The entire (massive) CPU business hinges on these machines in 2021 and given that I’m quite certain ASML hold substantial patents it seems like they should be offering these machines for per-wafer fees rather than selling them directly to TSMC and Samsung for them to print endless mountains of cash with them.
AMD and Apple are utterly reliant on these devices, and no-one outside of tech even knows they exist. If no-one sees the massive value your company contributes that’s a problem with PR.
So, in practice, ASML and the few companies are dependent on each other and probably found local optimum of coexistence. ASML does not want to push several top-level tech companies into search for alternatives and the top level tech companies do not want to waste their talent on replication of an already existing product unless necessary.
They are also so far ahead, after uv it will be X-ray, and even electron etching. They are developing tech 10 years ago that still will get them further. At some point there will be enough power, for general consumer products to function just fine. Once a laptop can run email/web/word/excel smoothly, which I think is like 90% pc use, on a whole day battery. I think we don’t need more speed. It might be at that point, that some players step in. Try to build machines that work at similar quality. However, why not just get a second hand ASML? As far as I know once a new generation get bought the old goes to a lower tier processor manufacturers.
Or we can build a more resource hungry software.
Thats not how it works, software is a gas that expands to fill all the available performance space until it all slows down and we need better hardware again.
In which case you are trading human (software development) time for hardware resources.
If that were the case, it seems like an okay-ish trade off to me tbh.
Conceivably they could run the economically important IT functions of the country using less powerful, easier to manufacture, less power hungry greener, and readily available semiconductors.
This feels like a bad deal for Apple.
If you wait - you’ll get a 20x machine at current prices.
This is what Moore’s law is about ? I.e. Density (~= proxy for compute) doubling at some fixed period.
It sounds like you'd be stuck producing old designs on those old machines.
Apparently, that equipment rarely gets sold off or excessed. Those machine keep producing for years and years and decades even, for the same companies, in the same fabs, in the same location. Even as they get older and older, they’re still so profitable to keep running that it almost never makes sense to turn them off or sell them to someone else.
They were actually quite proud of the fact that they had lots of old hardware in the field that was still running and producing wafers for the same customer so many years later. They were a little taken aback by what it cost them to keep those old systems around themselves, so that they could provide support for them in the field. But that’s what the customers were paying for, and so that’s what ASML did.
In an investor meeting, ASML told its audience that TSMC knows more about how to operate their EUV Twinscan then they do.
And no, Intel, Samsung, and TSMC are no longer ASML's shareholder, as somehow constantly being pointed out in various form of social media including HN. They only invested in I think 2011 / 2012 to help fund the development of EUV. They have since reduced their investment once this was done.
It is also funny you mention this, because I am sure if ASML were an American company they will absolutely do that to maximise their profits. But ASML is Dutch. And ASML got to where they are today with the help of TSMC. It used to a fight between Canon and Nikon. But that is another story.
ASML doesn't hold the core patents. The U.S. Department of Energy allowed ASML to join the EUV LLC. ASML bought Cymer for exactly that reason.
The other issue is that $150M for a stepper is more than an order of magnitude more costly than the previous stepper technology used for 20+ years. To make that work, very expensive CAD tools were required to grossly violate normal diffraction laws (but obviously not the laws of physics themselves). This is done with "multi-patterning" which uses multiple exposures of the same targets to attain sub-diffraction limit dimensions. It requires many additional process steps to do this and it's mathematically/CAD intense.
And finally for fabs, the throughput of ASML EUV is about 1/50th of the prior steppers (but the previous steppers required 5-10 additional multi-patterning stepper exposure steps in the process to attain similar results. The general rule of thumb is an additional process step is 1/2 day of throughput so the old technology would add a week or two of throughput but the lower throughput of EUV only saves about 50% of that so the throughput is worse or a wash.
As it is, the wavelength of EUV is 13.5 nm so the diffraction limit is still only 6.25 nm feature size (so 7 nm is clean without multi-patterning but below that you need to return to multi-patterning but not as many steps added (yet).
The "Net Net" is that yes EUV is the only way get deep nanometer dimensions (<25 nm) easily and cheaply BUT EUV is economically marginal for both stepper vendor and fab.
The former is why no competitor wants to jump in (plus patents as well). The latter is why it's certainly necessary if the commitment to deep nanometer is life-or-death for the fab. A LOT of fabs have decided that there are better markets above those dimensions that they can tap so they've written-off deep nanometer and switched orthogonal to different market value propositions.
Fun Fact 1: another common name of EUV is "soft X-rays". As such, conventional lens-based optics DOES NOT WORK, so the ASML system uses the only technology that works for X-ray optics: low-angle, glancing metal mirrors. A well known other application of this is X-ray telescopes in space.
Fun Fact 2: the optical power level at the mask is less than 10 watts which is barely enough to expose the photoresist. This requires far longer exposures with perfect mechanical stability to hold the mask-created image exactly still. This is why the throughput is so low compared to prior steppers. This is also where much of the capital cost comes from: stabilizing mechanical components far beyond what prior technologies required.
Massive capital investment and extraordinary long term planning are the things that are sometimes considered to be difficult for the private sector.
Same for much of US (and elsewhere's) private aerospace industries etc.
As the article points out, this company has been pressured at geopolitical scales to deny China access. The fact the US government recognizes the value in denying China access because competition becomes impractical says a lot. China is filled with technical and engineering talent no different than the rest of the world and has its fair share of capital as an economic powerhouse. There's a lot of incentives for the government and people of China to develop their own competitor but it's recognized that this is impractical due to the level of proprietary knowledge, instruments, and skills needed to recreate the device. I'm confident they can produce a competitor if they choose but it'll be decades without any real competitor.
Now pretend you're not the PRC but Joe Sixpack in the US and you want to enter the market. We need to drop these ridiculous ideas of competition regulating the markets. Even injecting VC money here is probably at irrelevant scale. I'm all for competition but we need to look for new models of market regulation. Competition just isn't it in many cases.