4 ms·

How does one start a semiconductor manufacturing company?

whatshisface 5y ago
First, find several dozen billion dollars. More seriously, find a low-performance niche application and make chips on old technology for it. Differentiate yourself on the basis of being easy for your customers to work with, a natural advantage for US-based companies. Then you only need several dozen million dollars.
- an_opabinia 5y ago
  > First, find several dozen billion dollars. Some electric car companies have not had trouble tapping into capital markets. Even Transmeta, a fabless IC maker, could tap $275m in 2000. If you were a super talented engineering firm with the ability to spend $2b on a fab, you could tap finance for it. > More seriously, find a low-performance niche application and make chips on old technology for it. This seems like the opposite of any winning strategy, and the opposite of what the government should subsidize. TSMC is so successfully because it relentlessly focuses on performance chips. If you are enjoying American subsidies it better be for the cutting edge, what the hell else are we paying for? > Differentiate yourself on the basis of being easy for your customers to work with The only customer that matters is Apple, and your only competition is Samsung and TSMC. Everyone else is using "foreign chip manufacturers" as a way to launder low Chinese labor costs, not because there is inadequate supply or poor competition in specifically semiconductor manufacturing. Perhaps that's what people want - a manufacturing hub, i.e., assembly of chips, not the chips themselves. I don't know if that's super important.
  - qbasic_forever 5y ago
    Starting from zero and competing directly with TSMC's top end processes right now is basically being at the "President Kennedy said we're going to the Moon" stage of landing on the moon. You are going to need many, many billions and billions of dollars and years of engineering and manufacturing (and failures) to get there. This is not something you just turn on the money spigot and solve the problem overnight. It is spectacular engineering at the smallest atomic level to produce in mass quantity the chips from TSMC today.
    - javajosh 5y ago
      The US has an overabundance of brilliant physics PhDs most of which are probably pretty sick of academia's bullshit at this point. Let's give 2000 of them a 100k/year job plus $5B, cheap/free access to federal land, energy, and water, and see what kind of fab(s) they could make. I bet they'd get to TSMC's level pretty quick. If we offered Taiwan a mutual defense pact, we might even get their help to get there.
      - an_opabinia 5y ago
        This is the right idea. But like I said, most of the value is in exploiting cheap Chinese labor, not the chips, which is why really none of that would work accounting wise - like you’re talking about a valuable economic enterprise but not one that would produce a lot of profits for shareholders.
      - qbasic_forever 5y ago
        They'll turn it down because they can make $250k a year writing code for Facebook/Google/etc. And it's really more talent in process engineering that you need, not in the physics of chip building. We already know what has to be done to make transistors. It's the how to do it at scale and with high enough yield that's impressively difficult.
        javajosh 5y ago
        Haha. No way. I don't think anyone who gets a physics PhD is excited by the prospect of solving ad market problems. Or financial market problems (because you forgot to mention the $500k quant jobs). Physics PhDs aren't, in my experience, particularly motivated by money. How could they be? A project like this would a) let them use far more of their training than coding ever would and b) it is for a good cause, effectively a peaceful version of the Manhattan project.
        mikebike 5y ago
        You might want to check the bio of Antonio García Martínez, who has a physics background, and subsequently worked in both finance and adtech.
  - ajmadesc 5y ago
    >What the hell else are we paying for? Are chips for industrial equipment or cars? When you say cutting edge I think desktop / phone / servers.
willis936 5y ago
This question makes me think of another question: is there a difference between a rich tax credit and a poor tax?
- lisper 5y ago
  Yes. The difference is in how it is perceived, but in an economy, perceptions actually matter. If you give a tax credit to the rich, the poor don't perceive it as a tax on them even though the strict monetary effect is the same in both cases i.e. a wealth transfer from the poor to the rich. But because the poor don't perceive rich tax credits as a tax on them they are less likely to object.
  - Ericson2314 5y ago
    That type of thinking is political suicide in the long term. Big meddling government that doesn't attack the fact that we have far more demand than supply problems is going to get thrown out.
- zdragnar 5y ago
  They aren't related at all, because the federal budget and revenues don't need to balance. They rarely ever do. On the other hand, tax credits can be (but arent guaranteed to be). That is, if a tax credit incentivizes creating jobs that would not have existed (in the US) otherwise, those jobs end up also stimulating other economic activity. Tax incentives at the state level tend to be closer to zero sum (or a net loss) if the jobs would have been created anyway, or when thr incentives aren't strictly tied to quantifiable completion metrics like factory completion, number of positions filled, etc. (Much ado about Foxconn in Wisconsin, etc)
  - Ericson2314 5y ago
    Huh? Yes federal debt doesn't matter but that's why they are similar—they are both shifting power to capital away from labor. The fact that governments try to do Keynesian by making manufacturing jobs in 2021 kills me. We're decades past the point where those sectors were sufficiently unproductive to benefit the workers as they did in the glory days. Boost demand directly (UBI), keep an ey on imports, and let private capital fend for itself.
bob1029 5y ago
Very carefully. And as of 2021, also with the aid of a time machine.
xbmcuser 5y ago
Starting a semiconductor manufacturing company is 1 thing starting 1 that will be competitive will require 100s of billions of dollars.
breck 5y ago
Copyright and patent laws make it near impossible except for a few thousand people.

First, find several dozen billion dollars. More seriously, find a low-performance niche application and make chips on old technology for it. Differentiate yourself on the basis of being easy for your customers to work with, a natural advantage for US-based companies. Then you only need several dozen million dollars.

an_opabinia 5y ago
> First, find several dozen billion dollars. Some electric car companies have not had trouble tapping into capital markets. Even Transmeta, a fabless IC maker, could tap $275m in 2000. If you were a super talented engineering firm with the ability to spend $2b on a fab, you could tap finance for it. > More seriously, find a low-performance niche application and make chips on old technology for it. This seems like the opposite of any winning strategy, and the opposite of what the government should subsidize. TSMC is so successfully because it relentlessly focuses on performance chips. If you are enjoying American subsidies it better be for the cutting edge, what the hell else are we paying for? > Differentiate yourself on the basis of being easy for your customers to work with The only customer that matters is Apple, and your only competition is Samsung and TSMC. Everyone else is using "foreign chip manufacturers" as a way to launder low Chinese labor costs, not because there is inadequate supply or poor competition in specifically semiconductor manufacturing. Perhaps that's what people want - a manufacturing hub, i.e., assembly of chips, not the chips themselves. I don't know if that's super important.
- qbasic_forever 5y ago
  Starting from zero and competing directly with TSMC's top end processes right now is basically being at the "President Kennedy said we're going to the Moon" stage of landing on the moon. You are going to need many, many billions and billions of dollars and years of engineering and manufacturing (and failures) to get there. This is not something you just turn on the money spigot and solve the problem overnight. It is spectacular engineering at the smallest atomic level to produce in mass quantity the chips from TSMC today.
  - javajosh 5y ago
    The US has an overabundance of brilliant physics PhDs most of which are probably pretty sick of academia's bullshit at this point. Let's give 2000 of them a 100k/year job plus $5B, cheap/free access to federal land, energy, and water, and see what kind of fab(s) they could make. I bet they'd get to TSMC's level pretty quick. If we offered Taiwan a mutual defense pact, we might even get their help to get there.
    - an_opabinia 5y ago
      This is the right idea. But like I said, most of the value is in exploiting cheap Chinese labor, not the chips, which is why really none of that would work accounting wise - like you’re talking about a valuable economic enterprise but not one that would produce a lot of profits for shareholders.
    - qbasic_forever 5y ago
      They'll turn it down because they can make $250k a year writing code for Facebook/Google/etc. And it's really more talent in process engineering that you need, not in the physics of chip building. We already know what has to be done to make transistors. It's the how to do it at scale and with high enough yield that's impressively difficult.
      - javajosh 5y ago
        Haha. No way. I don't think anyone who gets a physics PhD is excited by the prospect of solving ad market problems. Or financial market problems (because you forgot to mention the $500k quant jobs). Physics PhDs aren't, in my experience, particularly motivated by money. How could they be? A project like this would a) let them use far more of their training than coding ever would and b) it is for a good cause, effectively a peaceful version of the Manhattan project.
        mikebike 5y ago
        You might want to check the bio of Antonio García Martínez, who has a physics background, and subsequently worked in both finance and adtech.
- ajmadesc 5y ago
  >What the hell else are we paying for? Are chips for industrial equipment or cars? When you say cutting edge I think desktop / phone / servers.

willis936 5y ago

This question makes me think of another question: is there a difference between a rich tax credit and a poor tax?

lisper 5y ago
Yes. The difference is in how it is perceived, but in an economy, perceptions actually matter. If you give a tax credit to the rich, the poor don't perceive it as a tax on them even though the strict monetary effect is the same in both cases i.e. a wealth transfer from the poor to the rich. But because the poor don't perceive rich tax credits as a tax on them they are less likely to object.
- Ericson2314 5y ago
  That type of thinking is political suicide in the long term. Big meddling government that doesn't attack the fact that we have far more demand than supply problems is going to get thrown out.
zdragnar 5y ago
They aren't related at all, because the federal budget and revenues don't need to balance. They rarely ever do. On the other hand, tax credits can be (but arent guaranteed to be). That is, if a tax credit incentivizes creating jobs that would not have existed (in the US) otherwise, those jobs end up also stimulating other economic activity. Tax incentives at the state level tend to be closer to zero sum (or a net loss) if the jobs would have been created anyway, or when thr incentives aren't strictly tied to quantifiable completion metrics like factory completion, number of positions filled, etc. (Much ado about Foxconn in Wisconsin, etc)
- Ericson2314 5y ago
  Huh? Yes federal debt doesn't matter but that's why they are similar—they are both shifting power to capital away from labor. The fact that governments try to do Keynesian by making manufacturing jobs in 2021 kills me. We're decades past the point where those sectors were sufficiently unproductive to benefit the workers as they did in the glory days. Boost demand directly (UBI), keep an ey on imports, and let private capital fend for itself.

bob1029 5y ago

Very carefully. And as of 2021, also with the aid of a time machine.

xbmcuser 5y ago

Starting a semiconductor manufacturing company is 1 thing starting 1 that will be competitive will require 100s of billions of dollars.

breck 5y ago