Marcin Dukalski of the Aramco Research Center

Marcin Dukalski, of the Aramco Research Center, is interviewed by Yuval Boger. Marcin discusses his team’s focus on advanced computing for Aramco’s upstream business. He elaborates on the computationally intensive nature of subsurface imaging and how they are working to make it more manageable through quantum computing. He envisions a hybrid approach, combining quantum and classical computing, for the foreseeable future. By the end of the year, he expects end-users at Aramco to begin solving shallow subsurface anomaly estimation problems using quantum computers.

Full Transcript

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

Marcin Dukalski: Hello, Yuval, and happy to be here. Thank you for having me.

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

Marcin: Well, currently I’m leading a team of quantum computing researchers at the Aramco Research Centre in Delft in the Netherlands. So our center focuses on advanced computing for the upstream business of Aramco as a whole. A couple of years ago, our management recognized that Aramco is a very important emerging technology, and that’s why they dedicated resources for us. And here I am. My history with quantum goes back, I think, more than 15 years. I have a master’s in theoretical physics, which was on the high energy physics side of things, very much not applied. So after graduating, I decided that I would like to do something a whole lot more applied. And that was back in 2009 when I came to Technical University of Delft and I applied for a PhD position in quantum computing, which I was very happy to get. And for the next four years, I worked on trying to understand the mathematics or trying to model quantum hardware. Basically, we looked at NV centers and transmon qubits, and mostly aspects of what practical quantum measurement looks like, and how this relates to decoherence. I actually had a chance to see a quantum measurement in action because we managed to stretch it out in time, and then do some parity measurements and entangle some qubits.

That was a lot of fun. But that was 2013, and there wasn’t a quantum computing industry yet. And the other thing that I’m very fascinated about is energy. So I joined the energy industry. And I started working over there on some of the most computationally challenging problems that we have as a society. And that’s subsurface imaging, specifically for both oil and gas exploration, but also say storage of CO2 or hydrogen. A few years later, I got pulled back into quantum because the interaction between academia and industry has strengthened. And in 2018, I was a, what do they call it? An industrial advisor for the Quantum Delta organization. So it was really nice to see how quickly that industry has grown and that we can do things together. And then everything happened very quickly. In 2021, we started programming quantum annealers and started identifying some early use cases in geoscience. A year ago, we had our first in-house solver. A few months ago, we started processing gigabytes worth of data on our quantum computer. We even started making our first maps of the subsurface that came out of a quantum computer output. You don’t do that every day, that’s for sure. And yeah, we’re now looking at different quantum computing architectures, different strengths and weaknesses of different computers. It’s a really, really exciting time.

Yuval: You mentioned subsurface imaging. That’s like an ultrasound for the Earth, right? What is the problem? What is the computational problem that you’re trying to solve when doing subsurface imaging?

Marcin: Yeah, that’s actually what makes it quite a difficult question to answer, right? When it comes to subsurface imaging, You’re dealing with a very large amount of data in the process. And quantum computers are just inherently not really good at handling large amounts of data. I mean, sure, you can try to store it in a superposition. But technologically, we’re just not quite there yet. So there are quite a lot of bottlenecks that we have to overcome. So we’re looking at problems that are small enough, but at the same time, difficult enough such that we could benefit from quantum computers. And that process by itself is very difficult because there are just so many boxes that you have to tick on the quantum computer side. And you’re trying to find an application to that. So it’s a little bit like a solution trying to find a problem. And that’s what makes it a little bit of a challenging story.

But when it comes to subsurface imaging as a whole, In principle, it’s a very large optimization problem that we’re trying to solve. As a result, it’s a very good fit for analog quantum computing. The problem is that it’s a matter of figuring out exactly how you’re going to break it down into smaller pieces, and what aspect you’re going to optimize for first. So we take a two-pronged approach to this. One of them is that we try to identify a use case and we try to write it in a way that is quote-unquote quantum native. So we reformulate the problem and then hope that it fits with the the way the quantum computer would like to see it done. The other way of doing this is we see the overall process of computation as a big logistical problem that we’re trying to solve. So that’s where we’re trying to fall back on all those tools that have been developed already around quantum for operations research.

Yuval: And when doing subsurface imaging, you’re trying to detect surface anomalies, right? So that’s basically saying, is there a rock there? Is there oil? What’s the goal of subsurface imaging? And why is it difficult computationally?

Marcin: That is a very good question. You alluded earlier to ultrasound or medical imaging. Over there, you have data acquisition, data processing, imaging, and then interpretation by somebody conducting the ultrasound, all happening in real-time. When you try to do the same thing for subsurface imaging, data acquisition takes months, and you start collecting petabytes of data. And processing and imaging could take also many, many months. the entire project could run into a year or more. And there is a lot of supercomputing involved. This kind of problem is probably one of the top three global consumers of high-performance computing. And that’s why we actually have the supercomputer workshops, or high-performance computing workshops, where people from the industry come together and think about how to solve those very difficult problems.

And that’s why many, many companies in our industries own or at least lease a supercomputer. And Aramco has two computers in, I think, a top 40 of all supercomputers in the world. That’s how big of a problem this really is. You mentioned anomalies. And that’s indeed one of the first applications that we have developed. We realized that there is actually a small and difficult problem that we can solve at scale using analog quantum computing. And that specifically has to do with trying to identify anomalies that are below the resolution limit. So a seismic image is just a collection of black-and-white wiggles and it takes quite a number of years to actually start being able to make sense of what is what. But a lot of what is happening near the sources and receivers at the surface, because we can only image the subsurface from one side, of course, a lot of what’s happening close to the– in the near-surface can dramatically distort your image. So that’s where you have, for example, a cave or an underground river. And those sound waves that are propagating through the subsurface get either accelerated or slowed down in those intrusions, in those anomalies. And what you can do is you know what should be generally roughly the trend in the data, but you don’t know how to align it. And the alignment happens relative to each other. And that’s what makes it so difficult to do. And that’s what we tried solving. Once we found out what those time shifts were, we were able to then invert for the anomalies and get a map, and identify where the cave is or a shallow salt body, for example.

Yuval: Putting aside subsurface imaging, what other interesting use cases do you see in the energy industry?

Marcin: Well, there are probably many. but identifying a really, really good one is hard. I can tell you a couple of things that we have tried at Aramco. There is a team that looked into understanding chemicals, but those chemicals need to be quite simple. I think that the chemical use cases still require the hardware to mature considerably. When it comes to things outside of chemistry, one of those very fun projects that we have done together with a University and our colleagues in Paris, we looked at how to address a load-balancing problem when you have a lot of renewable energy in the system. So you’re thinking about, I would like to build a lot of infrastructure for the next decades as we’re transitioning into a greener economy. That means that we’re going to have a lot of intermittent energy sources. So for those, we need to think about what do we do with storage. That requires a fair amount of modeling, but also thinking about how much of the storage you want to be prepared to build, optimally so, such that you can address the intermittency of renewables. So that’s a project that we ran for almost a year now and with some really interesting results. Scale is still an issue but I think this is one of the big general directions that we’re going to continue exploring.

Yuval: You mentioned supercomputers and the large investment that energy companies have made and are making in supercomputers. Do you see quantum being, do you see primarily algorithms being hybrid algorithms where classical and quantum work hand in hand? Or do you see them primarily as separate things? Or maybe you don’t care, you just need the results?

Marcin: I think that most pragmatic people will tell you that they really don’t care. They would indeed like to just see the result. But I’m the person whose job it is to make it happen. So from my perspective, it’s always going to be a hybrid approach. There are two reasons for this. Or maybe two sides of the same coin, even. On one side, we’re just not there yet. So what can we do with the quantum computers that we have available right now? We can solve increasingly relevant optimization problems. And we’re just about to start being able to compete with classical computers. But to be able to solve a relevant problem of that size, we need to break it down into digestible chunks. This means that we’re going to continuously alternate between a classical and a quantum computer as we’re trying to solve a bigger problem. On the other side of the same coin is, you know, you give me a better quantum computer with more qubits that can simultaneously study a much higher number of combinations at the same time, I’m going to come up with a bigger problem and I’m still going to need to alternate between a classical and a quantum computer. I guess the question at some point is going to be, will the classical computers be the ones that will kind of start not being able to keep up with the quantum machines in the coming years? I think this will be interesting to see.

Yuval: Do energy companies care about energy consumption? I mean, this is obviously a little bit of a tongue-in-cheek question, but one of the things about supercomputers is that they take megawatts and megawatts of energy. And would you use a quantum computer if it produced the same results except that it needed much lower energy? Is that a consideration for an energy company?

Marcin: Oh, I think I definitely would. I think saving energy, there are two reasons. Typically, Supercomputers not only consume a lot of energy, but they also consume a lot of conventional energy because you’d like to make sure that you have a reliable stream of electricity powering your data center. And indeed they can consume at the orders of megawatts, tens of megawatts sometimes. So that’s a consumption of a small town. So if we can reduce that, then not only we save energy, but we also to reduce our carbon footprint. And that’s very important to any company these days. And energy companies are definitely among those. Now, here is a little bit of a problem, because we’re not going to replace a classical supercomputer with a quantum computer, simply because– well, let’s start with the fact that a quantum computer by itself, in my opinion, I think is a little bit of a misnomer. It really is a quantum processing unit of sorts. it, at least for the time being, will be acting as an accelerator for, for example, solving optimization problems. So it will not be replacing a supercomputer, it will be amending a supercomputer. So maybe we’ll be able to overall reduce power consumption by being smarter about how we solve optimization problems, but that energy consumption will be reduced to trying to solve only some of the computational tasks and not all of them.

What I have seen, and one of the things that we are trying to do is the way we’re trying to be clever about how we perform computation with the goal of reducing both the computational time and, also as a result, energy carbon footprint is that we could try to use quantum computers to help manage this process more efficiently. As a result, this would reduce energy consumption. When it comes to just looking at the computer and just comparing the specs, oh, one is 10 megawatts, the other one is 20 kilowatts, I think that’s not a very good comparison, because it just simply doesn’t work this way. And so that’s for the analog computers. When it comes to the gate-based quantum computers, I don’t think we really established how energy-hungry they will be because the amount of control electronics that will need to go into this. I’m not sitting in this area, so I shouldn’t be the one speaking out on it, but I have seen voices, I think it’s the quantum energy initiative, where people are quite concerned about whether those devices actually will consume less energy.

Yuval: You mentioned some results that you received on, I think the subsurface imaging-related problems. Do you think quantum computers are useful today for the energy industry other than sort of in the exploratory sense? And if not, when would you estimate that they become truly useful?

Marcin: I think, I think last Friday. Last Friday, we managed to run our very first job where we took seismic data and we managed to produce a subsurface image from within our production software environment. It’s still a developer’s setting, but if all goes well by the end of the year or maybe early next year, end users will be able to solve the shallow subsurface anomaly estimation problem using a quantum computer at our company. So that would already be useful. We’ve seen that the results can be better by a few percent. We can find a more optimal solution by about a few percent right now. So we’re still testing and trying to understand under what instances we get better solutions so that we can become more efficient like that. But that’s the first application, so that’s a drop in the bucket. I believe as far as business-relevant use cases, it is an industry first. There were people, of course, we were standing of course and building up on the shoulders of giants. People have done geophysics on a quantum computer before us, so we’ve learned a lot from them. When it comes to other people in our industry, there is interest. I think there are a lot of quantum computing companies that are actually hiring experts from the upstream industry to bridge that gap between a quantum computer and an application. I am personally involved in a workshop, or I’d say a small conference, with 50 to 100 participants. So experts from high-performance computing in the upstream sector coming together. It’s actually happening next week in Switzerland. And for the first time in the history of this workshop, we have an entire session dedicated to quantum computing. So the demand for knowledge about quantum computing I think has reached a threshold where we have a session and who knows maybe in a couple of years we’ll have our own conference. I’m quite confident that this is coming.

Yuval: So first results. That must have been quite a celebration internally, congratulations.

Marcin: Thank you.

Yuval: What kind of help do you need from companies outside the energy industry to achieve your goals?

Marcin: So you mean specifically software or quantum hardware providers?

Yuval: Software, hardware, algorithms, academia, anything that might help you.

Marcin: Yeah, so I think that very much depends on the technical readiness level. I mean, the journey starts by really identifying a use case, and then you prototype and then you develop and then you have to think about integrating with HPC, and then you benchmark throughout, and then you update because probably by the time you’ve gone through this process, there’s a new release on the hardware which can do so much more. And so I think there’s room for everyone. So when it comes to academia, we’re doing quite a lot of projects with students, both masters and then PhD students, where we’re offering internships. That’s– we have professor sabbaticals even in our office. So there are a lot of people out there who are really interested in trying to do practical quantum computing. And that’s something that we are proud to offer at Aramco in Delft. When it comes to service providers, I think they play a very important role in the technical readiness level, say, three to seven areas. And I think what is very important is to have a very good relationship and just work together very closely. I’ve seen too often a service provider interact with a big client, such as a big company. And somebody in the big company would say, “Oh, quantum computing, I don’t understand this very well, “But here is this very big problem that I struggle to solve. “Why don’t you get a crack at that? ” And the interaction sort of stops. The service provider goes away, works on it, and it later comes up back that there wasn’t sufficient alignment early on to make this work. So I think that this continuous engagement is very important.

But I think there’s a little bit of a problem with even that approach. Because when you have a sufficiently difficult problem in the industry, there’s definitely been people that have tried working on it for many, many years, which means that there’s so much knowledge in the company that simply people might not have the time to be able to share with you and continuously guide you through the process as you’re trying to make this quantum. So I think a very viable approach, and that’s an approach that I personally prefer, is to have a tool where a service provider comes and say, hey, look, I am really good at solving that class of problems. And it’s a sufficiently broad class. And then a person such as myself, a researcher in the industry, in collaboration with the service provider, is able to very quickly identify the use case that fits that class, and then with just a little bit of tweaking, things can progress. Seeing how we’re trying to solve NP or NP-hard problems, I think that identifying those classes and how efficiently things scale, is probably the most efficient way to go. hoping to see 10, 20, or 100 people start-up or scale up and work on a very important big use case of a 100,000 people company. I think there is a bit of a mismatch of scales to see this really come to fruition. And so here, what I was talking about, I was mainly focusing on trying to solve optimization problems with analog quantum computers. When it comes to gate-based computers, I think we’re still already from academia, and then as you go up the TRL chain, we need to know a lot more about what quantum software will be able to do from input all the way to output. Not just how we have some computational complexity scaling in the middle bit, but loading a particular bit of data into the quantum computer suddenly has this exponentially scaling complexity component to it, and those things cancel out. I think it’s really good to start communicating openly about the entire production chain, as those QPUs will be accelerators. Or maybe should they become a standalone quantum computer in the future? I think these are the two important things that are quite important to have.

Yuval: I believe you’re in Europe right now. So as we’re recording this, it’s getting close to dinner time. So let me ask you a dinner hypothetical. If you could have dinner with one of the quantum greats, dead or alive, who would that person be?

Marcin: That is a tough question. But I think I may have identified a very interesting, soon-to-be great, soon-to-be quantum great. And that person is not a physicist. I came across a very interesting article a couple of weeks ago. From, I forgot who wrote it, but it was talking about Susanna Glickman. I hope I pronounced it correctly. So she is a recent PhD graduate who wrote a very first PhD on the history of quantum computing. I think that just really speaks to the imagination and it’s a testament of how far we’ve come as a community, right, when we have our history books written about the effort of trying to build a quantum computer. How did it come to be that you have gone from a relatively speculative technology that came from just the idea of how do I factorize numbers with just a bunch of qubits, whatever they are, or whatever they do, all the way to a multi-billion dollar industry and really large governmental support across the board globally? I would really like to have a chance to chat with Susana, especially that I was trying to read her thesis and I saw it’s under embargo for another five years. I’m hoping she will publish this book. I’ll try to be the first one to buy it, but maybe something for you to think about, maybe you could invite her on the podcast first and we can get a preview.

Yuval: That’s an excellent idea. Marcin, thank you so much for joining me today.

Marcin: Thank you for having me, Yuval. It was a great pleasure.