Comments Locked

14 Comments

Back to Article

  • ishould - Wednesday, June 29, 2022 - link

    Honestly I like TSMC's iterative approach. Much more reliable node enhancements, even if incremental, vs Intel's 10nm delaydelaydelay because they were aiming for too high a target.
  • Khanan - Wednesday, June 29, 2022 - link

    Yep, Intels aggressive approach already failed once and that’s why I trust TSMC approach more and think it is the better one. Semiconductor is a risky business anyway, no need to make it even more risky.
  • name99 - Wednesday, June 29, 2022 - link

    "Nonetheless, it is a bit odd to see that TSMC is taking a rather long road to backside power delivery."

    Only if you haven't noticed TSMC's pattern since forever!
    TSMC is all about
    (a) not promising more than they are sure they can deliver
    (b) cautious steps, not huge leaps

    Let's see Intel actually deliver on their "this time, really, we promise, we can deliver GAA and BPD in time and on budget, just you wait"...

    There's something bizarre about the tone of this whole article. We all know the first customer for N2 will be Apple, and while Apple would surely love increased density, they'd be OK with lower power and a SoC for that year whose primary feature is lower energy (just like the A15/M2). So Apple is the primary customer for N2 as just GAA, then the next year Apple (and followers) get N2E with BPD and another 20% or whatever density reduction.
    Something like:
    2022 (A16) new micro-architecture, N3
    2023 iterate (within much the same area), N3E
    2024 iterate (probably larger area), N3E
    2025 iterate (energy as primary concern, like A15) N2
    2026 new micro-architecture, take advantage of density boost from BPD
    Everyone's happy!

    This is (once again) the huge advantage TSMC has by having so many customers with so many different requirements. At least some can get substantial value from just GAA and the power reductions, without the density improvements of BPD.

    Intel seems to have learned nothing from the past, not even on the foundry side! Why go to the trouble of these "internal-only" nodes like i4 and GAA-only i20? If you take foundry seriously, why not open up these nodes to others? Is there no-one in the whole world but Intel who could see some value in a high-performance-cell only i4, or a GAA-only i20? I just don't get it. It's like half the company insists "we can be a full service foundry like TSMC, just look at our Tower offerings", while the other half of the company insists "being a foundry is so much hassle, cleaning up all the edges and corners for outsiders! let's just not bother"
    Compare with, say, AMZ. Of COURSE cleaning up all the edges and corners to get everything on micro-services that could be sold to 3rd parties was a hassle. But AMZ went all in, and thus success. INTC still seems unwilling to ACTUALLY go all-in with foundry.
  • name99 - Wednesday, June 29, 2022 - link

    A different aspect to TSMC's decision is to consider the next step:

    After just a generation or so of GAA we have a generation of forksheet transistors, then we get to CFET. The question is
    - do we deliver CFET as a separate sliver wafer of Si mounted on an underlying sliver wafer?
    - which is in turn mounted on an underlying wafer of BPD?

    On the one hand, OMG!
    On the other hand, if you bite the bullet and create a full infrastructure for handling and stacking these sliver wafers, you're in a pretty good position, going forward, for dense monolithic 3D fabrication...
    Whereas if your BPD is a one-trick pony that doesn't materially advance this larger (ten+ year) strategy, that seems a lost opportunity.
  • xol - Wednesday, June 29, 2022 - link

    I'd think there's more profit for Intel if the chips they produce are their chips, and not someone elses. The exception to this would be if they get a big process advantage over TSMC/Samsung and can charge through the roof to discerning customers ie Apple.

    Strongly suspect beyond that, the "we foundry too now" messaging stuff is mostly business politics. (and that they won't be completive price wise/have a ton of customers lining up)
  • xol - Wednesday, June 29, 2022 - link

    typo [competitive not completive]
  • Khanan - Wednesday, June 29, 2022 - link

    I agree, Intel foundry is odd, they should open it up.
  • Freeb!rd - Wednesday, June 29, 2022 - link

    "Why go to the trouble of these "internal-only" nodes like i4 and GAA-only i20? If you take foundry seriously, why not open up these nodes to others?"

    It seems to me that it is doubtful Intel has the EUV machine CAPACITY to offer it to foundry partners and also have volume for its own needs, yields may also be an issue. They use it internally until they get yields and volumes up and then allow foundry customers access to it. It still looks to me they won't truly be able to "catch up" to TSMC until the end of the decade.
  • whatthe123 - Wednesday, June 29, 2022 - link

    intel can't deliver those nodes because they're incomplete and don't have enough capacity. it's not rocket science, even if you're optimistic and assume intel has working next gen nodes, it doesn't mean they can deliver all of them at the volume necessary for foundry services.

    whats with illuminati like conspiracy theory's surrounding intel? they've yet to prove they can execute, people aren't going to jump in on their foundry services unless they're competitive or offer serious discounts, it's literally that simple.
  • dotjaz - Thursday, June 30, 2022 - link

    "Is there no-one in the whole world but Intel who could see some value in a high-performance-cell only i4"

    What are they gonna do with it? There's absolutely no standard library for anything.
  • Vitor - Wednesday, June 29, 2022 - link

    So it really seems silicon progress will be at a dead end by 2035, right? After N2 thinks will get so tiny and tricky that only a new material will be able to provide substantial improvements.
  • Khanan - Wednesday, June 29, 2022 - link

    Tech advancement will never stop, could be that it slows down but it’ll never stop.
  • DougMcC - Wednesday, June 29, 2022 - link

    Meh. It will be challenging to get more dense than Hydrogen fabrication at 31pm, in a 3d layered gas I guess? We're rapidly closing in on the fundamental constraints of the universe we live in.
  • nandnandnand - Wednesday, June 29, 2022 - link

    They could do tiny node steps into the late 2030s, or move to a roadmap that emphasizes 3D stacking improvements like what was done with V-NAND.

Log in

Don't have an account? Sign up now