8 ms·

Historically, excess wafer fab capacity is what bankrupts semiconductor companies. An underloaded fab very rapidly eats up profit. Over the last 10 years, the i

Historically, excess wafer fab capacity is what bankrupts semiconductor companies. An underloaded fab very rapidly eats up profit. Over the last 10 years, the industry has become very cautious not to overbuild capacity. We have now arrived at the opposite end of the long term (~10 year) cycle and we have accumulated a serious underinvestment in capacity.

hoseja 4y ago
Why isn't it possible to do this industry at less than razor edge?
- sct202 4y ago
  It wasn't razor edge in 2009, when TSMC was running a 40% utilization rate. Forecasting is hard, and when you have in recent experience a massive glut it makes companies conservative about demand forecasts and investments.
  - xxpor 4y ago
    See also: oil "capital discipline" right now
- blihp 4y ago
  Because that's not where you end up in a world without numerous serious competitors.
  - Macha 4y ago
    I would generally assume a lack of competitors would lead to a decreased need to innovate and therefore fatter margins.
    - blihp 4y ago
      Right, which is pretty much where we are now: very fat margins largely because fabs have been banging up against capacity for a while now because they could. (i.e. to avoid the risk of overcapacity... why take a chance if you don't have to?) In a more perfect world with healthy competition, one or more of them would see the lack of capacity and see opportunity to take the risk to build a new fab to take some of that business. The fact that it's basically governments saying 'you will do this' tells you everything you need to know about how much the players are really driven by competition. It's a great time to be a semi manufacturer: TSMC has customers like Apple funding their expansion, Intel has the U.S. government funding theirs. Can you blame them? A decade of record profits isn't what it used to be...
- joenathanone 4y ago
  The lithography machines and factories are prohibitively expensive, it can take years of production to break even, if you over build you may just end up in the red. Additionally the fabs have a limited lifespan when they are cutting edge and able to produce premium chips.
  - rednerrus 4y ago
    Then shouldn't the government be involved in ensuring we have what's necessary? Seems like strategic reserves are in order here.
    - lsllc 4y ago
      It's not unprecedented -- during the cold war, the US Air Force spent $279M (~$3B in today's money) on building some of the largest presses in the world specifically to enhance the capabilities around building complex and strong aircraft parts: https://en.wikipedia.org/wiki/Heavy_Press_Program https://en.wikipedia.org/wiki/Heavy_Press_Program
      - stevenhuang 4y ago
        Wow. Attached is a fascinating doc going through the specs and construction process of this feat https://web.archive.org/web/20120227025023/http://files.asme.org/asmeorg/communities/history/landmarks/5488.pdf https://web.archive.org/web/20120227025023/http://files.asme... It has pictures!
      - blamazon 4y ago
        Fascinating link! Check out this titanium bulkhead for an F-15 before and after pressing: https://upload.wikimedia.org/wikipedia/commons/9/9e/F15_bulkhead_before-after_pressing.jpg https://upload.wikimedia.org/wikipedia/commons/9/9e/F15_bulk...
    - 55873445216111 4y ago
      In practical terms, the government is just too slow to actually help. They always shoot behind the target. The semiconductor market is cyclical. By the time the CHIPS Act, etc, actually get passed and implemented, the market will already be in oversupply and voters/politicians will complain about government handouts for these companies.
- deelowe 4y ago
  I'm not sure I agree with some of the responses. The issue is that the price to build or retrofit fabs goes up significantly with each generation. Only a few companies have been able to survive given the amount of financial risk involved.
- 55873445216111 4y ago
  It takes >$10B and >2 years to build a fab which will be effectively obsolete ~2 years after it opens. If it doesn't open on schedule or is underloaded, the ROI goes negative almost immediately. (I somewhat oversimplified this of course but it's to make the point.)
  - hoseja 4y ago
    Obsolete? Automakers etc are huuuungry for old chips.
    - 55873445216111 4y ago
      Yes but there is a huge technology gap. example: Automakers need a lot more small low-power MCUs built on ~130/90nm process. A 14nm fab that started 2 years ago is never going to be producing those 130/90nm MCUs. Furthermore, the semi toolmakers that produce the tools needed to build 130/90nm chips stopped making those tools years ago. This is a rabbit-hole that just keeps going. The problem isn't going away soon or easily.
      - mjevans 4y ago
        Maybe three different 'bands' of production: * cutting / bleeding edge (as now) * 80/20 trailing edge (5 year LTS of cutting edge) * Cost focused 'lifetime' LTS It seems like the cutting edge is built, and transitioned to the 80/20 already. It's that cost focused bulk components / entry level that lacks investment or even planning. Maybe that's where some industry development and co-ordination from regulators could help establish a market that's healthy.
    - StillBored 4y ago
      Not just automakers, and not just old fabs. Not everything needs the latest bleeding edge process. So, even then some of the old processes get updated to support larger wafers. Ex: brand new 65nm fab: https://www.electronicsweekly.com/news/business/bosch-opens-fab-2021-06/ https://www.electronicsweekly.com/news/business/bosch-opens-... So, 10-15 year ROI easy, and I'm guessing that might stretch even further as the trend to ever smaller nodes slows. AKA I'm betting 2nm doesn't look nearly as aged in 2040 as 90nm looks today.
mistrial9 4y ago
> accumulated a serious underinvestment in capacity says who? the world's rivers and oceans are polluted with micro-plastics and heavy metals from E*Waste, while thirty years of "wild west" profits and dog-eat-dog markets, COMDEX and single-use devices will take 1000 years to degrade in Nature. Calls for "more" to be viewed with earned-suspicion IMO
- btgeekboy 4y ago
  While I do agree that e-waste is a real problem, it’s not like we can just sit with old tech and say “nah, that’s good enough.” We need both more and better fabs, and better waste handling / recycling.
  - beamatronic 4y ago
    Maybe a certain level of tech IS good enough. A better world for people means education and healthcare, not a 1mm slimmer disposable phone.
    - peanuty1 4y ago
      Newer chips are also more energy efficient (at least, at the same level of performance).
      - beamatronic 4y ago
        Isn’t a recyclable, user-replaceable battery better for the environment in the long run?
      - KSteffensen 4y ago
        Does that improved energy efficiency outweigh the energy costs of replacing perfectly capable existing chips?
    - TrevorJ 4y ago
      A phone from 3 years ago, running software from 4 years ago would be as fast as anyone would reasonably need, but it gets obsoleted via software bloat. More frameworks, more layers of abstraction and no confining force that forces anyone to write software that isn't a bloated mess.
unethical_ban 4y ago
I feel like there was a lesson about just-in-time inventory for critical supply chains recently. Perhaps a product as complex, sophisticated and absolutely necessary for a modern society should have government guarantees.
- acchow 4y ago
  A futures market would give some income stream to anyone who held significant inventories. Wonder how soon we have something like a futures market for semiconductors
- 55873445216111 4y ago
  Honestly, the risks of just-in-time have always been know. The problem is that holding excess inventory and/or reserve capacity costs real money, and somebody has to pay for it. Investors/managers drive out wasted costs or eventually they go out of business. (Absent government support.)
- stubish 4y ago
  What will the lesson be though? Once the dust settles, it may well be that the savings over the last few decades outweighs the losses due to the supply shortages. I guess it comes down to how good the risk analysis was and if they acknowledge that once-in-a-lifetime events happen many times in a lifetime.
cjsplat 4y ago
The wash out in the memory industry was generally the opposite. DRAM was a boom and bust market and during bust/oversupply windows vendors that under invested in the next node technology got washed out when the next boom cycle happened. When the market is high volume, price per bit, and new nodes are every 2 or 3 years, you can't really compete it your are nearly 2x as expensive (30% linear improvement => 70% squared cost per bit). Of course you need to survive the dry times, but you make maximum margin when you are on the next fab node and your competition isn't. I think what you are saying makes more sense when applied to contract manufacturers than all semiconductor companies.
kube-system 4y ago
Maybe it’s time to implement policies that encourage reserve capacity, as we do with some other critical industries.

hoseja 4y ago

Why isn't it possible to do this industry at less than razor edge?

sct202 4y ago
It wasn't razor edge in 2009, when TSMC was running a 40% utilization rate. Forecasting is hard, and when you have in recent experience a massive glut it makes companies conservative about demand forecasts and investments.
- xxpor 4y ago
  See also: oil "capital discipline" right now
blihp 4y ago
Because that's not where you end up in a world without numerous serious competitors.
- Macha 4y ago
  I would generally assume a lack of competitors would lead to a decreased need to innovate and therefore fatter margins.
  - blihp 4y ago
    Right, which is pretty much where we are now: very fat margins largely because fabs have been banging up against capacity for a while now because they could. (i.e. to avoid the risk of overcapacity... why take a chance if you don't have to?) In a more perfect world with healthy competition, one or more of them would see the lack of capacity and see opportunity to take the risk to build a new fab to take some of that business. The fact that it's basically governments saying 'you will do this' tells you everything you need to know about how much the players are really driven by competition. It's a great time to be a semi manufacturer: TSMC has customers like Apple funding their expansion, Intel has the U.S. government funding theirs. Can you blame them? A decade of record profits isn't what it used to be...
joenathanone 4y ago
The lithography machines and factories are prohibitively expensive, it can take years of production to break even, if you over build you may just end up in the red. Additionally the fabs have a limited lifespan when they are cutting edge and able to produce premium chips.
- rednerrus 4y ago
  Then shouldn't the government be involved in ensuring we have what's necessary? Seems like strategic reserves are in order here.
  - lsllc 4y ago
    It's not unprecedented -- during the cold war, the US Air Force spent $279M (~$3B in today's money) on building some of the largest presses in the world specifically to enhance the capabilities around building complex and strong aircraft parts: https://en.wikipedia.org/wiki/Heavy_Press_Program https://en.wikipedia.org/wiki/Heavy_Press_Program
    - stevenhuang 4y ago
      Wow. Attached is a fascinating doc going through the specs and construction process of this feat https://web.archive.org/web/20120227025023/http://files.asme.org/asmeorg/communities/history/landmarks/5488.pdf https://web.archive.org/web/20120227025023/http://files.asme... It has pictures!
    - blamazon 4y ago
      Fascinating link! Check out this titanium bulkhead for an F-15 before and after pressing: https://upload.wikimedia.org/wikipedia/commons/9/9e/F15_bulkhead_before-after_pressing.jpg https://upload.wikimedia.org/wikipedia/commons/9/9e/F15_bulk...
  - 55873445216111 4y ago
    In practical terms, the government is just too slow to actually help. They always shoot behind the target. The semiconductor market is cyclical. By the time the CHIPS Act, etc, actually get passed and implemented, the market will already be in oversupply and voters/politicians will complain about government handouts for these companies.
deelowe 4y ago
I'm not sure I agree with some of the responses. The issue is that the price to build or retrofit fabs goes up significantly with each generation. Only a few companies have been able to survive given the amount of financial risk involved.
55873445216111 4y ago
It takes >$10B and >2 years to build a fab which will be effectively obsolete ~2 years after it opens. If it doesn't open on schedule or is underloaded, the ROI goes negative almost immediately. (I somewhat oversimplified this of course but it's to make the point.)
- hoseja 4y ago
  Obsolete? Automakers etc are huuuungry for old chips.
  - 55873445216111 4y ago
    Yes but there is a huge technology gap. example: Automakers need a lot more small low-power MCUs built on ~130/90nm process. A 14nm fab that started 2 years ago is never going to be producing those 130/90nm MCUs. Furthermore, the semi toolmakers that produce the tools needed to build 130/90nm chips stopped making those tools years ago. This is a rabbit-hole that just keeps going. The problem isn't going away soon or easily.
    - mjevans 4y ago
      Maybe three different 'bands' of production: * cutting / bleeding edge (as now) * 80/20 trailing edge (5 year LTS of cutting edge) * Cost focused 'lifetime' LTS It seems like the cutting edge is built, and transitioned to the 80/20 already. It's that cost focused bulk components / entry level that lacks investment or even planning. Maybe that's where some industry development and co-ordination from regulators could help establish a market that's healthy.
  - StillBored 4y ago
    Not just automakers, and not just old fabs. Not everything needs the latest bleeding edge process. So, even then some of the old processes get updated to support larger wafers. Ex: brand new 65nm fab: https://www.electronicsweekly.com/news/business/bosch-opens-fab-2021-06/ https://www.electronicsweekly.com/news/business/bosch-opens-... So, 10-15 year ROI easy, and I'm guessing that might stretch even further as the trend to ever smaller nodes slows. AKA I'm betting 2nm doesn't look nearly as aged in 2040 as 90nm looks today.

mistrial9 4y ago

> accumulated a serious underinvestment in capacity says who? the world's rivers and oceans are polluted with micro-plastics and heavy metals from E*Waste, while thirty years of "wild west" profits and dog-eat-dog markets, COMDEX and single-use devices will take 1000 years to degrade in Nature. Calls for "more" to be viewed with earned-suspicion IMO

btgeekboy 4y ago
While I do agree that e-waste is a real problem, it’s not like we can just sit with old tech and say “nah, that’s good enough.” We need both more and better fabs, and better waste handling / recycling.
- beamatronic 4y ago
  Maybe a certain level of tech IS good enough. A better world for people means education and healthcare, not a 1mm slimmer disposable phone.
  - peanuty1 4y ago
    Newer chips are also more energy efficient (at least, at the same level of performance).
    - beamatronic 4y ago
      Isn’t a recyclable, user-replaceable battery better for the environment in the long run?
    - KSteffensen 4y ago
      Does that improved energy efficiency outweigh the energy costs of replacing perfectly capable existing chips?
  - TrevorJ 4y ago
    A phone from 3 years ago, running software from 4 years ago would be as fast as anyone would reasonably need, but it gets obsoleted via software bloat. More frameworks, more layers of abstraction and no confining force that forces anyone to write software that isn't a bloated mess.

unethical_ban 4y ago

I feel like there was a lesson about just-in-time inventory for critical supply chains recently. Perhaps a product as complex, sophisticated and absolutely necessary for a modern society should have government guarantees.

acchow 4y ago
A futures market would give some income stream to anyone who held significant inventories. Wonder how soon we have something like a futures market for semiconductors
55873445216111 4y ago
Honestly, the risks of just-in-time have always been know. The problem is that holding excess inventory and/or reserve capacity costs real money, and somebody has to pay for it. Investors/managers drive out wasted costs or eventually they go out of business. (Absent government support.)
stubish 4y ago
What will the lesson be though? Once the dust settles, it may well be that the savings over the last few decades outweighs the losses due to the supply shortages. I guess it comes down to how good the risk analysis was and if they acknowledge that once-in-a-lifetime events happen many times in a lifetime.

cjsplat 4y ago

The wash out in the memory industry was generally the opposite. DRAM was a boom and bust market and during bust/oversupply windows vendors that under invested in the next node technology got washed out when the next boom cycle happened. When the market is high volume, price per bit, and new nodes are every 2 or 3 years, you can't really compete it your are nearly 2x as expensive (30% linear improvement => 70% squared cost per bit). Of course you need to survive the dry times, but you make maximum margin when you are on the next fab node and your competition isn't. I think what you are saying makes more sense when applied to contract manufacturers than all semiconductor companies.

kube-system 4y ago

Maybe it’s time to implement policies that encourage reserve capacity, as we do with some other critical industries.