Théau Peronnin, CEO and co-founder of Alice and Bob, a company making fault-tolerant quantum computers using cat qubits is interviewed by Yuval Boger. Theau and Yuval talk about what cat qubits are, why they protect against bit flip errors, the kind of software that will be required to take advantage of these new machines, and much more.

### Transcript

**Yuval Boger:** Hello Théau, and thank you for joining me today.

**Théau Peronnin:** Hi Yuval.

**Yuval:** So who are you, and what do you do?

**Théau:** So I’m Théau Peronnin. I’m the CEO and co-founder of Alice & Bob. And at Alice & Bob we’re building a universal fault-tolerant quantum computer to sell its computing power.

**Yuval:** Many companies have the goal of building a universal fault-tolerant quantum computer. What is unique about the way that you’re going about it?

**Théau:** Yes. At Alice & Bob we’re building a quantum computer using the well-known superconducting circuit platform. But within that subdomain we are an exotic player. We’re using what is called the cat qubit, which is a special superconducting qubit that is designed to protect against one type of error. So basically, by design, cat qubits are stable against bit flips. And that allows to design a full-tolerant quantum computer with much less redundancy, with much less complexity and overhead. So it’s really sort of a fast track to fault tolerance.

**Yuval:** Before we get into the technology, you mentioned “fast track”. How fast is this “fast track?” When do you expect to have a computer that’s tolerant against these common bit-flip errors?

**Théau:** Yes. The target for Alice & Bob is to deliver the first logical qubit, at least demonstrate all the key ingredients of it next year. So it’s very soon. And at that point, we’ll have already demonstrated that we can correct against bit flip, and now the challenge is to correct the remaining phase flip errors. In quantum computing, you have two types of errors, bit flip and phase flip, that are equally important to correct for. And so, the real challenge for us is to truly escape decoherence. To engineer a system such that it remains quantum, it remains coherent for a very long period of time. And this target for us is, let’s say, next year. And from there, while it’s about muscle, it’s about scaling up the system size and designing a large-scale fault-tolerant quantum computer out of those elementary blocks.

**Yuval:** So, could you give us maybe a popular science description of what cat qubits and cat states are?

**Théau:** Yes. The cat states are named after the thought experiment of Schrödinger’s cat. The idea of that thought experiment is quite fun historically because it was a thought experiment that was aiming at saying computing is kind of out of reality. So it was a thought experiment where you have a box, you put a cat in it, and you have some special trick that connects an atom to a mechanism that would trigger some explosive or poison that would kill the cat inside, but only if the atom decays through radioactivity. And so you could put the atom in a superposition because it’s a quantum particle, so in a superposition of decayed and not decayed. And that would, in turn, mean that your cat, a macroscopic living organism is in a superposition state. And so, that for the experiment, obviously, is not what we’re doing in the lab.

I mean, no cats are harmed in our lab. What we have are cat states within harmonic oscillators. So to put it in simpler ways, in simpler terms, the idea is to have a harmonic oscillator. So a resonator, just like a pendulum, if you may, is in the superposition of two sorts of classical states. Typically, a state with a well-defined phase and amplitude, those are called coherent states of an harmonic oscillator. It’s really the closest to classical that you can get. And so, a cat qubit, a superconducting cat qubit is really, in simpler terms, it’s just a resonator, an antenna on a chip where you have a superposition of two electromagnetic waves with the same amplitude, but opposite phases. And this is, in a very rigorous way, a cat state, hence the cat qubit name.

**Yuval:** And what makes it easy to correct for these errors if you do have a cat state?

**Théau:** Yeah, so the breakthrough that was made by both here and the Paris research ecosystem that we are a spinoff of, was about a decade ago, to realize that those cat states actually can be stabilized. They can be stabilized through a very well-engineered dissipation, some exotic dissipation that exchanges photons two by two with the environment. So basically, you have resonators that actually exchange energy with its environment, but by pairs of photons. This is not something you find in nature, you have to engineer it. And when you do so, and you go through the maps, you can see that if you start in one of those two coherence states, or let’s say, a given phase, an amplitude. While it’s exponentially hard to switch to the opposite phase, and this would mean switching from one phase to the opposite, one would be a bit flip in the Bloch sphere of the cat qubit.

And so, it’s really a shift of paradigm in the sense that, whereas most players in the industry try to isolate as much as possible their quantum bits. What we’re doing at Alice & Bob is actually trying to couple our quantum bit as much as we can to the environment, but through a very well-defined channel that is able to exchange energy, so continuously replenishing the resonator in energy while not exchanging information, while not inducing decoherence. And this is really the trick that was developed by those two groups and that we are pursuing at Alice & Bob.

**Yuval:** So your qubits are different from regular superconducting qubits. What about the gates? What about one qubit or certainly two qubit gates? Are they similar, or are they also very different?

**Théau:** Yes. Here we’re going to get a bit more tricky. So what I described previously was what is called a bias noise qubit, meaning a qubit that has very few bit flips but still has many phase flips. So it has a biased noise in that sense. And then the mindset of the error corrections scheme we have at Alice & Bob is to correct for the remaining phase flip through redundancy. And so, in that regard, what it means is that your error correction code is going to be a simple 1D error correction code just to correct for the remaining phase flip. But you have to be able to assume that you haven’t introduced any bit flips in the mix. And so, to answer your question, what that means is that all the gates that you’re going to do at the physical level have to preserve that noise bias.

They should not introduce some new uncorrectable bit flips. So, for example, we do not have a physical Hadamard gate. You see, the Hadamard gate, which is one of the most common gates in quantum computing, actually would change a bit flip into a phase flip and vice versa. And so, we cannot do that at the physical level. So we’re kind of different in the physical gates we can implement. We’re limited to what are called bias-preserving gates. And through there we are able to do that error correction code, and then at the logical qubit level, here we can have a universal set of gates and recreate the Hadamard gate. We actually published all this roadmap in 2020. And our competitors from AWS also built on that vision of bias-preserving gates on cat qubits in their roadmap.

**Yuval:** So you mentioned the single qubit and how you do two qubit gates. How about the other infrastructure? Do you also need cryogenic cooling to basically the same temperature as regular superconducting qubits?

**Théau:** Yeah, exactly. So we’re very strong, that when you look at our lab, it’s somewhat the same kind of infrastructure as any other superconducting circuit player, meaning that we have dilution refrigerators, racks of micro electronics, FPGA boards to generate signals. The main difference is the level of complexity or the number of physical qubits we’re aiming for. Given that each of our physical qubit, each of our cat qubits is already correcting for some of its errors, while we’ll need much fewer redundancy, actually, orders of magnitude fewer physical qubits to build logical ones. And that, in terms, kind of softens a lot the constraint on infrastructures in terms of cooling power, in terms of the number of microwave lines you have to input, in terms of the number of control electronics, devices, you need to buy a large-scale quantum computer. So we’re kind of the fast second in that approach, leveraging what has been done in the ecosystem in terms of scaling superconducting circuit platforms, but while requiring much less overhead, much less complexity. So that it’s definitely within reach of a startup. We’re definitely a smaller player than some of our competitors.

**Yuval:** So one of the leading superconducting qubit vendors or supercomputing is IBM, and they’re planning to release a 1,000 qubit chip. Now, given that you have dramatically different error correction, what is the equivalent number of qubits if you wanted to run roughly the same application of IBM’s 1,000 qubit chip on your chip?

**Théau:** Yeah. So one application, one algorithm that you might want to run on some of our competitor’s hardware, would be actually to do an error correction code and try to build a logical qubit that fixes 10 to the minus six to 10 to the minus nine error rates, which are the ballpark of where we believe some true exponential speed-up can be demonstrated unequivocally. And so, the numbers that are often cited for a surface-code approach that of usual transformer or usual superconducting circuits is typically about 30 times 30 physical qubits. So about 1,000 physical qubit to build one logical qubit out of a surface-code architecture. Well, with the same level of assumption with cat qubits, you actually only need to correct from one error. So you actually only need one line of physical qubits, so basically, only 30.

So, the rough idea is to say that you’re gaining a square root factor in terms of overhead. This is a very rough idea because then you have to talk a bit about how you’re going to do the non-default gate, how you’re going to do the magic state distillation and all the exotic things that come on top of the error correction. But the orders of magnitude are definitely within this ballpark.

**Yuval:** Tell me a little bit about the company if you can. How many people are you? What level of funding? When did you start operations?

**Théau:** Yeah, so we started Alice & Bob in February 2020. My co-founder, Raphaël Lescanne, and myself. At that time, we had just published the first demonstration in Nature Physics of that exponential suppression of bit flips. I believe, at the time, it was the first demonstration. And so, we started Alice & Bob with €3 million seed round with French venture capital firms. And a bit more than two years later, we’re now nearly 60, maybe 58, I believe, as of today, we raised a total of 30 million of equity to venture capital firms, mostly French. And we’re definitely rising toward that technological breakthrough or what we believe to be the Sputnik moment of quantum computing, the logical qubit. And this is our core goal to deliver on the cloud our first logical qubit next year.

**Yuval:** I read someplace that IBM invested until now about $1.5 billion in building their quantum program. I don’t suppose you’re also aiming for a square root of that amount to build your computer. So will you also need hundreds of millions of dollars to reach a maturity level?

**Théau:** I think it would be extremely arrogant to believe we can do with much less. The thing is, we’re picking our battles. Just like any startup, we have to go step by step and convince investors at each step that we actually deserve, or are the right fit, the right team, we have the right tech. And it’s all about tempo. Trying to time your growth with the opportunity, and not rushing too fast, too early, and crumbling under your own weight. So definitely, we’ll have to scale just like most of our competitors before us. But we believe we’re right on time in terms of tech maturity and growth of the company to deliver actual exponential computing power on time for our end users.

**Yuval:** A lot of the current software algorithms try to assume that there’s noise. We are in the NISQ era, and so you have hybrid algorithms, and you have all kinds of methods to say, “Well, I can’t run a super complex algorithm that’s pure quantum because of coherence and so on.” Do you think that your machine will spurn a development of a new class of algorithms or something that maybe doesn’t have to account for all the noise of the existing first-generation models?

**Théau:** So, you’re right, this is definitely the mindset of Alice & Bob. So we’re not sure of many things, but what we’re sure of is that once you reach a sufficient level of performance, once you have sufficiently good logical qubits, then you can go for algorithms that are guaranteed to have exponential speed up. And the other key ingredient we have in mind is the fact that classical computers are very good. So to beat them at a true real-world task, you actually need more than a quadratic speed-up. And it’s slightly that you’re going to require exponential speed up to actually demonstrate an advantage in the decade to come. So with those ideas in mind, the battle we picked at Alice & Bob is to really focus on escaping decoherence, delivering a platform that is both scalable and robust against noise, to first aim for those use cases, those algorithms that are guaranteed to have exponential speed up, and actually require a very deep algorithm or a very large depth of circuits.

**Yuval:** To that end, is applications a focus of yours at the company, or are you relying on collaborations, or at this time, basically, the main focus is just getting the hardware done?

**Théau:** So we’re a hardware company. We’re doing everything in-house, from the really low-level math of the controlled theory of our qubits, up to operating the devices and plugging them and coding the bias, if I may, of the quantum computer. But we’re not a quantum algorithm software. For that, we’re relying on partnerships with players that are focused and have very good expertise to leverage our hardware. So, it’s definitely the latter, in the sense that we are focusing on hardware, focusing on building a platform with which our partners can actually deliver value for our clients together. And we’re definitely open to more collaboration in that regard.

**Yuval:** As we get close to the end of our conversation today, I was curious, if you could have dinner with anyone in quantum computing, dead or alive, who would you want to have dinner with?

**Théau:** That’s a tricky question. I’m going to answer on a personal level. I’ve joined physics early on because I was a big fan of the character Richard Feynman. And I mean, every other talk in the field starts with a quote from Richard Feynman when he was helping his son, I believe, build classical computers at the time. And he realized it could be interesting to start playing with quantum mechanics. I really would be curious to have his updated point of view on where we’re going, and what advances for science we can hope are coming from quantum computing or quantum information in a more general sense, just like classical computing pushed forward statistical physics and a lot of field of sciences in a very generic way. So that would be my go-to dinner.

**Yuval:** Fantastic. Théau, how can people get in touch with you to learn more about your work?

**Théau:** Yes, on our website alice-bob.com, you’ll find a way to contact us very easily. We are on most social networks, and obviously, we’re a company created by physicists for physicists, so we’re still publishing quite a bit, staging for our academic papers, and you’ll be able to meet us in conferences for sure.

**Yuval:** Well, thank you very much for joining me today.

**Théau:** Thank you, Yuval.

Yuval Boger is an executive working at the intersection of quantum technology and business. Known as the “Superposition Guy” as well as the original “Qubit Guy,” he can be reached on LinkedIn or at this email.

December 3, 2022