Slashdot Mirror
First LEON Silicon Tested Successfully
uglomera writes: "LEON, the open-source CPU developed for the European Space Agency, has been successfully manufactured and tested by Atmel on an Atmel ATC35. Gaisler Research, whose CEO Jiri Gaisler wrote the VHDL model of LEON, also offers a real time kernel, simulator, a cross-compiler, etc. for this SPARC-family processor designed for space applications. Check it out." You can find more good information on the LEON processor on the Gaisler site, including diagrams and further reading. Open Source hardware running Free software -- wheee!
and LEON is getting larger!
NASA, in the USia, spends literally millions of dollars designing the space shuttle's computing system from the ground up. Their in-house coders pour over each microchip and line of code hundereds of times looking for even the smallest bug. This superior attention to detail is not possible using the limited resources of the open-source method. It is also why NASA can succeed in safe, reliable space flight time, and time again, while other space programs are struggling. Open source has it's advantages, but when 100% reliability is necessary, it may not be the best option.
Slashdot: Open Source, Closed Minds.
More stuff like this can be found here: opencores
I doubt they plan to send this widget up without a full seperate functionality audit, which they can do for an open source processor, but it might be hard to convince say Intel to let them pore over the VHDL for the celeron or whatever.
Just because it's open source doesn't mean they aren't going to put it through the same rigorous tests that they would put a commercial processor through.
---
Play Six Pack Man. I
Enjoy your job, make lots of money, work within the law. Choose any two.
Open source hardware seems a bit odd. The nice thing is, yes, we can have the specs to make apps really fast. Instead of, for example, having to wait for Intel to decide to contribute to gcc, kernelsm or (finally) release their in-house (Linux) compiler. So a more open spec on the processor would allow people to actually get things right. I guess the question that sits on my mind is what is truly more valuable, a truly wide open spec or a more open design so anyone can manufacture one.
I guess my feelings are known because I believe that hardware's design should not be completely open. Let a manufacturer keep it all to themselves. However, when you buy the processor you should have the _complete_ spec available. (which is way different than Intel does now)
---
My opinions do not represent those of my employer of course.
I was speaking to a co-worker the other day about Sun's UltraSPARC III processor and he was telling me about CPU manufacturing in general.
To actually create a fabrication facility to make CPUs, it takes about 20 billion dollars. $20,000,000,000 dollars. That's more than most companies can afford. Even Intel couldn't make very many new fabrication plants.
If a company can't afford to create their own plant, they have to schedule time at a fabrication facility. This is basically a window (say 48 or 72 hours) where the facility will crank out as many chips as possible. If they miss the window for some reason, they have to re-schedule and it can be months until there is another open time slot.
What I'm getting at is designing Open CPUs is a great idea. It allows developers to really get inside of the hardware and optimize the hell out of applications, which is a good thing. However, the actual cost to make these CPUs is staggering, and unless a big company put up some big bucks, I don't see it happening in the near future.
My co-worker also mentioned how low cost almost everything else is. Video cards, NICs, sound cards and the like. Wouldn't it be better to focus on products like these since they would work with all hardware (how about a video card that worked on Sun and x86 machines?)? With Moore's Law getting us faster and faster CPU speeds, perhaps it's time to make the peripherals first, and focus on a CPU once we have found sucess with smaller projects.
If I recall, the US space shuttle runs on something like a dozen underclocked i286's, each processor with something like 5 way redundancy. Each processor in a set of five will perform the same calculations. The solution presented by the majority of processors is deemed to be failsafe. The 80286 was chosen because it is simplistic, and it's reliability has been proven through years and years of experience in the field. In addition, Intel has spent millions on development and testing of this processor to ensure it is 100% reliable.
Slashdot: Open Source, Closed Minds.
negative spin.
LEON
The LEON model also exists in a fault-tolerant
version, capable of detecting and correcting
single-event upset (SEU) errors in any register
or RAM element. This is done completely in
hardware without any software intervention.
The area overhead for the fault-tolerance
functions is approximately 30% while the timing
penalty is around 5%. The fault-tolerant features
makes it possible to use LEON in the severe
space environment without having to develop specific SEU-hardened cell libraries. The LEON
fault-tolerant VHDL model can under some conditions be licensed from ESA -
Ethics II Axiom 2. "Man thinks." B. Spinoza
It sounds like you do no believe that the brits do anything other than run down to the local electronics shop, slap a bunch of stuff together and fire it off into space. I do not deny that space flight is dangerous, but the thought that a single component failure producing catastrophic results when talking about computer equipment is not bothering to think.
If they are doing anything like what NASA has done in the past, then there is not one, but three computers for each group of tasks. These three computers vote, if one is consistantly out of sync with the others it gets shut down.
I don't think that NASA has cornered the market on testing systems. As the public has even seen, Intel, and Motorola are not perfect. What the public does not tend to see are the less well known bugs which have existed and gone on with less public fanfare. The one I struggled with the most was a bug in the 386 chip which made it useless for network based semephors.
Equating Open source with bug ridden and unreliable is to look upon what is currently available with closed eyes.
In a place beyond time and space, in a land far better than this, look for me there...
Assuming that the processing demands of current and future space flights are increasing beyond what a dozen 286s can accomplish, what would you have NASA, ESA et. al. do? Use more and more 286s? Use a higher end PC processor? Use a normal workstation processor? Or use a radiation-hardened processor that is compatible with a popular, well tested processor? (SPARC)
Enjoy your job, make lots of money, work within the law. Choose any two.
Thousands of excellent designers work together from tens of nations, and they start mixing feet with meters again... Yikes!
Now where did I leave my babelfish...
Sofar.
The 286 (special version) was chosen because it was radation shielded relatively easily. They reused parts from previous missions, probably to save on R+D. I'm not totally sure, but I'm relativley sure thate the Hubble telescope runs on 386s (massive floating point performance here ;), that are radation shielded as well.
I was just wondering: are the shuttle and Hubble programmed in ADA? Just a thought. yeech.
Constitutional rights may be respected, repealed, or modified; but they must never be ignored.
The availability of an inexpensive, radiation tolerant CPU is a big win for space researchers. Right now there are darn few radiation-tolerant parts available for use in space applications due to decreased demand from the military. The International Space Station is using Intel 386s for embedded CPUs, as they are simple enough to be relatively rad-hard. More modern CPUs, such as in the laptops used on the ISS and Shuttle have about one lockup/day due to radiation.
The design requirements for software controlled systems in space are so stringent that to do anything sophisticated requires incredible redundancy, cross checks among the systems, and increased design complexity, all of which significantly drives up design costs (and causes all kinds of debugging problems). Tell me three times is not enough, you have to tell three controllers three times, three different ways and then they need to cross check. This could be a big step forward for software geeks in space.
I saw a number of projects on the net where people try to build a pure hardware codec using FPGA. It seems that the better approach is to build a CPU-based, hardware-assisted codec.
Here is a great opportunity to free music from MP3 license payments. If somebody creates an open-source reference hardware/firmware implementation, Far-East companies will start making cheap portable players/recorders in no time.
As for the development costs: many FPGA vendors provide their software for free or for a small price, because they make money on their chips. The only problem is a good Verilog/VHDL simulator. FPGAs themselves are pricey, but there are some one time programmable devices (Atmel, Quicklogic) that cost under 50USD.
$20 billion? Hell, an aircraft carrier only costs about $3 billion. IIRC the newest Intel fab cost about $2 billion.
Best Slashdot Co
"Their in-house coders pour over each microchip and line of code hundereds of times looking for even the smallest bug. This superior attention to detail is not possible using the limited resources of the open-source method."
Sure it is, here's how: Use exactly the process NASA uses know and that you are apparently comfortable with. Then ADD (not replace) more programmers by making the source available via FTP.
Adding openness to an existing project loses nothing. Yes, shifting the burden of quality OFF of some process ONTO openness may not always be a good idea (not in one go, anyway). But adding more checks doesn't lower quality.
--
324006
``Leon the pig farmer'' boards the shuttle? (-:
Got time? Spend some of it coding or testing
Open Source doesn't mean it was developed by the masses. It just means it's licensed under an open-source license. It means that if others want the design, they can have it, and use it.
You can bet that Nasa usest he same quality controls on these circuits as with any other.. they just release it to the public.
Well, we've got Open Source hardware in the news. I figure with all the theorizations, I'd toss in my two cents on what this will entail.
THE BAD:
1) Expect to see some of the cultic behavior we've seen that has affected Linux's reputation - people jumping on every note on LEON and related technolog as the Ultimate Thing to Save Us from Microsoft. Expect this behavior to be noted by non-OS manufacturers and used against OS hardware.
2) Expect serious reality cramps when people discover just how much fabricating chips cost. Expect conspiracy theories to emerge.
3) This is a first step, and there's a lot further to go.
THE GOOD:
1) OS processors are at least feasible. Let's face it - this is just cool.
2) The genie is out of the bottle - the idea is there. It will spread.
3) Intel has been made a fool of by AMD. Transmeta (associated with Linus Torvalds fortunately) has their new chips. Now we've got this. People are starting to rethink chips, processors, etc.
Do I think a revolution just started? No, though I expect some people will play it up as such. There may be a revolution, but it won't happen immediately.
However, a good idea is out there and its physically manifested. I expect good things to come of it - just not right away.
"The Sage treasures Unity and measures all things by it" - Lao Tzu
Typically it costs few hundreds of dollars. They make money on chips, and sometimes even give software away for free. Even for ASIC's, there are some low-cost alternatives, like Alliance ( http://www-asim.lip6.fr/alliance.html )
Now several electronics mfg's can band together to produce chips. The chip making club will still be small just not as small or as expensive.
By definition, a government has no conscience. Sometimes it has a policy, but nothing more. - Albert Camus
; there's no reason not to trust open source developers; they're much better than the money-hungry programmers hired by coprporations.
;-)
So why are "money hungry" programmers any less driven than freebie bedroom programmers?
They are probably more motivated on occasion (money, moolah, ca$h), and may actually have achieved a higher degree of academic excellence to get their high-paying jobs in the first place.
In saying that, I have been guilty of knocking out the odd Friday-afternoon bit of code!
(this is my first thing i wrote for slashdot!)
Congratulations! With a user number like that you must have been lurking for a considerable period of time. Expect the Grammer Nazi at you for that spelling anytime soon
Oh, please.
Often the open source developer and the "money-hungry programmer" is the same. Where do you think we get the money to make a living from? Here's a hint: We're not the voice in McDonald's box.
Je ne parle pas francais.
You'd be much better off using ASICs so far, but FPGA's are getting faster.
"wheee" indeed... The thing is running at 60 MHz. LEON may be "open-source", but I'll stick to Intel processors that are 10 times faster and probably cheaper, thank you.
What is actually open-sourced is VHDL code. That is a language that describes a hardware design. It can be written at various levels, I assume that for a CPU they'd carry it to the lower levels (registers and gates). But to actually get a chip, you've got to compile the VHDL for the particular chip-making processes. This isn't as easy as compiling a C program, it takes considerable human intervention to get a physical fit and correct for various process limitations. And then you've got to create the exposure masks for each step in the process (dozens of them), and finally make a batch of chips. It might take $0.5 million to get the first chip. But that's a few million less than if you created a CPU design from scratch, or licensed one from Intel.
Notice that the implementations in silicon which have got as far as test are quite slow (down to 35MHz), while the chips that are still in process are estimated at 90 to 150MHz. But maybe a Sparc-compatible at 100MHz running *nix will beat a P-III at 800MHz burdened by Windoze? At any rate, it is one heck of an improvement on the underclocked 286's and 386's that are now the peak of space-rated technology. And since even the best quality checks seem to be subject to blind spots, getting outsiders to pick away at the design it should improve the chances of finding bugs before they pop up in flight. (Blind spots: ESA: A new Ariane rocket blew up on the first launch because a calculation related to velocity overflowed, in software was carried over from an older, slower rocket. NASA: Lost a Mars probe because they forgot to convert pounds to newtons. Intel: Pentium divide bug.)
So why are "money hungry" programmers any less driven than freebie bedroom programmers?
Some reasons spring to mind:
They are probably more motivated on occasion (money, moolah, ca$h), and may actually have achieved a higher degree of academic excellence to get their high-paying jobs in the first place.
Academic excellence != good programmer. Even most of a degree in Computer Science you will never use in industry. I did some really weird and academic modules that were useless. As for high paying jobs, I've found that pay tends to be inversely proportional to the challenge. My choices between jobs have been pretty well divided along the lines of 'interesting' or 'well paid'. I always pick the former as it pushes me more and builds up skills I can always trade in later if I get wife/house/kids/etc.
Phillip.
Property for sale in Nice, France
"They should implement the processor in some sort of non-volatile programmable logic." That would be nice. And it has been done for the Z80 and maybe some other 8-bitters. You take too big of a performance penalty to do this with a 32-bit CPU, so far. That will change, but by then 64-bit CPU's will be standard in desktop computers... So for now, expect to see this implemented as ASIC's (semi-custom chips) and maybe full-custom.
Also, the cost of compiling VHDL of this complexity to fit into any particular chip is probably in the low six figures. Sorry, but they are going to have to get it right, then be sparing with the upgrades.
The LEON being a SPARC compliant chip will it then be able to SMP? Isn't it Linux/SPARC that scales to 64 cpus?
Even if this chip sounds slow to me (25 MHz) with 64 of them it will start to take off.
So the question is really _does_ this new processor do SMP?
The ESA is a gov't agency, they aren't concerned with competition. I think what they are really hoping for (besides getting enough eyes looking at the VHDL code to prevent anything like the Pentium divide bug from sneaking through), is that there will be sufficient commercial volume to drive the prices down for them. If they kept it closed source, it would simply remain a low-performance, high-cost chip built specially for space applications.
At least somebody thought of FPGAs instead of billion dollar plants (and the future ink jet ICs!).
But you can get free software from Xilinx that will program the Spartan 2 series and the Virtex 300. I'll bet that the LEON could fit in a US$20 Spartan 2 150...
For those interested, it was the testing instrument of the lens which failed, so the lens was polished correctly according to the specifications of the instrument. So one could say that the testing procedure was flawed, and not the quality of the testing itself. I guess this is a difference, that when you do something "in-house", it's difficult to get fundamental criticism, i.e. out-of-the-box criticism. In open source contexts, you almost always have someone pointing out fundamental flaws, because they are not afraid of criticizing "the boss".
Your recollection is wrong.
The shuttle uses five IBM AP-101S computers. They are not microprocessors. Architecturally, they belong to the IBM 360/370 family of computers. See http://www.fas.org/spp/civil/sts/newsref/sts-av.ht ml.
The shuttle's GPCs are programmed in HAL/S (high-level aerospace language/shuttle).
Alright! Now as soon as it is proven to be reliable,building the computers for a starship might not be so hard.;)
Geek Hillbilly
Seems pretty open source to me, they even have a shot of the sim running on KDE.... pity I use GNOME
If I was in some extremely hazardous environment, and relied on a computer to keep me intact, I'd go for one that'll do the job, whatever speed it ran at.
If you'd prefer the flat-out MIPS rating, go for it. Your choice. Just don't go into space, the Irish Sea, anywhere that's glowing, or within a few hundred miles of any EMP stuff. You'll be just fine. Bored witless, but fine.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
...the compiler is gnu and the money isn't there since the spec is open anyway.
Not so. There's plenty of money in porting gcc, the linux kernel, &c. to different architectures. Cygnus Solutions does just that and makes money at it; my company, MontaVista Software, has been responsible for getting Linux to run on several boards -- and free software or no, there are companies who are more than happy to pay us to do it.
Presumably this works for hardware too.
How fast do you people think that this processor could run ... like in the future once it gets developed further? How would it measure up to, say a Pentium III or a SPARC or whatever.
Open Source processors sound like a pretty cool idea and I hope it gets popular, but for now (or maybe 5 years from now), it seems like they'd fall way behind (speed-wise)traditional ones ... but that's speaking from a consumer angle. My understanding is that this would be a universal platform for the research and [aero]space community ... where shared development and not necessarily processor speed is the main issue.
-Christian
our written thoughts are gifts to our future selves