Transcript | From Qubits to Qumodes in Photonic Quantum Computing with QuiX Quantum

In late 2020, physicists in China generated controversy by claiming quantum advantage with a photonic system that’s technically not programmable. Other companies have been experimenting with photonic systems, including QuiX Quantum. How do these machines work? Should scientists redefine what “quantum advantage” means?

Join us for a light exploration of light-based quantum computing in this episode of The Post-Quantum World. I’m your host, Konstantinos Karagiannis. I lead Quantum Computing Services at Protiviti, where we’re helping companies prepare for the benefits and threats of this exploding field. I hope you’ll join each episode as we explore the technology and business impacts of this post-quantum era.

Our guest today is the CTO of QuiX Quantum, Dr. Jelmer Renema. Welcome to the show.

Thank you.

Right now, I’m in San Diego, at the IQT conference while I’m recording this, and I’m reaching to you all the way in the Netherlands.

Yes. I’m coming to you today from my home office in the Netherlands.

Excellent. Tell us a little bit about your background in quantum photonics.

I did a Ph.D. at Leiden in the group of Martin Van Exter and Dirk Bouwmeester on superconducting single-photon detectors. I then did a postdoc in Oxford in the group of Ian Walmsley and then moved back to the Netherlands on a fellowship and started there initially as an academic, and then founded QuiX Quantum in 2019 together with Hans van den Vlekkert.

It seems like your background helped direct the direction of the company with that focus on photonics.

Yes — absolutely.

QuiX Quantum’s relatively young. It’s a startup. It’s three years old now. Tell us a little bit about the company and why you started it.

QuiX Quantum was founded based on a piece of scientific serendipity. When I was in Oxford as a postdoc, we were looking for devices to do these quantum optical experiments. One of our hires in the group was someone from the University of Twente in the Netherlands, who said, “I know how to get our hands on these devices, because there is a group there that is good at integrated optics and that is working with these things.”

We organized that we could borrow one of these devices from the University of Twente, take it to Oxford and measure it. When we did that, we realized that we had something. When I came back to the Netherlands about a year later, I found the people who had fabricated this device and I said, “We have to start exploiting this technology of silicon nitride optical waveguides. We have to start exploiting this for quantum technologies.”

After some discussions, which took about another half year, we ended up founding QuiX Quantum in January 2019. The company, as this goes, was initially very small — started with a handful of people in the summer of 2020. About a year later, we secured a seed round with a few investors. Actually, the last hand that I shook before the pandemic was of the person who would end up becoming the lead investor, so that was an interesting way to go into lockdown. Literally, the last in-person meeting that I had was this kind of meeting.

We had that over the summer of 2020. We had that seed round, and we then grew. We’re now at about a dozen people. Then, we also have recently realized hardware sales. Something that I’m very proud of is — a lot of what quantum computing is, of course, is about selling a promise of the future, but something that QuiX is doing where I would say it’s ahead of the pack is that it’s already generating hardware sales. We’ve sold half a dozen of these quantum photonic processors to both academics and to other quantum startups worldwide. This is also why I say we are the market leader in photonic quantum computing hardware, because there are not very many companies that can say that they have sold half a dozen quantum processor units.

Your first processor came out quickly. It came out within a year of the company founding. I’ve heard it called record-breaking. Can you elaborate on what led to that moniker for that first processor?

First, I want to say something about why we were able to move so fast. We, at QuiX, are organized as a fabulous company. We tried to outsource as much of the fabrication as we can. We took on board the fabrication capabilities of the company that had originally designed this processor that went to Oxford — the one that was done as an academic collaboration. We simply iterated further on that technology.

A lot of what we needed to do to be successful was already there, particularly in the low-level components that we needed. A lot of the groundwork that you need in order to make a good quantum photonic processor is to have low optical-loss, optical waveguides. Our information carriers in our system are photons, and these photons, they want to move at the speed of light. You need to provide some channels for them to move over, and these channels are what are called waveguides. The thing that you don’t want to happen is that these photons are lost to your computation as they move over these waveguides.

The thing that made us realize that someone should get started on building quantum photonic processors out of silicon nitride is when we computed the component density that you can achieve in silicon nitride quantum photonic processors. It turns out that it’s not so much about the optical losses per centimeter that you can specify. If you talk to an integrated-optics person, they will say, “I can make waveguides, and they have losses of so and so many decibels per centimeter.”

It turns out that that’s not the important number. The important number is losses per component. One of the reasons why silicon nitride is good is because it sits at the sweet spot where you can both build relatively compact components but still have relatively low loss. There are other materials out there that have smaller components, but then the losses are so much higher that it still doesn’t work out, and there are other materials that have lower losses, but then the components are huge, so it still doesn’t work out. Silicon nitride is precisely at that sweet spot, and that’s what we realized in 2018. That’s why we said, “We have to do something about this.” That’s a little bit about why we were able to go fast.

It was that initial machine. That one that you borrowed you essentially improved upon.

If you say it like that, that’s maybe not giving enough credit to our engineers, because by now, there is a lot more cleverness in this device than there was in the original academic one. What’s maybe also worth saying is, the metrics for these kinds of devices are the number of optical channels that you have, and then the end-to-end losses. Those are, more or less, the two things that you care about.

Because we have this very good underlying technology, we were able to achieve what was at the time the largest processor in the world, with 12 channels — we call these channels qumodes; that’s something that we can go into later — then, with also the lowest losses in the world. Then, we put that out there in a webinar. If you go onto our YouTube channel, you can look up the webinar from 2020 where we present this 12-mode processor. Then, earlier this year, we improved on that further and we put out a 20-mode processor, and that processor has about 2.9 dB of optical loss. That’s about 50% of optical transmission, if you don’t speak photonics engineer. It has 20 channels, and this is the largest number of channels reported in the scientific literature. This is something that we are very proud of.

That’s exactly what I was going to ask you about next— the newest one having 20 qumodes. Did you want to take a step back and explain to listeners the difference between qumodes and qubits?

A qumode is, essentially, for linear optics, the analogon of a qubit. Linear optics is a restricted model of quantum computation where you do not have access to all of the operations that you have access to in a gate-based model. Of course, you have the big advantage that linear optics is a relatively straightforward thing to implement in an experiment. If you are aware of the recent quantum advantage demonstrations that have happened at the USTC in Shanghai, then this is also linear optics. It’s that model of computation that this QuiX Quantum photonic processor ultimately slots into.

The biggest criticism of that Chinese experiment was that it wasn’t solving a real problem, which is a theme that comes up in quantum-advantage experimentation. What types of problems can you apply your processor to?

Well, let me say something first about the point of quantum-advantage experiments not solving a real problem, because I feel that there has been a little bit of moving of the goalposts by some people in the community. This is not something that’s even specific to optics. After the 2019 Google quantum-advantage paper came out, there was similar criticism where people said, “This doesn’t actually do anything useful.” The countercriticism to that is very simple: “No one ever claimed that it did.” I don’t agree with the people who have redefined the word “quantum advantage” to mean useful quantum computation, because quantum advantage has a very clear-cut computational-complexity definition. I don’t see why we should be muddling with that. Of course, obviously, once you have achieved this, the next question is, “What is it good for?”

I’m a hardware guy. I ultimately make quantum photonic hardware. The question of what you can do with linear optics has not nearly received the amount of attention from theorists and from computer scientists that it should have received, but there are good ideas out there. For example, it turns out that these linear optical systems map onto the structure of graphs. It goes maybe a little bit too far to discuss exactly how this works, but there are certain problems within graphs where you’re looking for certain structures in these graphs or you’re looking to identify whether a graph has a particular hidden structure where you could do this with such a device and you could not do it with a classical device.

This maybe sounds abstract, but if you consider that any relationship between things, you can represent as a graph, whether that’s social networks or COVID-19 transmission, or even things like for protein folding, you even care about graph structures. There are lots of problems there. Then, other things that one can consider are machine learning applications, cryptographic applications, and there are a whole bunch of these kinds of things. One of the reasons we decided to start selling devices early on is for a very simple reason, which is that we wanted to unleash the creativity of the community onto this as a problem.

Another thing is that we, as QuiX, are not restricting ourselves to these devices. This is, historically, what we have been doing, and it’s something that we are very good at. We are building the best quantum photonic processors out there. We are also building up toward universal quantum computation using photonics, which is also a thing that is possible. You can map linear optics to the familiar gate-based model of quantum computation. Essentially, the price that you pay for this is overhead. What is nice about these experiments that, for example, have happened at USTC is that every single photon is a computational unit in the system, is directly adding to the complexity of the problem. That also makes it very implementable because you need “only” a hundred photons in order to get to a quantum advantage, and if you work hard enough on that problem, that turns out to be a solvable problem.

Yes. You’re not that far off.

What we care about, particularly at QuiX, is, we care about the programmability of this optical transformation. If I compare the larger systems that are out there, they have a more or less static optical transformation, or the opportunity to vary some phases but not the opportunity to completely control the transmission matrix of your optical system, which is, ultimately, what encodes the computation that you are doing.

With this kind of machine – and you sell it in multiple forms; you can get the core processor and then a complete setup with everything you would need to get it running — how would people interact with this at the current stage it’s in? You have six customers already. How do they then interact? Is there some kind of stack that still has to be created for them to manipulate, to program, etc.? What does that look like?

The stack is quite mature. We don’t just sell the processor as a chip, but we sell it as a very high-TRL device. The processor comes packaged in a control box that has all the electronics in it, temperature control, and so on and comes with a computer program that knows about all of the calibrations on chip. You, as a user, provide the photon sources. If we’re just talking about the processor, you provide the photon sources and the detectors. It’s a computer that has a program running on it where you type in the optical transformation that you want to do. You press Enter, and the system does it for you. It knows about all the calibrations. It knows about how to interact with all of the various subcomponents on the chip. It does all of that for you. In that sense, it’s a very plug-and-play system.

A lot of listeners are probably used to the concept of either annealing or gates, one or the other — gate-based or annealing. What does it look like to program this? How can these transformations be mapped to some kind of problem? Does it have any set language or gate system that it applies already?

We’re not yet at the level where we have this system in the cloud with some kind of interface where you can talk to it directly, remotely. If we had something like that, then, you can immediately point to it and say, “This is the thing — this is how it works,” but this is something that we are very much working on.” The goal is to have this in the cloud with all the peripherals in 2023 — so, next year.

To make a rough sketch of what such a system would look like, you can imagine that you can provide something about the input stage. You can say where there will be photons and where not. You can say what the optical transformation looks like, and then you simply get a certain number of shots of the experiment. You press the button and our hardware starts running, records the outputs of this system, puts them in a text file and sends that to you. That’s what that would then look like.

Similar to other approaches. Next year, we could imagine, then, this will be another photonic option available on the cloud. Customers who buy it now, do they then run some special program to talk to that other program that understands the calibrations, and things like that? Is there a local programming interface they use for the lab?

Ultimately, you’d have to ask our customers.

Today, you’re developing that one?

No. I mean, as far as they have told us, because of course, if someone buys a device from us and they are not happy telling us what they’re going to do with it, that’s also fine with us, ultimately. It’s a commercial business. It’s a finished product. We trust our product well enough to send it out there and have people work with it.

Of course, we do have customers who do provide feedback. A lot of the customers that we have so far are people who are domain experts in quantum photonics. They have some ideas about what they want to do with this thing, and they understand what is going on well enough to talk. We have some API where you can have your software running and it talks to our software. You can integrate our processor into the workflow of your system, much as you would any piece of laboratory electronics.

That’s the dream with the cloud too — integrating it in like you would select another GPU, or whatever. Now, you want to select another QPU to do something extra.

Exactly. That’s ultimately the dream of the cloud. That’s absolutely true.

You hinted at some areas where this would already excel. I wanted to ask you about if this machine is going to be, one day, the best at something, if you could predict. For example, we know annealers are always going to be probably the best at optimization, and then gate-base would be different use cases. Would you say there’s a chance that no matter what happens in the future, this type of machine will be the best at something?

Definitely. I want to say, in general, that the question of where NISQ quantum computing is going is, obviously, something that a lot of people are busy with. We are building these devices, but we are not just building these devices for the sake of building these devices. They are also a stepping-stone toward universal quantum computation, and they are a demonstration of the strength of the underlying technology. There are various ways that you can look at this.

You imagine the ability to continue scaling and adding qumodes to one day have enough to be like a universal gate-based computer like this, or do you imagine changing the approach?

That’s not quite how it works. If you want to go to universal quantum computation, you have to add additional capabilities. It’s not enough to have passive linear interference on a chip. This is something where we are evaluating components. We’re looking — and, obviously, this is something that I cannot say very much more about than this, but this is where we are exploring the space.

That’s what I was asking, because I was trying to get a sense of what kind of plans you have for your road map. Are you thinking along the lines of making some kind of modifications? If I were to ask you, “Three years from now, what would you expect to see?” Would you expect to see something more along the lines of universal gate-based?

The road map is to have the current nonuniversal system in the cloud by 2023 and then to have a universal system that maps onto gate-based by 2026. That’s more or less what the road map looks like. Then, at the same time, keep building these nonuniversal systems and make them bigger. By the end of this calendar year, we want to have out there a 50-qumode system and then push that toward an actual demonstration of a quantum advantage using our own hardware. That’s the plan, in that direction.

For that quantum advantage, would it be in the sense you described before — it proves that it’s better at something, or it wouldn’t be trying to worry about the critics and the whole, like, is this a usable application, or something?

Of course, it has to go to a useful application, but one also has to understand that these things go step by step. Right now, we are sitting on some of the most advanced photonic hardware in the world. We want to build that out into useful applications. Having a quantum advantage in this complexity sense of the word is a stepping-stone toward that. While we are doing these things in the lab, we are, at the same time, talking to our partners on the software side to try to build out the software stack on that side. We’re not sitting still there either.

Where you’re located right now, it’s a region in the Netherlands that’s considered a photonics ecosystem. There’s a lot around you in that space. Are you constantly, then, looking to recruit in that area and bring in fresh ideas? I was wondering how that impacts the day-to-day, like being surrounded by all that.

Well, in fact, being in this hot spot of photonic technology development is crucial for us. Something that we have going at the moment, for example, is an applied-research project, with another company called PHIX, which makes, essentially, packaging for photonics. They are very good at the questions of, how do you get lights on and off chip, how do you get electronic signals on and off chip? As these systems get larger and larger, this is a question that becomes absolutely vital, because at some point, it’s not so much a question anymore of “Can you actually get the processing done on chip?” It becomes a question of “Can you get the chip to talk to the outside world?” It turns out that that is maybe even the harder question at some point.

This is, of course, a theme that you see in many quantum computing technology platforms. If you talk to the superconducting guys, then, at some point, they also say, “One of our big problems is actually getting so and so many cables down to a fridge.” This is not something that is even unique to photonics. What is great for us is that these people from PHIX are located one building away. Then, the university with which we collaborate — for example, the photon source that we use to do the calibration of this 12-mode processor and this 20-mode processor is located at the university, which is a five minutes’ walk away.

This ecosystem approach has been absolutely crucial to the success of the company, and it’s also been crucial to the fact that we have been able to move so fast. We essentially function as an integrator within this ecosystem for all of these excellent pieces of technology that are out there. We put them all together with adding our own expertise on quantum.

For all you know, some problem you’re trying to solve, someone might solve it right down the block without even knowing you were worried about it.

Precisely right. It’s also a cost issue. If you consider that none of this machinery is cheap, it makes no sense — we are what someone who comes from the semiconductor industry would consider a very low production volume operation. We make, let’s say, a couple of dozen chips a year — that order of magnitude. It makes no sense to buy a multimillion-euro machine to attach fibers to these chips and then use it 25 times a year. It’s great if there’s someone down the road who has that machine and who has it running all the time and who you can commercially get a couple of dozen of chips a year, get fibers attached to that. That’s why this ecosystem works so well.

That’s great. You know how things like this go: Some synergies will create some wonderful surprise one day, and the next great revolution begins.

Precisely.

Hotbeds of thinking, like going back to something like the Solvay Conference of Physics. You never know. You get minds together, and magic happens. Jelmer, thank you so much for joining me. I appreciate it, and I look forward to seeing how that road map progresses.

Thank you.

Now, it’s time for Coherence, the quantum executive summary, where I take a moment to highlight some of the business impacts we discussed today in case things got too nerdy at times. Let’s recap.

QuiX Quantum is a three-year-old startup focused on photonic quantum computing. They’ve already sold half a dozen processors to academics and companies. A fabless company, QuiX outsources the fabrication to a company that already laid groundwork in silicon nitride waveguide technology for similar machines in academia. These optical channels are known as qumodes. QuiX has 22 qumodes in its latest machine.

Qumodes and linear optics are analogous to qubits. The linear-optics approach is similar to that used in the Chinese machine Jiuzhang, which caused some controversy with its claims of quantum advantage. Should quantum advantage be declared only if a usable problem is solved more quickly by a quantum computer than a classical one? Dr. Renema feels that this is a moving of the goalposts: No one claimed a useful problem was being solved, only that quantum advantage at a task was being demonstrated.

Linear optics do not use gates but can perform useful tasks. They map onto the structures of graphs well, so use cases could include finding hidden structures in graphs, solving problems in social network data and research in folding proteins. Even machine learning and cryptographic use cases are possible. QuiX is hoping that the devices they provide will spur creativity. Their customers are domain experts, after all.

QuiX believes that one day, linear optics will be used in gate-based quantum computing too, and they’re working on a universal gate-based system with added components for release in 2026. Programming the current machines is enabled by a mature software stack loaded into a control box. The software interacts with all the subcomponents on the processor and performs the necessary calibrations to enable computation. There’s no cloud stack layer yet for programming, but they’re working on an interface like that for 2023. Before that, by the end of 2022, they’re hoping to have a 50-qumode system that may demonstrate quantum advantage.

That does it for this episode. Thanks to Dr. Jelmer Renema for joining to discuss QuiX Quantum and their photonic quantum computer. And thank you for listening. If you enjoyed the show, please subscribe to Protiviti’s The Post-Quantum World, and leave a review to help others find us. Be sure to follow me on Twitter and Instagram @KonstantHacker. You’ll find links there to what we’re doing in Quantum Computing Services at Protiviti. You can also DM me questions or suggestions for what you’d like to hear on the show, or maybe share your thoughts on quantum advantage. For more information on our quantum services, check out Protiviti.com, or follow ProtivitiTech on Twitter and LinkedIn. Until next time, be kind, and stay quantum curious.