BY JEFF D. MONTGOMERY
Copper has always been embattled. Moving into the 1950s, building on World War II technology, microwave links started biting into the copper market.* Fiber-optic-signal transmission seriously attacked copper in the 1980s. By 2000, fiber totally displaced copper in new submarine and medium-to-long-haul terrestrial telecommunications links and gained a significant share in telecom access and business premises networks.
The last frontier for copper-signal transmission is in "intra-enclosure" and "inter-enclosure" (between cabinets of equipment) links. Hundreds of tons of copper go into twisted-pair cables and printed wiring board planar signal interconnects. These links are short-microns to a few tens of meters. Through the 1990s, fiber has increasingly penetrated this application and is now accelerating.
Steady progression
Tellabs was a pioneer in intra-equipment fiber interconnect in the late 1980s, with 46-Mbit/sec LED-driven multimode-fiber links within its TITAN series digital-crossconnect switch (DCS). As telecom data throughput exploded in the 1990s, DCSs grew to as many as 70 cabinets per system, requiring lots of twisted-pair copper interconnection and with inter-cabinet data rates moving up to OC-12 (622 Mbits/sec). This trend is continuing; copper interconnect has become unfeasible and DCS vendors globally have converted to fiber interconnect. DCS applications continue as the leading user of fiber internal interconnect.
The most recent battleground between copper and fiber intra-enclosure interconnect is in very-short-reach applications: from a few centimeters up to 10 m. This market is mainly in servers and routers. Designers in this field typically have evolved from a computer-versus telecom-background and are accustomed to moving high data rates over short distances by copper. As both data rate per signal and the number of parallel signals have increased rapidly over the past five years, the copper-transport-versus-reach challenge has multiplied.
With substantial U.S. Defense Advanced Research Projects Agency (DARPA) support, plus significant R&D in Europe, parallel-stream signal transmission via optical fiber advanced steadily through the 1990s. That was aided greatly by various vendors-led by Honeywell-driving vertical-cavity surface-emitting-laser (VCSEL) technology to commercial viability. By 1999-2000, numerous equipment vendors designed multifiber links into their equipment. The Infineon PAROLI, however, was the only high-channel-count multifiber transceiver to achieve high-volume commercial shipments through 2001. Demand substantially exceeded supply capacity through this time period.
The 2001 recession brought near-term demand more into balance with supply and provided an opportunity for other vendors to bring multi fiber transceivers forward. Infineon introduced its second-generation PAROLI-2, and developed Molex as an alternate source. Over a dozen multifiber transceiver vendors now offer evaluation units, and several will achieve commercial status by the end of 2002. Meanwhile, VCSEL technology and multiple-source availability have advanced rapidly, adding singlemode and 10-Gbit/sec capability.
Advancement will continue
Multifiber links with 40-Gbit/sec throughput are now commercially available, and 48-fiber links with over 100-Gbit/sec capability are available from companies like Xanoptix for evaluation. Before 2006, 256-fiber links, 10 Gbits/sec per fiber (2.56-Tbit/sec throughput), will be feasible.
The global consumption of multifiber transmit link components will accelerate from only $49 million in 2001 to $1.25 billion by 2006 as detailed in the Table. DCS equipment will remain the leading application but with faster significant growth forecasted in server and router use.
Packing 48 VCSEL emitters plus 48 driver optoelectronic ASICs, plus other devices and parts, into less than 10 cubic inches, yet designing for low-cost assembly and test, is a major task. Heat transfer and radio-frequency/optical crosstalk problems must be solved. Yield must be high. Fortunately, VCSELs and ASICs can be tested on-wafer, and known good die can be diced into Nx12 or other arrays, or 1xN arrays can be stacked. That greatly reduces device and assembly costs for volume production.
The global consumption of devices and parts in multifiber transmit link component production will accelerate from only $14.1 million in 2001 to $414 million in 2006. Multifiber cable, for shelf-to-shelf and rack-to-rack interconnect, will lead parts value.
Packages will also be a major cost contributor. The cost share contributed by active devices will drop slightly as volume increases.
Reference
C. DeCusatis, Handbook of Fiber Optic Data Communication, 2nd edition-Chapter 1, "History of Fiber Optics," by J.D. Montgomery.