Chris Ballance, co-founder and CEO of Oxford Ionics, is interviewed by Yuval Boger. Chris describes their unique trapped ion quantum computers using integrated electronics for scalability and low error rates. They cover development plans, including plans for scaling up to 50-100 qubit systems and the development roadmap towards a 256-qubit device. Chris and Yuval also discuss error correction, market strategy, misconceptions about trapped ions, and much more.
Full Transcript
Yuval Boger: Hello Chris, and thank you for joining me today.
Chris Ballance: Great to be here.
Yuval: So, who are you, and what do you do?
Chris: So I’m Chris Ballance. I’m the co-founder and CEO of Oxford Ionics, and we build the world’s best quantum computers.
Yuval: And this is using trapped ions, I assume?
Chris: Yeah, that’s right: we use trapped ions, and we control them using electronics, so we can integrate all of this stuff into a semiconductor chip. This is really important since if you want to build large-scale systems, you pretty much have to be doing it in a semiconductor foundry. That’s the only way you can really build out millions of things well in today’s world. And then, if you build your qubits into the chip, you suddenly start having to do quantum fab, and your whole fabrication team gets very scared about this; you spend lots of your time fighting atomic defects, single-atom type stuff, which is very misaligned with what the semiconductor industry does. But, if you use these trapped ion systems, and you can trap these ions some tens of microns above the surface of the chip, your chip is suddenly classical. So, in system architecture, you have a purely classical chip that you can build using really standard techniques that don’t look too scary, you can use techniques from 2005, 2010, and then your quantum stuff is purely these individual atoms, which end up being pretty close to perfect.
Yuval: There are several trapped ion companies out there, one public, one private, maybe a couple others that I’m not recalling right now. How are you different?
Chris: It’s all about how you control the qubits. So, every other player out there controls qubits using lasers, and if you want to build a 10-qubit device, a 20-qubit device, a 30-qubit device, lasers look great; they get you there fast. But if you say, ‘I want to build a thousand qubit device, a ten thousand qubit device, and a hundred thousand qubit device,’ then you extrapolate back, the gradient of this looks a lot, lot better if you don’t use lasers. In fact, lasers are the worst things to integrate. If you speak to anyone about integrating wires into a chip that you can build at scale, they’re pretty relaxed about this. If you say, ‘Now, how do I implement hundreds of thousands, or millions, of laser beam delivery devices into a chip?’ that’s basically building a new field from scratch. So, our USP is how we control our qubits using electronics, which means we can get down to incredibly low errors in our technology. We’ve shown coherence times of tens of minutes; we’ve shown single-qubit gate errors well below one part per million, so that’s a fidelity of over 99.9999%; and we’ve shown the world’s best two-qubit gate fidelities as well.
Yuval: How far is this from being public?
Chris: We have systems online at the moment that are accessible by a private cloud; we’re not in any of the public clouds at the moment. The way we see it, when we get out to 50 to 100 qubits, you really start generating significant customer value if your error rates are low enough. But for these 10-qubit devices, the only reason you’d want to use them is to learn how to build bigger devices. So, the customers we work with are the people who really give us useful input on our small-scale, sub-50 qubit devices to make sure our roadmap is well aligned with the customer value, and we have a small but perfectly formed selection of customers doing that. We don’t really want to build our systems and spend effort getting them online, you know, in order to have people running their school homework on them. That doesn’t add too much business value to us right now.
Yuval: And just between us, how far are the 50 to 100 qubit systems from being online?
Chris: So, at the moment, you know, we have designs for our 256-qubit device class, and we’re building prototypes of that. Whenever you ask about roadmaps, it always depends on who you ask. If you ask the scientists, they’ll be very optimistic; if you ask the engineers, they’ll be very pessimistic. The first prototypes of these things, where, you know, doing integration tests and putting these things together, are expected towards the end of ’24, so, the end of next year. Then, the difference between having the first operation of these devices versus having them working with error rates of 10 to the minus 4 per 2-qubit gates and the full any-to-any connectivity, with errors this low, with a full working compiler toolchain — that’s a while after that.
Yuval: Do you believe in error correction, or is that not necessary in your mind, given the low error rates that you’re describing?
Chris: That’s a really interesting question, and the answer depends on what time horizon you’re talking about. You know, it’s very easy to target a market in 10 years’ time, or target a market in one year’s time, and paint yourself into a corner. We think a lot about this at Oxford Ionics, and how you go about choosing the right path for the technology that gets you to market value early. So, you’re building our technology while making money, making significant money, rather than building it all out, funded by venture capital, and that being our only option. While at the same time, you know, avoiding falling into the pitfall of only building NISQ devices, and then, when the era of NISQ ends, avoiding building a company that also ends with that path.
So, the way we think about it, we build out our roadmaps of building out to millions of qubits since, really, in the end, hundreds of thousands, or millions, of logical qubits is where any technology platform needs to get to in the long-term. And this requires error correction, you know. I think very few people are going to say you don’t need error correction to get out to millions of useful qubits, but when you wind back the marks and look at what the direction looks like from now, over the next two, three, five, ten years, what you say is, ‘Well actually, getting out to a few hundred qubits, if you get down to errors low enough, you can build systems with quantum volumes of a hundred, two hundred qubits that starts to look really quite valuable.’ It’s difficult to say whether the real revenue turns on at 156, or 256, or 318 qubits, but saying somewhere between 200 and 2000 qubits, if you get your error rates down low enough, the value really starts to turn on.
I think, pretty uniquely in the quantum space, we have hardware already with error rates low enough that we can scale out to a few hundred qubits without having to improve our errors, using a technology path that will get us into this regime of starting to get useful value, and at that point, the market completely changes. Since when you can start selling stuff that people need to have to solve their problems, to get the edge they need to make their markets, you’re suddenly building a very different company. You’re building out the next roadmap of your technology with meaningful revenue, and that just changes the market dynamics strategy.
Yuval: What do you think is the quantum chat GPT moment? I mean AI is a technology that’s been in the works for decades and then an overnight success, right? What does that look like for quantum?
Chris: As with anything, you know, you can only really tell in retrospect. And I think there will be a few moments — when we look ahead and then look back — that we’ll identify as the pivot point. Whether it was six months ago, a year ago, or two years ago, it’ll be clear when we see hardware platforms that scale in a boring and reliable way. When you see people really hitting their roadmap and building out devices that scale into the regime of value, you’ll then witness that ‘overnight success.’ People will be able to look back and chart how it happened. And the way it’s going to happen is by building systems that can be system engineered properly, where all the individual bits are there, and then people methodically build out the engineering team that delivers this.
And this is the difference, you know, between science and engineering. If you’re building a useful quantum computer, the science should be way behind you; it’s all system integration. The science of building quantum computers is relatively easy, and if you’re doing it right, it was easy five years ago. It’s all now system engineering — how you build out systems where you can scale, where you can add stuff to it, where you can make it bigger and bigger, and where you can manage the risks of this, you know, portfolio approach. And in particular, where there’s no individual risk that kills you. For each of the different things that could go wrong, you have multiple risk mitigation strategies. This is not sexy, but it’s how you build a technology company rather than a company that’s doing science and R&D. And the difference is between saying, ‘I can build one device that works on the best day ever with the best science team behind it,’ versus ‘Here is a device that’s been assembled by not the best technician, but by the worst technician, not using the best sample cherry-picked out of the bunch, but the worst sample that passed quality control, and assembled on the day after the Christmas party.’ And that still meets these specifications — that’s when we really know we’re onto a winner in quantum computing.
Yuval: What do you see as the disadvantages of trapped ions?
Chris: That’s a very interesting question. If you do a straw man poll — perhaps not, you know, the man on the street, but say, the man at a quantum computing conference — people will typically latch on to a couple of things. ‘Aren’t they expensive?’ or ‘Aren’t they hard to scale?’ or perhaps ‘Aren’t they slow?’ Those are not any of the things we worry about. Speed is not an issue. First of all, if you’re talking about speed, you’re already assuming you can build something. And building a slow, awesome quantum computer is far more valuable than not building one at all. If you could build it, it would be fast. But also, in quantum computing in general, the problems we care about are embarrassingly parallel. We can scale out the number of processors we use – if we’re taking a thousand samples of one circuit, we can run it over a thousand processors, each taking one sample and speeding up a thousand times. So, you have this state space of the speed of the individual processor versus the reproducibility and scalability: how much does it cost to build ten times as many or twenty times as many? And for trapped ions, that looks really favorable.
The second is the cost and complexity, and this is really interesting. I think this is an engineering culture, or science culture, from different backgrounds. Because if you look at the backgrounds of, say, people doing superconducting qubits, they typically come out of places that have been touched by the semiconductor industry. From that, what people see, and the kind of the institutional attitude, is that you build a dirty prototype that barely works once. They have seen it go from that to something beautiful, that works and scales, with just scaling up engineering teams, and pulling in engineering workforces that already exist. This is obvious if you look at where trapped ion communities typically come from, and the trapped ion pool of talent, it typically comes from atomic physics and metrology. There, you don’t really talk about doing it if you haven’t reproduced it, have an error budget, really understand how it all works and why it works, and have proven you can reproduce it. This ends up being just a massive difference in attitude between how you frame the same results, and how you present the same kind of progress. And I think, also historically, in trapped ion systems, people have been a bit too conservative in their approaches. They try to make marginal improvements in engineering and physics by making things 50% bigger or 30% bigger each generation, while not really solving the big problems that limit scaling.
Yuval: I watched a presentation from another trapped ion company and when they talked about scalability I think they showed interconnected processors. Is that part of your vision as well?
Chris: Yes, at some point, as with classical computers, networking systems is better than building one big system. This is what we learned from data centers and supercomputers between 1980 and 2000. They just kept building the Cray model of building bigger and bigger processors. Eventually, it turns out to be better to build smaller processors that are worse, and each of the individual elements of the system ends up looking worse, but you can actually build the damn thing and build it at scale, and this always beats building one monolith that’s perfect.
That said, with quantum computing, the real question is the market’s metric: when should the transition happen from building single individual processors to then networking things together? I think that’s where we disagree with a few of the other companies out there, and I say this as, you know, personally, back in my academic career, before I started Oxford Ionics, my team and I built the world’s best quantum network of trapped ion systems. So, we had two trapped ion qubits, two trapped ion systems, and we showed we could build an incredibly high rate, high fidelity networking between the two.
So, I think I can pretty confidently say I built the world’s best quantum network of quantum processors to date, and there’s a very good reason that, despite the team having this heritage, we’re actively not pursuing this on a near-term roadmap. And that’s because, by doing the same thing on individual chips, we know how to scale out our technology to at least 10,000 qubits on the chip the size of a thumbnail. Probably a hundred thousand doesn’t look too scary, and I’d say we’re 50-50 on whether we can get out to a million qubits on an individual chip versus needing to use networking.
So, at the moment, you know, in our technology development roadmap, we have two axes: one is how many nodes you have in your network, and the second axis is how big each of those nodes are, and we know that we have a good few generations of technology where it’s definitely, definitely much easier to make bigger one-node systems than to connect multiple nodes together. So, we have this in our back pocket; we’re putting quite a lot of thought into how we make sure our technology is compatible with this at scale, but we don’t need to do that until we’re well into the, you know, hundreds of thousands of qubit regime. So, we’re pretty happy about that. Because networking is hard.
Yuval: Do you envision your computers being deployed in data centers or do you envision primarily cloud access?
Chris: That’s a really good question and the other interesting question is always who is your customer you know who you actually cashing the checks for versus who are you just reaching with your marketing activities and I think an interesting part of the space to look at is say the Nvidia you know how does Nvidia make money what does Nvidia do with their AI products like how does that work and in practice, they build an ecosystem around it but a lot of their revenue and cash flow comes from selling big black boxes that go into a data center that powers the machine learning back ends and that’s where I envisage a lot of quantum computing ending up.
We have this computer people don’t really care where it’s going to be some people want it in-house most people just want to say I can log into my cloud provider of you know flavor of the month and get access to this and where we want to be as a as a company is selling B to Big B into these kind of service centers and not servicing the customers ourselves and the value there ends up being about how you build out these systems how you build them out reliably how you make them deployable but in the end the customer cares about having access to these systems and the question about whether you host them in your facilities or in the customer facilities or how you run the colocation of this that becomes a matter of market evolution and in the near term very much we can deploy more reliable higher performance systems if we keep them in our own roof you know co-located our own facilities we are already selling systems out into customers in the alpha versions where they can deploy them on site and the difference there is just the time taken to go between our technician team can keep them humming nicely versus they don’t need anyone touching them to keep them coming nicely that’s like a year 18 months technology lag so then it comes down to whenever you ask the customer do you prefer access to higher performance systems or systems you can have on your own property 90% of the customers say I just want the best system possible to access and I prefer to lease rather than buy and that just sits really naturally inside hosting them at our own facilities right now.
Yuval: Tell me a little bit about the company where are you based how large are you how? are you funded, when you started sort of the basics?
Chris: Awesome, so the misty, dark origin stories come from myself and my co-founder and CTO, Tom Harty. We’ve been working together since 2009 on trapped ion technology, in particular. And our thought there always was, you know, we see how we can build a 10-qubit device like this, but, you know, we don’t want to build 10-qubit devices. We want to build million-qubit devices, and we didn’t see how to do that. So we spent seven or eight years in academia, working together on how you go about controlling the errors, since error is something people don’t talk about enough in quantum computing. If you go and speak to any person running quantum algorithms and say, ‘What would you like most out of the quantum computer you’re gonna get for Christmas?’ they always say, ‘I want low errors.’ Never, ‘I want more qubits.’ It’s always, ‘I want lower errors.’ Lower errors are king.
We spent about ten years in academia, between us, working out how we can go about building systems that allow us not only to control the errors and set these world-leading error rates but also that we can integrate at chip scale and actually build out. And it got to a point where we were pretty frustrated that no one else in the world was pushing this technology forward, and ended up having dinner with Herman Hauser, who’s one of the founders of ARM. In the end, he said, ‘Yeah, this sounds really interesting, you know, why are you just not building a company?’ That was, I think, about in 2017, then in 2019, we pulled our thumbs out and started Oxford Ionics to do this and take this technology to market.
We’re currently a worldwide company. We have people all over the world, but all of our hardware is currently based in our offices and lab space just outside of Oxford. And we’re a little over 50 people at the moment, predominantly technical. We’re really focused on building out the technology that scales out hundreds, then thousands, then millions of qubits.
Yuval: professionally speaking what keeps you up at night?
Chris: Systems integration — if you have a scientist and you go speak to a science team, they’ll always tell you, ‘We’re pretty sure we know what to do, but give us another couple of years of research and development, and we’ll tell you the precise answer of that specification. We don’t want to give you an answer just yet, but give us a couple of years, and then we can give you a concrete answer.’ If you speak to an engineering team, they say, ‘When we get concrete specifications, give us a couple of years, and then we’ll deliver on them.’ I’m building out system integrations where you can have teams that work together with uncertain specifications and evolving specifications, and optimize this to have the whole thing delivered as fast as possible, and managing the uncertainties here. That is a people problem and an organization problem, and that is the hard problem in quantum computing.
I’ve said it before, I think in this interview, but I’ll say it again: Quantum computing is system integration. It’s not quantum physics. Quantum physics is the easy part. If you have atoms as qubits, or ions as qubits, they just work, you know. We’re demonstrating it at tens of minutes of coherence time. The only thing that goes wrong in the qubits is what you do wrong as the person building out the processor and building out these control fields. All of the control is purely classical, and all of it is, you know, you can tolerance it and get it down perfect.
The hard thing is building out systems architecture of devices that allow you to scale this so you do the same thing on your hundred qubit devices as you need to do on your thousand qubit devices, as you need to on your ten thousands of qubit devices, and you keep that very back aligned. And how you then relentlessly deliver on that, and build an organization that architects these, designs these, and builds these in a way where you’re focusing on the hardest problem that it takes to deliver these things, which quite often means you end up with a scheme that looks inelegant on a whiteboard, you know, if you’re thinking like a scientist. But it ends up looking elegant to the engineers who actually need to implement it, and put these system components together.
And having that systems engineering focus — how we go about building out this system, tearing it into its component parts, specifying those as early as possible, and then building them and delivering them, and working out what the longest poles in that process are, and then relentlessly pushing on the details that are going to limit us, and building out the foresight in the organization of what’s going to limit us rather than just hammering at the obvious problems — that’s what keeps me up at night.
Yuval: as we get to the end of our conversation, 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?
Chris: it has to be Feynman so one can debate whether he’s a quantum great but he really did come up with the first ideas of quantum computing or at least caused to have the first ideas of quantum computing by starting to think about quantum chemistry and is you know how how far how small can you go quantum quantum quantum lectures and I just think he’s a fascinating person I’d love to ask her about his experiences in the the Columbia investigation sorry to the Challenger investigation as well and the really interesting insights he had into the NASA NASA organizational failings at the time which seemed so delightfully crisp in his recollection of them but seemed to be missed by almost everyone at NASA for a very long time
Yuval: Chris, thank you so much for joining me today
Chris: Absolutely. It’s been a pleasure.