Last time, if you’ll recall, I discussed the basic idea of a chemical vapor deposition system, and described how you’d use it to deposit silicon onto your wafer. Today we’re following rather directly from that post, where we answer some important questions. Questions like “What if I don’t want to put down silicon? What other things can you offer me?” Well, for starters
SiCl4(g) + 2H2(g) + O2(g) —> SiO2(s) + 4HCl(g) ~900 C
Put in your oxygen, hydrogen, and trichlorosilane gas, get out gaseous hydrochloric acid. What could possibly go wrong? That’ll lay down those oxide layers from the example. Another one:
WF6(g) —> W(s) + 3F2(g) ~400 C, or
WF6(g) + 3H2(g) —> W(s) + 3HF(g) ~400 C
That’s Tungsten Hexafluoride. Now, you’re getting some really nasty reaction products out the other end (I love me a good hydrofluoric acid), but you’re also depositing tungsten onto your wafer. Why do you want tungsten? As it turns out it’s really good at making nice, uniform vias. The high aspect ratio (there’s that vocabulary word again) on those features doesn’t bother it at all.
Tungsten, as a side note, is used as the filament in incandescent light bulbs, because of it’s resistivity, and because of how well it holds up to high temperatures. Neither of those characteristics is of much interest to the semiconductor fabricator.
And the traces? Could we put those down with chemical vapor deposition? Yes, although I’ve had some trouble getting a good description of the chemistry. People start using terms like “triisobutylaluminum” and “trimethylvinylsilyhexaluforoacetylacetonate copper (I)” and for some reason they get leery about describing the reactions explicitly. I’ll try with that last one though; the chemical formula for it is C10H13CuF6O2Si. (I’d make y’all a meat-and-cheese diagram of that one if I could even find a picture of it.)
2Cu(hfac)TMVS(g) —> Cu(s) + Cu(hfac)2(g) + 2TMVS(g)
Where, if I understand this right, hfac abbreviates “hexafluoroacytlacetonate” and TMVS abbreviates trimethylvinylsily. So basically the reaction leaves one copper atom on the surface and splits the rest of that honking big molecule into two parts, which float off independently. That leaves copper on your surface. The other common option, by the way, is to electroplate copper onto your wafer. That comes with its own issues.
Good enough for today? Not yet! Before I leave this topic in the dust, I’ve got a couple of quick notes on variations on your standard CVD reactor.
The first thing is going back to gas flow. Your chemical reaction depends on the material moving to the wafer, and how the gas is flowing through. Means you get into a lot of fluid dynamics calculations, which I try to avoid. The upshot though is that you can change your reaction characteristics by adjusting the mean free path of a molecule in the gas. Write that one down; words like ‘mean free path’ have a way of showing up on the test.
The ‘mean free path’ is the distance a molecule is likely to go before it smacks into another one. Here, let’s have a diagram:
How far does Neo have to swing his fist before it’ll collide with an Agent Smith? When there’s only one Agent out there, sometimes Neo has to fly all the way across the room to hit him. When you’ve got masses and masses of bad guys you can’t swing your fist without knocking someone’s block off. The more agents dogpiling on in a fight the shorter the mean free path of Neo’s fist. (You’ll note the fight goes worse and worse for him as the mean free path shortens. I don’t know if that observation can be generalized.)
Back in the real world, this is related to the pressure of the gas, which after all has to do with how many atoms are hanging around smacking into one another. If you drop the pressure in your average CVD chamber to one part in a hundred, then the mean free path of the molecule tends to be about a millimeter long, where at atmospheric pressure it’s better measured in nanometers. That shoves the rate-limiting firmly over to the side of the chemical reaction, not on the mass transport side. Which means you get good uniformity without carefully positioning your wafers in the gas flow. Stack ’em in vertical and ramp up production, boys!
Wait, what? Hold the phone; you take mass away and now mass transport isn’t the limiting factor? Why does dropping the mass by a factor of a thousand mean you have fewer problems with getting enough material to the surface to react? I respectfully request, sir, that you pull my other leg.
Drat; I was hoping I could slip that one past y’all without providing an explanation. Okay, here goes. Remember that the CVD process requires that you get your reactants to the surface, and also that you get your products away. Picture a man running a food truck. Business is good; there’s a line waiting for greasy greasy gyros. The first customer gets his lamb and starts happily munching on it. The second customer does too. But they aren’t going over to sit at the picnic tables; they’re just standing there. Where other hungry customers are grumpily trying to push past towards the food truck. Business slows down, not because it’s taking longer to apply cucumber sauce, or because there aren’t hungry patrons, but because there are impolite doofuses getting in the way. It’s mass transport limited.
Now, the food truck guy is going to solve that problem by shouting. Reminding his customers to shuffle on. But you can’t exert that individual level of control on atoms. (Would be neat…). We need to get the HF molecules away from the surface so more WF6 molecules can bounce down there. And so we increase the mean free path. If a HF molecule will hit another molecule half a millimeter away from the surface that’s a heck of a lot better than if it’s only bouncing half a micron before getting deflected. Effectively you’re mixing your waste HF with your process gases much more quickly than it’d do on it’s own, which means you’re also mixing your process gases with your waste HF much more quickly. You still have plenty of mass for the reaction game, but now you’re able to get it to the surface easily.
The word for that is LPCVD, for Low Pressure Chemical Vapor Deposition. You can get VLP as well, which means you’re going even lower with the pressure. Haven’t heard why that one’s worth it, so I’ll leave it at a mention. Moving on with other modifications.
Next wrinkle, you’ll note those are some high temperatures I’ve been describing. Aluminum (commonly used, or at least was commonly used as a trace metal) melts at 660 C. Does that mean we have to stop using these things once you get some Aluminum on there? What about other problems a real high temperature might cause?
In that case, you can use something called “Plasma Enhanced Chemical Vapor Deposition” PECVD. You use a plasma to bruise your gas molecules while they’re still a gas. Makes ’em much more reactive, and consequently you can get away with temperatures in the 300 C range. What’s not to like? Well, there is one interesting side effect. If you’re depositing a molecule (say, Silicon Nitride, Si3N4) you actually get a large number of similar subspecies. Which means you no longer get to say that it’s strictly three parts Silicon to four parts nitrogen. You write it something like SiNx, where you figure you get about 80% as much nitrogen as silicon. Or some such.
One final variant. Instead of heat or plasma to get your chemical reaction going, you can use photons. High energy photons (say, those produced by an excimer laser) will also ‘bruise’ the gas molecules, starting your reactions for you, and consequently allowing it to happen at lower temperatures. This allows you to ‘write’ on a wafer by shooting it with lasers. If you’ll excuse me a moment I’m just going to reflect on how awesome it is that you can build a computer by shooting things with lasers. Oh, right, the name for that one, it’s a PHCVD (PHoton CVD) chamber.
Right. Had enough about this subject? Because there’s more; you grow carbon nanotubes in a CVD, and you can put graphene on your wafer too. But we’ll move along. We’ll be tackling epitaxy. That’s the process of growing a crystal on top of your crystal, keeping the lattice the same but maybe changing the doping. Join us fortnight next when we cover epitaxy in “You Get One Part I’ll Take Nineteen” or “Same to You, Pal.”
This is part 37 of my ongoing series on building a computer, the Ricardo Montalban way. You may find previous parts under the tag How to Build a Computer. This week’s post has been brought to you by Khan. KHAAAAAN!