Josh Savory, product launch manager at Quantinuum, is interviewed by Yuval Boger. Josh and Yuval spoke about their recent H2 system launch,GHZ and non-abelian topological states, what the H2 launch means for customers, whether 32 qubits are more useful than 400+ qubits, and much more.
Full Transcript
Yuval Boger: Hello, Josh, and thank you for joining me today.
Josh Savory: Hi, Yuval, glad to be here and appreciate the opportunity to speak about the new product we just released.
Yuval: So who are you and what are you doing?
Josh: So I’m a product launch manager. I work at Quantinuum and my job is to, prepare for when we have a new product to release, I bring all the teams together and make sure that we have smooth sailing for the release process. So that includes our legal, marketing, technical teams, and I work with all of them to make sure we introduce the product appropriately to the market.
Yuval: And I hear there’s a new product that you just launched. Tell me about it, please.
Josh: Yeah, so we’re continuing on our roadmap. Right now we have, as most people are aware, our H1 processor, which is a linear QCCD ion trap that holds up to 20 qubits. We’ve just released a new product that expands upon that up to 32 qubits but keeps the same level of fidelities as well as the same feature set and this greatly increases the computing power of that device.
Yuval: So going from 20 qubits to 32 qubits is a 60% improvement. Is that the performance improvement or the utility improvement that you can expect from this new system or is there more to it?
Josh: There’s more to come! So we’re looking to increase performance within the H2’s lifecycle we hope to go beyond what you can classically simulate so that would mean going above 50 qubits. The H2 processor was specifically designed to be upgradeable. We’re initially releasing at 32 qubits, but we’re hoping in 2024 that users can expect to see this increase even higher, to go above 50, which is really that regime we hope to see some quantum advantage demonstrations.
Yuval: What can the H2 do that the H1 can’t or can’t yet? I mean what kind of demonstrations could you have for showing the power of this new product?
Josh: Well I mean the simple answer there is that we’ve done three demonstrations on this device and a lot of those would not have been possible , on the 20 qubit processor.
We’ve done three demonstrations. So if we look at one of the first one, which was a benchmarking paper, so this was, we looked at doing the most thorough evaluation of the H2 device possible. We did over 14 different benchmarks.
Another good example there would be in the H2 benchmarking paper, we set a world record for a GHZ state. So we were able to measure a 32 qubit GHZ state. So this is the largest on record and this would not have been possible on its predecessor.
Another example is maybe some of you have already seen the creation of this non-abelian topological state. This would not have been possible on the H1 device because at a minimum it would have required 27 qubits to do that demonstration. So just the qubit number in and of itself unlocks exciting new possibilities on this device when you combine it with our best-in-class fidelities.
Yuval: Let’s go back and clarify, what is GHZ and why is it significant?
Josh: GHZ state, so the GHZ state does not stand for gigahertz. These are the initials of the initial people who predicted this state. This is an entangled, fully entangled state of the up and down. And it goes across all 32 qubits. And what’s nice is it shows that you have a functional quantum computer across all of those 32 qubits.
Yuval: And continuing this path, what is a non-abelian topological state?
Josh: A non-abelian topological state. So this is really getting outside of my area of expertise, this really is in the domain of Henrik Dreier who works in our Munich office. He did this in collaboration with collaborators from Harvard University and Caltech. So they’re really focused on doing condensed matter demonstrations. And so what they’re really looking at here is quasiparticles? So these are where you set up different entanglement patterns and arrange the underlying particles in to give desired statistics. So when we talk about Abelian, that means these particles are quasi-particles where you can exchange those particles in space and you get a somewhat, you get a phase change in the overall wave function. The importance of non-abelian is that when you do this exchange of these quasiparticles, you actually gain information in the system. there’s been a lot of work on how non-abelian states could be used to realize more robust fault-tolerant quantum computers. So this demonstration is really important because it sets us up to explore non-abelian statistics, but it also sets us up to look at the path forward for how could these be used to realize a fault-tolerant quantum computer, which we believe is really exciting and unlocks a new set of experiments that you could do on our devices.
Yuval: When you say you launched the H2, what does that mean for a customer? So if I’m a financial services customer, could I now submit jobs on the H2, or is the time to submit jobs several months into the future?
Josh: Yes, so one of our demonstrations was JP Morgan who’s one of our customers. They did a soft launch project. That paper is on ArXiv. They were looking at another method to improve portfolio optimization.
Yuval: Some vendors report many more qubits, 120-something qubits, or more than 400 qubits. How does that compare? Is it really just the qubit number? I mean, how would, if I were an end customer, how would I assess which vendor is the best for me?
Josh: So at Quantinuum, we really focus on the quality of qubits. And when we deliver a product, our focus is really on providing the highest quality and most usable set of qubits that we can provide. And that’s why we really focus on this quantum volume metric. This is in another one of our papers that was done by our theory team, as well as all the technical teams at Quantinuum. We released with this device, we’re hitting a quantum volume which is a world record. No other device has been able to demonstrate something like that.
Yuval: Do these results, do they happen with a specific tailwind? I mean, is it sort of once in a blue moon that you get these good results from the machine, or do you feel that they would be consistent?
Josh: If I ran this experiment several times, I would get pretty much the same results. These are consistent results. So in our benchmarking paper, we’ve released, like I said, we did 14 different optimization problems where we’re looking at single qubit gate fidelities, two qubitgate fidelities, and our memory error. And so not only do we put that on ArXiv, but we’ve also released all the data behind that on GitHub. Users are free to download that as well as any potential customer can go and look at that data. So all the data behind the benchmarking paper is available on GitHub and so users can really see there’s really statistics behind these and these are repeatable measurements.
Yuval: You mentioned quality and you mentioned that Quantinuum is focused on perhaps not the largest number of reported qubits, but the combination of qubits and qubit fidelity and quality. Does that lend itself to one or more particular applications, you know, maybe particularly good for machine learning or particularly good for optimization, or is it truly universal in its applicability?
Josh: Well, so I just would say one thing on top of that, that it’s in addition to our quality number of qubits. I also think another fact that sets our products apart is the feature set. This allows demonstrations that you wouldn’t be able to do on other quantum computers. That would be mid-circuit measurement with very low measurement noise. We have all-to-all connectivity. This means that you can implement circuits with a smaller number of gates. Because I can connect any two qubits without having to do a swap gate. So this means you can implement circuits with lower overhead on our devices. And then we also have the mid-circuit measurement and qubit reuse. So this allows you to reuse a qubit, you reuse a qubit mid-circuit to effectively increase the number of qubits that you have in your machine. So with the combination of all of these, I really think that I don’t think it’s one area. It really is enabling across many different areas. And we see that with our use cases.
for instance, one of our demonstrations, I’ll go into this one was a team here at Quantinuum and they were looking at how to improve combinatorial optimization. They were using a heuristic method to show how we hope to solve these type of problems in the future once we have more qubits and they were looking at a new method that reduces the overheads required. We really feel that the device we’ve made is a testbed or is a sandbox for many different areas including condensed matter, quantum machine learning, quantum error correction, finance problems. There are a lot of different areas that we feel this device is applicable to.
Yuval: As a product release manager, what keeps you up at night? Or maybe now you sleep well now that the product has been released?
Josh: It actually has been a very enjoyable process at Quantinuum. So there has been, to be honest with you little, that really worried me. We have a great team and that doesn’t just extend the hardware side of things, the technical teams were great, but also that we have a lot of, we’re an integrated company, so we have a lot of support on the software side as well, and applications, middleware, so we go full stack. And I think that the big thing for this release was to show, to really demonstrate that. So for me, it’s really about having a launch that highlights our product. And having the company that we do has been more exciting than me being afraid. So just really how big we could go with our demonstrations. Andwe set an ambitious goal and we exceeded that. Like I said, we have three papers on ArXiv going out with this release as well as an external paper from JP Morgan as well.
Yuval: I’ve had the privilege of hosting several companies, several types of modalities, several manufacturers on this podcast. And everyone would come in and say my qubits are better than your qubits and so on and so on. Do you feel that there’s a need for you know quantum Olympics? Do you think that customers focus on quantum volume or do they just tune out this sort of marketing and look at it as marketing noise that vendors just have to say?
Josh: For different customers, you go with different arguments. And that’s really what we are trying to showcase here is, you know, in our benchmarking paper, if that’s what you’re interested in, the quantum Olympics, that’s what we did is we went through, we really kicked the tires on the H2 and this will allow anybody to compare to our device. So all the numbers are there, but then you have some customers who want to see applications, right? And you can quote fidelities but what they really want to see is see a new demonstration. And that’s really what the non-abelian topological state paper, that’s something that you couldn’t do on any other device right now. We feel that that shows an application that sets us apart from others, as well as the demonstrations by the QML team and JP Morgan, that speaks to the applications that you can run on this device.
Yuval: And to the best of your knowledge, when you think about JP Morgan, did they try running their algorithms on other manufacturers and then basically came to the conclusion that it works really well on the H2 machine? Or were they essentially focused on your product line from the get-go?
Josh: Yeah, so to be transparent, this was a soft launch demonstration for them. So the goal here was not for them to compare our device to other devices, but really to kick the tires on H2. So I can’t speak to what other devices that they’ve tried their product on.
Yuval: Quantinuum, I believe, has a software layer that is presented as being hardware-agnostic. Do you feel that that’s really the case? Or essentially you’re saying it’s a hardware-agnostic software piece, but in reality, it runs the best on the H2 machine.
Josh: Well, I would say that, that software is really hardware agnostic. And I would say it runs well (on our device) because we have a great device. Not that we’ve tuned that software in any way to just highlight our device. I do feel because we are an integrated company, it gives us the opportunity to optimize things better, but it doesn’t change the fact that, TKET, runs and incorporates many different hardware types. And there’s nothing that they’re doing to limit performance on other devices.
Yuval: As we get to the end of our discussion today, I wanted to ask you a hypothetical. If you could have dinner with one of the quantum greats, dead or alive, who would that person be?
Josh: One of the quantum greats. I haven’t thought about this question before. I think from a physics standpoint, I mean Richard Feynman would be the guy that I would like to have a conversation with. I don’t know if he falls into the category you’re looking for, but I certainly, from his work, I think he would be one of the most interesting physicists to speak with.
Yuval: Absolutely falls into the category and in fact, it’s probably the most common answer I get.
Josh: Well glad to know I fell within that mean.
Yuval: So Josh, it was a pleasure having you here. Thank you so much for joining me.
Josh: Thank you, Yuval. It was a pleasure to be here.