Comments Locked

19 Comments

Back to Article

  • Blastdoor - Wednesday, June 29, 2022 - link

    The most effective way to encourage customers to switch is to raise prices on the old nodes. Maybe this is TSMC sending customers a heads-up that such price hikes are coming. If so, that's very chivalrous of them.
  • ingwe - Wednesday, June 29, 2022 - link

    With relatively few places, immediately going to a price increase is not a great way to do things. Even if TSMC is the 600 lb gorilla, it is better to work with companies as a partner.

    Of course it could be that they have already raised prices and are just having trouble keeping the older nodes running. If you don't have enough engineering resources to allocate to alternate part qualification, all the money in the world won't necessarily solve that problem.
  • astroboy888 - Wednesday, June 29, 2022 - link

    Funny thing is that they might have to. Given the level of mass expansion on 28nm capacity, it might be cheaper for TSMC to manufacture in 28nm than maintaining older 40nm or 90nm technology. 28nm yield must be high 90% by now, same as 40/90nm nodes.

    Or cut price on 28nm nodes when full capacity comes on line.
  • StevoLincolnite - Tuesday, August 16, 2022 - link

    Contracts may bind them to a fixed price.
  • Kamen Rider Blade - Wednesday, June 29, 2022 - link

    With increased Fab Waffer demands on older nodes & limited capacity, you can't expect TSMC to build out older fabs.

    What exists will stay there, but if TSMC is going to build out older nodes, it seems 28nm will be TSMC's process node choice.
  • Jorgp2 - Wednesday, June 29, 2022 - link

    Do they have a cost reduced version of 28nm?
  • astroboy888 - Wednesday, June 29, 2022 - link

    TSMC is expanding 28nm capacity in Taiwan. Building a new 28nm fab in Japan, and expanding the Nanjing 28nm fab in China. Cost is going to be reduced because of scale.
  • back2future - Wednesday, June 29, 2022 - link

    smaller nodes are tending towards shorter longevity, generally (>1-2[-3]decades \-\> <1decade?)
  • Duncan Macdonald - Wednesday, June 29, 2022 - link

    For many of the smaller integrated circuits moving to a smaller node (eg 90nm to 28nm) would not result in any increase in the number of working chips per wafer as the size of the chips is limited by the space needed for external connections (pins) and/or the power dissipation and/or the required working voltage. (As an example the circuitry of a 555 timer would probably fit inside a 5um square on a 28nm process - but the 8 connecting pins would constrain the chip to being much larger - the minimum bonding wire size is over 17um.)

    With a new process requiring new masks (not cheap), more testing (to ensure that the new design still meets all the requires parameters (especially voltage and current ratings) and possible regulatory approvals (if used in a medical or nuclear field) there is not likely to be a stampede to the smaller nodes.
  • noobmaster69 - Wednesday, June 29, 2022 - link

    Duncan, all of that is true, but there’s additional considerations as well.

    First, depending on what node a customer may be migrating from, the wafer itself may be significantly larger at 28nm than the older node. If so, the customer is going to get many more chips/wafer even if the chip size sees no reduction in size.

    Second, we know TSMC is raising pricing across the board. News this week was a 6% price hike across the board for next year on top of previously announced 25% price hikes on all ‘legacy’ nodes - including all those customers already on 28nm. This is going to help pay for the cost of expanding the 28nm production capacity and possibly helping defray the cost to customers having to redesign their chips and make new masks. There’s huge economic upside for TSMC to retire as many of the older legacy nodes as possible and reclaim fab space.

    Lastly, to the extent they can reclaim fab space by completely shutting down some older process nodes, they can repurpose the fab for leading edge nodes. The lithography equipment may be disposed of, but the fab itself - the water filtration, clean room, automation tools, etc - can still be used for leading edge nodes. That is a not an insignificant cost savings to expanding production at the leading edge, and the economic benefits of doing so makes subsidies to moving customers off older nodes very compelling. Eg spend $500m to help transition older customers off legacy nodes and make available $5B of fab space that would otherwise require new construction.
  • thestryker - Wednesday, June 29, 2022 - link

    There will also never be any expansion in capacity at the higher nodes. If there are currently backlogs on a node they will never get better unless people switch to newer nodes.
  • DanNeely - Tuesday, July 5, 2022 - link

    Never say never. the 50% increase in a node as old as 28nm is nearly unprecedented but being driven by surging demand for ICs that don't need anything close to leading edge fabbing; and IIRC other foundries are also increasing legacy capacity as a result of it too.

    And while TSMC might not do so, there are other companies who only do larger and simpler processes. If TSMC eventually does wind down some legacy processes entirely, companies who can't compete on the high end can be expected to pick up the demand from customers with products that can't physically or economically be moved to smaller nodes.
  • name99 - Wednesday, June 29, 2022 - link

    You don't have to move EVERYONE over, you just have to move the low-lying fruit over.
    If you have serious difficulty in moving stay on 40nm, but if you can move, so so, and free up space for the laggards.
    And if you are a laggard, well, what's your long term plan? To be selling year 2000 tech in 2040? Seriously???

    ULTIMATELY this will prove to be a story we have seen time and again.
    Some companies prefer to remain fat and lazy, sticking with what they know how to do. And that will be OK for a *few* more years.
    Meanwhile other companies can see which way the wind is blowing, will design from scratch to match new reality, will get the new regulatory approval, and will replace the dinosaurs.
    Just ask Intuit or Blackberry how well it plays out to ignore where the puck is headed...
  • SkipPerk - Wednesday, June 29, 2022 - link

    This might be true in “tech” fields, but in other sectors continuity is prized, especially if a new chip requires new regulatory approval. That said, too many companies make these transitions needlessly complex by updating everything (there is no need for software changes when shrinking to 28h).

    Many products do not require cutting edge, and in many fields the decline in product quality over the years often leads to older products being more valuable that the latest new release. Sometimes a chip simply needs to meet the spec, and nothing more. This is increasingly true for software as well (and why Linus is so much safer to use than anything Windows given how tricky getting an old version of windows can be)
  • FunBunny2 - Thursday, August 11, 2022 - link

    To be selling year 2000 tech in 2040? Seriously???

    battery powered autos are more than a 100 years old. same for gas and diesel. until the supply of Mr. Fusion power plants picks up, we'll be using them for another 100 years.
  • Duncan Macdonald - Wednesday, June 29, 2022 - link

    Small devices (eg power management ICs) get no benefit from being on a smaller node and many devices CANNOT be made on a smaller node due to the required voltages. (When you have insulator thicknesses measured in a few atoms there is a hard limit on the maximum usable voltage.)
  • evanh - Wednesday, June 29, 2022 - link

    Now's also a good time for said customers to be quietly raising their own prices to cover such a transition.
  • HarryVoyager - Saturday, July 2, 2022 - link

    One other factor is node rot. I've notice old node seem to eventually start to have yield problems. I recall one ancient chip we used that the manufacturer simply decided to stop making because their yield had taken a nose dive.

    The chip itself was stupidly cheap, and the replacement was stupidly cheap, but the ripple effects were a massive pain. Had different voltages and different physical layout, so we had to respin a PCB and write new firmware to deal with it. Ended up stupid expensive. Would have cheerfully taken a last time buy at way more than the original price, but the process just futzed out, so redesign ho we had to go.

    I suspect consolating on a single node means they can have some additional staff focused on making sure that mode keeps working, as opposed to spreading their crew out over a dozen or so similar but different processes.
  • TomWomack - Saturday, July 9, 2022 - link

    Unless TSMC can provide some really quite elaborate level-shifting circuitry in the space that's freed up by moving to a newer node, there's going to be trouble producing a 28nm chip that has the same footprint and voltages as a 65nm one.

    And I think by 28nm you're using multiple patterning and have really rather tight design rules; you can't just put the 65nm masks into a 28nm stepper and have them projected with really crisp corners.

Log in

Don't have an account? Sign up now