Transcript | Bringing Interconnect to Trapped Ions with IonQ

We’ve been talking for quite a while about how interconnect, or entangling qubits between quantum computers, is the key to scaling and building fault-tolerant systems. IonQ recently doubled down on this approach by acquiring a company that specialises in the technology. Find out how trapped ions will soon be talking to each other across rooms or even greater distances 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 a repeat guest. He’s now senior staff engineer at IonQ, Aharon Broduch. Welcome back to the show.

It’s great to be back.

Yeah, when you were last on, you were with Entangled Networks. But in January, your company was acquired by IonQ — so congratulations are in order. The news coverage made the deal sound like a continuation of your work. Do you want to talk about that?

This is us continuing to do the same stuff we were doing before, except we’re more focused now. When we talked last, I said our mission statement, when Ilya and I started Entangled Networks, was to help build larger quantum computers — help scale up. That’s what every quantum computing company is trying to do. Our way to do this was to go modular, to build quantum interconnects. And we didn’t invent this idea. In fact, the pioneers were IonQ’s founders, Chris and Jung Sang. Continuing with IonQ was very natural. IonQ was also at the top of our ideal customer list whenever we were pitching.

The conversation was, “You guys very clearly want to go to a modular solution. We’re offering this stuff.” And that was, for us, extremely scary at first, because we’re going to sell technology to the people who basically invented this technology.

We waited until we built enough technology to impress. We started the journey with IonQ as potential customers. We must have done a very good job, because at some point they, instead of wanting to buy the product, they wanted to buy the entire operation. And that was incredible for us not only in terms of us continuing to do what we want with more resources, but also in terms of doing it in a culture that we absolutely love. And what was interesting about the conversations with IonQ was that IonQ is customer-focused. The idea is to deliver products that will help the customers get computational advantage.

And the question about interconnects had always been not if we need interconnects — that’s obvious — but what is the right timing? And what we had done for, in particular, the last year at Entangled Networks, but even to some extent right from the beginning, was build this entire stack of software that tried to answer this question, because when you’re trying to sell an interconnect, the first thing you want to tell the customer is, “This is when you’re going to need these interconnects.” And so now, you’ve got to start planning this much time ahead in order to have this in your device. And this is the advantage you’re going to get if you start today. This really struck a chord with IonQ, so we merged into this culture of very engineering-based decision-making and trying to scale.

It’s clear that IonQ is a company that gets the whole stack, from the software to the hardware, and the idea of bringing these machines together to make a bigger, more powerful one. It does seem like a very natural fit. That’s pretty exciting news.

It looks like this acquisition is helping launch IonQ Canada in Toronto. What does that look like on the ground? Are there new team members? Maybe you have access to machines now in the area.

This is a big thing for IonQ, having an office in Canada. It’s definitely a big thing for us, having access to talent across the border. There’s a ton of talent at IonQ. But Canada is an incredible quantum ecosystem. I arrived in Canada because I wanted to do a postdoc at the biggest quantum computing institute in the world.

There’s a lot of stuff happening in Canada in general, and in southern Ontario and Toronto specifically and further afield in Quebec, in Alberta, in B.C. This is an opportunity for IonQ to tap into a lot of talent and a lot of knowledge that exists in Canada in this fantastic ecosystem.

Are you going to be able to be experimenting with the machine more directly now? Will there be one physically located in Canada for you?

At the moment, there are no plans to have any hardware in Canada. IonQ has just come up with a big announcement about having a new facility in Seattle. And so the focus on where we’re going to build more hardware is in Seattle. But who knows — things can change.

That’s close enough — it’s just over the border, practically. How has this affected your research? Were there any roadblocks lifted by being part of such a big manufacturer of hardware that made this easier?

It’s definitely removed a lot of the uncertainty around what we were doing. We’re building a device that connects into a quantum computer. Now, if you think about the usual uncertainty that startups have, we had the bonus of needing to connect to a machine where there is no standard, there’s no USB cable you can connect to, there’s no established software stack — everything is very specific. This is like the early days of computers: You have to build everything very specific to the machine. And at Entangled Networks, as a business, we could not afford to focus on one customer, because it’s too dangerous, so we had to be fairly broad. We designed things that had broad applicability,

We were reaching the end of our ability to do that, especially with the optimisation tools we were building. At some point, you have to optimise for one specific machine — and the uncertainty about not only what the machine does now but also about what’s going to happen a year from now, what’s going to happen three years from now, what’s going to happen five years from now.

And having access to a lot of specific details, we were struggling to figure those things out and find ways to work around those issues. We also now have a lot of access to experts, the best people in ion traps, and the people who built the first experiments on interconnects for ion traps, they’re a phone call away or a Zoom call away. It’s been incredible — we’ve been able to accelerate development.

When we talked last time, it was the idea was that you pick one technology and connect that — transmon to transmon, trapped-ion to trapped-ion, etc. So now, you’re able to focus, I’m assuming, on trapped-ion, specifically because of this approach.

What happened in terms of Entangled Networks was, when we started, we said, “We’ll try to be pretty broad.” And then we had a lot of advice from very smart business people who said, “You should focus on one thing — try to solve one problem,” and so we zoomed in on ions because that was the most realistic technology for interconnects.

And then we felt “Oh, we’re too constrained, because there are, like, three companies that we can talk to.” So we started broadening out and thinking about neutral items and maybe transmons and other things. It’s nice to have that focus.

Right now, is the approach to a perfect interconnect, is it to get that in the same room first and then maybe branch out? When you’re talking about connecting these machines, ideally, you want them there, not over some extra layer of network.

What we want to do initially is to build a network of quantum computers that are close by. We don’t want to start worrying about photon loss in fiber.

Maybe it’s good to say a couple of words about how interconnects work, specifically with ion traps. With ion traps, you want to create entanglement between two ions in different machines. The way that’s done is by having each one of these ions emit a single photon, and then do some fancy measurement on those photons. Now, those photons, in order to get to the measurement device, they have to go through a fiber. And the hardest thing, specifically with ions, is getting that photon that was emitted by the ion into a fiber.

It’s going in all different directions, and you have the lens in one specific point. You want that to be your only worry — how do I get that into fiber? If we start going long-distance, now the photon can also get lost in the fiber. We need to do that one at a time. Also, generally, for computation, you want things to be as fast as possible. Even when you build a supercomputer, you don’t want to put half it Tennessee and half of it in some other place.

Are there additional concerns with hardware and software for long distance? It becomes about those physical blocks. It also becomes about, there has to be a whole other software layer protocol. If you’re dealing with any appreciable distance, you would need repeaters and things like that.

It’s a different story when you’re going long-distance. It’s also different use cases, but let’s start with the hardware constraints. You have to go long-distance — you have to start worrying about loss of photons through fiber if you’re going through fiber. If you’re going in free space, apart from losing photons, you also have to worry about that giant yellow ball that’s emitting a ton of photons that are just going to kill you in terms of noise. But if you focus on loss, you just have to get the photons. The photons are carrying entanglement from one point to the other, and you want them to arrive safely at their destination.

With standard photons, you don’t call it photons — you call it light, because there are billions of them. And you shoot it through a fiber, and then you put an amplifier every few kilometers or every few hundred kilometers, and you amplify the signal. And that works great classically. In quantum, you can’t do that — you cannot amplify the signal. And so you have to use other tricks, which are quantum repeaters. What that creates is another technological barrier. In fact, ion-trapped quantum computers make great quantum repeaters. But now you have to deal with putting a quantum computer at each node.

The other big problem is converting between frequencies. If you want to go long-distance, there are two best ways to go around it: One is to work in fiber, and then you have to use telecom frequencies. And ions don’t work in telecom frequencies. The other way is to work in atmospheric windows if you’re going free space, but, again, with free space, there are a bunch of other problems. And so, generally you’re dealing with two other problems: converting the light and then having repeaters.

By having this new only-local networking to make these machines work as one close by, have you already seen some kind of improvement because you’re focusing on one technology? Has that liberation process of focusing only on trapped-ion made it more effective more quickly?

It’s not just about looking at trapped ions. It’s about looking at one specific type of ion. It’s about knowing what the trap looks like, knowing what the size of the window is on that trap, so we know what kind of lens goes in there, and we can start having much better estimates of how many photons we’re going to get. Understanding all the different technologies that we can come up with, can we move ions around? Is that a legal operation? Is that within the roadmap? How far are the ions spaced from each other because of design constraints? Like I said, there’s no USB, and so now it’s not only that we can mold ourselves to the parameters we can also change the parameters to fit us. That changes everything.

It’s funny how sometimes, constraints can be liberating. If you have to narrow the focus, you get things done better. It applies here. I suspect that something like that might happen.

Obviously, the goal here is to one day have an even bigger IonQ machine, because they’re all going work together, in a sense. How does being part of IonQ affect plans for Q-Link? That was going to be in general use to one day connect different types of systems over distances, if I’m correct. Is that affecting any plans for that? Does IonQ envision experimenting with other types of quantum computers too in the future?

Right now, IonQ is very focused on doing ion traps and expanding to a certain size. But if we think, again, what does Q-Link, the interconnect that we designed, do? It collects photons into fiber, and then it fuses them together in order to swap the entanglement from the photons into the qubits — in this case, ions. And Q-Link had always been designed to work in optical frequencies. It would never have worked off-the-shelf for things like transmon qubits, which work at microwave frequencies.

Our dream, even at Entangled Networks, had always been that someone will come up with something that can convert from microwave to optical, and that’s an insanely hard challenge. We were asked repeatedly, “Will you be doing that?” We didn’t have the skill set to do that conversion. It’s very difficult. But once that happens, it will unlock a lot of things for industry. A lot of companies will be rethinking the next generations of quantum computers, because suddenly, you can use the advantages of different technologies to go from one place to the other.

The first step in seeing conversion between different technologies is going to be between ions and photonic quantum computers, because the relative shift in frequencies is much easier — you don’t have to go all the way to microwave. And there is a ton that can be done in that direction. Already now, we know of use cases where that can be interesting.

If there is one piece that’s missing right now in industry — and it would be fantastic if someone could build it, and there are a few companies doing that — it is converting between frequencies and single photons: taking an optical photon and converting it, say, from blue to infrared.

I’m sure one of those things is perfected — I’m going to have them on this show too, so we’ll hear from that. Let’s dig a little into the technical specifics here. How has performance evolved since we spoke? I remember you felt that one-kilohertz speeds of interconnect would be possible with trapped ions. Are you still seeing an order-of-magnitude performance hit for a two-qubit gate if it’s done over an interconnect?

What we saw when we were doing our initial analysis at Entangled Networks was that we can live with about 1/1,000 ratios between entanglement generation rates and T2. That allows us to very fundamentally run some circuits. We also looked at what happens with the ratios between how fast a two-qubit gate is and how fast it takes to build up entanglement and use that.

What we’ve done on top of that into the next phase is try to understand how everything works together when you start accounting for noise. And in the first analysis, the only noise we accounted for was T2 — natural decoherence. But once you try to go a bit deeper, you need to think about gate noise. And we discovered something that is not surprising, but it’s nice to see it in actual simulation — the thing that hurts you the most is actually the fidelity of the two-qubit gates. When you use an interconnect, you have the entanglement that you created, but now you need to somehow use that to do things like teleportation, and that costs a number of gates, so you need to add a few more gates to your system. And it turns out that those additional gates, with today’s performance, they’re a big hit. You don’t want to add more gates to your system.

Now there is this multifaceted challenge. Again, this is not unexpected, but you want to improve interconnect rates, you want to improve two-qubit gates, you want to know which one is going to be most in your favor in terms of getting overall performance. And that’s a lot of the analysis we’re doing — we’re trying to figure out which one is going to pay off more. Where should the focus be in terms of development.

In terms of the one kilohertz, between 500 hertz and two kilohertz, those are the boundaries for getting two single ions to get entangled using fairly standard technology that has been demonstrated, but pushing it to the edge of engineering. Beyond that, you need to start multiplexing — that’s the easiest thing — or going into very different types of technology. When you think about rates, you need to think about the system as a whole. It’s not just about how fast you’re going to go, it’s about how fast should you go, in order to maximise performance, and what are the trade-offs? How big is the change? Every single piece comes into play here. Again, this is a lot of clarity now that we’re in the system and understand what’s going on there.

I’m assuming that the general plan is to get an optimal number of qubits on a trapped-ion system one day. We’ll figure out what that is — I don’t think anyone knows yet what that is —and then connect these machines together. Is there going to be a technical limitation to how many machines we can connect? Do you think there might come a point where there’s an overhead or something that’s introduced? I’m not sure if it’s possible to know this yet.

Physicists — and I’m a physicist by training, we always make very ideal assumptions about the world. And then you try to build a system and you realise that there are all these small engineering constraints that are going to mess with you. With the kind of numbers we’re seeing for at least the next few years, we’re not expecting any major blocks in terms of how many devices we can connect, apart from needing a switch that will allow us to route the photons where we want them to go. But once we get to larger sizes, we’re going to have to start worrying about bandwidth, and how close together we can pack those computers, because as soon as they start getting further apart, it’s going to be a problem. And how many switches do we need? And when they build supercomputers, when they build Frontier, the No. 1 constraint is power. With quantum computers, we know power requirements are not going to be as high as classical computers. But we are going to see these new requirements pop up. I think there’s going to be surprises when we get to a lot of qubits.

It seems like something that’s going to have to just come up one day. It’s hard to figure from now.

What kinds of software advances did your team bring to IonQ? It sounded like, from the announcement, they were talking about some kind of boost to the stack as well that Entangled Networks brings.

A lot of what we did in Entangled Networks was software work. Again, we were trying to answer the questions “When are interconnects needed? How fast do they need to be?” Because you’re trying to build an interconnect —the first thing you need to know is, where are you aiming? What is the minimal viable product? And what are the next steps? And the first thing we built was a compiler, which works incredibly well for taking a programme and distributing it into multiple systems. That compiler was meant to work on a multicore quantum machine. It also has some capabilities that are applicable on single-core machines. So that adds value very quickly.

But then, we also built a ton of other tools that haven’t been made public. Those are analysis tools that let you figure out how well your system is going to perform under different constraints. Again, the question is, how many qubits do you need per module? How many modules is optimal? How do you get more performance out of your machine in the best way? Get these architectural considerations. They’re huge. There are so many questions we didn’t see anyone answering. And so we just had to build the software to answer them.

Your training then — obviously, you’re a physicist, and you go into engineering and software. We talk a lot on this show about paths people can take to becoming a quantum coder. It seems like making use cases work is the easiest path to entry for quantum. But what about networking? What kind of development path would you recommend someone who’s interested in this field follow if they want to work in it?

The bar for networking is definitely higher than for coding. You need some experience working in something that is related. But one thing that I feel is always strange, somehow, going into the software implications side of quantum has become popular in some cases at the expense of other areas. I talk to a lot of people who asked me for advice about what they should do. And if I see someone who has this incredible CV and has expertise in a certain subfield of quantum, I’m thinking this person would be perfect to fix this problem that I have because there are, like, five experts in the world, and he’s one of them. And then they say, “We want to go into coding. Well, it seems like this is where all the jobs are.” And it’s like, “No, you’re an expert in this field.”

People should sit and think what they can offer in quantum communication. Anyone with experience in classical communication is interesting. Anyone with experience in fundamental quantum computing — if you’ve built a piece of a quantum computer, whether it’s the ions, you’ve fabricated devices, you’ve created pulses for quantum computers — there are so many different small things that you’ve done and you’re an expert on. These experts don’t realise how much value they can add if they just find the right niche.

One thing that happens to a lot of people who are doing their Ph.Ds. is, you look around you, and you’re the stupidest person in the room, because everyone else has 10 years of experience, and you only have five years of experience. But you have five years of experience — more than 99.999% of the population. People should assess their own skills and figure out, “What do I know that someone else doesn’t, and what do I love that someone else might not be an expert on,” and try to find yourself in that niche rather than “How can I modify my own skills in order to fit something that’s not so interesting or not really something that I want to work on.”

I appreciate the insights in that answer, because a lot of people are thinking about coding as the only way in. But this is a full stack, from hardware up to many layers of software. There are a lot of unique skills that can help it all along.

And we need engineers. Quantum is just full of physicists and theoretical computer scientists. And we’re good at what we do, but to build a machine, you need engineers. You need people who have had that insight, people who’ve had experience, people who’ve done project management. There are so many ways to go beyond just writing code where you can fit in.

What’s the No. 1 thing you hear? After someone comes up with a scientific idea, they then say, “Now it’s an engineering problem.” That’s almost always said when they’re getting ready to roll something out. It does take all types to get this industry to move forward.

With that, I’ll thank you again for coming back to share some updates with us. Good luck in connecting these already-amazing machines to make even more amazing ones in the future.

Thanks very much for hosting me.

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.

IonQ’s founders published a paper about interconnect in 2011, so it’s safe to say the idea of connecting modular trapped-ion systems was always in the DNA of the company. As a result, it makes sense that the quantum computing giant would acquire Entangled Networks, which specialises in interconnect. The goal is clear: to build large-scale quantum computers by enabling computation across multiple distributed quantum processors.

The acquisition creates IonQ Canada, and gives Aharon and his team resources to continue and advance their work on this future-changing technology. In addition to being able to call up the creators of interconnect, the team can now have access to the deepest technical details of IonQ’s trapped-ion systems. This allows for levels of tweaking and experimentation that were not possible before. They’re focused on entanglement via photons (pun intended), as they seek to perfect the process of having trapped-ion qubits emit single photons that need to be sent by fiber and measured. This will be short-distance entanglement and networking for interconnect at first — IonQ systems in the same room, for instance.

There are lots of overhead considerations to figure out: How many trapped ions will be needed to make successful gates between systems? How many systems can be connected? The team is also working on other levels of the stack, having built a compiler that will make decisions about how to split a programme among multiple quantum cores.

That does it for this episode. Thanks to Aharon Brodutch for joining to discuss quantum interconnect at IonQ. 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 all socials @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. 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.