Kuro5hin.org: technology and culture, from the trenches
create account | help/FAQ | contact | links | search | IRC | site news
[ Everything | Diaries | Technology | Science | Culture | Politics | Media | News | Internet | Op-Ed | Fiction | Meta | MLP ]
We need your support: buy an ad | premium membership

[P]
Extreme Microtech

By jd in Op-Ed
Mon May 27, 2002 at 10:32:02 PM EST
Tags: Technology (all tags)
Technology

You often hear that the Space Shuttle uses aged computers, but you don't often hear why, or what they could use instead. You sometimes hear that computers can't operate outside of a narrow range of temperatures (much to the disgust of overclockers), but often the explanation is a bit thin on the ground. So, what DO extreme technophiles use on their systems, to get them to work under these sorts of conditions?


Let us first start by dividing the types of extreme condition into several categories, and examine each category independently.

Extreme conditions can involve one (or more) of the following factors: Extreme radiation, extreme temperature values, extreme temperature variation, extreme acceleration, extreme speed and extreme reliability.

Extreme Radiation: This is becoming an increasing problem, not just for NASA, but for the technical industry as a whole. As computers become faster, they need to operate over narrower voltage ranges, to keep the heat down. (It's not much use if the computer cooks itself.) But, in the process, you need less and less outside interference to flip the state of any of the hundreds of millions of transistors.

How much energy is required to interfere with a computer chip? Well, if you have a 3.3v line, operating at 2 GHz, the energy in a single "state" is 3.3/200000000 of a watt, which is absolutely insignificant. A good, solid thwack by a cosmic ray could easily produce energies much higher than that, at aircraft altitudes, and are probably a measurable hazard even on the ground.

(The air is much more hazardous, as the atmosphere absorbs many of the high-energy particles, long before they reach the surface. Also, your house is probably a bit sturdier than an ultra-light airframe.)

Ok, so radiation is important to users of mobile computers in the air, and (as speeds increase) to computer users in general. It's absolutely critical, though, to all those computers the aircraft uses to navigate and fly. Spacecraft (probes, the shuttle, the space-station, sattelites, etc) get absolutely drenched in intense radiation, though. If the computers in the Space Shuttle get fried by radiation, the pilot can't simply order a fresh system over the phone.

For the avionics and space industries, then, there are entirely seperate, distinct computer industries, designed to serve specifically those kinds of needs. The components have much higher radiation tolerences (>100 Krads, in many cases, and a few >200 Krads), but the penalty is often much slower, and MUCH more expensive, components. These things don't come cheap.

There are three techniques for handling radiation, and different companies use a different mix of them.

The techniques are: Hardening of existing components, Building your own components, and Buildng tougher casing.

Hardening of existing components often involves re-arranging the components, adding error-checking and other wonderful stuff of this kind. The "Top Of The Line" components in this field are things like the Radiation Hardened Pentium Classic and the Radiation Hardened Sparc V7. Definitely not your latest-and-greatest! NASA has a basic primer to Radiation Hardening. Materials Science and Engineering, at Virginia Tech offer a more comprehensive tutorial on Radiation Hardening

Building your own components can be fun. And it's not that expensive! The technique used is called a "Sea of Gates". The system is essentially a massive array of radiation-hardened gates, with no interconnects. The customer then specifies a connection mask, which can then be retro-fitted to the Sea of Gates, producing a hybrid Custom / Mass Produced component. It's not as fast as a "pure" custom-built chip, but if you don't have a few billion to spend, then it's a good alternative.

Building toughened casing seems to be a relatively new field. The idea here is to abandon hardening the silicon, and to harden the case instead, making it opaque to high-energy particles. Because it is the case that is hardened, this technique can be used on any modern electrical component, regardless of speed.

Who makes these "radiation-hardened" systems, anyway? Well, here are a few companies that do:

A glance at these will show that many also produce components for the military. Well, duh! They're the ones who have funded most space research, have put most of the hardware in space, and are the only ones rich enough to afford to buy the rockets needed to get there. (They're also about the only ones licenced to even have that kind of rocket power.)

The list is not meant to be exhaustive, but rather a good selection of the sorts of things people produce. By glancing through the data sheets, you can see the sorts of conditions these devices are designed to handle. And it's not pretty.

Ok, so let's get back to the initial question. Why do people like NASA use 60's and 70's technology? Partly, because it's already tried and tested under the hostile environment of space. NASA knows what to expect. Partly, it's because the chips are much slower, and the transistors are much larger. This means that you need a significantly more powerful blast of radiation to mangle things. Lastly, it's because these chips are cheap - in price and in overhead. An 8086 doesn't have any space used up on MMX instructions, for example.

Price? Doesn't NASA have enough money? Well, no, they don't. Their resources are very limited, and shrink with each budget, both in absolute and real terms. A rad-hardened Pentium is likely to cost more than the rest of the satellite's systems put together. If you don't need the speed, you don't need the price-tag.

(That NASA is actually shopping on e-bay only goes to show just how severe their budget has become. There are other processors out there that'll work for what they want, and they know that. If they have to worry about margins that much, then you are looking at an organization that hasn't the money to safely launch a bottle rocket, never mind a Titan.)

For extreme temperatures, many of the same companies cover that ground as well. However, a search also turned up Lake Shore Cryotronics, as a company covers specifically high-temperature electronics and microelectronics. Another company that deals with temperature is Mikro Elektronik gmbh, a German company. (And I thought it sounded Australian! :)

Extreme motherboard designs (I don't have a URL to hand) can handle up to 20G, and some fairly vicious shocks. This would be good for, say, the volunteer lifeboat services, where tiny boats are skillfully navigated through violent storms, in efforts to rescue survivors from shipwrecks or aircraft crashes.

You can buy components for protecting entire systems from severe shocks and stress (one such system is sold by CSA Engineering for protecting satellites from the stresses of rocket launch.

What sorts of OS do people run on computers like these?

Well, probably something light-weight and real-time. You can't afford to wait whilst the disk is syncing, if you must give a correct response in the next three tenths of a second, to survive. You also can't afford anything memory-hungry (you won't be able to add that much memory), cycle-hungry (you can't afford the power requirements) or disk-hungry (a spinning hard-drive ain't gonna survive the North Sea, or a rocket launch).

This reduces your choice, considerably. In fact, you basically have a choice of the following:

  • QNX - One of the "classic" real-time OS'. It's popular, it's effective, and it does the job.
  • VxWorks - a Unix derivative, specifically designed for scientific and other extreme conditions. It's expensive, as is the hardware required, but it's generally considered "the best".
  • Roadrunner/Pk - about the most basic RTOS you can get, but if you are tight on RAM, it's a good one to go with.
  • Exopc - an experimental high-performance OS from MIT. I'm not sure if it's strictly RTOS, but it does have low latency and a design that offers a lot of potential.

Temperatures: Most modern electronic components come in one of two temperature ranges: -40 to +65, and -20 to +65. (All temperatures are in celcius.) This is fine, for most desktop PCs, but what if you were in the middle of the Antarctic? If your components go out of range, above or below, they can be damaged or destroyed.

Again, we look to the suppliers of the military. Military standards are much more exacting than those the rest of us get to put up with, and the temperature ranges are much more severe.

Capacitors and Connectors, for example, sell military-grade components. The typical range rockets to -55 to +125. You can now talk about computers that can be used to boil water, and remain operational, or be used where CO2 falls like snow. Other companies selling extreme temperature hardware are Vishay and Microcapacitors.

Ok, so when would you need such components? Let's take a look at some examples. Ships in the North Sea are often in conditions so cold that the sea spray will freeze solid on impact. The Arctic and Antarctic visitor is likely to experience conditions many times colder yet. As for any future manned mission to Mars.... That polar ice cap is still believed to be frozen CO2, so anyone going is likely to need to wrap up warm.

That's a significant number of groups that experience extreme low temperatures. What about high temperatures? Well, I imagine most vulcanologists can name situations where high temperatures might be an issue. Nor would I want to run a standard air-cooled PC in Death Valley. Firemen, using sensors to detect trapped people, need electronics that won't fry, if the people aren't to fry too.

All in all, surviving high temperatures is an issue for a lot of people. And, as I said, most of the components designed for that kind of work are designed for the military.

What about high temperature variations? Materials expand and contract with temperature, and if the variability is too great, the material will simply fall apart. Components that have materials with different thermal expansion properties have a worse time. They can bend, break, and then disintegrate. Not good.

I've not found components designed for high variability, but I'm going to assume that the military gradings take this into consideration.

I'll deal with extreme speed and extreme reliability quickly. (Hmmm!) The colder you make an electronic component, the faster you can make it run. This is why supercomputers are invariably super-cooled. And this gets back to needing components with high temperature tolerance. The greater the tolerance range, the colder you can get the system, and the faster the system will then run.

The current "land-speed" record for a Pentium IV is about 3.66 GHz, using a 2.0 GHz P4 as the starting point. The current top-of-the-line P4's should scale even better.

The drawback of overclocking like this is that nobody really knows the impact on the components. Terms like "electron migration" are bandied around, but nobody has really studied this in depth. What is certain, though, is that the effects will be dependent on the amount of cooling versus the amount of overclocking.

Extreme reliability: This is something that older components (again!) do better, because there are fewer parts to go wrong, and it requires a greater cause to produce such a fault.

Typically, components have what is called a "Mean Time Between Failures". And, again, your computer is likely to be made up of parts with abysmal MTBF's, and the jet fighters are likely to have parts that'll likely survive into the next millenium after the one that comes after we invent Star Trek transporters.

Actually, this is the most realistic part of the old Buck Rodgers stories. It is entirely believable that the components in a Space Shuttle could be of a quality such that the MTBF exceeds 500 years. And I would be very surprised if there were not military systems in use today with MTBFs in that kind of order.

"Domestic" uses for such systems would include emergency medical equiptment ("Oops! The Life Support blew a fuse. I'll be back in an hour!") and systems for other comparitively critical systems, where a single failure in the operational lifetime is one failure too many.

This is why many computers, programs, etc, have large disclaimers, prohibiting any use in nuclear reactors, life support systems, etc. It's because the MTBF is just too low to take real risks with. But your home banking & recipe book isn't worth as much (to them), and customers don't care enough about the MTBF to warrant spending the extra bucks in building a more robust system.

(If you're going to replace your PC every 3-4 years, do you really NEED every component guaranteed to last 30-40 years, under maximum load? Or will something that could well burn out in 4 years be just as good?)

Sponsors

Voxel dot net
o Managed Hosting
o VoxCAST Content Delivery
o Raw Infrastructure

Login

Related Links
o NASA
o Radiation Hardened Pentium Classic
o basic primer to Radiation Hardening
o Materials Science and Engineering, at Virginia Tech
o comprehens ive tutorial on Radiation Hardening
o "Sea of Gates"
o Honeywell
o Atmel
o LEON processor
o Gaisler Research
o AITech
o Maxwell Technologies"
o Micropac Industries
o Lake Shore Cryotronics
o Mikro Elektronik gmbh
o CSA Engineering
o QNX
o VxWorks
o Roadrunner /Pk
o Exopc
o Capacitors and Connectors
o Vishay
o Microcapac itors
o Also by jd


Display: Sort:
Extreme Microtech | 31 comments (18 topical, 13 editorial, 0 hidden)
Web page - not news - material (2.14 / 7) (#5)
by dbretton on Mon May 27, 2002 at 03:08:43 PM EST

Hi,

   I found your article very interesting.  As someone who is involved in the Defense and Aerospace industry, I found what you had to say to be relevant and informative.
-here comes the "but"-

BUT, there's really nothing here that's newsworthy.  That is, the information is not new, and does not stir much in the way of conversation.

In short, this would make for a nice static web page somewhere, but probably isn't sufficient to make it front page news.
If you can read this, you are too close.

Uhhh (4.77 / 9) (#9)
by siobibble on Mon May 27, 2002 at 04:37:39 PM EST

Kuro5hin isn't a "News" site like <cough cough>.

You CAN post news here, but it must be able to start an interesting, intelligent, and thoughtful conversation.

Please read the kuro5hin mission statement here: http://www.kuro5hin.org/?op=special;page=mission

[ Parent ]

But (4.80 / 5) (#13)
by JanneM on Mon May 27, 2002 at 04:47:51 PM EST

As you say, you are involved in the aerospace industry. For you this is mostly old hat. For most people - even technically inclined - this is a new, different, way to think about the technology around us. While we have a vague idea that environmental factors can be a consideration outside cubicle use, it is not something we tend to express clearly and think about. And as for this not being news, well, kuro5hin isn't a news site, really. Rather, it's a place for informed discussion, and while the jump-off point may be news, there is nothing that says it should be.

/Janne

---
Trust the Computer. The Computer is your friend.
[ Parent ]

VxWorks (5.00 / 7) (#11)
by emag on Mon May 27, 2002 at 04:45:37 PM EST

I've worked with it in the past.  It's an...interesting system.  I'm not sure I'd call it a Unix derivative so much as an RTOS that had been mapped to Unix functionality.  In fact, IIRC, BSD-style socket calls are actually an add-on to the base system.  It's also inherently multi-threaded (we were using this a lot), without needing packages like pthreads.

Like other (most?) RTOSes, the base system is fairly generic, with the need for a BSP (board support package) to run on any particular hardware.  You build the kernel, either through mucking with Makefiles, or using the GUI they supply, selectively leaving out parts.  Don't need floating point?  It's gone.  Don't need Unix syscalls?  They're gone.  Don't need support for C++?  It's gone.  This allows for a very customized build, running as fast as possible.  Some of the apps we were designing needed hard response times on the order of a few milliseconds, at most.  We were aiming for significantly below that level though, just to account for anything that "might" go wrong and need the additional time for error handling.

As for the hardware, the BSPs support a vast range of architectures and platforms.  The only reason you might consider it to be "expensive", is usually you pay a price premium for the embedded versions of normal off-the-shelf equipment.  Our group was working with PC104/PC104+ (approx 4" x 4" stackable cards) Pentium-based systems.  We could have actually done a lot of the initial testing on a normal PC had we been so inclined.  Another group was using PowerPC, which is still somewhat used in the embedded world.

It's in interesting system to develop for.  And the Unix-like qualities definitely help a lot of people who wouldn't normally be embedded developers jump right in and be productive.  Of course, when I left that job, there was talk about investigating moving some development to RT/Linux, so as to not be locked into a single-source supplier.  I've no clue if the customer (military) ever let that go through, but it's interesting to note even places that have "traditionally" been hard-core real-time (original system was hardwired, then moved to AN-UYK systems, then the PPC/x86 world, with successive generations) are moving to more "modern" components, and experimenting with newer technologies.

--
"The urge to save humanity is almost always a false front for the urge to rule." --H.L. Mencken

Embedded warnings (4.00 / 3) (#20)
by Rasman on Tue May 28, 2002 at 04:18:11 AM EST

I used to work with embedded computers too, and I saw the same move you're talking about. We had some of them running Windows 2000 (and not even real Win2k, but Windows 5.0b before they came up with the "2000" naming scheme - scary!).

I always used to get a kick out of running a program and the first thing you see says, "Warning! Do not use this program for helicopters or any application where a failure could risk human life. We are not responsible for any deaths resulting from the use of this program!". It makes you wonder if your little cubicle is providing enough protection, like maybe I should start coding in a hard hat or something...

[ Parent ]
Shuttle Computers (4.80 / 5) (#18)
by Bad Harmony on Mon May 27, 2002 at 10:54:56 PM EST

The Shuttle computers (IBM AP-101) were selected in part because of their prior use as flight control computers in USAF aircraft. They were originally designed and built by 4Pi, and were based on the IBM 360 architecture. They were radiation hardened and used magnetic core memory. Like the IBM 360, they were optimized for huge amounts of I/O. Flight control applications require computers that have deterministic timing, low-latency I/O, high reliability and operation over a wide range of environmental conditions. The controllers for the Shuttle's main engines use dual-redundant Motorola 68000s, one controller per engine.

5440' or Fight!

RTEMS -- Open Source RTOS for missile control.... (4.33 / 3) (#19)
by ckm on Tue May 28, 2002 at 03:22:52 AM EST

You forgot to mention RTEMS (http://www.rtems.army.mil/rg4/rtems.html), a real-time OS original designed as a missile control system.

It's Open Source, runs on a wide variety of platforms and is proven in, er, production....

Chris.


Radiation hardened (3.00 / 1) (#22)
by Herring on Tue May 28, 2002 at 05:09:18 AM EST

This good enough for yah?

I would've thought that would take a quite a bit.


Say lol what again motherfucker, say lol what again, I dare you, no I double dare you
But... (3.75 / 4) (#23)
by dissonant on Tue May 28, 2002 at 11:12:58 AM EST

..no HAL-9000 unit has ever been known to fail. It can only be attributed to human error.

(-8

100 Krads (4.66 / 3) (#24)
by wiredog on Tue May 28, 2002 at 11:28:58 AM EST

Hardening to resist 100,000 rads? In aircraft? Why? If the 767 gets hit with 1/50th of that (2000 rads) it's going to crash, regardless of the hardening, because the flight crew will be dead.

"one masturbation reference per 13 K5ers" --Rusty
Do more research... (5.00 / 2) (#25)
by JonesBoy on Tue May 28, 2002 at 03:42:06 PM EST

I have a few troubles with this article.

1) energy in a single "state" is 3.3/200000000 of a watt
No.   Volts*Time != power.   Is a 1 year old 1.5v battery 31Million watts?

2) as the atmosphere absorbs many of the high-energy particles
NO!   Otherwise we would use air for radiation shielding!   Its the earth's magnetic field that keeps us safe.   The higher up you are, the fewer magnetic lines per given area, the less protection, the more radiation.

2b)Also, your house is probably a bit sturdier than an ultra-light airframe
No, dense, conductive metal would make a better rad shield than non-conductive sheetrock and wood.  Depending on radiation type, you either want grounded metal or a dense substance (lead, wax, etc)

3)  Building your own components
SOG components are not necessarily rad hardened.

4) Temperatures are broken down into 3 basic ranges.   Commercial, Automotive/Industrial, and Military.    The ranges are usually  -20 to +65, -40 to +85, and -55 to +125.

5) Terms like "electron migration" are bandied around, but nobody has really studied this in depth.
Yes they have, but not by joe "brute force" overclocker.   The Navy did a lot of research into this, and I am sure you can find information on this in a good library, or even possibly the internet.

6) What is certain, though, is that the effects will be dependent on the amount of cooling
No.   It is dependent on current.   Go do research.   Here is 1 if you are feeling lazy.   http://www.altera.com/literature/wp/wp_copper.pdf

6) entirely believable that the components in a Space Shuttle could be of a quality such that the MTBF exceeds 500 years
That whole paragraph is bogus.   I wouldn't count on wire insulation lasting that long, let alone a processor or any transistor (think: dopant mobility!)   Even military equipment has a mtbf measured single digit years.

No offense, but this article is unresearched heresay.
Speeding never killed anyone. Stopping did.

Speaking of more research... (3.50 / 2) (#26)
by a humble lich on Tue May 28, 2002 at 07:33:25 PM EST

I think you are being a bit too harsh. I believe the atmosphere provides quite a lot of radiation shielding. I for one am quite glad we have an ozone layer. While you are right that the Earth's magnetic field is what deflects most high energy radiation, the atmosphere is what protects us from the rest; e.g., all the high energy photons out there. Most "cosmic rays" which we see (if memory serves) are high energy muons which are created in the upper atmosphere. Muons are short lived, so when you are at a higher altitude more muons survive to interact with you.

The we don't use air for radiation shielding, is there are few cases where in is practical to have miles of shielding. If the atmosphere did nothing to protect against radiation, God help anyone living near the pole.

As far as radiation shielding goes, I can't think of any reason why electrical conductivity would help against high energy particles (you want a conductive shell to shield out radio). I know the big concrete apartment building I live in would shield against radiation far better then any weeny airframe.

[ Parent ]

Applications to topic (none / 0) (#31)
by JonesBoy on Thu May 30, 2002 at 01:23:44 PM EST

Ok, but the particles are decaying over time, not because of the aptmosphere.   You would get the same affect in a vacuum.   The air really is not helping.   Yes, the ozone layer helps with low level rad, but that stuff would never penetrate the plastic packaging of an IC, and therefore not applicable to the article.

Read up a little on the topic.   Yes, there is a lot more interferance if you live at the magnetic poles, but it isn't so bad because the poles are on a line normal to the sun.   The aptmosphere plays no part in the radiation shielding, except for low energy rad near the visible spectrum.
http://www-ssc.igpp.ucla.edu/personnel/russell/papers/earth_mag/

Electrical conductivity is important for RF level interferance.    Many times I have seen circuits go nuts when some bozo keys up a high powered CB.    Some old Bosch EFI units had trouble with interferance from HF ham radios.   Since it interferes, I figured it was applicable to the article, and I included it.  

Yes, concrete would be better than an airframe under some circumstances, but if you look at my comment, I said grounded conductor or DENSE SUBSTANCE.   Concrete can be dense, and therefore help protect you.

Speeding never killed anyone. Stopping did.
[ Parent ]

And another link (none / 0) (#32)
by JonesBoy on Thu May 30, 2002 at 01:32:57 PM EST

http://aoss.engin.umich.edu/earth_space/pdf/magnetosphere_2.pdf
This link explains the roles of shielding better, and the role of the holes is the poles.

Oh, and I forgot, nobody lives near the magnetic poles.   They are currently in the ocean.

Speeding never killed anyone. Stopping did.
[ Parent ]

Just a note on RTOS (none / 0) (#27)
by mindstrm on Tue May 28, 2002 at 11:15:16 PM EST

It is often stated by people that RTOS (realtime operating systems) are faster than non-rtos. This is not the case. In fact, they have longer overall latencies for system calls.

The point of an RTOS is that there are *guaranteed* maximum latencies for every operation, so you can design around these maximums. You know that each operation will take at the most a certain time, no matter what.

RTOS (none / 0) (#28)
by salsaman on Wed May 29, 2002 at 05:31:03 AM EST

Huh ? Why the heck would anybody think a RTOS would be *faster* ? Obviously there are more overheads with a RTOS, you will be checking interrupts much more often and switching processes in and out at possibly inconvenient times.



[ Parent ]

You're Both Right...You're Both Wrong (none / 0) (#29)
by dbretton on Wed May 29, 2002 at 01:04:54 PM EST

A Real-Time Operating System (RTOS) is designed to provide a definite behavior under all provided conditions, given a finite set of well-defined conditions.

If you tested a RTOS's response in all cases, ranging from trivial to the stressing cases, then compared these results against a non-RTOS, most of the time (probably around 80%), you will find the average performance of the RTOS is better than the average performance of a non-RTOS.

Of course, hardware design/retrictions need to be taken into account.  That is, a non-RTOS will most likely run circles around a RTOS is both OS's are installed on some hardware that is never really stressed...

If you can read this, you are too close.
[ Parent ]

Terminology (none / 0) (#30)
by Bad Harmony on Thu May 30, 2002 at 02:59:26 AM EST

Many people confuse real-time with real-fast. One of my real-time programming books uses a payroll system as an extreme example of a real-time system that has long, but hard, deadlines.

5440' or Fight!
[ Parent ]

Extreme Microtech | 31 comments (18 topical, 13 editorial, 0 hidden)
Display: Sort:

kuro5hin.org

[XML]
All trademarks and copyrights on this page are owned by their respective companies. The Rest 2000 - Present Kuro5hin.org Inc.
See our legalese page for copyright policies. Please also read our Privacy Policy.
Kuro5hin.org is powered by Free Software, including Apache, Perl, and Linux, The Scoop Engine that runs this site is freely available, under the terms of the GPL.
Need some help? Email help@kuro5hin.org.
My heart's the long stairs.

Powered by Scoop create account | help/FAQ | mission | links | search | IRC | YOU choose the stories!