>> One of the like most famous precise instruments now is the ASML stepper.
>> Oh yeah.
>> Right. which is now being sold, you know, for $300 million and so on. And it's kind of considered like that nobody can make one besides ASML and so on >> and and and the Chinese are desperate to make one and fail in.
>> Have you have you been tempted to make a stepper?
>> The thing is in order to make one you have to believe you have a better idea >> than others. and I don't have the knowledge to even know if if I will have a better idea. You know, you have to spend time and understand it because it's immensely complicated technology. I mentioned before that I worked on this nonlinear resist nonlinear thermal resist which I thought could be the basis of a higher resolution stepper and we did develop together with a university here in town a very good thermal resist and and we modeled it and tested it but
>> so it's a resist which you when you heat it with two
>> change solubility yes it's a photo resist but thermally activated which is totally nonlinear so that part is very good but with today's geometries you run into the basic optics problem.
So basically first there is two ideas one idea to approach it from making the ideal resist that has no cross stop no proximity effect and you can achieve that with special ideas and resist but you still don't get around the problem how do you bring the wavelengths down so much so that you can write nanometer lines and as you know in the ASML current machine a huge part of the money was to develop the UV source which is
>> and optics.
>> Well, the optics are mirrors but but the problem was to make them so perfect.
>> Exactly.
>> There is a very good video by Veritasio and that's which shows the difficulty gives you an idea of the difficulty. I visited ASML many years ago. But most people don't understand why did ASML succeed where everybody failed including China which put in billions into this effort and ASML succeeded where everybody failed because they went against common wisdom in manufacturing because common wisdom is always never rely on a single source because that single source lets you down you're stuck. So everybody was taught if you manufacture you have to have multiple sources for each part. But you know the usual story. ASML said no no this is not the way to succeed. The way to succeed is find the best supplier in the world for each component each critical component >> and make an exclusivity deal with them that they can only make it for you and you can never buy this part for somebody else. So ASML looked around and said who is the company which makes the best lenses and optics in the world and that was clearly Zeiss Carl in Germany the long tradition and they went to Zeiss and says you will make us the lenses and we'll never buy lenses for somebody else but you cannot sell the lenses to anybody and they signed a deal a multi-year deal and the lenses changed.
Original lenses were for 193 nanometer.
Now they are mirrors for EUV but they stayed with the same supplier. Then they thought who is the best methology place in the world and there was a group at Philillips in Einhovven and Philips is known for light bulbs and refrigerators and appliances. But Philillips had a phenomenally good metrology group and they went to Philillips and says we want to make a deal with your metrology group. We don't care what you do with the rest of your products but they will only work for us and we will not buy from anybody else and they did more than that. They said we will guarantee the profitability of each one of the suppliers.
>> It's like accumulative effect seems like. So for China it seems like in order to make it they need on every single step. So that's a that's a that's a barrier that ASML locked up the best vendor in the world for each component.
There was a company which made the eximer lasers. So ASML bought them.
So some companies they bought and some made this deal. So at the end they ended up that for every critical part of the machine they locked up the best vendor in the world who couldn't sell to anybody else. Okay? And that's why nobody can compete with ASML because to succeed you'll have to find a replacement for Zeiss, a replacement for Philips, a replacement for Sim, replacement for 10 other critical vendors.
>> It's like whole world efforts.
>> Exactly. And that's why the Chinese that's one of the few things they miserably failed and they gave up and now they started again. And you know that they had a disography project for 10 years which failed and now they started again to make an ASML like machine and it seems they got very little progress. They managed to make the source part and it's not as good as ASML after spending a huge amount.
>> Mhm.
>> Why it takes time to get to this level of precision? Can it be accelerated?
>> There is experience of people. There is trade secrets. It's not something you can look up in a book. Yeah, >> especially when you do you have to polish a mirror flat to one atom.
>> Mhm.
>> Okay. There's always some old guys there who know what to do and these are irreplaceable. All right. So that's an amazing story and what is so amazing about ASML is their philosophy that they have single source for everything, the best in the world and that's a barrier.
>> That's super interesting. I I don't think I've heard that analysis.
>> Oh yeah. Yeah. No, that's how they operate. Yeah. As I said, I visited them and I know them well.
>> The best bolts, the best knots, the best mirrors, you know, it's nuts.
>> No, bolts are not somebody else can also make a good enough nut. But if you look at some key elements like the mirrors, the the motions, the lasers, there's about 10 critical elements. If you lock those up be very very difficult especially Z with the lenses and the mirrors because Z you know it's a very long tradition of how to make these optical parts.
>> Yeah I feel like I can understand like well switching the fields like when you go to high precision >> it's like complete different universe like you need to take everything into account. It's true and there's a lot of materials knowledge >> that details you don't take usually >> and there is a lot of experience of older people who've been doing it for many years. That's why I mentioned to you before in this factory where they make the jig bers in that book everybody looks older than I am because these are people who have been doing it 50 years same job >> and and and you just don't understand how they can do it so fast because you know they scrape this thing to sub micron in minutes and you try it takes you days and it's still no good because you know if you've been doing the same thing 50 years you can do it with your eyes closed close to perfection.
>> So do you think that maybe now there's like a big bet on AI kind of pushing the limits on maybe high precision like something which people didn't realize but >> no but >> by trial and error see like >> I think people most people don't make the connection between methology the high precision and computing power because the limit to computing power is methrology. Sure.
>> Because if you you know if you can make denser chips you can compute a lot faster and the and the and the chips the chips take less power because the gates are smaller. You know if the gate is much smaller it takes less power per gate. So you can put more gates on the chip because they they dissipate less power. So, so what limits computing power is the methology of the stages and the methology of the optics and both optics and the positioning it's it's a mechanical limit. How accurate can you make a mechanical part in the case of the mirrors is how you have a shape parabolic shape. Yeah.
>> How accurate can you make this parabola?
It's a mechanical problem and the motion of the stages is you have to move these stages to nanometers. So basically if you had a 10 times better methology you can make computers 10 times more powerful >> but could it work in reverse way like let's say if it's let's say for optics we are not limited by just spherical surfaces. If we go to >> caspherics or more complex patterns which is kind of beyond the no you can have new ideas there but you still need the brute force accuracy.
>> Sure.
>> Because if you make an apheric lens it's even harder.
>> Yes.
>> Because a spherical shape is selfgenerating. Self-generating means if you rub two pieces of glass together there they'll become spherical.
>> Yes. But if you want it aspheric, it's much harder, but it's not self-generating. There is no averaging effect. So spherical is actually self-generating and it relies on averaging like lapping. Any lapping relies on averaging, okay? And selfcorrection. But aspheric is not self-generating. If you rub two piece of glass together, they'll never become a spheric. The shape you want. They'll only become spherical. Okay? So because of that the it doesn't help you can compute better you can model the lens much better with computers but you still have to make it to atomic tolerances >> well I mean like for the tools let's say you will have a control of how you >> uh so okay so this is a common fallacy adjustments >> okay this is a common fallacy for most people who didn't work in this field most people who didn't work in this field says why do we need this raise the precision. Why don't we close the loop >> and measure the error and and do closed loop control and fix it? The >> idea, right?
>> Yeah. The way you take out distortion from an audio amplifier, you close the loop and take out the distortion. So, it turns out you can't in meology because it turns out that the same reasons which limit the accuracy in first place will prevent you from closing the loop. For example, you can calibrate the system to try to take out the errors with a lookup table, but the system will drift. The calibration will drift >> because it will wear if something is not perfect. It will wear at the high spots and it lose its accuracy. So it turns out that using this adaptive or software or feedback or even all kinds of clever feedback, not just normal feedback, but predictive feedback of high order.
>> You can improve the performance 5x 10x but you cannot take something totally bad and add infinite power and make it good.
>> 10x would be great. I mean that's like already >> 10x is is about the limit. So, so all these systems use corrective tables. So, so when I say Philips makes these perfect stages, they still correct them with lookup tables.
>> Okay. But the starting stage has to be astonishingly good that you can improve it a bit with some lookup tables at the end.
>> Okay. And yeah, it does use feedback.
All these ASML machines, they use a lot of sensors and feedback. But the basic system say if you have a parabola which is not perfect parabola there is no software correction for that because it's loss of information. The way to understand it best is think of a low pass filter. If you have an electronics lass filter and you send information through it you lose some of the information. There's nothing you can do after that low pass filter to restore what you lost. And that basically what Shannon came up with saying you have a bandwidth there's only so much information you can pass through this channel nothing will help you you can code and code but we reach that Shannon limit >> no more information can be trans the channel now the same is true for optics the information meaning the sharpness of the image okay if you have some bad optics and you get some bad image so there's some tricks you can recover some of the operations but basically information is lost and that information lost cannot be recovered by any any tricks.
>> Mhm.
>> So, so if the parabolic mirror in ASML machine is not the perfect parabola, there will be some blur and that's it.
Instead of a sharp edge, you get a soft edge to the line. There's nothing you can do which will help you. So, there is a fundamental limit. However, if the line is very sharp and accurate, software can position it if there is some small error and uh you have also a problem with aligning the different layers in a chip and this can be done by feedback. You sense the position and align it.
