San Jose -- Michael A. Johnson aims to pick the microprocessor that will bring Americans back to the moon. It's a challenging job, and long overdue.
NASA has been lumbering along for five years on the IBM PowerPC 750, a 200-Mips CPU that was discontinued in Apple Macintosh computers back in 1998. But defining a world-class processor that can tolerate the hazards of space--and doing it on a slender government R&D budget for an ambitious lunar mission schedule--involves much more than a stroll down to Best Buy.
As principal investigator of the High Performance Processor Project at Goddard Space Flight Center (Greenbelt, Md.), Johnson will lead the evaluation of the key electronics for the Crew Exploration Vehicle Block 2, aka Orion--the spacecraft that will pick up where the Apollo missions of the 1960s and '70s left off. (NASA awarded Lockheed Martin Corp. a multibillion-dollar contract in August to build the new moonship.) The chips that will drive Orion would also be used in next-generation moon rovers.
Those space and lunar vehicles are still in a design stage deep inside NASA as part of the agency's overarching Constellation program. Johnson does not expect to get all the requirements from NASA systems developers for as much as a year, though some specs will be delivered this fall.
Nevertheless, he has already spent the better part of a year surveying the semiconductor industry's capabilities and laying out some basic guidelines for the next space CPU. The first-pass microprocessor survey should be complete in a few months.
Johnson's group targets a CPU with as much as 3,000 Mips of performance and 2,000 Mips/watt in energy efficiency. It must tolerate a total-life-cycle radiation dose of 100 kilorads (assuming 100-mil alumi- num shielding ) and be able to survive 10–12 soft errors per bit per day. It must also withstand 100 mega-electron-volts/cm2 per milligram, making it effectively immune to any "latchups," or hard errors, that can be caused when a high-energy particle fries a chip with junction back-bias and excessive current draw.
The chip should be available for use in prototype systems for an unmanned mission to the moon currently scheduled for 2008. It must be production-ready for manned lunar-exploration systems that will start construction in about 2011.
A manned mission to the moon is expected to depart no later than 2020.
Johnson's project is just one piece of NASA's Constellation program, kicked off by President George W. Bush's 2004 challenge to NASA to send explorers back to the moon and then to Mars.
"The talk about going back to put footprints on the moon and then going to Mars has reinvigorated a lot of people at NASA," Johnson said. "This is something people say will be hard or maybe can't even be done--and that gives engineers the drive they need."
While the president has supplied motivation, the government has not been as forthcoming with the means. R&D budgets at NASA were slashed in half in 2004-05, and some projects were canceled to compensate for the ballooning cost estimates for building a new low-Earth-orbit vehicle and accelerating the schedule for delivering it. Cost overruns for design and construction of the International Space Station have also drained NASA's coffers for space exploration.
"There were some projects that were relevant to our exploration mission that could not be continued," Johnson said.
The schedule contraction stems from government concerns about a potential four-year gap during which NASA will have no means of launching humans into space. The space shuttle is due to be retired in 2010, when the International Space Station is completed, but a replacement low-Earth-orbit vehicle may not be ready for manned flight until 2014.
In an effort to conserve resources, Johnson has held preliminary meetings with Space Command officials at the Department of Defense. He hopes to collaborate on one choice of microprocessor that could serve the needs of both NASA and military space programs. A similar collaboration resulted in the choice of the radiation-hardened PowerPC 750 about five years ago.
Johnson said his team has been "tracking work the DOD has been funding through Darpa," or the Defense Advanced Research Projects Agency. "I don't want to follow DOD, and they don't want to follow NASA, so the best approach is to find a common path," he said.
One impediment to collaboration is the military's desire to keep some of its needs under wraps. But Johnson said the DOD "can generalize and compartmentalize some of their requirements so security is not a major issue." Meanwhile, all sides are waiting for President Bush's 2008 budget submission, which is due next February.
Budget constraints probably will force the government to pick one microprocessor architecture, not two or more. That's difficult, because some applications would be best served with a general-purpose processor, while others--such as robotic vehicles--would be better served by a single- or multiple-instruction, multiple-data architecture. "Ideally we can find a processor that can do both effectively," Johnson said.
The IBM Cell has already garnered attention in the NASA survey as "a possibility," he said. NASA plans to hold more conversations about the architecture with IBM.
"There are some really attractive technologies available off the shelf, but many of them just aren't robust enough to meet our constraints," said Johnson.
It's not clear whether the government will opt for a standalone processor or will define a system-on-chip that could serve a broad range of platforms. "Given our constraints, we can't roll a new SoC every time there is a new application, unless SoC technology becomes more cost- and time-effective," Johnson said.
Low volumes further complicate the picture. To date, NASA has used only several hundred rad-hard PowerPC 750s, although the device has been the agency's workhorse CPU for about five years.
The high-end requirements and low volumes result in astronomical prices when compared with PC processors. Whereas desktop CPUs typically sell for less than $200, "I would not be surprised to find some high-performance rad-hard processors that sell for $20,000 each," said Anthony Jordan, director of standard products at Aeroflex Inc. (Plainview, N.Y.), a government contractor that sells a variety of radiation-hardened chips.
Even relatively pedestrian 8-bit rad-hard controllers from Aeroflex can cost as much as $2,000 each, Jordan said. "They can be used as the heart of a satellite and are expected to function flawlessly for its 20-year life span, in an environment where repair is not an option."
Other approaches are under investigation. NASA has experimented with using three partially hardened CPUs in a redundant architecture in which the three chips "vote" on results. Systems require the same results from two of three CPUs. The concept has been tried in some systems using PowerPCs running at 1 GHz, but for some applications the approach requires too much power and board space to be viable.
A project called Space Technology 8, in the works at Honeywell, is exploring whether space systems can use chips that are partially hardened and then en- cased in a sort of wrapper that protects them from the single-event immunity issues that cause soft errors. In this approach, the electronics are constantly monitored for faults and can be remotely reset as needed.
Two other research programs take a similar approach using Xilinx Virtex-4 FPGAs with embedded PowerPC cores and interconnect fabrics. They are the Space Cube project at Goddard and the JPL Computer Platform project at the Jet Propulsion Laboratory.
"You can also do radiation hardening by having hardware redundancy and voting on results at the circuit level. All these approaches are being studied, and some of them are being deployed and flown," said Aeroflex's Jordan.
Aeroflex is developing a rad-hard cell library for a standard 130-nanometer CMOS process. It expects to tape out its first chips by the end of the year, although qualification for stringent military specifications can take a year or more.
By contrast, contractors such as BAE Systems and Honeywell have a history of developing unique process technology nodes for rad-hard devices.
Two other NASA project managers work in parallel with Johnson on developing space electronics. One is exploring low-temperature devices that can operate at –180°C to withstand such environments as the dark side of the moon. The other is studying reconfigurable electronics that can be programmed for changing missions or to work around faults that crop up during a long mission.