Comment by JustExAWS
14 hours ago
Chip manufacturing is too important for the US. We can’t be completely dependent on Taiwan. Nothing against Taiwan, it’s one attack away from being obliterated by China.
No company is going to come out of someone’s garage and build a chip fab.
We can definitely offer subsidies for manufacturing in the US - we've already gotten TSMC to open several fabs.
> We can definitely offer subsidies for manufacturing in the US
The very subsidies Intel now has to pay with shares for? How is that a subsidy? Companies now and in the future would be very concerned before taking any US subsidies because the terms can always change after the fact.
Are we so sure Intel sees this as a bad thing? The US now has even more reason to prop them up.
And it’s still owned by a foreign country and Taiwan is restricting TSMC from manufacturing their most advanced processors from being manufactured in the US.
https://www.tomshardware.com/tech-industry/semiconductors/ta...
Same as the US is restricting sale of Nvidia chips to China.
this is not to say that intel will be manufacturing competitive chips to what TSMC is.
are you worried that china will invade taiwan, and then somehow taiwan will still be around to prevent the US fabs from making the best chips?
its a bit far fetched
2 replies →
So we give a bunch of money to a company with a history of mismanagement and out sourcing chip manufacturing?
[dead]
Nobody is going to swoop in and buy a distressed company that owns a bunch of fabs then turn it around if that company keeps getting bailed out.
Right it would make a lot more sense to let this happen and then restrict that the buyers be American (or European, I guess).
Nvidia has a market cap of 4.5 trillion dollars and everyone is committing hundreds of billions to AI CapEx in their direction - they can afford to organise chip fabs if it really came to it. Ok TSMC and ASML would need to be on board but it could be done. Should be done in fact because even a simple SWOT analysis would show the risk to their business.
No amount of money is going to create a new fab in a reasonable timeframe.
You can buy one, if a suitable one exists, but there isn’t spare stock sitting around; the lead time is long, especially for high end nodes.
If Taiwan becomes practically inaccessible, is there any way another country can setup a competing fab (for the latest generation of chip sizes) without years of R&D? As far as I understand, the practical knowledge of how to do it doesn't exist right now. (Neither does the prerequisite tooling)
Given there’s fabs doing essentially the same thing elsewhere then yes. Getting down to 3nm and the technology and secrets that involves would take a while though.
TSMC can’t do it either without xUV lithography machines made by ASML in the Netherlands.
Furthermore there isn’t anything magical about about the current generation of chips that couldn’t be replicated at at a scale of 12 or 15 or 20 nanometers - it’s just that scaling down to that small allows for a greater density of transistors per wafer and thus increased power efficiency. An AI supercomputer could be built with chips with bigger transistors than 3nm it would just run hotter.
And investing in intel aside, one of Nvidias great competitive moats is CUDA and that’s software not hardware.
1 reply →
Aren’t the actual machines used in the fabs still made in Germany?
What is the risk for Nvidia if TSMC diseaper? Wouldn't they simply switch supplier and pick the second best option?
GPUs would go backwards a few generations for 5-10 years. Also supply shock on other industries would double the prices of chips for vehicles. Eg covid 3.0
Then they'd have to use Samsung or Intel. Both are a bit behind TSMC, but the main issue is that TSMC has a massive amount of capacity so chips would become very, very expensive.
That would not only be incredibly expensive, but there would be a period while quality catches up.
"Someone's garage" is a straw man. There must be people here who could, with adequate funding, build a smallish but viable chip manufacturing company.
I’d love this to be true but the tech involved is Sci-fi level stuff. Neutron beams used to chop off atomically perfect slices of giant silicon crystals and wacky stuff like that.
TBF garage fabs -are- a thing but it’s in the hundreds of nanometers scale. Thin film technology is also promising for low tech tape outs, but neither of those is going to be practical for anything better than 1980/90s tech. A modern die would be in the square meters range on those process densities, and could never achieve ghz speeds.
That said, there are a ton of scrappy companies sending out designs to 30-100nm scale fabs, companies with 5-10 employees cranking out cool designs and custom silicon… but they are still sending their tape-outs to giant companies to fab, just on their old, obsolete machines.
Silicon foundries are incredibly capital intensive, and short SOTA process lifespans burn through that investment at a frantic pace.
Sadly no. There isn't really a single person who understands the entire SOTA chip fabrication process in enough detail. Think thousands of material science PhDs with master and apprentice style relationships inserted at every level of a massive tech tree.
It's not like you can just look at the plans for a chip fab and copy/paste it into a new location and hire people to fill in who will have any idea how to work it.
I think one of the people who got closest to that was Sam Zeloof.
He kind of had everything going: extremely clever and motivated, cooperation with his parents (who also worked in the industry), access to equipment. Kind of hard to improve on that.
He was able to replicate most of Intel's SOTA process... from 50 years ago. That's more than almost everyone else has managed in their garage but that's about the best you can expect without insane capex and ramp up (and again, it's not like he didn't have access to capital, it just wasn't monetary.) Even still it took five or so years to work everything out.
The SOTA today is really kind of insane. It's right at the frontier of what all of humanity is capable of. Of course as time goes on we'll push that out and today's SOTA is tomorrow's commodity but that won't change everyone's concern with being unable to replicate the contemporary best process.
The reality is all our "defense" needs (and arguably most other needs too) are far more than adequately met with processes a decade old now. It's really not the big deal everyone makes it out to be.
> The reality is all our "defense" needs (and arguably most other needs too) are far more than adequately met with processes a decade old now. It's really not the big deal everyone makes it out to be.
Right, this is why I think in-sourcing chip manufacture is totally viable (that is, if we were actually interested in that and not just using it as an excuse for corruption). The interesting exceptions I've heard about are things like, IIRC, high-power local AI for autonomous drones. But for SAMs and such, old tech will probably do it.
It is not a straw man.
There is no amount of scrappy cleverness that gets you from zero to manufacturing cutting-edge chips without shitloads of capital investment, years/decades of R&D, a huge manufacturing workforce, and big contracts.
There's no such thing as starting small and scaling in that business.
You don't think $8.9B would do it?
6 replies →
There is no such thing as a “smallish” chip manufacturer that can manufacture leading edge chips. It’s about scale.
If it were that easy, Apple, Amazon, Google AMD, Nvidia, etc who all design their own chips would have done it.
I agree but by the same logic intel could have done it and didn’t manage to so far.
Flip side: why would Apple, Amazon, Google, AMD, NVIDIA etc build their own when they can outsource it cheaper?
Companies are run to make a profit… they don’t care about sovereignty as long as the money is coming in.
1 reply →
[dead]