Yuval interviews Robert Wille, a computer scientist and co-founder focused on quantum computing software. They discuss the field’s transition from research to practical deployment, the need for heterogeneous and hardware-agnostic software stacks, and the integration of quantum into HPC environments. Robert explains the importance of design automation, open-source strategy, and AI-assisted development, arguing that quantum’s complex optimization challenges resemble those long solved in classical computing.
Transcript
Yuval: Hello Robert and thank you for joining me today.
Robert: Hello Yuval, happy to be here.
Yuval: So who are you and what do you do?
Robert: Yeah, I’m Robert, I’m a computer scientist by training and for more than 15 years now actually involved in quantum computing. We are building software for quantum computing, which is important because it is awesome to have great quantum computing hardware and have great quantum computing applications, but you definitely need software to connect the end users to the hardware and to make these applications work.
Yuval: I think you have at least two hats, right? One is an academic and the other is an entrepreneur. Could you tell us a little bit about that, please?
Robert: Absolutely. By heart, I’m an academic and a researcher. I work at the Technical University of Munich and at the same time we also figured out it is great to do innovation at the university, in an academic setting, but for real impact we also need production-ready software and this cannot really be done at the university environment anymore. That’s why a little bit more than a year ago we also founded a company, and there we now take all the innovation from the academic developments and make it production-ready in a commercial context.
Yuval: Since you’ve been doing it for so long, software in the context of quantum, where are we today? What is the state of quantum software today?
Robert: It is certainly in a transition phase. So most of my career, quantum computing and quantum computing hardware and software have been in the basic research state. But of course, as you all know, a few years ago a real momentum built up. We are not completely there yet that we really can exploit quantum computing, but we are transitioning from basic research to applied research and in the next few years we’re getting there for real quantum computing applications. It’s a very exciting time but also a demanding time because we now cannot linger, goof around and play around like researchers do, try out this a little bit, but we really have to make it work in practice. It’s exciting, demanding, but we’re really right now working towards making quantum computing a reality and a real application rather than an academic dream.
Yuval: There’s a simplistic view that there are only two quantum algorithms, Shor’s and Grover’s. First, do you agree? And two, if that’s the case, what does your company do if it’s not on the algorithm development?
Robert: It’s a very interesting question. So, of course, those are the killer applications everyone is talking about. And those alone would make it absolutely plausible to work on quantum computing, because those algorithms are so disruptive that it’s definitely worth having quantum computing just for that. I personally see more potential than that. In fact, I see a much more heterogeneous computing environment in the future. We’re not just having CPUs and GPUs as so far, but we’re going to have quantum computers, not just for the two aspects you mentioned, but for a broader aspect. And then we’re going to have AI processors and DNA computers. And then the real challenge is to really find the right computing technology for the corresponding applications. And quantum computing beyond the two examples you just mentioned could contribute to several of them, maybe not all, but that’s going to be definitely much more than just two examples.
Yuval: I think today the dominant programming environment is Qiskit. Do you expect that to remain the situation? Do you see it transitioning to a different type of programming environment?
Robert: I mean, Qiskit is definitely, I mean, IBM and Qiskit, they did a tremendous service to the community by providing everything in an open source fashion. They helped a lot building a community. And I think a lot of development still will rely on Qiskit. But as I said before, I think the environment is going to be much more heterogeneous. And I think this is also one of the challenges. Much more directions will pop up. Also, Qiskit focused on a very much quantum-first approach. We now have several other developments talking about MLIR, QIR for example, which goes more in the direction of LLVM. And all of them have their respective pros and cons and all of those developments make sense. And I think we need a software community which follows these different paths, which still, however, remains compatible to each other as much as possible. And then particularly, also recognize the different modalities we have. We have superconducting, but also neutral atoms, ion traps, and all that needs dedicated software solutions. But eventually the end user doesn’t care about the modalities. He or she cares about easy access, and so we have to also bring it together. So to some extent, huge platforms should be helpful and help to unify a little bit, but then we also need dedicated developments for the respective modalities. And the challenge is how to bring that together and to avoid that there’s some solution which dominates all, but also how to avoid that we end up with dozens of solutions and nothing works together with each other anymore.
Yuval: The notion of write once, deploy on any computer is very attractive to end users, but I think it doesn’t hold in today’s reality. Do you envision that happening or do you think that it still has to be very, very platform specific?
Robert: The vision definitely is to have a platform-agnostic access point for the end users because the end users don’t want to study in detail whether or not they should go for this technology, this modality. And we’re working on that. A good example is for example our QDMI, which is a Quantum Device Management Interface, which already is an interface where on the hardware end you can plug in different backends, a superconducting computer, an ion trap computer, a simulator if you want. And on the other hand we kind of have access for certain compilers supporting that. So we are trying to already make that work, that we are building up a software stack that supports different modalities. And this of course involves hardware-specific compilers and solutions, but the more we get towards the end user side or towards the application side, the more hardware-agnostic it’s going to be. In a time where in classical computing we talk about low code and no code, and at the same time in quantum computing we still deal with assembly-like solutions, that’s definitely not going to be the future. We need this assembly, we need these hardware-specific parts as a foundation for a software stack which eventually allows the end user to utilize quantum computers no matter what platform is underneath.
Yuval: Classical computing beyond low code and no code, now there’s a lot of vibe coding where, whether it’s Claude Code or others, you say, “Well, this is what I want to happen,” and now the AI generates code. Do you see that coming to quantum anytime soon?
Robert: I mean, it’s definitely a development which won’t stop at quantum. If in the classical realm the end user is used to not working on programming languages or basically using high-level abstractions, why wouldn’t he or she expect the same for quantum computing? We are not there yet. We still have to basically figure out how actual compilation should work for different modalities. That’s where we do a lot of research and development these days. But of course, if these kind of things you mentioned get established in the conventional realm, it definitely will be at some point a requirement. We’re definitely not there yet. We still have other problems to deal with. But on the other hand, it took us a couple of decades to get there in the classical realm. I don’t expect us to need another couple of decades to get the same level of maturity for quantum computing because we could learn and reuse all those developments from the classical realm. It’s just that we’re not there yet, but we probably won’t take as long as it took us for the classical realm.
Yuval: When we first met several years ago, we were talking about EDA for quantum, right? Just like you design a chip and you can take, here’s a USB controller block and everything gets optimized. Again, do you see this coming for quantum anytime soon?
Robert: I think it has to. Right now, particularly when you talk about, I mean, I’m of course an advocate for design automation, for electronic design automation, because this is where I originally came from. And we’re already utilizing a lot of methods and tools from that domain to develop very efficient compilers nowadays. But also, and I think this is where your question was heading to, also when it comes to designing the underlying hardware. The moment this hardware scales and you cannot manually design your quantum chip anymore, you’re definitely going to rely also on design automation methods for that. So I think it’s a necessity and I really believe that a lot of the problems you have in both the compilers, the software, but also in the hardware design, are very reminiscent of a classical routing problem, a placement problem with different constraints, different objectives, but the methodology, the complexity, and the nature of the design tasks are very similar to design tasks for classical computing, just with different specs or objectives. And for that, it would be a shame not to utilize these decades of experiences from design automation also for this new technology.
Yuval: I’m quite familiar with neutral atoms and in neutral atoms you can shuttle qubits around, you could do parallel operations, this also allows you to use new and different quantum error correction codes. And so when you think about coding an application or coding an algorithm, there are so many degrees of freedom that have to do with where are the qubits, what do I shuttle, what do I move when, how do I parallelize it, what is the quantum error correction code that best fits the application. So this feels like a multi-level simultaneous optimization. Is this just a transient thing because there are not that many qubits and you have to super-optimize to get the most out of your system, or is that life for us in the next decade?
Robert: I would say it’s life for us in the next decades because of course you want to have as performant and as efficient a solution as possible, and if you have all those options, all those degrees of freedom, it increases the complexity and makes it super hard to find the optimal solution. But the interesting thing is, this is also to me nothing new. Also 15 years ago or 20 years ago when I studied, we also there had multi-objective optimization problems when designing classical circuits and systems, just with different things. And of course we may not get optimal solutions because the complexity is way too complex. But we develop heuristics and methods that prune the search space we have and then we find a good compromise between an efficient solution and a solution which has good quality. And to me, to some extent, this is nothing different from designing a smartphone which has the longest battery life and still is super efficient. And these kind of trade-offs and multi-objective optimization happen all over the place. And now we’re just having different but still similar problems with similar complexity in quantum computing as well. And it’s basically a matter of how can we develop software and algorithms that understand the degrees of freedom, that unfold the entire complexity, and then apply methods, search methods or so, to prune the search space down to something which we can still handle, tackle efficiently, but still with really good, maybe not optimal, but close to optimal results. So I’m not afraid of that. I think that’s basically what we’ve done all over the decades. And now we’re just having another playing ground, if you like, to apply those methods and to get the best out of it.
Yuval: Let’s assume a CEO comes to you, a CEO of a quantum software company wants to do chemistry applications for quantum, for instance, and has got a blank sheet of paper. They can choose any modality they want. They can choose any software development they want. What would you recommend to that person to do?
Robert: Very good question, because this is also, I think, one of the main challenges for end users these days, that they somehow are confronted with, as you said, lots of different technologies. Then even once you choose a technology, you have different architectures. Once you have chosen an architecture, you have different compilers, different compiler options. So far, it’s really hard to guide the end user through all those options. What we developed, this is also still research and development, we developed initial tools that, I hate to say it, but that indeed use AI and data based on past experiences to learn a little bit what would be a good decision for this particular application. So this is kind of the initial solutions we already developed, which guides end users through all those options. I honestly believe the more we learn, we can also find more sort of solutions which are not just based on data or AI, but also can analyze the problem and then decide, okay, because of the nature of this problem, this technology, architecture, compiler, yada yada, is better. So we have initial, it’s definitely a big challenge. And to be frank, most end users right now, what they’re doing is they got access to this particular technology and architecture. So they’re just using that. They don’t embrace and utilize the options they have. But I’m pretty sure we’re ending up, as I said before, in an environment where you have a heterogeneous system, we have a lot of options. So not just quantum, but also other technologies. And then the end user needs some guidance. And AI-based methods can help, but maybe also structural analysis, things like that. And again, software is key here, because eventually, software can analyze that for you and guide you through all those options. And that’s also part of the stuff we are doing both in the university but also in our company.
Yuval: You came from an academic environment that typically is more open. But now the commercial side is stepping in. I just saw a paper about doing Shor’s algorithm with 100,000 qubits. Let’s assume that someone developed an algorithm or a way to do Shor’s algorithm with 10,000 qubits or 5,000 qubits. Do you think they would publish it? Is that exciting? Is that worrying to you?
Robert: It’s a very good question. It depends on, I mean, you have to ask this particular question to the ones who develop that. Our philosophy right now is indeed, we are a huge advocate for open science and open source. Even founding our company, we committed ourselves to do open source, which does not necessarily mean we give everything out, but we are actually not a closed source company because we believe with open source, you can have huge impact. And at the stage we currently are, this is more important. And it might be a little bit of a risk for us as a company, but as a community, I think it is really beneficial. And what we are owing to ourselves is a little bit, you see that in the classical realm as well. I don’t want to compare it explicitly. The differences are too big, but we have seen successful open source environments in other domains as well. Red Hat, Linux, there’s a lot of open source and you can still commercialize that by still trying to keep certain parts of it open source. This is our approach right now, being open source to have impact and to bring the community forward. And at the same time, still, we have to think about how we keep the show running, how we pay for that. And so far in our first year, we’re still experimenting, but we don’t think it is absolutely impossible to do. Other companies do it differently. And I absolutely respect that and appreciate that as well. It makes sense both ways. We chose the open source path and let’s see where this takes us. But I don’t believe it’s impossible. It’s an interesting experiment, if you like.
Yuval: We are recording this episode in February 2025. What are the most exciting developments in quantum software that you’ve seen, say, in the last 12 or 14 months?
Robert: What I really appreciate is that right now everyone understands and acknowledges that software is important. In the past years, it was always about the hardware, which I completely understand. Without the hardware, everything else doesn’t matter. But now people really acknowledge that without software, in the worst case you just have very expensive technological systems in your HPC center or in your cellar which no one can use. So the biggest development, it’s a very general one, is the appreciation that software is key. And we are really keen about developments like hardware-software interfaces, like our own QDMI, that we really develop software that clearly connects to the hardware providers. But also on the other spectrum, this entire development on QIR, MLIR is very exciting because it allows us to connect also to the HPC environments, to the classical environments, and eventually allow some real classical-quantum hybrid solutions, which is definitely a way to make quantum computing useful. So the past year or maybe one or two years have been exciting because it really broadened the scope of what software development and what the community is currently doing.
Yuval: You mentioned HPC and I think there’s a pretty wide consensus that truly useful quantum applications will be hybrid applications, classical-quantum. But the quantum people work in a different way. They use different programming languages. They don’t really think, or they haven’t at least thought too much, about billing and synchronization and reservation and so on. How do you see these two worlds coming together?
Robert: I’m quite optimistic. There’s a lot of momentum. I completely agree that right now a lot of development is very quantum-only focused. And I completely understand that because we had to build the machines and then do something useful with the machines. It makes perfect sense. But I also am not that pessimistic that everything is quantum only. There’s a lot of momentum, a lot of communities coming up right now who are doing HPC-QC. The supercomputing community is super active. When you attend those conferences, a lot of people are discussing about how to integrate quantum computing into HPC centers, into HPC environments. A lot of momentum is happening there. And I’m really optimistic that with things like I said before, QIR, MLIR, LLVM, there are a lot of developments in this integration as well. So I see a lot of work in this direction and this is definitely going to be impactful in the near future.
Yuval: As our time today comes to a close, I wanted to ask you a hypothetical. If you could have dinner with one of the quantum greats, dead or alive, who would that be?
Robert: A tough question, because I don’t know. I’m citing a lot of Feynman because I like his way of, you know, his entire didactic way. So if I should choose one, having a nice dinner with Richard Feynman would be, I think, really inspiring.
Yuval: Wonderful. Robert, thank you so much for joining me today.
Robert: Thank you very much for having me. It was a pleasure.
Yuval Boger is the Chief Commercial Officer of QuEra Computing .