Compressing the Timeline to Quantum Advantage

Summary

Daniel Volz, co-founder and CEO of Kipu Quantum, a software company working to accelerate the delivery of quantum advantage, is interviewed by Yuval Boger. Daniel and Yuval talk about what Daniel learned about quantum at McKinsey and BASF, their optimized approach to quantum algorithms, the expected timescale for quantum advantage, and much more.

Access the full transcript on The Quantum Computing Report

Transcript

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

Daniel Volz: Hello, Yuval. Very, very glad to be with you today, and thanks for having me.

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

Daniel: I'm Daniel. I'm a chemist by training, recently turned CEO of a quantum computing startup. We want to bring early quantum advantage to industrial end users by being application-specific and hardware-specific.

Yuval: In terms of background, I think you were originally at McKinsey and then at BASF, doing some or a lot of quantum in both of these places.

What have you learned about the needs of users in quantum technology on your journey to Kipu Quantum?

Daniel: Yes, you're correct. I worked with quantum in my past two jobs, at McKinsey and at BASF, a large chemical company in Germany. What did I learn? I think the first learning, which I mainly drew from McKinsey, where I looked at quantum from a cross-industry perspective, was the massive economic potential.

Back then, I was looking at all kinds of industrial users from very different industry verticals: finance, logistics, chemicals of course, which is still my home industry, pharmaceuticals, manufacturing, et cetera. And essentially, no matter where we looked, you could really identify disruptive use cases.

From that point on, it was very clear that there is massive value potential that quantum computing, if it becomes mature enough, can bring to these industry verticals. What I learned at BASF specifically, where I built up the quantum effort together with a great team of colleagues, is that the users are incredibly diverse.

If you think about all the different kinds of things you could do with quantum, at BASF alone we found about 100 different use cases. Users have extremely diverse backgrounds, so I think this is always something to keep in mind. Of course, mostly they have a problem and they want it solved.

But these can be people doing controlling, people with a bread-and-butter business background. They can also be theoretical chemists, and in some cases even quantum physicists, who have been hired recently by a lot of the advanced users. So whatever product comes online needs to cater to these needs.

And maybe a third learning, which was somewhat sobering, at BASF, is that quantum advantage with the conventional approaches is, I think, still somewhat far out. There is still a lot of work to be done on both the hardware and the algorithm level to make this great potential a reality and to start doing something meaningful.

Yuval: I read the press release about your funding.

Congratulations, by the way. I think it mentions that you're claiming you could deliver quantum advantage sooner. On one hand, I'm very happy to hear that quantum advantage could come sooner. On the other hand, I'm very curious: how can you do that?

Daniel: If you look at today's hardware, we have a number of qubits that depends on the system, on the order of maybe 100 if we're just considering gate-based quantum computing, give or take; some platforms have fewer than 100.

If we add annealing, we are in the thousands, more or less. With all of the conventional approaches, if you're targeting fully fault-tolerant quantum computing, you really need a number north of a million qubits to do something meaningful, because you will want to have error correction.

You will have that massive overhead of physical qubits that are busy forming logical qubits to prevent these errors. The reason for that is that current quantum computers are, as I said, small and still have a lot of noise.

But we realized, and this is mainly the work of one of my co-founders, Enrique Solano, a former professor with more than two decades of research track record in quantum computing, that you can shrink down these numbers and requirements if you're hardware-specific.

So what we are sacrificing in order to deliver on that claim is the notion of a hardware-agnostic algorithm. That was more or less the talk of the town when I started my own journey in quantum back in 2017, 2018. Everybody was claiming that hardware-agnostic algorithms were the thing to get.

If industry users had that, they could pay for it today, then just wait until the hardware catches up, and then start solving problems. We think that this is not going to work, because even if the number of qubits doubles every year, it's still going to take a while until we are north of one million qubits.

So that's not something for the near term. Instead, our technology is geared towards delivering algorithms that are very specific to a certain problem and adjusted to a specific hardware setup, a certain hardware machine, to be precise. With that, we can cut away some of the overhead.

We don't require error correction, and we can start solving meaningful problems much, much sooner than in the fault-tolerant quantum computing paradigm.

Yuval: On this specific hardware that you mentioned, is it still a gate-based computer, or is it an analog computer, or is it a quantum-inspired algorithm?

What kind of hardware are you anticipating running this on?

Daniel: We are partnering with hardware manufacturers working on gate-based quantum computing. One of the two technical pillars we stand on is digital-analog quantum computing, which is essentially a mix between analog and digital quantum computing.

We are using native gate operations. We are using analog building blocks, which of course require some more tinkering in our algorithms. But we think it's worth it, because doing so cuts away a lot of overhead. Right now, hardware is great, and there's super cool improvement happening essentially every year, but we are still limited by the number of gates we can run.

Saving there is, I think, really worth it, because it can bring quantum advantage closer to the present day.

Yuval: You mentioned chemistry, and I think VQE is one of the popular algorithms in chemistry. If you take VQE and make it, as you mentioned, hardware-specific to a particular machine, how much savings can you get?

Is it 10% fewer gates? Is it 10 times fewer gates? Can you give me an order of magnitude there?

Daniel: Well, that depends on the problem. It can be several orders of magnitude, depending. I think what's also relevant to understand is that beyond the digital-analog element, we also have something called digitized counterdiabatic quantum computing.

That's more or less a digital-only compression tool that would take an algorithm like VQE, for instance, and again cut away the number of required gate operations, as well as lower the amount of classical pre-processing overhead needed to get the problem done.

So it's really orders of magnitude. And if you translate this into how much sooner for industry users, it's essentially many years, depending of course on problem size and the pace at which hardware development progresses.

Yuval: Is that reduction in the number of gates, or in resources in general, done automatically?

I'm thinking about the recent Nature article where AlphaTensor found a better way to multiply matrices. Is that the sort of work you're doing, or is it more manual adjustment to the algorithm, going in and saying, "Well, if I have this hardware and this algorithm, then I can save a little bit here and a lot more there," and so on?

Daniel: No, I think it's more about doing this in an automatic way. Of course, if you worked long enough with a given hardware setup, you could find all kinds of ways, and people are doing this. It's also a valid effort. Until we achieve quantum advantage as an industry, we need to squeeze out every optimization potential we can.

But it's not about tinkering and squeezing out incremental advantages by cutting corners or optimizing something by hand. It's more about finding systematic ways of doing so. And of course, this requires some sophisticated machine learning technology and a lot of proprietary techniques that help us do both our universal compression and the mapping onto analog building blocks and native gates in a more optimized way.

Yuval: You mentioned that at BASF you realized there were about 100 different use cases. But earlier in our discussion, we were talking about chemistry. Is your focus on chemistry algorithms for now, or is it broader, covering other types of applications?

Daniel: Our technological pillars are applicable to essentially every industry problem.

We did a lot of benchmarking work trying to find out where the limitations of the technology are. Of course, there are some problems here and there that don't lend themselves particularly well to our approach, but I would assume that the vast majority of fundamental problems can be solved, simplified, and compressed using our technologies.

Of course, as a chemist, this industry vertical is particularly dear to my heart, so we think it's certainly where we would like to start, but we are not limited to it. We also did some recent work on portfolio optimization for the financial industry, for instance. And for us now, the question depends again on the use cases, since we are all about making things very specific.

We want to look at different problems, working with companies from chemicals and pharma, but also finance, logistics, and manufacturing, to identify where the compression potential of our technology, applied to real problems, is the biggest. That will then tell us which of these use cases is going to be the first to help us achieve quantum advantage.

That's more or less what we are going through right now. And the likelihood that this is somehow related to chemistry is reasonably high, from my perspective. I would like to see this, because at the very beginning of my career, which I didn't mention, I also worked in a startup using classical computing to do materials modeling.

I really enjoyed the benefits, but I also ran into some severe issues because of accuracy limitations when we were trying to find blue-emitting materials. So I know these issues exist, and I would like to see chemicals be first, but it could also be finance or machine learning or something entirely different.

Yuval: In one of the interviews you gave that I read, you mentioned the GPU, the graphics processing unit, playbook. What did you mean by that?

Daniel: I think the nice thing about GPUs is that they also embody a spirit of being very, very efficient. GPUs are about solving a problem that is also complex, like graphics rendering, using small chip architectures.

And of course, the rationale for GPUs is that smaller chips are less expensive than larger ones. So you can make a much more cost-efficient process if you're doing this kind of mix-and-match approach, tailoring the chip architecture as well as the algorithms to the problem.

In the quantum space, where we are of course still limited by chip size in a way, it's not so much about saving costs; it's about lowering the threshold for solving real problems to the point where it intersects with the hardware's ability to deliver.

So this is something I really like.

Yuval: One last question on the technology. If someone were to try to explain what you do, would it be correct to say that you're building an optimizing compiler or a transpiler, or would you say your technology is a completely different type of product?

Daniel: I think it's a dual step. It certainly has the compression element, the so-called digitized counterdiabatic quantum computing that we are doing. But the compilation into digital-analog building blocks is certainly also there. So it's more or less both of these things, I would say.

Yuval: You mentioned compressing the time to quantum advantage. So the obvious question is: how soon? How soon can a customer expect to get some quantum advantage when working with you?

Daniel: Again, that depends on hardware development, of course. But if I look at the current roadmaps, I think there are three front-running technologies: superconducting quantum computers, ion traps, and now, more recently, neutral atoms or cold atoms, which have been emerging.

If at least a fraction of those roadmaps materialize, then I think we can target quantum advantage for the first isolated problems, not broad quantum advantage, but first isolated problems, on the order of three to five years. That's our current assumption, if the hardware development roadmaps hold and if our compression holds the same way it has been performing on smaller problem sizes and, of course, on the smaller quantum hardware we've been testing it on.

Yuval: Because this is hardware-specific, have you decided which technology or vendor you're going to focus on, or are you working on multiple versions at the same time?

Daniel: We're now working on multiple versions. Of course, there will be one player who will have a hardware platform that is good enough first.

There always has to be a winner in a race. But we think there will also be a mix and match between hardware types and problems at the end of the day. Some paradigms, like ion traps, offer very cool all-to-all connectivity, and there are certain mathematical problems that lend themselves very well to that specific capability.

There are other problems where all-to-all connectivity is maybe more of a nice-to-have and not used to the fullest extent. So we think there will be platforms that are good, using our technology, for some kinds of problems, and others for other kinds of problems.

So I don't believe it will be a winner-takes-all kind of market. From the hardware perspective, it's going to be a portfolio, a number of hardware players claiming quantum advantage over the next few years, hopefully with our help, and always problem-specific.

Yuval: If you would, tell me a little bit about the company. How large is the company, and how do customers work with you? Is it a product that they buy and then use, or is it more project-based at this time?

Daniel: As you alluded to, though I hadn't mentioned it yet, we recently closed a three million euro pre-seed round.

I say euro, obviously, given times of high inflation and an especially weak euro. But we now have full coffers, and with that we have already put together a team of 10 FTEs: of course a lot of physicists, some full-stack software developers, some application domain experts, and some operations professionals with a strong track record to build the organization.

We want to grow further than that, of course, and we are already working with our first customers. As I already said, quantum advantage, even with our technology, is about three to five years out. So we don't claim that we can solve meaningful problems as of today.

So it's more or less a project-related business model at this point. We think that very soon, as it becomes more apparent which specific hardware platform and which specific problem type it's going to be, our full-stack developers already on board will quickly turn this into a scalable product with an enterprise software business model.

But for now, it's project-based. Our favorite customers, to add that, are the companies who already work with quantum computing. We are less interested in convincing novices, companies who have never worked in quantum before, to just test it. I think there's already a very vital ecosystem with a lot of companies who do this particularly well.

I did this myself in earlier days at McKinsey, but those are not the target customers we are approaching. We want to work with the front runners that have been emerging, which already have quantum teams on the ground, which have the capability to know what their use cases are, where we don't need to convince anybody that this is valuable, and where they have a clear understanding of use cases.

Because this helps us, of course, to have clear counterparts in these organizations who can tell us, "Let's focus on this use case. For this, we would appreciate an application and a specific solution," and then work jointly on it.

Yuval: Speaking of customers in various stages of their quantum journey, have you seen customers that are disappointed, that maybe believed too much of the hype, tried something, it didn't work for them, and now they say, "Oh, we're going to put it aside for a few years"?

Or are most customers still in the haven't-tried-quantum stage, or working on quantum right now?

Daniel: I think it's a little bit in between. I can attest that there are a lot of customers who have been disappointed by overpromising from the ecosystem. Back then, the expectations were maybe a little bit driven by overhype.

Some suspected there would be not a doubling, but, I don't know, a five- or tenfold improvement or further leapfrogging in the number of qubits we would have. That didn't happen. It's still a healthy pace, but of course, no miracles happened in the past three or four years. And some customers were led to believe that a miracle would occur, which didn't.

So far, I haven't seen a lot of customers give up on quantum, mainly because there's still this massive opportunity. The carrot is still big and juicy, in a way. So from my understanding, that didn't happen, but of course, people are now more wary. And I think this is also good, because it helps avoid hype if customers don't jump into the arms of everybody claiming miracles in a quarter or two.

Customers are more critical, which is good. Some are disappointed with past progress, but I've not seen anybody give up on quantum at this point in time. Not quite yet.

Yuval: As we get close to the end of our conversation today: you are based in Berlin, or in two cities in Germany. Where is the company based?

Daniel: Yes. The official headquarters is in Karlsruhe, Germany, which is about an hour south of Frankfurt. And we have a second hub in Berlin. The tech team is extremely international at this point, and for that we wanted to pick the most international city, which is arguably Berlin in Germany.

So this is where the tech team is located, and the commercial activity is mainly driven out of Karlsruhe, which, in my humble opinion, has much better weather and climate compared to Berlin, which is terrible during winter.

Yuval: How can people get in touch with you to learn more about the work you're doing?

Daniel: I guess email is the best way to reach me: firstname.lastname@kipu-quantum.com. Feel free to get in touch if you're interested in learning more about Kipu or the kinds of things we can do. We are not fully loaded with customers yet, and we are still hiring.

So if you feel that you are a quantum scientist, a software developer, or a business professional with a good background, and you want to make quantum advantage a reality by being application- and hardware-specific, please be in touch. And I think the same goes for all quantum users who are feeling that pang of disappointment because promises have not been met, and who still want to give quantum another shot.

Those customers, of course, are also more than welcome to reach out.

Yuval: Daniel, thank you so much for joining me today.

Daniel: Thanks, Yuval. It was a pleasure.