What Is Quantum Annealing?
D-Wave has built five generations of commercial quantum processor, and Murray Thom says the power draw has not moved. Our fifth generation system uses 15 kilowatts, but our first generation system also consumed 15 kilowatts.
In this episode · 13 chapters
I mean, the thing that I have learned is that no amount of computing power is going to overcome a product that’s difficult to use.
Murray Thom started at D-Wave in 2002 and has built quantum computers by hand. He explains quantum annealing through the metallurgy it borrows from, and why D-Wave pairs quantum and classical computers rather than asking either to work alone.
The examples are practical: a grocery group scheduling staff across departments, cargo packed into trucks, a Pokemon hand swapping six cards at once. Thom also explains why annealers avoid the error correction burden of gate-model machines, and why programming one needs no quantum physics.
Murray Thom, Vice President of Quantum Technology Evangelism at D-Wave, argues on the Thinking On Paper podcast that quantum computers create value now by working alongside classical computers on hard scheduling and optimisation problems. He explains how D-Wave's quantum annealing works and how Pattison Food Group uses its hybrid technology to schedule store staff.
How does quantum annealing differ from gate-model machines?
Quantum annealing uses quantum effects to move between solutions faster, which suits optimisation. Thom contrasts it with gate-model machines, which store information in fragile quantum states and need error correction. D-Wave builds both.
How do D-Wave's hybrid quantum and classical systems work?
Classical and quantum computers search in different ways, and the hybrid uses both. Thom says a classical computer changes one decision at a time, while a quantum computer can swap many at once; the classical side calls on the quantum side when it needs a different starting point.
How does Pattison Food Group use D-Wave's technology?
Quantum scheduling is already in production at a Canadian grocer. Thom says Pattison Food Group uses D-Wave's hybrid technology for in-store staff scheduling and expects 50,000 hours of savings a year, which frees senior staff to mentor junior colleagues and work with customers.
How much power does a D-Wave quantum computer use?
D-Wave's power draw has stayed flat across five generations. Thom says its system uses about 15 kilowatts, the same as the first generation, because the processor chip consumes less than a millionth of a watt. He compares this with roughly 100 kilowatts for a rack filled with CPUs or GPUs.
Why do classical computers struggle with yes-or-no decisions?
Yes-or-no decisions are where classical computers struggle. Thom explains that a football player is either on the team or not, so there is no slope to follow, and that D-Wave's machines take problems in a binary form called QUBO, which he says has the same structure as a Hopfield network.
Where Mark and Thom disagree
Mark presses on cost, citing figures of 10 to 15 million he says he has read, and asks what a large company would pay. Thom answers that it is “not at all $20 million to get started”, and that anyone in the countries D-Wave serves can sign up for trial access straight away.
Links from this episode
Questions
What is quantum annealing?
Quantum annealing uses quantum effects to move between solutions faster, which suits optimisation. Thom contrasts it with gate-model machines, which store information in fragile quantum states and need error correction.
Who is Murray Thom?
Murray Thom, Vice President of Quantum Technology Evangelism at D-Wave, argues on the Thinking On Paper podcast that quantum computers create value now by working alongside classical computers on hard scheduling and optimisation problems. Murray Thom started at D-Wave in 2002 and has built quantum computers by hand.
What problems can D-Wave's quantum computers solve today?
Murray Thom, Vice President of Quantum Technology Evangelism at D-Wave, argues on the Thinking On Paper podcast that quantum computers create value now by working alongside classical computers on hard scheduling and optimisation problems. The examples are practical: a grocery group scheduling staff across departments, cargo packed into trucks, a Pokemon hand swapping six cards at once.
Do you need quantum physics to program a quantum annealer?
Thom also explains why annealers avoid the error correction burden of gate-model machines, and why programming one needs no quantum physics.
Transcript
Jeremy Gilbertson Disruptors and curious minds, welcome to another episode of Thinking on Paper. My name's Jeremy. This is Mark. We get to talk to the people that are building tomorrow rooted in today. We've got a great guest to talk more about quantum. We're quantum curious. We're actually quantum aware. We're not quite quantum savvy yet. Hopefully, we'll continue to progress down the timeline and see where we get. But this particular guest has been in the quantum world for, I think he said 22 years in our pre-production chat. He's been with the company for a really long time. I'm stoked to dig in deeper. Mark, how are you feeling this morning? Yeah, also quantum fan boys and girls as well in our community. Total fan boys. Total fan boys.
Mark Fielding And of this company of D-Wave Quantum. Yeah, we're talking about the real world applications of quantum, drug discovery, optimization, material science, scheduling, fault detection, traffic congestion, and something I learned. I want to get into Liverpool Football Club and the Fantasy Football League, which I'll leave that as a little teaser for later, something I found.
Mark Fielding But yeah, our guest today is Murray Thom. He is Vice President of Quantum Technology Evangelism at D-Wave, at Quantum D-Wave. So, welcome to the show, Murray.
Murray Thom Hey, good to be there. Thank you for being here. Yeah, thanks for being here.
The carryover question
Jeremy Gilbertson Do we want to start with the carryover question, Mark? Do you want to start there? And then I've got a tee-off question that might be interesting for us to dig into some background as well.
Mark Fielding I will do that. So, for all our new listeners, we like to thread our shows together. So, we have a guest leave a question for the next guest and we start off from there. So, last week's guest, Yir Kleeper from Blockchain, he left the question for Murray, which is about incentives and how quantum technology companies can use incentives to foster collaboration to push forward the real utility of quantum over the coming years. Okay.
Murray Thom So, yeah, this is a really interesting point because I'm working with customers to help them build value with a new disruptive technology. And coming from a physics background myself, I've actually got an electrical engineering degree and an honors in physics and I'll say a bit of math in my background. I often think about the world in terms of physical processes, right? So, when we're trying to help our customers solve problems, we kind of want to set them up in a circumstance where water is flowing downhill, as it were. I don't want to be trying to push water uphill. I want the water to flow downhill. And so, a big innovation that we had in the development of our systems and a realization that we had was that no quantum computer was ever going to be used in an application on its own. And I mean that in a...
Murray Thom Yes, we have to use computers to get data into quantum computers and back I mean more that the tasks of applications were going to be shared between quantum and classical computers because they each have their strengths. They had kind of complementary strengths. And the brilliance about building a quantum classical hybrid technology for supporting customers and applications is that it provides a number of benefits related to these kinds of incentives. And it really focuses on aligning our incentives as a full stack quantum computing provider with their incentives in terms of trying to get quality of solution business value. So, for instance, they're working with us because they've got challenging problems and need workforce scheduling, maybe logistics routing or production scheduling.
Murray Thom And they're trying to get to better quality solutions that allow them to be more productive, to be more effective in their operations and more competitive. And we are trying to develop the technology in the way that provides that maximum positive impact for their organizations. When we're building a quantum hybrid technology, the benefit is that for us, the quantum computers that we develop are like the least expensive resource that we can push the work onto. So, D-Wave has got a whole program here of like several teams working together to maximize the impact of quantum computing, to impact those customer applications that we're working on. And we actually direct the development and design of the quantum computers in that direction, because we're working with customers.
Murray Thom So, our incentive is to develop them so they have greater impact and they develop as quickly as possible in that direction. And we're constantly trying to push work onto the quantum computers. And that's exactly what the customers are looking for. They're solving a problem. Ideally, they want the greatest impact from technology today and in the future. And that sort of aligns our incentives as we're trying to maximize how quantum computing is providing a role there and providing an impact. And they're looking to leverage that impact to bring value to their end customers. So, I think that in as much as companies are systems of people and technology working together, right? And towards common goals that we as individuals can't do. And as much as we can help those organizations and those organizations working together to have their incentives aligned, you can get very powerful results that emerge out of that. Yeah.
Jeremy Gilbertson And one thing too that is a theme, there's a lot of themes. We've been doing up over 100 shows, I think, in this podcast. And the themes that keep emerging are technologies that come out. A lot of people, when technologies come out, they have to be rooted in something we do today. So, there's a bridge that people can go, oh, that's kind of familiar. I can kind of do this and do that. But there's a step you have to take. And that step has to be... It can't be super complicated. Blockchain's dealing with this right now where they're like, oh, we're going to onboard a billion people into a very complicated process that does something kind of similar to what databases do. And people are like, why am I doing that? So, the incentive piece is really interesting. And I love the be like water analogy. That's super cool as well. Mike, go ahead, Mark.
Mark Fielding Did you have a... Well, no, I was just going to comment on your allusion to blockchain and the problem that blockchain has with the terminology and the narrative and explaining the benefits. And everyone's obsessed with the technology. And in fact, if you're not a developer, you don't understand. So, you don't understand 99% of these websites. And I have to admit that I thought that that was as bad as it gets. But quantum takes it to another level of not being able to understand. If you focus on the technology and all of your websites explain the technology, and you see this on LinkedIn and you see it with the people who... A lot of people who speak about quantum. It's intimidating. It instills fear. It can't... I can't go near this because it's too complex.
Mark Fielding And yeah, that's just my observation for what you're saying about blockchain. Well, let's... Murray, let's go here. So, a lot of people are like quantum, brand new thing that's happening. Quantum computing, like, oh, I'm hearing a lot about it. It must only be a couple of years old. I think you told me earlier, told us earlier, you've been at D-Wave working in quantum computing for like 22 years, right? Is that it? All right. I want to ask you one question. It's kind of a two-part. Tell me what you... Give me some details about your first day working at D-Wave, and then give me like two or three milestones that you hit throughout that or D-Wave hit throughout that 22-year period that were like massive aha moments in the development of quantum. I might zoom in on you too, because people are tired of looking at us. Okay.
First day at D-Wave
Murray Thom So my first, I mean, definitely my first day at D-Wave was September 16th, 2002. And it was just awesome to have a job in quantum computing. I was so excited. I mean, I'm glad you asked me that question because I haven't had a chance to reflect on that for a while. I mean, you've kind of like floating through the office and you're like, yeah, I'm working at a company that's building quantum computers. And I think there was... When you're in school and you start learning about quantum physics, it's just kind of like mind blowing. It's a little bit like, what does Morpheus say in the Matrix? He's like, you look a little bit like Alice tumbling down the rabbit hole, accepting what you see because that's just in front of you kind of thing.
Murray Thom So a lot of quantum mechanics are people who are basically like asserting like, this is the way that the world works. And you're just kind of like scratching your head being like, is that for real? And what I love about the idea of quantum computing is that it's about, hey, let's take what we know about quantum physics and the description of how the world works and let's turn that into useful work. If we can make an engine out of quantum physics that can do something for us, well, that's like a new opportunity to understand it in a way that we didn't understand it before. And that's actually very much the way that it's materialized. We understand the machines we're building and what their core capabilities are a lot better today than when we started.
Murray Thom And it's given me an opportunity personally, and I think a lot of people who now have access to our quantum computers around the world to start interacting with quantum physics. Part of the reason why it's so unfamiliar for us is because we don't get to interact with it in our daily lives. I mean, at a quantum level, I mean, all classical systems are quantum, but they behave classically. And so now that we have quantum computers in our leap cloud network, in several places around the world, available to 42 countries around the world, there's all sorts of folks who can now say, hey, I'm going to create a puzzle for it. I'm going to try to solve a problem with it. I'm going to see how it reacts and how it's different than classical computers and how I can get better solutions out of it.
Murray Thom So that's fascinating. And in terms of the milestones in the last 22 years, it certainly has kind of felt like it's actually been like working at several different companies. In the beginning phases, I was working on building quantum computers themselves, designing for them. I built several different subcomponents, some of the early generations of them, and I've built a few quantum computers by hand. So I think that there was this incredible validation moment when we had sold our first system to Lockheed Martin and they installed that system at the Quantum Computing Center at the University of Southern California and the Information Sciences Institute.
Murray Thom That was really a point where the rubber hit the road in terms of talking with customers and learners who are trying to understand how to program the systems and teaching them everything we could about quantum computing and working with them to try to bring them value. So in a technology company, that's always a major moment. And when I'm advising startups who are doing deep tech, I often try to talk them through how to get through that early technology development to get to that customer integration point. And then I think that when we progressed to the point where we put our quantum computers on the cloud, it was this opportunity where it was the first real-time cloud access to quantum computers. So people could go to our leap quantum cloud platform, they could begin programming it, and they would get an answer back in seconds.
Murray Thom So now it's becoming much more of a learning tool, a development tool where you can start putting ideas in, getting answers out. And that's really, I think, where engagement and innovation gets accelerated. So yeah, I think those are two key phases in the development of the systems in my career here. I'm glad you still remember the first day and the date. So it honestly made a big impression on you. Yeah. Yeah. The rabbit hole, Alice in Wonderland thing is that, that's super cool.
There is not one way to do quantum
Jeremy Gilbertson So quantum computing, there isn't just one way to do quantum computing, right? And based on what I understand about D-Wave, quantum annealing is kind of one of the main philosophies of how you get a couple of bits to entangle and make them do computing things, right? So annealing also comes from kind of metallurgy, right? When you heat up a metal and you heat it up to a point to increase its ductility, right? Ability to be used and flexible in different ways. So talk us through what quantum annealing is. Maybe try to relate it to the metallurgy stuff so we can extend that knowledge a little bit wider.
Murray Thom Yeah. Yeah, absolutely. Okay. So, and I'm also, you mentioned earlier about talking with experts in quantum computing. I mean, as a field and as an industry, I'd have to say that we tend to make it harder than it needs to be. There's a lot of explanations of quantum physics, which are kind of like, it took folks a lot of years to build up familiarity with mathematics to be able to describe them. And sometimes it's, you know, you got to go back to the beginning of that journey and be like, well, what tangible things would I relate this to to make it more understandable? So I'm going to try to do that by saying that, you know, a quantum computer is just a device that uses quantum effects to accelerate calculations. That's all a quantum computer is. I don't even need to introduce superposition into that.
Murray Thom It's just trying to use quantum effects as a resource. And that's kind of an important idea to think about. Like we've got a computer and they use, you know, chips as a resource to process information for us. So what kind of a resource does quantum mechanics represent? Well, it turns out there's different ways to use quantum effects as a resource in a computer and the different ways that you choose basically determine the model that you're building. So in one method, in quantum annealing, which we're talking about right now, you're using quantum effects to allow the computer to move between solutions more quickly, right? So it accelerates a transition between solutions. So if I'm doing the optimization of a schedule for a grocery delivery fleet, I've got a schedule right now.
Quantum annealing
Murray Thom I can just start with a random schedule. You know, that's my current state, but I want to move to a better schedule from that. And a quantum effects can help me move to a better solution more quickly. Okay. So that's, that's pretty understandable. And you know, the other, just to contrast that with the other model, with gate model quantum computing, because D-Wave is building both models. In that model, you're using the quantum effects to allow the computer to store more information. So you're putting more information that quantum states are actually containing information for you. The trick there is that quantum states are famous for being delicate and collapsing. And when the quantum states collapse, then information you put there is lost.
Murray Thom So if you're going to use the quantum states to store more information, then you have to have error correction to kind of keep maintaining that information there to allow you to process over it. And depending on which choice you make, that actually really strongly determines which applications you're going to be targeting. If you're moving between solutions more quickly, that's very useful for optimization problems and applications in artificial intelligence. If you are storing more information, that's very helpful for like quantum chemistry, for simulating electrons and molecules, or for like fluid flow calculations. So that gives you a quick taxonomy in relatively plain language about quantum computing. And at D-Wave, we're famous for building quantum and E-Link systems because we started so early.
Murray Thom We got there ahead of everyone else. And the idea came out of the Institute of Science Tokyo, formerly Tokyo Tech, and a couple of really, really brilliant professors there, Kadowaki-san and Professor Nishimori. So essentially, let's start connecting it to that metallurgical phenomenon that you were talking about. In metallurgy, we discovered metals, but to change their shape was very difficult to do unless we heated them up in a fire. So when you heat them up, just as you were mentioning, it becomes easier to change their shape by hammering them out. And then what happens is that when you cool it back down again, it actually, the metal takes on different properties depending on the rate with which you cool it.
Quenching
Murray Thom So if you quench it in like water or in like oil really quickly, and you just drop its temperature really quickly, then what happens is that its internal structure doesn't change. Everything kind of freezes in place. And so it goes from this soft state into a crystalline state. And the crystalline state, the crystals just occur and then basically the boundary walls hit each other and they're very internally frustrated. And that frustration makes the metal really hard. And that's useful if you want to make like a knife edge or a sword. Now consider if we cool it down really, really slowly, the crystals start emerging, but they can actually, their boundary walls can actually move because they're just on the edge between that sort of like forming and not forming those crystals.
Murray Thom And so what happens is that although you've got these crystals, boundary walls moving, the boundary walls can actually move through. And if they move all the way to the other side of a crystal, it actually absorbs the other like crystal site. And so you can, these crystals effectively grow and become longer and longer and longer than material. And that means that the internal structure of the material becomes less frustrated and it's softer. So that's great if you want to use it for wire to conduct electricity or for some sort of metal where you want to conduct thermal energy, transmission through the metal is actually a lot easier. And it's also a lot softer if you want to draw it into wire. So fundamental answer is the rate with which you cool those metals changes their properties and how you can use them.
Murray Thom So in quantum annealing, it's just recognizing that thermal, that the heat represents thermal energy and the thermal energy introduces randomness. And then you just reduce that randomness and then you can kind of like settle into an unfrustrated state. Well, quantum effects, when you're solving an optimization problem can be thought of similarly. They introduce randomness in the state. It starts exploring a whole bunch of different states. And as you gradually turn that back down and it starts to localize, if you do that slowly enough, not only can you get to a less frustrated state, you can do faster than any classical system can. So the speed limit for quantum effects to reduce frustration are, the speed limit is faster and it's significantly faster than classical systems.
Murray Thom So that was the idea that came out of Japan about building a system that could implement that. That idea was then tested in crystals in the United States. This was in like 1998, 1999. And then in 2005, we started, we were hitting on these ideas. We realized that if we built a quantum annealing system, it would give us three huge advantages. One was it had practical applications in the near term in optimization and business, which is incredibly important. The second is you don't need to know quantum physics to program it. You can actually program it on Python using classical ideas. So programming accessibility is much higher. And the third is that they're much easier to build.
Murray Thom Primarily, I mean, for a number of engineering reasons, but also because of the, you don't need error correction in the same way because you're not putting information into the quantum states. You're letting the quantum states move you to better quality solutions. And I mean, I'm just asserting that, but you can actually see that it's true because the largest, most highly connected quantum computers on the planet are quantum annealing systems. We've got 5,000 qubit processors online at a commercial scale right now with 15 way interconnectivity, which means every qubit is connected to 15 other neighbors. That's by far and away, the largest, most highly connected systems.
Three in the field
Murray Thom And we've got three of them in the field and here in Burnaby, Canada, in Los Angeles, California, and in Julep, Germany as part of our worldwide quantum cloud network called Leap. So I got an image on the picture. Yeah, yeah, yeah. That's one of our systems here. And there, I mean, it looks small relative to my hand, but it's actually like taller than I am. If you're in the US, it's like 10 feet by 10 feet by 10 feet. If you're in Europe, three meters by three meters by three meters would be the dimensions. Every episode, Mark's going to giggle.
Jeremy Gilbertson I always throw my data center background out there. So I worked in kind of data center capacity planning and more on the infrastructure side. So that to me looks like a hot, cold aisle containment cabinet system within like a traditional data center. It looks pretty similar to that.
Murray Thom Well, and we did the industrial design that way precisely so that it's easier for folks to integrate. It's much more familiar. And in fact, I mean, I've talked to folks who manage data centers. And when we tell them about the like water and power requirements for the system, it's like, oh man, this is going to be so easy to implement. It just takes like tap water basically to cycle through the system. And then the power requirements are like 15 kilowatts. So like one of those racks here that's in front of that system, sorry, for those who are listening in, you can't see this, but like a 19 inch rack that you would see in a data center, if you completely filled it with like CPUs or GPUs, it would consume like 100 kilowatts. Whereas our whole system consumes about 15 kilowatts, about a sixth of that.
Murray Thom And what's fascinating about that is that this is our, actually, this is a photo of an earlier system, but our fifth generation, we have five generations of commercial processors, and we're working on our sixth right now that we have a prototype for. Our fifth generation system uses 15 kilowatts, but our first generation system also consumed 15 kilowatts. And that's not the case for like CPUs and GPUs, like the GPUs are basically growing in their power consumption each year. It's like 30 GPUs is the same power consumption as our whole computer. The reason the power is flat is because the processing is done in a chip about the size of our thumbnail inside that computer room, and the chip consumes less than a millionth of a watt.
Murray Thom So it's the metals that circulate the current have no resistivity, so it almost dissipates no heat. So what that means is that as we're creating these new generations of systems, we have this really large growth in the performance capability of the new processors, but the power consumption profile is flat. And what's fascinating about that is that as we look towards quantum artificial intelligence, if we can take some of those workloads that are difficult and challenging for high performance computing systems, and then move them onto quantum computers, we can fundamentally change the power consumption profile. And that's a really big concern as people are looking to how to leverage artificial intelligence today. Oh, it's massive now.
Municipalities
Jeremy Gilbertson Mark, just to chime on this real quickly, there are municipalities that are heavy data center municipalities around the world, like Ashburn, Virginia is one of the hottest data center markets. And they ran into a power constraint issue where they couldn't build power infrastructure fast enough to accommodate the compute infrastructure that wanted to be there. And this could be really interesting as we... I mean, it's going to be a long time, I think, before quantum will be above classical, just in general, they'll have to be both for a while. But down the road, we could see it maybe even being a solve to the power issue in some form. Yeah.
Murray Thom I mean, I think that... Well, I mean, what we know now is that... Let's say, let's use an optimization example. And let me put it in terms that are very, very plain language. Let's say my son, for a long time, was really into Pokemon. So, we'd be playing Pokemon card games with one another. And that's a game where you're playing a field of players with strategy. And you've got a hand of characters that you're trying to optimize. So, a classical computer is very good at saying, hey, this hand's not optimal. I'm going to take a card out and pull another one from the deck, and then see if I can improve it. And then I'm going to take another one, and maybe it'll accept or reject that move one at a time.
Murray Thom So, one at a time, card changes are very, very easy for classical computing, and it can do that blazingly fast. But the trick is that if you have a lot of cards in your hand, and you're looking for a combination of cards that interact well together, that's a very difficult way to improve your hand. And a quantum computer is very good at saying, hey, I can actually take six cards out and replace them with these other six. And when I do that, they'll be interacting with nine other cards much better in my hand. So, it's doing these non-local updates. And it's able to do that because it's using the superposition across states to see how to make sacrifices and gains in a much broader way. It's really interesting. There was a book a while ago, I can't remember who wrote it. It was called The Adjacent Possible.
Murray Thom And it was all of these things on the fringe of one thing that could be combined in unique ways to make different things. So, say we got three boxes, three rectangles in this video, right? So, say there are three pieces in my box, three pieces in your box, three pieces in Mark's box, where a traditional computer will look at the boxes as a whole and go, we're going to replace Jeremy because he's not optimized with these other guys. But quantum can look at the little pieces within the boxes and go, well, these two bits work with his bit grade. It's like the color palette is seeing more possibilities in the color palette of something. Is that even close? Well, let's put this in the context. So, we have a customer here in Western Canada, Pattison Food Group, and they own 13 retail grocery brands.
Scheduling
Murray Thom And one of the applications that they're working with is in-store grocery scheduling. So, they've got a store and working at, let's say, the deli department. And they're trying to decide, is Alice going to work in the deli department or is Bob going to work in the deli department? And let's say they've only got one more space left, right? So, they've got decisions to make. Is it going to be Alice or is it going to be Bob? And the decisions are related to one another. So, they know that they only want one or the other to kind of go in. So, if they decide that Alice is doing it, then Bob has to be off, not assigned to that task. And so, when you're putting a problem into a quantum hybrid system or directly into the quantum processor, you're telling it those relationships.
Murray Thom So, it knows only one decision should be on, the other one is off kind of thing. And just on its own, that's relatively easy to work out. But what's happening is that that assumes that everyone else has been scheduled in the deli department and the rest of the store. And the question is, and you've got all these decisions all over the store. And so, you've got this question of like, well, what if I pulled three people out of the deli department? Would I be able to cover the floral department of the pharmacy better? In which case, I want both Alice and Bob to be in there. So, it might make a sacrifice in between those two decisions by making a much bigger change elsewhere. And if you look at the combination of possible decisions you could be making there, it's enormous.
Murray Thom So, let's say, amongst all those employees in all those different departments in the store, if you've got 300 decisions to make in assignment, well, the number of possible combinations is larger than the estimated number of particles in the universe. So again, if you're taking someone off a shift and then looking through and putting somebody else on a shift, doing that local update that a classical computer can do, even if you can do billions of those a second, that's nothing compared to the estimated number of particles in the universe. You're making a small impact searching through that space. And the quantum computer, it does this much better job of making these much more non-local changes.
Murray Thom So, when they're working together, what's happening is that the classical computer is taking in a very large industrial problem, like one that Pattison Food Group is working with. It starts doing those local updates and quickly looking for better quality solutions. It's basically running downhill. And then it makes a call to the quantum computer on the customer's behalf and says, based on where I am, can you make some substitutions that move me in the space to an equal or better quality solution, but that's significantly different. And then it can start running downhill in both of those places. And then if one of them gets stuck, it can continue with the one that's better. So, it's the searching combinations. It's actually not a big system itself. It's that it's all of the interactions and the decision space that it's searching for you.
Murray Thom It can deal with the complexity of options. It would consider, let's say, what were your two names? The two names of the people? Alice Bergman. All right. So, let's say Alice has been known, she's cut her hand on the deli Slice her four times in the last three months. We probably don't want her slinging sandwiches right during the day, but she's really good at detail and stocking in the back. And Bob is really forget. Are those some of the things like the, it's just the expansion of possibilities to consider. Is that why, is that why this system works so well? Yeah, it's precisely because as you're navigating through the space, there are these conflicting constraints. So, so for instance, in the real problem, not all the employees are trained to work in all the departments.
Murray Thom And then they know that the forecasting for like varies each day of the week, like the number of employees they want to have based on demand, and that varies each week of the year. So, so there's that aspect to it. And then, and then you've got like shift considerations. So you've got to schedule employees according to shifts by seniority. And if they're working a night shift, they can't work a morning shift the next day. So as you start gathering all of these things together, you know, you're creating a problem, which is very difficult for humans to solve. It's also very difficult for classical computers to solve. And so, you know, our quantum hyper technology was able to bring a solution to, to Pattison Food Group that's saving them a significant amount of time.
Murray Thom So the folks building those schedules are actually saving many, many hours. And, you know, they're in their first phase of deploying that in their stores. I think they're up to about 60 or 90 stores right now. Their workforce is accepting, expecting 50,000 hours of savings annually. And think about that in terms of its impact on people and processes. These are the most senior employees in those locations. So now they're able to get out from behind their desks or grinding through spreadsheets and actually work with the junior employees. And so they're mentoring them. They're teaching them how to, you know, teaching them new, new skills and about the business. They're working directly with the customers. Now, their most experienced employees are actually out interacting with the customers more.
Murray Thom So that's a huge benefit. And then, you know, also, you know, people are on a career journey. And if somebody moves on from that post that used to have a very negative impact on everyone else's schedule. But now it's much easier for a new employee to come in and learn that scheduling task and make sure that everyone's got a high quality schedule all the time. So it's just fantastic working with customers, seeing, you know, quantum computers used in production applications and then doing all of the engineering around that to make that easy. I mean, we talked about programming the systems, you know, and the complexity of blockchain you were talking about earlier. I mean, the thing that I have learned is that no amount of computing power is going to overcome a product that's difficult to use.
It has to be easy to use
Murray Thom You know, it has to be easy to use. We've got to take complexity out of businesses with quantum computing. That's what we're doing. We're not trying to introduce complexity. So we've got to solve the technical problems and and then help our customers solve their problems so they can get focused on what their businesses need.
Mark Fielding There was a maybe it's the same solution, but one of the very interesting D-Wave for me personally, reports I read was about using D-Wave to pick the optimal Liverpool Football Club in the 2021-2022 season for various formations. And there's all the calculations of how it works out. And then it gives you the optimal team. And obviously, that's a very kind of niche idea, but you think it's like a new money ball. You could use quantum computers to make the optimal fantasy football team, the optimal club teams in any sport. And it was the same processes, I guess, that you're speaking about. But moving on from the grocery company, is it the same solutions that you're using for shipping or for portfolio optimisation? Is it the same? Is it the same ideas solving those challenges as well?
Murray Thom Yeah, it's it's a it's the same quantum hyper technology. And if you're doing you know, in fact, there's actually a very close connection between picking players for a sports team and picking assets for a financial portfolio for portfolio optimisation. Those formulations are actually very close. And so what we do is that our customers are talking with us about this new emerging technology. And the focus here is that, you know, at the end of the day, we build solutions that help them with complex industrial challenges like workforce scheduling, like routing logistics and final mile delivery and supply chains, like production scheduling. Ford Otosan, for instance, is working with us on optimising schedules for the manufacturing of a thousand cars.
Murray Thom And they've managed to bring that down from like previously taking them 30 minutes or more down to five minutes, for instance. Each time we work with a customer on those applications, we'll work with their subject matter experts and they will say, OK, here's the description of the problem. Here's what I'm trying to optimise and here's the constraints I'm trying to meet. I've got a series of decisions and I need to orchestrate those resources, that network of resources so they can operate as effectively as possible. And ideally, in a way that, you know, is agile and reacts quickly to change. And then we will create a formulation, a description of the problem that is a computer programme that they can then run in their application on their systems.
Murray Thom And once it, you know, it takes the model of the business task they're trying to do and it takes their data about, you know, the list of available employees or the list of available manufacturing machines and combines those together. That is basically like a description of all the decisions that they need to make, not actually their business data. It's just like, here's a series of decisions I need to have made. And then they pass it to our system, our quantum computers and then and our quantum hybrid systems. And then they'll start working over the decision space. So they they see all problems as a series of decisions that are happening or not happening. Or like I have one set of decisions.
Murray Thom And, you know, here's where, you know, quantum and classical become very complementary, is that, you know, if you're optimising a financial portfolio, you can sort of be like, well, I'll buy 17 units or I'll buy a million units. Like you kind of scale it, I'll even buy a fraction of a unit. But when you're doing this, you know, the team, the players in the Liverpool football club, you can't get a 17th of a player, right? The players that they're on the team or they're not on the team. And that feature of things that are happening or not happening and the choice of which player you pick kind of affects the rest of the team. That's very difficult for classical computers. All classical systems struggle with that.
Murray Thom And it it's partly because if they if they could continuously vary it, if they could just take a scoop of something like a 17th of a player, they could start calculating gradients like slopes that would help guide them towards, oh, you know, a little bit of this player is good. Maybe I should get a little bit more of that player. That's what they might do in finance, for instance. And once you once you don't have that, the players there or they're not there, there is no slope to calculate. Then a key strategy is lost. And then because it affects the rest of the network that creates this very large interconnected decision space that they have to search through.
Murray Thom And so there's kind of two main strategies to tackle there, which is either, you know, try to reduce the number of decisions you need to make or study the problem very intensely to detect structure so that you can make big moves in the space. But no classical systems focus on moving through the space more quickly. And that's because they're all running on the same hardware, right? It's always transistors at the bottom, whether you're using a CPU or a GP or something like that. And quantum systems have this like proven ability to move through that solution space more quickly. And that's what I think really exciting. So, you know, working with customers to discover those applications, build a proof of concept that demonstrates that business value and then work with them for production, deployment and release.
Moving through the solution space
Jeremy Gilbertson I like that moving through the solution space quicker. That kind of kind of gets me I see this giant map of interconnections and like how to kind of make sense of all those possibilities quickly is really interesting. Can I ask you one thing related? I'm always really interested in and I want to I want to try to make this simple. Right. So there's a handoff between classical and quantum. You got to define the problem, right? Spend a lot of time figuring out what you're trying to solve. But is that transition point is how we map traditional classical to quantum? Is it involved in this like binary quadratic model? Is that like the encapsulation of the problem that you can package and translate to quantum? Is that how that works? Yeah.
Murray Thom So so the way to think about this is that when you're working on an industrial problem, let's say you're packing cargo into a shipping truck or something like that, you've got you've got some binary decisions to make, which is that am I going to put this box into the Alice's truck or am I going to put this box into Bob's truck? Those are binary decisions. And then you've got some category decisions, which is that I need to orient the box, so I need to rotate it, but it can only rotate a 90 degree increments. So there's that kind of aspect of how it has to move around. And then you've got continuous decisions, which is like I need to slide along the bottom until it meets the next box. So those are the elements of the industrial version of the problem.
Murray Thom And classical computers are very good at moving through the continuous variables. And our quantum computers, the machine instruction that they accept is a binary quadratic optimization form. It's called Qubo. It's actually got different names in different fields. The Nobel Prize for physics this year went to John Hopfield and Geoffrey Hinton. And John Hopfield proposed a neural network which has that same structure. It's called a Hopfield network. So Hopfield network and Qubo are the same. So when you're programming the quantum computer, you're just giving it an expression about binary variables that are related to one another. And it's working out how to set those. But your industrial problem has like a whole mixture of different things.
Murray Thom So what happens is that you're taking in the problem, the classical computer is using the strategies that it does best. And then it constructs a Qubo, a binary optimization problem for the quantum computer that helps it. So it's not like you're actually fitting the problem into the quantum computer. You're saying, I'm searching like the algorithm is saying, I'm searching through this space. You know, if you can help me, you know, configure these decisions, that will help me move through the space more effectively. And then the quantum computer says, OK, well, why don't you try, you know, rotating your Rubik's cube in this kind of a way? And then it's like, oh, hey, that's helpful. And then, you know, it continues on using its classical strategies.
Jeremy Gilbertson So this is less about like, just real quick, I know I'm commandeering a lot of the airtime, but just following up on that, that's almost like quantum computing helps you ask better questions. Is that like, or helps refine?
Murray Thom Yeah, if you were going to try to use that strategy and instead of calling the quantum computer, you were going to call a classical computer. Some of the queries you'd be asking it, it would struggle with, it would basically give you bad suggestions. And so anybody who was working on that before would have avoided that strategy because it's so difficult, right? They would like they would have would have completely cut them off from trying that out. And a quantum computer makes that possible. So it's kind of it really is the case that a number of folks are thinking about problems differently because they're using our quantum computers. And that's opening up all sorts of strategies that are bringing them quality.
Jeremy Gilbertson Shout out to Julio Tino, The Nexus, augment your thinking spaces. Check out the book club. You can learn more about that. Mark, you got some thoughts. Where are you going with this?
Mark Fielding I just wanted to say, I think we've got an understanding of what D-Wave are offering. So if we can make this a bit more tangible. So for some of our listeners, I don't know. OK, and I might mention some companies here you're already working with. Maybe they already have quantum solutions in place. But I don't know if like Oracle or Salesforce or Samsung or Netflix are watching this. Kind of like what's the D-Wave relationship like? I mean, everybody wants to know, you know, how much does it cost? I've read like 10, 15 million. I don't know. But OK, Salesforce, they have 20 million. They come in, they buy and if you buy, they buy a D-Wave computer. What's the process like? What's the next step? How does it look for them if their CMO is watching this or something?
If the Netflix CMO is watching
Murray Thom Well, if the Netflix CMO is watching this and they have a 20 million dollar budget, absolutely give us a ring. That would be awesome.
Mark Fielding Is that a good budget? Is that a good ballpark figure or is that too little, too much? It is. It's never too much, Mark. But the thing is that like it's not where you need to start. So the way they would experience this is that so first they would want to know like we're a full stack quantum computing provider. So we both we actually build hardware ourselves, but we also have a cloud platform for accessing it. So that's like an application platform that we're building that allows people to build applications that call an API and get high quality solutions. And, you know, if they have a team that are working on those applications or if they're working with our team, we're building those applications with a software development kit, an SDK that allows them to program on the system with the systems in Python.
The SDK
Murray Thom And the SDK is written in Python and C for speed. And then we have a team of folks whose expertise is formulating problems. So if they want to really, you know, we're frequently working with our customers and bringing them millions of dollars of value per year. So they want to get those applications built quickly. And our launch team is really about helping them move through that journey as quickly as possible. It's not at all $20 million to get started. In fact, you know, everyone who's listening to this program in one of those 42 countries can log in and get trial access to our systems right away. And our full scale systems, you know, when you log into our platform, you're going to see that there are demonstrations that you can run with no programming.
Murray Thom We've got open source programming examples that show you how to do bin packing. It shows you how to do workforce scheduling. It shows you how to do like production, production scheduling and logistics routing. There's open source examples. And you can start starting from those and then building off of those into your business space. And then you basically, once you're running those applications, then those systems are just calling our platform from time to time. And, you know, production applications really vary. You know, it's absolutely the case that the domain of providing, you know, cloud solution platforms for optimization, they need to scale and adjust according to the size of the production application.
Murray Thom So the goal is always that the costs of the platform provide a strong return on investment for the application, whether it's small or large. The key thing here is that, you know, the platform is actually doing it's not just software. It's actually doing it's using hardware. It's deploying the machines, both classical and quantum and passing jobs to them. It's managing the scheduling of all of those jobs. The algorithms are optimized for the hardware they're running on, whether it's CPUs or CPUs or the quantum processors themselves. So a lot of that engineering and complexity is basically contained within the service. And because it's a combination of hardware and software, it scales with the application.
Murray Thom So if the application is small, then it's using a small set of resources and they're basically spun up when you're running the application and then released when you're not. And if it's really large scale, then, you know, it basically spins up that many more hardware resources running the software along it. And then, you know, our service, our teams are constantly improving those algorithms. So, in fact, we've engaged with an open source community of physicists at the low layer, mathematicians at the mid layer and software developers at the top layer. So wherever those improvements come in, hardware or software and that technology stack, the folks who are running the applications get a compounding benefit. So, you know, we started the program talking about incentives. All of the incentives of the community in terms of their engagement and what they develop helps those businesses who are running applications on top of that platform.
Jeremy Gilbertson Makes sense. Thanks. Yeah, awesome. This has been a great discussion. I've got some great, you know, new nuggets that I can be swirling around and... Yeah, more dangerous. More dangerous, more dangerous. Let's talk about the carryover question. So, Murray, for our next guest, I think we're... Oh, go ahead, Mark.
Mark Fielding Can I just add like a precursor to that question before you ask the question? Because I think it's maybe the interoperability of quantum systems. And our guest is IBM. I was wondering what the communication is like between yourselves, between Microsoft, IBM, the bigger quantum and the smaller quantum. We had Horizon and I think they're maybe second tier. But what's the communication like? Are you sharing ideas? Are you sharing insight? Or is it very much protective at the moment?
Murray Thom Well, I think there's both aspects at play. So at D- Wave, we're building both bottles of quantum computers. We're building quantum annealers and we're building gate model systems. And so we absolutely don't want to reinvent a whole bunch of proprietary work that's solely around, you know, gate model that other people have already done. So we're absolutely trying to work with, you know, from a software development programming standpoint, systems that build on the research community, what they're doing, the compilers that folks have built, you know, the software community. So that people can get that kind of benefit. And we love to compete, you know, so I don't mind if people run on multiple platforms because ours is the best. So everyone's going to have the best possible experience.
Murray Thom So that's positive. And I think that's very customer friendly. When it comes to our annealing systems versus gate systems, though, I mentioned the point before that, like, you really need to know quantum physics if you're going to program a gate system at a low level. You don't need to know quantum physics to program quantum annealers. So, you know, take like quantum and artificial intelligence. We're seeing this really growing interest from our customers in that space. And they're thinking about, you know, how can quantum computers accelerate that? If you're building a gate model system, then what you're doing is you're trying to train an artificial brain, you know, to sort of mention it, you know, from a conceptual point of view, how to learn.
Murray Thom And you're breaking that down into a bunch of linear algebra and you're trying to accelerate the linear algebra with the gate model system. Right. So that's a way to tackle the problem. With a quantum annealer, the quantum annealer is the brain. So it is like the architecture is a quantum neural network. I mentioned, like, you know, John Hopfield and Geoffrey Hinton, they were describing neural networks and the way they can, you know, each neuron can excite or inhibit its neighbor from it being excited. And the way that they respond statistically, you know, which is which is part of the process of, like, trying to figure out how to wire them up so they can detect patterns. When you program our core machine instructions that you're actually setting the connections between a group of neurons and looking at their response.
Murray Thom So so in that sense, the programming can be very different. And in fact, I mean, that's part of the reason why we've got such strong engagement from the AI community is because that's that's a programming methodology that they're very familiar with. And it makes it easy for them to build in the platform. So it's it's it's a little bit of both. It's cooperation. You know, you want to. You're always trying to put your resources to the most effective use, and if the community, the research community has solved the problem for you, then bring it in, you know, and work with them. And if and if it's your core value proposition, then, you know, for that element, work on it internally and then and then contribute that as a as a service. But it has to provide a lot of value, you know, for it to be a business service. And that's what we're focused on.
Jeremy Gilbertson Makes sense. Makes sense. I think the question or the transfer question is there. Yeah. Yeah. What's the carryover question that you have for our post quantum cryptography expert?
The question Murray would ask
Murray Thom OK, well, I think that the probably the question I would ask is you're going to have different users who are thinking about cybersecurity from a different perspective, and it's the folks who are trying to control surprise who are really trying to watch the space most closely. You know, if they're working in health, then they're working on making sure that the people's protected data is protected for decades into the future. So, you know, what advice do they provide for health care professionals who are trying to control surprise and how they should be thinking about post quantum cryptography and how it relates to them? That's a good one. That's a good looking forward to asking that one. Yeah, awesome. We'll let you know. We'll let you know, Murray, what they say.
Murray Thom Well, you know, I listen to a whole bunch of you guys' podcasts, so I'm going to tune in. I'm going to listen and hear what they say myself. It'll be fun.
Jeremy Gilbertson Oh, outstanding. Not only a guest, but a listener. That's our favorite combination. It's favorite combination. Well, thanks for joining us today. This has been awesome as we continue to step into quantum, as we continue to let our audience learn a little bit more, get quantum dangerous from quantum curious and move down the path to maybe incorporating it into their business down the road. Mark, closing thoughts?
Mark Fielding Join our book club where we are going deeper and deeper into quantum. We're reading Quantum Supremacy by Michio Kaku. Jeremy has mentioned The Nexus by Julio Tino a few times. My favorite book of 2024, Complex Systems. It's fantastic. The convergence of science, art and technology.
Mark Fielding So, yeah, thinking on paper to X, Y, Z. If you like what you listen to, please comment, share with a friend who's also quantum curious. And we'll see you next week for some post quantum cryptography. Yes. Thank you, Jeremy. Be curious, stay disruptive.
Jeremy Gilbertson Keep thinking on paper.




