Rob Schoelkopf, co-founder and chief scientist at Quantum Circuits Inc., and Ray Smets, CEO, are interviewed by Yuval Boger. Rob shares insights on Quantum Circuits’ unique dual-rail qubit approach, which focuses on reducing error rates through error detection and correction at the hardware level. Ray emphasizes the company’s shift from scientific research to commercial implementation, highlighting its cost-efficiency and full-stack quantum computing solution. They discuss the scaling potential of superconducting qubits, the importance of error correction, the next steps for Quantum Circuits’ technology, and much more
Transcript
Yuval: Hello Rob, hello Ray, thank you so much for joining me today.
Rob: Hi Yuval, how are you doing?
Yuval: Living the dream. So Rob, who are you and what do you do?
Rob: Yeah, I’m Rob Schoelkopf, I’m the Sterling Professor of Applied Physics at Yale University and I’m the co-founder and chief scientist at Quantum Circuits, Inc.
Yuval: And Ray, how about you, who are you and what do you do?
Ray: I’m the CEO of a really impressive company called Quantum Circuits, Inc. and a partner with Rob, who is the founder of this company. I think we’re making some very important progress in the world of quantum computing and we’re really happy to talk about it today.
Yuval: Why does the world need another quantum computing company? I mean, there seems to be so many. What’s new and different about your approach?
Rob: Well, I think the thesis behind the founding of Quantum Circuits, Inc. is that we really felt that the approach that everyone was taking was probably a little too cumbersome to really scale to do useful problems. We would need to have a partnership between academic research and industrial development that was really focused on efficiently solving the problem of quantum error correction.
Yuval: Okay, so tell me how you got to this and what is the innovation in your technology?
Rob: Yeah, well I guess as part of the backstory, I started my lab at Yale in 1998 before anybody had solid-state qubits. Maybe trapped ions were already around but nobody had built a superconducting qubit, and it actually started as curiosity-based research. We were asking the question, can you ever build a man-made macroscopic object like a circuit that’s a millimeter on a side and have it follow the true rules of the quantum world with superposition, entanglement and all of that sort of stuff?
There were a lot of innovations along the way. For instance, we founded this field of circuit quantum electrodynamics where you use microwave signals to control and measure and connect various qubits. Then around 2008 or 2009 came the transmon, which made superconducting qubits really kind of stable. We were able to do some of the first quantum algorithms just in the university lab, and we started to see the growth of the industrial efforts with IBM, Google, and Rigetti, all doing subtle variations on this basic platform of transmons and circuit QED.
Yuval: How is your superconducting qubit different than IBM or Rigetti or IQM or all the other companies that are doing superconducting?
Rob: We’ve focused since about 2010 or so on a kind of flipped paradigm in superconducting devices. We don’t use the Josephson junction-based transmon qubits as the primary information carrier. We use true microwave photons in some kind of linear resonator. It can be an LC on a chip, or a three-dimensional cavity, and that offers a lot of advantages. These devices can have longer coherence times and they also have a simpler error model. They have fewer types of errors and errors which you can devise more efficient ways of detecting and correcting.
Yuval: Ray, I think you are an accomplished business person and you’ve joined the company not too long ago. Does that signal that you’re close to having a product that’s out of the lab and into users’ hands?
Ray: Yeah, it’s a great question and there’s a reason why I’m here. When Rob talked about starting this back in 2008, I was thinking about where I was at the time, working on security, IP, wireless LAN, mobility, and here I am working in probably one of the most interesting science areas that is about to go commercial.
I think the industry would agree we’re kind of at that tipping point, moving from the science world to the engineering world and ultimately to the commercial world associated with bringing quantum solutions to the marketplace. This company has been working very diligently building the science around the inventions that Rob just talked about, the innovations around the dual-rail qubit and the approach. This company has really, from the very beginning, been focused on correcting and then scaling.
That science has been proven to work, so now it’s time to take it to the market. That’s when I had a chance to meet Rob about a year ago and talked about the vision and the potential of this business and decided it was about the right time for someone such as myself. I bring years and years of experience working in Silicon Valley, taking technology to the market and building customers and revenue. I think the time has come for us to do the same here at Quantum Circuits. That’s why I’m here, to kind of bolt the superposition of science and business together at the same time and see if we can’t make the world a better place with quantum computers.
Yuval: So you started in 1998. I think I heard yesterday that Chris Monroe had the first trapped ion qubit two years prior, but it sounds like an overnight revolution, 25 years in the making for you. Give us some numbers. I mean, how do you measure the technical success? How many qubits do you have? What’s the single-qubit gate fidelity? What’s the two-qubit gate fidelity? What’s the coherence time? Something that we can compare to existing superconducting solutions.
Rob: So we have been following this kind of alternative approach and trying to really build devices that can detect and correct the errors at the hardware level because we don’t want to have to massively scale to thousands of physical qubits to make one high-performance qubit.
A year or two ago, we really had a breakthrough on this approach, which is called the dual rail. In the dual rail, you have a qubit, which is the superposition of a single microwave excitation in one 3D cavity or in another 3D cavity. It allows you to basically detect when a photon is lost, which is the dominant source of error in this kind of system.
The basic idea is that even if you’re making something which is at the physical level near the current state of the art where fidelities are 99% or something like that, you basically get to have the first round of error correction built into the devices and effectively get square that probability.
So, as a first instance, there’s this metric called SPAM, state preparation and measurement, which people don’t talk about as much. But of course, you’ve got to initialize the computer and read it out. In the superconducting platforms, that’s typically a few percent error per qubit, which means you’re kind of starting out with something that’s not super high fidelity, even before you’re doing gates. With these dual-rail qubits, you get to detect errors in the preparation. You can prepare the state, check it, then use it, and detect at the end and throw away or detect the shots that have an actual error in them.
We were able to demonstrate SPAM at the 10 to the minus four level, so really like 99% squared. What we’ve been doing now at QCI is scaling up the first versions of this qubit into the first small machine. The things we’re operating in the lab right now have a handful of these dual-rail qubits with gate fidelities that are pretty competitive. But they also come with this error model, which allows you to incorporate them into even smaller correcting codes and get some gains.
So what we’re looking to do in the near future is build machines with, say, a few dozen of these dual-rail qubits. And there are two uses. One is it’s a post-NISQ machine where you can operate it as some number of physical qubits and detect the errors, but then still keep the shots that are good, meaning you have much higher fidelity computation in the end. And the second use is to implement error-correcting codes on top of this more robust version of a qubit. The goal there is to have it that as you increase the distance of a code or you add another layer of redundancy in the hardware, you don’t gain a factor of 1.5 or 2, you might gain a factor of 10, or maybe even 20 or 100. And then if you can scale that sort of a machine, maybe you only need a few hundred physical qubits to make one very-high performance logical qubit.
Yuval: So when I listen to you describe this, it reminds me a little bit of what people say about cat qubits, right? That there are phase-flip errors and bit-flip errors, and maybe they’re kind of inherently immune to one type and therefore need fewer physical qubits to create a logical qubit. Is that true? I mean, is the analogy correct?
Rob: Yeah. The cat qubits that a few people are working with are also things that were innovated by my collaborators and me at Yale. They often use the same technique or always use the same technique of encoding the information in a cavity. Also the cat qubit is the first device that actually performed error correction in real time on the native errors and got a gain. That was an experiment we did in my lab in 2016. The gain was like 1.2 or something. More recently, gains have gotten up to about two.
The cat qubit approach is a little bit of a more complex thing, trying to stabilize and get a nice error model. But the dual rail allows you to just build these cavities and you get the nice error properties kind of for free. And the other thing that’s really hard with some of these devices, like the cat qubits, is how do you actually perform gates and entangling operations, single and two-qubit gates on them? It’s a much more complicated manipulation than just working with two transmons.
A nice feature of the dual rail is that is a simplification from cat or other cavity qubits. I would that the dual rail is to error correction as the transmon is to the Josephson junction qubit. So before the transmon, there were all these different varieties of superconducting qubits based on flux or charge or phase, and they all had their strengths and their weaknesses. The transmon was something that was much simpler than any of those approaches. It was this head-slap moment, like why didn’t we think of this earlier, which avoided all of the known problems and made something that was stable.
For me, the dual rail is like that same head-slap moment, like, oh, here’s the simple thing that takes this thesis of it being better to work with cavities rather than transmons directly, and makes it simple and easy. So in cat qubits, you can do some gates, but they also have a few percent error even for a single-qubit gate. With the dual rail, we get single-qubit gates that are in the 10 to the minus four, 10 to the minus five range. And they’re only 50 to 100 nanoseconds in duration, so they’re really quite competitive right away, even with the transmon, but they have this error-detection capability now built in.
Yuval: Ray, if we zoom out for a second from the head slaps and the cats and the transmons, tell me a little bit about the company. How large are you? How are you funded? Where are you physically based? What can you share?
Ray: Well, I’ll start with where we’re physically based. We’re sitting right next to campus at Yale in New Haven, Connecticut. This is ground zero for Quantum Circuits, and we think it’s ground zero for what the state of Connecticut calls the quantum corridor, which they’re working very hard to develop. So we’re very excited to be part of that physically in New Haven and next to Yale. The company’s about 70 people large. Obviously, a very large percentage of them are PhDs, many of which have graduated right out of Rob’s program.
It’s a group of very young, bright scientists working on what they believe is the ultimate best solution in quantum computing. Our investors aren’t too bad either, by the way. If you take a look at our investment group, we have some of the biggest names in the community investing in us. They believe in us long-term—Sequoia Capital, Canaan, Arch Ventures, F Prime and others. We recently added In-Q-Tel to our group of investors, and we feel pretty proud about that since we’re, in many cases, their only investment in quantum computing.
So one of the things I think is very unique about this business as we’ve been working towards working quantum computing solutions is we’ve burned the least amount of capital compared to others who’ve reached similar milestones. And I think that’s a pretty interesting and notable approach. It kind of backs up what Rob is saying—it’s a simpler approach to solving a bigger problem, and we’re hoping we can carry that forward. So a little bit about our business: this is a stepping-stone approach. We’re taking something to the market. We think we have something good. We’ve validated it. We’ve talked to other scientists to prove it.
Now it’s time to actually get validation by real users, letting real customers try it, modifying the algorithms that they may have written to work on other quantum solutions, adding our novel features, which are capable of doing new things in quantum computing that they couldn’t have done before, and just moving the needle in terms of their ability to discover something that they need to discover within their specific domains—whether it be in pharma, petrochemical, networking, financial services, or whatever the case may be. We’re going to get that validation and show it to the public. So that’s kind of the journey that we’re on.
We’ve been sciencing the heck out of this for years. We’re in the process of feeling very good about where we’re at and beginning the engineering process. We believe we have something that can actually scale.
Rob: Can I interject one thing there? One thing that’s maybe different about our model: I still have a dual role. I’m primarily at the university, but I’m a consultant here at Quantum Circuits, Inc. The idea is that we’re still doing fundamental discovery work in the right kind of environment to do that, which is graduate students asking open-ended questions and doing high-risk, high-reward explorations of novel things.
But at the same time, what’s different here at Quantum Circuits is that there’s a cadre of the quantum physics and circuit QED experts, but we’ve also added the software engineers, the mechanical engineers, the RF electrical engineers, the firmware engineers that it takes to build a full platform. One way we’ve been able to be capital efficient is we’re not burning investors’ capital creating everything from the start. We get to focus on the engineering and the scaling of things while the science is done on campus.
Yuval: When could someone access this machine?
Ray: It’s a very good question. We already have commercial customers that are engaged with the company, beginning to work through our full-stack solution. We have a cloud portal, and we have a solution where the software can be developed using these new novel features, leveraging the error correction capability that Rob talked about, and the error detection capability that is built into the qubit from the start. We’re moving that through the simulation environment to prepare the software for actual customer use on cold hardware running in the lab. So that’s happening imminently.
You’re actually talking to us at a very important tipping point in the development of this company, as we’re going from lab to commercial implementation literally as we speak. So it’s happening right now.
Yuval: You mentioned cold. Does one need a dilution fridge to operate this?
Rob: Like all of these superconducting technologies, we operate the quantum part at 10 millikelvin, and it’s all controlled by RF high-speed electronics that sit at room temperature. There’s still a long way to go with that basic approach. We’ve seen people brute-forcing things up to hundreds and now even thousands of transmon qubits. I think there are more elegant and simpler ways of doing some of these things that make the engineering challenges around operating a large-scale system a lot easier.
The cryogenics is not really the hard part. That’s another thing that’s changed a lot since I started back in 1998. Now they’re not only commercially available, but they’re kind of “push a button that says cool down,” and provided everything is good in the vacuum system, you’re just off and running a day or so later. We feel like this is not really a bottleneck.
Ray: And as we think what the future of this industry will look like, of course, we’ll build hardware, quantum computing systems that are in cryogenic fridges and ship them and sell them to customers who want them physically in their location. But the real market opportunity is going to come from providing those solutions through the cloud. So it doesn’t matter that they’re sitting cold in a fridge. The power of those systems will be available to anybody who wants access to them through cloud-based solutions.
So I often get a question about, well, who’s going to win? This is a race, and it’s like the new space race. Who’s going to get there first?
It’s a compelling conversation to have, especially over a cocktail or a beer with a few scientists and businesspeople in the room. My point of view on this one is that this is a race where there are going to be multiple winners. There’s going to be a lot of potential for this market to generate a tremendous amount of value across multiple domains—not just quantum computing, but all the domains that will use it. There are going to be multiple solutions available to that marketplace to solve a lot of problems. I think a company like Quantum Circuits, with its full-stack approach, its novel dual-rail cavity qubit approach, with its correct-first-and-scale approach, has an opportunity to be one of the winners. And we feel like we’re on our way to make that happen.
Yuval: You mentioned earlier that customers would modify their programs to use your machine, and now you mentioned full stack. So help me understand that. Do I, as a customer, just give you a QASM program and say, “Hey, run it on the dual-rail machine,” or do I have to do something different to take advantage of the hardware?
Rob: Right. So, we’ve demonstrated the canonical universal gate set on these dual-rail qubits with all the single and two-qubit gates. It’s perfectly possible to, and we’ve done it, take a Qiskit or QASM program, push it through the full stack, and run. The thing that’s different, of course, when you start thinking about error correction is you don’t want to just compile a program, play a pulse sequence, and measure what comes back at the end.
What you need to be doing, of course, is watching for the errors as they occur and then taking some corrective action and responding in real time to the detection of an error in the quantum system. One thing that’s unique about our control system and software stack is that it’s got built-in real-time detection, allowing you to branch and change what happens next, depending on an error. This is what I mean when I say post-NISQ. That’s not something you’re typically able to do on an ordinary machine.
One example could be you’re running an algorithm, part of the computation has been detected to have an error, but you can still measure some observables that aren’t in the region where the error could have propagated. You get something out of that shot, even though it contains a certain error. The simplest example is just recording the errors and then giving back to people a report: “Okay, we ran it a million times; this many times it had three errors, this many times it had two errors, here are the ones with one error, and this is the location in which the errors occurred, and here are the shots that don’t have any errors, and give you a much higher fidelity.”
So one of the things we’ve done, just as a demonstration of this potential with two qubits, is running the usual variational quantum eigensolver routine on a two-qubit H2 model. What we’re able to do with just detecting these errors is get something which is such a high-fidelity computation end-to-end, from state preparation to gates to measurement, that you reach chemical accuracy without any error mitigation or other techniques like that.
I think there’s a hunger in the market right now for machines that have new and experimental features that have some differentiated capabilities. We think there is a lot of interest, and indeed, as we’ve been talking to people they kind of perk up when they hear that it’s not just the same as running on one of the other machines that’s out there already.
Ray: I’ll just add to that, since Rob’s not always in the room when we talk to these potential commercial customers, we call them alpha customers today. It’s really enjoyable to watch data scientists who’ve been working in this space for quite some time, and you put on the table something they’ve never seen before. The ability to do real-time control flow and change the way the algorithm works based on real-time feedback from the error is something new and novel. It’s expanding their potential, and it’s really enjoyable to watch the eyebrows go up when we put that on the table.
Yuval: Let me play back what I heard and ask the next question here. So, superconducting qubits, new types of superconducting qubits, cryogenic cooling required, inherently lower error rate. I would guess that connectivity would be similar to what you see in other superconducting systems. It sounds like you would need fewer qubits to get meaningful results, but still, at some point, people want thousands or tens of thousands of qubits. How does this scale? Whether trapped ions or superconducting, other manufacturers say, well, we’re going to have optical interconnects, or every time we get more than a few hundred qubits, we’re going to have multiple networked machines. Does that also apply to your approach?
Rob: Yeah, I think the superconducting approach, and in particular our implementation of it, does really well with this kind of modular architecture. We’ve already done and published science experiments where you have some of these qubits in separate modules, and you just need a simple microwave cable that links the two modules. A really interesting thing with superconducting devices is that you can convert from a stationary qubit, including one of our dual rails, to a flying qubit in the same time that it takes to do a single gate.
The main problem you get with these kinds of modular connections is typically the quality factor of the wiring is not as good as what’s inside the processor. What that means is you just have a little bit more photon loss on those links, which is something that we can apply all these error detection and correction techniques to. I think there’s a pretty straightforward path in terms of scaling the quantum hardware.
An interesting question is, how big is the block going to be that you incorporate? My answer to that is always, well, it’s the biggest quantum computer you can reliably build, and then you make two of them and connect them together. But kind of more practically, I think it’s straightforward to get several hundred in a single block and then be connecting those together. I think compared to ions, it’s a lot nicer because we don’t have to invent the high-efficiency ion-to-ultraviolet photon link and all those sorts of things. I think the superconducting systems are in the lead.
Another interesting thing that’s a bit different about our technology is that it’s modular already. We don’t make a single monolithic chip. We make these three-dimensional modules, and they contain multiple chips that have the Josephson junction devices for the input and output and control. What’s nice about that is if you cool a system down and one or two of your devices are not quite in spec where you wanted them to be, you just can change out those individual devices, and you can rapidly converge to something in a way that’s less capital-intensive than building a foundry and cranking out hundreds of chips until you get the one that works.
Yuval: As we get closer to the end of our conversation today, I’m curious: if this dual-rail thing didn’t work, or maybe you determine in a few years that it doesn’t work, what’s your next best favorite quantum modality?
Rob: It’s kind of interesting. It’s hard to foresee the future. When we started, I didn’t actually imagine that we would get to a budding industry within my professional lifetime. We definitely didn’t expect many of the innovations or breakthroughs that have happened along the way. One of the things that happened around 2010 was discovering this three-dimensional approach and solving some of the materials problems, and having coherence times of transmons leap by two orders of magnitude. That was something we did in our labs at Yale.
I think we’re working on various kinds of innovations that can offer the same kind of transformational breakthroughs and can be applied to this basic architecture. There’s certainly room to make better cavities, better kinds of gates. It’s kind of fun because this is a fairly new approach. We’re doing things like inventing six different ways of doing a two-qubit gate all in the last half year or so. I think the path weaves and is a little hard to predict sometimes, but when you’re doing it right, you don’t have to throw everything out and start again. You can layer the innovations on top of what you already have.
Yuval: That was a very elegant non-answer because you’re basically saying, “It’s going to work. Don’t worry about it. But if it didn’t, I don’t care.”
Ray: That’s one of the best “stump the scientist” questions I’ve heard in a long time.
Rob: I don’t want to talk about the specific things that I think could be huge jumps until we actually show that they’re real. But I do think that this dual rail is the simplification, the differentiating factor, that’s really going to enable us to solve the problem of error correction. That’s the really important thing that we’re focused on.
Ray: From a less scientific point of view, my observation is this industry has been living on levels of enlightenment. The dual rail is another level of enlightenment. It’s a new approach, solving a problem that’s very complicated and has been perplexing other companies for years to get to. I think we’re going to contribute to the industry in a pretty interesting way as more and more people become aware of what we can do. At the same time, we’re going to learn a little bit more about how we can improve what we’ve got. I don’t think there’s a short window here of opportunity on dual rail. I think this one’s here to last for a long, long time. We’ll take a look and see what develops outside and see if it can apply to what we’re doing and we’ll all reach a new level of enlightenment together.
Yuval: My last question is a hypothetical, if I may. Maybe Ray first and then Rob. If you could have dinner with one of the quantum greats, dead or alive, who would that person be?
Ray: Well, I mean, for me, it’s going to be Einstein, but I’ve got the next Einstein here sitting with me on this call. I am thrilled every day to be a partner of Rob, to sit around and watch his vision come to reality. We’ve had many dinners that I would consider to be quite fascinating already. I’m looking more forward to that as well. I’m curious to hear what Rob’s answer is to this question.
Rob: Yeah, I think the first one that leaps to mind for me is Norman Ramsey. Everybody does a Ramsey experiment and talks about their Ramsey times all day every day. He was one of the real innovators of both NMR and atomic physics. Sadly, I never got to meet him. The other one I really would like to meet now that I think of it is Ed Purcell. We all know the Purcell effect, and he was both one of the drivers of microwave technology at the Rad Lab during World War II and then also all these kinds of innovations that came after. Those two would be nice to sit down and have dinner with. I think they’d be amazed about where we are today.
Yuval: Wonderful. Ray, Rob, thank you so much for joining me today.
Rob: Thanks a lot, Yuval. It’s been a lot of fun.
Ray: Yeah, thanks very much.