Japan on edge of EUV lithography chip-making revolution
asiatimes.com
asiatimes.com
[1] https://semiengineering.com/nanoimprint-finds-its-footing-in...
[2] https://www.electronicsweekly.com/news/business/china-to-bui...
Canon and Nikon don't need to be bleeding edge. Majority of the chip fabrication market is 20nm, 28nm, and 40nm along with analog chips (60nm+).
The sub-20nm chips will become more common as more and more fabs get turned on and EUV eventually gets commoditized, but the RoI isn't there yet for most electronics aside from high margins consumer electronics like GPUs, high end cellphones, etc.
> Chinese universities and companies are working on particle accelerator based EUV sources
Both the Chinese (SMIC) and the Japanese (OIST) attempts don't have great yield rates (the OIST project above is barely 50% of ASML's and SMIC's is reported to be similar).
Both will crack it eventually, but the order book for ASML's High NA EUV (3nm and below) is entirely Intel, Samsung, and the American fabs for TSMC for the next 3-4 years so by the time a competitor is ASML comparable, a significant subset of high end US and SK fabs will have transitioned to 3nm.
Surely the vast majority of fab income derives from these small nodes, just based on the sheer pricing disparity between complex semiconductors and everything else.
Few leading node machines = machine manufacturer/fab can charge high margins
Multiple trailing node machines/suppliers = machine manufacturers/fabs begin to compete on price, driving down margins
A good example of this principle is the diffusion of analog chip and memory chip fabrication in the 1990s to early 2010s (US/TW/JP/SK in 1980s-90s to MY/SG/PH in the 90s to CN in the 2000s).
The ongoing diffusion of specialized DUVL processes like 20nm, 28nm, and 40nm is another great example (US in the 2000s to TW/SK in the 2010s to CN in the late 2010s/early 2020s).
Nowadays companies like PSMC are selling the entire end-to-end Fab IP for 28nm/40nm processes to companies in JP, IN, ID, US, etc, further diffusing the know how.
There was a startup in Fremont, Lyncean Technologies, that had a 10 watt EUV source based on a small synchrotron.[2][3] They supposedly got to 10 watts. But that was too weak for EUV semiconductor work. They struggled for years to find a market, and went bankrupt in 2022. Here's the video of the equipment auction.[4] If, with this new optical system, only 20 watts are needed, that technology is worth reviving.
Something YC might be interested in.
An accelerator source is much cleaner than the laser tin-zapping thing AMSL uses. It's amazing that works as a commercial product.
[1] https://euvlsymposium.lbl.gov/pdf/2014/39af88b9d637418d86614...
[2] https://web.archive.org/web/20220119174924/https://lynceante...
The Science/Engineering to make this work is nothing short of magic! There are actually two lasers; the first one to reshape the spherical tin droplet to concave shape and the second one to vaporize it to plasma generating EUV range light.
The Extreme Engineering of ASML’s EUV Light Source by Asianometry - https://www.youtube.com/watch?v=5Ge2RcvDlgw
if A has $500 and B has $500 (and each put money away at the end of the year), tariffs on A are meant to 'eliminate' the compounding interest of the market. so over long horizons B ends up ahead. A still makes progress (but substantially further behind from where it could have been).
my example is more true of sanctions than tariffs. i've noticed many misguided tariffs from countries with less noble aims recently.
the US chip sanctions are viewed happily by the entire industry - except China. This means other countries have a chance to further displace China in the sector. Not agreeing with a certain sanction is fine but my point stands: they are meant to hobble countries long term.
Taiwan doesn’t consider itself part of China.
So the US view is in line with the practical facts on the ground.
And China wouldn’t need to saber rattle if they had credible “alternate” facts or didn’t feel they were the ones “hobbled”.
You mean like this? https://en.wikipedia.org/wiki/CHIPS_and_Science_Act
In fairness, I don't know all the inner workings of what was already in place and how much more they had to do initially, it just feels peculiar that the denied interested parties(large Japanese companies) never achieved it.
[0] https://www.bbc.com/news/business-65022389
[1] https://asia.nikkei.com/Business/36Kr-KrASIA/Semiconductor-f...
[2] https://www.scmp.com/comment/opinion/article/3206470/greates...
[3] https://www.scmp.com/tech/tech-trends/article/3126124/chinas...
(A lot of these sources mention the same incidents, they are not all separate, just wanted to give more than one source)
On top of that - an underappreciated aspect of lithography is yield. It's likely China can already beat ASML in a lab setting with <5% yield rates, but that's not a competitive process. What makes TSMC so valuable is their relatively high yield on cutting-edge node technology, alongside their fairly large capacity to supply multiple businesses at once. China has to design a scalable EUV alternative, bring it up to acceptable/profitable yield and quality control, and then scale it to replace the demand for most of the domestic chip clients. It's a lot of work, and it would not surprise me if we saw a lot of China's chip efforts echo their confident-but-slowed progress in engine design.
I full well expect that China is interested in producing silicon at-cost for the defense industry. It remains to be seen if they even have competitive designs though. Their premier domestic ISA, LoongArch, is based on 40-year-old MIPS architecture design. Cut off from international designs, it's not really clear how strong China's engineering chops are.
CATL, DJI, Douyin & post-grad engineering programs all over the US suggest the answer is "very strong." Underestimating China is not good for anyone wishing to remain competitive
To my knowledge, China's most advanced domestic lithography is a dubious 7nm chip with unknown yields made entirely with DUV, not EUV. This is really quite poor and I invite you to share more updated sources if you think that I'm underestimating them here.
Since when have EVs required high-density silicon? China's EVs are impressive, but it's 100% more of an endorsement of their lithium supply chain than their silicon one.
> They're at the top of telecom tech as well
That's a funny rendering of events. Didn't Huawei get forcibly removed from the market after they refused to let governments audit their security? I don't think that's an industry-leading attitude compared to Cisco. The only thing Huawei can win on is low prices, which really isn't a priority of telecoms when you're trying to sell backdoored hardware.
Additionally - telecom tech doesn't require advanced silicon either. Most WiFi routers and Ethernet switches are made with the same silicon the US already manufactures in GlobalFoundries and Intel fabs. The fact that China has 24nm DUV is not at all impressive when half the world was already exporting similar density silicon for cars and routers.
So... yeah, I do need to look at better examples if I want to trust their word. From both the examples you've given me it seems like China is focusing on cheap low-density hardware that anyone can make, but China can make cheaper thanks to their political prisoners and labor camps. Not only is it unimpressive, it's not even a slight example of how China can put modern silicon nodes into mass-production.
You are describing 2010-era fab technology, when talking about EV and telecom chips. I just hope you know that's what you're bragging about here.
Maybe they do turn things around, in the long-run. But the US can invest just as hard, and they haven't burned their bridges with ASML/TSMC. On top of that, the US can license IP from companies like ARM that previously weren't even shared with China to avoid unlicensed clones. The US wants their IP and global industry to prop them up here, and I don't think it's crazy to suggest they've got the upper-hand right now.
Important to contextualize "a while" is really last 10 years - post Tianamen weapons sanctions catchup was still in era where PRC tertiary production was fraction of current talent production with massive brain drain. Turbojet catchup was contingent on 10 years of generational tertiary catchup, 10 years of integrating talent into workforce, which semi is benefitting from out of gate/not as cripplingly bottlenecked on. Hence catchup gap/rate likely lessened under current conditions where there's increasingly less talent shortage, with trend of outright talent advantage, i.e. PRC IC white paper from 2018 targetted pumping 30k IC talent per year, now likely around ~550k/700k they think needed for entire domestic IC supply chain (was at ~400k in 2018). Meanwhile everyone else is projecting talent shortfall, so the ultimate catchup is post 2030s when PRC has 700k+/700k semi with likely surplus while western semi try to maintain lead with talent shortage, all the while PRC indigenous semi will chip away at the 300B+ IC imports currently being paid to western semi to support their R&D towards maintaining lead.
> forced to import foreign technology
Question is how long / at which point will PRC be fine with with relying indigenous, rumored indigenous 12nm production line coming online soon. Industrial policy can spam that... then what are the incentives? I argue they would be MORE incentivized to disrupt/degrade TSMC responsible for ~90% of high end nodes even with after CHIPS that disproportionately support the west. The fastest way for PRC to close gap then is to deny west advanced nodes, already largely denied to it. Some PRC planner is crunching the math that if everyone forced to make do with mostly 12nm chips, then PRC is going to have advantage in cost/scale in production AND renewable energy for operation costs. The flip side of the competitive coin and TSMC delimma is to make adversaries less competitive via denial. US can maintain peacetime advantage with lawfare, PRC can erode that advantage with warfare.
> US can invest just as hard, and they haven't burned their bridges with ASML/TSMC
But will it be as effective? Developments around Intel not promising. Even with CHIPs there's no sign US is going to plug talent shortage gap. US can buy the machines, build the fabs, the current bottle neck is still US talent production with respect to semi. Are they going to figure out how to get US fab workers to work as hard as TWnese even with proper compensation?
Also PRC hasn't burnt bridges with ASML/TSMC, US export controls attempted to burn that bridge against their wishes. But current ASML net sales to PRC due to stockpiling ~50% of their global total. They'd be happy to sell kitchen sink to PRC absent export controls.
> they've got the upper-hand right now
I don't think anyone's disputing that. I'm pointing out the contraints to catch timeframe for PRC is likely smaller than vs aviation, and ability for western semi to keep gap less due to current projected talent shortfalls. And once PRC establishes comprehensive semi supply chain, one that does not require multinational coordination, it will have certain competitive advantages. Also the entire point is to pivot away from western tech stack to avoid the ip/licensing fees and chip away US dominance in computing in general. PRC getting their semi up = starting condition for PRC eventually competing against trillions of value of US SaaS. PRC not just unhappy US control semi hardware, it's also unhappy US controls global software/appstore ecosystems. But that IMO much harder and broader battle, but one that is coming.
1. The original paper in Japanese claim to use only 4 mirrors to achieve the same results. Although this story only mentions >the EUV equipment’s optical system is greatly simplified
1. Japan isn't new to lithography. Both Canon and Nikon are in the market as well and eventually gave on EUV research and focuses on DUV.
2. Power consumption is reduced by a factor of ten is extremely important as modern EUV, and future High-NA EUV are insanely power hungry, with 30%+ of the raw cost of chip manufacturing goes to energy usage alone. And this isn't just TSMC but also NAND and DRAM maker as well.
3. ASML has something similar in the work as well. Although not to any extreme as factor of 10. But then they also dont need these sort of claim to get funding as the one here does.
Is the marginal cost of production at all relevant compared to the upfront capital outlay?
To be fair, I did notice were busy colouring in money with different felt tip pens.
“This is ‘blue’ money and it can’t be mixed with the ‘red’ or ‘green’ money. That would be silly!”
Those imply very different sorts of business.
I scoured for some other info as well: A single chip would go through a photo-lithography tool about 80 times. A photo-lithography tool can work on approx. 125 wafers per hour (5 FOUPS). A wafer contains ~230 chips. Current EUV light source requires a 20kw carbon laser.
> To eliminate this problem, Professor Shintake aligned two axis-symmetric mirrors in a straight line and used a total of only four mirrors instead of ten.
Are there any links to this data? Curious to read more about it, thanks.
The older and cheaper models of the ASML EUV sources would be good enough for this system that needs a much smaller power for the EUV sources.
The European partners of ASML are the only companies that could manufacture at this time the optical system.
Nevertheless, it is likely that the patent owners would prefer to license it to Nikon, to create competition for ASML.
Nikon should be able to make the optical system, but this could need a few years of development, unless Nikon already has a secret internal research project for making EUV mirrors.
The most difficult problem remains the EUV source, even if a much simpler source than the current sources is good enough.
It seems very unlikely that ASML would be willing to sell EUV sources separately to Nikon, providing components for competing systems.
The only deal that seems possible would be if ASML accepted to sell some of their older models of EUV sources to Nikon, in exchange for receiving licenses for this new improved kind of EUV optical system, which will then be incorporated in all new EUV systems, regardless if they are made by ASML or Nikon.
This would reduce the dependence on ASML and its partners, but ASML would still completely control the market of EUV sources, even if not also the market of complete lithography systems, like it does now.
Nikon's out of the EUVL game and doesn't plan on returning. They're reiterated this in just about every quarterly prospectus for years now.
They're migrating to licensing whatever EUVL IP they had to other players.
They've pivoted heavily to FPD Lithography (LCD, OLED, and potentially PV fabrication) - better margins, less political headache.
"As it turns out, however, EUV monopolist ASML and its optics suppliers Zeiss have already considered similar setups. “Zeiss has several designs available with similar/better performance than the Shintake design,” an ASML spokesperson told Bits&Chips. So far, these haven’t generated chipmakers’ interest due to the reduced resolution and smaller field size of the design. Should this change in the future, “we will explore whether one of Zeiss’ systems makes sense.”"
https://bits-chips.nl/article/despite-superior-efficiency-ja...
Ugh. The process naming which has nothing to do with size anymore still confuses people.
The US has made it clear that it sees semiconductors as a vector to attack power centres in Asia. Today it is China. Democracies are notoriously fickle, tomorrow it could be India, Japan, Korea or anyone really. There is a strategic need here to break the remaining European monopolies on equipment supply. For China the need is urgent but for all the Asian powers the need is real.
Given the manufacturing powerhouse that the collective Asia represents, I expect they'll figure all these details out and become technologically independent at least on paper. They've had more than a century of experience in dealing with the west, the risks of the status quo are too obvious once the US is overtly exploiting it.
Sorry but this sounds more like propaganda. China also sees supply chains "as a vector to attack power centres in Asia", that's applicable to anyone and anywhere. I'm not sure why Asia is a singled out here, sounds like one to trying to construe a racist motive when there is none.
The aformentioned India, Japan & Korea aren't worried about USA, their primary concerns are China and their neighbours. China's explicit objective to control the entire supply chain places them at a direct conflict of interest with those countries. They've already tried sanctioning Korea and Japan over petty issues like Fukushima water waste. Trying to appeal to some vague "Asian" solidarity as opposed to the existing OECD and actual shared norms sounds more like an attempt to divide allies rather than actually caring about those countries' interests.
Well China wasn't the only one with concerns, but probably the only one with the balls to sanction: https://www.nytimes.com/2023/08/21/world/asia/korea-japan-fu... - and why shouldn't they?
The main thing that won't be filtered is tritium, which already exists in the ocean, and has an ~12 year half life.
People get irrational where radiation is concerned, though.
If I pee in the lake you're swimming in, that's kind of gross... but it ends up being an incomprehensibly tiny fraction of the contaminants and gross stuff you're swimming in.
How long has Asia been a race? The continent contains 60% of all humans.
are they really though? At least in democracies things require oversight, debates and people are being held responsible.
It's exactly authoritarian regimes that can turn on a dime, if the supreme leader feels like it.
I'm scratching my head at this one. Why do you think they are doing those things if not because they are constantly re-evaluating their direction? If you're just going to keep doing the same thing over and over again there isn't much need for debate. The oversight and accountability mechanisms are all-weather good ideas, but typically the US is using them to constantly adjust and change policy responses.
Compare that to somewhere more authoritarian where they can't change policies precisely because those 3 aspects are quite weak. They find their groove once and stick to it for decades, and anyone trying to shake things up gets mired in institutional inertia. Sure a supreme leader could force sudden change in theory, but why are they going to do that? He's got everything set up how he likes it and his view of the world fossilised maybe 10, 20 years ago.
Democracies are absolute monsters. The moment they detect that their interests have changed they'll vote out the old mob, bring in a new one and do things differently. The US went Obama to Trump in one election cycle, which was probably a shock for anyone trying to negotiate with them.
I have faith in the institutions of our democracy, no matter how much you what-aboutists try to erode it.
It hasn't been. The US was one of the key players who built China up to what it is today, for a significant chunk of the past century they were strategic partners against Russia.
And it is worth remembering that post WWII the US was busy marginalising the Europeans, making sure Germany was prosperous but didn't execute an independent military resurgence and helpfully euthanising the British empire. There were enemies other than Russia and China over that period. It just happens that the other European powers were broken to the point where their resistance was ineffectual and the process was mostly amicable - if they'd been more competent it'd have been a lot nastier.
Can someone explain how they carefully rethought the theory to reduce the number of mirrors from more than 6 to just 2 (or 4)?
The yield rate is unimpressive (100 wafers/hr) compared to existing ASML EUV lithography machines (170-200/hr) or even DUV lithography (300/hr).
It also matters which are the energy and materials consumption, the occupied floor area and the acquisition and maintenance prices associated with that yield.
When all those are small enough, it may be worthwhile to replace a big and expensive fast machine with many smaller and cheaper slow machines.
That said, 10% reduction in power is decent along with the reduction in mirrors, but it would still take some effort for an attempt like this to be cost competitive compared to licensing from ASML.
At such a power reduction, it should be possible to use a much smaller and cheaper EUV source, allowing the use of multiple lithography systems, or increase a lot the throughput achievable with the current EUV sources.
I hope that they have modeled correctly the losses in the mirrors. Here the incidence angle of the EUV light on the mirrors is much closer to a right angle than in the ASML system. It is likely that this increases the absorption in the mirror in comparison with a mirror on which the light falls at an angle far from perpendicular, as used by ASML.
EDIT:
Unfortunately this paper might be a hoax.
I have read the 2013 paper that they give as reference [2] and as the basis of the claim for the 10 times reduction in power losses.
There the losses in the mirror are given only for incidence angles no greater than 21 degrees, very far from a perpendicular incidence. Moreover the data imply that at higher incidence angles the losses increase a lot.
Therefore the system with 4 mirrors at high incidence angles will have much higher losses than a system with 10 mirrors at low incidence angles, unless better mirror materials were discovered in the future.
So ASML must have analyzed this simple system and dismissed it as impossible to make with the known materials.
Not necessarily. It might just have performed at this peak capacity in ideal lab conditions.
This would need to be reproduced, but like you pointed out ASML and Zeiss most likely tested this configuration out at some point in the last 25 years.
That said, the English is very off on this paper - I did a re-read and it looks like they're proposing that they were able to cut the power down to 20W from 200W
In this book it is said even if EUV mirrors have typically very poor reflectance at high incidence angles, it has become possible to make mirrors with a 60% to 70% reflectance even for high incidence angles (without specifying how high).
They have computed their power advantage using the 65% value which is a middle value between those from the book, so it is OK.
However they have used the same value for the ASML mirrors, which use low incidence angles, so they may have a higher reflectance.
So according to the book, their system should be feasible nowadays, but the actual improvement in light losses is unknown, because the losses in each mirror depend heavily on both the mirror construction and the light incidence angle, which are not taken into consideration in the paper.