[rivo]

A few highlights from a conversation with a friend who works for Zeiss and regularly deals with ASML (take this with a grain of salt, I may have misunderstood some of it and he also may not be an expert in all of these areas):

- The lenses (actually mirrors for EUV) only demagnify by a factor of 4. So your wafer template is already extremely small and costs millions to produce.

- 13.5nm is pretty much the smallest wavelength you can reasonably handle. If you want to go smaller, you have to build a particle accelerator. (He heard that from a colleague, wasn't sure how much of that was actually true.)

- The prices in this Twitter thread are pretty spot on. The EU wants to invest into chip making to become more independent in this area. But the sums they're talking about wouldn't even pay for one such machine. Politicians don't seem to be aware of the dimensions we're talking about here.

- Also in the thread, ASML has a monopoly on this tech. Others have not invested into EUV and by now, it's pretty much impossible for anybody to catch up.

- Everyone's hiring like crazy atm. He started home office during Covid and now Zeiss says he can't even go back to the office because they don't have enough office space anymore. He doesn't mind. Everyone's distributed all over the place anyway so whether he does conference calls at the office or at home makes no difference.

- He said they're not very affected by the chip shortage.

[cromwellian]

> ASML has a monopoly on this tech

A lot of this tech is the result of more than 30 years of research, starting with national labs (DOE funded Sandia, LLNL, etc), then consortiums of Intel, AMD, etc (EUV LLC), in fact, ASML was a member of EUV LLV and specifically, because some of the techniques they're using was originally pioneered by DOE Labs and EUV LLC members, they had to get approval from DOE/EUV AFAIK.

Also, the core laser tech, tin droplets, was actually pioneered by Cymer, but they bought Cymer wholesale. So part of the issue is they vertically integrated a lot of the very difficult components of their machine by buying smaller companies.

This would be like TSMC buying ASML and preventing other fabs from getting access. It's definitely bad for the market, and EUV is so damn expensive it took 30 years of R&D and a consortium of companies to even get to this point. The competing technologies, using different Lasers, would need enormous funding to get them to work.

EUV LLC built a 100-nm EUV prototype almost 20 years ago using a different technique, but gave up on that path. There's also electron beam lithography (EBL) which is maskless AFAIK, but I don't think it ever made it to production either. I guess Intel and other players decided it was better to just own ASML stock.

So the world seems stuck with ASML, unless of course, China gets a hold of one of the machines and decides to just clone it and ignore patents.

[rasz]

China has one of the older ASML DUV machines, they got it thru a scam :o) and proceeded to pawn it for $90mil to a local bank for a quick scam exit https://chinatalk.substack.com/p/billion-dollar-heist-how-sc... https://interconnected.blog/chinas-semiconductor-theranos-hs...

but even non scam Chinese initiatives are failing https://techtaiwan.com/20210714/tsinghua-unigroup-bankruptcy...

[janmo]

> - Also in the thread, ASML has a monopoly on this tech. Others have not invested into EUV and by now, it's pretty much impossible for anybody to catch up.

Having basically the entire tech economy relying on one unique supplier is just insane. I hope their production facilities are very well protected and decentralised otherwise one natural disaster in the Netherlands or an Unabomber type terrorist attack and the semiconductor shortage we are facing now would be nothing compared to what would happen then.

[parsecs]

Except I'd imagine that most people won't feel the terrorist attack quite so quickly, giving folks likely enough time to recover. If they got bombed up, all the foundries would be fine and can continue production of chips as usual. Of course, the machinery need sophisticated maintenance and replacement parts, but most of those can be produced by non specialist equipment (like normal mills and lathes and stuff for all the mechanical structures). Zeiss plays a big part with the optics and stuff so that would mostly be covered as well.

[jychang]

There won't be any new foundries though. Moore's law would grind to a halt (moreso than it already has).

[parsecs]

I think the key is that, the machines that make extremely advanced machines are not that advanced in of themselves.

You don't have to go down that many layers from the machines that foundries use (arguably objects among the very apex of human technology) to bog standard fabrication machines like lathes and mills and whatever they use to cast metal. We have lots of those around the world. It will definitely take a bit of time, but I don't think it will be all that utterly devastating.

[pms]

Won't Netherlands be under water because of global warming?

[achow]

Aren't they system integrator and get their supplies from thousands of vendors worldwide. That actually makes them disaster proof.

[cutemonster]

If so, I wonder how many (if any) of those vendors are single points of failures?

> Aren't they system integrator

But the tweets mentioned ASML bought suppliers with unique technology that no one else has

[zsmi]

> Also in the thread, ASML has a monopoly on this tech. Others have not invested into EUV and by now, it's pretty much impossible for anybody to catch up.

This is essentially true, but people need to keep in mind this is not like when we say Apple has a monopoly, or something.

The capital equipment market is very different than consumer markets. For example in 2019 it's estimated ASML sold 26 EUV machines, in total. Analysis are super excited because this might go to 41 in 2021.

https://blog.visiblealpha.com/asml-euv-monopoly-may-stack-ch...

[BenoitP]

> - He said they're not very affected by the chip shortage.

You bet! TSMC/Intel/Samsung must give them for free; also give them the first of each batch they make, as samples to be analyzed.

[bradknowles]

ASML makes the machines that make the chips. They can do their own runs of whatever chips they need, because they have to fully test the multi-million dollar machines before they disassemble them and ship them out.

[atq2119]

That's not how it works. The light source and steppers are important, yes, but there's much more to a fab. Think of all the chemistry, for one thing.

What you claim is basically like saying a robotics company could just build their own trucks, because the truck companies use their robots for the truck factories.

[bradknowles]

They have to fully and completely test out the machines before they ship them out.

So, yes — they do have their own fabs in house. They built them.

Disclaimer: I worked a 6 month contract for ASML in Eindhoven, and I talked extensively with their engineers. They told me what they did as part of their testing process.

[hvidgaard]

Sure they could. But someone still need to design the chips and take them from silicon to a packaged chip. They obviously do that to a limited degree, but a test run to see if everything is within spec is very different to producing actual usable finished chips. Or at least they don't have to go the full on complex state of the art chip design just for testing.

[bradknowles]

They have the masks and all the chemicals and other materials, equipment, and devices necessary to run the machines that they build. They get masks and other information as needed from their customers. They have to be able to demonstrate full end-to-end performance of the machine to the satisfaction of the customers, before the machines can then be disassembled and shipped out.

They make very few machines per year (measured in double digits), and each one takes a team of people many months to make.

Each team spends months on-site at the customer to prepare for the machine to be made, then goes back to Eindhoven to actually build it (which takes months), then it gets disassembled and shipped out to the site, then the team goes back to the site and spends more months rebuilding the machine and tweaking it for final operations. The whole process takes years to go from start to finish. That’s for one machine. And if you want to move it a foot to the left, then you’ve got to go back and tear it down and go through the rebuild process all over again. These machines are that sensitive. Just you walking into the clean room changes the atmosphere where the machines are running, and risks causing them to become misaligned and need recalibration.

And that’s for the regular non-EUV machines. Each one of these costs about 50 million bucks or more, and that’s just for the core machine itself, and not all the ancillary equipment that is needed. EUV machines cost two to three times that amount.

The technology to do EUV depends much more on things that are not chips, than they do cutting edge chips that could only be manufactured by the likes of TSMC. Their actual requirements for the chips themselves that go inside the machines are relatively low. With ASML, It’s the physical engineering processes that are truly superhuman.

So, they can easily make the chips in-house that they need. It’s more a matter of whether they want to do such a small run to get the chips they need, because they don’t need that many chips per machine, and they make just a tiny handful of machines each year.

OTOH, they can also easily get supplies from anyone they want, because of their critical position in the chain.

[InvaderFizz]

> - He said they're not very affected by the chip shortage

My guess as to why is that they probably have a very predictable/steady chip supply pipeline and are low enough volume and high enough margin that they can easily outbid anyone and it wouldn't have a meaningful impact on the bottom line.

[cutemonster]

I guess they can get good deals with their customers too,... Edit: see another comment here: https://news.ycombinator.com/item?id=28270545

[detaro]

> Others have not invested into EUV

Not entirely correct AFAIK. EUV is something the industry pretty widely worked on, ASML just have the best tech in the end.

[01100011]

> you have to build a particle accelerator

That's not exactly difficult. Particle accelerators are all over these days, from medical applications(radiotherapy with liniacs) to materials inspection(portable cyclotrons).

[s0rce]

X-ray producing synchrotrons are generally pretty large and expensive. There has been thoughts of using them for lithography and microfabrication in the past (see LIGA) that haven't really panned out. There are companies working on more advanced laboratory sources, but mostly for analytical applications (EasyXAFS, Sigray). There is a company working on a lab scale synchrotron (lyncean tech).

[01100011]

Do you need to use a synchrotron? Is the radiation generated coherent or something? Otherwise why not use a liniac or cyclotron?

[tullatulla]

You would really need a free electron laser to produce sufficient photons at the desired wavelength.

The lower the wavelength, the less photons you have at constant power, so shot noise becomes an issue...