Defense Department and industry officials struggle to retain qualified suppliers of radiation-hardened and radiation-tolerant components for orbital and deep-space systems
By John Rhea
U.S. Department of Defense (DOD) leaders are struggling to find new ways to safeguard the dwindling supplier base of radiation-hardened microelectronics that are necessary to meet future spacecraft requirements. Officials are getting nervous about space systems support as previously reliable rad-hard device manufacturers defect to lucrative commercial markets.
Still, the booming commercial satellite market may take up some of the slack, a prospect which leads some observers to predict a doubling of today`s estimated $200 million annual spacecraft component market over the next five years.
The situation is getting top-level attention at the Pentagon. An in-house study by a rad-hard integrated product team delivered Dec. 20 to Paul Kaminski, undersecretary of defense for acquisition and technology, concluded that DOD officials must invest $60 million a year to push the technology. The current level is about $30 million, mostly through the services` system program offices and the Defense Special Weapons Agency (formerly the Defense Nuclear Agency).
"Without radiation-hardened circuits, a single low-yield nuclear detonation in lower space could rapidly degrade the performance or cause catastrophic failure of many critical satellite systems, directly impacting command and control and battlefield performance," the report states. "However, beginning in 1993 our domestic capability to develop and manufacture radiation-hard components appeared in possible jeopardy because of the decline in the DOD budgets for radiation-hardened microelectronics enabling technology and the decline in the domestic vendors capable of producing DOD-unique hardened microelectronics."
Declining support from the system program offices, which traditionally underwrote development costs and represented potentially large markets for production quantities of the new devices, is compounding the military availability problem. The former Strategic Defense Initiative Organization (SDIO), for example, was the driving force behind the RH-32 rad-hard 32-bit computer program, which leaders of the U.S. Air Force Rome Laboratory in Rome, N.Y., initiated nearly a decade ago.
But that momentum has been lost along with the perceived end of the Cold War and the apparent easing of national security threats from strategic missiles, and as the SDIO evolved into today`s Ballistic Missile Defense Organization.
The RH-32 situation is illustrative of these parallel trends of declining production facilities and the scramble to find new sponsors. Once planned for a wide spectrum of airborne as well as space applications - including the Air Force`s airborne warning and control system surveillance aircraft and F-22 advanced tactical fighter - the single-board all-purpose computer now depends on a single foundry, and user applications are still in early development.
The program is due to reach fruition this year as the two surviving development contractors, Honeywell Space Systems Division in Clearwater, Fla., and TRW Inc. in Redondo Beach, Calif., complete final acceptance testing of their advanced development models. Honeywell designers are due to accomplish this in June, but TRW experts have been delayed until September because their previous source of die, UTMC Microelectronic Systems in Colorado Springs, Colo., is no longer available for that purpose. Instead, TRW engineers have had to use Honeywell`s foundry at Plymouth, Minn.
Gene Blackburn, chief of the Electronics Reliability Division at Rome Lab, calls UTMC`s withdrawal to service Rockwell International`s consumer product needs a "shakeout," but sees "no show stoppers" in the program.
Regarding new sponsors, Blackburn points first to the Space Based Infrared System (SBIRS), which he says has been one of the "most faithful funders" and is likely to be the first volume user. And, although he regards using rad-hard computers in aircraft as "overkill," he does see some possibilities in the "high flyers": the unmanned aerial vehicles operating above 50,000 feet, where single event upsets (SEUs) could be troublesome. Rome Lab experts have been working with the Naval Research Laboratory (NRL) in Washington to quantify the SEU problem, he says.
DOD and National Aeronautics and Space Administration (NASA) officials once assured availability of space-qualified parts by establishing dedicated or captive production lines at willing vendors, and a few captive lines still produce specialized spacecraft components in low volumes. The captive lines, however, have become a thing of the past with the advent of streamlined qualification processes. One industry source describes them as "the path of least resistance ... to do what nobody else will do."
Among the government agencies with their own wafer fabs are:
- Sandia National Laboratories in Albuquerque, N.M., which has been working on devices for the Global Positioning System (GPS);
- a high-end CMOS line operated for the National Security Agency (NSA) at Fort Meade, Md., by National Semiconductor Corp. of Santa Clara., Calif., which is only now beginning to go to 0.5 micron feature size; and
- the Naval Command, Control, and Ocean Surveillance Center RDT&E Division, the former Naval Ocean Systems Center in San Diego, which is working on charge-coupled-device sensors and silicon-on-sapphire (SOS) and silicon-on-insulation (SOI) devices, also at 0.5 micron.
These in-house operations are typically two years behind the technology of their commercial counterparts, experts believe.
Today there are only four principal foundries making rad-hard die, according to the integrated product team report: two analog and two digital. The analog foundries are at Analog Devices of Norwood, Mass., and Harris Semiconductor in Melbourne, Fla. (which also has an extensive digital logic product line). On the digital side, in addition to Honeywell`s foundry in Minnesota, is Lockheed Martin Federal Systems in Manassas, Va.
Lockheed Martin engineers supply digital rad-hard die to UTMC and to other device producers as well as to their own RAD 6000 32-bit spacecraft computers based on the IBM RISC System/6000 microprocessor architecture. Other rad-hard suppliers include National Semiconductor, which has a foundry in South Portland, Maine, and Texas Instruments in Dallas. Hughes Space & Communications Co. in Los Angeles also has an internal SOS capability and is expected to reenter the rad-hard external market.
In fact, Lockheed Martin`s RAD 6000 became the first 32-bit rad-hard computer in space with NASA`s launch Dec. 4 of the Pathfinder mission to Mars. This is a 35 MIPS computer running at 35 MHz using entirely commercial off-the-shelf (COTS) technology, including the associated high-density hermetically sealed memory.
John Bendekovic, business development manager for space and advanced technology programs at Lockheed Martin-Manassas, is looking for new applications of the RAD 6000. The company teams with the former Rockwell space operations of the Boeing Co. in Seattle, for the GPS 2-F program, which is scheduled for flight around 2001.
Lockheed Martin has two RAD 6000-based product lines. The first product line has a 64-bit VMEbus backplane, such as the one used on Pathfinder and planned for GPS. The other is tailored to the Advanced Technology Insertion Module program of the Air Force Phillips Lab in Albuquerque, N.M., and uses the Advanced Spaceborne Computer Module (ASCM) backplane interface.
For now, Lockheed Martin designers are using 5-inch wafers of bulk CMOS at feature sizes of 0.5 microns, yet they have development efforts in progress on SOI and 0.35 micron resolution. Company officials are also talking with IBM about licensing the IBM PowerPC 64-bit architecture for a rad-hard processor to be fabricated at Manassas, Bendekovic says.
This progression to 64-bit rad-hard microprocessor architectures is logical since the military space market tends to trail the commercial market by one or two generations - analog as well as digital - says Lucien Debacker, head of the Johari Associates consulting firm in Colorado Springs, Colo. He also looks for gigabit random access memory chips in space as another logical evolution.
Anthony Jordan, standard product marketing manager at UTMC, contends that it isn`t necessary - or even desirable - for a company to have its own fab to participate in the space market. Jordan`s position is not surprising, considering that UTMC recently gave up its own rad-hard fab. When UTMC leaders sold their fab to Rockwell in 1995 they accumulated about five years worth of inventory to carry them through the switch to becoming a fabless supplier. The decision to become a fabless supplier has enhanced UTMC`s profitability, Jordan says.
"The Intels and Motorolas and TIs are looking at billion-dollar markets," Jordan notes. By contrast, the space market is on the order of tens of thousands of components a year; it makes more sense to amortize the wafer fab costs over the consumer product lines, he explains. Furthermore, tying themselves to the high front-end costs of a fab line, small companies can jump in with innovative products - even using offshore fab facilities, he says.
After all, COTS parts are becoming acceptable for space applications, Jordan says - if COTS means procuring via catalog part number to the QML-V standards for space usage. The QML, short for Qualified Manufacturing List, encompasses the MIL-PRF 38535 and MIL-PRF 38534 military quality standards for semiconductors, and represents the DOD`s first tangible attempt to embrace widely accepted industry standards based on statistical process control rather than exhaustively testing each part that comes off the line.
Jordan expresses doubts about plastic packaging for space, however, noting that packaging represents only 1 to 2 percent of total costs and the radiation effects are not adequately understood.
The streamlining of standards - particularly switching the QML to performance-based standards - has reduced paperwork and sped the availability of products, says Dave Allen, manager of space-level marketing at National Semiconductor. "QML is a marvelous step forward that benefits users and manufacturers and allows flexibility," he says. Allen, who has been with National since 1973, recalls that during the Class A days of the 1960s the users were responsible for most of the documentation burden. That burden had increasingly shifted to the semiconductor firms, first with the S-level qualification process of the 1970s and then with the JAN-S requirements, until the QML revolutionized the process.
QML minimizes government involvement, he says, and lets the OEMs and vendors capitalize on the latest technology. He recommends that they establish partnerships early in development. "Customers should not have an inferiority complex," Allen comments.
The president of one of the fabless suppliers, Space Electronics Inc. (SEI) in San Diego, agrees. "The lead time for satellites from design to orbit is two years," says SEI President Dave Strobel, and that means delivery time on parts has to be cut from 24 to 12 weeks.
One way SEI designers meet their cost and delivery imperatives is using their trademark Rad-Pak shielding process, which takes a variety of devices in dual in-line, and flatpack packages and hardens them to meet rad-tolerant or rad-hard requirements. This enables the use of plastic packages, for example, in Motorola`s Iridium system of communications satellites. However, it does involve some weight penalty.
This approach is adaptable to the entire spectrum of rad-hard parts such as A-D converters, field programmable gate arrays, interfaces, and all types of memories. It is not necessary for all parts to be at the leading edge of technology for space purposes, SEI officials say. Space Electronics engineers get their digital signal processors (DSPs) from Texas Instruments and their microprocessors from Intel; the most powerful microprocessors on SEI`s current catalog are the 386DX and 486DX 32-bit devices. The company even offers a rad-hard version of the 8-bit Intel 8086.
Strobel offers that, on a comparable function basis, each step up from commercial grade to rad-tolerant to rad-hard represents an increase in price of an order of magnitude, i.e., tens of dollars for commercial parts, hundreds for rad-tolerant, and thousands for rad-hard.
At Harris Semiconductor, where officials have leveraged their commercial capability to penetrate the military market, Steve Strickler, manager of military and space marketing, looks for the largest future growth in the civil satellite sector. He puts this at eventually 85 percent of the market, citing such current programs as Motorola`s Iridium, TRW`s Odyssey, and a host of future European and Asian communications satellite programs. Among the leading domestic military programs Strickler lists is SBIRS, GPS 2-F, and the upgrade of the Minuteman 3 nuclear ballistic missile.
While this growing international market raises questions about technology transfer, many of these overseas applications don`t need - and probably can`t afford - the state-of-the-art products planned for U.S. military systems, says Scott Moody, vice president for military and space products at Harris Semiconductor. The designs, for the most part, remain in American hands, and wafers can be fabricated almost anywhere.
For example, there are plenty of applications in low earth orbit for single-chip 1750A 16-bit processors, says Larry Wing, director of electronic systems at Honeywell Space systems. Although Honeywell Space is on the Rome Lab RH-32 and the Phillips Lab ASCM programs, company officials have qualified their 1750A Generic VHSIC (Very High Speed Integrated Circuit) Spaceborne Computer (GVSC) central processor using early 1.25 micron VHSIC processes. The Honeywell RH-32 is implemented in 0.8 micron VHSIC CMOS.
Honeywell officials will begin delivering their four-chip RH-32 multichip modules next year for qualification on the SBIRS program, scheduled for launch sometime after 1999. The software development tools include Ada and C compilers. SBIRS will use a MIL-STD 1553 serial bus with a pre-planned product improvement to 1773, but Wing looks for a variety of buses for the future, including the IEEE standard back panel, Futurebus, VME, and others in a SEM-E standard form factor.
What`s beyond today`s 32-bit rad-hard offerings? A 64-bit version of the Power PC or Pentium is a possibility, but the issue becomes one of need. Honeywell officials are claiming a performance of only "greater than 20 MIPS" for their RH-32, and Wing says that will do the job - with the necessary glue logic. "There are engineering MIPS and marketing MIPS," he quips, and the engineering MIPS can usually get the job done. He also maintains that, with COTS DSPs, there is enough scalar capability now to meet existing requirements.
One company that is cashing in on the rad-hard space market with a line of specialized memory products is Irvine Sensors Corp. of Costa Mesa, Calif. Keith Gann, director of high-density electronics, reports the company will begin sampling a five-high stack of 128K-by-40-bit SRAMs in July for possible use on SBIRS built with Honeywell SOI dice to rad-hard standards. Production will be on a build-to-order basis, and will be expensive. "These are not jelly beans," Gann notes.
Electronic Designs Inc. (EDI) in Westborough, Mass., introduced a 32 megabit Nand flash memory in February configured as 4M by 8 bits. Mark Hampson, EDI product marketing manager, says he sees a natural synergy for the product in aircraft and spacecraft. The memory has completed radiation testing at NASA`s Goddard Space Flight Center in Greenbelt, Md., and EDI officials expect to begin sampling this month.
The principal market will be military aircraft retrofits, including avionics, digital maps, and crash recorders, which will require as much as 4 gigabytes of memory, Hampson says. Yet similar, albeit smaller, requirements exist in low-earth-orbit communications satellites and in the NASA space shuttle. The price is $175 in quantities of a thousand. "Seven years ago we`d have had to sell it for $2,000 for all the paperwork and testing," he notes. Fabrication is by Samsung in Korea.
Designers from UTMC Microelectronic Systems produce radiation-hardened components without maintaining a chip fab by obtaining rad-hard die from TRW, Lockheed Martin, and other companies.
The UTMC RAD-PAL, which company officials say is industry`s first radiation-hardened programmable array logic compatible with industry standard 22VP10 architecture, is also QML Q and V compliant, and is intended for space applications such as system counters, decoders, and state machines.
The monolithic surface-mount 32-megabit NAND flash from Electronic Designs Inc. is for harsh-environment solid state storage applications such as cockpit data and voice recorders, satellite recorders, and digital maps.
The biggest challenge: 20 megarads
When it comes to the ultimate in severe radiation environments, not even experts at the Defense Department can match what the deep space probes of the NASA Jet Propulsion Laboratory (JPL) encounter. One spacecraft from JPL in Pasadena, Calif., that is still awaiting NASA approval will face levels as high as 20 megarads on a mission to Jupiter`s moon Europa early in the next century.
In preparation for the new deep-space missions, JPL designers are developing what they call their next-generation scalable spacecraft, the X-2000, which will rely on state-of-the-art onboard processors, shielding, and associated electronics.
Richard Kemski, mission assurance manager for X-2000, says no other mission to date will be as demanding as the Europa effort. By contrast, the earlier Galileo mission to Jupiter had to withstand only 175 kilorads. Europa is in a particularly hot zone.
Under the JPL approach, the primary radiation hardening is via a spacecraft shield. That`s good for 1 to 2 megarads, Kemski says, and an additional 100 mils of aluminum provide another 4 megarads of hardening. The RAD 6K 32-bit RISC processors are good for 2 megarads, and field programmable gate arrays are usually good for about 1 megarad, he adds.
But that still leaves linear devices and controllers, which need to be shielded by a factor of 40, by Kemski`s estimate. DRAMs and A-D converters are characterized to fail at 10 kilorads and spot shielded, with the "softer" parts put toward the center of the spacecraft and the "harder" parts outboard.
Kemski estimates he needs the 35 MIPS of the RAD 6K processors, but points out they are not cheap. They are running at least $100,000 with the necessary peripheral components, but that is not necessarily a problem. Over JPL`s history, electronics has run about seven to eight percent of the spacecraft development costs for labor and materials.
The bigger concerns are weight and long-term operation. Kemski estimates that radiation shielding will require about 10 percent of the spacecraft`s weight, projected at about 250 kilograms. One solution being investigated in cooperation with Space Electronics Inc. of San Diego, is the development of a new moldable polymer to encapsulate radiation-tolerant parts in multi-chip modules. This would be an improvement over the present machinable shields of tungsten and tantalum, he adds.
As in all space missions - and particularly deep-space probes - the tradeoff is between the weight of onboard processing and the performance loss if the raw data have to be processed on Earth. What Kemski says he really wants for the next-generation spacecraft is a 64-bit processor capable of gigaflops of throughput and fabricated to the extreme conditions the spacecraft will encounter. But for the meantime he`s reconciled to shielding. - J.R.
The overseas challenge
Development of the newest state-of-the-art spacecraft computers is no longer confined to the United States - nor, perhaps in a few years, to silicon.
Two French companies are challenging the U.S. lead: Thomson-CSF and Matra MHS. The French have tended to stress silicon on insulation (SOI) over silicon on sapphire (SOS) because there is more commercial potential for SOI devices and therefore greater economies of scale. SOI also offers better resolutions, currently 0.35 to 0.25 micron with 0.18 micron in the works.
Thomson designers are using SOI in their rad-hard components, including a 32-bit microprocessor built to 1 megarad total dose and a line of SPARC processors and memories aimed at pin-for-pin replacements of the Harris Semiconductor SPARC line.
Matra, with sponsorship from the European Space Agency, has developed a radiation-tolerant SPARC processor with 0.8 micron CMOS technology as will as 1-megabyte SRAMs and a 32-bit RISC microprocessor. Patrice Hamard, avionics, military, and space marketing manager for North America at Matra`s Temic semiconductor facility in Santa Clara, Calif., reports that his company has been working with Space Electronics Inc. of San Diego to qualify a 32-bit digital signal processor to the 100 Krad radiation-tolerant standard. The device was released in December and now is available for Beta testing, Hamard says.
The driving force in the international market is the post-Cold War shift toward commercial satellites, notably communications, meteorological, and resource survey models, according to a survey paper prepared by Lewis Field, a researcher at the National Ground Intelligence Center in Charlottesville, Va.
"Due to high launch costs, these space programs require rad-hard/-tolerant electronics to be reasonably cost-effective," Field concludes. "Some nations have increased indigenous production of rad-hard electronics as a result of bans from the previous producers or because they desired to conduct the space program as a purely indigenous effort."
One of these countries is Japan, which Field notes had not been involved in military devices but which is getting into SOI to meet commercial space needs. Examples include a 16-megabyte DRAM in SOI from Mitsubishi, early work by NEC of a 1-gigabyte SOI DRAM, and development of a 32-bit rad-hard processor by Hitachi under sponsorship of the Japanese National Space Development Agency.
Looking beyond silicon, the report cites work in rad-hard gallium arsenide, although it lags conventional silicon and SOI in complexity, and ferroelectric materials, which have been used in nonvolatile memories and which have inherent radiation tolerance.
Also of interest are two technologies aimed primarily at high-temperature applications, silicon carbide and diamond circuits. These also have inherent radiation-tolerant properties, but are expected to find more immediate applications in aircraft, where the heating problem is more severe. - J.R.