Correction, that's a tour of the Cymer EUV light source facility, the secret sauce that makes ASML machines the hot shit. That light source gets integrated into the ASML stepper machines.
Cannot recommend it enough, fair warning though, I have realized I watched a few videos too many on more than one occasion, Asianometry becomes a dangerous time sink way too easily.
Is the blur on the videos enough to keep the proprietary components concealed? Seems like a state actor could easily reverse this filter and learn the underlying shapes.
How much isn't revealed by this tour? This seems like it's giving away quite a lot of the EUV process. Or has all of the content of the video been publicly known (or stolen previously)?
I get that this might excite young people into entering the field, but if the US is trying to gatekeep advanced processors, this feels like showing off a lot of the secret sauce.
How long do we think it'll take a foreign actor to replicate EUV, and how far along do we think we'll be into the next process by then?
Pretty much all of this has been public knowledge for a decade, or even longer.
The technology behind EUV is quite simple on paper, the hard part is getting it to work reliably. It would take you years to reverse engineer it even if you had unlimited physical access to a working machine, and all the truly secret sauce is in all the measurement and calibration equipment around it.
> The technology behind EUV is quite simple on paper, the hard part is getting it to work reliably.
EUV machines are a terrible approach to the problem made to work through a huge amount of effort. Look at the size of those things. All to get a little burst of soft X-rays and focus it on a mask and wafer. With a light source that works by using lasers to vaporize a droplet of tin falling in vacuum. Creating an internal mess that has to be cleaned up.
For a decade, people in wafer fab hoped there would be some other way to do this. Small synchrotrons. X-ray lasers. E-beam machines. But those approaches all fizzled out. So what ought to be a physics experiment run occasionally was hammered into being a production technology.
Examples of a commercial technology that hard which went mainstream are rare. There was the image orthicon.[1] That's a TV camera tube. Previous technologies needed very bright light. Zworklin's iconoscope accumulated the image over an entire frame time, but photon to electron conversion was very inefficient. Farnsworth's image dissector had much more efficient photon to electron conversion, because it had a photomultiplier built into the tube, but only the light received from the pixel currently under the scanning beam contributed to the output - there was no full-frame accumulation.
So RCA developed the image orthicon, which combined both of those technologies. It was insanely complicated and expensive. The tube alone cost about US$10,000 in the 1950. Color required three tubes. The tubes required something like ten different power supplies, and control and alignment required hundreds of adjustments. Here's someone trying to restore one.[2] All those racks of equipment for one camera.
Does anyone know why synchrotron light sources didn't/don't work out for EUV generation? Certainly they seem affordable at fab scale (which is mind-blowing in and of itself) so there must be some other reason.
Not sure. As an experiment, part of the Stanford Linear Accelerator was used to expose some wafers. It worked. Design paper from China on a synchrotron ring for EUV.[1] Note lack of info about ring dimensions. But it looks like hundreds of meters, at least. I think it's been tried with a large research synchrotron ring. But for now, apparently building a synchrotron is even more expensive than those AMSL nightmares. Any news on progress from the VCs? If someone can make this work cost-effectively it will be very profitable.
That’s it in a nutshell. The current cost of a Synchrotron is more then the cost of the equivalent capability using non-synchrotron technologies.
It might seem crazy but it turns out Synchrotrons are pretty rare machines too… and they are less “operationalised” in an industrial sense than the crazy tech inside the EUV machines… between the vacuum levels needed and the various linear or circular accelerator technologies available, at the end of the day it’s just way more of a pain in the ass to maintain a large industrially pure vacuum machine operating at continuous round the clock fashion. which might seem nuts but talk to anyone who’s works with high vacuum systems about vacuum purity, out gassing and contamination and yeah… it will make sense pretty quickly.
I don't think high vacuum is the main problem. I used to work closely with people who did serious high vacuum heavy lifting (with RHIC at Brookhaven) and they always gave off the impression that it's plenty difficult, but it's engineering-project difficult, not research-project difficult. There are solutions, they are understood, and they work at scale. Though they might not be pretty.
No, I'm guessing it's something else. I can't imagine complexity is the issue, given the nightmare that is the ASML scanner, so I can only guess that some property of the light generated by synchrotrons is troublesome and not easily correctable.
If that's not the case... perhaps there really might be a business opportunity here?
IIRC Tsinghua SSMB ring design has circumference of ~140m for kW EUV light source. My understanding, at least how SSMB was/is being sold is: storage ring technology mature, price per tool / light source will be relative low.
> For a decade, people in wafer fab hoped there would be some other way to do this. Small synchrotrons. X-ray lasers. E-beam machines. But those approaches all fizzled out. So what ought to be a physics experiment run occasionally was hammered into being a production technology.
Now I'm imaging some retro-50s (ala Fallout) setting where the fancy chipmaking machines use bomb-pumped x-ray lasers. Swords into ploughshares taken too far? Maybe, but who cares!
For this to make sense you would need to expose multiple wafers on each detonation. But then you would need one mask per wafer, and masks are even harder to make than a chip (electron beam, ...)
Hi, I work in laser measurement and calibration. There is no secret sauce; you take a physical phenomenon that translates optical power or energy into an electrical signal, measure that signal, then find the sensitivity value of that phenomenon and account for any other effects such as ambient temperature, fill factor, linearity, parasitics etc. and then prove that your calculations are correct by comparing them with NIST/PTB.
We can’t make F1s because the supply chains that made the materials and the parts no longer exist. We could try to approximate one with new materials and techniques, but it would not be an F1 - it’d be a design derived from it.
That’s the thing with the chefs - you can’t make a mammoth dish because there are no more mammoth to cook.
Yes they do, the supply chains still exist, we still use many of those alloys, etc. We still make rocket engines. In the F1's case we were just going too fast and didn't write things down.
In the nuclear weapons case we deliberately dismantled it.
> In the nuclear weapons case we deliberately dismantled it.
I'd say that is actually a good thing. Nukes are meant not to be used. They are a deterrent, so it's important they exist, that they are functional, and that you are willing to use them to hit hard against an aggressor. MAD kept us alive until now.
Chinese fighter jet engines are supposedly already better than the best Russian jet engines they've been given, though. General perception is that the WS-10 is generally better than the AL-31FN, hence why they started replacing AL-31s with WS-10s, both in domestic and export aircraft. The WS-15, if it does get even close to it's design goals, would be significantly better than any Russian engine.
They seem to be struggling a lot with airliner engines, yes, but they've only recently even considered making them. Even then, the WS-20 has entered service, and if it matches claimed specifications, it's as good as anything Russia has either.
The recent leaked picture showing a J-20 with dual WS-15 puts their development on par with F-119/F-135. Lets see how quickly they can ramp up production. It seems they have fixed their supply chain issues.
It seems easier to make an engine that gives very high power at the expense of efficiency than an extremely efficient engine with reasonable power which is what airliners want.
In the Pacific, efficiency is the main goal for fighter jet engines. The goal of the WS-15 is to have F-35 level efficiency while being able to go relatively far in the supersonic regime and outrange every US plane.
And of course the engines for strategic lift planes are based around efficiency first.
I could believe the J-20 could match the F-35, and apparently the Air Force could believe it too because there’s also the F-22. And strategic lift planes use normal jet engines not supersonic ones so they face the same issue as Chinese airliners do.
The F-22's engine are not as good as the F-35 engine. It gets around that by simply being a larger plane and having a second engine.
Both modern fighter jets and airliners use turbofan engines, the difference is the bypass ratio (hence the size of the bypass blades). For both low and high bypass, Chinese engines are supposedly better than Russian engines. They aren't better than Western engines but they're not that far apart. As far as airliners, the airliner engine is a different design from the WS-20 because of different requirements. It started development far later than the WS-20 and it has been flight tested after only 5 years, which isn't bad at all.
In the sixteenth century Dutch merchants managed to bribe a Portuguese captain into drawing them some maps. This is why the Netherlands managed to reach Indonesia on the second expedition instead of taking the 50 years it took Portugal.
I was in a meeting with ASML a few months back and people were pretty much slack-jawed as they were taking us through some video and slides of their technology. It was definitely one of those "this is just science fiction" things.
Learn to ignore short-term market movements. End of story.
Honestly, modern-markets are just nuts short term, there are so many conflicting parties seeking short-term profits.
On the professional side you've got the hedge funds, the algo traders and people like that. They pretty much trade fully automated, so if their algorithm decided to sell ASML on the news after their text analysis of the RNS or whatever then that's what happens.
Then on the retail side you've got lots of places where you can go punt with leveraged trades (CFDs and spread bets). If you're A-book retail the broker will like-for-like replicate what you're doing on the market. If you're B-book retail (most people !) then your broker will take a position on the market based upon their net view and their appetite for risk.
In addition to the above, the final piece of the puzzle is that moves in "hot" stocks are always amplified due to the sheer volume. Nobody (should !) blink twice if a FAANG drops 4% in a day and its sort of the same thing with ASML.
Likely the same thing that's caused Apple's stock to drop significantly yesterday and today - China has banned iPhones for government officials at work.
While that in and of itself isn't a huge problem, the broader concern is that this is part of a government effort to push homegrown phones (and chips). That's compounded by the fact that Huawei just revealed a phone with a 7nm chip, which is technology they weren't expected to have.
If the Chinese population moves to Chinese-made phones with Chinese-made chips, it'll certainly have a negative impact on ASML.
While a common misconception it is not actually unexpected news, SMIC was reported to be able to reach 7nm years ago and their capabilities mostly track with those expectations over the years (not an expert in any way, but I have read those reports for the last 5 years).
What was blocked through sanctions is the EUV machines, not the DUV. And then there is plenty of reports of China stealing tech from TSMC/Samsung/others. And both TSMC and Samsung have done 7nm with DUV before. So this was expected.
To give some additional context - actual fully native capability of SMIC up until very recently has been 28nm, with this copied tech they can do a replica of TSMC 7nm. But that's it, without EUV it is very unlikely they can go much further with this. All of the new nodes that will be coming soon are to the best of our current understanding physically impossible to achieve with DUV.
Actually Korean industry experts expect that SMIC will probably be able to create 3nm and 5nm without access to EUV[1] machines. This guy claims that 7nm is the last feasible node with DUV only[2], but what does China care about the infeasibility of the node? They have over a billion people produce an order of magnitude more engineers per year than we do and print their own money.
DUV is blocked through sanctions as well. Please don't make misleading claims. The ban also goes way beyond just lithography and includes a lot of other tooling. On top of that the sanctions seem to have been extended to not only block access to the machines but also to after purchase servicing, which actually led China to threaten ASML to buy back the machines should that be the case.
And yes TSMC was indeed producing 7nm using DUV before using EUV, but that doesn't mean that our wonderfully talented politicians and think tanks understand any of that given that just last year these geniuses were gloating that the export controls would throw back Chinas access to advanced chips decades.
>They have over a billion people produce an order of magnitude more engineers per year than we do and print their own money
The semis industry is truly a "globalism" industry
ASML is a system integrator. It uses laser from cymer, glass from zeiss...etc.
For fab like TSMC to make chip require TSMC to develop a node process, chemicals from Japan (the trade spat between South Korea and Japan result in Samsung executes going to Japan) machine from various vendors like ASML, applied material..etc.
i don't think one country can do it all by itself. Not US, Japan, Dutch or China...etc.
>a billion people produce an order of magnitude more engineers per year
Is this quantity over quality? i was educated in Taiwan up to high school before i come to States. Let me tell you, the way they educated student is call "stuffing the duck". They spoon-feed you info until you're "full" (remember the answers). That kind of education does not produce Steve Jobs and the cram school is everywhere in Asia.
>i don't think one country can do it all by itself
Decades ago US basically wholey dominated semi and it was strategic lapse in judgement that industry has globalized so much, away from US control, something they're looking to regain and technically could if semi wasn't so entangled into US east asian security architecture. Current western semi industry is basically US+JP+DE+NL+SKR+TW. Aggregated effort of ~700M, where US is half that. And in US it's not well compensated job where industry had/has first dibs on talent. Nor extremely high industrial policy except in East Asia, but in general it hasn't gotten strategic priority until recently.
At the end of the day, PRC large enough that talent generation can match/exceed in terms of quantity. In terms of talent quality, TW, SKR, JP all have rote education which was still enough for them excel in different aspects of the supply chain, from equipment, to production, to inputs. Meanwhile even Chaebol and Zaibatsu culture was enough to produce absolute electronic giants. How much of Steve Job's vision was built by East Asian talent? Statistics alone will create outlier Steve Jobs, the hard part is building large pool of technical talent. Looking at current trends PRC pretty much only actor on trend to address semi/IC industry shortage by end of decade (most countries above projected to 100-300k shortage). PRC currently pumping about 30k IC graduates a year, still 200k short, 520k/720k out of # of IC talent 2018 white paper identified PRC would need for complete indigenous supply chain. Semi is a hard and large sector (like aviation), but (IMO) "do it yourself" is likely viable with talent base and proper industrial/talent policy. For reference US aerospace has about 700k jobs. They're currently the only major "do it yourself" aerospace power - they have the bodies to do it all essentially indigenously, vs next competitor, EU who has to work as a bloc.
> i don't think one country can do it all by itself. Not US, Japan, Dutch or China.
That is indeed what most people and I used to think, so if this is what's going to happen it's all the more impressive. German economy minister just said that German companies need to choose between the US and China. Germany's chemical industry leader BASF warned last year that the EU sanctions will decimate its industry[1](they recently announced a 76% profit loss[2]) and announced a cutback on its EU presence. TSMC founder Morris Chang said last year that Globalization is "almost dead" [3]
It doesn't matter if it's globalized or not, if China is locked out of it they will have to produce in on their own. They already have preferred access to the worlds biggest commodity and energy exporter now.
ASML by the way doesn't just create Lithography machines that people used. They are customized with the IP from Samsung for Samsung lithography machines, just like they're customized with TSMC IP for TSMC lithography machines.
> the trade spat between South Korea and Japan result in Samsung executes going to Japan
Can you elaborate on this? Would love to know more about that.
Kishida's government actually suggested even stricter restrictions on exports than the Biden administration did. The Chinese seem pissed and are actually starting to get litigative against Japanese companies. Let's not forget that on a company by company bases Huawei is actually the biggest single company 5G patent holder followed by Samsung, LG, Nokia and ZTE[1]. The same applies to RISC-V.
> Is this quantity over quality? i was educated in Taiwan from up to high school before i come to States. Let me tell you, back then the way they educated student is call "stuffing the duck". They basically spoon-feed you info until you're "full" (remember the answers). That kind of education does not produce Steve Jobs and the cram school is everywhere in Asia.
Maybe. But I was visiting the communication network society in Germany every year before Covid and the professors there certainly didn't think so.
But look at the board of directors of the biggest engineering schools in China. Look at board of NTHU in Hsinchu. I went to Semicon today and will go again tomorrow and I walk through those Halls with a big imposter syndrome, being thoroughly impressed by Taiwanese companies, but I still can't help but feel like Taiwan has lost its way. Too occupied with geopolitical games at a policy level. Look at the papers being churned out from the US top universities. 80% of the authors are of indian, chinese, possibly Taiwanese decent, 10% from other places and some from 3rd/4th gen US Americans.
With the documented surveillance of Chinese scientists many Chinese scientists have decided to return to their birthplace[]. Back in the day the founder of SMIC studied in Taiwan. Nowadays Taiwan leaks talent to China because they have had stagnating wages for 30 years.
By the way if you speak to people on the street you will see that many people have not forgotten what the USA did to Japans semiconductor industry in 1986.[6]
>Though it remains unclear exactly how far Tokyo could slow the export approval process or if it will shift towards a ban, South Korean chipmakers are worried the situation could develop into a full-blown crisis.
>“These materials, they are not something that we can find at another store and buy quickly,” said a source at one South Korean chipmaker, declining to be identified due to the sensitivity of the matter."
edit: Japan produce the best chemicals use for chip making. I believe no one country can do it all.
I see, one thing I noticed during Covid is that when China stopped exporting medical equipment due to the emergency ban, Japan immediately engaged companies like Iris-Ohyama to bring their production facilities back to Japan.
To my understanding Japan is the only country that said they want production back home, immediately worked on implementing their subsidy program and immediately executed on the production of facilities, such as the TSMC fab. The US talked about it and the EU talked about it,they came around for long subsidy planning and production planning. The EU as always being the slowest of the bunch.
>Actually Korean industry experts expect that SMIC will probably be able to create 3nm and 5nm without access to EUV[1] machines. This guy claims that 7nm is the last feasible node with DUV only[2], but what does China care about the infeasibility of the node?
Very well said actually, I didn't want to speculate further, but I certainly do think that it's probable SMIC will squeeze more than 7nm out of DUV if they can't acquire EUV, in that case it's no longer a matter of reasonable cost/yields.
>DUV is blocked through sanctions as well. Please don't make misleading claims.
Apologies for the poor wording, but this was neither supposed to be misleading nor is, to the best of my knowledge, misleading in the sense you imply. I am fairly certain I read about DUV deliveries sometime in 2020-2021 and according to this[1] the ban on DUV only effectively starts this year. My point was that that they weren't blocked from DUV, as in in the past, and that allowed SMIC and China in general a lot easier access to DUV. It's only with EUV that we could talk about actually developing a native capability alone.
>And yes TSMC was indeed producing 7nm using DUV before using EUV, but that doesn't mean that our wonderfully talented politicians and think tanks understand any of that given that just last year these geniuses were gloating that the export controls would throw back Chinas access to advanced chips decades.
Not the way I would phrase it, but I would be hard pressed to disagree.
> in that case it's no longer a matter of reasonable cost/yields.
This is an interesting point. Considering the extremely difficult engineering problem of the ASML machines, China might opt for an entirely different technology that, although more expensive and impractical for those with access to ASML EUV machines, could eventually allow them a better evolutionary path towards smaller geometries.
Blocking access to tech can have unintended effects.
SMIC can get a bit further with DUV, but they are certainly going to get eaten up with multi patterning. The additional cost and complexity from this will be significant.
My understanding is that Huawei was able to make a 7nm chip because they had ASML DUV machines.
From ASML's perspective they didn't have any choice. The US State Dept made the Dutch government do it. Rutte made a show of putting up a fight against the USA, but caved in about a week. Worst yet, the final deal was secret. We'll never know what deal was worked out for the Dutch politicians in the USA's pocket.
As a Dutch citizen it's scary to watch our government and the rest of the EU go against it's own best interests. We are vassals of the USA and it would appear our elites are completely captured.
As an American, I would love the EU to have superpower status as well, just so they would stop complaining about the US every time we fight authoritarian, undemocratic wannabe hegemons.
Please, enable Chinese military and technological superiority. The Chinese government is famously honorable and decent to western European political and social values.
To clarify, do you think it is in EU or Dutch best interests long term to have China on the leading edge of AI and computing?
> As a Dutch citizen it's scary to watch our government and the rest of the EU go against it's own best interests. We are vassals of the USA and it would appear our elites are completely captured.
I'm in the EU too. I don't like US hegemony either, but what's the alternative? We are not strong enough alone to avoid being under the influence of either the USA, Russia, or China. We could tell the US to withdraw from Europe with all their military bases, but I don't know if you would prefer Russians in Holland or their influence over the US. This is the untold price we pay for security. If we want to change that, we would need to divert a lot of money from social security nets in the EU and allocate it to the military. Then we could achieve independence, but I don't see the political will for that in any country besides Poland at the moment (which will have the strongest military in the EU in a few years but is still under heavy US influence).
I actually think that not too long ago the EU would have been strong enough to be a third/now fourth strong power, but decades of lacking a cohesive long term strategy have led to this point unfortunately.
Nope EUV is Cymer. TSMC, Apple and Huawei pooled talent and R&D on 7, 5, 3nm development so they could build their mobile chips. Probably because huawei had all that experience in 7 and 5nm they can guide SMIC in getting the yield up. Also dont forgot SMIC got a lot of top talent from TSMC.
Curiously enough the Dutch government (or what is left of it) extended the deadline for DUV equipment sold to China until year end. An apparent snub to the US.
ASML Says It Can Ship Restricted Chipmaking Gear to China Until Year End
(Bloomberg) — ASML Holding NV said it has licenses to ship restricted chipmaking machines to China until the end of the year, even as export curbs kick in from September.
I think the problem is that the Chinese alternatives to ASML DUV are close to being good enough. And a ban just would cause ASML to loose the market for good without actually hurting the Chinese industrial capacity.
Given that just a few months ago a lot of the HN crowd was frantically down voting anyone who was predicting such a thing and denouncing them as CCP shill, simply for being realistic I think it's a good thing that they are being humbled a bit.
Being humbled isn't a bad thing. I grew up in Germany, and until I moved to Asia I actually thought that most of everything that has a name that can remotely be pronounced in a German fashion was in fact German. I suspect that that is very much the case for large numbers of Americans and even more so in the European and American political class.
> Being humbled isn't a bad thing. I grew in Germany and until I moved to Asia, I actually thought that most of everything that has a name that can remotely be pronounced in a German fashion was in fact German.
I had a kind of funny example of that, a Chinese student named Weiman buying a cart full of Weiman brand cleaning products. Pronounced totally different.
> I grew in Germany and until I moved to Asia, I actually thought that most of everything that has a name that can remotely be pronounced in a German fashion was in fact German.
That requires to be pretty sheltered, culturally, though. I’m (also German) pretty much on the opposite end, and regularly surprised how many things are indeed German.
Well the benchmark result do show its close to Snapdragon gen 1.
So given those metrics it makes sense it has at least 6~7nm performance characteristic. But the SOC is not the special thing about the Mate 60 its the 5g and satellite communication module. China is now free to chart its own way again don't have to cooperate with US companies on 6g standards etc.
Given how diplomatically and economically the world is being split into a NATO and BRICS camps. So i wouldn't be surprised if the tech standards will also split into US, EU vs CN, RUS, IRAN sphere.
(Funny it has taken the market almost a full week to realise just how big news the Mate 60 is.)
Apple will take a hit on the iPhone 15 sales in China but it is not going to be banned. The banning chatter (which has been there before) is basically China saying hey US - now that you see we have alternatives maybe you wanna cool it on the tech war ... or else.
If you are taking about the new Huawei phone the theory is that it’s just repurposed Kirin 9000 chips hoarded before the sanctions and not a custom design. So basically a propagada performance. I guess someone will get hold of a phone and tear it down eventually.
TechInsights has already done a tear down and write up on what they found, they have electron microscope cross-sections for subscribers, and there are screenshots floating around on Twitter and the Semiwiki forums. It does seem real, but who knows the volume or cost at this point.
> the broader concern is that this is part of a government effort to push homegrown phones
I would say government officials carrying equipment they haven’t fully audited into sensitive meetings or using it to work on secret documents would be a HUGE no-no.
In the early 2000’s I was helping to get across the message it was a terrible idea for the Brazilian government (military included) to rely on (mostly) Microsoft products for office applications and that, instead, it’d be better to invest in local collaboration over open source software equivalents. For a lot of things, it stuck.
Apple stock is down, ostensibly on China’s iPhone curbs [1]. This appears to have carried through to TSMC stock, which might explain ASML’s movement. (Whatever replaces those unsold iPhones presumably won’t use chips built with ASML’s machines.)
Not the GP, but I guess that there isn't much of a predictive power of good or bad news that comes out of a company towards the movement of its stock price.
Obviously the company and its stock price are not "completely" separate, but what a company does in the short, medium, and long terms may not even account for half of the short, medium, and long-term changes in its stock price.
