Next in Tech | Ep. 263 : Quantum CEO Series: Dr. Theau Peronnin

View Full Transcript

Eric Hanselman: Welcome to Next in Tech, an S&P Global podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, Chief Analyst for Industry Research at S&P Global. And today, we have the next segment of our CEO series on quantum computing. And we're going to dig into how quantum capabilities are defined. With me to discuss it is Dr. Theau Peronnin, Co-Founder and CEO of Alice & Bob, a quantum computing pioneer, building fault-tolerant computing in a compact and efficient way. Theau, welcome to the podcast. 
 

Dr. Theau Peronnin: Eric, thanks for having me today. 
 

Eric Hanselman: It's great to have you here. We've talked a lot about quantum architectures and capabilities. But one of the things that we wind up running into with conversations with clients often is a focus around how you define quantum capabilities and what are realistic metrics and really thinking more specifically about when we start looking at the shift to quantum, how we actually characterize quantum capabilities. So often, we get focused on qubit counts, for example. But yet qubits are not a single thing despite a lot of the current perceptions. I'd like to get your perceptions on really what does the qubit represent? And how does that change with different approaches within the quantum sphere? 
 

Dr. Theau Peronnin: Yes. So indeed, it's tough out there because there is no clean standards yet. I think we're still somewhat in the early days of the industry, even though the whole field is moving very fast. So maybe to cover some of the basics. So Qantumbit initially is a physical system. could be an atom. It could be a super connecting antenna on a chip. It can be actually many things. What matters is that you can use it to encode one bit of quantum information. That was the initial definition of a qubit. The problem is that all -- not all those qubits are equal, especially once you start asking them to do computation. And what the community is now trying to solve is how to make those qubits extremely reliable. And this is where Alice & Bob Technologies will come into play, and we'll discuss that later on. But the whole point is that having what we call physical qubits, actual hardware stuff that makes your computer is not really what the users are interested in. What they want more kind of an abstract unit. Just like in your computer, when you're talking about one bit of memory, you have no idea how many transistors are actually involved in handling that classical bit. And it becomes the same thing when it comes to the current state of affairs in quantum that now the industry is shifting away from discussing the number of physical qubits or equivalent of physical transistor in a classical computer and starting to consider more of logical qubits, which are the abstract thing that is handled by software or quantum software programs. But even here, the definition is unclear because the whole point of having logical qubits would be to have qubits that are sufficiently protected against noise that manage to erase or correct for errors as they appear so that the users do not have to think about that noise that limits the performances. So what is currently happening is that we used to have a clean definition of what even a logical qubit is. And nowadays, now that the industry is on the fence, moving from physical to logical qubits, definitions are becoming more and more blurred. So I guess we're kind of in the fog of war where every player in the industry is moving their road map and trying to adjust the naming to fit their current state of affairs. But I guess in a few years, we'll see very clearly. And just to illustrate what I mean by that is that eventually, what matters is what can you actually do with your quantum computer? What kind of applications can you run? And one iconic example of that is to try and run short algorithm to break RSA 2048. -- not that we're especially bad guys or hackers in the mindset. It's just that it's a clean math problem, unambiguous. And that provides you not only with how many logical you must have, but also how good, how little remaining errors can you tolerate in that algorithm. And so this is why at Alice & Bob, we've always benchmarked our architecture against a use case to make sure we're comparing apples-to-apples. 
 

Eric Hanselman: So in this fog of definition, clearly being able to get towards some benchmark is critical. I think it's something we faced, of course, in the classical computing world. But we spent significant amounts of time sorting out the various benchmarks that are necessary. And realistically, today, we've got so many different benchmarks to choose from, whether or not those are high-performance floating-point benchmarks, core computational benchmarks, memory benchmarks. We wound up with this whole portfolio is working with one particular use case enough. How should we be thinking about how we evaluate quantum computers? 
 

Dr. Theau Peronnin: I would provide 2 answers here. I think as we are scaling the number of logical qubits and trying to decipher what are proper logical qubits from those that are near logical qubits, there is one key concept the listener must have, which is the fact that error correction, the thing that lets you move from physical to logical qubits only works if your physical qubits are good enough, meaning there is a threshold of performance that you need to achieve before you can just add more physical qubits, more redundancy so that together through some sort of majority vote, if you want, you can obtain one arbitrary good logical qubit. And today, there is very few companies out there that have demonstrated cleanly that they are under threshold. So I guess the immediate question to ask anyone who's talking about logical qubits is for me that you're below threshold and how below threshold are you because this is really what sorts the signals from the noise when it comes to logical qubits if I may. Now in a few years, we'll get to the first impactful applications. And we know the threshold here in terms of performances. You need at least 100 logical qubits with at most 10 to the minus 6 error per operation, error per gate. And once you enter that regime, then it's fairly simple. You just increase the size or the complexity of the problem you can solve and that provides you a native clean benchmark to compare different hardware. Until then, it's going to be a bit tricky to do proper one-to-one comparisons. 
 

Eric Hanselman: You talked about some of the limiting factors, qubit stability, fault tolerance. I guess I'd like to dig into those in a little more detail. I think we've talked about them a little bit on previous episodes, but stability is, of course, quite critical, error rates, the same. What are the aspects that we should be aware of? And how does that relate to the specific capabilities of actually performing quantum calculations? 
 

Dr. Theau Peronnin: So just to equip the reader with a few orders of magnitude, just to put things in perspective before answering that. At the moment, quantum bits typically, if you take them out of the box, make about 1 hour every, let's say, 1,000 gates. This is at least 18 orders of magnitude more than your classical computer. So 1 billion, billion times more errors than a classical computer. Just to give you a sense of how noisy a quantum machine is at the moment. The good news is we don't need to go all the way through 10 to the minus 21 to be sufficiently good for application. But we still need to be at least at 10 to the minus 6 and more likely 10 to the minus 9 for a lot of those use cases. So you need something else. You need this error corrections because we're not going to find a wonder material that is 1 billion time better than everything we've tried so far. And quantum error correction is really great, but it's actually nothing really new in the sense that every generation of telco going from 3G to 4G, from 4G to 5G before anything else is a change in the error correction embedded in your communication channel. So actually, engineering has done a great deal of progress when it comes to error correction in the classical world. We're just importing that into quantum to rescue the day in a sense. Now in terms of where we are at the moment, it's very interesting because many different modalities, whether you're building quantum computers with individual atoms, control with laser or electric field, whether you're using optical light on chip or like us at Alice & Bob using cryogenic or super connecting microwave devices. Well, all of the platforms are kind of at the same level of performance at the moment. around the threshold of error correction. So we're just at that tipping point where we can start to implement that error correction scheme. And this is exactly the reason why you see so much momentum in the field at the moment, so much investment and enthusiasm because the minute you can implement error correction, then you can go to arbitrate low error rate. If you want to go from 10 to the minus 9 to 10 to the minus 12 improve by 1,000x your qubits, well, you're just going to add a bit of redundancy in your hardware, pay a bit more of cost. The challenge, and this is where Alice & Bob's come into play, that when you run the numbers, actually, it's not just a little bit. You actually have to, with the standard approaches, use a tremendous amount of redundancy. Typically, if you want one good bit of quantum information, you're going to need about 1,000 copies or 1,000 physical qubits to achieve that. And what we've done at Alice & Bob is that we took the time to redesign the building block, the qubit so that it directly has a first day of error correction hardwired within the design. The same way you can engineer a mechanical feedback loop to stabilize, I don't know, any toy out there. We did exactly the same kind of tricks, but with the equations of quantum mechanics. That gave us to the CA qubits. And what the CAD qubit does is that slashes the complexity of error correction by nearly 200-fold. So that makes the whole system dramatically more compact and reduce tremendously the amount of redundancy required. 
 

Eric Hanselman: So we hear a lot about AI applications looking to help that. A lot of these are focused on error correction but postprocessing from the results rather than trying to do error correction or error correction early on in the process actually within the qubit itself. 
 

Dr. Theau Peronnin: Yes, exactly. And I think so initially, we really -- when we started the company, if I'm really candid, the goal was pretty motivated about the feasibility, trying to find an elegant way and very feasible way to build large-scale, high-impact machines. But now that we've made so much progress on our Rmap, now the way we read into our technology is completely different. What we've disrupted are actually the economics of building a quantum computer in the sense that when we say we reduced the size and complexity of the system by 200-fold, what we actually should say is that we reduced the cost of building a quantum computer by 200-fold, not only the CapEx, but also the OpEx of running it. And that changed the forecast from machine that would cost tens of billions of dollars to machine that costs a few tens of millions, if not a couple of hundred at most or the largest one, and that would consume only a few hundred kilowatts. So machines that are definitely within the expectation of data center if you want. So this is really the thing we cracked at Alice & Bob is to take the time to build in the error correction within the design, which funnily enough has a perfect analogy in classical computing. If some of your listeners are experts on low-level electronics, what we've invented is actually the equivalent of the static RAM, the type of memory you have within the processor that is always powered, always stable meant for compute. It's the same thing here. 
 

Eric Hanselman: That's a really interesting comparison. I'm a semiconductor guy from far enough back, and that's a really good point. You can build stable cells, stable capabilities or you can build a lot of supporting infrastructure around it to ensure that it stays stable and also replicate and build in extra circuitry for error correction. And I guess the question is, do you spend the effort to build a more stable capability upfront? Or do you spend the effort to replicate, build in redundancy, and provide supporting capabilities around it? 
 

Dr. Theau Peronnin: The paradigm of quantum computing is very different from the fundamental architecture of classical computing. In quantum, we're doing computing directly in memory. So when it comes to storing information in a quantum computer, it's not as well separated from running programs as in classical computers. So hence, the inspiration to look for static RAM equivalent. 
 

Eric Hanselman: Interesting because, in fact, the state is the fundamental element of the computational process. I'm curious, so we talked a little bit about stability, speed of computational work is also another limiting factor. And we wind up with some significant differences between superconducting qubits, ion trap in the rate at which the computation can actually take place. 
 

Dr. Theau Peronnin: Yes. It's really fascinating that this topic is not brought up more often because I think this is an elephant in the room here in the sense that at the moment, you have different modalities that are at the same level of performances in terms of errors per gate, but absolutely not living at the same clock speed. And when you're building computers, speed actually tremendously matters because this is the productivity of your machine. It's how many jobs can you run per unit of time, per unit of depreciation of your hardware. And so indeed, you have basically a simple dichotomy, which is that machines with massive moving parts, so typically atomic systems that moves atoms around that neutral atoms or track clients typically at clock speeds that are between 100 and 1,000x lower than solid-state qubits, meaning spin qubits, super connecting qubits and photonic. -- which is fun because it's not the first time in the history of computing, you have the same kind of match. If you go back into the 50s or 40s and look at hardware made classical computers back then, you'd find a wide range of modalities. You had vacuum tubes, electromechanical relay, the very beginning of solid-state transistors and all kind of analog devices to run computation. Eventually, the transistor won, mostly because it was dramatically cheaper than anything else, 1 million times cheaper than all the alternatives per unit of computation, but also because it was so much faster. Electromechanical relays were extremely slow in the few tens of hertz, if not hundreds of hertz at most. And when the first transistor came around, that dramatically changed the speed and so the reach of those machines. So today, it's, again, of a great relevance. 
 

Eric Hanselman: It's so many advances, switching time is still a critical part when we start thinking about the fundamentals. I'd like to know what you see as some of the important milestones we should be looking for as we measure progress. You talked about the process of getting to 100 logical functional qubits. Are there other aspects of quantum capabilities and infrastructure that we should be looking for? 
 

Dr. Theau Peronnin: I think there is 2 missing ingredients beyond the deep logical qubits meaning beyond being able to have one unit of memory that escapes the coherence for arbitrary law. Those 2 elements are: first, the ability to do the magic Gate, the Magic State Factory. It's the former name of this gate, but it's if your listener are more accustomed to classical computing, in classical computing, you only need basically one gate, say the or. And from there, you can rebuild any logic. In quantum, you need a slightly more -- slightly bigger set of gates. And there is always one that will be reluctant to combine with error correction. -- nature gives and takes in a way. And so we need a special workaround, which is called Magic State Factory. It's a whole science in itself to compensate for this challenge. And so this is one missing piece of the puzzle. There's a lot of proposals. There's a lot of proof of concepts at the moment. But I think we should put that as a strong milestone on our radar in the next 2 years, I'd say. And then the final piece of the puzzle is how do you interconnect modules. I'm not talking about Quantum Internet. I'm just saying, well, we're not trying to build cray supercomputers. We're trying to build the equivalent of H100, something you can scale if you want 100 of them, 1,000 of them, you pitch your size and you have one large computer as a result of all of them working together. So you need a way to interconnect. And so at Ais Bob, we've published quite a few papers on how that could be done in terms of architecture, but it still remains to be properly demonstrated to do logical gates between interconnecting different chips while not introducing any errors through that weak link between different chips. So those would be the 2 extra milestones to look out for. 
 

Eric Hanselman: The communications problem, even at small scale, is a challenge, but many things to look for. I want to get into maybe where enterprises should be looking as they evaluate readiness -- it's something where clearly, there are a whole host of quantum problems that are being addressed today, but still a relatively small scale, but I think ones that show a tremendous amount of promise. If we think about simple chemical reaction dynamics, some very fascinating things seem to be going on there. How should enterprises think about this? And what's the point at which they're diving in and really looking at how they start their journey and engage with the potential of quantum? 
 

Dr. Theau Peronnin: I think here, there is a great deal of arbitrage at play, meaning the first ones who will be able to leverage quantum capabilities in the very first years of having impactful hardware out there will be able to secure a lot of IP, no materials, no catalysts, and that can reshuffle the cards pretty significantly in various industries. I think a lot of the enterprise we're discussing with are now starting to understand how much is at stake and that the cost of being too early is extremely small compared to the cost of risking of being too late here. Now I think the way I now advise companies to look at or to think about that is rather different from how I used to pitch it 2 years ago in the sense that in the meantime, you have all those AIs and agentic workflows that enter the game. And so now it becomes, I would say, rather unlikely that it's going to be humans reading the dock or the manual on how to program a quantum computer, most likely going to be machines and that. And so the challenge for enterprise is to make sure they've documented their use cases their business logic, where is actually the value to guide their agentic R&D on to where most of the value is. Say, you're working on, I don't know, a new way to make fertilizer, you have a clear target. But if you're a broad industry in pharma, have many different targets, where do you start? Where do you start putting these quantum capabilities at work? And so you need to capture all this business intelligence so that your R&D that will be more and more relying on to compute. And I think we're definitely entering an era of compute-based science and compute-based innovation. In a way, the quantum computer can be thought as the wet bench of the agentic chemists. -- replacing completely actual lengthy, expensive experiments with the access to ground truth to questions that were never asked to novel data that gives them an edge compared to their competitors. Because very soon AI will mix all the current know-how and value will be captured in exploring new territory. This is where Quantum comes into play, providing answers or lifting rocks that no one has ever lifted. 
 

Eric Hanselman: And hopefully finding interesting things underneath them. 
 

Dr. Theau Peronnin: Exactly. 
 

Eric Hanselman: I also -- we're just on the other side of the RSA conference, the big security conference. And Alice & Bob, of course, is a concept in cryptographic protection. It would be great to get some background into where Alice & Bob's name come. Is there a security background? 
 

Dr. Theau Peronnin: Actually, no, it's -- we discovered that afterwards. Michael Rana and myself were physicist by training. And actually, in physics textbooks, especially in fot experiment about quantum mechanics, you have the same placeholders. So for point A and point B, you have the characters Alice and Bob showing the spookiness of the measurement in quantum mechanics basically and the fact that you have correlations between the experiment and so on and so forth. But the reason we picked that name, even though the security aspect is very fitting because we're building a machine that comes with its own set of challenges when it comes to crypt analysis and making sure, yes, everything is migrated in time, and we should hurry up on that front because quantum computers are coming fast, and I don't want to be in that position. So the reason we picked that name was really just to avoid the word quantum. The word quantum is so polluted by pop culture to mean magical. There's nothing magical about what we're doing. Quantum mechanics is the best description we have of nature, whether we like it or not. And building a quantum computer is difficult, but those textbooks exercise, if you want. So yes, we love that. 
 

Eric Hanselman: And I appreciate the plug for people getting moving on post-quantum cryptographic work. We've spent a lot of time talking with that as well. But this has been great. Thank you for all of the insights and all the perspectives. 
 

Dr. Theau Peronnin: Thanks. 
 

Eric Hanselman: That is it this episode of Next in Tech. Thanks to our audience for staying with us, and thanks to our production team, including Sophie Carr, Feranmi Adeoshun, and Kyra Smith on the marketing and events teams. If you like this episode, please subscribe like us. Please keep in mind that statements made by S&P Global employees represent their own views and are not necessarily the views of S&P Global. I hope you'll join us for our next episode because there is always something, Next in Tech!