By John Rhea
Open-systems standards continue to emerge that encompass new commercial off-the-shelf technologies, which presents electronics designers with a fundamental decision. Should they keep their products within the realm of proprietary architectures and reduce the size of their potential markets, or should they move to open standards and take a shot at vastly expanding their markets?
Sandards in the military and aerospace electronics business emulate one of the most basic laws of physics - the one that explains the underlying principle of inertia. This law states that bodies in motion tend to remain in motion, while bodies at rest tend to remain at rest.
The bodies in motion in this case are the standards that new technologies spawn, increasingly those based on commercial off-the-shelf (COTS) technologies. Dominating the bodies at rest are the time-honored infrastructures, which extend the lives of old systems and provide aid and comfort to designers and end users alike.
Somewhere in the middle is an overlapping zone, and it is here that tension exists. Technology partisans who share an eye toward eating away at the markets of dominant suppliers are backing the new standards. In this area there is frequently a high infant mortality rate.
"Clean-sheet-of-paper" designs should be the logical homes for new standards, but even here the advantages of proven technologies often outweigh the promise of performance improvements. In the U.S. Air Force's top-of-the-line F-22 fighter, for example, Smiths Industries in Grand Rapids, Mich., is using well-established commercial aviation standards in the aircraft's power distribution system.
Bruce Flater, senior standards engineer at the firm, estimates that many similar systems, such as crash recorders, flight data recorders, and health monitoring systems for jet engines - perhaps 60 percent of the cases - can benefit from the proven ARINC standards embodied in commercial aviation.
Does this mean that old standards are hanging around so long that they impede new technology? Not necessarily, responds Richard Jaenicke, director of product marketing at Mercury Computer Systems, Chelmsford, Mass. "One of the reasons that standards exist is precisely because technology changes," he maintains. "Different technologies move at different speeds, and standards provide a means for those changing technologies to work together."
Jaenicke makes the distinction between the rapidly changing computer processor architectures and the more slowly evolving protocol standards for the buses that connect the processors and thus accommodate the rapid changes. What this process adds up to, he says, is an infrastructure to allow the two disparate technologies to coexist.
Since Mercury is a leading VME supplier, Jaenicke naturally favors that approach over its rival, CompactPCI, noting that VMEbus has evolved with new technology and managed that change in a backward-compatible way.
"Unlike VME, as PCI evolves, it gets its speed boosts at the expense of the number of devices that can be connected," Jaenicke asserts. "Eventually, new technology will need to be grafted onto PCI or it will indeed impede new technology and will need to be replaced."
VME's leading proponent, Ray Alderman, executive director of the VMEbus International Trade Association, or VITA, in Scottsdale, Ariz., defines standards as an "extrapolation of the state of technology at the time," further subdividing them into three categories: physical (the pins), logical (protocols), and electrical (the interface). He calls the last one "the anchor" that enables the systems to work.
"What standards don't do is tell you how to do it; that's left to the designer," Alderman says. "We tell you how to get in and out, not what goes inside [the box or board]." Moreover, he urges that standards be as short as possible, and define only the minimum number of physical, logical, and electrical criteria needed.
Beyond the technical meaning of standards, Alderman explains, is what he calls the business definition: "a mechanism for the creative destruction of invested capital." Implicit in this definition is that it is somebody else's capital that is to be destroyed. The purpose of strategic marketing plans, he says, is to force competitors to spend exorbitant amounts of money or time in order to participate in the market.
The mechanism for this is a proprietary architecture that compels customer loyalty. Familiar examples include the dominance of the Big Three automakers in the days before the competition from imports, and IBM's stranglehold on the mainframe computer market before the advent of the personal computer. Personal computers ultimately became commodity products with their own, expanded customer base.
Alderman likens the decision between building a proprietary or open-systems architecture to the choice between trying to capture 100 percent of a $100 million market or 1 percent of a $100 billion market, Alderman contends. The latter is more lucrative by an order of magnitude.
A more recent example, he recounts, is the jockeying within the industry leading up to the new InfiniBand standard. The process began with Intel Corp.'s definition of the next-generation input/output interface in early 1998 and was quickly followed at mid-year by an incompatible standard, the future I/O promoted by IBM, Hewlett Packard, and Compaq. The two were merged last year to create InfiniBand, but that standard is already under attack by the new RapidIO switched-fabric interconnect architecture being championed by Mercury and Motorola.
Alderman says the tradeoffs inherent in standards revolve around two poles. On the one hand, they can be a negative factor because they can diminish creativity. On the other hand, they create markets. "In fact, both are true," he adds.
Ron Hehr, strategic marketing manager at UTMC Microelectronic Systems Inc. in Colorado Springs, Colo., contends that the biggest factor in sustaining standards, sometimes for decades as is the case with the company's work in MIL-STD 1553, which is the aircraft infrastructure that precludes swapping out cabling and couplers. "You have to redo an entire airplane, not just one box," he says.
This is not to say that standards are irrevocably frozen, Hehr adds. The Society of Automotive Engineers in Warrendale, Pa., has a committee looking at the next generation 1553, and there are other approaches. UTMC engineers have been looking at using the same cable harness to upgrade bus structures, but sometimes the old ways are best. "Ten-megabit-per-second Ethernet has been around for a long time," he says, "and it's not worth ripping stuff out."
Hehr cites another evolutionary technological approach - low voltage differential signaling (LVDS) devices - which are widely accepted in the commercial world. Hehr expects UTMC to have a line of LVDS transceivers for space applications by this fall. These operate at data rates point-to-point of 155 megabits per second, and the company is working on a box-to-box next-generation device capable of moving data faster than 400 megabits per second.
The standards that succeed and survive are the ones that are "shareable," says Rodger Hosking, vice president at Pentek Inc. in Upper Saddle River, N.J. By this, he means that the standards must be sufficiently open for enough vendors to participate.
The standards that have not done well are those based on proprietary architectures, with the accompanying licensing fees and other methods that some companies use in an attempt to corner a market. This is bad enough in the commercial world, but it is even worse for military customers, Hosking says. For products to be sold to those customers, Pentek requires at least six vendors of its own to assure availability. Hosking notes that the U.S. Navy has been particularly concerned about lifetime issues and has been afraid of obsolescence.
Another reason that standards fail is they do not deliver what Hosking calls a "sufficient delta in benefits." This has to be built into the standard from the beginning and, if done properly, solves problems on many levels, not just performance. He cites CompactPCI, which supposedly was invented by Nortel and was not readily available until telecommunications companies picked it up.
"Commercial telecommunications is a huge magnet," he says, and Pentek and others are using CompactPCI to achieve a critical mass for government customers. Nonetheless, Hosking expects VME to be around for at least another 10 years in a variety of forms. "Innovation within a standard is a huge factor," he declares.
Hosking also raises the issue of software standardization, focusing on the vector signal and image-processing library that originated with the Defense Advanced Research Projects Agency in Arlington, Va. This is now a standard platform for writing such algorithms as fast Fourier transforms that can be run on any processor.
A significant newcomer is Linux. Customers are asking for it, Hosking says, and he suggests it may become the operating system of choice for embedded applications in three to four years.
Craig Whitelock, vice president for marketing at Arrow/Zeus Electronics, an electronics distributor in Purchase, N.Y., points out the military desire for seamless connectivity in a communications environment. This, he says, offers another rationale for standards. This was demonstrated in the operations in Kosovo, as the U.S. Airborne Warning and Control System aircraft, satellites, B-2 bombers, and smart weapons received direction from compatible systems.
As dedicated distributors to the military and aerospace markets, Arrow/Zeus officials are concerned with what Whitelock calls the "circle of life" of new designs for integrated circuits, discrete components, and even subsystems such as flat panels integrated into ruggedized chassis. As programs move on into low-rate initial production and then into full production, the participants must stay on top of end-of-life and diminishing manufacturing sources issues.
Cost reduction efforts are conducted in parallel, and Whitelock discloses that one innovation has been a software tool known as Tactrac from TacTech Inc. of Yorba Linda, Calif., which captures data on where a part is in the life cycle.
Steve Paavola, product marketing manager at Sky Computers Inc., a subsidiary of Analogic Corp., stresses the infrastructure issue as the factor that determines which standards succeed and which fail. VME and PCI, for example, each have at least 200 available vendors. "You don't convert overnight," he says. "The customers go with existing technology because it makes them comfortable."
Nonetheless standards do evolve and do complement each other. Paavola notes that InfiniBand started out as a system I/O and, like Sky's own Skychannel and the other new standard, RapidIO, is a packet-switched architecture. He claims a first for Skychannel, the use of LVDS links. This is advantageous for distances of 5 meters between chassis.
He expects the quest for greater speed to continue, using off-the-shelf silicon wherever possible, adding that what is needed is a bridge to the processor and switch technology (for example, cross bars for multiple processors). "We also need bridges to the I/O and PCI technology and to FibreChannel for access to disc drives," Paavola concludes.
Harris RF in Rochester, N.Y., faces a similar problem in meeting existing standards, such as the old Single Channel Ground and Airborne Radio System (SINCGARS), while implementing new technology and still assuring radio users that they will all be able to talk to each other. Steve Elvy, director of engineering, says this has been accomplished with the company's new line of Falcon multiband radios, using commercial standards wherever possible.
The military is already using commercial standards in its radios, he says, and it is essential that suppliers be Internet Protocol (IP) ready and Ethernet capable. Using such devices as the new Motorola Altivec processors, Harris has combined HF for long haul and VHF for short range tactical with long-range satellite capability.
The net result, Elvy says, is that it is easy to hook up and take advantage of COTS because the military is buying COTS for its command and control infrastructure. However, he makes the distinction between what can and cannot be COTS standards. COTS is acceptable for what is inside the radio, the "black box," but not for the actual signaling protocols used in combat situations.
Elvy puts the dividing line between COTS and military standards at the antenna. Beyond that point military-unique standards must prevail for security reasons. "We spend a lot of time on over-the-air waveforms," he notes.
New standards are evolving. One is MIL-STD 188-141B for the third generation of HF networking radios that automatically link to each other on the best frequency and correct transmission errors. The Army is the lead service on this effort, which is centered at the Communications-Electronics Command (CECOM), at Fort Monmouth, N.J.
A parallel effort is in progress with a standards group at the Air Force's communications center at Scott Air Force Base, Ill., which is working on a new waveform to permit 9,600 bits per second bandwidth at distances more than 2,000 miles.
Taking the process a step further, new standards are being developed by NATO under what are known as "stanags," or standardization agreements, in order to provide IP-ready protocols (with a built-in Ethernet interface) for interoperability among the allies. Elvy says he expects those standards to be completed by the end of this year.
RapidIO poised to challenge InfiniBand for embedded markets
CHELMSFORD, Mass. - A dozen of the founding companies of the new RapidIO standard are scheduled to meet this month to firm up a switched-fabric interconnect architecture capable of competing with the InfiniBand standard established last year.
The primary driving forces behind the RapidIO effort are Mercury Computer Systems of Chelmsford, Mass., and the Motorola Semiconductor Products Sector in Austin, Texas. The two companies initiated parallel development programs three years ago and merged their efforts two years ago, says Craig Lund, chief technology officer at Mercury. RapidIO was unveiled in February at the Embedded Systems Conference in Chicago.
Lund maintains that the principal difference between RapidIO and InfiniBand is that the former aims at chip-to-chip applications, and the latter, box-to-box. Driving the new architecture is the consumer community, he adds, and end users are heavily represented in the RapidIO Trade Association, a 501(c)6 nonprofit organization committed to adopting the technology as an open standard.
InfiniBand, which has been supported by Intel Corp. of Santa Clara, Calif., and the principal manufacturers of computer servers, was approved last year. This month's expected final approval of RapidIO should put the two architectures on a collision course for the large embedded markets, such as networking and wireless communications, which require increased connectivity and throughput.
*In addition to Mercury and Motorola, founding members of the trade association include Cisco Systems of San Jose, Calif.; Galileo Technology in San Jose; HAL Computer Systems Inc. of Campbell, Calif.; Lucent Technologies Inc. of Naperville, Ill; Nortel Networks of Billerica, Mass.; Seagull Semiconductor of Hertzelia, Israel; Tundra Semiconductor Corp. in Kanata, Ontario; and Xilinx of San Jose.
Within a month of the February announcement, one of Mercury's principal rivals in this market, Sky Computers Inc., also based in Chelmsford, Mass., announced that it was also joining the association.
RapidIO will be a "99.99 percent commercial" standard, Lund says. Expected applications beyond the embedded market include high-performance digital signal and image processing functions such as digital video broadcasting, medical diagnostic imaging, radar, sonar, signals intelligence, and semiconductor wafer inspection.
Unlike system-to-system protocols, its proponents contend, RapidIO provides the high bandwidth and low latency necessary for intra-system communications. The interconnect is designed to fit inside a field programmable gate array with room to spare for additional functionality.
What this means, Lund explains, is that users can build a system that is not bus-based but fabric-based. The focus is on backward compatibility, he adds, and one of the objectives is to protect the PCI legacy. Software development is independent of the interconnect architecture, so the Rapid IO technology is expected to be transparent to the existing software base.
As part of the company's announcement in March that it was joining the Rapid IO Trade Association, Bob Hoening, vice president of engineering at Sky, stressed the similarities with the company's own interconnect architecture. "RapidIO uses a non-blocking packet switch architecture, which is one of the key components of our existing SKYchannel," he says. However, the announcement also noted that Sky was continuing to evaluate InfiniBand and other complementary emerging I/O protocols.
Lund at Mercury says he expects the organization of the RapidIO Trade Association to be similar to that of InfiniBand, in which membership is required to obtain patent rights and the actual management of the organization will be contracted out to an administrative group.
The technical description of the architecture provided by the association spells out a packet-switched protocol implemented on standard printed circuit boards. The first version requires 40 pins per port with each port having a data path 8 bits wide and fully duplex.
Physical signaling is via standard low voltage differential signaling (LVDS) operating at 250 MHz. Data are transferred with a source synchronous clock and sampled on both clock edges. Total bandwidth per port is 10 gigabits. The draft specification also envisions an advanced version twice as wide and twice as fast, or a total bandwidth of 40 gigabits per second per port.
Additional information about RapidIO is available on the trade association's web site, www.RapidIO.org. - J.R.
