ok i dont usually subscribe, but since i watch most of your clips and you seem to go to great lengths to research your topics well, next time i cross your channel i will if it helps you
So is it meaning tsmc buy high-na EUV to replace current 5nm speed, also tsmc will use High-na EUV to multi patterning for 3nm(2x 5nm transistor density ) and 2nm(4x 5nm transistor density)?
I did my undergrad in physics and masters in quantum optics, and man, even then optics so often went over my head. I think you do a great job making such a complicated topic palatable.
@@sunriseadventures5459 I actually flunked algebra in highschool as well. I hated math back then, I was an art student lol. I didn't switch until after I went to uni
I really like the added humour you threw into the mix with the weird analogies and the more laid back tone throughout the video. Excellent content as always :)
I feel like you're the only channel I know that can get me really engaged in this kind of stuff without dumbing it down to the point where I'm hardly even learning anything. Really exciting tech to see. Thanks for the vid!
Oh one thing! UA-cam subtitles tend to overlap your funny little comments at the bottom of images. I don't want to say to throw off your visual style too much but if you could move that text slightly that'd be awesome :)
@@Asianometry why Brazil or any south American countries fail in chips manufacturing? Is a good idea to make a video about it!!! (I know it is bit out of Asia but it's like your videos about European and Americans chips manufacturing companies you did!!!!)
It's worth noting that ASML's wafers per hour are usually quoted at 20 mJ/cm2. But most processes uses 30-40+ mJ/cm2. Wph decreases linearly. ASML was going to quote numbers at 30 at one point,not sure if they made the jump yet.
They let me order one, maybe it's just your area. Are you in the Netherlands? I had my option of business or overnight air. Naturally a picked overnight air, but I need a new litho machine for my garage based semi-fab.
@@kaloryfer99999 f/1.8 refers to the aperture of a lens, the number 1.8 is the ratio of the focal length to the aperture. The Smaller the ratio the greater the numerical aperture f/0.8 > f/1.4.
@@kaloryfer99999All you had to do was a simple online search or even a UA-cam search. What you did is equivalent to calling a friend who might know the definition of a word when you had a dictionary within arm's reach the whole time. Willful ignorance isn't funny.
@@vigilant_1934 Calm down, he's just asking, if you can't contribute to the thread you don't need to reply. He's asking so someone can reply here and others can see, not hoping for someone like you to get butt hurt.
imperfect english, im thinking this guy is german af and legit lol. please explain how those fkn oval holes work. is it something to do with "lensing" due to aperture effect? Heisenberg uncertainty exclusion or something? wouldnt those holes mean you need a lot more light to start out with then? aint there some kind of issue with just cranking up the light? lol very curious
Boy, you dudes are doing pretty unbelievable stuff…. I can’t see us getting much smaller….. possibly famous last words, but what tricks and new light wavelengths are left in the toolbox.
@@peterwoolliams1283 X rays and Gamma rays are the only known electromagnetic waves with a smaller wavelength than UV. And their wavelengths are a couple of magnitudes (1/10) smaller than UV.
@@peterwoolliams1283 1) Embedded Cooling 2) Vertical Transistors to reduce electronic crosstalk and enhance transistor density 3) Significantly simplified CPU Instruction set with numerous dedicated accelerators for significant power reduction 4) 3D integrated power delivery via deep trench semiconductor capacitors and integrated GaN power delivery circuitry 5) 3D IP segmentation 6) AI optimized circuit/system design 7) SOC's with built in FPGA to speed software to hardware acceleration. 8) Nonvolatile UltraRam integrated L2 through L4 caches to further reduce power 9) Backside Power Delivery. I think these might be some of the new innovations that are either close to mass production or in mass production to hit the next 1000x in system performance.
Love your videos! You are the only channel on UA-cam that goes into the details of chip manufacturing and you also explain the information in an entertaining way. Keep up the good work.
@@Asianometry Yes, Asionometry, please make more of those machines. TSMC really needs more of them and it's taking sooo looooong. I'll watch you make them if that really helps.
About the 1 pm of aberrations of the mirrors: One atom is several 100pm in size. So the 1pm is not the maximum aberration as this is physically impossible but an average form error. Don't remember the specifics from my Zeiss days but they "average out" the random fluctuations that come from individual atoms missing or sticking out
It's kinda ironic that the most in-depth information on YT about this Dutch-German-US technology collaboration comes from a channel covering Asian economics 😉
I work at the nanometer scale, but that size-of-the-earth:human-hair comparison still blew my mind! With napkin math I get a few orders of magnitude below though..
So you’re telling me that to have a more relevant comparison to human scale, we would need to say something like “if you blew it up to the size of Jupiter…” (Which would not exactly be helpful either…) How the hell are we dumb monkeys making these sorts of things.
This is scary. I get very anxious watching your video about this new technology. Higher numerical apertures, larger mirrors and finer resolution. I have an ancient degree in physics and know a tiny bit about optics. Up until your recent videos I had no idea how, say, 7 nanometer features are built. Seemed impossible. Still does. Yet you're getting bigger fatter larger mirrors to increase the NA angle, among other things. Oh my God. I'm completely serious here. I'm 10 min into the video, paused, and my pulse is going up. Not a joke. I don't know if I'll finish watching it. This is an extraordinary, Herculean technical accomplishment.
Now imagine you are the product manager responsible for on-time delivery of that previously-impossible piece of hardware (and it has to work, no we will fix it with an over-the-air update).
Would love a vid maybe expanding a bit on the alternatives to EUV such as nano-imprint (pushed, as it is, by the Japanese) or advances in packaging/stacking (?) which I've read may be one good way to reduce the overemphasis on and/or need for ever-more expensive lithography equipment. Sorry if there is already a video on this
@@Asianometry Its more about the alternatives really. Industrial upgrading within the Semis space is slowing down and its very much because of the issues related to the development of 'super' machines like this which you outlined: Greater costs, longer timelines and fewer end-use customers/cases. How many of these machines will TSMC/Samsung/Intel even need? Will that in any way justify ASML's and other SME players' massive R&D costs? Who will fabs even need to make 1-2nm chips for? These are just some of the questions i've been pondering for a while. An alternative or a workaround that performs the same function and/or solves the same problem as an EUV system but costs far less, is certainly something the Chinese at least should be pouring tonnes of money/research into. Quite frankly, I'm very interested to see what comes out of it. I also don't think Nikon/Canon have given up on nano-imprint. Can it be made to work? And will it suffer from the same ridiculous cost curve as EUV?
The "obscuration" thing is works because it reduces the angle of incidence to be within the range that the multilayer mirrors have good reflectance, since the reflectance depends on the angle of incidence. A review paper that I am referring to speaks of "something something Zemike polynomials not orthogonal.... tatian polynomials needed ...". Thank you for these amazing videos on VLSI. Because of these, i've taken EUV lithography as the topic for one of my UG papers.
It's also sort-of required for any high NA - see 200-yr-old telescope designs. This mirror dispersion problem for off-angle incidence is also is in effect for all such Bragg mirrors at any wavelength, which, like cored mirrors, is also not new math.
The holes in the mirrors actually reduce transmission in and of itself. But it allows the incident rays to hit within the narrow allowed angles of the multi-layered mirrors, avoiding an even worse loss. Result is increased transmission compared to what a laterally-offset mirror arrangement (as previous machine had) would provide, given the increased NA. It may have a compactness benefit as well.
At the downside of much higher complexity and cost? I mean, you're literally drilling holes into a mirror which cannot have more than a picometer of deviation per meter
Having no experience with this topic, I don't know accurate but I imagined how reflecting telescope works to compensate that the receiver or viewer is located in the middle of the light entering the telescope.
Great video as usual! I’ve been wondering, how do you produce so many videos so quickly? I’m always surprised to see highly researched content put out so frequently, on a variety of topics no less. I hope you won’t get burned out!
@@timbo5663 A few month ago he actually announced he would take it a little slower and I was relieved, especially after he talked about his daily schedule working the research and editing around his full-time job.
Man this video is so good. I’m taking some graduate level cmos design courses right now but we seldom touch fabrication techniques. It gives me a whole new appreciation for what we are designing in class.
20 years ago when I was about to graduate from graphic arts, there was a big alarm on mechanical photography being replaced by scanners, lithography becoming part of the past and the whole concept of plate making. I never heard of EUV or ASML machines until now. This is fascinating.
Your work is outstanding. When a new video appears, I stop what ever am doing to watch it. I have lived in Thailand, Malaysia, Indonesia, Taiwan and Philippines and have learned so much from your videos that I did not know. One small comment, and since I am a somewhat new viewer, maybe I have missed it?? But I would be interested in seeing these same kind of videos on both Thailand and Philippines (which I may have missed???) if you get the time.
One can think of the holes as a similar to Cassegrain telescope or other folded reflective telescope flipped around. The hole is light lost (The effective obscuration) but it is a very small fraction of the total light. It doesn't affect the image quality. I'm surprised they were not always doing this.
They likely didn't got for it before they had for the same reason the anamorphic mirror is coming only now: it's extremely difficult to manufacture and qualify at this level of precision
Because in Cassegrain optics, or in a Schmidt-Cassegrain telescope, the final light path needs to pass through the centre of primary mirror with a curved secondary mirror. The improved image resolution over Newtonian mirror optics is a real improvement though. Unfortunately, unlike the Newtonian telescope, you probably don't have the finesse to grind the secondary mirror to the required grade.
11:46 The anamorphic lens' point is not to project on weird screen sizes but to fully utilize the 35mm film even if your film will end up being 1.85:1 or 2.39:1 ...you stretch your image on a 35mm so when they project it again they will get a picture that has more vertical resolution
I worked at lawrence berkeley lab as a mechanical technician and is is the first time I've heard and understood what the program in the next building over, CXRO was working on. Very very cool to have these loops closed.
Well, thanks for this explanation. I had to watch this video 3 times before I understood it at all. My goodness, what an industry, what a tremendous achievement by ASML engineers.
I was a printer and press mechanic for 30 plus years. We used to run the web through a long gas oven (heatset web) then around “chill” rollers to set the ink. The heatset ink is difficult to work with. Plus, switching to coldset involved a total wash. Technology changed and Heatset essentially became UV. The UV unit is only 4-5ft and width depends on press width. A huge advance…no more exhaust from an oven (horribly polluting). However, the ink is similar heat set. As thick as a custom colour (we can make any colour, yet the customer ordered ink) and it arrived late. Foreman made the call to leave it until morning. This was winter in ON, so the ink in the truck took 24 hours to thaw😂. I put the ink kits on top of the drive motors. Warmed them enough I could use a drill with a long t-shaped Allen key to finally get some viscosity.
I really enjoyed this video. Before I saw this video I had a grain sized understanding. NOW, I have about 5 grains ;-) Seriously. Really enjoyed it. the importance of this technology can not be over stated at this time for this time. Until we get involve in bio-neural networks this is the horse we are riding. This is not your first video that has impressed. I'm subscribed.
Anamorphic lenses: I think it's the other way around. Cinematographers use an anamorphic lens to squeeze a wide view onto standard film/sensor format. Then anamorphic projection lenses unsqueeze the squeezed images to a wide screen. If you project widescreen aspect ratio movies on to non-widescreen you get a squashed up image. (Sorry if this has already been covered; easier to repeat than scroll through all comments.)
I discovered your videos on semi conductors supply chain. I'm hooked. Very little science background but I just love learning new things. And you make it good
5:09 Can you post the Amazon link to the EUV machine in the description? I searched and couldn’t find it on Amazon. At least, not with two day shipping and a "buy again" monthly subscription option. Thanks!
Most of your videos had covered existing state of art technologies so is great to have one on up coming technologies like advanced EUV technology. It would be great to have a video comparing performance of various nodes and from different vendors. Given that fabs are obfuscating the definition of geometry size (ie Intel 10nm is equivalent to TSMC 7nm) it would be good to understand the true dimensional size and how each technology scale in size/power/performance especially into the 5/3/2 nm nomenclature.
node specs has been getting more and more complex as before we moved to finfets, the transistors were completely planar, so it was easier to define what your "resolution" was and thus your density, with finfets adding a vertical component things got more complicated, we are now adding cobalt interconnects, gate all around soon, imo density is the only metric that still holds up. Then there is the problem of libraries used, they usually advertise the numbers of the dense lower power libraries but cpus rarely use those (amd was able to nearly double the density of their cache in the same node just by changing libraries (talking about normal zen 3 vs zen3 3d's cache die)), To make matters worse at least from what i have seen intel has mostly stopped to share their more detailed node specs with the public, the latest i can find is for the first 10nm that was barely usable.
@@cj09beira Thanks for the info. I'm just wondering with claims of 3/2/1 nm process, what is the actual minimum feature size used verses achieved by 3D structures. Trying to understand the minimum feature size supported by DUV verses EUV and how much of the process scaling is due to 3D and other optimization.
Excellent and sophisticated presentation! NA is not free. The economics scales by non-linear functions that are pretty complicated, so the cost of optics increase between the square (for processed mirror area) and the cube (for material to support mechanical rigidity). You need to recover that cost by the speed up in throughput. It’s a difficult challenge.
As usual and typically of this channel, a quality and informative video peppered with nano noodles and nano-sized buddies! LoL and you can buy the $150m machine on Amazon. The only problem is that Stripe is unable to process your payment!
You deserve a visit around the ASML factories, I don't think that it was easy to gather this much information about the EXE machines! In any case, now we know that the new machines have a bigger dose of light and an amazing resolution. I'd love to know what are now ASML, Zeiss and all these companies cooking for the next iteration. Throughput will be soon close to the XT - NXT machines, and NA is just so expensive and hard to get.
That is a very silly comment. ASML doesnt allow people to enter for such reasons. Because those rooms are kept 1000% more sterile than clean air. Going around there means putting machines at risk for a sole reason of documentation which is secretive to begin with. The Netherlands develops a lot for the world, but we're not in a walk-in company. You can not easily get access to this. Reasons like this video are a no brainer for them to refuse. It can endanger information they want to keep secret and you risk damaging property you cant even afford to pay.
One aspect of the sound I liked about this video was that the levels were higher than usual, on par with the regular media I consume. For previous Asianometry videos I always have to turn up my mixer quite a bit for the duration of the video as 100% volume on UA-cam isn't sufficient.
You forget to state that Resolution in the formula is "better being smaller" as it is not resolution in amount of detail or something per area, but resolvable detail size or distance between resolved points. Very important for understanding the formula.
If you think about what has to be *around* the M5/M6 mirror system, the diagram makes more sense. M6 is focusing light onto the wafer, so the wafer must be down right below M5. Similarly, M5 reflects light from M4 and mask up on M6... so the mask and the rest of the projection system must be above M6. Mirrors with holes in them is a pretty standard trick in astronomy!
When I heard that "water tension destroys the line.... " I truly got a sense of the scale of these things. It blows my mind thinking about how far we've come as a species
Well done. Now, with the world in continuing short supply of IC manufacturing capability and bills being passed to fund more factories stateside, this video helps greatly in the understanding of how it's not just a simple matter of building a new factory and equipping it with 10-year-old IC tech. Also, why it's not a cheap thing to do, either.
One thing that offsets the lower throughput in wafers is the smaller die. So, as far as actual chip production it probably offsets. Going from TSMC N7 to N3 is about a 3X increase in transistor density. While that won't = 3X the number of die per wafer, it's probably something closer to 2.5X which is still excellent.
Outstanding analysis, a remaining curiosity, what is the total cost of ownership for the ASML EUV systems. 150MM or 300MM, this is just the start on a 4 or 6 year investment. Your video does raise a further question, past videos implied ASML is producing 50-60 of these complex machines/ year, with the additional level of complexity, can ASML still produce at that volume?
What you are describing is pretty much how I expect the end of Moore's law to play out. It doesn't end with a bang; there's no failure to make better processes; they are just more expensive to manufacture and the gains over last generation are less meaningful. Eventually they will overshoot, realize "N3" or whatever the node will be called was a step too far and just stay on the "N5" or whatever the process will be called, and try to make progress in some different dimension like chip stacking or through put or whatever and this will grind on while they find something to replace silicon CMOS (probably a couple of decades out since there's nothing great on the horizon right now).
There was a presentation from TSMC a few years ago about future predictions and they pretty much admitted they plan to innovate in different ways, like stacking, 3D chips and then changing the architecture of memory-compute model to make it basically one thing.
In purely optical terms, it seems Zeiss is using a shorter focal length mirror (wider light cone) and somehow controlling optical aberrations (not easy). An example of overcoming some of the optical limitations of mirror designs (generally occurring due to obstructed secondary/tertiary...mirrors) is a Schiefspiegler optical system. The more general case of tilt and decenter techniques combined with using only a part of a full mirror for optical systems seems to be what Zeiss is doing here. As with the Schiefspiegler design, using a portion of a mirror can give more degrees of freedom in an optical design at the cost of light capture. Clearly some fancy engineering going on.
Great content. I was looking for a video which parallels the complete supply chain of chip parts with main countries/companies involved along with how the components fit into the overall architecture of a computer. If you can make one that would prepare your audience actually understand what you explained in this video and videos to come. Please make one explaining the birds eye view of a chip architecture, how they fit into a computer overall and which companies are involved in the components to get together to work in a PC or any device. Thanks.
@@Gameboygenius Team Blue has gotten themselves a CEO who's a silicon designer and not a bean counter. Patrick Gelsinger knows that Intel lives and dies with their ability to fabricate. Personally I think they should've stuck with Bob Swan for another two-three years to make sure they *really* got down on one knee before they get up again... but alas, it is what it is, and Intel is looking to monopolize the market again... That they'll have the 5200 before TSMC is actually kind of worrisome...
@@andersjjensen Prices for enterprise X86 chips will for sure rise again as the market won't be able to catch up with demand any time soon. ASML's limited output and monopoly position is a big factor in the shortages of high end chips, is there any competition for them on the horizon?
Works both ways, you use an ana lens to capture a compressed image onto the film and then another ana lens on the projector to decompress it onto the screen
I suspect Zeiss's "obscurations" are based on a principle similar to the transparency of some butterfly wings (significantly more transparent than glass). I think someone at the U of Rochester is working on applying that to solar panels.
I always love how its like a surprise that once they are done they are already working on a new machine. From working in the sector i can guarantee you the list of "design flaws" gets pretty long, but addressing all of them would cost too much. Better finish up what you started and instantly start working on the revision addressing the list.
The bleeding edge of this tech is so wild. Just manufacturing any single component of one of these machines is an astonishing accomplishment. It's also kind of funny knowing how many of the chip made by these things will go into phones for people to make low quality junk for Tik-Tok. 😅 What a world we live in.
So... They wanted to make EUV "better and faster", so they increased the NA from 0.33 to 0.55, but that makes two light cones overlap by 1° which is not allowed (for some reason), so they doubled the magnification in one dimension which translates to half the wafer coverage. It seems like there *has to be* a better way of shaving off that 1° discrepancy than reducing the wafer coverage by half! Can't they target a slightly smaller NA or just place a constraint on the mask that says "this tiny sliver of the mask has to be black, so you lose x% of your wafer and mask, but the machine goes twice as fast". I don't know. This is crazy tech...
They are doing weird things to the light source to get the highest contrast possible and this puts a lot of constraints in the optics geometry. There is just no other way. The main constraint is the intensity of the source itself. Brighter source allows faster exposure. But the intensity is already fix because of how the EUV is generated so they have to resort to optics jujitsu.
@@kazedcat medieval alchemists would look at this light show and burn you at the stake. This is good progress even if it uses hacks and slows down wafers per hour. Going smaller undoubtedly wont be peaches but those praise the sun wizards at optics labs are chanting spells alright.
@@kazedcat If I'm not mistaken, the light source comes from Cymer in San Diego (acquired by ASML) - and so, presumably, one needs to look for information from them. I once had it described to me as high-energy lasers hitting a molten drop of tin - which produces 13.5 nm 'light.' Not light that any of us can see. Electromagnetic radiation. Wonder if Cymer has a relevant youtube video? ua-cam.com/video/5yTARacBxHI/v-deo.html
Well the "two light cones overlap by 1°" means the the next mirrors (before and after the mask) would have to physically overlap. That's not possible - you can't have two things in the same place. Jon's graphics here is not accurate, he should draw actually overlapping cones instead of keeping the same cone and just putting the word "overlap" there.
![[36] TSMC Technology Interview: A16 Node, System-on-Wafer, and High-NA](/img/n.gif)
Hope you enjoyed the video. Subscribe and check out the playlist: ua-cam.com/play/PLKtxx9TnH76RiptUQ22iDGxNewdxjI6Xh.html
Is the image from the reticle to the wafer rotated 90 degrees? Or is that an error in your image? 9:26
ok i dont usually subscribe, but since i watch most of your clips and you seem to go to great lengths to research your topics well, next time i cross your channel i will if it helps you
its worth for all the gaming performence. I take asml entire stock
wouldnt that be lanzhou beef noodle？ Ive never heard of taiwan beef noodle
So is it meaning tsmc buy high-na EUV to replace current 5nm speed, also tsmc will use High-na EUV to multi patterning for 3nm(2x 5nm transistor density ) and 2nm(4x 5nm transistor density)?
I did my undergrad in physics and masters in quantum optics, and man, even then optics so often went over my head. I think you do a great job making such a complicated topic palatable.
Feynman techniques are great to learn how to learn. Goodjob by the way.
optics was my worst university physics module. We didn't have a lecturer for much of it but I'm not sure it would have made a difference. I got 44%!
DAMMM Can't even imagine the homework!!! (As someone that FLUNKED Algebra!!) !!! LOL
@@sunriseadventures5459 I actually flunked algebra in highschool as well. I hated math back then, I was an art student lol. I didn't switch until after I went to uni
@@Brandon-oc8lr ROCK ON!! Did you get a job in the industry?
I really like the added humour you threw into the mix with the weird analogies and the more laid back tone throughout the video. Excellent content as always :)
5:01 Its amazing what you can buy on Amazon these days...
@@matthiasbecker5064 same day delivery is gonna be tough for this one
@@dienelt5661 Don't worry, those Prime drivers will find a way (unless Prime is not available in Taiwan).
I feel like you're the only channel I know that can get me really engaged in this kind of stuff without dumbing it down to the point where I'm hardly even learning anything. Really exciting tech to see. Thanks for the vid!
Oh one thing! UA-cam subtitles tend to overlap your funny little comments at the bottom of images. I don't want to say to throw off your visual style too much but if you could move that text slightly that'd be awesome :)
I'll keep that in mind. Thanks.
You, as a viewer can actually move the subtitles yourself. At least that's the case while watching on a PC
@@czechgop7631 I didn't know that.
Every day's a school day.
Cheers.
@@Asianometry why Brazil or any south American countries fail in chips manufacturing? Is a good idea to make a video about it!!! (I know it is bit out of Asia but it's like your videos about European and Americans chips manufacturing companies you did!!!!)
It's worth noting that ASML's wafers per hour are usually quoted at 20 mJ/cm2. But most processes uses 30-40+ mJ/cm2. Wph decreases linearly. ASML was going to quote numbers at 30 at one point,not sure if they made the jump yet.
you are my fav channel that does deep dive into semiconductors, even though i don't have any knowledge in the topic
No checkmark yet, huh?
pleasant surprise comment from Dr. Ian Cutress!!
Hello I love you :D
Yes they made the jump to 30
DARN IT, I checked Amazon and the Twinscan NXE 3600D is not available!!! LOVE YOUR WORK!!!
They let me order one, maybe it's just your area. Are you in the Netherlands? I had my option of business or overnight air. Naturally a picked overnight air, but I need a new litho machine for my garage based semi-fab.
@@jandrewmore In the US. LOL Can i get it to fit in my moms Toyota!!!! I WANT to see your garage!!
Buy me one.
@@suntzu1409 I'll need you to wire $150 million USD to my bank account.
@@jandrewmore I got mine from a Nigerian Prince....
For the photographically inclined, going from NA 0.33 to 0.55 is equivalent to ƒ/1.4 to ƒ/0.8. Very bright indeed.
And what does it mean for the rest of us who just speak English?
@@kaloryfer99999 more light
@@kaloryfer99999 f/1.8 refers to the aperture of a lens, the number 1.8 is the ratio of the focal length to the aperture.
The Smaller the ratio the greater the numerical aperture f/0.8 > f/1.4.
@@kaloryfer99999All you had to do was a simple online search or even a UA-cam search. What you did is equivalent to calling a friend who might know the definition of a word when you had a dictionary within arm's reach the whole time. Willful ignorance isn't funny.
@@vigilant_1934 Calm down, he's just asking, if you can't contribute to the thread you don't need to reply.
He's asking so someone can reply here and others can see, not hoping for someone like you to get butt hurt.
I can congratulate to this video. Well done and perfect explanations !! Keep on doing this
imperfect english, im thinking this guy is german af and legit lol.
please explain how those fkn oval holes work. is it something to do with "lensing" due to aperture effect? Heisenberg uncertainty exclusion or something? wouldnt those holes mean you need a lot more light to start out with then? aint there some kind of issue with just cranking up the light? lol very curious
Boy, you dudes are doing pretty unbelievable stuff…. I can’t see us getting much smaller….. possibly famous last words, but what tricks and new light wavelengths are left in the toolbox.
@@peterwoolliams1283 X rays and Gamma rays are the only known electromagnetic waves with a smaller wavelength than UV. And their wavelengths are a couple of magnitudes (1/10) smaller than UV.
@@himanshusingh5214 they won't work as they are ionizing rays.
@@peterwoolliams1283 1) Embedded Cooling 2) Vertical Transistors to reduce electronic crosstalk and enhance transistor density 3) Significantly simplified CPU Instruction set with numerous dedicated accelerators for significant power reduction 4) 3D integrated power delivery via deep trench semiconductor capacitors and integrated GaN power delivery circuitry 5) 3D IP segmentation 6) AI optimized circuit/system design 7) SOC's with built in FPGA to speed software to hardware acceleration. 8) Nonvolatile UltraRam integrated L2 through L4 caches to further reduce power 9) Backside Power Delivery. I think these might be some of the new innovations that are either close to mass production or in mass production to hit the next 1000x in system performance.
Love your videos!
You are the only channel on UA-cam that goes into the details of chip manufacturing and you also explain the information in an entertaining way. Keep up the good work.
We need more videos on machines like this. You do a great job of making them
If people watch then I'll keep making them!
@@Asianometry Yes, Asionometry, please make more of those machines. TSMC really needs more of them and it's taking sooo looooong. I'll watch you make them if that really helps.
About the 1 pm of aberrations of the mirrors: One atom is several 100pm in size. So the 1pm is not the maximum aberration as this is physically impossible but an average form error. Don't remember the specifics from my Zeiss days but they "average out" the random fluctuations that come from individual atoms missing or sticking out
It's kinda ironic that the most in-depth information on YT about this Dutch-German-US technology collaboration comes from a channel covering Asian economics 😉
I work at the nanometer scale, but that size-of-the-earth:human-hair comparison still blew my mind!
With napkin math I get a few orders of magnitude below though..
RIght? I couldn't help but do the calc too... I got 1-2 orders of magnitude lower! Insane.
So you’re telling me that to have a more relevant comparison to human scale, we would need to say something like “if you blew it up to the size of Jupiter…” (Which would not exactly be helpful either…)
How the hell are we dumb monkeys making these sorts of things.
Diam. Earth is 12,756 km = 12,756,000 m. Div that by 1 pm (1E-12 meter) = 1.276E-5m. Mult. by 1 million to get microns and you get
It's false. The distance between silicon atoms in a mirror is more than 200 picometers. There's no way to have 1 picometer accuracy.
@@jess_riedel I kinda wondered about that - people get pico and nano mixed up.
This channel is a national treasure
Me: "Mirror Mirror on the wall. Who is the fairest of them all?"
Zeiss Mirror: "I am."
Me: "That you are."
This is scary. I get very anxious watching your video about this new technology. Higher numerical apertures, larger mirrors and finer resolution.
I have an ancient degree in physics and know a tiny bit about optics. Up until your recent videos I had no idea how, say, 7 nanometer features are built. Seemed impossible. Still does.
Yet you're getting bigger fatter larger mirrors to increase the NA angle, among other things. Oh my God.
I'm completely serious here. I'm 10 min into the video, paused, and my pulse is going up. Not a joke. I don't know if I'll finish watching it.
This is an extraordinary, Herculean technical accomplishment.
Now imagine you are the product manager responsible for on-time delivery of that previously-impossible piece of hardware (and it has to work, no we will fix it with an over-the-air update).
Thanks for the industry update and the clear explanations of the technology. But oh boy, the humor was on fire in this episode as well.
I think I had a glass of wine first before starting on this one.
Would love a vid maybe expanding a bit on the alternatives to EUV such as nano-imprint (pushed, as it is, by the Japanese) or advances in packaging/stacking (?) which I've read may be one good way to reduce the overemphasis on and/or need for ever-more expensive lithography equipment.
Sorry if there is already a video on this
Nano imprint sounds like a good idea. I will think about it.
@@Asianometry How long do you think Moore's law can last and what are the alternatives after it ends ?
@Arkajit Beats me. I'm just a deer.
@@Asianometry Its more about the alternatives really. Industrial upgrading within the Semis space is slowing down and its very much because of the issues related to the development of 'super' machines like this which you outlined: Greater costs, longer timelines and fewer end-use customers/cases. How many of these machines will TSMC/Samsung/Intel even need? Will that in any way justify ASML's and other SME players' massive R&D costs? Who will fabs even need to make 1-2nm chips for? These are just some of the questions i've been pondering for a while.
An alternative or a workaround that performs the same function and/or solves the same problem as an EUV system but costs far less, is certainly something the Chinese at least should be pouring tonnes of money/research into. Quite frankly, I'm very interested to see what comes out of it. I also don't think Nikon/Canon have given up on nano-imprint. Can it be made to work? And will it suffer from the same ridiculous cost curve as EUV?
@@orkkojit Every year, the number of people predicting Moore's law is about to end doubles.
The "obscuration" thing is works because it reduces the angle of incidence to be within the range that the multilayer mirrors have good reflectance, since the reflectance depends on the angle of incidence. A review paper that I am referring to speaks of "something something Zemike polynomials not orthogonal.... tatian polynomials needed ...". Thank you for these amazing videos on VLSI. Because of these, i've taken EUV lithography as the topic for one of my UG papers.
It's also sort-of required for any high NA - see 200-yr-old telescope designs. This mirror dispersion problem for off-angle incidence is also is in effect for all such Bragg mirrors at any wavelength, which, like cored mirrors, is also not new math.
Humour so on point, it passes the Starlith 5000's aperture.
The holes in the mirrors actually reduce transmission in and of itself. But it allows the incident rays to hit within the narrow allowed angles of the multi-layered mirrors, avoiding an even worse loss. Result is increased transmission compared to what a laterally-offset mirror arrangement (as previous machine had) would provide, given the increased NA. It may have a compactness benefit as well.
At the downside of much higher complexity and cost? I mean, you're literally drilling holes into a mirror which cannot have more than a picometer of deviation per meter
Having no experience with this topic, I don't know accurate but I imagined how reflecting telescope works to compensate that the receiver or viewer is located in the middle of the light entering the telescope.
Great video as usual! I’ve been wondering, how do you produce so many videos so quickly? I’m always surprised to see highly researched content put out so frequently, on a variety of topics no less. I hope you won’t get burned out!
Tbh, yeah it is a little too much.
Don't encourage him to take a break ;) We need good tech content on youtube!
@@timbo5663 A few month ago he actually announced he would take it a little slower and I was relieved, especially after he talked about his daily schedule working the research and editing around his full-time job.
@@Asianometry - take a break - you deserve it.
Man this video is so good. I’m taking some graduate level cmos design courses right now but we seldom touch fabrication techniques. It gives me a whole new appreciation for what we are designing in class.
This totally went over my head, but I still enjoy trying to learn about it.
20 years ago when I was about to graduate from graphic arts, there was a big alarm on mechanical photography being replaced by scanners, lithography becoming part of the past and the whole concept of plate making.
I never heard of EUV or ASML machines until now. This is fascinating.
ASML is just the company making EUV lithography machines.
I am proud developing software for these machines.
Your work is outstanding. When a new video appears, I stop what ever am doing to watch it. I have lived in Thailand, Malaysia, Indonesia, Taiwan and Philippines and have learned so much from your videos that I did not know. One small comment, and since I am a somewhat new viewer, maybe I have missed it?? But I would be interested in seeing these same kind of videos on both Thailand and Philippines (which I may have missed???) if you get the time.
I did a video on Thailand's hard drive industry before and I have another one on their auto industry coming up soon. Keep an eye out
@@Asianometry Thank you. I will go search it out. Your channel, and the Cold Fusion channel, are THE BEST
@@Asianometry Please also make on Philippines IT outsourcing business
One can think of the holes as a similar to Cassegrain telescope or other folded reflective telescope flipped around. The hole is light lost (The effective obscuration) but it is a very small fraction of the total light. It doesn't affect the image quality. I'm surprised they were not always doing this.
Like a satellite antenna?
They likely didn't got for it before they had for the same reason the anamorphic mirror is coming only now: it's extremely difficult to manufacture and qualify at this level of precision
@@MrMischelito yes why would they take it to the next level useless it’s needed
Not sure either. Why don't you go and tell them Zeiss engineers. They could use an armchair optician opinion pronto.
Because in Cassegrain optics, or in a Schmidt-Cassegrain telescope, the final light path needs to pass through the centre of primary mirror with a curved secondary mirror. The improved image resolution over Newtonian mirror optics is a real improvement though. Unfortunately, unlike the Newtonian telescope, you probably don't have the finesse to grind the secondary mirror to the required grade.
At the FAB I work at production is at the limit with meteorology of anything with a target value below 10nm, even with +/-3nm tolerance.
11:46 The anamorphic lens' point is not to project on weird screen sizes but to fully utilize the 35mm film even if your film will end up being 1.85:1 or 2.39:1 ...you stretch your image on a 35mm so when they project it again they will get a picture that has more vertical resolution
Love the dig at Ethereum gas fees at the end. Nice video and very interesting topic.
7:07 “Wow it’s so thick”. Lol, love these subtle gems. Well played, Sir.
The people who figures these things out are seriously incredible.
Makes it seem like anything is possible given enough time.
This are work by hundreds of engineers from all over the world. Improving fabrication technology is a global effort.
I worked at lawrence berkeley lab as a mechanical technician and is is the first time I've heard and understood what the program in the next building over, CXRO was working on. Very very cool to have these loops closed.
very impressive you can release this information right when it hits the mainstream. you are ahead of the curve my friend!!
Well, thanks for this explanation. I had to watch this video 3 times before I understood it at all. My goodness, what an industry, what a tremendous achievement by ASML engineers.
I found this channel a year ago and I’m still blown the F away. Crushing it dude. Love the on point humour too.
This was the most incomprehensible video so far, keep up the great work
brilliant video, as usual, but i think at 9:25 the 26mm should swap positions with the 33mm. at 11:06 the ratio is normal again
I was a printer and press mechanic for 30 plus years. We used to run the web through a long gas oven (heatset web) then around “chill” rollers to set the ink. The heatset ink is difficult to work with. Plus, switching to coldset involved a total wash. Technology changed and Heatset essentially became UV.
The UV unit is only 4-5ft and width depends on press width. A huge advance…no more exhaust from an oven (horribly polluting).
However, the ink is similar heat set. As thick as a custom colour (we can make any colour, yet the customer ordered ink) and it arrived late. Foreman made the call to leave it until morning. This was winter in ON, so the ink in the truck took 24 hours to thaw😂. I put the ink kits on top of the drive motors. Warmed them enough I could use a drill with a long t-shaped Allen key to finally get some viscosity.
It's good to hear that the sound volume of your channel is now raised. I don't have to crank the volume up anymore.
I absolutely love the dry humor in this.
I really enjoyed this video. Before I saw this video I had a grain sized understanding. NOW, I have about 5 grains ;-) Seriously. Really enjoyed it. the importance of this technology can not be over stated at this time for this time. Until we get involve in bio-neural networks this is the horse we are riding.
This is not your first video that has impressed. I'm subscribed.
I like the explanations and overview. Very concise and understandable. I work in the reticle handling side of the EUV system.
Anamorphic lenses: I think it's the other way around. Cinematographers use an anamorphic lens to squeeze a wide view onto standard film/sensor format. Then anamorphic projection lenses unsqueeze the squeezed images to a wide screen. If you project widescreen aspect ratio movies on to non-widescreen you get a squashed up image. (Sorry if this has already been covered; easier to repeat than scroll through all comments.)
As someone who works for ASML. I enjoyed this video.
Love your sense of humor and keeping us up to date on this technology and keeping it simple (kinda).
I discovered your videos on semi conductors supply chain. I'm hooked. Very little science background but I just love learning new things. And you make it good
The techs, the explanations, the jokes, the sneer, and the snarks...
Man, I love this channel lol
As always, I'm amazed how concrete and light your vidoes are. Keep it up ;) !
5:09 Can you post the Amazon link to the EUV machine in the description? I searched and couldn’t find it on Amazon. At least, not with two day shipping and a "buy again" monthly subscription option. Thanks!
Most of your videos had covered existing state of art technologies so is great to have one on up coming technologies like advanced EUV technology. It would be great to have a video comparing performance of various nodes and from different vendors. Given that fabs are obfuscating the definition of geometry size (ie Intel 10nm is equivalent to TSMC 7nm) it would be good to understand the true dimensional size and how each technology scale in size/power/performance especially into the 5/3/2 nm nomenclature.
Yeah I’ve finished that video. It’s sitting in early access right now
node specs has been getting more and more complex as before we moved to finfets, the transistors were completely planar, so it was easier to define what your "resolution" was and thus your density, with finfets adding a vertical component things got more complicated, we are now adding cobalt interconnects, gate all around soon, imo density is the only metric that still holds up.
Then there is the problem of libraries used, they usually advertise the numbers of the dense lower power libraries but cpus rarely use those (amd was able to nearly double the density of their cache in the same node just by changing libraries (talking about normal zen 3 vs zen3 3d's cache die)),
To make matters worse at least from what i have seen intel has mostly stopped to share their more detailed node specs with the public, the latest i can find is for the first 10nm that was barely usable.
@@cj09beira Thanks for the info. I'm just wondering with claims of 3/2/1 nm process, what is the actual minimum feature size used verses achieved by 3D structures. Trying to understand the minimum feature size supported by DUV verses EUV and how much of the process scaling is due to 3D and other optimization.
Wow...this is a very impressive video with clear understanding of the physic behind the tool.
Why is this channel so good and so full of topics?
Amazing just amazing the science behind this . Some very clever people at ASML and others .
I don’t understand how these machines can be built in the first place to be so precise that is some truly amazing engineering
They are all standing on the shoulders of giants.
Thanks for the very understandable explanation of the functioning of such a complex machine!!!
Excellent and sophisticated presentation! NA is not free. The economics scales by non-linear functions that are pretty complicated, so the cost of optics increase between the square (for processed mirror area) and the cube (for material to support mechanical rigidity). You need to recover that cost by the speed up in throughput. It’s a difficult challenge.
As usual and typically of this channel, a quality and informative video peppered with nano noodles and nano-sized buddies! LoL
and you can buy the $150m machine on Amazon.
The only problem is that Stripe is unable to process your payment!
How do you make a mirror with 1 pm of aberration? You could do a video on that. Maybe one on mask manufacturing too.
Can't have enough of you technical video. Luv it.
As always a fascination and well-constructed description of highly advanced technology. The added humor fits perfectly.
You deserve a visit around the ASML factories, I don't think that it was easy to gather this much information about the EXE machines! In any case, now we know that the new machines have a bigger dose of light and an amazing resolution. I'd love to know what are now ASML, Zeiss and all these companies cooking for the next iteration. Throughput will be soon close to the XT - NXT machines, and NA is just so expensive and hard to get.
That is a very silly comment.
ASML doesnt allow people to enter for such reasons. Because those rooms are kept 1000% more sterile than clean air.
Going around there means putting machines at risk for a sole reason of documentation which is secretive to begin with.
The Netherlands develops a lot for the world, but we're not in a walk-in company. You can not easily get access to this. Reasons like this video are a no brainer for them to refuse.
It can endanger information they want to keep secret and you risk damaging property you cant even afford to pay.
The sound is a bit harsh in this episode... But the content is top dollar as usual. You're like the EUV version of tech journalism! :P
I did this video while traveling. My bad.
One aspect of the sound I liked about this video was that the levels were higher than usual, on par with the regular media I consume. For previous Asianometry videos I always have to turn up my mixer quite a bit for the duration of the video as 100% volume on UA-cam isn't sufficient.
You forget to state that Resolution in the formula is "better being smaller" as it is not resolution in amount of detail or something per area, but resolvable detail size or distance between resolved points. Very important for understanding the formula.
There's no other channel on UA-cam can produce this type of videos.
Nice info , thanks for the detailed explanation , now i can make my NA EUV machine by myself at my backyard
If you think about what has to be *around* the M5/M6 mirror system, the diagram makes more sense. M6 is focusing light onto the wafer, so the wafer must be down right below M5. Similarly, M5 reflects light from M4 and mask up on M6... so the mask and the rest of the projection system must be above M6.
Mirrors with holes in them is a pretty standard trick in astronomy!
When I heard that "water tension destroys the line.... " I truly got a sense of the scale of these things. It blows my mind thinking about how far we've come as a species
Great video! Perhaps many do not understand what ASML is capable of doing...
Great video, Jon! Many thanks for your hours of great educational entertainment ❤️
What a metallic voice! Let alone the frying.
Holy crap I’m so hyped to watch this 🤓
Thanks for this. Up until a month ago I knew nothing about micro chips. Although I didn't understand all of it, I certainly understand a lot more.
Well done. Now, with the world in continuing short supply of IC manufacturing capability and bills being passed to fund more factories stateside, this video helps greatly in the understanding of how it's not just a simple matter of building a new factory and equipping it with 10-year-old IC tech. Also, why it's not a cheap thing to do, either.
One thing that offsets the lower throughput in wafers is the smaller die. So, as far as actual chip production it probably offsets.
Going from TSMC N7 to N3 is about a 3X increase in transistor density. While that won't = 3X the number of die per wafer, it's probably something closer to 2.5X which is still excellent.
It's so incredible this technology actually exists
Outstanding analysis, a remaining curiosity, what is the total cost of ownership for the ASML EUV systems. 150MM or 300MM, this is just the start on a 4 or 6 year investment. Your video does raise a further question, past videos implied ASML is producing 50-60 of these complex machines/ year, with the additional level of complexity, can ASML still produce at that volume?
What you are describing is pretty much how I expect the end of Moore's law to play out. It doesn't end with a bang; there's no failure to make better processes; they are just more expensive to manufacture and the gains over last generation are less meaningful. Eventually they will overshoot, realize "N3" or whatever the node will be called was a step too far and just stay on the "N5" or whatever the process will be called, and try to make progress in some different dimension like chip stacking or through put or whatever and this will grind on while they find something to replace silicon CMOS (probably a couple of decades out since there's nothing great on the horizon right now).
There was a presentation from TSMC a few years ago about future predictions and they pretty much admitted they plan to innovate in different ways, like stacking, 3D chips and then changing the architecture of memory-compute model to make it basically one thing.
This is nothing short of pure black magic.
In purely optical terms, it seems Zeiss is using a shorter focal length mirror (wider light cone) and somehow controlling optical aberrations (not easy). An example of overcoming some of the optical limitations of mirror designs (generally occurring due to obstructed secondary/tertiary...mirrors) is a Schiefspiegler optical system. The more general case of tilt and decenter techniques combined with using only a part of a full mirror for optical systems seems to be what Zeiss is doing here. As with the Schiefspiegler design, using a portion of a mirror can give more degrees of freedom in an optical design at the cost of light capture. Clearly some fancy engineering going on.
Interesting topics you have about EUV. Could be also interesting to check out the history on ASML
Hi, great video. i work for ASML on the nxe exe
Great content. I was looking for a video which parallels the complete supply chain of chip parts with main countries/companies involved along with how the components fit into the overall architecture of a computer. If you can make one that would prepare your audience actually understand what you explained in this video and videos to come. Please make one explaining the birds eye view of a chip architecture, how they fit into a computer overall and which companies are involved in the components to get together to work in a PC or any device. Thanks.
Intel is the first buyer of the EXE:5200 per their announcement today.. The cost of EXE:5200 systems would be "significantly" above $340 million
Team Blue is finally playing catch-up, are they?
@@Gameboygenius Team Blue has gotten themselves a CEO who's a silicon designer and not a bean counter. Patrick Gelsinger knows that Intel lives and dies with their ability to fabricate. Personally I think they should've stuck with Bob Swan for another two-three years to make sure they *really* got down on one knee before they get up again... but alas, it is what it is, and Intel is looking to monopolize the market again... That they'll have the 5200 before TSMC is actually kind of worrisome...
@@andersjjensen Prices for enterprise X86 chips will for sure rise again as the market won't be able to catch up with demand any time soon. ASML's limited output and monopoly position is a big factor in the shortages of high end chips, is there any competition for them on the horizon?
Incredible presentation. I learned so much. Wow.
iirc anamorphic lenses compress/decompress to standard sized /film/ rather than screens. Apart from that, amazing stuff! Many thanks!
Works both ways, you use an ana lens to capture a compressed image onto the film and then another ana lens on the projector to decompress it onto the screen
Thanks for the deep dive.
I suspect Zeiss's "obscurations" are based on a principle similar to the transparency of some butterfly wings (significantly more transparent than glass). I think someone at the U of Rochester is working on applying that to solar panels.
He said 'obturation' wrong.
Blows my tiny mind🙏🙏🙏🙏
I always love how its like a surprise that once they are done they are already working on a new machine. From working in the sector i can guarantee you the list of "design flaws" gets pretty long, but addressing all of them would cost too much. Better finish up what you started and instantly start working on the revision addressing the list.
Excellent, I keep learning, while in awe!
NA 0.55 in vacuum is about 33° half-angle or 66° full angle of the light cone. (NA = n * sin alpha, where n=1 (vacuum) and alpha is the half-angle).
Really good video! Technical inaccuracies are minor, judged by someone from the industry.
Really great content. Please don't stop making these videos.
The bleeding edge of this tech is so wild. Just manufacturing any single component of one of these machines is an astonishing accomplishment. It's also kind of funny knowing how many of the chip made by these things will go into phones for people to make low quality junk for Tik-Tok. 😅 What a world we live in.
Dont feel bad they profit off those folks the same they profit off you, dont forget
Well it also goes into MacBooks for me to make videos about how it is made. Meta.
@@1Snouser Oh, I'm not saying it's a bad thing, really. Just kind of funny.
@@BRUXXUS tis tis, semiconductor industry is plain magic
I am afraid without that massive customer base this technology might be unfeasible on a economy level so impossible to make.
So... They wanted to make EUV "better and faster", so they increased the NA from 0.33 to 0.55, but that makes two light cones overlap by 1° which is not allowed (for some reason), so they doubled the magnification in one dimension which translates to half the wafer coverage. It seems like there *has to be* a better way of shaving off that 1° discrepancy than reducing the wafer coverage by half! Can't they target a slightly smaller NA or just place a constraint on the mask that says "this tiny sliver of the mask has to be black, so you lose x% of your wafer and mask, but the machine goes twice as fast". I don't know. This is crazy tech...
They are doing weird things to the light source to get the highest contrast possible and this puts a lot of constraints in the optics geometry. There is just no other way. The main constraint is the intensity of the source itself. Brighter source allows faster exposure. But the intensity is already fix because of how the EUV is generated so they have to resort to optics jujitsu.
@@kazedcat medieval alchemists would look at this light show and burn you at the stake. This is good progress even if it uses hacks and slows down wafers per hour. Going smaller undoubtedly wont be peaches but those praise the sun wizards at optics labs are chanting spells alright.
It's rather clumsy isn't it?
@@kazedcat If I'm not mistaken, the light source comes from Cymer in San Diego (acquired by ASML) - and so, presumably, one needs to look for information from them. I once had it described to me as high-energy lasers hitting a molten drop of tin - which produces 13.5 nm 'light.' Not light that any of us can see. Electromagnetic radiation.
Wonder if Cymer has a relevant youtube video?
ua-cam.com/video/5yTARacBxHI/v-deo.html
Well the "two light cones overlap by 1°" means the the next mirrors (before and after the mask) would have to physically overlap. That's not possible - you can't have two things in the same place. Jon's graphics here is not accurate, he should draw actually overlapping cones instead of keeping the same cone and just putting the word "overlap" there.
I love this channel...... Keep up the amazing videos
We need a collab between Asianometry and Huygens Optics.
Great channel, another great video