Forgot your password?
typodupeerror
Businesses Intel Hardware

ARMs Race: Licensing vs. Manufacturing Models In the Mobile Era 54

Posted by timothy
from the everyone-knows-salsa-works-better-on-chips dept.
MojoKid writes "The semiconductor market for mobile and hand-held devices has changed dramatically in the past six years and ARM has had to evolve along side it. ARM's IP focus allows it to dedicate all its resources to building a great design rather than committing to any single manufacturing process node, customer, or foundry. Architectural design and implementation is done very much in partnership with both foundries (TSMC, GlobalFoundries) and licensees like Samsung or Qualcomm. The difference between the way Intel goes to market and ARM's model is more nuanced than the simple ownership of manufacturing facilities. Owning its own fab means that Intel can tweak process technology to match the particulars of a given architecture (and vice-versa). It also gives the company far more flexibility when planning future nodes. If Intel feels that integrating Peanut Butter Silicon on Insulator (PB-SOI) is the best way to hit its performance and power consumption targets at 14nm, for example, it can make that happen internally. ARM, in contrast, is limited by the decisions of the foundry manufacturers it partners with."
This discussion has been archived. No new comments can be posted.

ARMs Race: Licensing vs. Manufacturing Models In the Mobile Era

Comments Filter:
  • Intel's ARM license (Score:5, Interesting)

    by Anonymous Coward on Thursday July 04, 2013 @05:29AM (#44186561)

    intel still retains the royalty-free license it obtained from ARM when ARM were running out of cash. the deal was that intel would feed back any improvements made. unfortunately, just as my associate told ARM when he was working for LSI Logic - advice which ARM completely ignored - the designs ARM had at the time were so poor that intel was forced to use their *own* super-scalar Harvard architecture and to put an ARM-compatible front-end on it. intel never gave back any modifications to the design.... because they never made any. ARM were pissed, Intel were embarrassed at having a non-x86 SoC that outperformed both their own cores *and* ARM's, so sold it to Marvell... *minus* the royalty-free license. Marvell had absolutely no qualms about out-performing ARM and immediately ramped it up to 1ghz.

    • by Anonymous Coward

      You're talking about StrongARM, right? I used to have a RiscPC souped up with one of those. A 200MHz ARM chip, when PCs were running up to 120MHz Pentium at the time, was pretty bloody awesome.

      It's hardly surprising that Intel could do a better silicon design than ARM. But if they took ARM's ISA alone, and managed to outperform their own flagship product, it's not exactly clear which company counts as having "poor design" :)

    • by drinkypoo (153816) <martin.espinoza@gmail.com> on Thursday July 04, 2013 @08:20AM (#44187075) Homepage Journal

      ARM were pissed, Intel were embarrassed at having a non-x86 SoC that outperformed both their own cores *and* ARM's, so sold it to Marvell... *minus* the royalty-free license.

      Intel was embarrassed that their ARM core had the highest power consumption on the market, and that they had no idea how to reduce it except a process shrink that wouldn't come soon enough.

    • by MrNemesis (587188)

      Wasn't the StrongARM (IIRC Intel's only dabble in the ARM world, which begat XScale, which was then sold to Marvell) originally a DEC creation? DEC was one of the earliest partners for ARM not long after Apple came out with the Newton.

      • Yeah, when DEC settled its lawsuit w/ Intel over the violation of Alpha patents, part of the settlement was that the Hudson fab, as well as the StrongARM were sold to Intel. Compaq sold the rest of their semiconductor IP to Intel much later.

        One thing I don't get is that DEC was one of the MIPS licensees as well at the time - their early Ultrix workstations were based on the MIPS R3000s. So couldn't they have taken that and made designs that would be low on power consumption? Especially once MIPS switch

    • by slew (2918) on Thursday July 04, 2013 @09:53AM (#44187661)

      There are 3 from scratch Arm designs where you are talking as if there were 1.

      StrongARM: DEC's original ARMv4 compatible designs (SA-110, SA-1100). DEC sold the design/business/name to Intel who sold if for a while and made an upgrade (SA-1110), but later came up with...

      XScale: Intel's ARMv5 redesign which had 3 generations (PXA1xx..3xx) and some of their own instruction set instructions (like wireless MMX). Intel sold the business/name to Marvell who sold chips manufactured by Intel, but Marvell already had...

      Feroceon: Marvell's own superscalar ARMv5/v6/v7 compatible design that they used as an embedded processor for their Ethernet and Storage business(88Fxxxx, and later PXA9xx). Marvel is in the process of transitioning all of the old Intel business to their own Arm Core (w/ the PXA9xx)

      Also, as an architectural licensee (like Intel and Marvell), there is no requirement to "give-back" improvements to Arm. In fact, as an architectural licensee, you can't actually start-with/use/modify Arm's designs. You must design your own from scratch and it must pass their compatibility test suite (earlier architecture tests for v4/v5 allowed for instruction set extensions, but later tests for v6+ do not), but that is all.

      On the flip side, if you are a regular licensee, you must use the design Arm gives up pretty much As-Is (although you can make timing fixes and ram wrappers and similar adaptations that don't change the functionality). Although you can request Arm to make modifications for you, they are free to share these modifications with other licensees...

      The main reason Intel sold Xscale (for which they had to pay Arm a royalty and to which they couldn't add new functionality) to concentrate on pushing x86 (free from royalties and free to innovate) into the mobile space. Time will tell if this was a good move (although Intel retained their Arm license in case they want to reverse course)...

      • if you are a regular licensee

        Out of curiosity, what do you get w/ a "regular license"? Is it synthesizable RTL, netlist, or what?

        • Pretty much a regular ARM license allows you to use the core as-is. You may put things around the core in a SOC to modify the frequency for example but the core is pretty much identical for every regular licensee. Initially nVidia's Tegra used an ARM11 core (1136) which was the same core as TI's OMAP2. The main difference was the Tegra used GeForce ULV as the GPU while OMAP2 used PowerVR. There were other differences in the SoC.
    • by expatriot (903070)

      DEC StrongArm was released as Intel XScale but was then was sold to Marvel (http://en.wikipedia.org/wiki/XScale)

      According to reports, Intel has an architecture licence but does not intend to use it. http://www.zdnet.com/blog/computers/intel-we-have-arm-license-no-plans-to-use-it/5845 [zdnet.com]

      Intel also brought Infineon which had an ARM licence (http://www.gomonews.com/intels-1-4-billion-infineon-mobile-chips-purchase-2g-3g-lte-wimax/) but I don't know what the status is of that going forward.

      The comment above ra

  • by dltaylor (7510) on Thursday July 04, 2013 @05:45AM (#44186599)

    Spend some time looking through the Linux kernel archives, or actually USING one, and you'll see that quite the opposite is true.

    What was a tolerable architecture for low-complexity embedded designs has serious flaws in cache coherency and clock-for-clock is painfully slow compared to a 464 (or, even, 440) PowerPC.

    Because the ARM lacks the useful complexity for cache coherency and memory, and memory-mapped IO, barriers, and a quite small page table entry cache, it does have a power consumption advantage over the PPC, though.

    Maybe (hopefully, really) the 64-bit versions won't be quite so crippled.

    • What does that have to do with the ARM ISA? If you find that the existing ARM processors lack a useful implementation of cache coherency, design an ARM implementation that does have it, or pay for one. I'm pretty sure that the 80386 sucked in roughly the same manner.
      • by slew (2918)

        The problem is that since cores available from ARM don't have a usable I/O cache coherent mechanism, and nearly all users of ARM (except Apple, Marvell, and Qualcomm) use cores from ARM, that means most platforms will necessarily have this flaw.

        However, in actuality, all platforms have this flaw because since ARM didn't do it, Apple, Marvell, and Qualcomm implemented the same crappy cache coherency since all the device drivers written for Arm architectures assume that is the standard and any expensive coher

    • ARMv6 had the option to be cache coherent between cores. ARMv7 adds optional coherent bus support. Most manufacturers I have seen opt for just coherency between the cores which means you still have to flush cache by address for some DMA operations.
  • by YoopDaDum (1998474) on Thursday July 04, 2013 @06:35AM (#44186733)

    Owning its own fab means that Intel can tweak process technology to match the particulars of a given architecture (and vice-versa)

    That may be understood as an Intel exclusive, but it's not entirely true. Even in the fabless world the big shots (Qualcomm, NVidia, AMD & co) have very early access to new process nodes and can certainly tune their design to it, and have their own specifics tweaks made. So they can do both kind of adaptation too, although it's not as integrated as for Intel. If you draw a line, Intel is at one extreme being able to have close integration, the small fabless companies are at the other extreme taking the stock TSMC or GF or UMC or else offering as-is. But the big fabless guys are somewhat in the middle.

    ARM, in contrast, is limited by the decisions of the foundry manufacturers it partners with.

    It's also a bit misleading. ARM has early access to all big fabs (Globalfoundries and TSMC), and because ARM is so pervasive there is a very very high pressure for a fab to provide the best ARM implementations on their process. So sure, it's the fab making the decisions on their process in the end. But you can bet they will pay a lot of attention to any ARM feedback gained during the early access co-work.
    ARM doesn't only provide processor IP, they do the whole range now from memory cells to GPUs to interconnect to memory controllers. And they work with the fabs to optimize their design for them and provide their customers "Process Optimization Packages" (POPs) that summarize how to get the best of a process for their IP. So ARM has the know-how, the access and the pull to have a big say in what happens in the fabs roadmaps.

    • by slew (2918)

      So ARM has the know-how, the access and the pull to have a big say in what happens in the fabs roadmaps.

      However, ARM doesn't have the focus to do the best job on every fab and process. A singular company can focus to get the best out of a single fab and process, where ARM needs to balance their resources (and they aren't a huge company to start with ~2K employees vs ~100K for intel)

    • by fermion (181285)
      Also look at the travails of Apple to get microprocessors over the past 30 or 40 years. They started with MOS, went to Motorola, then IBM and then Intel, and now custom designed chips for mobile devices. These changes were made because the firms that "could do everything" could not make the chip that Apple needed. When one is just a customer, even a major customer that pays for special services, one has to compete for services and compromise based on the standard product.

      So Apple now makes it own chip t

      • by Dcnjoe60 (682885)

        So the issue is not who manufacturers where, but who can deliver the microprocessor that a firms need to run what are increasingly embedded devices(which is why the silly argument about 'can I upgrade' is becoming increasingly silly, it's like can I upgrade the inkjet in my printer to a laserjet, no one cares). What is all too soon going to be legacy laptop and desktop may require a traditional processor, it is unclear if Intel can survive in such an environment, while ARM, and maybe to a lesser degree AMD, should thrive.

        ARM beats Intel in two areas, power consumption and price and sacrifices raw computing power to achieve it. Intel can easily throttle back any of their chip lines to reduce power consumption and they have so much spread between the selling price and the cost to manufacture that price is not actually a barrier, either.

        Intel's biggest problem is the same that IBM had in the 80s and that is with lower end products cutting into profit margins on higher end product lines (the old PC vs Min vs Mainframe). Ultima

      • by evilviper (135110)

        AMD are not as likely, if not more likely,

        You're a politician, aren't you?

        I noticed when my head started hurting reading what you wrote, and by the fact that you wrote several paragraphs that, at the end, say absolutely nothing...

    • by Dputiger (561114)

      " So ARM has the know-how, the access and the pull to have a big say in what happens in the fabs roadmaps."

      As the author:

      I've talked to a number of people about this, on all sides of the equation: Foundry folks, Intel engineers, ARM, etc.

      The universal consensus is that yes, Intel has an integration advantage. Even ARM admits that. Being able to directly tie silicon and process implementation to SoC architecture is a real benefit. On top of that, Intel has a technology advantage.

      That said, there's a *reason

    • have very early access to new process nodes and can certainly tune their design to it, and have their own specifics tweaks made

      What sort of tweaks might ARM or another outfit want made to a new process? I know there is the basic "low speed but low static current" vs. "high speed but high static current" tradeoff, but it sounds like you're talking about more detailed things.

      • It's "process" in a very general meaning here. There is a whole stack before getting to a physical embodiment on the silicon die from the design RTL. Small clients will just use the approved tools (from Cadence, Synopsys, Mentor...) with ready made libraries of cells/components (many providers, and ARM is one). The big guys have the man-power to define their own cells based on their own trade-offs to get higher performance (at the price of larger area / higher power) or lower power depending. They can also
  • "allows it to dedicate all its resources to building a great design rather than committing to any single manufacturing process node, customer, or foundry"

    Being the old cynic that I am, I feel tempted to say that we owe this to the design process by Lynn Conway and Carver Mead, not to ARM Holdings.

    • by slew (2918)

      Being the contrarian I am, I feel tempted to say that we really owe the ability to create a design independent of a single manufacturing process node to the standard-cell asic design process created by VLSI and LSI-Logic and logic synthesis tools (rather than the lambda scalable design rules of Mead-Conway for custom logic which is not really used at all any more).

      In the old days, portability was merely done using analog parameters (scaling transistors in a fixed circuit topology). Now days, the design is

  • by khakipuce (625944) on Thursday July 04, 2013 @07:32AM (#44186909) Homepage Journal

    Owning the Fab is a big capital cost and means the beast has to be fed but allows for more late-stage tweaking. Not owning that Fab means potentially more flexibility in choosing a process for the design. Of course ARM ahs to work with it's partners, that doesn't mean it's partners are in-flexible and far from cutting edge.

    Seems like no great news, just the same old in-house vs. out-sourced debate...

    • by Dcnjoe60 (682885)

      Owning the Fab is a big capital cost and means the beast has to be fed but allows for more late-stage tweaking. Not owning that Fab means potentially more flexibility in choosing a process for the design. Of course ARM ahs to work with it's partners, that doesn't mean it's partners are in-flexible and far from cutting edge.

      Seems like no great news, just the same old in-house vs. out-sourced debate...

      That is only partially true. Yes owning the Fab is a big capital cost, but Intel Fabs chips for others, too, so that capital cost is borne not only by their products but others, too. In addition, they are not paying the built in profit that would be required if they out-sourced the Fab and finally, they get significant tax write-offs for depreciation. So, yes, there is a big capital cost, but there are also big real world benefits besides being able to have more late-stage tweaking.

    • by gl4ss (559668)

      you talk about choosing a process like arm is the party doing the choosing.

      let me tell you how the process goes: mr. manufacturer goes through his quarterly reports and chooses the process they happen to have reliably running at the moment.

  • by Anonymous Coward on Thursday July 04, 2013 @08:21AM (#44187077)

    Looking at it objectively, the ARM ecosystem benefits from a diversity of process technologies funded by its customers, whereas Intel needs to get its process bet right every generation. And even to the extent that its graphics cores allow it to claim that it is used to producing SoCs, not just CPUs, it's still just one supplier -- it can't compete on customisation -- at best it could offer a few standardised SoCs.

    Intel will suffer from margin compression by entering the mobile space (and the more chips it sells there the worse that problem will get) while simultaneously seeing increasing ARM competition in servers shave its margins there too. Intel may not have the money much longer to stay a generation and a half ahead in process tech.

    • by TheRaven64 (641858) on Thursday July 04, 2013 @08:37AM (#44187151) Journal
      Intel does currently have two big advantages over ARM. The first is that they are typically a generation (or, at least, half a generation) ahead of other fab owners. This is the same advantage that allowed them to compete with superior designs from AMD for the late '90s: they could get higher clocks at the same power, and more chips per wafer (meaning lower cost). The other advantage is that they are a much larger company and so can afford to make half a dozen guesses about the state of the market at the end of the five-year chip design cycle. They can then have different teams working on chips for those market predictions and only actually ship one or two of the final microarchitectures. In contrast, every ARM chip has to be a success, often for several generations (for example, the new A7 is an A8 with a number of refinements and a lot of tweaking for optimisation and some slight ISA tweaks to make it instruction-set compatible with the A15).
      • Re: (Score:2, Insightful)

        by Anonymous Coward

        And ARM has one huge advantage over Intel - everybody else except Intel has their own ARM SOC these days, so designers can shop around to get the chip they want, from the supplier they want. And they can easily switch to a different supplier without having to throw out their entire investment into the architecture.

        That being said, I cannot see the GPs world where ARM seriously threatens Intel in the server/high performance computing market, where Intel holds most of the cards, not just the aces.

        • And ARM has one huge advantage over Intel - everybody else except Intel has their own ARM SOC these days, so designers can shop around to get the chip they want, from the supplier they want. And they can easily switch to a different supplier without having to throw out their entire investment into the architecture.

          Not really. The ARM part is pretty similar currently between all of them. With the exception of Marvell's line, they're all either stock ARM designs or slight tweaks (e.g. the improved floating point pipeline on Qualcomm's offerings). If one SoC vendor offers something better than another, it will typically be in terms of other cores on the SoC, and these are generally not compatible between vendors.

          This is something that ARM is trying to fix with ARMv8. They intentionally delayed the release of the

        • by gl4ss (559668)

          And ARM has one huge advantage over Intel - everybody else except Intel has their own ARM SOC these days, so designers can shop around to get the chip they want, from the supplier they want. And they can easily switch to a different supplier without having to throw out their entire investment into the architecture.

          That being said, I cannot see the GPs world where ARM seriously threatens Intel in the server/high performance computing market, where Intel holds most of the cards, not just the aces.

          and as said a million times even if they did come up with the end of all super killer arm design that scales up in multi ghz.. intel could just start making those. sure, some x86 investment would be down the drain but they wouldn't be out of business like amd..

        • by Dcnjoe60 (682885)

          And ARM has one huge advantage over Intel - everybody else except Intel has their own ARM SOC these days, so designers can shop around to get the chip they want, from the supplier they want. And they can easily switch to a different supplier without having to throw out their entire investment into the architecture.

          That being said, I cannot see the GPs world where ARM seriously threatens Intel in the server/high performance computing market, where Intel holds most of the cards, not just the aces.

          What makes you think that Intel doesn't have ARM designs that they could turn around and Fab? Just because they don't have a commercial product that they sell does not mean they don't have something on the back burner that they could bring to market if they felt the need to do so.

          • by Pulzar (81031)

            What makes you think that Intel doesn't have ARM designs that they could turn around and Fab? Just because they don't have a commercial product that they sell does not mean they don't have something on the back burner that they could bring to market if they felt the need to do so.

            If they did, wouldn't that be exactly what ARM wants? ARM doesn't care who makes the ARM server chips, it gets paid either way. And having Intel develop ARM server ecosystem so that other partners can jump in would be downright fan

            • by Dcnjoe60 (682885)

              What makes you think that Intel doesn't have ARM designs that they could turn around and Fab? Just because they don't have a commercial product that they sell does not mean they don't have something on the back burner that they could bring to market if they felt the need to do so.

              If they did, wouldn't that be exactly what ARM wants? ARM doesn't care who makes the ARM server chips, it gets paid either way. And having Intel develop ARM server ecosystem so that other partners can jump in would be downright fantastic for them.

              Intel would have to be close to bankruptcy to do something that desperate.

              But Intel doesn't care if ARM gets paid. Intel cares that Intel shareholders get paid. If they can make a return on producing ARM chips that meats their internal ROI requirements, they will produce ARM chips. Intel is interested in profits, it doesn't matter what chip line it comes from.

  • That's a new one on me. Never even heard of it before.
    • by Dputiger (561114)

      An AMD staffer actually came up with it as a joke about ten years ago. It's become my go-to when discussing a concept. It's clearly ridiculous (so no chance of either leaking an actual technology or sowing FUD by appearing to leak something that isn't happening), but it refers back to real technologies as an anchor.

      We have partially depleted and fully depleted SOI (PD-SOI and FD-SOI respectively) so PB-SOI is a good fit. ;)

  • by turgid (580780) on Thursday July 04, 2013 @03:09PM (#44189735) Journal

    Why pay royalties to ARM when you can have an OpenRISC [wikipedia.org] for free?

    OK, ARM is way ahead of OpenRISC just now, but would it be justified to make an analogy between ARM and OpenRISC like Linux in its infancy compared with Windows NT 3.1?

    I'm not familiar with CPU development (yet) but I've met people who have worked on commercial projects that used it instead of ARM to save on licensing costs, and although it's only suitable for low-end projects at the moment, it works well enough. With some more development, it could be very successful.

  • Spend some time looking through the Linux kernel archives, or actually USING one, and you'll see that quite the opposite is true. What was a tolerable architecture for low-complexity embedded designs has serious flaws in cache coherency and clock-for-clock is painfully slow compared to a 464 (or, even, 440) PowerPC. Because the ARM lacks the useful complexity for cache coherency and memory, and memory-mapped IO, barriers, and a quite small page table entry cache, it does have a power consumption advantage

It's later than you think, the joint Russian-American space mission has already begun.

Working...