IonQ, Inc. (IONQ) Earnings Call Transcript & Summary

Hi, everyone. Welcome back. This is our fourth session of the day of Day 2 of the Artificial Intelligence and Disruptive Tech Conference here at Bank of America. We're delighted to host Chris Monroe, Founder of IonQ in this session which is addressing advances in quantum computing. So Chris is the Chief Scientist at IonQ. Dr. Monroe is a world-renowned pioneer in quantum physics and the original inventor of the ion trap quantum computer. With David Wineland, he demonstrated the first logic Quantum Logic Gate in 1995 at NIST, which contributed to Dr. Wineland's Nobel Prize later in 2012. And later at the University of Michigan and Maryland, Dr. Monroe demonstrated the first ion trap on a semiconductor chip pioneered qubits networking using photons and developed ultrafast quantum gates for trapped ion qubits. He is a Professor of Physics and Electrical and Computer Engineering at Duke University and a College Park Professor at the University of Maryland. Dr. Monroe is also a member of the National Academy of Sciences and the National Quantum Initiative Advisory Committee. And Dr. Monroe is IonQ's co-founder and has served as its Chief Scientist in September of 2015. So with that, Dr. Monroe, welcome. We appreciate you taking the time to talk about quantum computing with our investor base.

Thank you. It's my pleasure to be here. I look forward to this discussion.

Excellent. So maybe to kick it off for investors who are not as familiar with IonQ, can you give a quick introduction of the company, please?

Indeed. IonQ, we formed in 2015 as a start-up, spun out of university research over the past few decades, pioneered by myself and Jungsang Kim, who's a co-founder. And IonQ is now a public company. We had an IPO about a year -- a little over a year ago. We build quantum computers, and that doesn't mean just the hardware, but we call ourselves a full stack quantum computer company that is not only do we build the basic level hardware that can store quantum information. I expect we'll talk more about that. We build the engineering on top of it to control the system, the software, the applications, the interface, so users can run algorithms in this new mode of computing. So we sort of have to do all of that to be successful. You can't -- all of those different levels have to work very tightly together, and we're one of the few companies really doing that at a very high level.

Okay. That's great. So maybe just to level set the discussion a little bit for the investor community that we have on, can you just talk about just why is quantum computing needed? What are some of the key things that you can accomplish via quantum computing that you can't through a traditional supercomputer, for example?

Good. Answer, it's a great question, and we can actually appeal to a little bit of history going back to the development of the conventionals, so we call it, classical a non-quantum computer back in the starting in the 30s, 40s, 50s and 60s. And of course, back then, it wasn't clear what the use cases would be for a computer, what was needed. They were built for very specific and targeted applications, something to do with military use, calculating missile trajectories, things like this. Transistors themselves were first envisioned as being a way to make a hearing aid amplifier that could fit in your pocket. So of course, but the concept of computers is much more wide ranging than that, and we really had to have the devices to see what their full reach would be in the marketplace. So I don't want to say history will repeat itself exactly like that. But quantum computers have a basic advantage over classical computers in that they can compute over multiple inputs at the same time. And that's very hard to get your head around. And it stems from the foundational aspects of quantum physics, which is also very hard for almost anybody to get their head around, but these rules are very clear, and they work. And if you have quantum information, you can store so much more data at the same time that you can attack problems. You can start to -- you can start to crack problems that are impossible using classical computers. And some of these problems will always be impossible using classical computers. The numbers get pretty amazingly high. And let me give you a great quick example of that. We're used to the bit as the fundamental unit of information 0 or 1, on or off, heads or tails and can be substantiated in many ways. In quantum computing, the analogist information carriers called the qubit, the quantum bit, and a quantum bit can be in both 0 and 1 at the same time, with arbitrary weightings, it could be 50-50, 30-70, 90-10. And the best part is when you add quantum bits together, when you treat multiple quantum bits, the number of possibilities doubles every time you add 1 bit. So when you go from 1 bit to 2 bit, the possibilities go from 2 states to 4 states and then to 8 states and then to 16 states. So this is exponential growth. Okay, in the banking industry, everybody is very familiar with this, but this is doubling every time you add 1 bit. So here's the amazing statement, really. With just 300 quantum bits, 300 qubits, there are 2 to the power 300 possibilities all stored at the same time. And why I pick 300 is 2 to 300 is about 10 to the 90 or 1 with 90 0s in front of it. 10 to the 90 is more than the number of atoms in the universe. So even if every part of the universe is part of a classical computer, it still cannot deal with that amount of information. So to finally get to your question, there are families of problems, big very important problems that we tend not to do now or we approximate , we take a guess. These are generally optimization problems that have to deal with lots of configurations. I can give you some examples. But these are problems that we tend not to solve very well right now. And when we have quantum computers, they have great promise to be able to attack these types of problems. Now not just banking, but also big pharma, oil and gas, logistics, self-driving vehicles, it's hard to believe there's a sector that will not be touched by quantum computing.

Yes. No, that makes a lot of sense. And I appreciate the analogy there. So when we think about qubits, there are different approaches to quantum computing that people are using different types of qubits. So maybe you can help demystify a little bit what these different types are, which is the path you've chosen and why.

The good news is, at a high level, what we learned, I hinted on this earlier, is that the concept of a bit is independent of hardware. We think about data storage. We used to store data on tapes and then records and CDs and now it's flash memory. The concept of storage and processing is independent of the hardware. So there are many potential hardwares that can work. And the secret to good quantum hardware is -- again, it's getting to the strange part of quantum. These super positions, these multiple numbers at the same time, they only exist as long as they're not -- as long as they're perfectly isolated from observation from the environment. A quantum system, when you look at it, it sort of gets destroyed. And that's kind of a big problem because we need to look at it eventually at the end of the day. But when we're doing a quantum computation, we need the system to be nearly perfectly isolated. So this helps inform what good quantum hardware will look like. So at IonQ, our quantum hardware, these are naturally quantum systems. These are individual atoms. It's a little exotic, I'll use that word again in a little bit. But individual atoms, the great thing about individual atoms is that they can be isolated. At IonQ, our atoms are suspended like a magnetically levitated train, or only the train are atoms and the train tracks are electrodes. And it's in a vacuum chamber, so there's no air. So these atoms are -- they're sitting there. They're not part of a surface. They're not part of the solid. They're nearly perfect quantum systems. And you can replicate them with perfection. You don't worry about yield because if we have a given isotope of a given atom, they're perfectly replicable no matter where you are. So our hardware is exotic. It involves laser beams that actually push around these atoms to wire them together is you can't use wires, if you use a real wire real -- real wire is made up of, I don't know, 10 to the 20 atoms and every wire is different, and it's not a good quantum system. So here's the rub. And what most of the producers in quantum technology are doing now is they're trying to extend silicon or solid-state conventional electronics to be quantum. And I can understand why we should do this as a society because silicon was incredibly successful. You can't argue with Moore's Law and the wonderful advances in computing over the last few decades. But I would say that solid state is nowhere to store -- that's not a good way to store quantum information because a solid system is big. It has way too many complexities, and you have to get it very cold, nearly absolute zero. At IonQ, our system is a little more exotic to the untrained eye, and therefore, I think a lot of the big players missed this technology. But it's the leading technology in terms of performance. And by any measure, we sort of have this platform that we know how to scale up, and that's another unique feature of our system. We're not relying on physics breakthroughs to scale up. It's more about the control engineering, getting these atoms lined up and laser beams. It's going to be an optical system that allows us to control these things. It's very well established, how to do everything like that in a laboratory environment. And in IonQ, we're bringing research out of the laboratory over the last few decades into commercial production. And I don't think any other player in the solid-state community can claim that. I think they're doing great research, but they can't yet -- they don't really know how to scale their systems up.

So to follow up on that, scaling, where do we stand today in terms of what is the scale at which we're operating today? And then when you talk about scaling, you mentioned for solid state, there are issues around cooling and the temperature at which they need to operate. It's very challenging and hard to do that at scale. But then you also mentioned the ability to drive maybe the control electronics closer to your qubits in some way. And so what are some of the things that you're doing that allow for scaling that is maybe a challenge in other systems?

Well, I guess I'll set the scale. I mentioned the number 300. If you had 300 qubits and you could control them at will and do the equivalent of logic gates we call them quantum logic gates. That's how you build a quantum algorithm. If you have 300 qubits and you have enough and they're clean enough to do a computation, that's amazing is you're going to be doing things that you could never think of doing classically. But you don't need 300. If you have maybe just say, 60 or 70. There are enough atoms in the universe to deal with that size of a system, but not every atom in the universe is part of the computer we control as we know. So when you get to 60 or 70 qubits, we call them algorithmic qubits. They are qubits that are good enough that you can run programs on. That's when we're going to start to see advantage. We're going to start to see algorithms coming out of the woodwork in many different economic sectors that are creating value. And that will pave the way to keep scaling. So where are we now? At IonQ, we're at an algorithmic qubit number in the 20s. But as you were asking, we know how to scale these things up. And it comes down to not quantum mechanics or quantum physics, it comes down to largely optical systems, miniaturizing lasers, getting them integrated on our chips. The chips are not the quantum part, the chips are like the train tracks. They hold the atoms in place. And those electrode, the electrodes that do that are on a chip inside of the vacuum chamber, and we'd like to get optics on that chip, integrate the optics using wave guide, using techniques that are out there right now but haven't been put to use in this way. So we're sort of -- I don't want to say we're heads down, we're certainly heads up. We need to work with users that have algorithms. But in terms of the hardware, we're heads down and we have a very clear path to get to the 50, 200 algorithm in qubit numbers that will allow them to be deployed. But it's very important that our heads not be down because we have to understand the pattern of algorithms of applications, and that can inform how we build our systems for a particular sector or for a particular application. All we really need to do is get one. And if it creates value, we pretty much know that we can replicate that in different context. So I'm not sure I answered you exactly, but we really do have a concrete scaling plan going forward, and it's not physics research.

Okay. That's actually very helpful. Maybe you could talk a little bit about given that a lot of the investment community is not as familiar with quantum physics. Maybe it's to be helpful to how can externally people look at milestones? What are the things that people should be monitoring to understand success of both your company as well as of the industry? And what is it that -- what are reasonable time frames for some of these milestones?

Yes. I think, again, I touched on the metric that we like to use is called algorithmic qubit number. And I don't [ add ] too technical, but I did hint that if you have some number of qubits, you're going to need to also run some number of gates on those qubits. And clearly, the more qubits you have, you're going to need to run more gates. Now unlike classical computers, quantum computers are very sensitive to errors. And if you run too many gates, then the noise accumulates. It's almost like an analog computer if somebody listens to on what that is. And analog computers are not inherently stable because errors add up. They have a finite depth, and that's why we don't typically use them to compute. While quantum computing has an aspect of analog computers, but the good news at the end is this doubling with every qubit, it still gives you more power. So algorithmic qubit numbers basically how many qubits do you have if that number is in, you can also run n squared gates. Now why n squared? Well, n squared comes because if you have n of something like think about if you have n footballs, there are n squared pairs of footballs roughly. And so -- and you need to run gates between all pairs of qubits in general. And so that's a good sort of a proxy for how deep your circuit needs to be. If you have n qubits, you need to run n squared ops. Now if you can only run 10 ops, and that's it, then things sort of fall apart after 10 ops. You're wasting your time if you have 1 million qubits, if you can only run 10 ops. If you only have 10 ops, you only need 3 qubits because 3 squared is 10, about 10. So the one thing to keep in mind is, don't pay attention to just qubit number only. It's easy to put 1 million qubit on a chip and say these are quantum systems, but they're so noisy that you can't run many gates. You have to talk about not only qubit number, but the depth or the number of gates that you can apply. That will tell you how powerful your system is. And algorithmic qubit neatly captures both of them. you don't put too many qubits if you don't need them. And also if you have really good -- if you have really good gates, but you only have 3 qubits, that's not very interesting because you're spinning your wheels all the time. So you need to increase them both very deliberately. And at IonQ, our road map sort of honors that. We're increasing qubit number, but only as they get better. So that's why it's difficult to scale. You have to add more qubits and they have to get better per qubit. And in a solid-state platform, I don't really see any physical way to do that. When you make a system bigger, it becomes noisier. It becomes harder to isolate it. That should be sort of clear. A big system is not very well isolated. So at IonQ, we have this very concrete path to scale, and this is really absent in almost all of our competitors.

Okay. That's helpful. I think one of our investors have a question over here around your comment about analog computers and sort of the noise that could create. But I think the question is really, can analog computers do some of the things that digital computers cannot? Can there be -- can they be a limited bridge before quantum computers since they'd be much easier to build than quantum computers?

A truly analog computer, I would say no, because a classical analog computer doesn't have this advantage of doubling its computational power in a sense, every time you add a quantum bit. But I think the question is deeper than that in the sense that a quantum computer does have an analog component to it and can we still use that. Now one thing I didn't talk about at all is a topic. It's a -- it's a very important aspect of quantum computing in the long run it's called error correction. Now some of you might remember back in the '70s, we had parity checks on ram, on memory. And this was just to make sure -- I mean the errors were more prevalent back then, you had to know whether there's an error and if so, you just do it over again. Now classical error correction has been very efficient, and it allows us to basically not worry so much about errors. In quantum computing, there's also quantum error correction. It's much harder because there are many more ways errors can get into the system, but there exists quantum air correction codes, but they're very expensive. In some hardware, you might need a 10,000 to 1 overhead. In other words, to get one good noise-free qubit, you might need 10,000 regular qubits that have noise. And it's all about redundancy and encoding. It's a little like the parity check only. It's really bad in terms of this expense. Now at IonQ, we're very mindful of air correction, but because our system is so clean, the air correction overhead is not so bad, maybe it's 10:1 or 100:1. That's still a lot, but we don't -- I think to get to your listeners question, we don't really need to deploy error correction yet. I think to get to an algorithm qubit number of 50 to 100, we're going to have to do some amount of error correction. So that's part of our scaling strategy. And with atomic qubits, they're cheap, we can easily put lots of atoms together. We need to control them all. And with error correction, to get 70 algorithmic qubits, we may need about 1,000 physical cubits, a little more noisy qubits. And we have that 10:1 or 15:1 type overhead that gets us 70. So -- and the reason I bring that up is that, in a sense, it becomes less analog, it becomes more digital, it latches. When you have air correction, you can get systems to be more stable. And that's why we use digital systems because they're much more stable than analog. They don't -- errors don't accumulate and that's the idea behind quantum error correction. So regular analog computing that you might be gears and motors and so forth, that's not going to get us very far because it doesn't have the advantages of quantum.

Okay. That's very helpful. Maybe you can talk a little bit about the different offerings that you have. I notice you have IonQ Harmony, Aria, Forte. What are these different systems? And can you just talk through what these -- what the differences there are?

Sure. We've -- at IonQ, we've built 8 generations of quantum computers. And I would say, certainly in the early days, these quantum computers were prototypes. Every new system is quite a bit different than its -- well, there's still atom. So fundamentally, they're the same, but we're improving. We're learning when we build a system, the deep laws of engineering, how to make it perform better, how to make it perform autonomously on the cloud. So each generation is generally better than its previous and the Harmony systems are on the commoditized cloud of Microsoft Azure, AWS bracket and also Google Cloud platform. So people can get access to those. But they get access to those that there's sort of a buffer between the users and IonQ. And so it's good because at least people can start to learn the language, the low-level language how to [indiscernible] devices, but we also don't get any information on what they're doing or if they have questions directly to us. So we also have our own cloud, of course, that we can't send out to thousands of customers because we want to handhold every one of them and learn about their applications. So we have very deep collaborations, for instance, with Hyundai Motor. They're interested in battery -- simulating battery material science and there's quantum programs for that. But also they're interested in image recognition, pattern recognition for future autonomous vehicle applications. So that's -- Hyundai is one example of several of the customers that we work with. Our applications team about 10 people strong or so, and they're growing. They sort of work with the customers and help them understand our system, and we also learn from their type of application. So it's really good for both sides. So the Aria system, that's our fifth generation is very recently been installed on the Azure cloud. So I should have corrected myself, as just in the last month or so. And that one is quite a bit more powerful than the Harmony systems. It has an AQ algorithmic qubit number measured in the low 20s, which is, again, state-of-the-art, nobody, no other system as far as I know, can reach that. The Forte system is we haven't yet tested it on a system level algorithm, but the components, the gates and the quality of the gates looks to be even better than Aria. And so we're still working through that. And eventually, that will get on the cloud. It's a different -- we use a little bit of a different optical addressing system that's details. But again, these are prototypes that are sort of advancing the ball significantly around each. Now I should also note that we're -- there is demand for our machines to be installed in a customer's warehouse or their office. And so we have to start to move away from prototypes to a production model, and we're just starting to do that now. And we expect in the next couple of years to be able to have a production model where now we can talk about yield. If we build 10 systems, are they all the same? And that's production engineering unit at IonQ is now fast in motion to do exactly that. So again, we've had 8 generations of production line coming. So again, we're growing very fast. And we have a big engineering staff, and that will also double in size in the next year or two.

Okay. That's fantastic. So for -- I think you said you don't have too much visibility on what's going on with the hyperscalers in terms of the -- what customers might be using that for. But as you think about the use cases that you expect will really tick up over the -- let's say, the commercial systems that you might be offering 2 years from now, where do you see the most opportunity and most sort of pent up demand?

I think there's one word. It's a little bit of a serious word, it's called heuristic. Heuristic is something that -- it's a tool that we use, but we might not know exactly why it works so well. But they're very well-established standards and so forth. We use heuristics all the time to optimize models. And machine learning itself, I think, in classical computing isn't heuristic. We don't know exactly when it's going to work, when it's going to fail, but on average, it can offer a huge advantage in certain applications. So there's something called quantum machine learning that allows us to expand -- sort of expand the research, if you expand the direction of machine learning to be able to again, sample much larger pieces of data and act on that. So those are largely heuristics, but we're starting to see some very interesting results when it comes to aspects. I mentioned image recognition or natural language processing, how to recognize handwritten numerals and so forth. Again, these are still small applications on small computers. You could probably simulate what's going on in the quantum machine with your laptop. But as we learn the structure of that problem and we build bigger and bigger quantum computers, your laptop is going to fall flat in a few years. So you mentioned, what's the time line for some of these things, I think, and maybe the last question, we're not looking at 10-plus years. We're really -- we think we're going to -- given our technical road map, we think it's measured in a few years, not decades or longer. So machine learning is one, another one surprisingly, and I'm not just saying that because this is a BofA event, is the financial quantifying financial models. We've had some interesting work with the Fidelity Center for Applied Technology and others, Goldman Sachs, and so forth that allows us to deploy our quantum computers for things like multivariate sampling. When you have correlated brand of variables, and there are very many of them, it's very hard to sample or stimulate sampling from that distribution. And by having something called quantum entanglement that qubits have, it allows us to have these sort of implicit connections between random variables. It's a very interesting application. The one that might have most resonance with listeners here is just optimizing financial models or portfolio optimization, even if your model is not correct, you want to model the Dow Jones Index, you have 10,000 variables and you have a nonlinear model. What setting of those 10,000 variables will make the biggest Dow Jones Index. Well, even if your model's correct, you still can't solve it. But if your model's incorrect, you would like to solve it. And again, there's too many configurations. And so there is work that's looking very interesting in very small versions of that type of application using our quantum computer. And so that was a little bit of a surprise. I think the financial sector kind of came up real strong. And we worked with a few other quantum start-ups, quantum software startups to get us kind of informed on some of those applications. So I think those are the early ones. I think I mentioned chemists, maybe I didn't -- materials and chemistry simulating how molecules bind. These are problems that are vexing very hard on classical computers, and we're -- we have an active area of research with several customers on quantum chemistry. That's probably next in the line in a few years. And then after that, boy, it's just hard to tell.

And it sounds like -- I mean this would also be a good fit for something like drug discovery and protein folding problems, those sort of...

Yes. Those are obviously both in the same market, but totally different types of problems. One is the logistics problem, and you have 10,000 compounds and you're pretty sure 5 of them together are going to attack some -- for some kind of therapy, but 10,000, how do you find out which 5? So that's logistics problem, but then protein folding is -- it's a very complex molecular dynamics problem that you can actually -- there are ways to program it on a quantum computer. And many companies are really thinking about that, both those things.

We're just about coming to the top of our time limit here, but I was hoping maybe you could talk about 2 more things quickly. One is, you mentioned briefly about exposing programmers to the more low level sort of programming concepts, which are generally, a, you've had a limited set of people to do that, and it's quite specialized. But I'm sure that there are companies or maybe you are -- who are trying to extract that in a higher level and be able to help drive better software faster software development around this concept. So a, your thoughts on that? And b, where do you think this industry is going to be 5 years from now? Would love to get sort of a longer-term outlook from you?

Yes. Great questions. And both of them, to me, it's just a very exciting time. I mean when I was a kid, I programmed a little bit an assembly language a very low-level language. I haven't touched assembly code since the 1970s, though. And it's a good thing because it's a very low level, we use much higher level programs now. And so I don't need to know all the details. I don't need to know about semiconductor physics every time I pull out my laptop. So indeed, as you say, there are great efforts in making software libraries available, maybe a hybrid type of computing where you have your computer on the cloud at AWS or whatever it is, and you press the quantum button and then it tries to solve this problem using a quantum computer. The programmer really doesn't need to know this. So wrapping up much higher level libraries or algorithm families. This is definitely on the horizon. It's a little chicken and the egg though, who's going to write those and what are they going to do? They need to create value before they get widespread use. So we are at lower levels now. And all we want to do is to accelerate the time where we have these libraries available for everybody. We do need to work with the customers, just to find one app that's a kind of a killer app or something that creates value. And probably we'll be able to replicate that. To some extent, we'll have to change this and that. So it works better for a particular problem. But you're right, it's a little challenging to get people willing to do that. On the other hand, I look at the young people coming out of the universities and they come to us. Wearing my university had at Duke, I mean, we get tons of students in engineering, computer science, even physics that they know all about quantum computers, and they want a career in this. So it's great that we have companies. It's a big ecosystem Google, the big behemoths IBM, Amazon, Intel, IBM, they're -- these companies are all spreading the word, and it's great. So I think it's going to happen. And I think it's a matter of a few years before we start to see this hit [indiscernible]. It's going to be very specialized for a little while. It's going to make value. And that's important economically, that's going to pay for the development after that, just like Moore's Law. We found applications for regular computers that paid for the further miniaturization in the very expensive fabrication, high-yield fabrication of chips, where the economics is such that each chip now costs only $50. But if you -- the first chip costs $10 billion. So that economy of scale is absolutely important. We have to find these applications.

Chris, unfortunately, we're out of time, but we really appreciate you taking part of your day and spending it with us over here. It was very informative, and we look forward to continued dialogue as you make more progress around these initiatives, and we look forward to an update at another time. So thank you so much for taking your time

It's my pleasure.

And for all the investors dialed in, if you have any questions, please do reach out to us. We'll do our best to get them answered for you. Our next session starts at 1:30 Eastern -- it's on next-generated augmented reality. We're going to be speaking with Magic Leap. So that starts in about 10 minutes from now. So please join us for that.