Coincidentally, i have been recently watching/reading a bunch of videos/articles on ASML and their technology.
Can some experts/knowledgeable folks here actually explain the technology in ELI5 (and above) terms? As i understand, a laser (what are its characteristics?) is fired at Tin droplets in a vaccuum chamber causing it to emit light in "Extreme UV" wavelength range which is then focused using a set of Zeiss mirrors to do the actual photolithography. Wikipedia (https://en.wikipedia.org/wiki/Extreme_ultraviolet_lithograph...) is well over my head. What i am unable to bridge is how this EUV wavelength maps to transistor sizes (in nanometers) via High-NA/Hyper-NA
technology.
A major limitation comes from the laws of optics. German physicist Ernst Abbe found that the resolution of a microscope d is (roughly) limited to the wavelength λ of the light used in illumination:
d = λ/(nsin(α)) ...(1)
where n is the refractive index of the medium between the lens and the object and α is the half-angle of the objective's cone of light. For lithography, substituting numerical aperture (NA) for n sin(α) and adding a factor k to the formula (because lithographic resolution can be strongly tweaked with illumination tricks), the minimum feasible structure, or critical dimension (CD), is:
CD = kλ/NA ...(2)
This formula, which governs all lithographic imaging processes, makes obvious why the wavelength is such a crucial parameter. As a result, engineers have been looking for light sources with ever-shorter wavelengths to produce ever-smaller features.
Diffraction sets a limit on how small the features can be patterned using photolith. When the wavelength is larger than the feature size, diffraction causes the light to spread out and blur the edges of what you're trying to pattern. The Rayleigh criterion shows how the ability to separate features depends on the numerical aperture of the system and the wavelength used. The explanation under 'Resolution in projection systems' on the wikipedia for photolithography is a better explanation than what is talked about under the EUV article. https://en.wikipedia.org/wiki/Photolithography
Going further and further into the UV makes the wavelength smaller and smaller and thus the feature size smaller and smaller. But making light that is controllable in a way for photolithography techniques to work that far into the UV is the difficult part.
Thank You! The section "Resolution in projection systems" in the above wikipedia link (https://en.wikipedia.org/wiki/Photolithography#Resolution_in...) contains the essential info. It lists the same exact equation which i have linked to in my comment and adds further details, to whit;
The minimum feature size that a projection system can print is given approximately by:
CD = k1 ⋅ λ / N A
where CD is the minimum feature size (also called the critical dimension, target design rule, or "half-pitch"), λ is the wavelength of light used, and NA is the numerical aperture of the lens as seen from the wafer.
k1 (commonly called k1 factor) is a coefficient that encapsulates process-related factors and typically equals 0.4 for production. (k1 is actually a function of process factors such as the angle of incident light on a reticle and the incident light intensity distribution. It is fixed per process.) The minimum feature size can be reduced by decreasing this coefficient through computational lithography.
According to this equation, minimum feature sizes can be decreased by decreasing the wavelength, and increasing the numerical aperture (to achieve a tighter focused beam and a smaller spot size).
Thus the NA being increased in Hyper-NA (0.75) from High-NA (0.55) results in a smaller "feature size" i.e. smaller nanometers.
I need to read some more but i think i now get the basic Physics concepts involved.
Not an expert, but I've always understood it as (LI5):
You may have heard the effect of light is like a wave, higher frequency wave-lengths have a shorter wavelength, which means it can 'reach' smaller features without impacting the rest of the surroundings. Photo-lithography is done through a mask, this effectively means you get a crisper image projected onto the wafer.
If you go too far into/past UV it becomes hard to deal with in terms of heat and optics from my understanding. Which is why we keep getting new $prefix-UV rather than something like X-rays (which are past UV).
Absolutely Brilliant! The animation of diffraction physics and mapping them to the terms in the given formula is just superlative. Scientific explanation through video at its best! Not too complicated and not too dumbed down but just the essence in a very understandable way. I now need to consult my Physics books for a Wave Optics brush up :-)
Thanks for pointing me to this. This truly deserves all the views/recognition from HN/larger Internet.
The waves emanating from from that point would be spherically symmetrical (think a 360deg "field of view"[0], whereas most lenses are <<90deg).
Now, since optical paths are two-way, this also implies that forming a perfect point image requires perfectlu spherically symmetric wavefronts[1] converging to that point, causing all the waves to perfectly cancel out each other everywhere except at the image point.
If you take away a slice of the wavefronts (i.e. block light with an aperture), the cancellations is no longer balanced, producing stray excitations at places that should be silent. (Think of it like squeezing a beer can with your hand causing it to spurt out of the sides)
The large the slice of wavefronts you are missing, the greater the imbalance. The resulting artifact are oscillations on the size order relative to the waves' frequency.
Basically, high NA means trying to capture as complete of the total wavefronts as possible to minimize the imbalance, and short wavelength means trying to keep the size of whatever artifact you do end up getting to be as small as possible.
[0] In air quotes because FOV != NA. The main distinction is that FOV refers to the span of principal directions (i.e. how many points can you see), whereas NA means, given a point object, how complete of its total wavefronts are you capturing it, (i.e. how bright is any given one point)
[1] Up to 2pi phase differential. If your signal is CW then multiples of 2pi is indistinguishable from being in phase, think Shannon Limit. This is why lenses work despite having path differential, because all that's important is that it's back in phase for the given wavelength even if shifted by multiple cycles.
> Basically, high NA means trying to capture as complete of the total wavefronts as possible to minimize the imbalance, and short wavelength means trying to keep the size of whatever artifact you do end up getting to be as small as possible.
What's nuts to me is the time scale on the bottom. That is not a lot of time for something so complex.
Then again, if they do an Intel 10nm+++++ and fail to maintain velocity even for a couple of years, their competition will close that gap very quickly.
At the top nodes and volumes, indeed there is no competition. They have a huge headstart handed to them basically by Japan screwing it up in the 90s and no one else being able to step (get it?) up quickly before the gap opened.
However, if they can't keep moving, competition that was trailing behind on larger nodes could get close enough that they become competitors for the high end. At first it would be slightly bigger nodes, but cheaper, which might be an acceptable tradeoff, as not everyone in the world is chasing performace over price. CPUs are so gruesomely overpowered these days that it may not really matter that much in many end applications.
Think AMD coming up behind Intel while Intel was thrashing around.
How long the AI hype continues may be important: if AI capabilities are needed in end-user equipment and that requires the real cutting edge processes, ASML/TSMC keep the advantage.
ASML is going to remain a monopoly and whatever Chinese alternative that there may be is going to stay in China for the foreseeable future.
However because the Chinese electronics industry is so big and they rely so much on imports, when they are building alternatives to the US-controlled semiconductor supply chain (ASML, Applied Materials, TSMC etc.), it will matter at one point for ASML even if the Chinese tools and final products are never exported.
Simple question, but the article doesn’t define this: what does “NA” mean here, and what do the does increasing NA from 0.33 to 0.55 mean? The gist I got from the article was “bigger number better” but I have no idea what these actually mean.
If I were a G7 leader, the state of global semiconductor production would make me very nervous.
ASML is a Dutch company and is essentially a monopoly now that they first commercialized EUV lithography. They can essentially decide what companies live and die.
TSMC is probably ASML's largest customer because it reportedly produces over half of the world's chips. It is a Taiwanese company (and Taiwan accounts for two-thirds of global chip production). There is Intel too but TSMC have been way more successful in commercial chip fabrication in recent years.
TSMC is of course in Taiwan, which is way more politically precarious than Western Europe. A major disruption to Taiwan's production could be absolutely devastating. This is probably why the US is pursuing domestic chip fabrication (eg in Ohio with the CHIPS Act).
But having two companies with this much potential market influence has to make a lot of people very nervous.
Both TSMC and ASML are just the tip of an iceberg. Each has thousands of supplier companies spread across many countries. Many of these supplier companies are peerless themselves in their respective areas.
It's a lot less concentrated on those two countries than it seems, but at the same time things are even more fragile than your post would imply.
The lithography/EUV optics comes from Zeiss. The laser for generating the plasma from Trumpf. Both key elements that ASML would not be able to build in-house or get from a different supplier.
I’m not sure Intel and Samsung are behind TSMC to the extent that switching to their CPUs would be, like, catastrophic. There’s a big advantage to using the newest node because people who buy iPhone want to be assured that their iPhones were made using the best, newest tech. But a 2 year old iPhone is not totally junk, it actually is still a much more powerful computer than almost anybody really needs.
TSMC has huge value, their R&D is hugely important. But if they vanished overnight, your G7 nation wouldn’t start to starve or anything.
Also, Samsung's fab is genuinely behind. Exynos and Samsung-built tensor chips are consistently hotter and more power-hungry than Snapdragons. It's pretty sad for us in Europe because we pay the same for our phones but get a worse product. We also miss out on features like 5G mmWave and MST (though I believe that's now gone from the US too)
Samsung is consistently behind in the sense it's difficult to compete for them in the high end mobile chips. But this difference is virtually meaningless in the strategic sense. Samsung can produce any capability TSMC can. The loss of TSMC would be strategically important only in the lost capacity.
Yeah I don't like big phones so it's more difficult for me. I have an S23 now which I'm really happy about. I just don't know what to do when it breaks because the S24 went back to exynos :(
Why’d you cut off the second half of my sentence? It contained the important caveat, really the main point of the thing—I was responding to the “if I were a G7 nation leader I’d be worried” sentiment.
Intel and Samsung are indisputably behind TSMC, but not in a way that would, like, risk social cohesion or something like that if we had to switch to them.
These observations while good are probably old news (about a decade old) by now.
The US has viewed tech as strategic and a realm of competition with China since at least the late Obama admin (so 2015 or so).
Semicon as a key strategic goal for China is regularly mentioned in their 5 year plans in an extremely public way.
The US has leaned heavily on ASML to go along with export restrictions against China, along with a wide array of restrictions for e.g. leading edge GPUs and even CPUs. These restrictions have ramped up recently but they’ve been in place for years now.
Not only does this situation make policymakers nervous but they’ve been taking action. It's funny that the public has finally started noticing.
You are making the situation sound much less precarious than it is. The reality is that the US has only really done something in the last few years, and those changes and attempts to hedge will take a very long time to materialize.
Odds are the US and Europe will never enjoy the comparative advantage Taiwan has in the semiconductor supply chain.
What does that mean? One of the west’s key strategic advantages is the ability to lead in frontier compute technologies. China, in addition to rapidly working its way up the value chain and nodes, can easily mitigate much of this advantage if it deems the calculus worthwhile. It need only disrupt Taiwan and basically all leading edge fab capacity is off the table. Outside of the work Intel is doing, the US does not have leading edge fabs.
We do not have the skilled workforce or supply chain to take advantage of leading edge fabs.
Why is this situation with Taiwan so difficult to unwind? In the West there is a fantasy that if war breaks out, we will just load all the Taiwanese onto a starship and bring them over to the US where they will happily resume their work in OUR fabs. But they don’t want that. They want their home and ideals to be defended, and they’re willing to do the work… from their soil.
The reality is the fabs are bargaining chip with the West, far better than any iron dome. The question becomes, how did we get in this position? And the answer is deliberately. The US saw putting semiconductor production in Taiwan and a way to reduce cost and challenge the Japanese, but also a way to imbed incentives into our foreign policy for protection of tenuous Taiwanese democracy and independence.
> The US saw putting semiconductor production in Taiwan and a way to reduce cost and challenge the Japanese, but also a way to imbed incentives into our foreign policy for protection of tenuous Taiwanese democracy and independence.
That's such a silly statement, and completely taking away the ingenuity of the Taiwanese. Taiwan(and Korea)'s semiconductor rise came specifically because the US hammered Japan's semiconductor industry in hopes of taking over their market share. It also happened in part, because the US massively attacked the RoC(Taiwan) nuclear industry, which left a lot of those engineers without a future, add on top of that a national strategy that caused two different strategies between UMC, TSMC.
This is a culmination of heavy engineering focus in Government, higher education and national support, along with a market gap created by clobbering the dominant player in the industry at the time.
What the US is doing right now does the opposite. It creates a market gap in the industry due to Taiwan being completely beholden to the US. The only reason for the rise in domestic semiconductor production and supply chain in China is specifically because the US tried to restrict China's access to these goods and in turn inadvertently created a similar situation for Taiwan that it did for Japan 1986. Leaders with cooler heads like the Mediatek CEO acknowledged as much and warned that this is what is going to happen.
> The US saw putting semiconductor production in Taiwan and a way to reduce cost and challenge the Japanese
Considering AMD, Nvidia, Supermicro and many others are all headed by Taiwanese CEOs etc I think this has more of an influence than otherwise. The community and ecosystem was built way before anyone saw it as a thing.
Japan had no chance with the conglomerates like Fujitsu still stuck in their ways.
> Why is this situation with Taiwan so difficult to unwind? In the West there is a fantasy that if war breaks out, we will just load all the Taiwanese onto a starship and bring them over to the US where they will happily resume their work in OUR fabs. But they don’t want that. They want their home and ideals to be defended, and they’re willing to do the work… from their soil.
Who is saying or even implying this? I've never heard this anywhere. Otherwise there wouldn't be all these folks gaming out war scenarios because the US would never need to get involved.
Oh I fully agree the situation is precarious. It's just fun seeing people independently start waking up to that fact.
I mean, you could probably see the latter being a cause of the former (if people generally realized the world was no longer that of the 1990s maybe they would on the margin encourage even more proactive/effective policy).
Could if been interesting if the article included price in some manner; as I understand it the total cost pr. transistor is still lower at 28 nm lithography and will be so in the foreseeable future.
Dichalcogenides in a 2D layer like WS2 or MoS2 for example. These can already be grown on silicon wafers but are hard to integrate with other materials.
Can some experts/knowledgeable folks here actually explain the technology in ELI5 (and above) terms? As i understand, a laser (what are its characteristics?) is fired at Tin droplets in a vaccuum chamber causing it to emit light in "Extreme UV" wavelength range which is then focused using a set of Zeiss mirrors to do the actual photolithography. Wikipedia (https://en.wikipedia.org/wiki/Extreme_ultraviolet_lithograph...) is well over my head. What i am unable to bridge is how this EUV wavelength maps to transistor sizes (in nanometers) via High-NA/Hyper-NA technology.
From https://www.laserfocusworld.com/blogs/article/14039015/how-d...
A major limitation comes from the laws of optics. German physicist Ernst Abbe found that the resolution of a microscope d is (roughly) limited to the wavelength λ of the light used in illumination:
d = λ/(nsin(α)) ...(1)
where n is the refractive index of the medium between the lens and the object and α is the half-angle of the objective's cone of light. For lithography, substituting numerical aperture (NA) for n sin(α) and adding a factor k to the formula (because lithographic resolution can be strongly tweaked with illumination tricks), the minimum feasible structure, or critical dimension (CD), is:
CD = kλ/NA ...(2)
This formula, which governs all lithographic imaging processes, makes obvious why the wavelength is such a crucial parameter. As a result, engineers have been looking for light sources with ever-shorter wavelengths to produce ever-smaller features.
Explain the above?