Podcast | Topological Qubits are Here! Discussing Majorana 1 — with Microsoft

Konstantinos Karagiannis: This episode was recorded on-site at Microsoft headquarters, where I was invited to see the launch of the world’s first quantum processor powered by topological qubits, Majorana 1. I have to admit, I was super emotional being able to hold this chip in the palm of my hand. I’ve been waiting for this modality to become a reality for about a decade now, and the interview you’re about to hear happened just a few feet away from a working Majorana 1 system.

Find out all about this amazing breakthrough, and how it will house a million qubits, in this episode of The Post-Quantum World. I’m your host, Konstantinos Karagiannis. I lead Quantum Computing Services at Protiviti, where we’re helping companies prepare for the benefits and threats of this exploding field. I hope you’ll join each episode as we explore the technology and business impacts of this post-quantum era.

Our guest today is a technical fellow at Microsoft, Chetan Nayak. Welcome to The Post-Quantum World. This is a weird episode because we’re in person and I just got through the most mentally exciting day I’ve had in a long time.

Chetan Nayak: Thank you very much.

Konstantinos Karagiannis: This was amazing. I got to hold in my hands a topological mini system, Majorano 1. Let’s start with why that’s amazing. That little thing I held in my hands, how many qubits will that one day represent?

Chetan Nayak: What’s amazing about it is that it’s a different type of qubit architecture and a different approach to quantum computing than the first generation of quantum machines. Something of that size, a single chip can one day hold a million qubits.

That potential result is so different from what people are talking about now with multimodule machines where you’re connecting hundreds or more cryostats, each holding some number of qubits, potentially requiring complex 3D integration. Our approach is to not do any of that, to sidestep all of that. In order to do that, we had to invent a new kind of qubit to be at the core of it. That’s the topological core of our quantum processing unit, hence the name Majorana 1, because it’s based on Majorana 0 modes. It’s a reimagining, a reinvention, of quantum processors based on a new state of matter: topoconductors, short for “topological superconductors.”

We’ve created a new state of matter. We’ve harnessed it into a new type of qubits, and those are the topological core of this kind of quantum processing unit, which we envision, if we look at our roadmap, all being on a single chip.

Konstantinos Karagiannis: We’ve been talking topological for about 10 years now on the outsides of the industry, and you’ve been working on it for twice as long as that. I’ve been very excited for this, and I’ve seen it take different shapes over the years, and of course, there have been the ups and the downs, and more recently, it’s been all ups. I’ve reported on it on this podcast a couple of times. Let’s go through those steps. One of the biggest ones is creating an entirely new state of matter.

Chetan Nayak: We’re all used to states of matter like solids, liquids, gases. We grow up learning about those. But there are other states of matter that are some more and some less subtle. There’s superconductivity, which occurs in many materials. When you cool them down to very low temperatures, they’re able to carry electrical currents without resistance. That’s not some physical or mechanical difference, like the difference between a solid and a liquid. But that’s a difference that’s more subtle, that occurs with the electrons that inside the solid. They do what’s called Cooper pairing: They form these pairs. Their wave functions become rigidly aligned with each other. It’s quantum mechanical, and an effect that manifests itself in the electrical properties.

What people have realized over the last two decades or more is that there are other kinds of superconductors: topological superconductors. Those have a remarkable property: Unlike a regular superconductor, which cares whether it has an even or odd number of electrons because in the odd case, one of them is unpaired and there’s an energy penalty — you’ve got an unhappy electron somewhere — in a topological superconductor, an odd number of electrons is just fine. It doesn’t mind at all. There’s no unpaired electron that you can find anywhere. It managed to hide that unpaired electron in the underlying ground state of the system.

This, in turn, is an example of an even more general class of what are called topological phases of matter, in which the distinctions are these very subtle things that are very discrete — like even oddness is very discrete. These very subtle distinctions have to do with topology, which is the study of those properties of geometrical shapes that are invariant under what are called smooth deformations, which is making small changes. You make small changes to an untwisted band and it’s still untwisted. You can stretch it, you can pull it. But if you take a Möbius band, which has one twist in it, there’s no way by smooth deformations to get from the untwisted band to the Möbius band — to get from a wedding ring to a Möbius band and vice versa.

Those kinds of topological distinctions, the quantum mechanical wave functions in these topological phases, have those kinds of distinctions and the associated rigidity and the notion of how hard it is to tear a band and then twist it and reglue it. That’s a property of these states called the topological gap. It’s an energy gap and energy scale that tells you how robust and how stable these states of matter are. The larger you can make that, the more robust your topoconductor will be, the more robust your topological phase will be, the better your qubits will be, the more robust, the less error-prone they’ll be. But also, they’ll be smaller and you’ll be able to operate them faster.

Konstantinos Karagiannis: Is that gap what allows you to take an accurate measurement of the state of a qubit?

Chetan Nayak: It is one of the important features, but it’s not the only important feature. To do an accurate measurement of the state of the qubit, you first have to figure out this degree of freedom that’s been hidden, the parity. You’re going to have millions of electrons in one of these topoconductor wires. You need to be able to tell the difference between a million and a million plus one. That’s a very subtle difference.

The way it’s done is through something called interferometry, where in quantum mechanics, there’s this remarkable property that electrons can take multiple paths around a circuit and they interfere in the similar way that light waves behind a double slit form an interference pattern on a wall — a purely quantum mechanical effect. Interferometers can enable you to measure topological information precisely because it involves something going around a loop.

In order to do that, we take a topological superconductor and couple it to what’s called a quantum dot. Through that coupling, the quantum dot’s properties get changed. It’s sensitive to the quantum state. Our ability to connect the quantum dot to the qubit and the stability of the qubit rely on that topological gap. But it also relies, for instance, on the rest of the readout chain because our ability to make an accurate measurement does depend on our ability to interpret the signal that comes out. The fundamental limits on device performance are ultimately the size of that topological gap. One way of putting it is, you want as big a topological gap as possible so the underlying qubit is as good as possible. Then you want to give away as little of that as possible in all the subsequent steps you have to do to measure.

Konstantinos Karagiannis: The electronics to do this kind of measurement, when you mentioned before that a million qubits will fit on this tiny little talking, graham-cracker little chip, how does that scale? What happens there?

Chetan Nayak: There are two pieces of it. There’s the control, and there’s a readout. The control is extremely simple because almost all the control that we do is, you have a quantum dot, you couple it to the wire and you decouple it from the wire. While it’s coupled to the wire, you’re measuring. The quantum dot is measuring the state. When it’s decoupled, it’s not.

The cool thing is, we have different quantum dots we can couple in different ways to our qubit. Some of them can couple to an individual wire, and some of them can couple to across wires to two wires within the same qubit. Some can couple between qubits.

We can do measurements of qubits natively in different bases. We can do a measurement in the x basis, and we can do a measurement in the z basis. We can also do two qubit measurements — zz or xx, for instance — in the same way. We have the ability to do these measurements. All we’re doing is digitally turning on and off these measurements.

The input bottleneck you would have of trying to control a million qubits is substantially reduced. In fact, we send relatively little information down to the fridge. That information is then by a cryogenic CMOS controller. A classical controller is then sent out to all the qubits. Another way of saying it is, there’s a fundamental challenge in controlling qubits at scale, which is that there’s nothing like Brent’s rule, which says that you may have a billion transistors on a chip, but only a few thousand control lines, because the signals can fan out. On the quantum processor itself or on the quantum chip, there isn’t that kind of fan-out. Every single qubit has to have classical control lines.

The way that we get around that is to have something in the cold, very close to the qubit, that effectively does that fan-out for us — that receives digital messages and then sends out the voltage pulses. It’s the fact that these pulses have carried very little information with them. You turn on a measurement, you turn off a measurement — they’re very digital. That enables us to avoid that input bottleneck.

In a sense, that is very closely related to topology, because topology is a very discrete thing. All the continuous deformations you do to shapes, topology doesn’t care about that. It’s focused on the discrete information you have. That’s true also of the electronic wave functions and of the device configuration — that discrete aspect to it.

There’s also a readout side. You’re turning on and off the couplings to the quantum dots. You then are going to also read out what was the effect on the quantum dot. That readout is much more analog because you are measuring a phase shift. But we then discretize that at room temperature because we can substantially multiplex those microwave signals that are coming back out. We need to just be able to tell, was it a 0 or 1, the result that you got.

Konstantinos Karagiannis: Which could be handled in that other end of the bandwidth.

Chetan Nayak: Exactly, and doesn’t involve any complicated pulse shaping or timing or anything like that.

Konstantinos Karagiannis: In theory, maybe even the readout is less prone to error.

The chip I held in my hands contains a grid of four by two qubits, correct?

Chetan Nayak: Yeah.

Konstantinos Karagiannis: That’s what we have right now, and that form factor will not change when we go to a million.

Chetan Nayak: If you look at the chip, if you look at the processor, the underlying qubits are smaller than 10 microns by 10 microns. That’s a pretty nice Goldilocks regime. If the qubits were smaller than that, it’d be almost impossible to get the wiring in. On the other hand, if they’re too big and you need a million, you’re going to be on many chips, potentially many modules, maybe even an entire warehouse in some of the roadmaps that I’ve seen. But if you’ve got something that’s 10 microns by 10 microns, 1 million qubits is 1,000 by 1,000. Even though you will need some space for the wired and control structures and so on, it’s still putting you in a few centimeters by a few centimeters. Those are still roughly that form factor that you saw.

Konstantinos Karagiannis: That one little thing hanging off a fridge, and it’s all there. Incredible. That prevents your need for any kind of interconnect or anything like that for a long time.

Chetan Nayak: We didn’t want to both solve the problem of quantum computing and solve the problem of quantum networking in order to make something useful. There may be certain situations in which it’s advantageous to have a quantum network and it’s advantageous to do some kind of distributed computing. You don’t want to shut the door on that. But we also don’t want to be, and I believe we’re not in a case in which we are, entirely dependent on quantum networking to make a system that can solve useful problems.

Konstantinos Karagiannis: This year, we’re expecting 50 logical qubits from Microsoft’s work with Adam. All the qubit virtualization that we’ve covered on this show. Those learnings, they’re going to be applied here, I’d imagine.

Chetan Nayak: We’ve been learning a lot about quantum error-correcting codes not just in the asymptotic limit, not just for some idealized error models, but also for specific system sizes and for specific hardware. All these hardware modalities, some things are harder, some things are easier, and the types of errors that occur are very dependent on the hardware. We have been learning how to tailor the code and codesign the error-correcting codes with the hardware. All those learnings in the same discipline we’re learning from that, we’re also applying to our own hardware.

Konstantinos Karagiannis: This first chip, I see some here running and everything. What’s the plan for that being live in some form on the cloud? What does the first Majorana access look like? Is it something this year even on a simple test space?

Chetan Nayak: I don’t think we know for sure yet. I don’t know exactly what there’s a market for. We know we have a much better idea of what the end state is, which is that we want utility-scale quantum computers — fault-tolerant quantum computers that are large enough to solve otherwise-unsolvable materials and chemistry problems. We are a cloud computing business. That is Microsoft’s core business, and we expect that to be an integral part of Azure. Between now and then, it’s a little unclear, exactly, what there’s a demand for.

There definitely are people who have more of a research bent who want to be able to play around with systems that are available that exist right now to learn with them, maybe to peek under the hood and understand how these things work at a physical level. But there are also people who would rather say, “I want to be many abstraction layers away from all that stuff. I just want to be up here and write quantum software and not worry about what’s happening down at that lower level.” For those people, they are going to be much more interested in that end state. It depends. We don’t have a plan right now to release into the wild any of these smaller systems prior to fault-tolerant utility scale. But that could certainly change depending on what there’s demand for.

Konstantinos Karagiannis: I was thinking along lines of research like, people like to see papers and what they ran on. It’s reasonable to expect that any papers that come out will be coming out from your stuff done here, not just folks tapping in. Do you have any guesses at what kind of ratios of qubits would produce what kind of logical yield?

Chetan Nayak: We have pretty good ideas of what kinds of codes — what distance surface code for instance you’re going to want at the scale of useful fault tolerance and/or, ultimately, utility scale. The interesting ratio when we get out to utility scale is going to be in the hundreds, just to be clear, of physical qubits or logical qubits. Exact numbers depend on details. My way of thinking about it is that that point of where you would say, “I have a useful fault-tolerant computer” is, your logical error rates are less than 10–6 — one error every million or so operations — and logical qubits in the 5–20 count. That’s where you start to get interesting. Of course, 50 logical qubits, 100 logical qubits, is really interesting. But that’s the point at which you have crossed the threshold to where you can say, “From here on, we’re scaling up.”

Konstantinos Karagiannis: Microsoft has some pretty rigid definitions of logical. They’re not playing fast and loose. These are very usable qubits.

Chetan Nayak: That’s right. It’s important not only that you’ll be able to detect and correct errors, but also that you need to be able to do computations with them. It’s not just quantum memory — it’s logical qubits you can operate on.

Konstantinos Karagiannis: My listeners could never feel this personally, but having dreamed of topological qubits for so long and to be here today and see it with my own eyes was insane. It was a moving moment.

Chetan Nayak: Thank you. Believe me, I feel the same way.

Konstantinos Karagiannis: And I’m still here. You and I have talked before, and we joked about Majorana. After he discovered the Majorana fermion, he disappeared about a year later. I’m still here. Look at that.

Chetan Nayak: We haven’t suffered the same fate.

Konstantinos Karagiannis: We lived to talk about it.

Chetan Nayak: That’s reassuring.

Konstantinos Karagiannis: Let’s talk about use cases for a second. It’s interesting that Microsoft’s going so hard and heavy into chemistry. Do you want to talk about that?

Chetan Nayak: If you think about Feynman’s earliest papers, his motivation was, it’s hard to simulate quantum systems, because Hilbert space is huge. If you want to do a 100-site solid or a spin system with 100 sites, you need this huge Hilbert space, which is 2100-dimensional, that you’re never going to be able to do on a classical computer. Once you get into thousands, there aren’t that many atoms in the known universe. That’s definitely not something you do with the classical computer, except for the interesting case in which we discover something unusual about quantum mechanics that tells us that there’s some classical theory hiding behind it.

Konstantinos Karagiannis: That it’s not the lowest level of reality.

Chetan Nayak: That it’s not the lowest level of reality, and there’s something even more fundamental. But let’s assume that’s not the case. Every indication we have is that quantum mechanics is the right theory. We’re not going to be able to simulate those things on a classical computer. What a quantum computer is ideal for is those types of problems.

You develop processors that are ideal for certain things like GPUs. What are they? They’re doing matrix multiplication. With GPUs, you’re trying to optimize for something that’s good at matrix multiplication. CPUs are very general-purpose, but they do things like arithmetic logic and certain kinds of operations.

Konstantinos Karagiannis: They’re Turing tape machines, basically.

Chetan Nayak: They take some set of instructions, they figure out some dependencies and then they break them into chunks and execute on them. They do that a fairly flexible way based on some ideas of what kinds of instructions they’re getting. What is it that a quantum processing unit is tailor-made for? It’s tailor-made for simulating operations in a Hilbert space. But it’s not some weird niche application. Just as GPUs turned out, it’s not just for gamers. It’s not just for good graphics.

Konstantinos Karagiannis: AI, mining, Bitcoin.

Chetan Nayak: Mining, Bitcoin, AI.

Konstantinos Karagiannis: Lots of things.

Chetan Nayak: Quantum computers, simulating quantum systems. that is the language of nature. There are many things — materials, chemistry — that are very difficult to otherwise simulate. My career in physics started with me thinking about high-temperature superconductors. Those were discovered in the late ’80s, and they were a big revolution. People knew, for over 100 years, about superconductivity. Mercury was the first superconductor, discovered by Kamerlingh Onnes in Leiden back in the early part of the 20th century. People knew about aluminum, lead, lots of other things.

It was a revolution in the late ’80s when these other materials, like atrium barium, copper oxide, lanthanum, strontium, copper oxide, were discovered with these very high superconducting transition temperatures over 100 Kelvin, above the liquid nitrogen temperature. You didn’t need liquid helium. You could cool them down with liquid nitrogen. At the time, there was a lot of excitement that the temperatures are going up — maybe we’ll even have room-temperature superconductors.

Konstantinos Karagiannis: We just tasted that recently when literally everyone in the world was trying to recreate it.

Chetan Nayak: But the honest answer is, we don’t know exactly why those materials are superconducting even at 100 Kelvin, and we don’t know what the fundamental limits are. Could there be room-temperature superconducting materials, or even just a little bit higher-temperature? Those are fundamentally systems in which out of simple constituents — electrons and ions and the mutual interactions — these complex behaviors emerge out of them. They emerge out of the quantum mechanical interactions between these collections of electrons and ions, which have pretty simple underlying laws and equations. Yet a very complex behavior emerges out of them. It’s fundamentally quantum mechanical. We can’t simulate those systems with a classical computer. That’s not just true of superconductivity. It’s true of magnetism. It’s true of other kinds of interesting and fundamental materials properties.

Early in my career, I was looking at those systems, those kinds of materials, and trying to understand them, trying to come up with theories of looking at the data, trying out different approximations that give us some classical insights into how these materials work. But the way to solve these kinds of problems, to understand materials, is to use a quantum computer. That’s the direct approach to understanding these. These kinds of materials are super important for the future of technology. There are lots of things we could look at technologically that we could do. The laws of physics don’t preclude them. It’s just that they’re hard to do. In many cases, it’s because the properties of the materials that we have aren’t good enough to do those things.

If we had a deeper and better understanding and richer understanding of materials and the potential properties and the ability to design materials, that could open up a new set of possibilities. I’m thinking about the materials context, but that’s also true in chemistry. Right now, a lot of discovery of things like catalysts or drugs, there’s a huge trial-and-error component to it, and simulations, because many of these things have an underlying quantum-mechanical basis and are molecules that are a bit too complex to solve. Our simulations are of limited use in many cases. Having quantum computers would change the way we operate all those things, narrow down the possibilities from a large search space to a much smaller set that’s most promising, which you would then, in the lab, approach.

Konstantinos Karagiannis: A lot of this is working in tandem with AI. I know that that’s been a big thing — the Quantinuum announcement and other approaches.

Chetan Nayak: All these things are part of the puzzle together. Quantum processors are good for specific things. What people don’t always appreciate is, a quantum computer, or quantum processing, is not just a 10X faster version of a classical or 100X faster or 1 million X faster. It’s not just a faster. It’s fundamentally different. It’s overused to say it’s apples and oranges, but many things that a classical computer does now, a quantum computer will be terrible at. It’ll be extremely slow and bad.

Then there are things that a classical computer today could never do, that a quantum computer, when it gets that scale, would be relatively easy to do. When you look at a quantum computer versus a classic computer, there is a little bit of, the hard is easy and the easy is hard. There are a lot of things that are easy for today’s classical computers that are hard for a quantum computer. A lot of things that are super hard for a class computer would be easier for a quantum computer. They are, in a sense, very complementary. You’re going to need both. With any given workload, you will be giving some parts of that problem to the quantum processor and some parts to the classical processor.

Konstantinos Karagiannis: I held Majorana 1. It’s feasible that Majorana 12 will find a way to improve the materials to make Majorana 13 or something. It’s possible.

Chetan Nayak: You’re not the first person to come up with that wisecrack. That’s something we’ve said. Decades ago, I remember joking about that: If we just did a quantum computer, we’d know how to build a better quantum computer.

Konstantinos Karagiannis: But it’s amazing that we’re even here. Like I hinted at before, you’ve been at this 19 years.

Chetan Nayak: I’ve been at Microsoft for 19 years, all of it within one version of the quantum program or another.

Konstantinos Karagiannis: How does this feel for you? This is quite a moment.

Chetan Nayak: It’s exciting. I can tell you, the moment I saw some of the first data from some of these devices that we showed you today and earlier generations, it was spine-tingling. There was that sense that here are these Majorana 0 modes, these topological superconductors. The math works, but are they going to exist in the real world? That’s a higher hurdle. When we first started seeing some of the data that looks exactly like what the theory predicts, that’s a special moment.

Konstantinos Karagiannis: I got to see some of the art that’s going to be in Nature. The Nature article is live online, and of course, there’s a Microsoft blog and everything that’ll be live. All that’ll be linked in the show notes because I strongly encourage everyone to take a look to fully grasp this.

Thanks for making this a reality. It’s incredible. I cannot believe I’m here and I saw what I saw.

Chetan Nayak: I’m glad you came, and it was great spending time with you and discussing this.

Konstantinos Karagiannis: Now, it’s time for Coherence, the quantum executive summary, where I take a moment to highlight some of the business impacts we discussed today in case things got too nerdy at times. Let’s recap.

Nature released a paper validating how Microsoft created a new state of matter to get topological quantum computing to work. The new material stack of indium arsenide and aluminum, built literally one atom at a time, is a topoconductor, or topological superconductor. Majorana 0 modes, or MZMs, are created on the topoconductor and can store quantum information based on whether an odd or even number of electrons are present. To take a measurement, a digital switch couples the ends of the nanowire to a quantum dot. Microwaves then reflect off the dot to tell the odd or even state very reliably.

The resulting qubits, called tetrons, due to their four-ended, H-like shape, are less prone to error, which should lead to better error-correcting ratios of physical tetrons to logical qubits in the future. Think hundreds to one for rock-solid logical qubits with one in a million or fewer errors. The Majorana 1 chip holds a 4 x 2 array of tetrons, but as each qubit is only 10 microns wide, it can hold a million qubits in a couple of centimeters one day without changing the form factor of the module.

I held this QPU in my hand, so the miniaturization is amazing. There is a control chip on board to prevent the need for excessive wiring running down the fridge and into the board.

Microsoft has already announced that DARPA advanced them to the next stage of the US2QC program for creating a fault-tolerant system. Microsoft is expected to build out fault tolerance with 1 million qubits in this architecture within years, not decades. Microsoft believes the first killer application for such a machine will be in chemistry and other simulations of the quantum universe. This is very much in line with the simulation work Feynman envisaged back in 1981, when he came up with the idea of a quantum simulator, and will be made possible by an older, fascinating discovery by Ettore Majorana in 1937.

That does it for this episode. Thanks to Chetan Nayak for joining to discuss Microsoft’s first topological QPU, the Majorana 1. Thanks for listening. If you enjoyed the show, please subscribe to Protiviti’s The Post-Quantum World and leave a review to help others find us. Be sure to follow me on all socials @KonstantHacker. You’ll find links there to what we’re doing in Quantum Computing Services at Protiviti. You can also DM me questions or suggestions for what you’d like to hear on the show. For more information on our quantum services, check out Protiviti.com, or follow Protiviti Tech on X and LinkedIn. Until next time, be kind, and stay quantum-curious.