5 ms·

The US is literally doing for the Chinese foundry industry what nations often do to develop an advanced industry: force the local economy to use it so it has a

by ajb 3y ago

The US is literally doing for the Chinese foundry industry what nations often do to develop an advanced industry: force the local economy to use it so it has a revenue source and is protected from foreign competition, so it isn't snuffed out before it can really compete.

bhouston 3y ago
100%. China was going to achieve this technology at some point but we we sort of forced them into Manhattan style projects for this as a result of the sanctions. So they are likely going to achieve technology independence from the US faster as a result and then may surpass it clearly if they continue this level of effort/investment. To be fair the state of the art is 3mn as of now (eg iPhone 15 releasing next week or so.) But 7mn is pretty darn close.
- throwoutway 3y ago
  I struggle to follow along with this. I understand Moore's law, but at what point can we just keep using 7nm for most things instead of always moving down an integer 3>7. How many things will require a 2nm and not be satisfied by a 7nm fab?
  - bhouston 3y ago
    The issue is you can make this claim anywhere over the past 50 years. When you just achieve a new node, you think it isn't yet useful. But what happens is that software starts to take advantage of it and it gives you new types of capabilities you didn't have before. For example, if we never got down to where we are now, we likely wouldn't have LLMs or viable real-time ray tracing or the ability to run various types of photo improvement algorithms on your iPhone to make all your pics look good. Those require significant processing and it wouldn't have been achievable if technology here just stopped a decade ago. If we had stopped CPU progress 25 years ago we would have never gotten smart phones at all. It is hard to know what will be possible in a decade with all the additional compute resources we get from moving down more nodes. But I do expect there to be a lot.
  - ajb 3y ago
    Depends how you count. Per design we already use >7nm for way more things because older processes are cheaper. But per $ (and perhaps also in volume) the leading-edge processes get the lions share.
- formerly_proven 3y ago
  Consider however that 7 nm (TSMC N7, Intel 10ESF/Intel 7, Samsung 8/7 nm) is universally a DUV (193 nm light) process using multipatterning. The nodes below are universally EUV. That's a step-change in complexity. If "The West" manages to contain EUV to friendly countries, then it doesn't seem unlikely to me at all that Chinese semiconductors will run into the EUV wall and get stuck for a significant amount of time trying to get over it.
  - bhouston 3y ago
    Yes, but it is likely China can over come it. China has lots of educated people and as close to infinite resources as you can get on this planet. So it is just a matter of focus. Blockading a small country that doesn't have a lot of resources can be very effective, even long-term, look at Cuba or Gaza Strip, but blockading China just slows them down temporarily until they overcome this challenge. The main issue China has is just that they have a population decline problem which will continue to get worse over the next 50 years. That to me is the only thing that truly threatens China long-term and will sap its innovation and productivity.
    - xvilka 3y ago
      > The main issue China has is just that they have a population decline problem which will continue to get worse over the next 50 years. It can be (and likely will be - just see at corresponding market figures) mitigated by ubiquitous automation and robotics.
      - ajb 3y ago
        They also have over a billion people total. The top percentiles of the rural population are likely not tapped out yet.
    - est 3y ago
      > The main issue China has is just that they have a population decline problem which will continue to get worse over the next 50 years cough Industrial Revolution cough Black Death cough
- 3pac 3y ago
  Won't errors due to cosmic rays be an increasing issue at these process sizes?

by ajb 3y ago

bhouston 3y ago

100%. China was going to achieve this technology at some point but we we sort of forced them into Manhattan style projects for this as a result of the sanctions. So they are likely going to achieve technology independence from the US faster as a result and then may surpass it clearly if they continue this level of effort/investment. To be fair the state of the art is 3mn as of now (eg iPhone 15 releasing next week or so.) But 7mn is pretty darn close.

throwoutway 3y ago
I struggle to follow along with this. I understand Moore's law, but at what point can we just keep using 7nm for most things instead of always moving down an integer 3>7. How many things will require a 2nm and not be satisfied by a 7nm fab?
- bhouston 3y ago
  The issue is you can make this claim anywhere over the past 50 years. When you just achieve a new node, you think it isn't yet useful. But what happens is that software starts to take advantage of it and it gives you new types of capabilities you didn't have before. For example, if we never got down to where we are now, we likely wouldn't have LLMs or viable real-time ray tracing or the ability to run various types of photo improvement algorithms on your iPhone to make all your pics look good. Those require significant processing and it wouldn't have been achievable if technology here just stopped a decade ago. If we had stopped CPU progress 25 years ago we would have never gotten smart phones at all. It is hard to know what will be possible in a decade with all the additional compute resources we get from moving down more nodes. But I do expect there to be a lot.
- ajb 3y ago
  Depends how you count. Per design we already use >7nm for way more things because older processes are cheaper. But per $ (and perhaps also in volume) the leading-edge processes get the lions share.
formerly_proven 3y ago
Consider however that 7 nm (TSMC N7, Intel 10ESF/Intel 7, Samsung 8/7 nm) is universally a DUV (193 nm light) process using multipatterning. The nodes below are universally EUV. That's a step-change in complexity. If "The West" manages to contain EUV to friendly countries, then it doesn't seem unlikely to me at all that Chinese semiconductors will run into the EUV wall and get stuck for a significant amount of time trying to get over it.
- bhouston 3y ago
  Yes, but it is likely China can over come it. China has lots of educated people and as close to infinite resources as you can get on this planet. So it is just a matter of focus. Blockading a small country that doesn't have a lot of resources can be very effective, even long-term, look at Cuba or Gaza Strip, but blockading China just slows them down temporarily until they overcome this challenge. The main issue China has is just that they have a population decline problem which will continue to get worse over the next 50 years. That to me is the only thing that truly threatens China long-term and will sap its innovation and productivity.
  - xvilka 3y ago
    > The main issue China has is just that they have a population decline problem which will continue to get worse over the next 50 years. It can be (and likely will be - just see at corresponding market figures) mitigated by ubiquitous automation and robotics.
    - ajb 3y ago
      They also have over a billion people total. The top percentiles of the rural population are likely not tapped out yet.
  - est 3y ago
    > The main issue China has is just that they have a population decline problem which will continue to get worse over the next 50 years cough Industrial Revolution cough Black Death cough
3pac 3y ago
Won't errors due to cosmic rays be an increasing issue at these process sizes?