As pointed out in Part 1 of this two-part series (see Laser Focus World, March 1999, p. 59), the development of vertical-cavity-laser (VCL) technology to produce VCL arrays has led directly to the fabrication of parallel optical links for digital-signal processing. This article looks at developments in VCL markets and technology.
The development of parallel optical-interconnect links has advanced rapidly over the past five years. Substantial funding of several programs in this field has been provided by the Defense Ad vanced Research Projects Agency in the USA and by ESPRIT in Europe. Nearly all of this effort has been based on monolithic arrays of VCLs and photodiode monolithic arrays, as well as electronic integrated-circuit (IC) chips in hybrid optoelectronic integrated-circuit formats.
These parallel links are aimed at two major markets: physically parallel, functionally serial, minimum-footprint links, mainly for intershelf and intercabinet interconnect in communication equipment, and physically and functionally parallel links-transmission of one overall signal group originated in parallel format, such as a central processing unit, output to its remote memory and/or memory bus.
These links and supporting components initially were designed for 860-nm-band multimode transmission. Gallium arsenide VCLs for this format are much easier to design and fabricate than 1310-nm-band single-mode indium phosphide VCLs. There is now, however, significant development proceeding on 1310-nm-bandwidth VCLs. Meanwhile, massively parallel 1310-nm transmitters based on single-mode edge-emitting laser diodes have been commercially introduced.
The data-rate-versus-distance parameters and applications of the primary massively parallel link developments in the United States are all designed for relatively short reach: 0.5 m to a few hundred meters with per-channel data rates in the 200-1000-Mbit/s range.
By mid-1998, the POINT link (AMP; Harrisburg, PA) and OETC (Honeywell; Minneapolis, MN) were becoming commercially available. The POLO (Hewlett-Packard; Palo Alto, CA) link is awaiting company management decisions, and the Motorola (New Orleans, LA) OptoBus is in the process of management restructuring and product reorientation.
Shipments of parallel-link modules in 1997 consisted of a few hundred trial or evaluation units provided to prospective users. Commercial consumption accelerated, continued in late 1998, and will continue through 1999 and beyond. Current VCL producers include AMP, Cielo, Emcore, Furukawa Electric, Hewlett-Packard, Honeywell, Mitel, Siemens, Sumitomo, and W. L. Gore & Associates.
Within the 840- to 980-nm range, VCL design, materials processing, and other elements are broadly similar, and developers have demonstrated excellent devices at 840, 850, and 980 nm. The 850-nm link, however, has won the position of standard for data interconnect in Fibre Channel. This is giving it a strong lead in the current design-in. Nearly all VCLs, to date, have also been designed and fabricated for multimode transmission. Single-mode VCLs, however, are feasible.
It is also possible to fabricate VCLs at longer wavelengths. VCLs for the 1300- to 1350-nm range have been demonstrated by several laboratories, and sample evaluation VCLs at 1300 nm will be available in 1999-multimode and single-mode. VCLs for the 1500- to 1550-nm band are technically feasible, but the economics do not appear to favor them for commercial production for at least another four to five years.
In addition to the intracomputer applications and premises local-area-network interconnects such as Fibre Channel and gigabit Ethernet, there is very rapid growth of the requirement for high-data-rate interconnection at rapidly increasing data rates at distances in the 1-30-km range for campus and metropolitan computer network interconnect. Many of these links are now operating at a few gigabits up to tens of gigabits, leading to high-end application of dense WDM.
In the 1300-nm band, single-mode VCLs will find a substantial market in these applications. There also is a rapidly expanding need for intershelf and intercabinet optical-interconnect links, data rates up to a gigabit and upward, and interconnect distances up to several hundred meters.
Coincidentally, there is great pressure to reduce the space ("footprint") required in the shelf board/modules for these transmitters and receivers. This demand now is primarily in telecommunication central-office equipment, which uses several hundred fiber links per system. This application presents a strong opportunity for 1300-nm VCLs.
The global consumption of massively parallel optical interconnect components in 1998 is estimated at $59.5 million. This will accelerate to $893 million in 2003. Dramatic growth will continue, driven by the continuing trend to higher-data-rate input/output per equipment unit, the growing complexity of networks, demand for higher processing throughput, and pressures for lower cost per channel and reduced size. Strongly rising quantity growth will be partially offset by a continuing rapid decline of average prices.
Optoelectronic transmitters dominated global massively parallel (masspar) optical interconnect components in 1998, with a combined 51% share of $30.5 million. The transmitters value share will expand to 65% or $582 million by 2003 (see figure on p. 57). Currently, nearly all masspar links use ganged individual transmit/receive units of conventional discrete design. Over the next five years, a rapidly increasing share of transmit/receive (T/R) functions will be accommodated by highly integrated array T/R units with most transmitters based on VCLs.
The average per-channel prices of all masspar components will fall substantially over the forecast period. The integrated T/R per-channel prices for a specific T/R configuration will fall less rapidly than the prices of connectors and cable. This will be due to the T/R trend to more-expensive, higher-data-rate, synchronous (low skew) arrays. Calculated over all T/R types, data rates, and sizes, the average price of T/R modules will rise while the average price of connectors and interconnect fiber falls. This will cause the increase in the T/R share of total value.
Not All VCLs
Although vertical-cavity surface-emitting laser diodes have dominated masspar development investment over the past six years due to potentially very low cost per diode, other emitters will find substantial use. The VCL will dominate synchronous transmission, but nonsynchronous links will be significant users of other emitters in addition to VCLs. There is now significant use of light-emitting diodes (LEDs) in low-data-rate links, up through OC-3 (155 Mbit/s). In 1997, Tellabs (Lisle, IL) used approximately $40 million of LED-based parallel links in its Titan 5500 digital-crossconnect-switch production. (This is not counted as massively parallel in the ElectroniCast data, because the module has only two T/Rs.) At the other end of the scale, network-equipment developers foresee a substantial need for 10-Gbit/s per fiber (OC-192) interconnects by 2001 for some of their cabinet-to-cabinet links. In this application, multifiber links will compete with dense WDM-based single-fiber links.
Massively parallel optical-interconnect links transport signals that are synchronous or nonsynchronous. Synchronous signals are involved when a single overall signal (such as from a microprocessor) consists of multiple, interrelated signal streams that must be reassembled into a single stream at the destination (such as a memory). For error-free reassembly, the transport timing of the subsidiary signal streams must be tightly controlled (low skew).
Nonsynchronous massively parallel signal transport, in contrast, consists of a number (four or more) of functionally unrelated serial signal streams transported over physically parallel paths, such that substantial cost and space reduction can be achieved by consolidation into a single link with single-footprint transmit/receive modules. There is now a substantial market for nonsynchronous masspar links, estimated at more than $50 million global consumption in 1998. This excludes more than $300 million of physically parallel nonsynchronous links in 1998 that are not counted in the masspar consumption values because they do not meet the single-footprint T/R module criteria. o
JEFF D. MONTGOMERY is chairman and STEPHEN MONTGOMERY is president of ElectroniCast Corp., 800 South Claremont St., Ste. 105, San Mateo, CA 94402; [email protected].