By John Keller
Integrators of central processing units — the brains of sophisticated military and aerospace systems — are dividing into two camps: those who continue to specify traditional silicon chips, and those heading in the new direction of implementing microprocessor "cores" in programmable devices or in application-specific hardware.
Choosing the right microprocessor for military and aerospace applications involves navigating a dizzying array of options that can leave confused systems integrators wondering if they have to consider hardware, software, or something in-between.
The dilemma revolves around whether to use microprocessor chips, standard or custom peripherals and interfaces, a new breed of microprocessor "hard cores" and "soft cores," and a broad array of hardware that is extremely programmable, sort of programmable, or not very programmable at all.
It was not always like this. A decade or so ago, selecting microprocessors involved a lot of manufacturers and many different kinds of processor architectures, yet in the midst of these variables was at least one constant: nearly all microprocessors were packaged as individual silicon chips.
Then, as now, systems designers faced a raft of alternatives when shopping for the most appropriate microprocessors for their applications. Among the available chips were names such as MIPS R3000, AMD 2900, Intel 80960, Motorola 68000 and 88000, Sun SPARC, and even lingering interest in the 16-bit MIL-STD-1750A processor. Many of these were available in specially manufactured "militarized" versions that met military standards for ruggedness and electromagnetic characteristics.
One of the hottest issues surrounding microprocessors for defense applications — the so-called "RISC versus CISC" argument — split manufacturers and users into two separate camps. The issue revolved around whether to use what was the new crop of reduced-instruction-set computer (RISC) architectures, or stay with the established complex-instruction-set computer (CISC) approach.
CISC proponents favored a large and varied instruction set that attempted to improve performance by reducing the size of the software code that the processor needed to execute. RISC advocates, meanwhile, favored simple, single-cycle instructions that executed quickly, but might need to execute relatively large amounts of software code.
Today the issues surrounding choosing microprocessors for military and aerospace applications are different, and so are the groups of competing microprocessors. The RISC versus CISC debate also is far less relevant with today's complex microprocessor architectures and inexpensive memory. Also gone are most of the militarized microprocessor versions as commercial off-the-shelf (COTS) architectures have come to dominate this segment of the electronics industry.
Curiously, however, the community of microprocessor manufacturers and users today is still split into two camps — those who sell and integrate microprocessors as silicon chips, and those who provide microprocessor "soft cores" and implement these cores in field programmable gate arrays (FPGA) or application-specific integrated circuits (ASICs).
Microprocessor chips vs. cores
A microprocessor chip is straightforward and well known — a complex integrated circuit packaged in silicon that performs control functions and numeric processing. For many of today's military and aerospace applications, choices for traditional microprocessors narrow down to a few general families — the x86 Pentium series from Intel Corp. in Santa Clara, Calif., and the PowerPC microprocessor family from Motorola Inc. in Austin, Texas, and IBM Microelectronics in Hopewell Junction, N.Y.
Low-volume niche users who need powerful radiation-hardened microprocessors often choose the RAD6000 chip from BAE Systems in Manassas, Va., and those who need embedded workstations can choose the Sun SPARC implemented in single-board computers from companies such as Themis Computer in Fremont, Calif.
Microprocessor cores, however, are something different. Cores are pieces of hardware-description-language software, or "intellectual property" (IP) that contain functional, synthesizable hardware descriptions of microprocessor architectures.
Those who license these cores can implement the microprocessor architectures as FPGAs for low-volume applications, or as ASICs for large-volume applications. Microprocessor core descriptions often are contained in hardware description languages such as the VHSIC Hardware Description Language — better known as VHDL.
It would not make technological and economic sense to make every known microprocessor architecture available as a core, and not all microprocessor manufacturers make their architectures available as cores. Intel is a prime example of a company that does not make its most advanced processors available as cores because company officials believe their profitability lies in providing silicon chips.
Other companies, however, have found that divorcing themselves from manufacturing silicon microprocessor chips is in their best interests. Rather than selling chips, they sell licenses for microprocessor cores, and those who license the cores arrange for semiconductor fabs to implement the architectures in FPGAs or ASICs.
Microprocessor chips
Systems integrators choose microprocessor chips over cores when they need a lot of processing power, and when they need a wealth of off-the-shelf software and peripheral support. Sometimes using off-the-shelf processor chips such as PowerPC and Pentium can offer more peripheral support than the application actually needs, but the benefits of performance can outweigh the drawbacks of unnecessary features.
"For high-end processing, a high-end general-purpose chip is likely to have the things you need — and is likely to have some things you don't need — but on high-end applications you are usually willing to take that extra stuff," says Richard Jaenicke, director of product management at Mercury Computer Systems Inc. in Chelmsford, Mass. "A fully integrated everything-but-the-kitchen-sink kind of processor is usually acceptable for most military applications."
Mercury, which specializes in demanding multiprocessing computer architectures and computer subsystems, offers a Raceway-based multiprocessor architecture that is able to accommodate a wide variety of microprocessors, yet for the past couple of years has used the PowerPC Altivec exclusively in the company's military and aerospace work, Jaenicke says.
Of today's two prevalent microprocessor chip families, the PowerPC from Motorola and IBM dominates military and aerospace applications over its Intel Pentium rival. "The PowerPC for embedded applications is by far the most prevalent because of its familiarity and longevity," says Frank Willis, vice president of government business development for SBS Technologies in Albuquerque, N.M. SBS specializes in single-board computers, computer systems, and computer integration.
The PowerPC tends to be physically smaller and uses less power than its Pentium rival, which often makes PowerPC the default choice in embedded military and aerospace applications. The PowerPC also handles digital signal processing (DSP) better than Pentium, experts say, and military and aerospace systems integrators often find they can use PowerPC as an alternative to dedicated DSP chips.
Mercury's exclusive use of the PowerPC could change, however, because of the way the PowerPC exchanges data with other devices off-chip through what Jaenicke calls its "front-side bus." This data path is crucial for real-time multiprocessor applications because it is how the processor receives data; it does not matter how quickly the processor can crunch data if the front-side bus cannot keep the device adequately loaded.
The PowerPC's front-side bus runs at 133 MHz, while the most modern Intel Pentium processor chips run at nearly 1 GHz, Jaenicke says. "This is a big deal in image- and signal-processing applications, and the problem gets worse when the processors get faster. It might drive us toward the Intel chips."
null
Still, the relatively large size and power consumption of the Pentium microprocessor chips ultimately may cause integrators from Mercury and other companies to shy away from the Pentium. "Intel is not really in the embedded space for power consumption," Jaenicke says. "It is not something where we can stick four Pentiums on a 6U VME card. Where we might use Pentiums is where people don't have stringent power or weight issues."
With the microprocessor chip discussion revolving primarily around PowerPC and Pentium, the casual observer might believe that far fewer microprocessors are on the scene today than 10 or 15 years ago, and this would be correct where chips themselves are concerned. "Processors are more complex and it is much more expensive to develop them and enter the market," says Benoit Robert, executive director of product marketing at embedded computer provider Kontron America in San Diego.
Yet the total number of actual microprocessor architectures may be even larger than it was a decade or so ago because of the growing popularity of microprocessor cores. "We don't have as many microprocessors but we have more microprocessor cores," says Ray Alderman, executive director of the VMEbus International Trade Association (VITA) in Fountain Hills, Ariz.
Microprocessor cores
Two of the best-known microprocessor core providers are MIPS Technologies Inc. of Mountain View, Calif., and ARM Ltd. (formerly known as Advanced RISC Machines) in Cambridge, England. MIPS in the early 1990s manufactured microprocessor chips, but since then has switched over to licensing microprocessor cores exclusively. ARM started licensing microprocessor cores in 1991.
The MIPS and ARM cores allow licensees little freedom to customize the inner workings of the microprocessor architecture itself, but in exchange offer a broad variety of off-the-shelf software support. Another player in microprocessor cores is the Nios embedded "soft core" from Altera Corp. in San Jose, Calif. Nios users can customize the microprocessor architecture any way they like, but then must write custom software to run the device.
Perhaps the biggest advantage of microprocessor cores is the flexibility they bring to the systems designer. When implementing a microprocessor core as an FPGA or ASIC, designers can choose exactly the processor peripherals they need, and nothing more.
In addition, designers who use microprocessor cores are not stuck with the peripherals and interfacing that come with processor chips; they can customize to best advantage for their applications. "I don't want to pay for what I don't need, or have to go out and get what I need for the chip," says Bob Garrett, Altera's marketing manager for the Nios embedded processor.
"The beauty of the FPGA is a platform where you can choose the right peripheral set; it is a reconfigurable platform to adapt to your needs — especially in the military and aerospace market," says Yankin Tanurhan, senior director of applications and IP solutions at FPGA provider Actel Corp. in Mountain View, Calif.
"There is no standard microcontroller today, for example, that has a MIL-STD-1553 implementation," Tanurhan says. "To do that with something else, you could use an ASIC that would cost an arm and a leg, buy three or four different chips, or do the FPGA implementation. In the 20-year cycle of a military system, things change. FPGAs give you flexibility; you can update the processor or the peripherals."
It is a big plus when FPGA and ASIC vendors offer rugged devices for demanding environmental conditions. Actel, for example, offers devices that operate in the full-military temperature range of –55 to 125 degrees Celsius.
Designers who choose cores also are not beholden to any particular microprocessor fab. "The customer has control of the intellectual property," Alderman says. "They can do ASIC, or FPGA, and are not dependent on a fab to make them.
Chips or cores?
When choosing a microprocessor chip or microprocessor core, Willis at SBS says price, performance, thermal aspects, and environmental ruggedness are in the forefront of his company's decision. In addition, he says the processor's high-speed interconnects to peripherals and I/O, and how much data loading the processor can tolerate are big concerns.
SBS specializes in multiprocessor single-board computers, and Willis says that board designers can place only so many processors on a board before they reach the point of diminishing returns. "With 20 processors on a board, you should be able to come up with a general-purpose FPGA-based platform. You could put 50 processors into an FPGA if you wanted."
Engineers at SBS have a rule of thumb — 50 to 100 megabytes per second of bandwidth, and more than 100 operations on that data — when they consider substituting FPGA processors for processor chips. "Instead of brute-forcing with more PowerPCs, it might be a more elegant solution to go with FPGAs," Willis says.
Customizing with cores
One of the most useful algorithms for military signal and image processing is the fast Fourier transform, better known as the FFT. Designers who rely heavily on FFT processing can use FPGAs with embedded microprocessor cores to create custom FFT engines, Alderman explains. "You can write an FFT algorithm in C, spit out the VHDL code, and it will be a hardware model that will become an FPGA," he says. "You can trim it down only to the functions you need to do an FFT."
Alderman says it makes the best sense to implement microprocessor cores as FPGAs in low-volume applications, and in ASICs in high-volume applications. In fact, microprocessor cores can make it possible for designers to use both.
"In one scenario, a customer designs with a standard that has not settled out yet, but looks like it will be specified, explains Altera's Garrett. "He can design with an FPGA and get the design out to market quickly. If there are changes to the spec, they can make those changes in the FPGA. But if the goal is a high-volume product, such as tens of thousands a year, the customer can go to an ASIC — depending on the his price point."
At QuickLogic Inc. in Sunnyvale, Calif., experts combine FPGA and ASIC technology on the same device. "The MIPS core is on the ASIC side of the chip," says Ian Ferguson, vice president the QuickLogic QuickMIPS division. "The advantage is we can write turnkey software; everything is ready to go for the customer on the CPU side. Most customers don't want to play around with the CPU core itself."
The programmable side of the device allows systems designers to enhance security with unique interfacing between the processor and proprietary subsystems or peripherals, or by giving each processor its own unique identification. "As soon as you have everything on a single chip, you can do secret-source stuff because you reduce the number of off-chip accesses," Ferguson says. "Each device could have a unique identifier — a 64-bit unique ID so that the software will only work on that chip, so software might only run on an authenticated piece of hardware."
Component obsolescence
For military and aerospace systems designers, a particularly attractive benefit of microprocessor cores is their ability to help alleviate one of the most pressing and longtime problems involving military electronics designs — rapid component obsolescence.
Even before the COTS era began in 1994, military systems designers struggled with how to field military electronic systems with up-to-date components such as microprocessors. With military design cycles stretching out for as long as a decade between original plans and system fielding, combat aircraft, armored vehicles, surface ships, communications systems, and other important platforms inevitably contained years-old electronic components by the time they joined the active forces.
The Lockheed Martin F-22, for example, was designed in the early 1990s, and its integrated avionics suite originally called for the Intel 80960 microprocessor. Today the F-22 avionics is being redesigned, in part, because Intel has not manufactured or supported the 80960 in years.
The new COTS paradigm and the resulting demise of military-specific component manufacturing lines raised this problem to crisis proportions. The military's reliance on COTS means that military systems designers must use microprocessors that are aimed at the desktop computer market.
The upside of that is the military has access to the latest computer technology; the downside, however, is the average life cycle of desktop computer is 18 months or less before new, more powerful models hit the stores. That life cycle just gets shorter over time, and microprocessor chip manufacturer support of chips that are more than a generation or two old can be lackluster or nonexistent.
Military systems, however, must remain viable in the field for decades, not just a few years. In fact, the trends in life-cycle durations of military systems and personal computers are heading in opposite directions — the lifetimes of PCs are getting shorter, while the lifetimes of major military platforms are getting longer.
To keep technologically up to date, those who manage military ships, aircraft, and ground vehicles must redesign or overhaul electronic systems many times over the platform's lifetime. This often involves expensive and time-consuming system recertifications with each new redesign — especially for life- and mission-critical systems such as aircraft weapons management and navigation subsystems.
Until recently, systems designers had been forced to take these problems in stride, but the combination of microprocessor cores, FPGAs, and ASICs, is giving designers another option: rather than designing in and recertifying generation after generation of new microprocessor chips, they can simply upgrade FPGAs with up-to-date microprocessor cores.
"When you change a CPU, it is a big deal. A customer doing nuclear power, for example, will have many certifications, so they can't change the CPU in a snap," says Benoit Robert, executive director of product marketing at embedded computer provider Kontron Embedded Modules GmbH in Deggendorf, Germany.
In this way, designers can ensure that the upgraded microprocessor cores can run existing system software, and can avoid the need to recertify computer boards by making sure that new or upgraded FPGAs are pin-for-pin compatible with existing boards.
Not a panacea
VITA's Alderman cautions, however, that programmable hardware and microprocessor cores cannot always replace microprocessor chips. First, he says, today's FPGAs are not large and complex enough to contain the most advanced microprocessor architectures. Second, he says, FPGA and ASIC design-and-development tools have a long way to go before they are 100-percent reliable for life- and mission-critical applications.
"Implementing in FPGAs is quirky, and that's why they still want microprocessors in silicon," Alderman says. "It's a whole different set of design rules, and the expertise is not there, nor is the motivation. A VHDL model can be just as buggy as software." In addition, validating a VHDL model can be expensive and time-consuming . "Validation takes forever. There's a host of different variables, and that's what's messy."
The emergence of a standard FPGA architecture will also take time. The lack of FPGA standards also may be holding this design approach back in its competition with microprocessor chips such as the Power PC, says Doug Patterson, director of marketing, at Vista Controls Corp., a single-board computer and computer subsystems manufacturer in Santa Clarita, Calif. "There is an FPGA opportunity now, but there is still not standard FPGA architecture," he points out.
FPGA technology is somewhat new to military and aerospace applications, and prospective users may need assurance that these devices will be supported in the long term, says Colin McCracken, director of market development at Ampro Computers Inc. in San Jose, Calif.
Experts in the programmable hardware industry are putting continuous effort into perfecting development and validation tools, yet Alderman calls today's tools "Neanderthal."
"The tools will get better every year," Alderman says. "VHDL compilers get better every year, and eventually they will be automated, but it takes time for that evolution. It will take another five to seven years before those tools are not quirky anymore."
