Why are spacecraft computers obsolete at launch?

Question

One might think that spacecraft would be on the cutting edge of technology. However, when looking over details of spacecraft, it seems their computer systems are often very much behind the times. For example, the Curiosity rover was launched in 2011, when consumer laptop systems ran in the GHz and had GB's of memory. Curiosity's CPU runs at 132 MHz and the system only has 256 MB of RAM (source: http://en.wikipedia.org/wiki/Comparison_of_embedded_computer_systems_on_board_the_Mars_rovers ). I realize there may be some lag from obtaining the parts for the spacecraft before it is built and ultimately launched, but this seems extreme. Why don't spacecraft have more up-to-date data systems?

Answer 1

There are a number of reasons why spacecraft electronics typically lag what is commercially available by several years.

Radiation tolerance

Electronics are very susceptible to radiation phenomenon that terrestrial electronics are largely protected from by the Earth's atmosphere and magnetic field. Common radiation-based failure mechanisms are Single-Event Event/Upset (SEE/SEU) — most commonly thought about as a flipped bit, latch-up — where a bit gets stuck in a certain state and the part needs to be powered down to be reset, burn-out — where a high energy particle (e.g. proton or neutron) destroys the part, and total dose — where long-term exposure (rather than a freak event) degrades the part. As chips and circuits advance and pack transistors more tightly, the probability of these events increases.

Several techniques and testing methods exist to demonstrate if electronic assemblies are robust in space radiation environments, but these tests are expensive. So once they've been done for a part, component, or assembly, the trade is often made to live with less performance and save the cost of re-testing and avoid the risk of complete mission failure.

Reliability

It is easier to do maintenance on a terrestrial computer, and the costs of failure are often much lower than for spacecraft. Ground systems also don't have the same tight power, size, and mass budgets that space systems do, which limit the amount of redundancy that is feasible. A solution is to continue to use parts that have been shown to have high reliability. Another way to increase reliability is to perform parts screening and to perform lots of electronics testing (e.g. bake-out to find infant mortality, random vibration testing to mimic launch environments, shock testing to mimic pyrotechnic events like fairing jettison, and thermal vacuum testing to mimic space. This testing takes time and is expensive. The time delay alone puts most space systems at least one Moore's law cycle behind the latest consumer parts.

Build time for satellites

To say nothing of the avionics, satellites take a long time to build. Even when the computers are done the rest of the vehicle has to be assembled and tested. For large spacecraft this can take years. Meanwhile, the computer isn't getting any younger and an aversion (often justified) to risk means upgrading it would require many of these tests to be re-done.

Power consumption

Over time Moore's law helps chips to increase in processing power and decrease in power consumption, but generally speaking, when comparing contemporaneous parts more powerful chips consume more power. Spacecraft are almost universally power starved, so there's little incentive to use a more power-hungry chip than is absolutely necessary. Everything in a spacecraft is a trade-off: a Watt of power used for the main flight computer carrying around unused cycles is a Watt that can't be used for RF communications, or providing power to a payload (when that payload isn't communications), etc.

Paperwork

Paperwork and process can be as dominate as any of the other reasons. The aerospace industry as a historically high barrier to entry. Once reason is the human capital required to build and launch spacecraft, but equally as important is the space heritage of the software and components that go into them. Space environments are more challenging that terrestrial ones in a variety of ways and often require unique solutions (for avionics, rejection of heat without convective cooling is a good example). Launch environments were discussed above. Qualification of components is a real-world hardware-centric task, but there is a paper trail that backs up this analysis and provides confidence to a spacecraft builder's customers and the launch provider that the vehicle will be safe during ascent and that it will operate in space. This is proven through a combination of test and analysis and demonstration, but most of the people who care don't witness or oversee these activities directly, so they rely on excellent paperwork to provide that confidence. Once you've gone through the trouble of getting buy-in on widget X — the effort associated with a obtaining Δ buy-in for widget Y or even X+ is harder to justify if the older part still works. Aerospace suppliers (prime contractors and all the way down the supply chain) are often also required to have rigorous quality processes in place — i.e. more paperwork and process. All of this slows down the pace of innovation and change for in exchange for predictability.

Launch delays

Then once the spacecraft is ready it needs to be launched, and launches can slip months if not years.

Answer 2

A big part of it is reliability. NASA could probably put in an Intel Xeon chip made in 2012 that has a crazy high amount of processing power.

However, the chip that was used, the RAD750, has years of experiments and usage behind it, such as being used in a variety of spacecraft including:

• Deep Impact comet chasing spacecraft, launched in January 2005 - first to use the RAD750 computer.
• XSS 11, small experimental satellite, launched April 11, 2005
• Mars Reconnaissance Orbiter, launched August 12, 2005.
• WorldView-1 satellite, launched Sept 18, 2007 - has two RAD750s.
• Fermi Gamma-ray Space Telescope, formerly GLAST, launched June 11, 2008
• Kepler space telescope, launched in March 2009.
• Lunar Reconnaissance Orbiter, launched on 18 June, 2009
• Wide-field Infrared Survey Explorer (WISE) launched 14 December, 2009
• Solar Dynamics Observatory, launched Feb 11, 2010
• Juno spacecraft, launched August 5, 2011
• Curiosity rover, launched November 26, 2011

Because of use since '05, NASA can be fairly confident that the chip won't fail due to radiation problems, etc.

Why? Well, I would say that John Besin's answer summed it up pretty well, and I won't try to top it:

I wouldn't think this would be the case at all. If anything, NASA would want to use hardware (and software) that has been extensively tested throughout years of use, both in NASA and in industry as a whole. The last thing NASA wants is to find a bug in a spacecraft's system at an inopportune moment, and when you’re talking about devices that need to travel potentially hundreds of thousands of miles through space, there are many inopportune moments.

Answer 3

One might think that spacecraft would be on the cutting edge of technology.

I wouldn't think this would be the case at all. If anything, NASA would want to use hardware (and software) that has been extensively tested throughout years of use, both in NASA and in industry as a whole. The last thing NASA wants is to find a bug in a spacecraft's system at an inopportune moment, and when you’re talking about devices that need to travel potentially hundreds of thousands of miles through space, there are many inopportune moments.

You might also find this question on Programmers.SE interesting; it addresses the programming languages, hardware, etc. used to construct the Mars Curiousity rover.

Also, I imagine that the lower-spec hardware NASA uses has lower power requirements than cutting-edge, high-powered hardware. For example, if the rover doesn't need a faster processor to operate, why waste space and weight by powering such a processor when a lower-spec one will suffice?

Answer 4

Another big reason is there simply isn't a need to do anything more powerful. There exist many applications on Earth where reliability is more important than speed. For instance, a vending machine contains a simple computer. You don't want that to crash and take your money.

The vast majority of the processing used today by computers is in the graphical interface. As there is no satellite that runs a graphical interface, it doesn't really make that much of a difference.

The point of a satellite's computer is to keep the satellite alive, pointed in the right direction, manage power, and collect data for use on the ground. Thus, they don't need to have gigahertz processors, they just need to be a data pipe. They need to do this with a high level of precision. You can't go and push the power button on a spacecraft, you need its systems to work flawlessly at all times.

Computers are regularly used on the ISS by the astronauts, but these are done for non-critical systems. It's only when the computer has to do significant processing of the data that the speed matters, and except for some compression, most of that is still done on Earth. Furthermore, most of the image laden systems out there have custom on board chips that assist with processing the images quicker, allowing less of the work to be done on the main processor.

Answer 5

There is an anime called "Rocket Girls" in which the protagonist asked this same question. The answer she got was that they only use Classic Technology; technology that has built up a reputation of success over time. This is true for medicine and general aviation as well. In fact this is true for most branches of engineering, its primarily software engineering that keeps using the "latest" stuff.

Also, CMOS is more susceptible to Radiation than TTL, so when you are doing Radiation Hardening it may be better to have a slow 100 mhz TTL based chip than a fast 3.4 Ghz CMOS based chip.

Answer 6

A couple of things I might add to the good answers already here:

• Selection timeframe. The decision as which hardware to use for a vehicle is made long before (years?) the vehicle is launched. Thus, at launch it is probably obsolete.
• Radiation hardening. Often, these comparisons focus on one or two specs that are interesting for terrestrial uses: CPU clock speed and RAM for example. While these are important, fault tolerance in a radiated environment is more important while flying by Jupiter than it is when playing Doom. This tolerance creates a trade-off that doesn't help the other specs.

Answer 7

• Selection Time: spacecraft are designed and built years prior to launch. The selected processor at build time, even if top of the line, will have been eclipsed by launch time.
• Vibration Tolerance: launch of spacecraft requires vibration tolerant computer systems; many newer processors are not yet rated at design time.
• Radiation Resistance: Smaller circuits are more subject to radiation induced errors than larger circuits. most more advanced processors use smaller circuitry to reduce energy costs, thermal loads, and operation cycle times.
• Price: Older processors can be bought for far less than current leading edge processors; prices drop noticeably once patents expire.
• lack of need: Not all satellites need highly robust processing solutions.
  The entire Apollo mission was run with a processing power equivalent to a couple of high end linux workstations... this includes the mainframes at JSC and Cape Kenedy. The on-board computer on Apollo was about as powerful as many digital watches. (80kB total memory; that's 37kB words of 2B each for ROM, plus 2K words of RAM.) It ran at 1 MHz, rather fast for its day. I've bought $20 calculators with better specifications than the AGC.
  The tasks of most satellites can be reliably run with older processors without mission compromise.

Answer 8

Same thing happens in aviation as you've indentified for space technology. Major factors would be reliability, "hardness" and development time frames, but there are other considerations.

Any life-critical system has to be trustworthy, and when you can't get at it to fix it if it breaks (like robotic space probes), reliability becomes paramount also. The longer a thing has been around and accumulated experience, the more it can be trusted. Also, the more complex a system is, the more difficult it can be to verify that all "working parts" are working as they should. The newest technologies are always pushing boundaries of one form or another, challenging the limits of what can be done. That can put a thing on the edge of catastrophe - not a good place to be when lives are on the line. Newer computing technology is always more sophisticated (more complex) than what it replaces, making verification/validation more difficult.

Airplanes and rockets operate in harsh environments; the vehicles themselves create harsh or perhaps extreme environments for some of their own components. It's difficult to build electronic components and systems which can operate in such conditions - temperature, shock/vibration, EMI, radiation, etc. without some challenge to reliabilty.

It takes a long time (years) for a new aircraft or space system to make it from the initial design to "first launch", and design of subsystems (including those employing computers) has to be frozen at some point in the process. Computer technology moves much faster, so the designs get frozen with what is trustworthy (perhaps already becoming obsolete), and computer technology marches further along before the aircraft or rocket takes flight.

It really might not be a wise thing to try to do it any other way. When your life is in the balance, much better to have an old, crude but dependable system than something new and snazzy, but not fully proven.

Answer 9

Interestingly, this does not apply to all spacecraft. The Flock satellites by Planet Labs are actually pretty cutting-edge as stated by one of the developers on The AmpHour podcast. In fact, the testing of new satellite designs was slowed down by the time it took to actually launch the satellites once they were manufactured.

I suggest listening to the podcast, this episode was quite interesting.