% market share two years ago. The perceived high cost of fiber optics versus unshielded twisted-pair copper has prevented the use of fiber in nearly all applications except those that mandate higher performance, higher reliability, security, interference immunity or wide bandwidth.
Part of this perception may be real: Evidence suggests that some equipment manufacturers, value-added resell ers and installers are charging premium prices for fiber-optic ports/links in the mistaken belief that the LAN market will pay a premium to get the higher performance of fiber.
Fiber distributed data interface technology, the local area network standard written specifically for fiber optics use, was touted by leading industry analysts as the answer to the world`s need for more data communications bandwidth. Unfortunately, FDDI technology has not captured a significant share of the LAN market.
The reason is its high cost of implementation. The FDDI standard defines a technical protocol that is difficult to install simply and inexpensively, as interpreted by network providers. Furthermore, it specifies an expensive fiber-optic connector and relatively costly long-wavelength technology for opto-electronic devices.
In the telecommunications market, most of the capital expenditures for fiber optics have been made, and demands for additional fiber and optics are flat or declining. Suppliers to that market are beginning to feel the pinch and have turned to the data communications arena to cover their shortfall.
Dual markets
Data communications comprises two markets--long haul and short haul for wide- and metropolitan area networks, respectively. Obviously, the long-haul data communications market needs to make the same kind of cost/performance tradeoffs executed in the telecommunications market. By contrast, the short-haul or LAN market is cost-sensitive and invokes a different set of decision values.
The LAN market has experienced rapid growth during the last five years, averaging more than 20% per year. Approximately 65% of the LAN installations are Ethernet, and less than 30% are token ring. FDDI comprises approximately 3%, and high-performance parallel interface, Fibre Channel and asynchronous transfer mode make up the balance.
Products sold in the LAN market are based on gallium aluminum arsenide double-heterojunction light-emitting diode, silicon p-type-intrinsic-n-type photodetector and silicon integrated circuit technologies. There has been little demand for higher performance, long-wavelength solutions. Manufacturers of opto-electronic devices that function at higher data rates are including the low-cost assembly and packaging of current product offerings, but will have to control semiconductor component costs to remain competitive in the LAN market.
Avoidance of exotic and expensive raw materials is obvious. Indeed, 1-gigabit-per-second short-wavelength modules at or below present prices for 155-megabit-per-second transceivers are possible, with lower bit-rate modules correspondingly priced at less cost. Greater cost savings will be achieved at short wavelengths now that GaAs integrated-circuit receiver technology has been successfully transferred from the laboratory to production.
Local area network users are striving for additional bandwidth. But these users have already made formidable investments in unshielded twisted-pair copper and multimode fiber. They have consistently accepted solutions that minimize acquisition cost and use as much of the already installed material as possible.
An example is the collapsed backbone concept in which a virtual LAN provides the full bandwidth to every user. It uses existing cable and fiber, and calls for changing only the hub or router; it does not change the data rate or the type of circuitry needed for implementation.
In the LAN market, several standards are being proposed to implement Ethernet at a 100-Mbit/sec data rate. These proposals give users an upgrade path through a familiar protocol, but, on the debit side, they still remain unscalable broadcast protocols.
A scalable, switched protocol such as asynchronous transfer mode is an improved approach to meet LAN future needs. It could totally supplant 100-Mbit/sec Ethernet and FDDI if it becomes affordable.
Manufacturing advances
Competition to sell opto-electronic devices into the LAN market is keen, and prices have consistently fallen between 10% and 20% per year--a trend that does not appear to be slowing. Most cost reductions have resulted from assembly and packaging improvements, but not from basic semiconductor technology changes.
For example, long-wavelength lasers and LEDs are made of InGaAsP, a quaternary compound that until recently was difficult and expensive to manufacture.
Advances in semiconductor processing technology now allow production of long-wavelength LEDs at costs comparable to, but still considerably more expensive than, conventional 850-nanometer ternary (GaAlAs) LEDs. Even so, worldwide, only approximately 10 suppliers offer 1300-nm LEDs suitable for use at 155 Mbits/sec, and perhaps three suppliers offer products suitable for use at 622 Mbits/sec.
Lasers that operate at 1300 and 1550 nm are available as edge-emitters, which means the lasing cavity within the wafer is formed as the various layers of the laser structure are built up on the wafer substrate. To expose the laser cavity, the wafer must be cut to expose its edge. In practice, the wafer is cut into thin ribbons that are then cleaved into dice. The cleaving process leaves mirror surfaces on the ends of the die, producing a laser cavity.
Each laser die must not only be separated from others on the wafer, but also be partially fabricated into a finished packaged product before it can be tested. Bad dice have some manufacturing value added before they are identified and discarded, so good lasers are always more expensive than good LEDs.
Further adding to the cost of the laser assembly is the need for an optical power monitoring detector. Edge-emitting lasers have a large temperature coefficient of threshold current that can vary output power widely over temperature for a given operating condition. Consequently, they require feedback circuits to maintain their operating point. Although a laser-power-monitoring photodiode is a relatively low-cost PIN, its cost and that of the associated feedback amplifier incur costs not required by other solutions.
VCSEL technology
The discovery of vertical cavity surface-emitting lasers might prove as important as the invention of the LED. They are made using almost the same process as that of making double heterojunction LEDs, and their cost to produce is essentially the same as that for LEDs. The vertical cavity surface-emitting lasers can be designed to have a low thermal coefficient of threshold current; consequently, they may not require optical feedback control. Moreover, they can be made inherently eye-safe, eliminating the need for open fiber control loops.
The relatively low coherence of vertical cavity lasers eliminates problems associated with modal noise. Because of their low threshold currents, they operate without the need for heavy heat-sinking, enhancing reliability. Beam shape is symmetrical so coupling to fibers is simplified. These devices can be made to emit at 850 nm, so inexpensive, silicon, moderate-speed detectors or GaAs high-speed detectors can be used in corresponding receivers.
As backup, inexpensive and readily available 780- to 850-nm compact-disk-type lasers can be used instead, although they require optical feedback. Both solutions are compatible with the synchronous optical network protocol over at least 300 meters of 62.5- and 50-micron fiber at ATM bit rates through optical carrier level 12 (622 Mbits/sec) and higher.
Three companies are producing 850-nm vertical cavity surface-emitting laser-based products--Motorola Inc., Vixel Corp. and Honeywell. Hewlett-Packard Co. has a product working at 980 nm, and several other companies are pursuing vertical cavity surface-emitting laser technology.
Even if semiconductor manufacturers were vertically integrated and made their own material, the relative cost of a high-speed, long-wavelength detector would be many times that of a short-wavelength detector consisting of gallium arsenide or silicon.
High-speed, long-wavelength detectors are InGaAs structures manufactured on InP substrates. The InP substrate material cost on the open market is approximately $80 per square inch. By comparison, GaAs material costs $15 per square inch, and silicon is $1 per square inch. InP, GaAs and opto-electronic silicon are available in 2-inch, 3-inch and 4-inch wafers, respectively
The lower data rates and shorter distances associated with LANs--combined with a desire to use low-cost technology--has resulted in an installed base of predominately 62.5-micron multimode graded-index fiber for data communications applications.
With the advent of higher speed data communications systems, cost incentives have arisen to revisit the issue of multimode graded-index fiber for laser-based applications. Designers have known for some time that it is possible to obtain enhanced modal bandwidth when using lasers, but the issue has never been fully studied and quantified.
Preliminary results from an ongoing study, though, indicate that reliable and robust communications across distances in excess of one kilometer at 1 Gbit/sec are possible in 62.5-micron multimode fiber at an 850-nm wavelength. Even with a full understanding of the technology available, analysts claim it is difficult to forecast that there will be a place for long-wavelength fiber optics in the cost-sensitive LAN environment.
International fiber standards
The U.S. and European LAN markets have a large installed base of 62.5-micron multimode fiber. Japan has standardized on 50-micron multimode fiber. Solutions that employ short-wavelength 850-nm vertical cavity lasers and HS GaAs detectors support these needs at an attractive cost.
However, the long-wavelength solution cannot support 300 meters of transmission distance with fiber diameters below 62.5 microns at higher data rates (greater than 500 Mbits/sec) with LED sources. This technology requires 980-nm vertical cavity lasers or costly edge-emitting lasers. A dual standard could be adopted, but market research reveals this approach is precarious with serious interoperability obstacles.
Most installed fiber is operating at 10-Mbit/sec Ethernet speeds. Only a small portion of the market operates faster--essentially none above 155 Mbits/sec. There is a small market for 100-Mbit/sec FDDI, but it is not expected to grow substantially because ATM is just around the corner.
Manufacturers are beginning to introduce 155-Mbit/sec ATM equipment, but little has been installed to date. Present demand for very high-speed links is limited. Above 155 Mbits/sec, only computer-to-computer, workstation-to-workstation and computer-to-storage applications appear to demand data rates to 1 Gbit/sec. u
Roger Ady is application engineer for local area network products at Honeywell Inc.`s Micro Switch Division in Freeport, IL.