Slashdot Mirror

← Back to Stories (view on slashdot.org)

Project Aims To Build a Fully Open SoC and Dev Board

Posted by timothy on from the reducing-the-riscs dept.

DeviceGuru (1136715) writes "A non-profit company is developing an open source 64-bit system-on-chip that will enable fully open hardware, 'from the CPU core to the development board.' The 'lowRISC' SoC is the brainchild of a team of hardware and software hackers from the University of Cambridge, with the stated goal of implementing a 'fully open computing eco-system, including the instruction set architecture (ISA), processor silicon, and development boards.' The lowRISC's design is based on a new 64-bit RISC-V ISA, developed at UC Berkeley. The RISC-V core design has now advanced enough for the lowRISC project to begin designing an SoC around it. Prototype silicon of a 'RISC-V Rocket' core itself has already been benchmarked at UC Berkeley, with results results (on GitHub) suggesting that in comparison to a 32-bit ARM Cortex-A5 core, the RISC-V core is faster, smaller, and uses less power. And on top of that it's open source. Oh, and there's a nifty JavaScript-based RISC-V simulator that runs in your browser."

31 of 47 comments (clear)

  1. expecting performance from IE? by raymorris · · Score: 2, Insightful

    Are you expecting high performance from Microsoft IE, in their JScript engine?

    One of the reasons Chrome EXISTS is to provide a high performance platform for Google Docs, Gmail and similar large JavaScript applications. These are the applications that intend to replace Microsoft' s cash cow, Office. It would be better for MS to stop shipping IE at all than for them to provide an excellent platform in which to run Google Docs.

  2. Re:linux by NotInHere · · Score: 1

    Yes it does, but it needs three minutes to boot to shell, at least in firefox.

  3. Re:linux by Anonymous Coward · · Score: 1

    I just ran Linux on "it", inside Chrome, which is running in a Linux VM on my MacBook Pro.

    Yeah. That's not good somehow.

    Based on the reported IPS of the RISC-V CPU emulator, it takes 1839 actual hardware instructions to cycle the RISC-V once.

  4. Where are those chips baked? by Mister+Liberty · · Score: 1

    That's what I was about to ask.
    But then I wondered -- what actually was the motivation for this all out Open Source SoC?

    1. Re:Where are those chips baked? by sixoh1 · · Score: 3, Insightful

      From the article they are using TSMC, which is one of the largest silicon foundries (ASIC manufacturing) in the world.

      As for the all out open-source, they also make clear on the project page that hardware patents on the chipset instruction is supposedly strangling innovation for processors. I'm not sure I buy that, ARM, Intel and IBM have moved their architectures along pretty well. Even poor little MIPS has made strides despite losing market share.

    2. Re:Where are those chips baked? by Anonymous Coward · · Score: 1

      Cross-licensing? The strangling is probably that the barrier of entry for any new player is incredibly high, which sucks big time. It's like this in all sufficiently large tech markets it seems. The theory is that the one who makes the product people want wins. In practice, the one who kills the competition (using patents for example) wins.

    3. Re:Where are those chips baked? by Casandro · · Score: 1

      Well you can fix all the problems existing in current SoCs. For example you could build an architecture which enables you to have multiple SoC boot up from the same image, just like the PC does. You could have basic hardware support without binary blobs.

      In essence you could create a new portable platform where you could, for example, swap out the operating system on your mobile phone just by putting another OS onto your SD-card. That way even if your vendor doesn't support your device anymore, you can still get the latest version of whatever operating system you want. In fact as it'll greatly simplify making distributions, you could even get specially tailored operating systems for your needs.

      I hope the people doing this will understand their chance.

    4. Re:Where are those chips baked? by Anonymous Coward · · Score: 2, Informative

      As the blog in EE Times ("The Case for Free, Open Instruction Sets") argues, an ARM license costs $1M to $10M and takes 6 to 24 months to negotiate and then they take a small royalty per chip.
      http://www.eetimes.com/author.asp?doc_id=1323406

      The proprietary instruction sets (ARM, IBM, Intel) have indeed evolved; that is not the problem. The problem is that you're not allowed to share implementations of the proprietary instruction sets with others. Thus, the lowRISC project is using a design from UC Berkeley for free without having to take the time or money to negotiate a contract, and they can modify it as much as they desire. Can't do that with ARM.

    5. Re:Where are those chips baked? by asbradbury · · Score: 1

      Just to be clear, lowRISC is a separate project to the RISC-V instruction set architecture, though we are lucky enough to have Krste Asanovic on our technical advisory board and are working with the Berkeley team. The results for the 'Rocket' core from Berkeley are using TSMC 40nm, but that isn't necessarily what we will produce lowRISC on.

    6. Re: Where are those chips baked? by asbradbury · · Score: 1

      You are correct that the standard cell library is going to be proprietary to the fab and therefore out of our control. It will also be produced using proprietary synthesis and place and route tools, which is currently the only feasible way. That said, some people like efabless are using Yosys for synthesis on older process nodes (180nm+).

    7. Re:Where are those chips baked? by AdamHaun · · Score: 1

      But then I wondered -- what actually was the motivation for this all out Open Source SoC?

      There have been a few projects like this posted to Slashdot over the years. For some people, it's like climbing Mount Everest -- "because it's there". Some people want to extend the open hardware community down into chip design, possibly encouraging new start-up companies. (lowRISC seems to be in this category.) And some people think the semiconductor industry is a stagnant patent-choked wasteland in need of a Linux-style revolution. (These people are idiots, and do not know anything about hardware manufacturing or the semiconductor industry.)

      The big thing I don't understand is why they all want to make chips with high-performance CPUs with tons of modern peripherals. Okay, I do understand -- they want to run Linux on their product. But what's wrong with making an AVR clone? Surely it would be much easier and much cheaper to make an 8- or 16-bit CPU with a few low-end comm peripherals on an older process? Is making a fully open hardware Arduino somehow less of an accomplishment?

      Regardless, none of these projects have succeeded yet, because making an SoC is much harder than making software. From easiest to hardest, the main obstacles to making an open IC product are: development skills, design tools, prototyping and manufacturing costs, testing, and logistics. I'm not an expert on the full process, but I can try to give an overview:

      * Development skills: Designing high-quality digital integrated circuits, even with an HDL, is not trivial. You need people with EE or CE training, not just basic programming skills. Hardware has non-ideal behavior that must be accounted for in the design. It's also not cheap, so the design needs to (mostly) work the first time. This means you need real expertise, not just random volunteers. lowRISC has some experienced people (although not in IC design?) running it, and they're hiring a couple of EE Ph.Ds right now.

      * Design tools: You can edit an HDL with a text editor, but physical design and simulation require nasty, expensive, proprietary software packages from companies like Cadence or Mentor Graphics. These are not cheap, so you won't be running a copy at home. Your work will happen at an organization with money, like a university or corporation. lowRISC is a nonprofit associated with the University of Cambridge, so they can probably negotiate lower rates.

      * Prototyping and manufacturing costs: You can prototype an IC design on an FPGA, but large FPGAs are pretty expensive. Again, you'll need money for this, thousands of dollars per FPGA board at least. At some point, you'll want to make real hardware. You might be able to get a few prototype units for tens of thousands of dollars, but for real manufacturing in a modern process you'll need a proper mask set. This will probably be on the order of $500k. Small revisions (metal-only) will cost perhaps a tenth as much. If you need to move transistors around, you'll have to go through the physical design process and pay the full ~$500k again. An important side effect of this is that most design bugs will be fixed in the physical layout, not the HDL. This requires expertise, and you'll need to design for the possibility in advance. After all that, you'll have to spend thousands of dollars per wafer for manufacturing, plus more for testing (see below). lowRISC's nonprofit/academic status will help reduce these costs, and obviously they're getting funding from Cambridge and maybe their founders.

      * Testing: Even if your design is perfect (and it never is), you will need to test and qualify the hardware before you can sell it. This is where open IC projects fall flat, because nobody even talks about testing. You'll need to include test features like ATPG in your design, probably using even more expensive design tools. You'll also need to write test cases for functional verification, and generate test patterns for your automated test equipment (ATE). Once you've made the tests, you'll

      --
      Visit the
    8. Re:Where are those chips baked? by tlhIngan · · Score: 1

      As the blog in EE Times ("The Case for Free, Open Instruction Sets") argues, an ARM license costs $1M to $10M and takes 6 to 24 months to negotiate and then they take a small royalty per chip.
      http://www.eetimes.com/author....

      The proprietary instruction sets (ARM, IBM, Intel) have indeed evolved; that is not the problem. The problem is that you're not allowed to share implementations of the proprietary instruction sets with others. Thus, the lowRISC project is using a design from UC Berkeley for free without having to take the time or money to negotiate a contract, and they can modify it as much as they desire. Can't do that with ARM.

      A more hidden advantage is that a lower complexity chip is easier to fab.

      A modern ARM core that you get spending $1M would be easily a 6-7 Metal process, and masks aren't cheap ($100K each. And a 6-7 Metal would easily need 15-20 masks, so $2M outlay before the first chip is made).

      If the low complexity design can get away with a 4Metal design, that can easily halve the cost of just starting up. Plus if you use lower end technologies that they're about to retire instead of the latest and greatest, that'll cut fab costs down even more.

  5. Stallman by Anonymous Coward · · Score: 3, Funny

    Finally, he can upgrade his computer...

  6. Big names involved- RISC creator, rPi leaders by raymorris · · Score: 3, Insightful

    For those who didn't read TFS, the project is led by people with a track record of getting things done. One team member helped design, and named, the RISC architecture. Others are leaders of the Raspberry Pi project. That suggests these people know how to do this sort of thing successfully.

  7. Re:linux by Anonymous Coward · · Score: 3, Insightful

    Yeah. That's not good somehow.

    Based on the reported IPS of the RISC-V CPU emulator, it takes 1839 actual hardware instructions to cycle the RISC-V once.

    So? The speed of an emulator isn't a reflection on the quality of a CPU. A cycle accurate simulator is designed to function exactly like the real thing in simulated time, not real time. It doesn't reflect either way on the actual silicon.

    A better (but still poor) proxy for the performance of a CPU design on silicon is it's elaboration in an FPGA. The generator should be out real soon (I am not affiliated with the project, but it says on the github), then you can test it out on your FPGA board.

    Unfortunately without a $250k (I'm guessing) or academic license, you can't elaborate a Cortex A5 for comparison, and neither core design seems like it is optimised for FPGA, so once again a poor comparison, but at least a fair one.

    As a practical matter, a mask set at 40nm costs around $60k*, so an opensource SoC implemented on a closed source FPGA is still more open from a practical matter for me than an ASIC which I have to use expensive closed source tools to tape out and pay $60k* each time I run a "build".

    -puddingpimp

    * Maybe this number comes down through wafer sharing ala MOSIX.

  8. Forth? by Anomalyst · · Score: 1

    Is there a FORTH interpreter for it?

    --
    There is no right to feel safe thru security vaudeville at the expense of everyone's freedom, privacy and tax money.
    1. Re:Forth? by nurb432 · · Score: 1

      If there isn't after the project stabilizes, and you cant port it yourself, then you shouldn't be asking about FORTH in the first place.

      --
      ---- Booth was a patriot ----
  9. Re:Arms market by TechyImmigrant · · Score: 1

    For a clean sheet CPU design, ARM's instruction set is a horrible choice. It's carrying the bloat of 4 different architecture widths (24, 32, 16 and 64 in that order).

    I'm all in favour of people designing their own CPUs. It's not that hard and good ideas come from it.

    --
    I should use this sig to advertise my book ISBN-13 : 978-1501515132.
  10. $60k mask set for 40nm by Anonymous Coward · · Score: 1

    $60k for 40nm? Where???? I will place my order tomorrow.

    It is about $60K for (the very old and commodity) 180nm process, but there are other costs as well... still this is 'affordable'. 40nm is lots of $$$ for masks. Mosis at about $60k for a few 'TinyChip' (if they still call it at) parts in 40nm sounds about right though...

  11. Raspberry Pi was putting a chip on a board by YesIAmAScript · · Score: 1

    Well, securing supply of an already existing chip at a good price and putting it on a board.

    That doesn't relate to designing an entire SoC and getting it fabbed.

    --
    http://lkml.org/lkml/2005/8/20/95
    1. Re:Raspberry Pi was putting a chip on a board by thegarbz · · Score: 1

      Well, securing supply of an already existing chip at a good price and putting it on a board.

      That doesn't relate to designing an entire SoC and getting it fabbed.

      You mean securing supply of an existing chip, at a good price point, building exactly what people were asking, delivering it at the price people wanted, marketting it well to the right customers, building up the correct user base, and all in all selling 2.5million units.

      You can try and downplay their successes anyway you want, but the reality is that there are MANY SoC devices out there which are better than the RPi but none of them managed to do what the team did with that device.

      If you want someone with a good track record for delivering a project, or a track record for turning a simple thing that many others have done into a wildly successful ecosystem then the Raspberry Pi is a great example.

  12. Re:linux by ChunderDownunder · · Score: 1

    which probably says more about a JS-based emulator in a browser than the platform itself.

    No qemu support yet?

  13. Re:Hackers? by jpatters · · Score: 1

    Get a grip.

    You lost interest in this project because the summary of the slashdot post decribed them as "hackers"?

    FYI, a "hacker" is someone who finds uses for a technological item that were not intended/anticipated by the original inventor of that item. Not sure that really applies here, but it doesn't matter, because the wording chosen for a slashdot post summary should have zero impact on weather or not a project is interest-worthy.

    --
    "Remember, there never were pineapple-almond cookies here."
  14. Re:Hackers? by gtall · · Score: 3, Informative

    Ah, you youngins...its original meaning was anyone who whacked away at software or hardware, it had nothing to do with finding unintended uses or any other borderline technical behavior. What happened was Hollywood and the media picked up the term to apply to people for whom they had no name. Those people were originally called crackers. But Hollywood and the media couldn't tell the phonetic different between the two terms, hacker was easier to pronounce, and had few letters. Then some babies were born and now use the term in its present meaning.

  15. Real comparisons? by tomhath · · Score: 1

    Are the performance comparisons actual results? I don't see that they have a working chip yet. Plus they appear to have compared their 64bit design against an older 32bit ARM chip that's already in production. It wouldn't surprise me if the chip ends up close to what they are claiming but it still seems a bit premature to be making those claims.

    1. Re:Real comparisons? by asbradbury · · Score: 1

      There are comparisons made by the Berkeley RISC-V team made using their test chip. There was a good discussion on those numbers on the recent eetimes article that might interest you (http://www.eetimes.com/author.asp?section_id=36&doc_id=1323406&_mc=RSS_EET_EDT).

  16. Re:Hackers? by radl33t · · Score: 1

    the real point here is that your brain is so hard-wired by age that you cannot embrace the change or dynamism that drive human progress. it is time to wander off into the woods. alternatively accept that your opinions and thinking are outmoded and allow yourself to be reduced to some menial skill set until such time you can wander off into the woods.

  17. Real comparisons? by Caelius · · Score: 1

    I don't see that they have a working chip yet.

    Yes, it's real silicon. There are 8 silicon implementations so far (from Berkeley at least, not from LowRISC). - Berkeley RISC-V user.

  18. Re:linux by AdamHaun · · Score: 1

    As a practical matter, a mask set at 40nm costs around $60k

    Maybe one mask costs $60k, but a mask set? I think you left off a zero.

    --
    Visit the
  19. Re:Hackers? by Megol · · Score: 1

    Because we should all bow down to simplification and opinions of common people. Really? Go f*** yourself and then jump of a cliff.

    Myself and other sane people will keep calling a computer a computer and not a "CPU".

  20. Re:Hackers? by danknight48 · · Score: 1

    You lost interest in this project because the summary of the slashdot post decribed them as "hackers"?

    Pretty much.

    Slashdot Editors:
    "mm, what to call people who design and manufacture SoC chips. Nope, cant think of anything, lets just call them hackers because we can use it in anything, whilst looking cool!"

    Everyone else:
    "SoC designers and engineers"

    I'am with everyone else.