ASML's High-NA EUV Lithography: A 2024 Update
Вставка
- Опубліковано 1 тра 2024
- I plan to be in Antwerp, Belgium for IMEC's ITF World 2024 in May 21st and 22nd. It's my first trip to Belgium. If you are in town, please let me know. Would love to speak to you. Email me.
Links:
- The Asianometry Newsletter: www.asianometry.com
- Patreon: / asianometry
- Threads: www.threads.net/@asianometry
- Twitter: / asianometry
I plan to be in Antwerp, Belgium for IMEC's ITF World 2024 in May 21st and 22nd. It's my first trip to Belgium. If you are in town, please let me know. Would love to speak to you. Shoot me an email!
I have a friend in belgium. Have not talked to him in years. It is nice to see technology progress but everyone is overlooking one thing. It takes just want worldwide cme or emp from WW3 to wipe out all chip making capacity. The question is what will you do if and when that happen. Just food for thought. Maybe you can do a video on this topic since ww3 is in the air.
I m close to town but i m quite ignorant and thus probably boring to talk to ^^ ASML is literally next door but i don’t work there for some reason x)
I live in Belgium. Tickets cost 550 euros however...
So you're not a deer AFTER ALL THIS TIME?!?!
It would be great if you could add a little visit to Veldhoven to your trip to Belgium. I, and so many of my colleagues, enjoy and regularly talk about your channel.
Imagine that your products is so good that the other competitor is... another product still made by you. One of the most important company in the world
There are several examples of this being the undoing of many companies in a book called the Innovators Dilemma.
Imagine if that company were to be destroyed in a war. Kind of scary .
It's no exaggeration to say that massive leaps in computation were a primary factor in the USA breaking out of stagflation (who benefitted is another matter) whereas the USSR who never figured out integrated circuits continued to stagnate into the 1980s ultimately leading to collaspe.
I find it hard to pinpoint any one player in the semiconductor industry as the most important as its all so interdependent, but it's definitely the single most important industry of humanity
@@googacctIsn’t that still the only way to survive as a company though? If you stop making products that compete with your other older products, some other company will take over your market share with a new product in a few years.
Well, what happens, like always, is that patents prevent other companies from being able to compete
For pronouncing Zeiss correctly
Of CAUSE we want to hear about DSA!
I second the motion. Video on DSA please. ❤❤❤❤
6:23 Obviously, that would be the TWIN²SCAN.
Nah, TWINSCAN²
Actually, it should be the 2TWINSCAN.
2 TWIN 2 SCAN
@@6.28I almost like this a lot but all these in the comment chain, aren’t options provided
TTS. TLA rules!
i love how your sense of humor is so deadpan that it is barely discernable, yet beneath the surface lies some of the most maniacal hatred i have ever heard
🙊
It's professional hating and I'm all here for it :3
All of this is a one big mascarade, they building them for living.
I'm always excited for an asianometry upload. You're so informative and funny af. I'll be a lifelong subscriber.
That because you are a geek nerd like him.
I'm honestly amazed that this much information is in the public domain. Another fantastic video.
Another thing Lithography is used for. is to create the microscopic UV mirrors in seren models of resin 3D printers.
I wonder how long until they are so precise we'll literally won't be able to tell the layer lines.
2x2 = Quadscan or Quac 🦆 for short
Quac - it does what twinscan't.
Ur mom for short
Woah, caught this one posted 15 seconds after upload.
John, watching this now, and four minutes in, great content, as always.
Excuse me for a few moments while I pick the bits of my brain off the floor, walls, ceiling, and furniture.
Getting these glimpses into the technology of chip making and other areas you cover fascinates me to no end.
At least with the SQL videos, I understand what you're talking about, since I have some knowledge of the subject.
Hardware is a whole nother level of abstraction.
60 kilodrops of Sn/second? That's so insane! Must be pretty small drops.....
Or just really fast
Yup, and they're shot with 2 pulses of 2 separate lasers. I worked on the laser power control at Trumpf. Fun stuff.
@@Vinzmannn
Man... these computers 💀💀💀
Great stuff as usual. Also cool to hear you as a guest on other podcasts. Have a safe and productive trip!
Wow. I caught one early. Have an absolute blast in Antwerp. Hope you meet interesting people to tell their stories or just you know have fun in Belgium.
So happy that this community has grown to almost 700,000 individuals. 1M soon! Keep up the fantastic content
The Zeiss optics video trauma still seems to run deep. ^^
Why is it a trauma and why is Zeiss watching?
What did zeiss do
He's not the only one being traumatized by optics
@@Termodramatisch I remember he was saying in at least one of the videos about Zeiss that the topic was mind bendingly difficult to grasp. And Zeiss watching, well I think that's just a joke because optics, you know.
@@Quast Thank you, this makes sense.
Re: Synchrotrons. Synchrotrons are Bremsstrahlung sources, they have very poor spectral brightness ("brilliance"). Total brightness in the rough spectral area doesn't really matter that much as the process relies heavily on having a near monochromatic beam.
In terms of well established light sources you wouldn't want a synchrotron or wiggler, but rather an undulator. I work in developing a new medical imaging technique that also has issues with spectral brightness in the x-ray regime, and so far undulators are the best we can run (but we also run higher energies, so can't use optics to focus as well). I suspect that undulators are a combination of too big, too expensive and too dim for ASML, you basically need a particle accelerator to drive them and only get intensity proportional to sqrt(N) where N is the electron number. For context: the beamline my group uses most often uses a 53keV undulator attached to PETRA III. It produces ~1W/cm2 in (very nice) xray photons.
There is an emerging technology that might work though - xray lasers. More specifically free electron lasers (FEL). European XFEL achieves 10**18 W/cm2 of even nicer photons. Buuuuut the beamsize is pretty small. In the FEL literature there are quite a few people talking about trying to make it usable for lithography though. So I'd keep an eye on that. Maybe when/if we move to 10keV and beyond and optics get more and more difficult it'll be back to lasers
Amazing info, thanks for sharing 👍
Great stuff👍 Also, he has mention FEL at least he talked about a particle accelerator in conduction to the EUV light source. Probably in the essay last year on the same subject I believe without checking.
This information is so impressive!
Always love these videos. Never ceases to amaze at how much semiconductor production is basically just damn near pure fucking magic lol.
PhD levels of understanding of a lot of subjects. That's why if semiconductor manufacturing was a movie the credits would go on for a long time. Lots of hands involved.
as always! love it
Love to see you in Belgium. Or at least in veldhoven. The story you were telling about throughput needs an update!!! I will see if I am allowed to give you the info.
You really enjoyed writing this one, it comes through and despite the difficulty of the subject matter your humor and enthusiasm for it makes it enjoyable to listen to. You explain how things work...and don't work very well. Honestly, I have no experience in the least with any of this tech, but I enjoy learning about things that I don't understand , which often can take me down a rabbit hole of looking something up to find the connection, only to realize I didn't understand the explanation of the thing that I needed an explanation for to begin with, but you explain things well enough that I can grasp the idea by context and that's where your enthusiasm and humor keep me interested. Even when I don't understand some of your references that I'm sure are quite clever, I'll just tell myself, "maybe I'll get the next one". Until I found your channel, I had only the smallest understanding of computer chips, basically that they came from wafers, I had no idea what it required to produce that wafer. I have really enjoyed learning how much I don't understand, thank you.
i also get those vibes, one of the greatest on my subscription list tbh
Well stated, as it's the same for myself.
Your guess about MOR is correct, Sn has a strong absorption for EUV because EUV is basically it is generated by Sn plasma.
Yup. Spectral emission lines are also absorption lines.
Amazing!!
Another great video about how we keep routinely doing the impossible.
Thanks for the video. Fun to learn about the bleeding edge of tech
You are the gold standard of tech. 3 mil subscriber deserved by now. hope you start a podcast as well.
Very good content as always. Thank you.
Yes! You pronounced Zeiss correctly (Ts-eiss). Thank you! Thank you so much! ❤❤❤❤❤❤❤❤❤❤❤❤
Not like Dr. Zeus?
I love geeking out on semi! ❤❤❤
TSMC doesn't need to buy the machines now. They were the first to get onto EUV and did so without too much pain. They can use EUV to get to N2 nodes and that's good for the next 5 years. They don't need to be on N2 next year but they do need to be on it for 2026. They can improve N3 to put out different processes using that base node. And it's OK because they're still working to improve defect rate for N3. So, Apple can move to an improved N3 for 2025. Nvidia will be down to a custom N3 next year and AMD will be able to get their next gen products out on N4 and N3, all of which will be out later this year and early next year.
So, once again TSMC can put out N2 in 2026 and I believe they've said they're on track for that. It's after N2 that TSMC will probably need High-NA EUV, for 2028 and that gives ASML time to work out any issues.
This may give Intel an advantage but it won't matter much other than TSMC maybe needing to bring costs down a little, and Intel won't have that advantage UNTIL they're producing on High-NA EUV. Intel is REALLY behind right now in spite of being at Intel 4 now. The issue is Intel 7 wasn't very power efficient and customers are tired of it, and getting onto Intel 4, their FIRST EUV product line has been challenging. So they are very behind. TSMC is going to be putting out improved N3 while Intel is only capable of laptop parts being on Intel 4.
I'm sure ASML will work with TSMC just like they are now with Intel to get them into production as soon as possible once TSMC gets the new machines.
Perhaps a Boeing episode? They've had some poor dealings in Asia.
Thanks for putting the video together. Some of this is plucked from the San Jose SPIE in March (which I attended). You mention in the video that a synchrotron isn't being taken seriously as a potential light source. At SPIE, xlight gave a well attended talk proposing its first cousin, the Free Electron Laser. Their proposal is a site-wide FEL to act as a central light facility for a fab. I found the proposal a little fantastical and almost crazily ambitious. However, given that someone is putting some money behind xlight, the proposal is not without backing.
Also - as someone who is interested in DSA at work, I believe this would be a good future topic.
Thanks!
Feels to me a big limiting factor these days is the laser source used and therefor the myriad steps that need to be integrated in the process to work around it's limitations.
If some guy managed to build a more powerful and/or more efficient laser source of the right kind of wavelength needed he would be rich af. They (tool-makers) could dispense with the whole shooting at a solid to create the right wavelength phase and a lot of the mirrors.
Fascinating stuff! Do you have a video that's a kind of "semiconductor lithography for dummies?" If not I think it would be greatly useful.
Beautiful as always. Please wanna hear more about intel's DSA. Thank you.
Sick, very cool
welcome to Belgium🎉
Come for the technology explanations, stay for the ultra dry wit.
100%
I would love to see a video covering an overview and history of direct self assembly from you!
Its in ASMLs best interest to make High NA EUV work and cost effective so they will make it happen.
Good to know we can still expect some upgrades to our hardware for a while yet.
Echoing another comment on this thread, "one of the most important companies in the world"...and watching Intel receive the new machine, saying it will take 3 years to bring on line, gives real perspective.
With the Nobel Prize for Physics in 2023, Attosecond Spectroscopy, the possibilities of graphene and other amazing tech becoming real, it really opens the mind to the possibilities, all while having that same mind blown away, by the people who create these machines.
Thank you for the awesome vid. 🙌
PLEASE do a vid on DSA !! That stuff is super cool and right up your alley
Would love an explanation of syncatron viability
I want to know more about DSA please.
I stopped and read all the lit you had in the video.....multiple reads spent to understand this topic. Fascinating technology.
I wonder how well the China projects are moving along?
Are we getting to the thin edge of the wedge now in improvements? We already have 1TB Micro SD cards. Most of the components we use today are not using cutting edge 3nm circuits for instance. Imagine when they do - and that is existing litho already in use on CPU's. Seems like getting CPU's using even smaller transistors and circuits is heading toward zero sum game. Am I luddite? Even after championing the race to where we are today?
We still haven't seen mainstream GPU's adopt PCIe 5.0 yet and right now PCIe 6.0 is being ratified. DDR5 is already seen at being near the limits of production due to interference from rest of PC motherboards RF interference - future gens going to be contending with more RF control driving up costs.
We never had it so good! Hehehe.
AI demands more advanced nodes and they have the money to force going forward.
This subject has been presented in an incredibly educational and engaging manner. The subject matter is so far over my head I can’t see it from here but love the video nonetheless.
An element will absorb the same spectral lines it creates. As they are using tin to create the EUV light it would also be the element to use to make the resist.
Wtf is this channel. This is amazing. You are so legitimate and intelligent.
Be sure to visit the diamond district when you are in Antwerp.
14:57 TSMC has declared higher than 100% utilization occasionally by purposefully deferring long term maintenance to maintain maximum output for a period of time.
I understand that N3 is important for energetic efficiency and miniaturization but reaching this point, the paths to more transistors goes to advance 3d packaging, 3D pose a challenge to thermal management . How 🥵 temperature affects the deterministic behavior of electrons in small nodes like N3 N2? This makes the Ansys acquisition lot more sense
The role of computational lithography in enhancing lithographic processes is crucial.
DSA sounds cool, more info please.
Babe Wake up a new Asianometry High-NA video dropped.
16:16 Now I'm curious if there are particular technical objections to using a synchrotron light source for this purpose, if it it's just that it makes no economic sense (for now) due to the cost and space requirements
That hah at 8:03 sounded like an AI voice artefact and considering you add subtitles to every video either means you are either a very hard working individual with rare "hahs" or you have been replaced by AI already :O seems I will have to track you down in Belgium to verify :D
60000 per second you say? Holy shit. The EUV machine was already the most complicated machine in the world, but they double (or well, sextuple) it down for the next evolution.
5:59 SCAN^2^2
उच्च संख्यात्मक एपर्चर आधारित तकनीक और प्रणाली का स्पष्ट अवलोकन, प्रज्ञावान विश्लेषण , सुन्दर छायाचित्र और बहुत सुंदर व्याख्या 👌👌👌👏👏👏।
🎵 SCAN-SCAN-SCAN-SCAN-SCAN-SCAAAAAN, WONDERFUL SCAAAAAN! 🎶
13:56 "lithography hotspot detection" by Jea Woo Park, *doctor of philosophy in electrical and computer engineering* ! I had no idea there was a PhD for philosophy in CE, but a quick search shows it's not limited to portland state university.
Maybe he can predict the semiconductor cycle? A valuable skill to have.
Madness.
I'm actually curious if these computerized mask design modifications can somehow be used for less high tech things, such as etching metal using a laser burned mask using electrolytic etching? I have tried to etch small details into metal this way but undercut is always a huge issue for me. I think it seems clever that you can get smaller details than possible by modifying the mask itself so that it ends up printing what you want.
❤❤
When you are in Belgium try the many Belgium's Beer! They have brewery olde then 1000 years.
They have heavy beers stronger than wine... 😀😁😏
Just a drink from heaven!
❤
can you explain why synchotrons aren't viable? the power issue makes synchotrons sound like a legitimate possibility.
6:00 Duel Twin Scan, or Quad Scan
Once these companies get down to nodes like 17A and 14A which is what High-NA should allow for, Moore's Law in terms of transistor density isn't going to matter.
I've said this before, I'll say it again. The transistor density should be around 500 MTr/mm2. PCs won't need anything much more, small devices like smart phones won't, most servers won't.
The bigger problem is going to be data throughput. When you can shove that many transistors in such a tiny space you get the bigger problem, and I mean a PHYSICAL problem of squeezing data from the outside world into such small spaces fast enough.
For instance I can make a CPU on 17A I'm sure very easily with 1000 cores, or a 1,000c/2,000t CPU. You still have the issue of getting the data into that CPU package to FEED that many threads.
For PC, you'll see the trend continue to MCM (multiple chip modules), but then as nodes get down to N2 or 20A, 17A and 14A you'll see that start to revert back to monolithic die, or monolithic for almost all compute functions and another die for moving data onto/off of the CPU. And it's because that chip with the cores for the CPU or graphics cores or ray tracing cores or AI cores will easily fit on a single die. That doesn't mean for very powerful graphics but for APU type products for laptops/game devices/lower power PCs. You will CERTAINLY see the number of cores for a CPU chiplet increase so that only Workstations would need a CPU with more than one core chiplet.
Interesting and valid point 👍
TwinTwinScanScan-TwinScan, or TTSS-TS to make it easy. Or TS², assuming someone lazier hasn't already taken it.
I bet this machine will be on the top of Intels shopping list this time!
How about ScanTwinScan, or Scanny McTwinface?
Twinny McScanface? Or just Scanny McScanface
I still like twinscanscan though.
A wild Dr Cutress at 10:26
I heard ASML next most powerful machine will only take an upgrade which they can just swap out of the machines already built(however many the build when they make their next upgrade)
ASML's High-NA EUV Lithography gets down to 1nm. Atoms are less than 1 atom. ASML says they can do yet another major upgrade. I can't believe they can get down to atomic precision assembly, but, i"m sure they can get down to some atomic resolution ability - atomic resolution 3d printing?
Process nanometer is marketing. 1nm process is actually 20nm pitch. We are nowhere near atomic resolution except for layer thickness.
@@kazedcat good to know; thanks for the reply!
Where do we vote for twintwinscan? I don’t see the vote.
twinscantwinscan +1
Wake up babe Asianometry just dropped a new video
Sure there is a... "babe"... .
Could as well be a time machine.
Quadscan is likely
I studied under Steve Campbell at UMN, and I remember him saying something like, "Y'know, it's never been straightforward or obvious, but we've always been able to 'clever' our way past the roadblocks." And he's right, it seems that if physics doesn't prevent it, and there's enough demand, we'll figure out a way to make any device.
the easiest voice to replicate with AI
🙏🙏🙏🙏🙏
When does HEUV officially finally switch to SXR? Cymer was doing everything they could to steer XSR back in the last 90's, with little or no luck.
it is a monopoly but you can't stop progress and you can't really rush progress either - everything in due time - these industries will keep chugging for a long while but the sands do shift and when changes happen they may get devoured and devoured quickly
0:56 ah! i thought it stands for "Highly Not-Available"
how does cerebras do their wafer size chip? is stitching required there?
Yes there is. They have some custom IP for stitching. More i don't know.
+1 for scan scan
It's about time....
And throughput...
I could never imagine ASML, or anyone else for that matter producing such a complex tool. I actually worked on their oil burner in 1986 thinking it was advanced. Die stitching 2 reticles with a twin twin scan sounds fantastic.
amazing comment.
TwinTwinScan gets my vote.
How many twin scans can a twinscanscan twintwinscan?
you are the only person. mainstream. that is doing this. you have a monopoly on it. soak it up.
TwinScan(squared)
Maybe make a video about IMEC, very under reported what they do.
This technology is insanely complex.
What does sodium have to do with chips?
Whoever is ahead end of December 2024 will hold that lead for a decade, I bet. Liquidity will dry up, R&D invrstment will slow to a stop and there will be a push to economical solutions, even possibly diversification of existing tech solutions. Due to the fragility in all global supply chains.