5 ms·

While TMSC is no doubt ahead of competition, it is just by margin of 1-2 generations at max, which accounts for max 2-3 years for friendly country like Korea to

While TMSC is no doubt ahead of competition, it is just by margin of 1-2 generations at max, which accounts for max 2-3 years for friendly country like Korea to catch up. So if some Taiwan politics comes into play, we will just lose the progress for some years of semiconductor fab tech, not a small thing but it is not that the chips couldn't be built if not for Taiwan. We would just lose something like 10-20% of performance in the chip.

Buraksr 6y ago
I would argue the opposite, that it is a huge deal for defense. Imagine if the situation is missile defense, both the missile and the defense systems are computerized and not only the milliseconds but also the microseconds are the difference between a hit and a stopped missile. Whoever had the newest fab node would have a notable advantage in being able to push their system harder. It could be better ML in missile to lock on and dodge, vs the defense system predicting paths as fast as possible with as low latency as possible to defend a ship or similar asset. Its asymmetric in the example of course, but both sides would be doing both shooting and defending.
- FrojoS 6y ago
  Your scenario is not founded in reality. Few problems that involve computers are limited by or can primarily be solved by computer technology. In your example, one of the main challenges is that every missile defense system can be overwhelmed by shear number of attack missiles. Have a look at what type of microchips are used in “cutting edge” military equipment like fighter jets. They are often decade olds. More speed is simply not important enough.
  - xupybd 6y ago
    Not only that but smaller cores are less radiation resistant. Meaning more likely to get interference from natural radiation sources. So less reliable.
  - Buraksr 6y ago
    I think what you say is correct, its so much easier to fire a missile than it is to stop one. Honestly, reliability and cost are bigger drivers than raw performance. The scenario is a bit bad I can admit, and did admit. But if we were in desperate times, I think the limits would matter and fab tech would be an important strategic asset for what it enables. The nature of it is that its easier to attack, but we hopefully have something to defend. Intrinsically larger process nodes are better for analog and have uses even as new fabs come along. In 7nm your analog signals are not treated very nicely, but these are important for sensors. Also power electronics are not really on the newest nodes either. The microchips are old because they get the job done and I don’t think we feel so threatened to push the limits.
- blackoil 6y ago
  Are defense systems on such cutting age tech? I would have assumed they are using 40-200nm as development cycle for such system is usually in decades.
  - Buraksr 6y ago
    Generally old chips are good, depends on the purpose. Some things like the f35 would like to use TSMC’s fabs. But not every chip is for only processing and larger process have their advantages for certain applications like power, analog, and em hardening
    - megous 6y ago
      Also the fab for a 20year old process is probably easier/cheaper/faster to re-build in case the enemy destroys it, than the fab for 5nm process you've been building for 4 years. Nevermind the nation probably has more than 1 of such old fabs, so it can't be technologically decapitated by destroying a single precious 5nm fab it has.
simonh 6y ago
You can't simply swap out the guidance and control systems of state of the art kit for a bunch of chips 5 years older. That's not how high performance embedded systems engineering works. Older chips need more power so bigger batteries, produce more heat, take up more space, have lower performance, all inside a missile, munition, satellite or fighter jet with extremely limited space and exacting design tolerances. The best you could do is go back to making older systems for which the factories may not exist anymore, or design and build new systems using the old CPU tech and new everything else. Both options would impose a severe economic and military cost and take years to implement. It doesn't matter that 5nm chips aren't used in military hardware today. You can bet military hardware is being designed now that will use 5nm chips. Therefore access to that production capacity is strategic.
phkahler 6y ago
>> While TMSC is no doubt ahead of competition, it is just by margin of 1-2 generations at max No, they're ahead because they mastered EUV lithography. It's not just another shrink, it requires a lot of changes and they all have to work and play nice together. The fact that Intel can't seem to get it right, and global foundries voluntarily stopped short is very telling. Having cleared the hurdles TSMC is rapidly going through a bunch of new nodes: 7,6,5,3,2.
- YetAnotherNick 6y ago
  Hasn't samsung already started production with EUV lithography?
  - sanxiyn 6y ago
    Yes, Samsung's EUV process is in production since 2019.
  - mardiyah 6y ago
    IMHO the latest is 8nm, not ensuring true or half
- fpoling 6y ago
  EUV machinery was delivered to TMSC by a firm from The Netherlands.
  - memorybadger 6y ago
    ASML
  - phkahler 6y ago
    And Intel has purchased such equipment from them as well. A machine is not a fab-process-in-a-box or everyone with money would be in production already.
    - mardiyah 6y ago

Buraksr 6y ago

I would argue the opposite, that it is a huge deal for defense. Imagine if the situation is missile defense, both the missile and the defense systems are computerized and not only the milliseconds but also the microseconds are the difference between a hit and a stopped missile. Whoever had the newest fab node would have a notable advantage in being able to push their system harder. It could be better ML in missile to lock on and dodge, vs the defense system predicting paths as fast as possible with as low latency as possible to defend a ship or similar asset. Its asymmetric in the example of course, but both sides would be doing both shooting and defending.

FrojoS 6y ago
Your scenario is not founded in reality. Few problems that involve computers are limited by or can primarily be solved by computer technology. In your example, one of the main challenges is that every missile defense system can be overwhelmed by shear number of attack missiles. Have a look at what type of microchips are used in “cutting edge” military equipment like fighter jets. They are often decade olds. More speed is simply not important enough.
- xupybd 6y ago
  Not only that but smaller cores are less radiation resistant. Meaning more likely to get interference from natural radiation sources. So less reliable.
- Buraksr 6y ago
  I think what you say is correct, its so much easier to fire a missile than it is to stop one. Honestly, reliability and cost are bigger drivers than raw performance. The scenario is a bit bad I can admit, and did admit. But if we were in desperate times, I think the limits would matter and fab tech would be an important strategic asset for what it enables. The nature of it is that its easier to attack, but we hopefully have something to defend. Intrinsically larger process nodes are better for analog and have uses even as new fabs come along. In 7nm your analog signals are not treated very nicely, but these are important for sensors. Also power electronics are not really on the newest nodes either. The microchips are old because they get the job done and I don’t think we feel so threatened to push the limits.
blackoil 6y ago
Are defense systems on such cutting age tech? I would have assumed they are using 40-200nm as development cycle for such system is usually in decades.
- Buraksr 6y ago
  Generally old chips are good, depends on the purpose. Some things like the f35 would like to use TSMC’s fabs. But not every chip is for only processing and larger process have their advantages for certain applications like power, analog, and em hardening
  - megous 6y ago
    Also the fab for a 20year old process is probably easier/cheaper/faster to re-build in case the enemy destroys it, than the fab for 5nm process you've been building for 4 years. Nevermind the nation probably has more than 1 of such old fabs, so it can't be technologically decapitated by destroying a single precious 5nm fab it has.

simonh 6y ago

You can't simply swap out the guidance and control systems of state of the art kit for a bunch of chips 5 years older. That's not how high performance embedded systems engineering works. Older chips need more power so bigger batteries, produce more heat, take up more space, have lower performance, all inside a missile, munition, satellite or fighter jet with extremely limited space and exacting design tolerances. The best you could do is go back to making older systems for which the factories may not exist anymore, or design and build new systems using the old CPU tech and new everything else. Both options would impose a severe economic and military cost and take years to implement. It doesn't matter that 5nm chips aren't used in military hardware today. You can bet military hardware is being designed now that will use 5nm chips. Therefore access to that production capacity is strategic.

phkahler 6y ago

>> While TMSC is no doubt ahead of competition, it is just by margin of 1-2 generations at max No, they're ahead because they mastered EUV lithography. It's not just another shrink, it requires a lot of changes and they all have to work and play nice together. The fact that Intel can't seem to get it right, and global foundries voluntarily stopped short is very telling. Having cleared the hurdles TSMC is rapidly going through a bunch of new nodes: 7,6,5,3,2.

YetAnotherNick 6y ago
Hasn't samsung already started production with EUV lithography?
- sanxiyn 6y ago
  Yes, Samsung's EUV process is in production since 2019.
- mardiyah 6y ago
  IMHO the latest is 8nm, not ensuring true or half
fpoling 6y ago
EUV machinery was delivered to TMSC by a firm from The Netherlands.
- memorybadger 6y ago
  ASML
- phkahler 6y ago
  And Intel has purchased such equipment from them as well. A machine is not a fab-process-in-a-box or everyone with money would be in production already.
  - mardiyah 6y ago