TSMC, ARM team up on 64-bit chips with non-planar transistors

Unlike Intel, vendors of ARM-based SoCs can’t yet reap the benefits of 64-bit addressing and “3D” transistors. But soon, they will. Oh, yes. TSMC and ARM say they’ve joined forces in a new, multi-year partnership, which will involve the manufacture of 64-bit ARM processors on sub-20-nm fab processes with fancy new FinFET transistors.

Now, ARM isn’t a direct client of TSMC’s. Rather, ARM develops intellectual property, which system-on-a-chip makers like TI, Qualcomm, and Nvidia subsequently license and integrate into their SoC designs. Those companies then enlist the services of independent foundries like TSMC for manufacturing.

The freshly announced ARM-TSMC partnership is all about facilitating that process, so that ARM’s newest CPU architecture can be more easily built into real products. Or, in the words of the two companies:

The collaboration will enable sharing of technical information and feedback between the two companies, enhancing the development of ARM IP and TSMC process technology. ARM will leverage process information to optimize the power, performance and area (PPA) of the overall solution to reduce risk and encourage early adoption. TSMC will use the latest ARM processors and technology to benchmark and tune advanced FinFET process technologies. The combination of TSMC FinFET technology and ARMv8 architecture provides the fabless industry with solutions for continued innovation across diverse market segments. The collaboration will result in improved silicon process, physical IP and processor technology that together will enable new system-on-chip (SoC) innovation and shorten time-to-market.

FinFETs are a type of non-planar transistor similar to Intel’s tri-gate transistors. (FET stands for field-effect transistor, and FinFETs get their name from the fin-like structure of the conducting channel.) ARMv8, meanwhile, is the 64-bit successor to the ARMv7 architecture. The Cortex-A9 core found inside Nvidia’s Tegra 3, Apple’s A5, and a number of other, high-profile SoCs is based on the 32-bit ARMv7 architecture.

According to EE Times, this partnership won’t bear fruit until the second half of 2015. That’s when the site says TSMC will ramp its 16-nm CMOS process. Of course, it’s worth stressing that Intel already offers 64-bit chips with 3D transistors today—the Ivy Bridge family. I’m sure it won’t be long before we see some low-power Atom SoCs built using the same 22-nm tri-gate transistors, too.

Comments closed
    • Martian
    • 7 years ago

    [i<]"I'm sure it won't be long before we see some low-power Atom SoCs built using the same 22-nm tri-gate transistors, too."[/i<] The Atoms they try to sell for smartphones maybe. They always squeeze the last cent out of every process tech, Atoms for PCs, netbooks, tablets etc. will not see that tech for a good while..

      • NeelyCam
      • 7 years ago

      If they go through the trouble to design a fast, efficient Atom chip in 22nm, why not use it in netbooks/tablets/etc?

      To me it almost seems like Intel is now “squeezing the last cent” out of a process with server products (that have pretty much zero competition now)

    • HighTech4US2
    • 7 years ago

    Nvidia’s Project Denver is also 64 bit and will be released sometime in 2013. That would put it on 28nm first with future versions on 20nm.

    [url<]http://www.xbitlabs.com/news/cpu/display/20111110230832_Nvidia_Project_Denver_Is_On_Track_Does_Not_Interfere_with_ARM_s_Own_64_Bit_Tech.html[/url<] So Nvidia will have at least a year or two jump on 64 bit ARM competitors.

    • HisDivineOrder
    • 7 years ago

    TSMC will be late.

      • fredsnotdead
      • 7 years ago

      Now there’s a safe prediction.

    • ronch
    • 7 years ago

    So basically ARM is putting all their eggs in one basket. If TSMC doesn’t deliver then everyone in the ARMs race is screwed. Surely any other foundry would love to work closely with ARM on something similar because they can then court ARM licensees touting their capability to do what TSMC can, probably at a lower price. I wonder what the other foundry guys have to say about this.

      • Airmantharp
      • 7 years ago

      Like UMC? Are there any other foundries taking orders that are up to date with those two?

        • kalelovil
        • 7 years ago

        Samsung. Globalfoundries.

      • BobbinThreadbare
      • 7 years ago

      Where did you get one basket from? Was there another press release I didn’t see where they said ARM companies are not allowed to use any other fab?

        • NeelyCam
        • 7 years ago

        The point is that it takes a lot of design effort to make an efficient, optimized ARM core in a Restricted Design Rules process (like the TSMC 16nm FinFET is probably going to be).

        Processes from various foundries are likely to be different, so if ARM wants to offer efficient, optimized IP in all foundry processes, they’ll have to have multiple, individually optimized versions of the core, and the resources to design all of those. It’s more cost effective to just make one super-optimized version (with the largest foundry, of course), and have other foundry versions sort of sub-optimized.

        It’s the right move overall; ARM knows they need process/design co-optimization to be able to compete with Intel in the future. And if you can afford to do it with only one foundry, TSMC is the obvious choice, as that’s where most of the customers are.

          • BobbinThreadbare
          • 7 years ago

          Ok, I still fail to see how this is one basket. Nothing is stopping Samsung or TI from doing their own R&D and coming up with their own way to fab chips.

            • NeelyCam
            • 7 years ago

            True, but without ARM collaboration, Samsung process won’t get optimized for ARM cores.

            Oh, and TI doesn’t develop its own digital process technology anymore; they use foundries just like Qualcomm and NVidia

        • ronch
        • 7 years ago

        Everyone will probably benefit and want this manufacturing technology baked into their chips unless they’re selling cheap ARM-based products. So, my point being, if TSMC doesn’t pull it off (and they’re past track record hasn’t been exactly amazing), ARM devices using this thing will not see the light of day, not in 2015, at least.

    • derFunkenstein
    • 7 years ago

    Intel doesn’t even have their entire desktop/mobile lineup on 22nm yet (not that there’s much reason to at this point) and won’t have a 22nm server/workstation product for a year. So let’s not go predicting they’ll have Atom SoCs “before long”.

      • codedivine
      • 7 years ago

      Well Intel has said that 22nm atoms will come out next year. They have said previously that they now intend to be very aggressive with Atom rollouts on newer process technologies. 32nm in 2012, 22nm in 2013 and 14nm in 2014 (with Atom being one of the lead products with 14nm). I remember them calling it “Faster than Moore’s Law”.

        • derFunkenstein
        • 7 years ago

        well that would be nice for sure.

        The minus was an accident, sorry. I guess the screen on the Nexus 7 isn’t as good as I thought or I have bad aim. Sorry about that.

      • NeelyCam
      • 7 years ago

      Intel doesn’t have to have everything on 22nm; those 32nm fabs are running cheap and chips are still more than competitive in, say, servers or desktop.

      Meanwhile, mobile chips can really use the extra help from 22nm trigate; I would guess that they are focusing on that – getting a 22nm Atom chip out as soon as possible… and they have publicly stated that as a goal.

    • blastdoor
    • 7 years ago

    Smart.

      • NeelyCam
      • 7 years ago

      Co-design of products and process is a thing that Intel has trumpeted as their key advantage against foundry customers. This AMD/TSMC collaboration should level the playing field a bit

        • dpaus
        • 7 years ago

        [quote<]This AMD/TSMC collaboration....[/quote<] Freudian slip, Neely? Actually, I'm glad that AMD isn't a primary partner in this; they have too any other balls in the air right now, and they'll benefit from the results anyway, since TSMC will need to market any resulting technology to all takers to cover the investment. Intel has a culture of partnership (too often leading to acquisition, but that's another story), as does ARM. TSMC? Not so much...

        • ronch
        • 7 years ago

        [quote<]This AMD/TSMC collaboration...[/quote<] AMD? Edit - Overlooked dpaus' earlier comment. It's (AMD instead of ARM) too easy to spot.

          • NeelyCam
          • 7 years ago

          Oops.. iI blame the tablet keyboard. This is why we need Ultrabooks with actual keyboards

    • chuckula
    • 7 years ago

    That’s great but bear in mind that 2H 2015 is the estimate that [b<]TSMC[/b<] is giving. Their past track record of predictions have been... shall we say... optimistic when it comes to getting processes into widespread production. Go ask Qualcomm if you don't believe me.

      • NeelyCam
      • 7 years ago

      Good point. And Intel will have 15nm products out in 1H/2014 (and they don’t miss their targets), so TSMC is still about two years behind Intel.

      UMC might have a golden opportunity to take on TSMC now if they are in fact going to FinFET at 20nm

        • blastdoor
        • 7 years ago

        My impression is that Intel does not typically use their latest fab technology for the types of chips that compete with ARM chips. But perhaps this will change?

          • bcronce
          • 7 years ago

          “But perhaps this will change?”

          Probably not. There is more of a margin on desktop/server chips than low end SoC.

            • NeelyCam
            • 7 years ago

            I don’t know… Intel is looking for ways to grow beyond the PC market (especially since PCs are dead, right?). Cell phones and tablets are the big growth areas, and there power efficiency matters a lot..

            I think it would be a smart move for Intel to put a ton of effort into developing uber-efficient smartphone chips leveraging the best process available, and they should do it [i<]now[/i<]. They have roughly a three-year window before others get their FinFET processes online.. clock is ticking tick tock tick tock

            • TO11MTM
            • 7 years ago

            Actually, Tick-Tock INDEED! I’ve read in other places that Intel has beten planning to put Atom on Tick-Tock later this year… this article is a bit old but it seems to go with that reasoning; I’ve seen it elsewhere but am not in the best position to look at the moment…

            [url<]http://liliputing.com/2011/05/intel-atom-chips-to-hit-14nm-by-2014.html[/url<] I bet some of this may be contingent on how well Medfield starts off in the market. If there's interest there will be a LOT more pressure to use an aggressive process, not just for power consumption but also for cost.

        • Unknown-Error
        • 7 years ago

        I thought Intel’s next node is 14nm and not 15nm?

      • blastdoor
      • 7 years ago

      The biggest challenge for all chipmakers not named Intel is a lack of access to adequate capital. Now, if TSMC could only find a company with vast financial resources and a willingness to pre-pay for the construction of an entire fab…

        • nanoflower
        • 7 years ago

        And a willingness to lose their investment if things don’t go right.

Pin It on Pinterest

Share This