.svg)
Podcast with Trevor Lanting, D-Wave
My guest today is Trevor Lanting, Director of Science at D-Wave Systems. Trevor and I discuss what quantum annealing CAN and CAN NOT do, how the quantum annealer is maintained, what customers worry about when they deploy quantum solutions and much more.
Listen to additional podcasts here
THE FULL TRANSCRIPT IS BELOW
Yuval: Hello, Trevor, and thanks for joining me today.
Trevor: Hello, Yuval, nice to meet you.
Yuval: So who are you and what do you do?
Trevor: My name is Trevor Lanting, I'm an experimental physicist at D-Wave Systems. I work on our processor development team. So I've been at D-Wave for just over 10 years, and I've been involved with many of the aspects of the development of our quantum annealing processor technology. And now, more recently as we build up a gate model effort, I'm involved with that as well.
Yuval: So would you give me a really quick intro on what quantum annealing is?
Trevor: Yeah. So quantum annealing, in some ways you can think of as the quantum analog to parallel tempering or to basically simulated annealing, which is a heuristic that you can run on a classical computer to solve optimization problems. So quantum annealing is the quantum analog of that. Where instead of, as with simulated annealing, you turn up and down the temperature and explore your solutions space by slowly decreasing your temperature. With quantum annealing, you're turning on quantum mechanical fluctuations. And so you're exploring a large solution space for solving an optimization problem via quantum fluctuations. So you can actually put yourself in a superposition over all possible solutions to the answer, and then slowly turn down tunneling until you localize into what you're hoping is the ground state or a low energy state of an overall system that encodes the answer to a problem that you're trying to solve.
Yuval: So if I go to the D-Wave website, I'm sure it'll tell me about all the problems that can be solved with quantum annealing, and you've got plenty of customers and plenty of use cases, but what kind of problems cannot be solved with quantum annealing?
Trevor: So what we're building with the quantum annealing technology is not a universal quantum computer. It's very as much a special purpose technology that solves optimization problems. So I think a lot of your listeners are probably familiar with an algorithm called Shor's algorithm, which is an algorithm that was developed and shown to be very effective at factoring large numbers. Shor's algorithm is an example of an algorithm that cannot be run on our quantum annealing processors. You can run inverse multiplication problems on a quantum annealing processor, but you can't explicitly run an algorithm called Shor's algorithm. So, in general, any of the gate based algorithms specifically that were developed in terms of like with a gate model system, applying a series of gates to solve a problem, those don't run directly on quantum annealing processors.
Yuval: If I'm a supply chain expert, a logistics expert, for instance, I have optimization problems, obviously.What does it take for me? What do I need to know to be able to express the problem in a way that would work on our quantum annealing machine?
Trevor: So there're different interfaces to our technology. And what we're finding is customers are most often accessing our technology through our hybrid solver service. So there you can pose a problem as a constraint satisfaction problem, which is sort of a common way to express these types of optimization problems. And then, our software stack takes care of translating that into something that can be run on the quantum annealing processor. So you just need to pose a problem with some number of constraints and some interaction between variables and submit them to our hybrid solver service. And that will try to do the translation into something that can be run on the quantum annealing processor. There is sort of a more direct access point, which is that if you can formulate your problem as an ising spin problem, where you have pairwise interactions between variables, then you can express that directly as a machine instruction on the quantum annealing processor, but our current quantum annealing processes have sort of fixed topologies.
Trevor: So there's finite connectivity between each one of the physical qubits in the processor. And what we're finding is that when customers are coming to us, they're coming with problems that have many more connections that can be represented natively on the processor. And so there is a translation step that is required. So I guess there's a short answer to your question, which is to use our hybrid solver service, but you can really dig in and pose those problems at basically the bare metal level if you want, but you need to know a little bit more about the topology of the chip and be a little bit more familiar with the circuitry that's on the chip.
Yuval: When we look at gate based vendors, and we'll talk about D-Wave gate based announcement in a second. But when we look at gate based vendors, sometimes they say, "Oh, now we have 127 qubits", like IBM announced short while ago. "And therefore you can run these type of applications. And when we have 400 qubits, you'll be able to this. And when you have 10,000 qubits, you'll be to do that." Is there sort of a back of the envelope formula that says the existing D-Wave annealing computer can run something that would take X number of gates on a gate based machine?
Trevor: I mean, probably the clearest way to compare is to use an optimization algorithm called QAOA. So this is a quantum approximate optimization algorithm that was developed for this NISQ era, so this near term noisy quantum computing era. The QAOA algorithm is really designed to solve optimization problems on a gate system. And so there you can basically pose problems to quantum annealing solvers and pose those problems to gate based solvers. And so, I mean, we've been very effectively solving problems that are much larger than what can be posed to the current size of the gate based computers.
Trevor: So there is some way of, at least for specifically the optimization space, a way of comparing annealing and gate. But really what we're finding is that they're complimentary technologies in a lot of ways. And that's one of the reasons why D-Wave is excited about our gate efforts. For optimization problems, it really makes much more sense from what we know now to solve those problems with a quantum annealing platform, but there're areas like quantum simulation and quantum chemistry where really gate based systems are... That's where we think a lot of the early applications and attention will be on for those systems. So there are ways to compare them, but they really are complimentary technologies.
Yuval: So for QAOA what is the equivalent? So you can run a certain problem in today's quantum annealing machine. How many gates would I need roughly on a gate based machine to run a problem of the same size?
Trevor: That's a great question. I mean, and it really does depend on sort of the quality of your processor, so the so-called gate depth of your circuits, as well as the individual qubit quality and the coherence times of your qubits. So it really comes down to gate fidelity. So I can't give you a comparison, like a one to one comparison, unless I have a bit more information about sort of where the gate model systems are right now and what this hypothesized gate model system would need to be to compete with us on QAOA.
Yuval: Got it. Now, you've made an announcement that you're working on a gate based computer and that makes a lot of sense. Obviously, some of your customers might come to you and say, "Well, we love your annealing machines, but we've got these set of problems that the annealing machines don't work for." Are there benefits to you working in quantum annealing that would make you especially qualified to develop a really good gate based machine?
Trevor: Yeah. The answer is definitely yes. We are pursuing a superconducting based gate model approach. And so we've had a lot of experience in building up medium and large scale, superconducting processors and superconducting circuits. So in a very real way, what we have is a VLSI superconducting capability, very large scale integration capability. Our current advantage in yielding processors have a million Josephson junctions on them and very sophisticated wiring structures to actually run those processors. So that experience in building up large scale superconducting control circuitry and superconducting circuitry carries over immediately to our gate effort. We will be building our devices in a superconducting stack. And so we have a lot of experience on how to fab, how to test, and how to actually develop this capability. Also, calibration and characterization of these processors is extremely important.
Trevor: So we need to measure and characterize the individual qubits and couplers on our annealing processors so that we can effectively pose problems to the processor and those calibration steps we've had to sort of co-evolve how efficient our calibration and characterization is over time as the process has become more sophisticated. And this is something that will absolutely be a key part of the gate model effort is not only building the circuits, but being able to run them, and control them in a way that scales. And so our experience with our annealing development, we are quick carrying over to our gate development, and both the manufacturing design, and then really the running of these processors and the characterization of these processors.
Yuval: What's your best guess to when and how large would your gate based machine be?
Trevor: That's something that is hard for me to answer. So we know is going to be challenging. We have sort of a roadmap for producing gate circuitry, but we're not giving any dates out externally in terms of the roadmap for when customers can run things on our circuits. We do realize that this is a hard technology to develop and so this is one of the reasons why we're not being public about any dates yet.
Yuval: What do you think, if anything, is holding quantum computing back these days?
Trevor: That's an interesting question. So, I mean, I think there's some fundamental scaling challenges that the field no matter how you're implementing quantum computing needs to contend with. So we need to get to the many hundreds of thousands to millions of physical qubit scale to over the next five to 10 years in order to start running algorithms that really are practically useful. So there is a scaling challenge. I think for a lot of the development we've needed to, we know that we need error correction, and we need to be able to build logical qubits that are much more long lived out of large ensembles of physical qubits, and error correction overheads are large. And so I think there's a big challenge in coming up with techniques for error correcting physical qubits that don't have as large an overhead.
Trevor: And that's one way to help start attacking the scaling problem. So from my perspective, scaling these circuits, or these implementation out to a size where they're really competing with sort of the best supercomputers at solving, say quantum chemistry problems, and coming up with better ways and schemes and strategies for error correcting these physical qubits, those I think are the two big challenges in quantum computing. At least that my perspective is as a hardware person who is building the technology. I do think if you ask someone on the algorithm side, they'll say there is a similar set of challenges, which is coming up with connecting quantum computers to applications. So this is something where we've done a lot of work internally at D-Wave. But I think there's still so much development to be done, to come up with sort of software and algorithms that can run on quantum computers. If you look at where classical computing is, like the software is far outpacing the growth hardware. And I think that there will be a similar development over the next decade in quantum computing.
Yuval: The hardware side of your answer is a little bit idealistic. I mean, on one hand, it's encouraging and the other hand discouraging because you're saying, "Well, if we had a million qubits and they're error corrected, then you could do all these wonderful things." But wouldn't you think that if I have 10,000 qubits and they're not as noisy as today, but I could least measure and characterize and run hybrid algorithms, then there would be business value in running algorithms in quantum computers that cannot be done today on classical?
Trevor: Absolutely. I think there definitely is a near and a medium term value in the circuits that we're building. But again, this is where the second half of my answer comes in, which is that we need more people thinking about those algorithms and those applications. So that these will be co-evolving as the hardware grows and gets better. But before we hit kind of the, say the million qubit mark, we expect that this technology will be very, very useful, but there's got to be a co-evolution of algorithms and software along with the hardware.
Yuval: If I may ask you a business question, I think that you guys are probably unique in the quantum annealing machines. And so when customers come to you, aren't they worried that it's just one supplier and there's no alternative, and if you guys stop being available, then all my algorithms are going to go to waste?
Trevor: I would say, it's not the single source. So the short answer is no. I mean, I'm not as customer focused on the technology side, but from what I'm hearing from our professional services and sales team, we haven't heard that. But what we have heard is customers that really want, as we move applications to production, they want some guarantee of uptime. So they want to make sure that this technology, if they pose a problem and submit a problem with a cloud service, there's a very, very high reliability that that problem will be solved. And so it's not as much the single source of quantum annealing so far, although that could be a concern as we develop more applications and more people start depending on the technology, but really the feedback we're getting is your service is valuable.
Trevor: We want to know how is it going to be guaranteed to be up. In some ways can we treat it like Amazon Web Services where we really are relying more and more on cloud resources to run large parts of our business. So I don't know if that answers your question. I think people are a lot more focused on sort of uptime and the fact that we have solvers and production right now, and sort of have some guarantees, and from that perspective, and less from if D-Wave way is the only supplier right now and that's a concern. That could change, for sure. So to my knowledge, that hasn't been part of the discussions I've had with customers.
Yuval: How long does it take to get an answer? I mean, I know it depends on the complexity of the circuit, but if I submit the circuit and the system is up and it's correctly coded, how long before I can get the response?
Trevor: So that's also a hard question to answer, because, again, it depends on sort of at what level of the software stack and the interface you're accessing the technology. If you want to make a single ising call to our quantum annealing processor so you're calling directly an optimization problem and asking to be solved on the processor with all the, and again, it really depends on kind of network latencies. But if you're anywhere in the world, you have access to our cloud service, it's going to be on the order of several hundred milliseconds and maybe up to just under a second. So again, it really depends on a bunch of things, like how heavy is the usage of the chip, are there a lot of people that are actually submitting problems with the chip, and how good is your internet from, say the location where you're submitting the problem. But sort of rough sort of order magnitude estimates like a few hundred milliseconds.
Yuval: Is there a notion of cycle time in terms of how many iterations you run in a second?
Trevor: Yeah, I mean, we definitely pay attention to the number of problems we solve per second. And again, I mean, usually the way the workflow works is that there will be multiple processes that are calling the chip and asking for answers back. So there would be four or five, many dozens of users that are submitting problems. Even a single user, they could be solving or submitting a series of problems. And typically, what we recommend is if you're running a lot of problems that you batch them, or you can call the chip asynchronously. So in some cases isn't possible, but if there's a way where you can kind of batch call all of your problems to the chip, then there's some threading and some advantages that we can take.
Trevor: If we get sort of a hundred requests to use the chip all at once, that's a lot more efficient than if you're coding and running a for loop on your computer and calling those hundred problems one at a time. And sometimes you definitely need feedback from the processor and from the previous call of to the chip to decide what you're going to do next, but not always. And in those cases where you don't, it's better to asynchronously send all of those jobs to the processor.
Yuval: So we get close to the end of our discussion. I'm just curious how many of these computers are out there, if you can tell me?
Trevor: I cannot right now. I mean, so we have our legacy 2000 Q lower noise solver, and then we have at least one other solver that's like our current advantage solver, that's in the cloud. I don't know that I can talk too much about the other systems that are online. But I think there's two primary systems that users can use right now, which is our legacy 2000 qubit technology, and then our newer 5,000 qubit Advantage technology. I think that if you want to dig into this more, I would have to check in with our sales team and figure out what am I allowed to say about the other solvers and stuff.
Yuval: Sure, I understand. Does a computer require a lot of maintenance or, I mean, I'm sure that from the API perspective, I just submit a job up and get a response. But behind the scenes, are there tons of people who are fine tuning the qubits or cleaning or cooling or doing anything like that?
Trevor: No fine tuning really. I mean, so I spoke about our calibration procedure, which is a procedure from when we get the chip and we put it in a refrigerator to when customers all problems. So there's a procedure that gets us to that point. Once we're at that point, there's very little tweaking or hand tuning that we're doing. So there're suites of diagnostic tests, of course, that are running in the background, giving us health checks and consistency checks on the chip. And then, the chip is actually housed in a cryogenic enclosure so it's kept at cryogenic temperatures.
Trevor: And so, there is some fairly regular maintenance we need to do on the refrigerator just to make sure that all the cooling systems are healthy and operational. But I think we're evolving to the point where a lot of that is just automated. So the system will tell us when something is going wrong. But for the most part, the automated procedures are checking, making sure that it's healthy. And we don't have someone sitting by the computer turning a dial in any way, that's just not sustainable or scalable so.
Yuval: I really appreciate you answering this spectrum of questions for me. How can people get in touch with you to learn more about your work?
Trevor: There's two different ways. So definitely you can access and get in touch with us through our main company website. But I'm also happy to take questions at my email address. So my email is tlanting@dwavesys.com. And I'm more than happy to answer technical questions or questions about D-Wave, anything, it's fun to talk to people about quantum computing.
Yuval: That's excellent. Thanks so much for joining me today.
Trevor: Thank you very much, Yuval. It was a great chat.
My guest today is Trevor Lanting, Director of Science at D-Wave Systems. Trevor and I discuss what quantum annealing CAN and CAN NOT do, how the quantum annealer is maintained, what customers worry about when they deploy quantum solutions and much more.
Listen to additional podcasts here
THE FULL TRANSCRIPT IS BELOW
Yuval: Hello, Trevor, and thanks for joining me today.
Trevor: Hello, Yuval, nice to meet you.
Yuval: So who are you and what do you do?
Trevor: My name is Trevor Lanting, I'm an experimental physicist at D-Wave Systems. I work on our processor development team. So I've been at D-Wave for just over 10 years, and I've been involved with many of the aspects of the development of our quantum annealing processor technology. And now, more recently as we build up a gate model effort, I'm involved with that as well.
Yuval: So would you give me a really quick intro on what quantum annealing is?
Trevor: Yeah. So quantum annealing, in some ways you can think of as the quantum analog to parallel tempering or to basically simulated annealing, which is a heuristic that you can run on a classical computer to solve optimization problems. So quantum annealing is the quantum analog of that. Where instead of, as with simulated annealing, you turn up and down the temperature and explore your solutions space by slowly decreasing your temperature. With quantum annealing, you're turning on quantum mechanical fluctuations. And so you're exploring a large solution space for solving an optimization problem via quantum fluctuations. So you can actually put yourself in a superposition over all possible solutions to the answer, and then slowly turn down tunneling until you localize into what you're hoping is the ground state or a low energy state of an overall system that encodes the answer to a problem that you're trying to solve.
Yuval: So if I go to the D-Wave website, I'm sure it'll tell me about all the problems that can be solved with quantum annealing, and you've got plenty of customers and plenty of use cases, but what kind of problems cannot be solved with quantum annealing?
Trevor: So what we're building with the quantum annealing technology is not a universal quantum computer. It's very as much a special purpose technology that solves optimization problems. So I think a lot of your listeners are probably familiar with an algorithm called Shor's algorithm, which is an algorithm that was developed and shown to be very effective at factoring large numbers. Shor's algorithm is an example of an algorithm that cannot be run on our quantum annealing processors. You can run inverse multiplication problems on a quantum annealing processor, but you can't explicitly run an algorithm called Shor's algorithm. So, in general, any of the gate based algorithms specifically that were developed in terms of like with a gate model system, applying a series of gates to solve a problem, those don't run directly on quantum annealing processors.
Yuval: If I'm a supply chain expert, a logistics expert, for instance, I have optimization problems, obviously.What does it take for me? What do I need to know to be able to express the problem in a way that would work on our quantum annealing machine?
Trevor: So there're different interfaces to our technology. And what we're finding is customers are most often accessing our technology through our hybrid solver service. So there you can pose a problem as a constraint satisfaction problem, which is sort of a common way to express these types of optimization problems. And then, our software stack takes care of translating that into something that can be run on the quantum annealing processor. So you just need to pose a problem with some number of constraints and some interaction between variables and submit them to our hybrid solver service. And that will try to do the translation into something that can be run on the quantum annealing processor. There is sort of a more direct access point, which is that if you can formulate your problem as an ising spin problem, where you have pairwise interactions between variables, then you can express that directly as a machine instruction on the quantum annealing processor, but our current quantum annealing processes have sort of fixed topologies.
Trevor: So there's finite connectivity between each one of the physical qubits in the processor. And what we're finding is that when customers are coming to us, they're coming with problems that have many more connections that can be represented natively on the processor. And so there is a translation step that is required. So I guess there's a short answer to your question, which is to use our hybrid solver service, but you can really dig in and pose those problems at basically the bare metal level if you want, but you need to know a little bit more about the topology of the chip and be a little bit more familiar with the circuitry that's on the chip.
Yuval: When we look at gate based vendors, and we'll talk about D-Wave gate based announcement in a second. But when we look at gate based vendors, sometimes they say, "Oh, now we have 127 qubits", like IBM announced short while ago. "And therefore you can run these type of applications. And when we have 400 qubits, you'll be able to this. And when you have 10,000 qubits, you'll be to do that." Is there sort of a back of the envelope formula that says the existing D-Wave annealing computer can run something that would take X number of gates on a gate based machine?
Trevor: I mean, probably the clearest way to compare is to use an optimization algorithm called QAOA. So this is a quantum approximate optimization algorithm that was developed for this NISQ era, so this near term noisy quantum computing era. The QAOA algorithm is really designed to solve optimization problems on a gate system. And so there you can basically pose problems to quantum annealing solvers and pose those problems to gate based solvers. And so, I mean, we've been very effectively solving problems that are much larger than what can be posed to the current size of the gate based computers.
Trevor: So there is some way of, at least for specifically the optimization space, a way of comparing annealing and gate. But really what we're finding is that they're complimentary technologies in a lot of ways. And that's one of the reasons why D-Wave is excited about our gate efforts. For optimization problems, it really makes much more sense from what we know now to solve those problems with a quantum annealing platform, but there're areas like quantum simulation and quantum chemistry where really gate based systems are... That's where we think a lot of the early applications and attention will be on for those systems. So there are ways to compare them, but they really are complimentary technologies.
Yuval: So for QAOA what is the equivalent? So you can run a certain problem in today's quantum annealing machine. How many gates would I need roughly on a gate based machine to run a problem of the same size?
Trevor: That's a great question. I mean, and it really does depend on sort of the quality of your processor, so the so-called gate depth of your circuits, as well as the individual qubit quality and the coherence times of your qubits. So it really comes down to gate fidelity. So I can't give you a comparison, like a one to one comparison, unless I have a bit more information about sort of where the gate model systems are right now and what this hypothesized gate model system would need to be to compete with us on QAOA.
Yuval: Got it. Now, you've made an announcement that you're working on a gate based computer and that makes a lot of sense. Obviously, some of your customers might come to you and say, "Well, we love your annealing machines, but we've got these set of problems that the annealing machines don't work for." Are there benefits to you working in quantum annealing that would make you especially qualified to develop a really good gate based machine?
Trevor: Yeah. The answer is definitely yes. We are pursuing a superconducting based gate model approach. And so we've had a lot of experience in building up medium and large scale, superconducting processors and superconducting circuits. So in a very real way, what we have is a VLSI superconducting capability, very large scale integration capability. Our current advantage in yielding processors have a million Josephson junctions on them and very sophisticated wiring structures to actually run those processors. So that experience in building up large scale superconducting control circuitry and superconducting circuitry carries over immediately to our gate effort. We will be building our devices in a superconducting stack. And so we have a lot of experience on how to fab, how to test, and how to actually develop this capability. Also, calibration and characterization of these processors is extremely important.
Trevor: So we need to measure and characterize the individual qubits and couplers on our annealing processors so that we can effectively pose problems to the processor and those calibration steps we've had to sort of co-evolve how efficient our calibration and characterization is over time as the process has become more sophisticated. And this is something that will absolutely be a key part of the gate model effort is not only building the circuits, but being able to run them, and control them in a way that scales. And so our experience with our annealing development, we are quick carrying over to our gate development, and both the manufacturing design, and then really the running of these processors and the characterization of these processors.
Yuval: What's your best guess to when and how large would your gate based machine be?
Trevor: That's something that is hard for me to answer. So we know is going to be challenging. We have sort of a roadmap for producing gate circuitry, but we're not giving any dates out externally in terms of the roadmap for when customers can run things on our circuits. We do realize that this is a hard technology to develop and so this is one of the reasons why we're not being public about any dates yet.
Yuval: What do you think, if anything, is holding quantum computing back these days?
Trevor: That's an interesting question. So, I mean, I think there's some fundamental scaling challenges that the field no matter how you're implementing quantum computing needs to contend with. So we need to get to the many hundreds of thousands to millions of physical qubit scale to over the next five to 10 years in order to start running algorithms that really are practically useful. So there is a scaling challenge. I think for a lot of the development we've needed to, we know that we need error correction, and we need to be able to build logical qubits that are much more long lived out of large ensembles of physical qubits, and error correction overheads are large. And so I think there's a big challenge in coming up with techniques for error correcting physical qubits that don't have as large an overhead.
Trevor: And that's one way to help start attacking the scaling problem. So from my perspective, scaling these circuits, or these implementation out to a size where they're really competing with sort of the best supercomputers at solving, say quantum chemistry problems, and coming up with better ways and schemes and strategies for error correcting these physical qubits, those I think are the two big challenges in quantum computing. At least that my perspective is as a hardware person who is building the technology. I do think if you ask someone on the algorithm side, they'll say there is a similar set of challenges, which is coming up with connecting quantum computers to applications. So this is something where we've done a lot of work internally at D-Wave. But I think there's still so much development to be done, to come up with sort of software and algorithms that can run on quantum computers. If you look at where classical computing is, like the software is far outpacing the growth hardware. And I think that there will be a similar development over the next decade in quantum computing.
Yuval: The hardware side of your answer is a little bit idealistic. I mean, on one hand, it's encouraging and the other hand discouraging because you're saying, "Well, if we had a million qubits and they're error corrected, then you could do all these wonderful things." But wouldn't you think that if I have 10,000 qubits and they're not as noisy as today, but I could least measure and characterize and run hybrid algorithms, then there would be business value in running algorithms in quantum computers that cannot be done today on classical?
Trevor: Absolutely. I think there definitely is a near and a medium term value in the circuits that we're building. But again, this is where the second half of my answer comes in, which is that we need more people thinking about those algorithms and those applications. So that these will be co-evolving as the hardware grows and gets better. But before we hit kind of the, say the million qubit mark, we expect that this technology will be very, very useful, but there's got to be a co-evolution of algorithms and software along with the hardware.
Yuval: If I may ask you a business question, I think that you guys are probably unique in the quantum annealing machines. And so when customers come to you, aren't they worried that it's just one supplier and there's no alternative, and if you guys stop being available, then all my algorithms are going to go to waste?
Trevor: I would say, it's not the single source. So the short answer is no. I mean, I'm not as customer focused on the technology side, but from what I'm hearing from our professional services and sales team, we haven't heard that. But what we have heard is customers that really want, as we move applications to production, they want some guarantee of uptime. So they want to make sure that this technology, if they pose a problem and submit a problem with a cloud service, there's a very, very high reliability that that problem will be solved. And so it's not as much the single source of quantum annealing so far, although that could be a concern as we develop more applications and more people start depending on the technology, but really the feedback we're getting is your service is valuable.
Trevor: We want to know how is it going to be guaranteed to be up. In some ways can we treat it like Amazon Web Services where we really are relying more and more on cloud resources to run large parts of our business. So I don't know if that answers your question. I think people are a lot more focused on sort of uptime and the fact that we have solvers and production right now, and sort of have some guarantees, and from that perspective, and less from if D-Wave way is the only supplier right now and that's a concern. That could change, for sure. So to my knowledge, that hasn't been part of the discussions I've had with customers.
Yuval: How long does it take to get an answer? I mean, I know it depends on the complexity of the circuit, but if I submit the circuit and the system is up and it's correctly coded, how long before I can get the response?
Trevor: So that's also a hard question to answer, because, again, it depends on sort of at what level of the software stack and the interface you're accessing the technology. If you want to make a single ising call to our quantum annealing processor so you're calling directly an optimization problem and asking to be solved on the processor with all the, and again, it really depends on kind of network latencies. But if you're anywhere in the world, you have access to our cloud service, it's going to be on the order of several hundred milliseconds and maybe up to just under a second. So again, it really depends on a bunch of things, like how heavy is the usage of the chip, are there a lot of people that are actually submitting problems with the chip, and how good is your internet from, say the location where you're submitting the problem. But sort of rough sort of order magnitude estimates like a few hundred milliseconds.
Yuval: Is there a notion of cycle time in terms of how many iterations you run in a second?
Trevor: Yeah, I mean, we definitely pay attention to the number of problems we solve per second. And again, I mean, usually the way the workflow works is that there will be multiple processes that are calling the chip and asking for answers back. So there would be four or five, many dozens of users that are submitting problems. Even a single user, they could be solving or submitting a series of problems. And typically, what we recommend is if you're running a lot of problems that you batch them, or you can call the chip asynchronously. So in some cases isn't possible, but if there's a way where you can kind of batch call all of your problems to the chip, then there's some threading and some advantages that we can take.
Trevor: If we get sort of a hundred requests to use the chip all at once, that's a lot more efficient than if you're coding and running a for loop on your computer and calling those hundred problems one at a time. And sometimes you definitely need feedback from the processor and from the previous call of to the chip to decide what you're going to do next, but not always. And in those cases where you don't, it's better to asynchronously send all of those jobs to the processor.
Yuval: So we get close to the end of our discussion. I'm just curious how many of these computers are out there, if you can tell me?
Trevor: I cannot right now. I mean, so we have our legacy 2000 Q lower noise solver, and then we have at least one other solver that's like our current advantage solver, that's in the cloud. I don't know that I can talk too much about the other systems that are online. But I think there's two primary systems that users can use right now, which is our legacy 2000 qubit technology, and then our newer 5,000 qubit Advantage technology. I think that if you want to dig into this more, I would have to check in with our sales team and figure out what am I allowed to say about the other solvers and stuff.
Yuval: Sure, I understand. Does a computer require a lot of maintenance or, I mean, I'm sure that from the API perspective, I just submit a job up and get a response. But behind the scenes, are there tons of people who are fine tuning the qubits or cleaning or cooling or doing anything like that?
Trevor: No fine tuning really. I mean, so I spoke about our calibration procedure, which is a procedure from when we get the chip and we put it in a refrigerator to when customers all problems. So there's a procedure that gets us to that point. Once we're at that point, there's very little tweaking or hand tuning that we're doing. So there're suites of diagnostic tests, of course, that are running in the background, giving us health checks and consistency checks on the chip. And then, the chip is actually housed in a cryogenic enclosure so it's kept at cryogenic temperatures.
Trevor: And so, there is some fairly regular maintenance we need to do on the refrigerator just to make sure that all the cooling systems are healthy and operational. But I think we're evolving to the point where a lot of that is just automated. So the system will tell us when something is going wrong. But for the most part, the automated procedures are checking, making sure that it's healthy. And we don't have someone sitting by the computer turning a dial in any way, that's just not sustainable or scalable so.
Yuval: I really appreciate you answering this spectrum of questions for me. How can people get in touch with you to learn more about your work?
Trevor: There's two different ways. So definitely you can access and get in touch with us through our main company website. But I'm also happy to take questions at my email address. So my email is tlanting@dwavesys.com. And I'm more than happy to answer technical questions or questions about D-Wave, anything, it's fun to talk to people about quantum computing.
Yuval: That's excellent. Thanks so much for joining me today.
Trevor: Thank you very much, Yuval. It was a great chat.
About "The Qubit Guy's Podcast"
Hosted by The Qubit Guy (Yuval Boger, our Chief Marketing Officer), the podcast hosts thought leaders in quantum computing to discuss business and technical questions that impact the quantum computing ecosystem. Our guests provide interesting insights about quantum computer software and algorithm, quantum computer hardware, key applications for quantum computing, market studies of the quantum industry and more.
If you would like to suggest a guest for the podcast, please contact us.
