HN Theater @HNTheaterMonth

Cant answer the question and my knowledge is only from 'one interested' but from how long (~20 years) it took to build the current EUV source and how complicated the alternatives look this is for sure no easy feat.

Can recommend [1] as the title say about the light source. Very interesting even for a non professional (and [2] as on high level overview of the machine and [3] as longer documentation inside ASML).

[1] https://www.youtube.com/watch?v=5Ge2RcvDlgw "The Extreme Engineering of ASML’s EUV Light Source"

[2] https://www.youtube.com/watch?v=f0gMdGrVteI "The Extreme Physics Pushing Moore’s Law to the Next Level"

[3] https://www.youtube.com/watch?v=zQu_TMgHO98 "ASML's Secret: An exclusive view from inside the global semiconductor giant"

⬐ Beldin

As a student, I did a tour of ASML. One of the coolest things was one of the tour guides bragging how they would inspect the prospective location for a new plant and deny sale if they didn't like it.

Why? Because their machines are incredibly sensitive... and equally expensive. And your customers get angry if the machine doesn't work reliably - and rightly so.

I don't know if they do that still, and it likely was somewhat hyperbole, but the core of it is actually correct: the machine is indeed that sensitive, and is indeed that expensive. Every second of lost operation is expensive, and creating a fab around one of them is not quite as straightforward as just pouring some concrete somewhere and adding walls and a roof.

⬐ AceJohnny2

I recall when STMicroelectronics built their new fab [1] near Grenoble, France, it was built on a kind of stilt-and-spring system to insulate it from vibrations from a nearby highway.

(and similarly, the nearby ILL Synchotron [2] had issues that were eventually tracked to trucks passing on its nearby highway)

[1] https://goo.gl/maps/jNESRpcdMkAtVURM9

[2] https://goo.gl/maps/4oXRXw6pxkxMFCUt5 Note that the satellite view is blurred because that whole area technically belongs to the french CEA, their Nuclear Administration!

⬐ lights0123

It's interesting to see that Mapbox blurs its highest zoom level there, but it's second-highest zoom still provides quite a lot of detail to the point where you can make out every car and their color and rough shape.

⬐ Powerjaxx

For Those looking to learn more about ASML. The book ASML's architects by Rene Raaijmakers is a great book! (FYI: I've read the Dutch version, so I am not sure how good the translated version is)

⬐ tullatulla

Can also highly recommend the translated version!

⬐ churnedgodotdev

ASML is the closest thing to the planet Dune in our world. Chips made with their machines extend life. Chips made with their machines expand consciousness. And he who controls the chips, controls the universe. The information universe, at least. I just wish for a messiah to arrive who leads the jihad against FAANG. The Kwisatz-RISC-V-Haderach. The Voice from the Open Source World. Fortunately, he need not fell Mark Zuckerberg in single combat with crysknife in order to lead our people to true freedom.

⬐ robbiet480

Asionometry has a fantastic series all about ASML that I highly recommend, starting with this video https://youtu.be/CFsn1CUyXWs

⬐ sylware

Well, the video does not show the ppl actually designing the blueprint of the machine and doing the R&D, the real brains... we are showed mostly assembly workers (all above averagely skilled), customer service operators, etc.

I was surprised to agree with most of what ASML bosses said, and I would have given exactly the same answer than the assembly operator, namely not providing any timeline... because it is complete garbage to do so on so much complex machines, not to mention the machine is a prototype, then the first one. I may have answered "yesterday".

But noway they will make me use a doz computer, or an android phone...

I am _seriously_ curious on how they want to get their pitch below 7nm with a 13.6nm EUV light. Do they plan to go massive multi-patterning with sub-nm xray positioning? For below 7nm pitch, what would they use: xray light? electron/ion beams? pattern printing? And what about the purity of the wafer dopped silicon for smaller features? xray light: what materials to use?? Quantum interference tricks?

I wonder how many silicon atoms from a wafer crystal we have on a 1nm row...

All that is to get chips which consume less energy, get faster telecom, play magnificient games, store more in flash ram.

⬐ mike_hearn

The documentary spends a lot of time interviewing their head of R&D who is described as the 'mad professor', he talks extensively about the problem of the tin exploding onto the mirror and how he encountered that problem etc, so presumably he should count as someone doing R&D for real.

⬐ sylware

The head of R&D, but he is the only "real brain" who is presented on the show. Could have presented maybe 1 other "real brain".

I have heard about "the 2 times lazered tin droplet to generate "enough" EUV light" already in an older ASML documentary (years ago).

⬐ hnuser123456

If I understand correctly, chips down to 14nm were made using ~190nm light, not sure how that worked but if that ratio of 10x+ holds, seems like 1nm fab using 13nm light should be doable with the same techniques, not to mention improved ones?

⬐ sylware

Idem, if I understand correctly, to go below the light pitch with DUV litographers, multi-patterning had to be used: a huge burden at chip design time and "pressure" on nm-grade alignment equipment (heard about coupled piezo electric motors with xray interferometers?), which is supposed to be from Taiwan.

In other words: photomask madness.

Photomask manufacturing is another critical part of silicium chip manufacturing. It seems EUV photomask are special (no more transparent crystals, but instead reflective "patterned" surfaces?), wonder if they are still using electron/ion beams to engrave patterns.

Wonder where are manufactured EUV photomasks, heard about Japan a lot.

Oh, and I realised that I am really curious on how they keep the EUV mirrors clean from the tin droplets :) Wild guess: EUV mirrors far away, H2 gas reaction, etc?).

⬐ MayeulC

7nm feature size or process? 7nm processes are on ~20 nm feature sizes, but have tighter alignment constraints, as they move into the 3D world.

⬐ adrian_b

The so-called "5 nm" CMOS processes have minimum pitches around 30 nm.

I do not know what minimum pitches are planned for the future so-called "2 nm" CMOS processes, but it is likely that they would not be smaller than 15 to 20 nm.

So there will be some time until multiple patterning could be needed again, and by that time there are chances that the transistors will have minimum sizes determined by other causes than lithography, so for further progress a switch to different semiconductor materials will be needed, not a further improvement of lithography.

⬐ MayeulC

Gate length is already determined by tunneling current, you can't really go below 20 nm on Si. Width is determined by the current density you want, and can be reduced with better electrostatic control (FDSOI, FinFET, GAA...)

⬐ sylware

EUV High NA can "draw" 7nm features, with heavy design constraints and alignment equipment, they can sort of go "below", but even with new electrostatic transistor design, they are sort of limited at 20nm for the gate lenght?

Then the litographer does not seem to be the limiting equipment anymore. I guess there is heavy R&D going into electrostatic transistor design and crystals to deal with the gate current below 20nm.

I wonder at which size, silicium doping does not work "enough" anymore.

⬐ MayeulC

At some point the electons start tunneling through the channel, regardless of its state... I don't remember my semiconductor theory that well, but possibly unaffected by the height of the potential barrier? In which case better electrostatic control or changing materials does nothing. Hot carriers (high voltages) are more affected, so you can lower voltages to an extent.

You can go below 20 nm, but you get this constant tunneling current, which deteriorates your on/off ratio, and increases power consumption (and TDP). It's doable if you can turn that part of the die off (dark silicon).

⬐ sylware

Alright, then if they manage to handle properly this tunneling current in order to get clean on/off, they still will be able to shrink the current transistor feature size... which is actually huge if we don't think about the marketing "Xnm lie", unless another limiting factor hits.

In the silicon crystal lattice, without considering "doping", the atomic valence lenght is 2.35 angstrom or 0.254 nm, and unit lenght 5.44 angstrom or 0.544 nm.

So a feature of 7nm is hardly 14 silicon crystal lattice units. With doping, I wonder if it will be ever required to have a better pitch to have something to work with silicon, unless near "atomic perfect" doping in near "atomic perfect" silicon crystals can still be used for on/off gates.

It makes me pessimistic on the amount of remaining transistor shrink steps. "2nm" maybe "1nm" then you would need to assemble atom per atom perfect lattices frozen in time.

⬐ tablespoon

What's the purpose of this corridor the workers walk through? Some kind of cleaning process, or a security scan to make sure they're not bringing in unauthorized devices?

https://youtu.be/zQu_TMgHO98?t=34

⬐ bob1029

It's an air shower designed to get any last bits off your bunny suit before you go into the cleanroom.

⬐ wofo

Hi there! This is a totally off-topic comment, but since there is no way in HN to send private messages, I though I might as well reply to your most recent comment. I have been enjoying your comments about being a craftsman and working in a niche, and I think I am heading in the same direction. I could use some advice and would love to ask you a few questions, if you don't mind (email is in my profile).

⬐ throwie_wayward

I absolutely despise the ancient traditional ways they practice: referring to their techniques of 'mystery', secrecy, and occultation of knowledge (which they do in order to construct knowledge as authority and then wield it as power)

they've overdone it. by this point they have made it way to fucking difficult for people to truly come to grips with the complexity of modern life mostly becuase in their zealous protection of 'technical knowledge' they've made a life such as my own impossible to realize (this in turn, stunts the advancement and development of ways to cope with the consequences of modernity and high-technology in the culture at large, becuase people cannot learn to undesrtand enough that they can figure it out; they move us to a scenario in which only their own chosen selected few are in a position from which something could be done; but they select for obedience and dullness; not for spark)

Maybe my reasoning is leaping a lot; but as I see things, this attitude towards knowledge and understanding is the cause of the now troubling distrust in the academic-scientific apparatus by a largely ignorant public (an ignorance which is somebody's profit).

this attitude has stunted my own personal pursuit of happiness. that I cannot easily come to understand their 'propietary' techniques (higher-order logical systems, advanced litography, etc) that they use to do what they do is just frustrating.

I'll end by adding that there are many more instances of this attitude around knowledge in other subjects I've studied... even some which are no longer considered 'sacred' (in more 'developed' countries) but which were still making people act all mysterious and in awe of 'the knowledge' in my own famously ignorant native country.

⬐ mensetmanusman

What you are describing is tacit knowledge. Something obvious to the workers- but not to outsiders.

Tacit knowledge is hard to pass down in written form because a reader would be overwhelmed by details. It is easier to pass down over video, but video loses information about smell, vibration, static charge, etc. that the human body can feel and learn from.

⬐ jdmdmdmdmd

>construct knowledge as authority

Can you explain why you used the word construct here?

⬐ throwie_wayward

because 'knowledge' on its own will not work to wield power over others just by itself. there are other ingredients in ensuring that knowledge can be used as power; these usually look like 'technology'.

I should also say that it's only due to the internet (and its novel aggregation-theory dynamics) that these old attitudes I am ranting against become truly noxious.

⬐ skrebbel

I have no clue what you’re on about. ASML has hardly any culture of secrecy to speak of, or at least in comparison to famously secretive companies like Apple.

Talk to a random ASML engineer at a party and they’ll enthusiastically tell you everything about what they’re working on.

⬐ rjzzleep

To summarize the video the secret is just how the people collaborate. Some parts seem oddly manual and others interestingly esoteric. Interesting little detail is how that researcher knows the political game of not presenting to Intel but instead having the VP himself present.

At least there is a lesson for society here about how to keep people together instead of dividing them. I'm both fascinated by the company, and like others in the company drawn by the technology, but also wonder if I would have a problem working in this cult like setting.

⬐ doikor

> Some parts seem oddly manual and others interestingly esoteric.

They don't really make that many machine for heavy automation to make much sense.

By the end of 2021 they had made ~140 EUV machines in total with commercial production starting in 2011. So if you average it out they make ~15 per year. Though the production rate has most likely ramped up over the years so most of those have been built closer to today then 2011.

And in those ~140 EUV machines there is many different versions with improvements. So you would have had to redo your automation lines multiple times over the process of building them.

In total they make around 300 machines per year at the moment but only a fraction of them are the highest end ones.

⬐ rjzzleep

baybal2 you seem to be shadow banned.

⬐ baybal2

The bottleneck is not TSMC, but Zeiss which makes the mirrors "more perfect than that of space telescopes"

The yield on those mirrors is very small. They throw away 19 out of 20 mirrors.

⬐ rjzzleep

Maybe, but then again the demand is far higher than what they can deliver and their solution is to hire thousands more people. While that’s a simple scaling mechanism they were limited by Covid restrictions and will probably be limited by the worsening energy crisis

⬐ himlion

They are also very much limited by the housing crisis in the Netherlands. They are perpetually looking for places to house their employees.

⬐ doikor

Even if they 10x their production you are talking about 200 of the same machine max per year. Nowhere near the numbers where proper industrial automation would make much sense.

Also who would be buying these machines if they 10x their production? They go into fabs that costs 10 to 20 billion to build. Let’s say you put 10 of them in each fab. It would still require 150 billion in fab investments every year. There is demand but not that much demand (or capacity to build the fabs)

⬐ jve

All these numbers remind me how people looked at rocketry pre-SpaceX. "Oh, the most complex, expensive, etc thing in the world" (Rightly so)

And then they come in to commoditize the market, make it accessible for lower players by aggressively trying to bring down cost-per-unit via simplicity and process automation.

Not saying it could apply to ASML, but I wonder... :)

⬐ doikor

Lithography equipment isn't some government funded cost+ industry with 0 innovation or competition like what SpaceX had (has?) as its competition. Nobody is using equipment from 70s (Soyuz) or even early 2000s (Atlas or Delta).

It is actually the opposite. Everything is really expensive because you have to be constantly making the newest/greatest thing. A 5 year old lithography machine is pretty much useless for manufacturing high end chips (there is its own market for their output but phones and computers are not it)

Basically if you tried to do the SpaceX way of making a "good enough" rocket (Falcon 9) and use it for a decade+ you would be bankrupt very fast. Nobody wants to pay top dollar for old tech.

edit:

> aggressively trying to bring down cost-per-unit

And what makes you think ASML is not trying to do that? Every cent they save on cost-per-unit increases their profit margin.

⬐ bradknowles

I worked at ASML for several months in the early 2000s. One of the most surprising things to me was the very long tail they had for old equipment in the field that was still being actively used, and therefore had to be actively supported. Like, stuff that was ten or even twenty years old. I seriously doubt that this situation has materially changed.