00:00We've heard so much about the GPU chips AI needs, which have catapulted NVIDIA to the top of the
00:06leaderboard. But what about quantum computing? How are they different? And who's making them?
00:12Jerry Chow of IBM gave us a tour at the Thomas J. Watson Center in Yorktown, New York.
00:19So this is our IBM Quantum System 2.
00:22Which is the most recent one.
00:23The most recent one, it is also an infrastructure that is built with modularity and scalability in mind, built for data centers.
00:33So the way to think about it is really this is a modular architecture.
00:37What's right behind us right here, this is actually the cryogenic part of it.
00:41So inside here, there's actually a cryogenic system which cools down.
00:45That chandelier that we saw in the other room.
00:46Chandelier, yeah.
00:47Cools down three, and at the moment there are three Heron processors, 156 qubits each.
00:53That are culled inside of this cryogenic infrastructure.
00:57And the idea is that over time, that kind of cryogenic infrastructure can be modularly scaled.
01:02The footprint can be increased.
01:04And then hanging off on the wings over here are actually classical control electronics.
01:10So the pieces that generate the signals to manipulate the qubits, read out the qubits, they're all in here.
01:18And they're actually located right next to it because we need to make sure that they have short time scales for the control and latency.
01:30But similarly, as you need to scale to a larger footprint, we can scale the amount of control electronics that are located right around it.
01:38But overall, you know, this is really designed with that in mind.
01:41But if I look behind there, it would look like a rack like I see in a data center, like a classical computer.
01:46It would look like classical racks of electronics we have.
01:51At the moment, these are based off of FPGA control board.
01:55So this is actually our latest Nighthawk quantum chip, right?
02:01And so the Nighthawk is 120 qubits arranged in a square lattice.
02:07So it's more connected than we had in the Heron processor in the previous generation.
02:13But what you're looking at is the chip mounted into a printed circuit board.
02:18This printed circuit board, you can see it has space for where the connectors come down.
02:22These connectors come down to bring the signals to manipulate the states of the qubits or to read them out.
02:27And then this is basically bolted and mounted at the bottom of our dilution refrigeration systems at 15 millikelvin.
02:34But is this the chip right here?
02:35That's the chip right there.
02:36And other than this, this is the printed circuit board on the flip side.
02:39Yeah.
02:40So it's mounted up into this printed circuit board.
02:42And then, you know, the signals are being carried out to all these various connectors.
02:47And does one of these chips go into one of those chandeliers?
02:50Yeah.
02:51So effectively, we're able to fit a varying number of these into cryogenic infrastructure.
02:58A big part of that comes down to not necessarily limitation of the size of this, but the size of the kind of wiring that comes down.
03:05And so a big push of how we're scaling infrastructure comes down to miniaturizing some of the wiring and the cabling and miniaturizing other components that are within.
03:16So that we can get more control for larger numbers of chips within the same kind of cryogenic footprint.
03:22So I'm going to go way past what I understand.
03:25Back in the old days of sort of electronics, you talk about whether they're rigged up in series or in parallel.
03:31Right.
03:32If you put multiple chips inside one, is there an analog to series or parallel?
03:37That's actually a funny, funny point there.
03:38Like, so you can place these multiple chips inside.
03:42We actually are going to be connecting them with cryogenic quantum links over the next year.
03:50So we actually have this technology that we actually demonstrated at first last year as part of a different architecture that we have.
03:57We call it Flamingo.
03:59But it actually allows us to actually scale these and network in a quantum way these various chips together.
04:05So that's one thing.
04:06There's already this kind of, you know, this strategy for scaling, you know, module by module.
04:13Another way is actually that there's classical communication networks in the control electronics that also lets you communicate between these chips, too.
04:23So that you can actually perform a circuit on one, measure the qubits, and in almost real time decide to do something on another chip.
04:32So this kind of dynamic circuit piece is also another capability that we have built as part of our overall infrastructure and platform.
04:38Does adding more chips reduce error?
04:41So that piece is, so that's not how we're approaching the error correction strategy, right?
04:51So error correction strategy comes down to actually a fundamentally different layout of how we are arranging the various qubits on the device.
04:59What that does in terms of having multiple chips really allows us to run a variety of different types of quantum circuits.
05:09So you have different kinds of applications and algorithms that might be possible by leveraging this kind of tight coupling of different chips.
05:17So this chip is actually what's on, so this is, this is actually a stack of two chips.
05:23So there's actually a butter, a butter side down and another chip that comes on top.
05:28That's what you're seeing in the backside of that.
05:30Right, right, right.
05:31And this is the piece that gets mounted up onto the bottom of this board.
05:35But there, but there are several of these, right?
05:38Yes, this is several of these.
05:39There's a lot of these.
05:39Yes, these are many Nighthawks, right?
05:41So we would cut, you know, we'd dice these out and package them and then put them in.
05:46Back to front, potentially, on top of each other.
05:48Yeah, and you can see this is, you know, 300 millimeter scale.
05:51This is silicon?
05:53Very good.
05:53And then it's etched with a sort of lithography processor?
05:56All kinds of lithography tools and techniques that you would, you know, certainly be familiar with for your standard CMOS processes.
06:06For GPUs or something.
06:07So this is actually the inside of one of our testing fridges.
06:12This lab actually we're in is our quantum test and characterization lab.
06:16Testing the chips?
06:17Testing the chips.
06:18So a lot of the either Heron or Nighthawk processors, right, after they come out of FAB, we go through a series of qualification steps.
06:28Some of them are room temperature tests.
06:31Some of them are actually cryogenic tests when they're ready.
06:33So that's what this lab is for, for us to actually qualify and characterize the performance of some of those devices before we decide whether it's good enough to go into, for example, a data center for deployment.
06:47So this basically you recreate the, you recreate the temperature from what you'll use in the quantum here.
06:53Right, right.
06:53We effectively recreate the environment in which it would be deployed in, but in order to get out all the salient parts and characteristics of the device to judge that, you know, this is going to be a good one.
07:06And how cold does it have to get?
07:0815 millikelvin, right?
07:10Okay, so explain that to those who don't think in kelvin.
07:13Yeah, you don't think in kelvin.
07:14I guess zero kelvin is absolute zero, right, minus 273 Celsius, right?
07:24And then zero, we're at 15 millikelvin, which is 0.015, right?
07:31And if you think about, we always say colder than space because space is in around several kelvin, right?
07:38Cosmic Mercury background is in the several kelvin range.
07:41So, you know, orders of magnitude colder than that.
07:43And you get there through liquid helium?
07:45You get that through two stages of cooling.
07:48Okay, so there's liquid helium itself, if you just had it as a liquid, it's 4 kelvin.
07:54We actually don't use it in its liquid form, per se.
07:57We have what's called a pulse tube cooler as part of this.
08:00It's actually the noise you hear, driving a compressed circulation of helium.
08:07That part of it gets a large part of the infrastructure down to around 4 kelvin.
08:12It's actually up at the top stage there that gets down to 4 kelvin.
08:15But hanging off that is another closed-loop refrigerator.
08:19It's called a dilution refrigerator.
08:22And it uses a different cryogen.
08:24It uses a mixture of helium, which is, you know, balloon helium, and helium-3, which is an isotope of helium.
08:31So helium typically has two protons, two neutrons.
08:35You knock off one of those neutrons, you have helium-3.
08:39It's a rare isotope of helium that doesn't exist in a lot of places, but we have enough of it in the world.
08:47But it is part of it, part of the mixture that is used to make these machines actually cool down to that 15 mili-kelvin.
Be the first to comment