Tommaso Demarie, CEO and co-founder of Entropica Labs, is interviewed by Yuval Boger. Tommaso and Yuval discuss Entropica Labs’ focus on quantum computing middleware, aiming to support fault-tolerant quantum error correction, their software development tools, potential timelines for quantum computing advancements, and much more.
Full Transcript
Yuval Boger: Hello, Tommaso. Thank you for joining me today.
Tommaso Demarie: Hi Yuval, it’s a pleasure to be here. Thank you for inviting me.
Yuval: Of course. So who are you and what do you do?
Tommaso: My name is Tommaso. I’m the CEO and co-founder of Entropica Labs. And I’m calling you from Singapore, actually. So with Entropica, we are a quantum computing software company. We started in 2018, and our mission is to build software development tools so that we can accelerate the arrival and adoption, which is very important, of useful quantum computers.
More specifically, we really focus on the middleware layer. So if you think that you have the users at the top of this, for now, still rather imaginary, stack for quantum computing, and the hardware at the bottom, we really love to work in the middleware layer with the goal of supporting fault-tolerant quantum error correction.
And a little bit more about myself. I am a physicist by education, as my name I guess gives away, I’m originally from Italy. Although I did my PhD in Australia on quantum information theory and I moved to Singapore in 2014. So originally I was supposed to be here for two years for a brief postdoc experience and as it often happens it became a much longer journey than I planned. It’s been almost a decade now and I moved here to work at the university and at the Center for Quantum Technologies.
Yuval: When you say middleware, do you mean sort of a library like, for instance, if I wanted to do QAOA optimization, then I call your library and you perform that task? Or is it middleware, meaning that you try to create an agnostic layer for all types of quantum computers? Could you explain what you mean by middleware?
Tommaso: Yeah, of course, of course. There’s a lot in that word. This word is a word that I love very much. You mentioned QAOA, which is a good example and it’s something that we have done for quite some time. In fact, now I’ll digress a little bit and then I’ll come back to the main area of focus of our work, which really is fault-tolerant quantum error correction. But since you mentioned QAOA, something that we have done is pretty much what you suggested, building a library that allows users to deploy QAOAs. And if anyone who’s listening is not familiar with the term, that stands for the Quantum Approximate Optimization Algorithm, an algorithm that can run on today’s quantum computers and can be used to solve combinatorial optimization problems. We released OpenQAOA, which is the library I just mentioned, last year. And the idea was indeed to facilitate users’ work so that they could deploy QAOAs on multiple hardware by streamlining the creation of the quantum computation for the user.
Now, this is some of the work that we have done. When I think about middleware today, I’m really thinking about specifically all the blocks of the software stack, issues like that we need to connect everything that happens at the user level. And I simplified that by simply wrapping everything around the word “algorithm”, and what happens at the hardware level, at the machine level. And again, I go back to the main concept, which is, for us, fault-tolerant quantum error correction. And when I speak about software development tools for fault-tolerant quantum error correction, what I really mean is offering the users the ability to design and implement quantum error correction schemes on a variety of quantum computers. You use the word “agnostic”, I wish I could tell you that yes, we’re going to be able to work with all quantum computers from day one. It’s obviously much more difficult than that, even if it’s our ambition to get to that point. Now we do have our favorite modalities that we work with.
Yuval: When you talk about error correction, it sometimes sounds like it needs to be very, very close to the hardware, both because of the speeds or the coherence time in which things happen, as well as the access to parameters that the hardware gives you. How can you do error correction if you’re not a hardware manufacturer?
Tommaso: Oh, this is an amazing question. There are a few ways to answer that, and it really depends on at what point in time you’re thinking. So today the hardware is indeed in the hands of hardware manufacturers. They can get as deep as they like, they can control it entirely. I mean, this is their job, right? Most of them are building not only the hardware, but they’re also building the electronics and all the control layers that they need to use quantum computers. There are other companies as well that are specifically working on the control layer only, and I’m specifically referring to hardware companies here. What I believe is going to happen in the next few years is that there will be, I think it’s fair to call it a synergetic approach between hardware and non-hardware, or let’s just call them software for simplicity, companies. And what I think will happen is that hardware companies will start to open up some of the stuff that they have to simplify the implementation of error correction because it is for their own benefit as well. They are spending a lot of money to build incredibly complex devices that are still noisy, that are inherently noisy. And I am seeing more and more interest and efforts specifically targeted or directed at the implementation of error correction. So I think it is in everyone’s benefit for hardware companies to really facilitate the implementation of such schemes. And we’re starting to see some of what I am envisioning here at the moment.
Yuval: Different companies are coming up with new features that seem to be specifically oriented at error correction, like mid-circuit measurements or fast classical logic and so forth. I want to dive into that a little bit more if it’s okay. Some companies are talking about error suppression, and some talking about error mitigation or optimal control. Are you more of a software solution that tries to create an algorithm that suppresses or reduces the number of errors? Or is this more of a pulse-level control of hardware once the manufacturers open that capability to you?
Tommaso: I’m glad you asked that. Let me take a step back and share with you our thesis at Entropica. So I mentioned that we’ve been around since 2018, and I’ve been working on quantum computing or quantum information theory topics since 2008… So, it’s been a long time, and if you’re around for some time, you see different ideas, you see changes in the industry, you see new approaches and so forth. When we started the company, we actually had quite some hopes for QAOA. It’s an algorithm that we really love, and I personally was hoping that there could be a path to demonstrate the utility somehow of QAOA and similar algorithms without the need for quantum error correction.
Today, I don’t believe that to be the case anymore. So, if I describe our thesis to you, I will tell you that there are really three steps. The first one is that most likely commercial applications of quantum computing will require fault tolerance. The second step is that fault tolerance will require quantum error correction. And when I say quantum error correction here, I really mean the ability to suppress errors on quantum computers. And the last step is that quantum error correction will require a new software stack. Something that today doesn’t really exist yet. People know that we need it, the industry knows that we need it, but it still feels like this huge challenge that everyone is aware of. And so some people are tackling it and maybe some folks are hoping that it will be solved by someone else, which is also fair.
That said, to answer your question, the big dream that we have is the ability to have a software infrastructure – and it goes back to that middleware part that you were asking just now – that enables a user or us to input a target computation and generate or output a series of instructions that can run on a quantum computer and are provably fault-tolerant. I hope that makes sense. Let me know if… It’s a bit of a mouthful. There are a lot of concepts there. I am aware of it. But that sentence encapsulates our vision, encapsulates what we are doing. And it is also the answer to what you are asking, right? It’s also much about pulse control, and it is really about even the ability to go from a target logical computation to a series of instructions that are provably fault-tolerant.
Yuval: Today there are many competing quantum modalities: superconducting, neutral atoms, trapped ions, and many more. And every manufacturer would tell the story about why they could get faster to large-scale error-corrected machines. What is your opinion? Is there a favorite technology that you say, “Well, I, Tommaso, think that there’s a better chance that they will actually make it,” or you don’t have a horse in the race?
Tommaso: It’s a very tough one to answer. I think all modalities have pros and cons, and so far I’m not saying anything controversial. Maybe I can try to not answer your question directly by highlighting what I see as the three main issues when it comes to hardware. So we have this idea of fault-tolerant quantum computing. What are the obstacles? Why don’t we have fault-tolerant quantum computers today, despite the fact that pretty much all hardware manufacturers will tell us that they are getting closer and their technology is the best and so forth. So I see three obstacles. The first one is that we need to be able to build much higher-quality qubits. When I say that, I am obviously saying something that everyone knows and it’s also an incredibly difficult thing to do. At the same time, the improvements in qubit manufacturing capabilities in the last 10 years have been astonishing. Today is fairly easy to build a qubit. I just love how if you go to pretty much any university where they have an interest in quantum computing or quantum information theory, new groups can very easily get started and in 6, 9, 12 months, they have a chip with a few qubits. And even just a decade ago, it would have not been possible. It would have not been that easy. So building qubits today is not difficult. Building higher-quality or high-quality qubits is incredibly hard. So this is the first challenge I see.
The second challenge is the need for a physical or hardware infrastructure that enables on-the-fly classical logic and integration of the classical logic with QPUs, which is really a key component for quantum error correction. What that means is that you need very fast computing capabilities that allow you to decode and correct for the errors that happen on a quantum computer. And this has to happen in parallel, at least in today’s view, it has to happen in parallel to quantum computation, to the quantum computation.
And the third point is when it’s scalable architectures and scalable manufacturing capabilities, which is actually a huge challenge, right? Because if it takes a long time to build a single QPU, even if it is of very high and exquisite quality, that alone might not be sufficient because today we all have computers and the computing industry is massive because we can build millions if not billions of chips at high scale. So rather than answering your question, I’ve highlighted what I believe are the three big challenges, and today, honestly, I don’t think any modality is able to answer all of them with high satisfaction. It doesn’t mean they won’t be able to, just my way to find a way out and not answer your question.
Yuval: You’ve articulated the vision of the company, but today, as you mentioned, maybe the qubits are not of high enough quality, or the on-the-fly classical logic does not exist or exist only in limited cases. What does Entropica Labs do in the interim while you wait for hardware manufacturers to provide the quality and access that you need to do true error correction?
Tommaso: It is true that we still don’t have all the hardware ingredients to enable error correction in a way that we want, which really means correcting errors, right? Demonstrably showing that we have a logical qubit, and we have two logical qubits, and we have a thousand logical qubits, and so forth. We are not there yet. It doesn’t mean we cannot test error correction schemes. So let me tell you what we do. Let me bring it back to what I said at the beginning. So what we are building today at Entropica is QCD. This is actually the first time I believe we talk about QCD publicly in this way. So QCD stands for the Quantum Circle Designer. And one thing which is quite funny about Singapore is that in Singapore, for some reason that I cannot explain yet, we all love acronyms. It is a bit of a national game. Pretty much everything becomes an acronym immediately. So our quantum circuit designer is QCD and Entropica Labs is EL. So we call it ElQCD, right? Which is also a bit of a funny joke because we have a few Spanish speakers in the company. And if you’re a Spanish speaker and you’re listening, well, you will appreciate why ElQCD I think it’s quite funny. I will also say that we love the name but it might change in the future.
So what is QCD? QCD is software to design, analyze, and output executable fault-tolerant quantum computations. What I just described is the vision that we have for QCD. And we are not able to do all of that just yet. Now what we can do and what we can actually test on some of today’s quantum computers, the machines that are available today, is to start producing, for example, variations of known quantum error correction circuits and then go and check whether they actually maintain error correction properties. So let me elaborate a little bit on that point because it’s actually an important point. So having quantum error correction schemes is actually not enough because the moment you are performing operations on a quantum computer, in this case you are performing quantum error correction, what you are implicitly doing is introducing more errors into your computation because you are performing something on a quantum computer. So the difficulty with any quantum error correction scheme is that they need to correct errors without introducing errors faster than you can actually correct them. And this is a requirement to maintain the fault tolerance of the quantum computation. And this is something that we can already start checking today. We already know the answers, right? We’re not under the illusion that we are going to find, you know, we’re going to demonstrate fault-tolerant logical qubit in the next three weeks or anything like that. But by doing the kind of work that I’m describing, what we are actually looking for is better implementations of known quantum error correction computations or schemes on real hardware. Because there is still a big divide between the theory of error correction and the implementation of error correction. And what we are working on today, one of the lines of work that we have is really to explore how we can reduce the usage or resources on real quantum computers, for example, given a quantum error correction scheme of your liking. I can go on because there are a lot of things that we are doing that I love talking about, but I’ll stop and I’ll pass it back to you, Yuval.
Yuval: When a customer uses this circuit designer, how do they start? Do they start from Qiskit code? Do they start from a high-level functional model? Do they start from Python, like Horizon might provide to you? What is the workflow of someone who uses ElQCD?
Tommaso: Nice. At the moment, we are the main users of QCD. So we are really in the middle of development. We want to start opening up access to QCD for users early next year. This is the roadmap that we have. And for now, the workflow is still a work in progress.
There’s a great question, the one that you’re asking. And we still have open questions with respect to that. Right now internally, yes, we are pretty much working in the way that you are describing. So you will start with a high-level classical language, we could be Python, but what we are really focusing on is the ability to have intermediate layers of abstraction. Ideally, these layers of abstraction will be able to work.
Again, earlier you hinted at the idea of being agnostic. It’s not going to be fully agnostic just yet, but at least to be able to integrate with some of the frameworks that are available in the market at the moment. The best answer I can give you right now is that building the workflow is very much a work in progress is one of the most fascinating aspects of what we are doing because there are quite a lot of open questions. What I can tell you is that as CEO of the company, I don’t want to work in a silo. I don’t want to spend the next two years building a solution internally without asking for anyone’s opinion and then showing up one day and telling the world that we have this fantastic QCD only to find out that this is not how people want to use it. So a lot of the work that we are doing internally is to get very close to the hardware companies, who will be the natural users for QCD, especially at the beginning, and make sure we get as much feedback as we can so that we can build a workflow that is perfect for them.
Yuval: You’ve obviously been tracking quantum error corrections, you’re tracking the hardware manufacturer and the algorithms. So what is your best estimate of when quantum computers will deliver truly useful results?
Tommaso: Okay, we can play a game. Let’s play a game, right? Because the reality is that I don’t know. The reality is that nobody really knows. And I could tell you, “Well, it’s going to happen by the end of the century,” which seems like a pretty safe statement to do, right? But it doesn’t carry a lot of information. So we can play this little game because I’ve been wondering the same as you can easily imagine. So let me give you a few numbers here. It actually took about 170 years to go from Babbage conceptualization of a computing machine to a pocket supercomputer, which is the iPhone. So about 170 years. It also took about 100 years to get to useful physics calculations. They were like simulations of physical systems with a classical computer. That happened in, if I’m not mistaken, it was about 1940.
Then it took 110 years to the transistor, which was ‘47, and about 130 years to get to really the critical technology of today’s computers, which is the integrated circuit and the microprocessors. So this was in 1971. So now we have quite a few dates, quite a few numbers in mind that in a bit of a simple way, but hopefully not simplistic, highlight what happened in classical computing. And my claim is that if you take that and you map it to everything that happened in quantum computing, today we might be at the 1940s of quantum computing. We can be a bit more specific and we can say that we are in 1939 of quantum computing because current machines, in my opinion, they seem to resemble one of the first devices built by Bell Labs, which was called the complex number calculator or CNC, if you are in Singapore.
You can do a couple of fun things and this is where the game starts. You can extrapolate and if you say it took 100 years to go from Babbage Analytical Engine to the CNC, but it only took 40 years to go from Menin, Benioff and Feynman’s ideas about a quantum computer to Google, Sycamore or to the supremacy experiment. So 40, let’s say half. So if today we’re in 1939, given all the dates that I just shared with you, we can try to extrapolate and we could say that in four years we’re going to have our transistor moment, which would be 47. In about nine years we are going to have general scientific applications, which doesn’t mean fully-fledged quantum computers, but still implies useful quantum computers. In 15 years we might have our IC and micro quantum processor moment. 20 years, personal quantum computers, 35 years pocket quantum computers. Now, do I believe everything that I just said? Well, it’s a game, right? So I’m not gonna bet, you know, everything on that. But I think it’s an interesting game because it gives you a little bit of perspective. At the very least, we can look at the past and we can make this very simple extrapolation of what could happen in the future. Now, even everything that I described and my answer to you in the context of the game would be that it’s going to take about nine years to have our first scientific applications, which honestly seems like a very reasonable target to me.
Yuval: As we come to the end of our conversation, you mentioned history and you went back many years in time. So if you could have dinner with one of the quantum greats dead or alive, who would that be?
Tommaso: That is the most difficult question you have asked me today. Okay, I’m very biased by the fact that I’ve just watched the Oppenheimer movie. So that is very fresh in my mind. And the character that has intrigued me the most is actually Einstein. It’s not only because of the book. He was an incredible character. But I’ve always been fascinated by how… Anticipate a lot, especially in the quantum information community. Often there is a little bit of… How can I call it? It’s a bit of an attitude towards Einstein’s ideas, because people say, “Oh, you know, Einstein didn’t believe in quantum mechanics, and it was spooked by entanglement,” and all of that, which is a massive simplification of the incredibly complex and beautiful thoughts and ideas that Einstein had. So I would love to go back in time and have dinner with him because I think his criticisms of quantum mechanics were incredibly profound, as all the conversations and the dialogue and arguments he had with Niels Bohr demonstrated. So, in fact, if I could expand on what you asked, I would love to have dinner with Bohr and Einstein, that would be absolutely fantastic, I believe. And really get from them their views and everything they believed about quantum theory at the time. Excellent.
Yuval: Tommaso, thank you so much for joining me today. It was my pleasure.
Tommaso: Thank you, Yuval. It was really fun.