back to article My chemical romance drowns tomorrow's money, warns TSMC: Chip maker's yields rocked by bad batch

Taiwan-based chip maker TSMC said on Monday that its chip production has been adversely affected by a bad batch of chemicals at a manufacturing facility that provides components for other chip makers. In a statement obtained by Nikkei Asian Review, the biz said, "TSMC has discovered a shipment of chemical material used in the …

  1. Voyna i Mor Silver badge

    First sign of a downturn

    When downturns start, people stop replacing expensive stuff that is good enough.

    Phones and laptops for two.

    It's noticeable around here that few houses are being sold but an awful lot of people are building extensions. Partly this is stamp duty but I think partly it's improve what you've got rather than embark on the choppy waters of a move.

    1. a_yank_lurker Silver badge

      Re: First sign of a downturn

      More likely a maturing markets for phones is the most likely cause of the slow down. As the market matures the replacement cycle often lengths as the newer devices only have more bling rather than useful features. So a phone that got replaced every couple of years might not get replaced for 3 or 4 years. Laptops and desktops the cycle is even longer.

    2. Anonymous Coward
      Anonymous Coward

      Re: First sign of a downturn

      I'm not sure this is a sign of a down turn - TSMC were likely to have their best year on record due to being the first mainstream producer of 7nm components while their competitors lagged or faltered and demand for their products was at an all time high following significant investments in new processes in 2018 (and previous years).

      1. jmch Silver badge

        Re: First sign of a downturn

        Not sure that's a sign of a downturn either... but a downturn is overdue. World economy used to run on longer cycles of 20-30 years but the cycle is shortening to about 10 years. Last one was 2007/08.

        China growth is slowing down, US got a one-off artificial boost with Trump's tax breaks last year (that it's clear are not affordable even in the medium let alone long term), and EU is certainly going to be impacted by Brexit.

        1. Alan Brown Silver badge

          Re: First sign of a downturn

          "US got a one-off artificial boost with Trump's tax breaks last year "

          Cycles aside, every time the USA has elected a populist president (twice in the 19th century and once in the 20th) there was a short term uptick in the economy and then a crash with prolonged recession.

          The 1870 crash/recession affected Europe so badly that it stayed in recession whilst the USA went through another boom/bust cycle with a crash in 1890. That didn't ease until 1903. Hoover did the same thing (although it's arguable this policies may have been effective if he'd been given another six months, as things were turning as he went out)

          It's fairly expected that the USA is going to crash again as it didn't fix the problems which led up to 2008 (that crash were pretty predictable when the spam for house and car loans started pummelling servers in the late 1990s. Offering credit to people with a history of defaults isn't a good move) or the problems and extreme hardship at the bottom end that have been caused since (Being poor isn't usually the poor person's "fault", so stop treating them like it is. That's how you breed criminality and extremism) - and with a populist president at the helm giving payoffs to his rich cronies (tax breaks) it's likely to crash _hard_ and stay that way for quite a while. This is why so many people have been jittery about a trade war on top of that.

          The next decade is going to be "interesting" as in the kind of history period you'd prefer to study, not live through.

  2. Wellyboot Silver badge

    Duff batch spotted

    At least QC seems to be working for TSMC and they're being open about this.

    Best part of 20 years ago there was a similar problem with IDE controllers where (I believe) the plastic used to encase the chip was slightly off spec. Nobody noticed until months later when after a few hundred hours of normal running the drive failure rate went through the roof.

    1. J. Cook Silver badge

      Re: Duff batch spotted

      Oh, the thermal issue with a rather *lot* of drives using a cirrus logic controller? bitten by that one, managed to resurrect one drive long enough to get the data off it; the other wasn't as lucky.

    2. Doctor Syntax Silver badge

      Re: Duff batch spotted

      It's not quite clear from the wording whether they discovered it before it was used or after. JIT manufacturing doesn't give much lee-way even if it's discovered in advance.

      1. Anonymous Coward
        Anonymous Coward

        Re: Duff batch spotted

        It's not so much JIT manufacturing that's to blame but rather the complexity of newer process nodes. Theres now around 1000 steps in the production of chips (ref: https://electroiq.com/2015/07/process-watch-increasing-process-steps-and-the-tyranny-of-numbers/) and limited testing possible between many of the steps. As a side note, speculation is that Intels 10nm process was killed by the number of steps and particularly long cycle times which maybe the real reason behind delays, as full capacity was going to be much lower than anticipated, meaning either a new process was required or more $10B fabs - the yield issues were just the final nail in the coffin that made it not worth even trying.

        Because of the number of steps in the pipeline, finding a problem results in all of the products in the pipeline being defective and discarded plus additional downtime for cleaning/recalibration. A typical high performance process will take around 3 months between a wafer entering the manufacturing process and a final product being tested - depending where the issue lay, that may mean a significant amount of product has been lost.

        Combined with the announcements by nVida and AMD all reporting an expected 20% profit hit (I think TSMC and designers share risk in production process - i.e. if there is a process failure, the designer is liable to a portion of the costs regardless of fault and if the design proves to be faulty, they are liable for full costs).

    3. Anonymous Crowbar

      Re: Duff batch spotted

      There was a motherboard capacitor issue back around then as well where the caps has some dodgy industrial espionage, china clones of chemical formulae, and Taiwan manufactures buying up cheap knockoff electrolyte solvent that was missing some additives that stopped the issue.

      https://youtu.be/EUsd-Ri0VCo

  3. whitepines Bronze badge
    Facepalm

    nVidia's datacenter business isn't doing as well as expected....after making people pay extra for a license to use their (mandatory) proprietary drivers on their (locked and signed) hardware. Just because it's used in a datacenter.

    Meanwhile AMD is busy providing open drivers that, for the most part, just work.. Brilliant strategy there, nVidia....

    1. Alan Brown Silver badge

      "Meanwhile AMD is busy providing open drivers that, for the most part, just work.. "

      It's an uphill struggle to break habitual use of CUDA in favour of OpenCL though - and that's not helped by enterprise distros *ahem*REDHAT*ahem* not pulling their collective fingers out of the arses and making their distros work with _current_ hardware and _current_ libraries.

      Nividia is the new Intel or Microsoft in many ways (hearts and minds) and enterprise software vendors are encouraging them. It was a nice feeling when after spending months trying to get a Nvidia video wall working in Linux I took advice from the Linux Nouveau developers and on their recommendation dropped in a Radeon card - which simply worked first time with no fuss.

  4. DJV Silver badge

    "adversely affected by a bad batch of chemicals"

    So, someone spilled their cuppa, then?

    1. VikiAi Bronze badge
      Facepalm

      Re: "adversely affected by a bad batch of chemicals"

      I said *hold* my beer! Not drop it in the batch!

    2. Version 1.0 Silver badge

      Re: "adversely affected by a bad batch of chemicals"

      This has happened before, remember power supply capacitors going POP? I wonder if this is going to be a similar cause?

      1. Alan Brown Silver badge

        Re: "adversely affected by a bad batch of chemicals"

        "remember power supply capacitors going POP"

        Yup, but that was down to the stolen electrolyte formula (pinched from Japan) being incomplete and not being gas-generating resistant when it got to Taiwan. Nor was it complete when China stole it from the Taiwanese and repeated the problem.

        Industrial espionage is everywhere. The USA is one of the largest practioners of it - and subtle industrial sabotage is also more common than you might think. Why blow things up (and end up with fingers pointing back at you) when you can make Iran's gas centrifuges pulse their speeds, or lightly tweak the chemicals coming out of XYZ plant in order to nobble the competition in $foreign country of choice? - all with complete plausible deniability.

        Bletchley park and Ultra should show that governments do know how to keep things close to their chests - and local manufacturers, etc _are_ a valuable resource that you want to do better than other countries' ones, so why not even the odds a bit without resorting to those pesky tariffs that raise so many hackles?

  5. The Oncoming Scorn Silver badge
    Pint

    Welcome To The Fracked Cascade (Failures)

    Best I can do (Its been one of those days) - Sorry.

    Icon = I really need a lot of those, but I'm On-Call tonight.

  6. Anonymous Coward
    Anonymous Coward

    Hmmmm

    Mostly 12/16nm processes affected - so mainstream products i.e. current nVidia/AMD GPU's and AMD CPU's, probably a number of network vendors and Huawei's telecoms business may struggle for volumes for a few months.

    For AMD's sake, I hope 7nm is unaffected - if Epyc Rome chips have a chance to challenge Intel in the marketplace rather than just in benchmarks, they need to avoid delays.

    1. wurdsmiff

      Re: Hmmmm

      The article on Tom's Hardware suggests that 7nm is fine and that it's mainly AMD's APUs for PS4/XBOX One that will be impacted on their side.

      Fingers crossed that's the case because the competition would be really welcome on the consumer tech side too.

    2. Wayland Bronze badge

      Re: Hmmmm

      It looks like a strategic move to me. Helps those customers products on 7nm and knobbles the encombant products. This may have the effect of making NVidia products scares and even more expensive whilst boosting AMD.

      It's not something they could do on purpose but they can get away with it by accident. Notice they are careful not to pin the blame on the actual supplier of tainted material.

  7. Anonymous Coward
    Anonymous Coward

    Bad chips

    There are rumors that Bitmain was also affected by something like this, as there have been a lot of S7 and S9 failures attributed to bad 28/14 nm ASICs.

    Seems that heat speeds up the failures and there is no fix: the chip just outright dies with no repair possible similar to the fault on some WS2812s.

    As they are all speed matched the fix is not simple.

    Incidentally I have ordered an S7 board to verify why the chips are failing and will post a teardown soon!!

    Decapping just the bad - chips is doable and hopefully I can also find out how to test them in circuit.

    1. Anonymous Coward
      Anonymous Coward

      Re: Bad chips

      I just dropped a load off at recycling earlier this month - they would boot up, run for a while and then quit. Sometimes a chain would restart but never for long.

  8. Mephistro Silver badge
    Facepalm

    "Nvidia on Monday lowered its fiscal Q4 2019 financial guidance..."

    In my experience -being the (un)happy owner of one of their graphic cards- it's a miracle they manage to sell anything at all, at least in the gaming market. Whether a game -new or old- will work with their cards or not is a Russian roulette. Updating their drivers sometimes is totally useless for new games, or breaks installed games, or both!. Draconian restrictions to the "approved use" of their graphics cards and lack of open-source drivers also help.

    To make things worse, a year and a half ago, they devised a slurping scheme called "GeForce Experience" where lots of usage data supposedly needed to make the damn cards work was included with their drivers. The option was "on" by default, and for a while it was mandatory or it wouldn't allow the installation of updated drivers. Thankfully GDPR put an end to that, but their drivers are just crap still.

    Add to that the stratospheric prices they charge for their cards -for a not so big increase in throughput- and this lowered income forecasts start looking terribly optimistic.

    Fuck'em with a cactus!

  9. imanidiot Silver badge

    Sounds a bit like a bad batch of resist resulting in poor litho performance. Or possibly something got contaminated with copper. That's a really insiduous contaminant in how it affects the chip. Could work fine for a while until higher temps allow the copper to diffuse somewhere it shouldn't be.

    1. Duncan Macdonald Silver badge

      Chemical purity

      Many of the chemicals used (even the water!!) are supposed to have impurity levels in the low parts per billion range. If, due to a supplier fault, one of the chemicals only has a purity of 99.9999% (1 part per million impurity) this could mess up a large batch of wafers before the problem was detected. The fab would then need to clean out all the affected equipment from the storage tanks on before refilling with fresh (tested!!!) chemicals The fab would then need to run some test wafers to check that the problem has been cleared before resuming production

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019