 Welcome Ari. Oh, actually one thing, I sort of don't have my notes here, they are in my backpack. Thank you very much. Great, and now I hope we are actually ready. So, this talk is about the making of a chip, and I sort of don't see the slides here. Oh, there we go, finally. Great to see you all. So, we are talking about fabrication of a microchip, an integrated circuit, not exactly cutting edge technology, but something that could build this 80286. Now 34 years old, 35 years old, 8 years older than I am, but it has something like, oh, it has something like 135,000 transistors. There should be a die shot over here, and I hope the rest of my presentation is not the same, because on my laptop it worked. Anyway, so on this die we have, so there's a die in here, there are 135,000 transistors, and the minimum feature size, so the minimum size that you have there inside there is 1.5 micrometers. So, that's already something. So, just let me quickly give an overview of how field effect transistor works. So, down here you have the silicon substrate, and on itself silicon is not very conductive. But you have two ways to make it conductive, and one is to implant charges into the silicon, which creates positive, in the case of a substrate, or negative charges, and they can move around and transport current in that way. And so you see here that there's the source region and the drain region, and they are conductive by themselves, but they are separated from the substrate P region by the depletion region where there are no free charges, and so there is no connection anywhere. But if you now turn on the gate, apply a voltage, this voltage on the gate here, it attracts charges down here in the substrate, and these charges are the same or similar ones than in the surrounding source and drain areas, and so there's a channel that can conduct current, and you have turned on the transistor. And so from this we see already the things that we need to put together to form transistors and in the end to form a tip. So we need to have a way to oxidize silicon, we need to place polysilicon, that's the material of which the gate is made of, we need to dope to implant other charges into the silicon substrate, we need to put insulators because later on we'll place connections that consist of metal and we don't want to short circuit everything, and lastly, and that's maybe half of it, we need a way to pattern, to form the materials that we created there, because the usual ways to pattern stuff that we know in the macroscopic world, things like printing or milling or molding, they don't scale down to these 130,000 features and not down to this one micrometer size, but we have a way to pattern light, we know that pretty well, and we also know materials which undergo chemical reactions and we'll use both of that to selectively take away or add layers of material. So enough with the theory, we'll start to go to the clean room, first of all we'll put on clean room garment and clean room shoes and also glove and this is in order not to contaminate the room, we have all these particles on the clothing that's dead skin cells and we don't want to fall them everywhere. We also take the silicon wafer over here and that's where we build our chip on. So first of all photolithography, that's sort of the answer to the question how we can produce a pattern on this film of silicon, which is here in yellow, silicon dioxide, sorry, and not into this silicon substrate, which is gray here. So first of all you coat the substrate with a layer of photoresist and then you put on a mask. This selectively blocks light in some areas and other where it has holes which light can go through. Then you shine light here through this mask and this gives you areas of photoresist which undergo some reaction which makes them soluble in some developer solution. Which is exactly what you do, you place it in developer solution which takes away developed, sorry, exposed resist and leaves the silicon dioxide in some areas bare and otherwise coated. And then you can at last make your actual step of taking away the silicon dioxide and afterwards dissolve away the resist and you have your hole here in your wafer, in the silicon dioxide layer in the wafer. Great, so let's look at the equipment and the steps in detail which make up photolithography. So first thing is you coat your wafer and that works by placing the wafer into a spin-coater, place liquid photoresist on top of the wafer and spin it quickly. What happens there is that the photoresist that sits just atop of the surface of the wafer, this will stick to the wafer but the excess wafer which sits away from the surface, this is accelerated and spun off the disc. You can see here nicely droplets that spin just off the disc and this leaves you with a very even layer of photoresist. Next thing you need is the mask and here you see an example mask which has the patterns that you want to transfer onto your device and this one has five times four devices on them and in each of them you can sort of see the geometry that you want to transfer. You take both the mask and the coated wafer to this mask liner and the important part is here the center here. So in this thing here you put the mask and this is the chuck where you put the wafer and here you see a microscope. So what you do here is you can move the chuck and you move it so that it's aligned with the mask on top especially if you already have patterns on your wafer you need to align them to the patterns on the mask. When you're fine with the alignment end you have to be really careful there because I mean you have details which are one micrometer so you need to have that order of accuracy in placing the mask to the wafer. When you're fine with that you switch the microscope to your light source and that's a light source that emits ultraviolet light. It's contained in this box here and it shines ultraviolet lights through this mirror here on your device partially blocked by the mask and so you end up with your resist exposed. To kick it up notch nowadays you don't use these mask liners anymore but usually stepper scanners. Boxes like these and you don't really see what's going on there. So what happens inside is that the mask is not placed on top of the wafer anymore but it is separated by a lens. And so you step by step expose all the chips that you have on the wafer and this gives you several upsides. You don't need several copies of your pattern on the mask and it also allows you to scale down the image here which sort of takes away some of the well necessity to have this really small structures well defined. But on the other side you need to now move the mask and also the wafer because you have this line of light and you need to do this with absolute accuracy. So I mean nowadays features are just 20 nanometers anymore and so you need to have the accuracy of these both movements here well just has to be perfectly aligned. So let's actually develop the resist and that's now easy. You just put it inside. Okay next thing that doesn't work sorry about that. So you just put it inside the developer solution and this will dissolve away the exposed parts of the resist and it might leave you with this thing here. So the darker areas here that's areas where you don't have resist anymore. The lighter areas are silicon covered with resist and you also see these crosses here they're used for alignment especially when you have details already on there you can use them to align the mask to your details. You also see here that there's quite a bit of contamination particle here something like 5 micrometers in size and if you imagine this would be here obstructing this connection here then the device would be unusable so that's why we need to have this level of cleanliness inside the clean room in order to have working devices. So all in all you have the resist taken away here by the development step. So let's go on to the actual material take away of material and you could use the wedge etching process that you probably know just some liquid acid and just put your wafer inside of this but rather we use this plus matching machine so you place your wafer inside of this and a fluorine inside of the chamber and then ignite a plasma. So the situation is now this you have fluorine ions here and they are accelerated to the wafer below and etch away the exposed oxide here so the oxide gets away where it's not protected by resist. If you had used wet etching so just placed your wafer inside liquid then you'd have the problem that liquid etching is isotropic and it would further go and etch the oxide below the resist and with this plasma etching you have the ions that just go down here and this is really fine motion and there the resist is capable of really well covering and protecting the oxide so this gives you in fact a better resolution and that's why you're using it. So need to dope our silicon now this happens inside this ion implanter and again you can't really see anything so we're turning to schematic so you start with this ion source here this emits some ions which are accelerated fly away into this magnetic field over here and this magnetic field sends them on a curved path and different types of ions they follow differently curved paths so you have this slit here which selects only the ions that you want so you apply ion to your substrate and in this case they create a negatively charged well inside your positively charged substrate you now continue to grow a thin oxide that separates your gates from the substrate and that's quite easy you just place them into a furnace just throw away first flow oxygen through it and then the silicon oxidizes and forms the gait oxide so you can now vary in the thickness by varying the time and oxygen pressure and also the temperature so next part of your gate, the actual gate material this is deposited in one of these chemical vapor deposition machines so you see here that it's another chamber you flow silicon containing gas through it you have the ability to heat the chamber with these heating elements and then whenever a molecule of this gas hits some surface it gets converted into atomic silicon and in case of the wafer you have now a covering of silicon everywhere on your wafer so you need now to form the gates just etch everything away you already know this plasma etching device so you take away the gait oxide and the gait polysilicon everywhere that you don't want it and for this you also have a photolithography step which I'm not going to show here next thing is you construct the source and the drain regions and for this you already seen the ion implantation machine and you form here positively charges inside the substrate you might wonder why we have deposited this well of negative charges first and then continue to positive charges and that's because actually there are two types of transistors inside the chips these are complementary so they use either negative charges or positive charges to conduct current that actually gives you the possibility to actually switch things otherwise you'd have just turned on and you need something to turn it off and that's the complementary transistor you also have here the situation that you don't need photolithography to define the regions which are doped but you just use the oxygen which is already there and also the gates so everywhere where you have a naked substrate showing you change the doping by implanting ions and that's called a self-aligned process so you don't need to align any photolithography step to your existing patterns so this gives you much higher accuracy because you don't have any alignment errors and well this basically allows you to scale down your devices which is obviously what you're going to do now we are nearing the end of the fabrication we have up until now we have working transistors but they are not connected really anywhere and well they are not really useful if they are not connected so we want to place metal but until now they are a conductive part exposed everywhere so first of all we are depositing some insulator that's again done in this chemical vapor deposition chamber but now not with silicon but with silicon dioxide forming gas and you see what happens here form a layer of insulator and then afterwards now you use a step of photolithography to edge holes inside the insulator to actually connect your gate sources and drains almost ready, you now deposit the actual connection layer and this is the last piece of equipment that I'm going to show so this is now again a vapor deposition device but this time it's a physical vapor deposition you load through this load log here your wafer, transfer it into the process chamber here and turn on the deposition it's a sputtering machine which means that you have argon ions which bombard some material source and you can see here that due to this ion bombardment there is a plasma building up and these ions they sputter out atoms into all directions and this creates a covering of metal on our wafer so we almost have a finished chip we have basically hopefully a working chip but it's still we have several copies sitting on our wafer so we need to dice the wafer that means cutting it into the single chips that we have and then we put it into some package connected to the outside world that's by placing these wires that you partly see here then close the package and well, then you have your chip just sort of have to test it whether it actually works but hey, you're finished so what would be different in actual fabs? so I mostly showed you research equipment which basically works the same way but in industrial fabs you basically have robots doing all the work because I mean it's repetitive work and humans in this process are, well, they contaminate the clean room so you take the humans out of the clean room and have robots just doing all the work and then you're better off so this sort of leads to that you don't see everything you just see some robot carrying your wafer into the next process chamber alright, and I am now already finished somehow this was much faster than in my testing so we can already have Q&A okay, you all know the Q&A game there's four microphones please queue up behind them and if you're on the stream our signal angel will relay your questions as our human interface device so let's start with mic 2 please thanks for the great talk my question is can you do multiple layers I thought there are chips with multiple layers and I wonder how this is done so there are sort of two answers to that so in newer chips than this one you have multiple metal layers and this is done by basically repeating these two steps over and over again so you build an insulator and then on top of that place the edge contacts and place the connections and nowadays you have on the order of 10 connection layers and I don't know if your question goes on to three-dimensional flash chips or things like that no, but it's only the connections in multiple layers it's only the connections in CPUs and stuff it's only the connections okay, thank you mic number one please yes, good evening what kind of techniques could you list that would enable a user to check the conformity of what was less printed on a dice with the original x-ray microscope electronic microscope what kind of technology do you know and my second question would it be possible according to you to invent or develop a very low-cost technology that would allow all of us to check the integrity of the integrated circuits we would have in our devices like we do a SHA-256 on a piece of code alright, so first part of the question is how can we analyze the functioning and this is really a really hard task to do so nowadays you're talking about billions of transistors and they don't exist on their own they only function through the connections in between and so you can check whether the fabrication produces the transistors that you want and how this works is you first dissolve the package and then take successive microphotographs or microscope images of the layers so you start with the topmost metal layer and then selectively edge away every layer and you basically can reconstruct the layers that your device is made of is it possible to ensure that some parts were correctly doped or not? Yeah, that's possible I mean the doping level you can actually see that the semiconductor just looks differently but I mean it's still possible to just change the functioning a bit and have completely different functioning of your device this is something that other people actually have demonstrated that you can sort of change a couple of transistors and that creates a factor in your CPU Doping attack Exactly My second question Do you think it would be possible to invent a low-cost technology to check? No, I don't think so. I think that's a lot of reach Okay, thank you Mic number four, please Ask one question, please Thanks for the task I'm just wondering myself how the photo mask is going to be printed how this is working So you also need to use some lithography but this time you don't use rays of light but rays of electrons and electrons you can very accurately shine somewhere You can make a ray of electrons go into some exact location and that actually works with much more precision than optical lithography You could somehow use this to create the CPU themselves but this doesn't really scale because you have just well, it basically is very similar to these old cathode ray tubes the big monitors that we have before a ray of electrons going someplace but it's just one very tiny dot of maybe 10 nanometers and you need to scan all of your surface with this electron beam and then you have resist which are sensitive to this electron beam change their chemistry there and you can dissolve it partially away Thank you very much Number three, please Hello, why wafers around when dice are square? That's because the wafers are cut from the cylinder of silicon and that's one crystal and it's just the most simple way of producing single crystals in this big size I hope I can open some... Is the square crystal complicated? I have not seen growing of a square crystal actually I don't know but I mean it's definitely more simple to have it in this round size Thanks Mic number two, please Sorry So you said the light source used for photo lithography is UV but in the modern CPU the size of the die is so small what light source is used there? It's actually still ultraviolet light and that's really a wonder how you can make patterns so small So they use 193 nanometers light to produce features the size of 20 nanometers and usually there's a sort of rule of thumb that says that you can only have features half of the wavelength so a little bit over 90 nanometers and you use some techniques which basically can overlap two beams sequentially the image of two beams and then well multiple time expose your resist and this if done correctly can create smaller features Thank you Mic number one, please So in the plasma etching how do they modulate how much material is actually removed and is the photoresist destroyed in the same step? So first part well you can control and measure the current of plasma let me check where the slide is this one here so basically you can I mean these are also charges right and you have plates here which have a potential difference and you can measure the current through this and this will be proportional to the amount of ions that act on your device What was the second part? Is the photoresist destroyed in the same step? No it's not necessarily destroyed so you have some attack on the photoresist but if you have the photoresist thick enough you don't care I mean this is a sacrificial layer it gets dissolved away afterwards anyway Okay, thank you Mic number three, please Thank you Hi, I would like to ask you about vertical stacking you just mentioned and the technology of through silicone via how do you make the holes in the chip? You know I'd say also with etching but actually I don't know yet, sorry Okay, excellent And back to one, please Yeah, regarding just before the small feature size what I've heard before is that you actually do also photolithography not in air but in other mediums Is this actually true or not? So that you do it like underwater or within a liquid where the wavelength changes Exactly, you can use water and this has a dispersion Well, it changes the wavelength of your ultraviolet light by a factor of 1.44 And so now you have 1.44 times smaller wavelength and you can image things that are 1.44 times smaller That's correct, yeah But it's not the whole answer to how can you produce patterns that are 20 nanometers with 193 nanometers light Thanks Number three, please I have a question regarding the bonding In a modern CPU you have up to a thousand contacts on a two-dimensional lattice Is the bonding done in a two-dimensional way as well or is it squeezed around the edges as well? No, it's done in two-dimensional way nowadays So basically it looks somehow like your ball grid array package But I mean the ball grid array package is also some kind of substrate And on the other side of this ball grid array you have a similar array And this time it's not, well it's similar to solder but it's actually bonding balls And you use, well you put the wafer upside down on this And bond this by actually a technique somehow like soldering You heat it up and then the balls will attach Thank you Do we have any more questions? No, then please thank our speaker We have a bit of time left, no? Yeah, I mean if you want to continue Actually I have a question myself And this question was is it somehow possible to recreate something like this process The sort of do-it-yourself way And I think it actually might be possible So I think, well I mean I'm a physicist So I think the most, well significant step to this photography And well turns out actually they sell devices that create some kind of pattern in this microscopic dimensions And that's called a Blu-ray burner So maybe we could use Blu-ray burner to create a pattern not on a Blu-ray disk But in, well, photoresist And that would be sort of the first step that we're necessary to create integrated circuits of our own Which are, you know, we can trust these things So because we have developed them all of the way If someone wants to comment, oh microphone one please So that raises the question, what is the size of a, I guess the maximum size of a usable IC If one were to try to create this at home How small would it have to be to actually be usable? Well I mean you saw this 8286 and this has patterns of the size of 1.5 microns And I mean Blu-rays have much smaller and the other 400 nanometers Smallest feature size So well it sort of points in the direction that you could create patterns this size And I mean this would create a functional if not very fast chip, no? He's nodding Mic number four please If you want to create a chip at your home brew equipment How will you do the doping? Because I think you need some kind of vacuum for it and all that stuff And it's not that easy to handle That's not the only part when you need a vacuum by the way But well you can just buy vacuum pumps for the implantation thing You actually, and that's how it was done in the old days You can just flow gas over the wafer And this will partially diffuse into the silicon under the right conditions I mean you have to heat up the wafer And then gas atoms will diffuse into the silicon So that's much easier than this ion implantation device But it's also not as accurate You can control the doping to that accuracy Do we have any comments from the internet by the way? I haven't seen the signal angel in ages Mic number two please Concerning the self-made stuff What are the tolerances for the thicknesses of the substrate Of the removing and so on Because if we get it within like 10% or so Then probably we can do it But I don't know if I can do it by hand with the accuracy needed Well actually that's what you need photolithography for You don't rely on the etching going something like 20 nanometers plus minus 0.1% But you rely on the next material being different And the etching stopping at the layer between the material change And so that would be actually easy I hope Thanks At mic number four are you queuing or? Okay cool go ahead So you're talking about making your own chips But I don't really think it would be very economical Or feasible in a reasonable way to make it at home There are a lot of fabrications out there We can just send your design and make it there Wouldn't it be more interesting to really create a platform to make that more affordable So we can just order our own chips made custom at the fab Actually there are people actually more companies that go into that direction You may have heard of one of this risk v processors out there I think they're called sci-fi or something like that And actually one of the things that they want to do is offer Well fabrication as sort of as a service I mean you can if you're a company and come up with your own design You can already do the manufacturing as a service That's what one of the well basically all of the big fabs offer But if you're just some people then it gets easier but it's still well expensive Can we get mic one please? Yes I also wanted to comment on making these chips at home And yes I agree it would probably be well at least very hard to do this Because you would need to miniaturize the equipment and stuff And it would need to be cheap And well if you could actually make chips that small at home You'd have a lot of problems I think with well as you said You need dust free environments and stuff And I don't know might be possible but I think it's probably extremely expensive Or it would take a really really long time to mature the process that much And I don't think it would have that many people in the world interested in making their own chips And if you don't have that kind of economic factor driving it I don't think it would ever work And regarding the other guy saying well you can just send your own plans to the fab lab Well if you're talking about modifying these chips Or preventing the modification of these chips and preventing someone from installing backdoors You can't really send it away and be sure it's not modified I mean you need some way to actually validate that your chip is not modified at all And as you said you only need to change a few transistors on a scale of well 20 nanometers And you have a backdoor and you wouldn't be able to prevent that if you just send your design away Yeah that's it I'm not really on the same side of you but well just leave it as a comment I guess I don't know Yeah well I don't know you can answer of course I mean just also as a comment I think I mean I For the fabrication I think for me as a person it would be Extremely rewarding to get to that point to have my own chip made It's sort of the final frontier of making you know Yes yes I don't doubt that I was just I don't know I don't say it's impossible and I don't say you shouldn't do it If you want to do it do it you can do anything But well there's a lot of really high obstacles to get there I think Definitely yes And concern the other thing well it's much simpler actually to edge a chip and then See at the micrographs and well see whether it sort of checks out So it's I think it's quite hard still to put a backdoor like that It's definitely much better you're in a much better position than when you just buy some Intel chip Oh yeah I mean yeah you would have to take the Intel chip apart of course I well I got the impression it was hard from what you said earlier Because I understood it that you needed to take every layer apart to actually verify it And not you can't just use a microphone a microscope but if well if you can then of course it's Well it's easier I mean it's sort of both is true It's just You're not in the best position by Matching up a chip and then sort of hoping that what you want what you're checking Resolves to well when your check turns out true that you're actually fine But it's also well the the other side is not really you know good power either Yeah I just got an idea you could maybe implement some kind of check in your own chip design Check if it's been tampered with you know some kind of check some What what the other guy who was a bit rude said earlier but well not like on the hardware level by Implementing some gates that return a different value once your design has been altered by a few transistors Yeah but you need to prove that this actually works out It's also very hard I think there might be people who are doing this but I don't know Whether this says actual you know positive results yet Yes it's it's pretty interesting I'm gonna go ahead and cut you off now sorry Mike number four please Hi um having at home is not gonna be possible because of the vibrations You have been in a lab where you manufactured them themselves and you remember it's probably a vibration Isolated or dampened basement and only your neighbors using the staircase is already vibrating Your equipment so much that at modern nanometer size it's just not gonna be feasible For nanometer size you are too yeah right but the the clean room I was at was actually not vibration free vibration dampened So I mean we colleagues of mine produced features of 10 nanometers or so and they were fine So this is probably not really repeatable but I mean I'm not really wanting to produce 20 nanometers But I would be fine with the micrometer size and I think at that resolution it's not that much of a problem Mike number two please I am I'm asking is it possible to buy used equipment so something which was used some time some years ago And some fabric and which is now thrown away because they have better stuff For research equipment that's definitely the case But I mean you're still looking I think at a couple of ten thousand bucks For I don't know maybe other universities which are not that well off Okay to again In case we could organize equipment like rent equipment would you be in into helping developing a process for making trips with this blue ray Let's talk about this afterwards So the thing that you have to know is that we know each other and while there is a previous discussion going on Do you have anything to add Apparently not Oh actually someone at mic number three Hi so silicon is not the only process you can use to make chips Can you speculate maybe if there's a possible future other than silicon Other than silicon So for As produced chips basically No for processors I think silicon will be there for a while You have processes which enable you to communicate with light to the outside world And that's sort of where today's processor are well reaching a boundary with electrical communication So once this is out there Silicon based devices will be fine for a while I mean sort of the next thing That drastically would change Once you go For well Not switching based on charges but on photons Well everything changes but I don't really see that happening That's where your question was going right I mean just as another semiconductor material So yeah there are other semiconductor materials but I don't think for processing For basically Well data stuff there is something else coming up I mean for power devices there's silicon carbide And obviously for light emitting diodes and stuff you use other materials Okay someone's walking to microphone 4 Nope Ari do you have anything to say? No I'm fine Okay then let's thank Ari Thank you for this great discussion we had Also thanks for the guy with the laptop