EP01. Hardware and Architecture

Summary

Over the last 5 years, we’ve seen an increase in private equity investment into quantum computing companies, showing a conviction in the long-term viability of the industry. More than 70% of this investment goes into hardware companies, showing the importance of getting quantum hardware right, as the fundamental step in having quantum computation at scale.

To discuss “Hardware and Architecture”, we invited two leading experts making an impact not just as scientists but also as founders of companies pushing the limits of computing technology:

• John Morton, Founder and CTO of Quantum Motion
• Chris Monroe, Co-Founder and Chief Scientist of IonQ

The first barrier to useful quantum computing: scalable quantum processors

Gwendolyn Regina, moderator

Why did Horizon choose to start this series Quantum Well with hardware and architectures?

Joe Fitzsimons, CEO of Horizon Quantum Computing

I think it’s a natural choice for us. Basically, the purpose of this series as we see it is to foster a better understanding within the community of what the real barriers are to quantum computing and how they're being overcome. So that people have a better understanding of where the technology is actually at. So that they are not caught up either in hype or in pessimism but rather have a realistic view of where the technology is. Quantum computing is a fundamentally new form of computation and that means that we need to build new technology, new hardware in order to get to implementation. It works in a completely different way from the way conventional computers do. And we need hardware to do that. It’s okay to talk about algorithms and we will later in the season. We’ll talk about fault tolerance and correcting errors as they occur. But if we have no hardware then we have nothing. So it’s the natural starting place for us.

Guests of the “Hardware and Architecture” episode: John Morton and Chris Monroe

Gwendolyn Regina, moderator

John, you’ve worked in a number of competing quantum technologies, what’s so exciting about silicon?

John Morton

I have worked on different types of qubits for coming up to 20 years. When I first got into this business straight out of university into my PhD, I was looking at molecular-based qubits. But always looking at the electron spin, this magnetic degree of freedom for encoding the qubit. And then, for about the last 10 years, I’ve been focused around realizing such spin qubits within silicon. I could give a 2-hour lecture on why it is a fantastic platform but maybe I could make a few brief remarks. It’s really an amazing material. Partly it’s around the efforts that mankind has put into engineering it. It’s the purest solid that mankind has made. Why? Well, because of the 70 years and trillions of dollars of investment into this semiconductor industry which is making the shift that’s in your smartphone, your watch, your dishwasher, etc. So this huge engineering effort into purifying the material and technology with which we can fabricate devices in silicon is a major benefit. But then there is also nature. Nature has elected to give silicon dominant isotopes that have zero nuclear spin so it makes naturally a quite environment to encode qubits that are magnetic in nature. So we are really excited about to really trying to leverage those kind of processes. The manufacturing methods that have been really pushed to their limits to make billions of transistors on a single chip with high yield and high uniformity and try to divert that process towards engineering qubits as well.

Gwendolyn Regina

Thank you, John. Another 2-hour lecture, another time. I will be interested. Thank you, again.

So, for our second guest speaker today, Chris Monroe. Chris, how did you get started and interested in working on trapped ions?

Chris Monroe

I sort of backed in this field a long time ago. About 25 years ago, I was making atomic clocks. In fact, I was a staff scientist at the U.S standards laboratory, where we make atomic clocks out of atoms. And this was 1995. I don't want to get too technical. We were actually entangling atoms to make a better clock. We didn't call it a quantum gate. We didn't call our system a quantum computer, but we in fact built the first quantum gate in any platform with these atoms when I worked with David Wineland in Boulder, Colorado. So I will say I kind of backed in the field, but the science of making quantum many-body entangled states, which is the sort of the backbone of a quantum computer, it became more than research for me. I collaborated with Jungsang Kim an engineer who got in this field about 20 years ago, and we've been building systems for the last 15 years out of individual atoms. So I think it'll be a nice counterpart to John here. I think we're going to look at two quite different platforms in quantum. They have majorly different challenges and maybe different opportunities going ahead. And yeah, these early days are very exciting.

Why don’t we have quantum computers yet?

Gwendolyn Regina

Why don't we have quantum computers yet? What has taken us so long?

Chris Monroe

I think the answer to your question has to do with the exotic nature of this type of computing. It's just not the next generation of standard processor. It makes an end-run around Moore's law by changing the laws of physics for computing. Now these laws, again, we can have a six-hour lecture on the laws of quantum physics I suppose. But they're quite exotic in the sense that they don't have analogies in everyday life. The idea of superposition, you could do things with multiple states at the same time with one device, sort of like parallel processing with one processor. And the requirements on systems to be able to show that behaviour that they be extremely isolated from the environment is if you look at it while it's computing, it gets destroyed. And I'm just being a conscious being, looking at it, but any unwanted interaction with the environment. So, in, say John's system, that means really cold temperatures, almost absolute zero. In systems I work with, individual atoms, I'm sure we'll talk more about the differences. In our systems, they're in a vacuum chamber, so the atoms are not connected to the environment because there's a vacuum in between. There's always some type of exotic feature that allows the system to be quantum and the hardest thing is, okay, we've isolated it. Now we have to control it without looking. That's possible, but very, very difficult. So it's going to require extreme techniques and engineering to really get these devices working, so that's why we don't have one now. I will put the onus on people like Joe, we need more apps. But I'll still say, well, we need machines to run the apps. And this is why the field is beautiful. I think it's a big community and everybody is working together to make this happen. Both on the algorithm side, and people like me on the hardware side building device API.

John Morton

In some ways you can say we already have quantum computers, they're just not very good yet. People are already making, including IonQ and making prototype quantum processes that you can access in the cloud. You can write algorithms for them, get back the answers. Some of them have even demonstrated this thing called quantum supremacy, where they've been shown to be impossible to emulate, using even the most advanced computers that we have on the planet. So I think I agree critically they're not yet at a scale that's useful to solve problems, but things have come a long way from the types of experiments that Chris talked about demonstrating an entanglement between two qubits towards developing devices that you can program, and that can be accessed and already starting to be a really useful playground, a sand pit to develop new techniques and new algorithms. And that's been incredibly exciting to watch even over the past five years, but yeah, as to why it's hard to improve, it's exactly this kind of challenge that Chris mentioned that on the one hand you have to incredibly isolate qubits, so they can't interact with their environment, but you still need to couple them together. And so you're always trying to sort of find systems where you can get good interactions between qubits, but not between their environment. And I guess another reason why it's it's much harder, there's the quantum aspect, but there's also the analogy with analog computers. So in the early days of computing, people would use, would encode information as, for example, any potential voltage within some range. And that would be your information. Analog it, sort of, I guess with the way the sort of early audio processing took place. But very quickly we realised that this was very prone to errors. Nothing to do with quantum, is just the fact that any noise would change your information because it's analog. It can be in any of these states. And so we moved to digital information, ones and zeros. We said, look, your information is either at this level zero or this level one, and any perturbation would keep you around zero or one. But the problem is with quantum information, you have in many ways, a kind of analog state, the qubit can be moved around between these different points. It's actually points on the sphere of any perturb, any noise can change the qubit state. And so, in fact, in the early days people say, well, okay, this is a fabulous idea on paper, this quantum computing, but because you can't do error correction, because you can't do this kind of noise suppression that you do in digital computing, then it's just an academic curiosity. Then the major breakthrough came from realising that actually, no, because when you measure quantum systems, you project it either onto the zero state or onto the one state, it is also kind of digital in nature and you can perform the same kind of error correcting methods. And the downside is that it needs redundancy. It needs many, many copies of an imperfect qubit in order to create a perfect one. And that number can range from tens to hundreds, to even thousands of copies of a qubit that you need in order to correct for these errors. And that means that if you want to use this error correction, if you want to perform this kind of digital noise suppression but on qubits, for the first applications you're going to need hundreds of thousands, if not millions of qubits. And that's in many ways, a daunting challenge. The noise levels in qubits in the lab are already low enough to satisfy that requirement, but you need many more qubits. So in a sense, the two approaches being followed now in quantum hardware are either to, let's say just focused on scaling up the number of qubits and looking at ways to get towards that sort of millions of qubits that you need, or trying to suppress the errors even further so that you can get squeeze out some useful quantum applications, even without performing this error correction before the quantum computer folds over.

Barriers to building scalable quantum computers

Gwendolyn Regina

What other barriers are there to building scalable quantum computers?

Joe Fitzsimons

Well, there's definitely a large number of barriers - hardware is the foremost one, and there's a lot of challenges associated with hardware. And John and Chris are really the experts here. I think one thing that's happened in recent years –to me, it seemed to happen around 2015 or 2016 – is that the emphasis went from making progressively better qubits to starting to scale up at systems. I guess this is around the time that IonQ was founded. So Chris may have some views on this, and why quantum hardware looked good to him at that time? But there was definitely a period where the quality of quantum gates in a lot of systems, the fidelity of entangling gates, which is basically how well you can entangle two qubits, how good you can do a gate between two quantum bits, without an error occurring, that has come down a lot since the early days of quantum computing, it was initially high, and the error rates are now coming down where there are less than 1% coming down to maybe less than one 10th of 1%, which is really an extraordinary achievement. But it seems to me that this has brought around a emphasis towards getting to larger scale systems. And you have some problems with making individual qubits well behaved, but as you start to scale up, there's all sorts of engineering problems that are involved as well. And I guess I will leave it up to John and Chris to talk more about that! Over to you guys.

Gwendolyn Regina

Yeah. So let's dive right into the kind of challenges that Chris and John, you've been facing in building tech. As all of you have pointed out this, a huge bunch of like barriers and things to be, alot of nuance right? That needs to be done and detailed to get that precision, so maybe back to Chris, what challenge are you facing in building your tech?

Chris Monroe

Good question. I could speak for an hour on that! I think we're going to all say this at the intro to every one of our answers. In the technology I work with, again, I come from a kind of a qubit physics point of view in my early life. But five or six years ago, this timing that Joe mentioned, a few things happened. We noticed that our system, I call it a system, let me define for about a minute what that is. Our qubits are composed of individual atoms. These are charged atoms they're ions, and they are electromagnetically confined with a bunch of electrodes. Think of a magnetically levitated train if you want, only these are atoms and it's electric, not magnetic. But apart from that, the chip behind me can see, has about a hundred electrodes. And we applied voltages to those electrodes and the atoms float above that chip in a vacuum chamber. These atoms are affected, they're measured through laser beam. So it's a very big optical system. So what we realised five or six years ago was, and this will get to the global challenges are going forward, what we realised is instead of doing atomic physics, we're not really doing physics at all. Our systems are so well behaved. These are atoms. You're not going to learn anything more about the atom, we sort of know everything we need to. Atoms are perfectly replicable. There are no errors when we make two atoms of the same isotope for the same element, there is no yield issue. Yield is exactly 100%. We have to make them, we have to control them, but it's really an optical engineering challenge. And what we noticed five or six years ago, in the morning we would tune up the system. You know, you have to aim laser beams, a lot of technology there, but in the afternoon we were running algorithms. We were sitting in our theatres with our regular computers and running circuits, right on our systems. We were not, we were very far removed from atomic physics. We weren't thinking about the qubits anymore. We were thinking about piling the algorithm, making it work better, because there are errors in the gates and we can only run it so far. We want to make it more efficient to do some type of a task. Some small algorithm, to make it some particular target quantum state. So we've been paying attention, over the last five or six years at IonQ particular on making the system more reliable, making it smaller, not to be cute, but when you make an optical system smaller it performs better, and also we can talk about scaling. If you're going to scale it up, we really need to make it small and modular. We know exactly how our system will scale. And what I mean by that is that we don't worry about any manufacturing issues. It's not about figuring out some fundamental physics of how its surface will interact with our individual models that really doesn't exist. It's all about engineering controllers making an optical system better. So the reason IonQ formed five or six years ago is these are things that are very hard to do at a university. Universities are not very good at making commoditised devices that can be made standard. That's really what industry does. So we felt there was a great need for an industry in our field. So again, I mean, it sounds super general here, but the challenges are many fold, but they're not physics. It's not the physics of the qubits, certainly not atomic physics. It's much more about engineering, and this is what gives us huge confidence going forward in our scaling. We don't know exactly all the technologies we will bring in, but we sort of to know what it will look like in the end. Integrated optics is a hint. Bringing optical systems right on the chip itself where the atoms are located. We don't really do much of that yet, but I think the whole community's moving in that direction. And so we're as an atomic physicist, I'm not an expert in integrated optics and integrating optical wave guides on silicon chips, you see silicon behind me. So we're indeed going to leverage the fabrication community out there that has been useful for so many solid state qubit architectures and we are going to use that will help us integrate optics.

Gwendolyn Regina

And John, do you face the same kind of challenges where it is more engineering rather than physics?

John Morton

Absolutely. I think the field in the hardware area has certainly moved in that direction. I would like to say it's been elevated to an engineering problem, though my colleagues in physics kind of dismiss engineering problems as the point at which it's no longer relevant, but I think that's where it gets exciting. But of course there's a whole set of different engineering problems in a sense, a lot of what we think about is trying to turn some proof of principle demonstrations that are performed within silicon using very bespoke fabrication methods. For example, in university clean rooms, and trying to understand how we can create that kind of qubit functionality within advanced CMOS processes. So for example, a two qubit gate has been demonstrated in a silicon device fabricated in a university clean room now with over 99.5% fidelity, which is well above the fault tolerant threshold and comparable to superconducting qubits, if not, Chris has nice ions. But bespoke process is not the same kind of CMOS process performed on 300 millimeter wafers, which is used in industry. And it's very advanced, you can produce billions of devices, but there's a large rule book that tells you what you can and can't do. And so understanding how to get that kind of functionality within these kinds of more industrial processes is a major challenge. But it also opens up the opportunities because if you're using exactly the same or similar processes that are used to make computer processes, well then actually, you can incorporate some of that technology on the same chip and you can really integrate the control and readout electronics with the qubits in a way which is much more natural, should we say than in other systems. But I sort of strongly echo that message from Chris, that doing all of this very difficult to see how it can be done within a university environment, because you do need to pull together such different expertise; I see engineers, of course quantum architecture theorists, but quantum engineers running measurements and optimising everything else that you need from the software control to bespoke circuit for the readout. So it's a very interdisciplinary engineering challenge and a startup has been a fantastic environment in which to bring together that sort of team and make progress.

Gwendolyn Regina

Fantastic. We'll hear more about a progress and a short bit. At this point, I wanted to bring up the poll that we had earlier. We now have the poll results and okay. So, 48%, the majority! I mean, almost half think that it was less than two years ago that they started considering quantum computing is potentially useful. Joe are you surprised by this?

Joe Fitzsimons

I think that's not surprising at all. A lot of the attention has come into the space recently. I guess the seeds of this have been sown quite a long time now, both with progress on the academic front, as well as what large corporations have now with an increasingly large number of startups. And it's easy for people to perhaps see the startups as something new, but I guess as we've seen, some of them have been working on this for a long time. And a lot of these efforts have grown out of academic research groups that have been working on quantum computing going back 20 years at this point. But it's really only recently that it's seen such large investment and seen such large focus from an industry perspective.

Quantum hardware types and their characteristics

Gwendolyn Regina

So we do have one question from the floor. Peter Morrison, thank you so much for the question. So he asks, “I'd like to understand more about the trade-offs between various different types of qubit, superconducting versus trapped ions versus silicon and so on.” Thank you so much, Peter. So I guess this is a very broad question. And throughout this conversation today, we also want to dive into a little bit, but maybe Joe, do you have some broad commentary? An answer to this question first? Then we can go back to Chris and John for a deeper dive.

Joe Fitzsimons

Well, there's a number of different technologies and each of them have their own characteristics. There's quite a nice report that came out about a year ago on the threat that quantum computing poses to cryptography. And it has a pretty good survey from, I think, 43 or 44 physicists in the field, I think perhaps Chris may have been one of the respondents, I was one of the respondents. There have been quite a few input from a number of experts from different areas. If you look at how things break down from that, you see that ions and superconducting qubits are generally viewed as the nearest term bet, as the most likely to succeed in the near term. And those are quite different technologies. There's some very nice things about ions in particular, they are optically active often, which means you can get photons out. You can in principle, think about entangling devices at a distance using them as part of a quantum network, as well as computers in their own right. They're also extremely good at storing information for long periods of time.

Joe Fitzsimons

So there's different properties to each of the devices. There's different timescales for the interaction between qubits, there's different levels of decoherence and there's different levels of control. Optics, photonic qubits are special in their own right. Because photons don't really interact with one another. So there's no easy way to do entangling gates with them. For those, you need to make all of your entangling operations through projective measurements. So you need to set up some measurement that projects your system into an untangled state, rather than being able to just interact two qubits. So, each system has its own pros and cons. At this stage, I would say it's probably not clear yet who's going to get there first, although I'm sure our guests have their own views on that, but we'll take the lead.

John Morton

Yeah, of course. It's very difficult to boil down this very complex parameters that Joe's mentioned when assessing these platforms. People are trying to look for a magic number, like you know, the number of megapixels in a camera to measure each one. And there've been various attempts at this. Obviously, sometimes people talk about qubit numbers. Look how many qubits are in my processor? And we've already talked about errors and how important it is to keep the error very low so the quality of each operation is very high. But each of those numbers on their own doesn't tell the full story, and there had been attempts to try to look at combinations of those. I think, in terms of assessing quantum processes the sort of prototype processes as they are today, it's really about algorithms. Picking different classes of algorithms and seeing how complex a problem can it solve. But if you're then trying to make predictions about which one is going to, and that's just trying to work out, which process we have today is more powerful. If you're then trying to work out which one is going to be most useful in the future, that becomes very difficult. You have to look at different ways to measure scalability, the networking, the optical link aspect of iron traps is very nice. So the network ability is important, integration, and my own view is that there will probably be different waves of dominant platform. We had the first computing devices, they were mechanical, right? They were gears. We had computers, the first real computers, general purpose ones were based on vacuum tubes and now it's predominantly silicon. So I think there's every reason that you will have different forms of quantum computing hardware at different stages in their development.

Quantum hardware scaling: engineering challenges

Gwendolyn Regina

Thank you, John. So maybe we can go now to how you're trying to overcome the different challenges and in building a tech and your companies. Chris, so engineering challenges, how are you trying to overcome them?

Chris Monroe

Well, hiring engineers, it's simply put. Now, if it were that simple! It's been a very interesting many years. I think, we owe a debt of gratitude to places like IBM, Google, Intel, Microsoft. They had, these are big behemoth companies that had early quantum. They sort of rose awareness in the field on the industrial side. And frankly, when we started our company, it was based on a university blueprint and my collaborations with Jungsang Kim, he's now at Duke University, I was at University of Maryland at the time. We had a pretty outsized research grant at the university that allowed us to start to do a little engineering, to get third-party companies, to start to make subsystems for us to integrate. And as we started building the system, we realised that again, I said this before, we're not really doing physics anymore. We're sort of doing systems engineering and those two words I'm getting, I think some of that in college, systems engineering, but it takes a decade or more in your career of doing it, to really understand what that is. Systems engineering is a little holistic, it's like the grand system, not just the sum of its parts. And in physics, we love the idea of some of the parts. We understand every little theory. So if somebody else can do all the engineering and John mentioned that, does this tend to dismiss the engineers. But systems engineering is a really subtle thing. It's like a jet engine. It's such a big thing. We have people that think about the entire point as a sort of a living being. And this is really coming to roost in quantum engineering. Finding people that have sort of that expertise, they don't have to know quantum physics, really. I think they have to be a little bit risk averse in the sense that it is a risky field. We don't know exactly where it is going, but we're all pretty sure it's going somewhere good. And so, I think the big challenge is getting those people into this field. People that have conventional engineering experience. So we love people that maybe, have worked on aircrafts. It's a totally different field, but they understand these big systems, the level of issues and integrating that with product controls. And then yes, there is in our platform, there is atomic visit, easier as an office, but I see that to be, it's actually kind of a small part that we have under wraps because we have no problem hiring those people. From the labs here at University of Maryland, we're just off campus. Maryland is a huge community of quantum researchers or professors all the way down to students that are over 200 people. In the U.S, there are many sort of meccas of quantum expertise on the academic side. So getting the physics side is not hard. It's really getting those that maybe have experience with integrated optics in a lithography system, like silicon, these, a lot of optics there. Or, in chip fab, solid state physics or RF, radio frequency and microwave control. So, that been the real growth over the last many years. I won't say we've struggled, I mean, we're not a big company. We have about 85 employees now, but we're in a growth stage right now. If you're out there listening, do look us up!

Gwendolyn Regina

You might be competing with John here and Joe as well for talent!

Joe Fitzsimons

I think we all are.

Gwendolyn Regina

So, John, how about you?

John Morton

We're hiring as well.

Gwendolyn Regina

There we go. Competition for talent. Everyone listen up.

Joe Fitzsimons

There's this tiny talent pool that everyone is competing over on the quantum side.

Gwendolyn Regina

The solution is to grow the pie, grow the funnel. Okay, John, onto you, how are you trying to overcome your challenges apart from hiring?

John Morton

Yeah. I just found myself agreeing strongly with everything Chris has said. In terms of the funnel, and talent of course, that's been a big advantage of having and maintaining close links with universities. So Quantum Motion was founded by myself and Simon Benjamin, and we've kept close links with the university of Oxford and UCL, funding PhD students, and so on, on their various quantum programs. And that's been really important in terms of getting a pipeline of talented people into the company. But also, we found that what bright students have, I guess, the sort of the startup route is now a kind of a third route for bright PhD students that are into this area. It used to be, if you wanted to keep working quantum computing, it was basically look for a career in academia. And now, of course, there are big efforts within large multinational companies, but also the startup environment. And we found lots of students who found the combination of working with a startup, but also working in a fun area like quantum computing to be a really exciting combination. Yes, absolutely, talent and getting good engineers is key to solving these challenges. And we found one way to address that is through close links with the universities.

Chris Monroe

And another challenge I think for all of us is to get people that don't know anything about quantum, to start thinking about the hard problems in their business. And how they need to know, I guess, a little bit about the opportunities in quantum, but for instance, the optimisation problem, this is a little bit of a catch all word, but we tend to ignore them in life right now. They're too hard. Like a traveling salesman type problem, a logistics problem. We make approximations, they work okay. Maybe they're kind of obvious, they make guesses. At least the quantum may be able to really help on these and getting the community to start to think, but not ignoring those problems anymore, but maybe learning a little bit about the thrill of quantum and working with Joe's company and designing algorithms. I think that's a big challenge too. We need to spread, we're going to open our own funnel, and we need to get the rest of the community involved.

Gwendolyn Regina

Chris, perfect kind of segue because we also have a question from the floor by Tommaso and he asks the question, maybe this time I can direct it to John and Joe as well. Given the great engineering challenges ahead, what business opportunities do you foresee in the quantum hardware space beyond what we might consider the most obvious ones?

John Morton

I think to answer that I need to know what are the most obvious ones, but I think certainly the obvious one might be that you need to build a QPU, or you need to build some kind of quantum processing unit. But if you want to deliver quantum computing capabilities, it's about much more than that. It’s about, we call the classical or the conventional computer that's behind the scenes, running the whole thing and the interface, which someone is then using it and developing quantum programs to run on it. So it's all that being part of the stack, but certainly if we're talking about achieving, developing useful quantum computers, then that's indeed a problem that needs to be tackled from both ends. Those developing the quantum hardware will make the best quantum hardware with the lowest error rate and the largest number of qubits and so on. It's also about finding and optimising applications to fit on those. Some of the first estimates for trying to use quantum computers to simulate versus basic quantum chemistry problems called for astronomical numbers of qubits and circuit depth since the length of the quantum code. And that's come down by many, many orders of magnitude. So the developments there have been far faster if you're in sort of, pure order of magnitude reduction than what we've seen in hardware. And of course that will slow down now that you know, lots of the sort of big gains have been had. I think, as Chris said, it's identifying the applications is important, but also trying to understand how the efficiencies can be found, so that they can run most usefully in quantum computing is important. I think there's a bit of a debate to be had around how application specific, to make the hardware. I think certainly that's something that you see around the community, different entities, making a bigger deal out of it than others. It's certainly quite risky because it's a lot of investment to develop any kind of quantum processor. And so, of course, you want to make it as general as possible. Nut at the same time, at least understanding the applications, understanding how at least hardware can be tuned towards best meeting those application needs. I think there's certainly an argument to be doing that as you're developing the architecture.

Architectural choices for a scalable quantum system

Gwendolyn Regina

Thank you, John. So let's kind of go back to really the hard questions regarding hardware. So a question to both Chris and John, again, are there specific architectural choices you have to make when designing a scalable system?

Chris Monroe

It's hard to prove this, but we're talking about efficiencies. You can use the architecture to make the system more efficient. And I kind of hinted earlier to compress a circuit so that the system can run long enough to do something useful. That you could argue, that's architectural. You can take advantage of, for instance, our system, we have very high connectivity that is we can do, we can do an operation between any pair of bits in the system even to scale. But I think the one architectural feature that I think will be necessary is modularity. Being able to stamp out small module quantum computers and hooking them up together. And once you do that, and again, this is nothing new, I think any complex system has modularity. Look at the airline hub system, where the airline doesn't fly between every pair of cities, they have hubs, to save costs, maintenance, and so forth. I think quantum computers will, of course, the bigger the system, the more open it is to having errors being observed by the environment. So by making it smaller, but somehow allowing connections that maybe you don't use so much, because you're limiting connectivity by doing this, at least you can still scale. So I think that's an architecture that will probably find its way into algorithms. When you look at an algorithm, you'll notice, oh look, only this 10% of qubits are doing a lot of work. Let's make that one module. So that's a black art on how you tie a particular algorithm application to the hard work. And I think we're going to be absolutely reliant on that in the early years, until qubits are a commodity and cheap, we're going to have to co-design the algorithms and the architecture with the hardware we have at hand. Only by doing that, you'll be able to get to a point where qubits, and gates are a commodity and you can just scale it. Who else sort of like, we, we take 20 megabyte pictures on our phones every day. I mean, that's a total waste. We do it of course, as memory is cheap, and the silicon behind it is early high yield. And so we're not there yet, but architecture is going play a huge role and I'm not an expert in this field. This is, you should ask somebody like Joe and who knows much more about architecture. But I do realise as a hardware provider, we can't be ignorant of hardware, of the architecture and people running and inventing algorithms, they cannot be ignorant of the hardware. They really, really have to have this vertical flow. And the people at the very top might not care about silicon, might not care about quantum physics. People like me at the very bottom normally might not care about what's above the API.

Joe Fitzsimons

So, actually, if I could jump in with a question I wanted to ask Chris and John, what do you see as the role of modularity? Do you see the effort to build a quantum computer five, 10 years from now as focusing on monolithic processors as we are effectively today? Or do you see there becoming more specialised parts of a system? So at the moment, most systems tend to treat bulk qubits equally. But if you look at a modern computer system, you have different systems for storage, then for processing, you also have specialised parts of a processor for arithmetic or different tasks that require different connectivity. How do you see the technology evolving, say, over the next 10 years?

John Morton

I think it's a really interesting point and yeah, absolutely we change our information between magnetic domains to charging semiconductor, to photons down a fibre, without even thinking about it in conventional computing. And certainly in the future, maybe 10 years, maybe beyond in a mature quantum information industry, certainly one could envisage that sort of fluid transfer of quantum information. But I'd like to stress, some people think when the kind of 1980s era of quantum computing, I think we're more in the sort of 1950s era of quantum computing where you had valve computers and mercury delay lines. I mean I wasn't there, but I've heard of it these things.

Joe Fitzsimons

I think it's like 1947 or 1948, If we are in the fifties, that’s doing well.

John Morton

Yeah. So I think all of that we can see is to come. But really now, I think that for developing the sort of first useful quantum computing applications, I don't see that really as a problem that's going to be solved by using quantum memory in these kinds of things, perhaps. And it's really about taking some of these systems and optimising them as well as we can. I mean, the modularity is an interesting point. I can see some advantages, that to go back to the jet engine example, that's not modular. You kind of build a jet engine. Of course, you can strap a few of them together and make the plane go faster. And you can do that with separate quantum processes without having to make them modular in a quantum sense. It can help with scaling, but there are very many extremely complex systems, including computing systems, which don't necessarily have that modularity. And so I think it's not a fundamental requirement for quantum computing. There are different ways that you can enable that sort of scaling.

Joe Fitzsimons

So in the context of something like an ion trap, this doesn't necessarily mean that you would have multiple different technologies connected together. And ion trap to two superconducting qubits or something like that. But it might be as simple as multiple species in a trap, but I say simple, realising that this is not simple at all. I'm just wondering, I guess, Chris, what your thoughts are in terms of, if we're moving towards things like segmented traps, you can think of perhaps of different trap regions for different functionality and how that kind of thing goes into your thinking or whether, or it doesn't at all, and if it's better to avoid those complexities?

Chris Monroe

Oh yes. We of course think about that sort of phase one of the scale-up is to, we deal with a single core, we call it of up to maybe 32 or so qubits and they just sit there. They never move. We have a bunch of laser beam photons. And we have full connectivity between all those 32. And of course, with 32 of anything or something like 500 pairs can do any in 500 qubit operation and have to get well over a hundred, we think about having multiple cores on a single chip, like the one behind me, where we actually move atoms to clean them. Now when you do that, you are limiting on activity. You can only start to hit the end ions, the end qubits, go back and forth. And that should get us maybe 2000 or a few thousand qubits on a single chip. But then the launch to this modular photonic connection allows us to go off chip to another chip. And you might think that limits kind of activity as well, but actually, no, we retain full conductivity because each chip has one or more fibre lines that comes out, and they all go into a cross-connect switch. So picture the old telephone operator that's taking any input fibre matches. We can match any air chips. So maybe we have a thousand chips all on a single rack mount like a data center, and we'd have full conductivity, any qubit in our system. So in our platform, that's the natural way to scale. And you mentioned, okay, so there's shuttling. We need multiple species. And this, when you shuttle that process heat, in a sense, we have to quench that. We also need multiple species to make the photon, making the photon is a process that can have any, even the single photon in memory nearby. So we have communication qubits, and memory qubits. So this has been done. All these demonstrations have been done in university labs, mostly, at IonQ we're doing all that stuff. But I would say that when you say different systems, a photonic qubit is something we actually don't use in a single core on a processing unit, but it is a quantum system that has the advantage of being able to go through fibres and room temperature without loss, or small distances. So that that systems are definitely going to juxtapose with our atoms going all the way to solid state. And there's wonderful research on doing something optically active. Interfaces, super conductors can be needed to migrate cavity that has a mechanical gear that moves. This is all researchy. The fidelities are very low. It's a wonderful research. And eventually I think there are ways to optically link even superconducting qubits, but I think we're a long way away before these different technologies will make one big processor.

Joe Fitzsimons

Yeah, there's some really interesting work from Andreas Waldorf on superconducting qubits, where they've made this giant tunnel between two dil refrigerators, which was not the way I was expecting to chop the two chips in different dil fridges.

Going from dozens to thousands of qubits

Gwendolyn Regina

So this might be a really great link to the next question, which is along the lines of what all of you kind of just discussed. How'd you go from demonstrations with dozens of qubits, so in many labs now, to production level systems, so thousands of qubits, I guess, some of you have said that with a level of reliability and reproducibility for use at scale. John?

John Morton

Our approach here is been tried to follow as closely as possible a manufacturing process that already produces large numbers of devices, right? Not thousands, but billions of transistors within a narrow range of operating parameters on a single day. And that for us has been core to our strategy, trying to use that sort of platform to develop the qubits, looking at what are the right building blocks, the sort of cells, qubit cells within that process. And see that as a route towards scaling up. But like all of these things, there's a lot of risk involved. These processes weren't designed to develop qubits. They were designed to do something quite different. So it's certainly an engineering challenge, but we think it's, we can show it to work. Then it's a very promising route.

Gwendolyn Regina

Thank you, John. Chris, how about you? Any thoughts on going to production level systems, linking more stuff?

Chris Monroe

It’s easy for me to answer this way, but again, we don't worry about qubit reproducibility, after all they are atomic blocks. What is a block, if it's not reproducible, that's the definition of a block and these blocks are way better than we need. We have indefinite idle coherence, within historic coherence for as long as you need to. So the manufacturing, we don't manufacture atoms. Joke is, I have vial of Terbium metal here. So I have a million, trillion qubits sitting right here on my desk, and they're all exactly the same. They're the same isotope, Terbium 171. So that's really not the point, either. Any kind of manufacturability is definitely in our roadmap at IonQ. We have a very credible, we're very confident to hit this roadmap and ability to make these modular quantum processing units. So we don't worry about the quantum part of it. It's a pattern I've said it for the last hour that it's really all about manufacturability of the optical system around it has nothing to do with quantum mechanics. It has everything to do indexing optical fibres on chips like the one behind me and getting the manufacturing process to be reliable. We have a long path ahead, but it is very well defined.

The role of quantum software in addressing hardware challenges

Gwendolyn Regina

So, kind of one last quick question to link hardware and few of the stuff they talked about and also software, right? So, you’ve spoken about the engineering challenges, errors and stuff like that. Do you see the role of software in addressing some of these problems, or are they purely hardware challenges?

John Morton

We’ve already talked about software in terms of developing the applications and more efficient applications. I think many times there are other ways we use software in terms of optimising qubit control, feedback and end calibration to optimise the qubit behaviour and interpret the data cell. And indeed there are other startups that focus on this, like Q-CTRL, Benchmark and others. So softwares in various forms enters the different layers of designing, operating and optimising quantum processes as well as of course within the quantum algorithm side.

Gwendolyn Regina

Thank you, John

Chris Monroe

I’ll add to that. At IonQ, given our platform, we don’t have wires. Atoms are floating in space. There are no wires. There’s nothing hardware. Everything is controlled by software. Absolutely everything. So I’m fine with saying around here that in few years we’ll be a software company. It’s true that we’re going to have this exotic atomic physics and optics going on inside. But everything from error correction to how many qubits you want per core, how many cores you want per chip, how many chips you want in your whole system, that all will be controlled by software entirely as there’s no hard wire. So this flexibility will be a real bonus in some years in the business world, in software companies, very less cost and very high margins that’s going to play a role in quantum computing as well, and we already have a big software team at IonQ. We didn't talk much about error correction but how much error correction do you want to do and we can control that in the software world.

Gwendolyn Regina

Thank you, John and Chris but before we conclude, I would love to figure out and bring back a quote. So back in 1943, so earlier you know both Joe, John, you were kind of like joking we=here were we in terms of quantum computing, 1950s, 1940s. So, 1943, we all know this quote but Thomas, president of IBM said, “I think there is a world market for about five computers”. So the question at all 3 of you, what’s the world market for quantum computers? Oh, John is shaking his head. Impossible question right?

John Morton

I think that quote proves that it’s a very dangerous question to answer in 1943 and probably today as well.

Joe Fitzsimons

Which is the reason to get you to say this in a recorded form so we can bring up 20 years, 30 years from now

Gwendolyn Regina

Exactly, exactly Joe, Chris, no thoughts?

Chris Monroe

It’s going to be way more than six.