Peter Chapman, CEO of IonQ, is interviewed by Yuval Boger. Peter and Yuval discuss IonQ’s plans to scale their computers and deliver business value, the best application-development partners for quantum customers, IonQ’s new quantum manufacturing facility, his hiring philosophy, and much more.
Yuval Boger: Hello, Peter, and thank you for joining me today.
Peter Chapman: Hello, Yuval, thanks for having me.
Yuval: So, who are you, and what do you do?
Peter: My name’s Peter Chapman. I’m the president and CEO of IonQ. A little bit, maybe, of my background. I’ve been in the high-tech business for, I guess, 45 years or so now. Started in the mid-’70s. The very first computer I learned to program on was a PDP-11. It has been an interesting journey going from PDP-11 to today’s quantum computers in that timeframe.
Yuval: Absolutely. As a kid, I worked on PDP-11 as well. So we share that in common. What made you jump into quantum computing?
Peter: That’s a complicated question. When I looked on Wikipedia, that section hadn’t been filled out yet. And so I thought to myself, “Well, I’ll come and work in quantum computing.” We’re at that place where I’m sure 25 years from now they will come, and it will be very full with a bunch of names, and I just wanted to be part of filling out Wikipedia and hopefully be part of the story for quantum computing.
Yuval: Customers usually talk about larger computers and error correction, scaling up existing technologies. I’ve had different guests on the podcast, and everyone is trying to convince me that their technology is the best way to go, whether it’s superconducting or trapped ions and so on. How do you answer that question? What are your plans for scaling up the machine?
Peter: That’s a great question. I’m quite sure first that every CEO tells exactly that, and I’ve heard it many times. And this is not unlike, to be honest, the 1980s in Silicon everyone was battling each other. Companies were saying, “Was it yield that was most important, transistor count, gate speeds, manufacturing costs?” All sorts of different things were being used at that time. So today, the question is, how do you get to scale, and when do you need to scale because definitely error correction is something that will be needed, but the longer you can hold off to getting to error correction, the better because error correction is going to have a tax because you’re suddenly going to have to consume lots of qubits to be able to do error correction. So in the NISQ era, the era that we’re in right now with these noisy systems, what we’re trying to do is to eke out enough performance to be able to get to quantum advantage.
And so that’s all about reducing the noise in the system, the error rate, the average two-qubit gate fidelities. And I want to be careful there, too, because what really matters is the average two-qubit gate fidelity, not your best two-qubit gate fidelity. One pair of qubits that do really well does not make an algorithm. You need all of them to do fairly well. So we talk about average two-qubit gate fidelities and it has this other kind of nice benefit, which is the amount of error correction you need is a function of the error that you’re trying to overcome. So our goal is to get to roughly 99.99 in kind of native gate fidelities without having to go to error correction. And then, after that to push the boundaries using error correction. And that kind of makes us different. Ion traps today have the best gate fidelities in the business.
Yuval: And with regards to scaling, how large do you think a computer needs to get before it delivers enough business value to a sufficiently large segment of customers?
Peter: So there’s an academic question which is when do you get to a point where you can do things that classical computers can’t do? That’s one. That for us is not as interesting. What the real question for us is: can our quantum computers and a quantum algorithm provide business value? And so that we’re already starting to see, in particular in machine learning, is that we’re seeing that even using our existing quantum machines, that we can come up with algorithms that roam around on our hardware that beat the best classical algorithms. Even though the hardware yet is not better than the best classical supercomputers. But I think if you’re a customer, you don’t care. The academic question doesn’t really matter. The question is: do we have a better mouse trap?
Yuval: And to deliver customer value, you need software beyond the hardware. Now, some companies out there have more of a full stack offering, an army of consultants, a big ecosystem, hardware, and so on. How do you compete in that market, and what’s your view on the full stack versus just hardware?
Peter: Sure. So full stack up to the application layer, that part of it we excel at and we have an operating system that runs on top of our hardware and a set of compilers and tools that also run that people might not even know they’re being used today when they use our quantum computers. Then above that is kind of the Qiskit layers and the libraries that sit above. In that, IonQ is very unique in that even all of our competitor’s frameworks all run on top of IonQ hardware. So I think that IonQ is the only place you can go where you can run every one of our competitors’ tool sets on top of IonQ hardware and that’s unique.
You can’t get Q# to run on some of these other platforms as an example. But you can in every one of our competition. So we made a decision not to compete in the kind of language area for now, but instead to make sure that our hardware works well with all the existing software out there. And people have been anxious to make sure their software works with IonQ Hardware because today, you can do things on IonQ hardware you can’t even get close to on the nearest competition.
Yuval: And how about the services layer? You mentioned machine learning. Machine learning is generic. You have to make it specific to a particular customer’s problem. How do you go about doing that?
Peter: So some of it comes by the fact that we do have kind of complete software compatibility with all the software platforms out there. If there’s a breakthrough that’s in Qiskit or something, that will work well on our hardware, but also we work with the three cloud providers as well. So whatever it is they’re doing, we get advantage of that. And we have companies like Accenture that is working and doing applications, and we work closely with them. And in addition, in-house we have a small, but growing applications team. And so they have been working directly with customers as well.
Yuval: I spoke with one of the industry analysts recently, and he told me there’s some concern for customers to work with the hardware or quantum software vendors because they need application-specific expertise. So if I have a supply chain problem, maybe I should go to a supply chain company that may or may not use quantum. Not to a quantum software or hardware company. How do you see that?
Peter: Well, it is true. When we looked at this, we ourselves said we can’t be experts at everything. And so you do have to figure out what it is that you think you’re going to specialize in. And so we made the conscious decision to focus IonQs applications, efforts, in machine learning and in chemistry. There are many other areas that one could focus in, but at least today what I would say is you need to have both quantum, you also need to have some domain expertise to be able to work on these problems. There’s even a little more nuance to this, which, as I mentioned, I go back into the 1970s. I remember getting my first Apple II on a 6502 processor, and my dad and I were going to do our first video game and my dad was an astrophysicist.
And so we decided we were going to do this game, which was called Rock Belter Hounds, and it was to do a space game about the solar system. And so we plotted out these ovals on the Apple II that used floating-point to be able to represent the planets and it took hours to generate that oval on an old 6502. And when we replaced all the floating point math with integer lookups instead and integer math, and it went like this. And so that’s kind of where we are with quantum today is that the machines themselves today, we’re not at a place where you can just throw away lots of qubits. You need to be smart about the algorithms and to be able to take advantage of the hardware and really get down to the iron. It’s an unfortunate statement, but it is currently the truth.
So what I would say too is, and we’ve seen it time and time again where people bring applications, might not run the first time, and then we will sit down with them to optimize their algorithms to do a better job taking advantage of the hardware’s capabilities, just the same way that we did back in an Apple II some 40 years ago. So I think it’s really important that people who are working on applications be working with a hardware vendor, hopefully, IonQ, but with a hardware vendor to optimize their particular applications to really get the best performance.
Yuval: One last question on applications. You’ve worked with quite a few customers. Which case studies are you most proud of?
Peter: Ooh. Well, let’s see. I’m excited about what we’ve been working with Accenture for a chemistry model for breaking down PFAS, the forever molecules. So I like that one just simply because it looks like PFAS is a large problem, and we might be able to contribute to the solution, and that impacts everyone. So that’s cool. On a personal basis, I really like the MLs stuff because it’s kind of also my background and it’s really, really fascinating to see the advancements that are going on in ML. We have shown already that we can create ML models, both on the quantum computer and also for classical hardware, where the black boxes actually capture more of the signal than you can do classically, which is very cool.
And the other piece is that we’ve seen now in a few examples where the features that’s being submitted, we can beat the best classical algorithms, but with 1000 fold less feature set. And that’s really interesting. You look at things like chat GPT, it’s super expensive for them to create that model. And so it suggests that maybe with quantum computers that, in the future that, you’ll be able to create those kind of complex models, which only can be created by the world’s largest corporations. And now, suddenly, everyone will be able to do that with a much smaller data set. And so that’s really cool as well. We’re kind of democratizing big ML.
Yuval: Do customers worry about power consumption? Is that one of the reasons they might opt to use quantum computers?
Peter: Yeah. We joke, it’s a little bit of an exaggeration that we run off a standard wall socket. I think it takes a couple of wall sockets to actually run the thing. But in terms of order of magnitude, today’s supercomputers, the reason we don’t build bare ones, it’s not because you can’t. It’s easy to put more Intel processors into a supercomputer is it consumes the power of a small city. And so who can afford to run them? So if we’re going to get to more computational power, we can’t have the power consumption be linear. And so that is certainly one of the promises. Somebody told me, I don’t know if it’s true or not, but somebody said some of the largest supercomputers are starting to consume the power of the Hoover Dam. And if that’s true, that’s horrific.
And clearly you do know that there are a lot of statistics that a significant portion, or maybe say it the other way, a not insignificant portion of the electrical grid at this point in the United States is powering data centers. So just in terms of climate change, we might be able to help just simply by replacing lots of data centers with quantum computers that consume a lot less electricity.
Yuval: We’re recording this episode on a week that you announced a manufacturing center. Why do you need a manufacturing center, what are you manufacturing there, and how many, if I may ask?
Peter: Yes. We announced in Seattle an opening of a 65,000-square-foot new center. The reason we’re doing it is we need to expand our manufacturing operation to meet kind of future demand that we’re projecting. So that’s one. One is demand that we’ve run out of space in College Park to build more quantum computers. So we need to expand in that sense. But the other thing is that the group in Seattle is going to be working on the productization of the quantum computers. Many quantum computers today are very fragile bespoke systems that are lovingly created by physicists with, we kind of joke, with screwdrivers out back and whatever. But all quantum computers to get to scale, it doesn’t matter if it’s ours or superconducting or anything else, sooner or later you can’t put more qubits on a chip, and so you’re going to have to go across multiple QPUs and you’re going to have to network them together.
But that also means you need to be able to reduce the cost of the quantum computer to get to scale. So it’s the same thing that we’ve done with Moore’s law and classical computers and today’s supercomputer is made up of 20 million processors. And so we should be thinking about how are we going to get to a place where there’s a data center that has thousands of quantum computers that are all networked together, that are all like a supercomputer, all working as one. And if the cost of those individual units is really expensive, then quantum computing will be like CERN, a country will be able to afford one. But IonQ’s vision is we want to have thousands of quantum computers. Maybe I’ll start to coin the phrase on a quantum computer on every desk to borrow Microsoft. So the Seattle space, they’re going to be working on a rack-mounted quantum computer that’s a much smaller footprint than what we have today and also starting to take into consideration like serviceability.
You don’t want to be at a place where if a power supply goes down, then you have to rip apart the whole machine. So starting to think about those kinds of things and to start stamping out much larger numbers of quantum computers. We’re right now working on networking on quantum computer. Not for communication, but for computation, to do entanglement across QPUs. To get that working, you need a lot of quantum computers. How many should you test to say that you’ve got networking? Hopefully, more than two. Four? 10? I don’t know, 100? So to really test that out, you need to build a lot more quantum computers and that’s what this is all about.
Yuval: I want to ask you a hypothetical question, probably the only hypothetical I could ask a CEO of a public company. If you could have dinner with one of the quantum greats, dead or alive, who would that person be?
Peter: Ooh. Well, first is I’ve had dinner with lots of quantum greats, which is very interesting. I’m going to go with Feynman himself. My dad went to his lectures, and I was too young at the time. I wish I had had the opportunity to go to his lectures. That would’ve been awesome.
Yuval: As we get close to the end of our conversation today, I wanted to ask you about executive hires. What type of people are you looking to hire, or are you hiring at IonQ as you think into the future?
Peter: So we’ve been very lucky to have a great team today. If you look, we’ve picked up from the best companies and managed to lure away some executives from exciting companies like Blue Origin and Google and Facebook, and all the rest. So generally, I have a philosophy which is similar to at least where Bill Gates was in the 1990s when Microsoft was king of the hill in the 1990s. And somebody asked Bill, “Well, what is your secret to success?” And he said, “I hire people, and I trust them to go do the work that I’ve tasked to them, and I try not to micromanage them. And if I get to a point where I’m micromanaging, we probably have a problem.” And so I’m looking to hire people that can take a piece of our business and run with totally independently without a lot of oversight on my part. And I’ve been lucky so far in putting together the team we have to get exactly that.
Yuval: And my last question today, what would your advice be to customers? What to do and what not to do when thinking about quantum computers.
Peter: That’s a great question. So if you look at the question, the question is: when will quantum computers and the applications for them start to be better than what you have today? And so you have to make a prediction. You yourself have to make a prediction is when you think that will actually be. Some companies are out because they don’t plan to have hardware for a decade or more or saying, “quantum computing is 10 or 15 years away.” IonQ has produced a roadmap that’s showing that in a mere two years from now is that we will have 64 algorithmic qubits, meaning that you should be able to create applications that have roughly 3,600 gates to it. And we expect that at that number of qubits with that fidelity is that you’ll be able to exceed what you can do on today’s classical machines. When I was working for Ray Kurzweil, we had a research team that always was working on predicting the future for his books.
There were several people who were doing that, but we also used that information in determining where we would invest our software resources to create applications. So we wanted to do optical character recognition on a cell phone, but the cell phones weren’t good enough. You needed to have the focal length of the camera, the processor, the memory, and the resolution of the camera had to be much better than what was commercially available. So we made a prediction that said, “At two years hence, that somebody will come out with a cell phone that has the hardware we need.” And we said to ourselves, “It’ll take two years to create that software. So the time to start working on that application was now.” And then that’s what we did. And then sure enough, literally to the trade show, to the month, to the day is the Nokia at the time, this is a while ago, came out with the exact phone that we predicted and our software was ready to go and that application won best mobile application of the year at Mobile World Congress.
And we were ready for the hardware when it came. And so what I would say to customers today is if you look at IonQ’s roadmap, and we’ve done a very good job of hitting our technical milestones that we’ve published several years ago, we’re on track to hit 64 algorithmic qubits, which is roughly two years from now. It will take two years for people to create production ready applications. If I was selling classical services and I came to a CEO and said, “Hey, we want to redo your manufacturing thing for a classical thing,” it wouldn’t surprise any CEO that it was going to take two years to write a classical application. Well we’re now within that timeframe. So we’re now at a place where people, if you believe IonQs roadmap and we’ve got a good track record, we’ve got data behind it now, if you believe that, you are actually already behind because it’s going to take two years plus because this isn’t classical.
It involves more than just creating a standard application. You’re going to have to find the algorithms. You can’t go out to GitHub and download a bunch of algorithms for your classical thing. Your team’s going to have to go create those algorithms. So anyway. What I would say now is that large Fortune 1000 companies are already now at risk at being behind. And as time goes by, that will become more critical.
Yuval: All right. Our time together went by quickly. So Peter, thank you so much for joining me today.
Peter: Ah, well thank you for having me. It’s been a pleasure.
Yuval Boger is an executive working at the intersection of quantum technology and business. Known as the “Superposition Guy” as well as the original “Qubit Guy,” he can be reached on LinkedIn or at this email.
January 26, 2023