Programming Throwdown - 147: Quantum Computing with Yonatan Cohen

Episode Date: November 28, 2022

Yonatan Cohen – Co-Founder & CTO of Quantum Machines – joins us in this episode to tackle quantum computing!  Did you know anyone can run quantum programs on Amazon Web Services for mere dollars? Learn about this field early to take pole superposition in the race to understand and use quantum computers!00:00:45 Introductions00:01:20 Yonatan’s beginnings00:03:49 The simulation question00:05:51 How physics led to quantum computing00:14:56 Richard Feynman00:16:44 On the irreversibility of normal computers00:21:25 Logic gates00:25:04 Qubits00:30:11 An example of qubits00:38:19 Why simulating a quantum computer matters00:42:23 NP-complete problems00:48:57 More people at a higher development level are needed00:54:16 Quantum machines in the middle layer01:02:56 Working at Quantum Machines01:05:05 FarewellsResources mentioned in this episode:Links:Quantum Machines:Website: https://www.quantum-machines.co/Careers: https://www.quantum-machines.co/careers/Yonatan Cohen:Linkedin: https://www.linkedin.com/in/yonatan-cohen-10076b113/References:Getting Started with Quantum Computinghttps://builtin.com/software-engineering-perspectives/how-to-learn-quantum-computingIf you’ve enjoyed this episode, you can listen to more on Programming Throwdown’s website: https://www.programmingthrowdown.com/Reach out to us via email: [e-mail] can also follow Programming Throwdown on Facebook | Apple Podcasts | Spotify | Player.FM Join the discussion on our DiscordHelp support Programming Throwdown through our Patreon ★ Support this podcast on Patreon ★

Programming Throwdown Episode 147 Quantum Computing with Yonatan Cohen. Take it away, Patrick. Hey, everyone. Welcome to another episode. Today, I'm excited to be able to have someone to talk to about quantum computing. This is something even dating back many years that I've been intrigued about. I've read various superficial descriptions of, but never really been able to go in depth about with quantum computing and I'm excited that we're here today with Yonatan Kohin, co-founder and CTO of Quantum Machines. Welcome to Programming Throwdown. Hi, it's great to be here. Thank you for inviting me. Awesome. Well, we normally get started by a little bit of an introduction about yourself, how you got into technology. Some people have like that one

time they remember that, you know, maybe their parents brought home a computer or they went to school and the Oregon Trail was put in on a floppy disk. Do you sort of remember your first like exposure to computers or technology? To computers in general? Yes. So yeah, I did. I did get a computer at a very early stage. And I was also very interested in physics, in fundamental physics, since I was very young. But really, I got into quantum computing when I was at university. So I did my bachelor's degree at the University of Washington in Seattle. And I had a great teacher. So it's all about teachers, you know.

So a professor at at University of Washington. His name is Boris Blinov. And he was this great guy and he was teaching quantum mechanics 101. And that kind of blew my mind because it kind of connects computing and fundamental physics, which are two topics that I truly loved from a very early age.

And that kind of like from a very early age. And that kind of like brought things together for me. And since then, I'm hooked on this subject. Oh, nice. You know, it's interesting. There's actually like a lot of people we talk to who kind of have this physics, computer science, like hybrid, you know, kind of crossover, I guess, like the ways of thinking maybe are same. I mean, there's the obvious, like computers use physics to work, but like, I mean, I feel like still,

there's a lot of, a lot of crossover there. That's like a pretty common background. Right. I mean, it's, it's interesting because, yeah, I mean, the reason I went to physics and not to computer. So when I was, when I was very young, I was obsessed with, so how does a computer work? You know, like when you're a kid, you know, why do I see those things on the screen? Yeah, like how does it all works? And then I took a few courses and at some point,

and I was obsessed also with how the universe works, you know, like why do I see all these things around me? You know, so it's kind of like the same thing, just replace, you know, everything that you see around you and the screen of the computer. So apparently, i think it's easier to understand how computers work than than how the universe works um so i feel we're gonna go deep right out the beginning yeah so sometimes during high school i realized okay yeah i think i understand how a computer

works so i better just go in the physics direction. But once I learned about quantum computing, I think it's interesting because I think that quantum computing is really, so some of it is about, you know, what's possible to compute, you know, is to compute, you know, with certain amount of time, certain amount of resources. And that sort of goes back into the kind of fundamental how the universe works thing you know uh for me oh man are we gonna ask that question are we all in a simulation is it are we headed to the end so we're just gonna tee it up right at the beginning for sure um yeah i mean no i mean i don't know if it's simulation but you you could view you know the laws of physics is kind of like uh you know some kind of a very sophisticated program and you can ask what, the laws of physics is kind of like, you know, some kind of a very sophisticated

program. And you can ask, what are the rules of programming? And these are the rules of physics, right? And quantum computing builds on that. And another interesting thing is that there are a lot of interesting thinkings right now, which might be the first applications of quantum computers as to how quantum computers can allow us also to understand deep concepts in physics. So now it's going back, the direction goes back, and maybe these computers can help us understand fundamental physics better. Wow. Oh, that's cool. Yeah. It's interesting sometimes how those things go in cycles and iterations and sort of one helps the other, the other helps back, and it's kind of fun that the other, the other helps back. And it's sort of,

it sort of, it's kind of fun that way to, to see those cycles happen. Yeah. Sorry to jump into the philosophy of quantum right away. Yeah. We're just going to roll with it. So what I'm going to say is like,

I think a lot of people maybe have those questions that, that you're saying, like I think, and a lot of people listening to this podcast are self-selecting as a potentially the people who ask those questions that you're saying. Like I think, and a lot of people listening to this podcast are self-selecting as potentially the people who ask those questions. But I don't know that everyone pursues the answers with the same, you know, sort of vigor, or you can't pursue every question you have equally, right? Like you would just get nowhere, just navel gaze, right? So like, but I think a lot of people have that. So how for you did you go from like, probably a relatively common question of, you know, why is my screen lighting up to or how is my computer choosing these pixels to like, you know, actually going in and starting to learn physics and quantum mechanics and then rolling that into choosing that to be a career? Do you feel like you sort of was that an intentional thing?

Did you sort of like fall into it? Was it just pursuit of those questions? Well, sort of. I mean, you know, I believe that you kind of roll in life. So it wasn't really, I don't see it as a single choice, but I think, you know, things lead to one another. So, you know, I was very interested in physics. I went to study physics. And then once I was exposed, I said to quantum computing, I felt that this is, you know, really a way to look at fundamental physics in a kind of eye-opening way.

But then I actually went and did my PhD, which was in condensed matter physics, which is related to quantum computing, but not exactly. And in parallel, at some point, I also, together with my partner, who is today the CEO of the company, we wanted to be entrepreneurs. So we felt like we got to understand that as well. You know, how do you start a business around technology? And one thing led to another. And then, you know, we realized that the only thing that we really know is quantum stuff.

So that led to, you know, starting a company in quantum computing. And that's what really got me into, you know, choosing it as a career path, you know. Well, I feel like when most people say, I'm going to make a startup, they mean like, I'm going to make a website. Like I'm going to go and I'm going to download a LAMP stack. And maybe even today I'll run on an AWS, a lot easier today.

And again, a lot of people probably have that entrepreneurial, maybe I could do this. Maybe I can make an app or a website. But you said, well, the thing that I know best, I'm going to go start a quantum computing startup. Help me understand that. That feels like such a big leap. Maybe saying condensed matter physics as a PhD. Maybe that's the thing I'm missing. I didn't go that route. But help me understand, is that something you felt that there were particular problems you knew that needed to

be solved? How did you decide that that were particular problems you knew that needed to be solved? How did you decide that that was going to be a successful endeavor? Well, it's an interesting story. So basically, it's never like a straight line, right? So we knew we wanted to start a company. In fact, we didn't start from quantum computing, to be honest. We had lots of ideas, machine learning, you know, like the usual stuff, as you said. But at some point, well, we realized that, well, we realized two things.

First of all, it was about five years ago when quantum, you know, the quantum industry started to evolve. And then there were a few things that happened in the quantum community that kind of got us thinking, oh, well, Quantum is really starting to happen. You know, something is happening. So, and really at the same time, we realized, hey, this is really the only thing that we know. Like we don't really know how to build the best website. That's never stopped anyone before. That's true.

But yeah, we realized this is our expertise. We still didn't know what we wanted to do in quantum. And that's when we called our third co-founder, who was the chief engineer of the company, who did his postdoc at Yale University in one of the top quantum computing groups in the world. And that's when we kind of started sitting together, the three of us, and thinking, you know,

what do we want to do? So Nissin, who went to Yale, also was the first person in history, actually, to perform the first experiment that demonstrated what's called quantum error correction. So Nissim was the first author on a paper that is really one of the biggest milestones, I would say, I don't know,

the last, let's say, decade in the field, because quantum error correction is the kind of like mainstream way that the community believes that we should take in order to scale up quantum computers, build truly large-scale, full-stack quantum computers. So Nissim did that. And then we were sitting together and thinking, what are we going to do? And some of the things that were required to do this experiment were things that we felt are really bottlenecking the field. That's what's called... So when you look at a quantum computer, it's built out of layers. There is the quantum hardware, and then there is the classical hardware

that talks to the quantum hardware. And then there is a software. And the classical hardware that talks to the quantum hardware is not quantum, but it's very complicated classical hardware that needs to be specially designed to talk to the quantum hardware. And we felt that everybody's working on the quantum hardware, but there are many bottlenecks, especially if we look at two, three, five years, 10 years from now, in this classical hardware, what's called the control system also. So we realized that's where we want to focus on. And that's what we did.

Awesome. Maybe to roll it back a bit, I guess, like you're saying, I have this feeling too, not as being anywhere near the field, but the number of articles that make sort of mainstream media, you know, so-and-so company has X number qubit new record or had a quantum entanglement for X time or all of these things I feel are coming out, you know, more and more frequently where people are beginning to say, hey, maybe we're on the cusp of this, I don't want to say like a revolution, but like this fundamental change where now it goes from an experiment

to a practical usable thing. And I'd love to get your thoughts on there, but maybe we queue them up for sort of later in the thing. But like maybe rolling it back to, you know, kind of like the beginning, I think like people kind of understand or maybe they do the,

there was the ENIAC and there were the code breakers in World War II building. But I would say leading up to where we are with the computer we're all using right now, the sort of classical computer, the architecture, I think the term would be the von Neumann machine. We have like a CPU and it has registers and pulls from RAM and processes things in a pipeline. You know, the thing that we all, you know, learn to program when we go to school and learn computer science. But at that same time, like some of those things that unlock that we're coming out of the field of physics, things like vacuum tubes and transistors. And that was around the same

time, I think, like when that quantum stuff, I guess, for me, it's always those quotes back to, you know, Niels Bohr and Albert Einstein, not that I know anything, but it's like, oh, God doesn't play dice with the universe, right? Like quantum mechanics has to be wrong. I think that might be a bit of a superficial take on his belief about that. But I mean, you hear those arguments about these people that today are sort of shortcut names for people who are smart, right? Einstein, Niels Bohr, these are like the father.

These are just stereo archetypes of like geniuses, right? Creating these fields but maybe kind of walk us how we got from like that there is some kind of physics that isn't classical or or kind of newtonian or relativistic like i'm saying but this quantum thing and how it kind of ended up getting us to where we now kind of fuse these two together quantum computing that was a lot sorry yeah yeah no i mean that mean, I think that's one of the most interesting topics to talk about. I mean, so the history of this field is very interesting because, so as you said, I think that quantum mechanics was, you know,

was developed pretty much during the same, you know, several decades where the first concepts in computing were developed. And that was independent. And, you know, sometimes in the 80s, these concepts started to come together, which I think is amazing. But if we go back to the beginning of quantum mechanics, so quantum mechanics, what it really tells us is that, you know, okay, so we have this, what we thought the the laws that we believe that that that the nature obeys um so call it you know if we want to talk in in programming um programming

language then you know what what are the what can i program you know what can i even program what's possible you know and then monomechanics it expands that. Expands it in ways that are a bit weird to grasp when you learn about it. But it actually expands it, I think, in a very elegant mathematical way that makes very complete mathematical structure. Maybe more complete, in fact, than classical physics, even though there is still a lot to discover. And then so people, you know, explore that for many years, while at the same time, kind of computers happened. And I think these two, so quantum mechanics, you know, completely revolutionized how we think about nature, while classical computers completely revolutionized our day-to-day lives,

right? And what we can do with nature, with the classical nature. Computers completely revolutionized our day-to-day lives, right? What we can do with nature, with the classical nature. And then sometimes in the late 70s, beginning of 80s, some people started thinking, okay, these concepts, you know, they're related in some sense. You know, it came from several directions. One was, can I simulate a quantum system using classical computers? And that turns out to be very, very difficult. So if you're trying to simulate certain quantum mechanical systems, it just takes forever on a computer. So you ask,

okay, so nature simulates itself very quickly. So why can't I do it with a computer? You know, we have this concept of a Turing machine, you know, that is a machine that's equivalent to all other computation machines, right? But no, I cannot do it with quantum mechanics. So immediately, well, not immediately, it required Richard Feynman, which is one of those people

that sounds like someone smart. And so he asked this question, you know, okay, so we cannot simulate nature efficiently with classical computers. But he took the next step also, and he asked, so should we build a quantum system that can simulate nature for us? And that is sort of the first concept of a quantum computer. It was also Paul Benoff that was giving talks about the computer

as a physical system. Thinking about the computer as a physical system, what are the fundamental limitations of a computer in terms of generating heat, the fact that classical computers are irreversible. You cannot reverse the computation while quantum mechanics is actually reversible. Everything that happens in quantum mechanics, you can reverse. So all of these concepts sort of brought people to start thinking about, can we build a machine that is by nature quantum and is a general computation machine? And the first person to actually bring it down to the language of computer science, I think was Eliezer Deutsch, who proved that there is a concept

of a universal quantum computer that could basically simulate any other quantum system, just like kind of a Turing machine does to any other classical system, right? And that sort of just expands all these concepts well there's a great tour there's a couple i have a lot of questions but i mean i think like no that was that was awesome overview one thing that you said i actually hadn't heard before but i guess like now i'm trying to puzzle it out in my head which is i think maybe maybe people don't know but this thing you mentioned

about irreversibility of normal computers, right? I think that when you have, if you think about logic gates, which underpin this, I'm going to try, you can tell me if I get it right or wrong, but like have like a CPU and the CPU is composed of logic gates. And if you send, you know, two bits into an AND gate, you're an OR gate, one and zero, one and one, zero and zero. And then you get, you know, the truth table, what comes out the other side, that having a one on the output of an org gate doesn't tell you what the inputs were. So some amount of information is sort of lost. I think that's related to sort of that Maxwell's demon and this like, you know, Oracle allowing, you know, and

the information theory, like you're destroying that information, some of the information is lost, because these things aren't reversible. And it tries to set, I guess, a lower bound on the amount of energy it takes to compute certain equations, because you have this amount of information loss. And it's not that you can compute it for that, like we actually, it takes way more energy than we than energy than we are calculating that way to do it. But there's this idea. So maybe folks hadn't heard about that. But you know, I think you get into that if you study computer science and how CPU architectures work and sort of ideas for getting around thermal limits and these kinds of things. But I actually hadn't heard what you were saying, which is quantum computers are reversible that this idea

that actually they don't have that same issue yeah exactly so let's start from classical computers classical computers have you know they have two main building blocks right we have the bit that's how we or the bits that's how we store information then we have the gates uh which is how we process information right and so every program, essentially, at the end of the day, fundamentally, is a sequence of gates performed on a bunch of bits, right? And as you said, those gates are, the way we do it at least now is, yeah, they're irreversible. So we lose information along the way. And that actually generates heat that goes back to you know computer computer as a physical system whenever you lose

information yeah you generate it yes we are very far away from this limit by the way right now I think we're several orders of magnitudes from reaching you know so most of the heat that the classical computer generates by far is not coming from that but that sets sort of the fundamental limits of how we do currently classical computer. And people started asking whether you could build classical or quantum systems that are reversible. Now, quantum systems, they replace the notion of a bit with the notion of a qubit, which is not a system that can be either zero or one. It can actually be in both zero and one at the same time with certain weights. We call those amplitudes.

So we say it's in a superposition of zero and one. It's, I don't know, a little bit in zero and a lot in one and vice versa. In fact, those amplitudes, these waves can also be negative numbers. So it can be minus 0.1 in the zero state and whatever, plus 0.9 in the one state. But then you operate on these qubits with quantum gates. And quantum gates, yes, they're always reversible. So we call this unitary transformation. But so every quantum gate has, you know, a certain number of bits going in, the same number of bits need to go out. And along the entire quantum circuit do because it's reversible. So that was the first thing to prove that a quantum computer could do everything that a classical computer can do. And it can.

That's a good answer. I was going to ask. The answer is that yes, it can, which also shows you that you could build a reversible machine that does, that's, you know, equivalent to a Turing machine. So you could do that, which is a great, I think, very interesting concept by itself, unrelated to necessarily, you know, quantum computing. So you can build a reversible machine that does whatever an irreversible machine does.

And then in addition to that, we can also show that because of the quantum nature, actually because these amplitudes are not just complex numbers, we can also take advantage of this and do more than a classical machine can do. So, yeah, I have a ton of questions about qubits as well. But actually, oddly, I'm going gonna skip those for a second and ask about the gates because so i guess like on this journey we talk about logic gates and most people have seen the picture of an ore and i shouldn't say most people hopefully you have if not go google it and you just sort of see the picture but it's also maybe you know 15 minutes you could explain

someone like dope semiconductors in certain ways and have them reason through how like a higher voltage potential and a lower voltage potential and a third, you know, and how a not works or how an or works or how an and works. So maybe people don't know that I shouldn't assume, but like, you know, I feel like that's readily accessible. Like I don't have a background in semiconductor physics, but like I have a rough working understanding as a practitioner in the field today that like there are these semiconductors and there's gates and they're doped and they're you know how i i it would take me a long time but you could sit down and sort of draw gates and figure it out and work your way through it but when you say that you have a qubit

in a superposition and then a quantum computer gate i just don't even know like is that it what is that gate is it a physical thing is it semi kind of like how like what does that actually look like yeah uh that's a great question people don't talk about it enough so it sounds like uh so it sounds you know inside a cloud of fog but it's actually pretty straightforward so the the way it works so okay there are different implementations of qubits today but but most most qubits are what i call stationary qubits so this is a physical system that sits somewhere in space you know it could be for example an ion or an atom that's trapped with laser beams so imagine you have a bunch of ions sitting in a row, some vacuum chamber, and these ions are the qubits. Actually, what the zero and one of the qubit are two orbitals in which the electron can be. So if you remember from high school chemistry, you have an atom.

So inside you have an electron and the electron can be, it circles around, but really it can sit on different orbitals, right? It could be one orbital or the other, but it can also be in a superposition of those orbitals. So this is our qubit, basically. We take two of these orbitals, and one of them is our zero, and one of them is our one state, okay? So we have two states. And now, as you said, we want to start to apply gates to the system. Now, you know that physicists like to shoot laser beams. So we do it as well here. And so we shoot a laser beam. Here, I actually do it for a reason.

So we shoot a laser beam on the ion. And that actually causes the electron to move from one orbital to another, okay? And if we do it just in the right way, we shape the laser beam correctly, it can also make the electron go into a superposition, and it basically applies the gate that we want, okay? So we can apply any quantum gate by different shaping of these laser beams that hit these ions. Okay. So it has to interact with two though, right? If you're going to have like not operation on one or you only perform it on one.

Like I guess, is it gate here, a single qubit and you're applying like, maybe I lost it. Maybe I... No, no, no. Perfect question. So that's that. Now what I described was exactly a single qubit gate. Okay, okay. So when you think about it, so think about the qubit.

One way to think about the qubit is, so a bit is zero and one, right? So you can have an arrow that points down or up, yeah? A qubit is like an arrow that can point anywhere on the surface of a sphere, okay? Okay. So that actually, this is an accurate mathematical description. So I can describe the state of the qubit as a point on the sphere. Yeah? So in contrast to the classical bit where the only single bit gate that you can perform is a not gate, right?

Just from zero to one one to zero um a a single qubit gate we have basically infinite number of single qubit gates because yeah so we can perform any rotation on the surface of the sphere right um so so actually the power of the laser beam as well as the phase of the laser that we apply to this qubit, to this electron inside the ion, that's going to determine how fast we are rotating around the sphere. And the phase will determine basically around which axis we're rotating. So then you can really create any rotation on the sphere. Oh, fascinating. Yeah.

And this is just a single qubit gate. And of course, as you said, now I need to, okay, of course we want qubits to interact. We want to, this is one of the most important things in quantum because we want to create entanglement. We want to have the entire system go into this like single quantum state. And for that, we need two qubit gates. But that's, you know, it's a similar idea, you hit those ions with some laser beams, and you bring them to interaction. So you have to

have some physical mechanism in which they interrupt. But this physical system mechanism depends, you know, on external forces like these external magnetic fields, electromagnetic fields, and so on. And this was about ions, but I could do the same with superconducting qubits, which is another leading platform where, you know, you create a small circuit from superconducting material like aluminum, for instance. And when you cool it down to very low temperatures, it starts to behave quantum mechanically, and then you actually can create a system which is kind of like the ion. It actually, it's a circuit, it's a macroscopic circuit that we can fabricate on a chip,

just like you fabricate transistors on a chip. You fabricate it on a chip, but it's like an artificial atom in the sense that it has those energy levels, quantum energy levels, just like the orbitals in the ion. And I could send a microwave. Now, it's not a laser pulse. It's a microwave pulse because it's just different energies, but it's the same concept. I shoot a microwave pulse through a wire that's connected to this chip,

and this microwave pulse basically hits those qubits, those superconducting qubits, and perform the gates, and I can perform two-qubit gates, single-qubit gates, and then I could create what's called the universal set of gates, which is a similar concept to the NAND gate in classical computing, which says that from a discrete number of quantum gates, I could actually create the universe of quantum computing. Well, you just built them all for us. I mean, you said a lot of things that are pretty sophisticated.

I don't want to get there quite yet, but I'm seeing your setup about the control systems for this, you know, lasers and controlling phase and intensities and I assume frequencies. And like, I mean, I think we're going to, I see where this is headed. But before we go there, you mentioned that there's sort of like superconducting, that there's ion, there's like different approaches. And it sounds like you're saying that it's not clear yet, necessarily, that one is better. Are there trade offs that are being explored? Like, one is better for this thing, the other is better for this, or it's just, everybody's just in a race to build the first one that works?

Yeah, exactly. I mean, we don't have kind of the silicon transistor yet of quantum. So there is a race now between many different kinds of qubits. I'd say there's at least five, six leading candidates to be the kind of dominating qubit. So that's happening right now. And some of them have, yes, advantages in some areas and some have advantages in other areas. So it's hard to say who's going to win this race. And there could be also combinations.

So when you think about a modern computer, actually, there are many different ways in which we implement bits. So some implementations are better to create, you know, a hard drive, and some implementations are better to create the bits that go into the CPU and so on. So it could be that in the future, we're also going to see some hybrids, you know, quantum computers that are based on different types of qubits. Yeah, I never thought about it that way, but I guess you're absolutely right. Like an SSD or flash and DRAM and a computer all have like slightly different kind of mechanisms for representing like zero or one. So, yeah, I guess from that standpoint, yeah, you're absolutely right.

And they have tradeoffs. It's not that only one is correct. There's a variety of reasons you may choose one or the other. Right. I can give you an example, for example. So a trap ion system is the lifetimes or the time that the qubit can actually preserve quantum information are much longer than superconducting qubits. But superconducting qubits are much, much faster. So you can perform the gates about three orders of magnitude faster. So, you know, so that's a huge advantage, right? Because, I mean, okay, let's say, okay, so it's true that a quantum computer should be at the end of the day, exponentially faster. And we don't care about those three factors, you know, the theories, they don't care about those three factors you know the theory the theories they don't care about these three factors because it's not fundamentally a part of you know whether it's in this computational class or another but but these pre-factors matter like

if we build the first quantum computers that can actually do something useful and it can do something that a classical computer does in 10 days maybe it can do it in one day well that's great i solve a problem you know 10 times Well, that's great. I solve a problem, you know, 10 times faster, but that's on a superconducting qubit computer. On a trapped ion computer, it will take a thousand days, right? So at least for the first, you know, decade of the field, I think that this is going to be important. But then on the other end, as I said, so maybe some qubits are better for memory and some for computation and so on.

I know you already kind of, I guess, seeded my thoughts there, but the error correcting, so everyone kind of knows like in a normal, or maybe not, I keep saying everyone. I feel like I have been well, like heard it over and over again in a normal computer, like, oh, you can have a cosmic ray comes in and strikes your memory and flips a bit.

But we also know like, you know, electrostatic discharge or, you know, just like depending on time, something like DRAM needs to be refreshed or you can have corruption. So I think we tend to not think about errors too much, except when your hard drive gets corrupted. But like you tend not to think about, you know, I guess errors too much except when your hard drive gets corrupted but like you tend not to think about you know i guess errors too much and it gets handled and there you know you learn about like crc checks or hashes you know things that are either tell you if there is a problem or tell you how to fix the problem is that's like kind of like the classical computer thing um when you're talking about quantum computers i assume it's sort of analogous. Is it like lifetime measured in units

of time or a number of operations? Like what are the things that cause you to sort of, you talk about like very cold temperatures and high vacuum. So I start to think about things like, you know, well, you can't really have it exactly pure. So there's always like something floating around, you're not exactly cold enough. Is it just system, things like that? Or is there like more fundamental things that are creeping in that cause you to kind of error out and not get the thing you thought you were going to get? No. So, yeah, the errors can come from many different mechanisms and sources of noise. As you said, it could come from simply the fact that there is a finite temperature.

So, you know, finite temperature tells you that, okay, if you're in this energy state, you could jump to a higher level energy state. So you try to isolate more, to cool down more. As you said, cosmic rays or even photons that wander around could give you errors and then the control system itself can give you errors as well because we're not sending exactly the right pulses that

we want to send to the system and there's noise, there's electronic noise coming from the classical electronics and so on and so forth so there are many, so I think there isn't exactly a fundamental thing and I think we could build better and better qubits in the future. I think that's super important to do that. There is, however, a challenge in the fact that on one hand, you want to isolate.

So for a system to be quantum and to be protected from the noise from the environment you want to isolate it from the environment right but then on the other side on the other end we do want to speak to it from the control system right this laser puzzle so we do want to couple it to the environment so there is this challenge where you on one hand you want to couple it to the control system but you you want to isolate it from everything else and even you even from the control system, but you want to isolate it from everything else. And even from the control system, you want to isolate it when you're not sending the passes that you're intending to see. So it's an engineering problem. I wouldn't say it's a fundamental issue, but it is, I think, a big challenge that's going to be important. And again, one of the main ways to deal with it

is to say, okay, I'm good enough now. Let me just now start doing quantum error correction. And that basically deals with this noise. Because what quantum error correction does, which I think is interesting to note, is that it digitizes the quantum information to some extent. So I talked about this qubit, which is kind of like a vector on a sphere, right? So it's, you think about it, it's an analog system. And one of the things that bother people is that, okay, so how can we build this thing? You know, it's analog. So I can have an error, which is a tiny angle that the qubit is not exactly, you know, pointing down, it's pointing

one angle to the right. So that's a challenge. But quantum error correction actually allows you to digitize those errors. It digitizes the errors. So when you do quantum error correction, you either get an error with a very low probability or you don't get an error at all. And that's very important because it kind of does what we do with digital. It kind of does what we do with digital electronics where we're protected from a

certain threshold and then it's very hard to get errors you know so you kind of can lower and lower and lower the error rates to as much as you want basically Do those errors creep in like you mentioned the qubits needing to be in superposition and then they can sort of

I guess like fall out of superposition during the calculation or is it more i don't know i don't know if that's the right analogy but like a measurement thing just like when you take a measurement there's some noise and so if the thing isn't exactly where you thought it was and they're just sort of measurement errors yes all of the things yeah so it's both so the so when we say that the the system falls from a superposition, one of the reasons it falls from a superposition is because it interacts with something in the environment. And in some sense, you can think about it as sort of

it's being measured by the environment. So the environment kind of measures the system and sucks the quantumness out of it because it kind of collapses it to one of the states right and we don't know to which one i mean if we if we could then measure the environment maybe we we know to which one so we can fix it but but but we don't so the so whenever we get coupled to the environment we kind of get get get measured by the environment and that and that is one reason for the noise then you you also have Ls.

When you measure the system, maybe you measure it in the wrong way. So these are called readout Ls. So you get all kinds of Ls you have to deal with. So we've been talking about, I guess, the fundamental qubits and gates. And I'm not going to ask you to describe algorithms that you compose from those things. But they exist, as you mentioned. You've proven that you have the equivalent of like the fundamental one. So you could just, I guess, as a backstop build, like the naive old way of doing things in the new approach. But you also mentioned this sort of like trying to use them to do, I guess, I would kind of take it as like fundamentally quantum approaches to solving problems. And I, you know, I mentioned even in the opening, like, you know,

Hey, we're going to factor large prime numbers. Everyone says, Oh, quantum computers are just going to like factor large prime numbers. All encryption is going to be broken or at least the ones that rely on, you know, prime testing or factorization. But then you mentioned something which I have, I've heard before, but I never really kind of understood, which is actually trying to have the quantum computers simulate quantum computers as like a proof that you're doing like a fundamentally different thing.

Could you maybe like dive in like why that matters? a lot of people in the field nowadays believe that simulating other quantum systems is perhaps going to be the first application of a quantum computer. Because these are sort of also the easier ones, so we can do those. There are interesting questions to ask about how quantum systems behave that we cannot ask because

we cannot simulate quantum systems with many degrees of freedom that we could probably do with relatively low number of qubits. So maybe a few hundreds of qubits, you can already simulate quantum systems, molecules, like phases in condensed matter and things that are actually interesting. So at the beginning, I think they're going to be interesting for science, which, by the way, I think it's already a good application to start from. Let's build a machine that can help us. So first, quantum computers are exploring themselves, right?

Just can I build this thing? How can I build better? How can I scale it up? Then I think they're going to explore other quantum systems, answer some scientific questions that we have. Then I think those things are going to probably evolve into something that is already of some industrial value. Like I can simulate a molecule

and I can maybe build better drugs or better nanomaterials or better materials for all kinds of usages in industry. And in parallel, there is then optimization problems that we hope to get some advantage from quantum computers. And I think those applications like breaking on encryption codes are a little bit more down the road. I think they just require a lot of qubits and a lot of error-corrected qubits. So some of those more, you know,

near-term quantum simulation applications, I think also require less quantum error correction and maybe noisier qubits so i never that was thank you that's super eye-opening i guess i have fallen into this trap about like thinking of traveling salesman problem or np complete problems in general and like quantum computers going and attacking you know all those things that are really hard for computers to do today and maybe opening up new avenues but this thing you mentioned i guess not my field so i don't traditionally think about it as a computer field or my computer field which is which is eye-opening like how i guess if i elaborate a bit like drug modeling or interaction like how a molecule of a certain shape interacts

in your body is insanely hard on computers today. And in fact, like, I guess, looking at the whatever biotech startup failure rate, apparently, if it was easy, they would be doing it. And so that it's not, I mean, this must be pretty hard. But you're kind of mentioning that it may be earlier in the quantum computer thing that you can begin to make progress against those sorts of things which are are not really even i want to say solvable by computers but not practically solvable by computers today so they aren't even really i don't know i don't know if they're used or not but aren't aren't sort of like using to sort of fundamentally solve the problem and so now you're saying maybe quantum computers things

like materials and molecular interaction this kind of thing that's actually that's encouraging because i always get sad this thing you say like, oh, sure, you can solve a 10-bit, you know, traveling salesman problem. So like, that's like, you know, it's not very long on my computer, even though it's slow, it doesn't matter. But you're mentioning, hey,

that actually still could be very useful in a different domain. Yeah, for sure. This is probably the first direction to explore. And these are things that you know. So today, you look at high-performance computers like supercomputers, a lot of what people are trying to do are things like that. So I think that's where we should look into.

Yeah. So the thing with NP-complete problems, by the way, so with NP-complete problems, by the way, so with NP-complete problems, we don't think that quantum computers are going to give an exponential speedup. Okay, so for some of these NP-complete problems, we know that quantum computers could give a speedup, but it's a quadratic speedup. Oh. This actually goes down to how quantum algorithms work and so the fundamental...

If a quantum computer could just explore all of the different possibilities, like sometimes people describe it and say, okay, hey, here is the right path then we can solve NP-complete problems. But that's not the case because at the end of the quantum algorithm,

you actually also end up with a superposition of all the results. And you don't want to measure, you don't, and you can measure only one, right? When we measure, we only measure one state. So in order to get advantage from a quantum system, you need to do more than just explore all the different possibilities at once. And so that leads to the fact that when you have a problem that has completely no structure, which is the nature of these NP-complete problems, then it's also very hard to get an advantage from a quantum computer. So we know we can get a quadratic speedup, which could be very, very useful if we had millions of qubits that are error- and all that stuff that could be amazing you know practically getting a quadratic

speed up for you know ai right now or this would have been amazing so but but we're not there yet so i think we are looking for some of those things in which you can get an exponential speed up at least in the short term and these these and some of the interesting things, I think have to do with simulating quantum systems, other quantum systems. I think I had fallen into that trap. I had heard and I knew it was wrong, but I'd never heard like kind of why,

which is that, yeah, you just put in the superposition, you run the traveling salesman program and it in parallel, it parallelizes the entire algorithm and then out the other end, you know, pops the shortest path. But I think this thing you're pointing out, I guess I don't maybe I miss it, but it feels like it helps is like, you could you can run all of those things in parallel. But then at the other end, when you go to try to find out your

answer, you end up kind of back in the same problem, which is you have to keep rerunning it and keep like looking at it. And you sort of like, do it, but then you have to undo it by like, it's not in a form that you wanted. It is not in the sort of like normal algorithm output. Exactly. So the only way that you can actually get something from the fact that you're running all of those things in parallel is if they form what we call interference. And that's where I go back to what I said. It's very important that these amplitudes, so each path that you go has these weights, right? Like what's the, so you think about it as probability,

but it's not really a probability. It's what we call probability amplitude. It's this kind of like square root of probability. It's a complex number. So it can be negative, for example. And then when those come together along this parallel computation, they interfere with each other. So that at the end, the result that you get at the end is the result of this massive interference. And if you could build the algorithm in such a way that all the wrong paths leading to the wrong computational results cancel one another.

It's kind of like waves in the ocean. They can flatten one another while the path leading to the right computational results amplify one another. Then you get a speed up. This is actually the only way we could discover quantum mechanics, right? Because we always measure at the end. How do we know that quantum quantum systems behave quantum you know to begin with it's because we can do you know probably you've

heard like the the double slit experiments and things like that where you create this interference and that's how we know that the electron went into two slits at the same time and so on and so so yeah yeah that on that one i mean that's a fascinating like not knowing anything but armchair following the kind of drama every time there's a press release from i won't name companies like x company has a quantum computer that did y thing and then you see you know on the the nerdy like hacker news or whatever scott aronson's blog says like yeah bs like that guy we actually can't prove that what they did, like you could have done it classically, like we don't know. And it's like, what, how do you have

this existential crisis that like, you have this thing you're claiming as a quantum computer, you lower all these lasers into a giant liquid nitrogen, but I don't even know it's liquid nitrogen, this giant, like, act of super cool, you know, stuff, and you run this thing, and then you don't know at the end, if what you did was even the thing you attempted to do. Like, it have that super cool, you know, stuff, and you run this thing, and then you don't know at the end if what you did was even the thing you attempted to do. Like, it's sort of this funny, like, in my head, I always, like, oh, man, I guess I don't know enough to know. But it's sort of a schadenfreude, I guess, to watch other people struggle with that. Yeah. No, and I think it's very important to clarify. It's very important for the field to clarify

that there are many things that quantum computers would not be able to do, okay? Or, you know, a test that we cannot prove. It's great by itself. You know, we're trying to build a machine that nobody built before and that we know can do certain things faster than classical computers, which is the craziest machine that we know can do certain things faster than classical computers,

which is the craziest machine that we've built so far, right? There's like trillions of dollars that went into the development of classical computers. I think it makes sense that we will build this machine and we will test it

and we will see what it can do and what it cannot do. And we know certain algorithms like Shor algorithm that can do stuff. But so why do we need more than that? Like, why do we need to generate hype that is based on grounded enough? You know, I think that we have all the right reasons to get funding to build these machines, right? Yeah. Yeah. I won't claim to know the trade-offs of marketing, but I hear you as like an engineer. I say the same thing. Like, why can't we just say truthfully the thing it does like it's pretty amazing so we we kind of talked about like a

lot of really really really low level stuff and and some of these bigger questions and then i know we're talking about like algorithmic design right so people thinking as you mentioned like these things are amplitudes trying to design them to cancel out like are there people who have i don't know like the equivalent of computer science jobs today, like working in these algorithms or programming these things? Or is it, it's still too early? Like, are people working at that high level? And what does that kind of look like? What does it mean to kind of like program quantum computers today? Yeah, so there are people working on it, but not enough, I would say. And I think that's actually one thing that I hope would change.

But the reasons are clear. First of all, it's very hard to come up with a quantum algorithm that actually solves something faster than a classical computer on a fundamental level. Like solves it exponentially or even quadratically faster than a classical computer. That's a very hard problem. And many very smart people tried. And we have, you know, some examples, but very important examples. I mean, if you break the RSA tomorrow, it's huge, but it's very hard to find new ones. The other reason why I think that not enough people are trying is because we don't have

the machines. So, you know, it's a different thing. Like, think about classical computers. Like, think about, like, do you think, you know, classical computers now are what, like, I don't know, so many years, so many decades exist, and still people come up with new paradigms in not just programming, but algorithms, you know?

So there are still new types of algorithms. There are new types of heuristics. There is all kinds of stuff that people come up with, but they have a big advantage. They could just try it out on the computer and see if it works, right? And it's a game changer. So that's why I think that to really have enough people

trying to program new algorithms, I think that we do need more quantum computers, which is another challenge to build those things. talking about the differential engine and Ada, you know, kind of figuring out programming of these analytical engines and trying to kind of work through that. You can talk about the ENIAC computer, you know, or the early computers, you know, cracking the Enigma or ENIAC, these kind of things, like on the spectrum of like, hey, we have a theoretical thing and people are thinking, I don't want to say mathematically, but abstractly about these algorithms. So we mentioned things like Shor's algorithm or these things that people had even before maybe the hardware was there to even try them, you know, up to programming languages are invented. And then, as you mentioned, we keep iterating the ideas, frameworks and programming languages and paradigms and you know functional versus object oriented like we still haven't figured out what we're doing today by any means but there's been this sort of explosion of approaches on that same i don't know that it maps but in that same

kind of like context where would you say like quantum computers lay today like there are some you could run some algorithms on them but like maybe they're like you said not universally accessible where do you feel like we are on that kind of spectrum? Not at the very, very beginning. So that was during the 90s. People had to think about these quantum algorithms completely detached from hardware. So now there are quantum computers and people can access them and program,

but it's still very small scale. So I think that it's still hard to do that if they don't get enough, you know, and, you know, there's a lot of noise, so you don't get the right answer. And I think we're very early, but I think, yes, I think that's why it's great that we start having those, you know, quantum cloud computing services that people can log in and can try. It's not that tomorrow, you know, you're going to have like 10 million programmers,

you know, trying to develop new algorithms, but I think it will increase the number of people who are trying and it helps, you know, educating people about this. So then you kind of start having more and more people going into this field. But I think this is very early stage, yeah. So I have seen those before, you know, log on and submit your job. Does it actually run on a

quantum computer or is it like a VM? Like you're running on the simulation of the quantum computer slower, but the results should be the same? No, you can run today on a real quantum computer. So you can do that using AWS Bracket. That's their quantum computing service. IBM Quantum. Azure has quantum service. So in all of them, you can run either on a simulator or on a real machine. So you could run on a real machine as well. And then you can compare what you get with a real machine and a simulator and you can see the noise level. And yeah, it's great. See how good the simulator is.

So I think we set up this sandwich like, you know, a bunch of low level stuff, a little bit of the high level algorithm stuff. And then if I sort of listen to what you were saying in the beginning about where you guys kind of set up, it sort of feels like it lies in the middle there. Like the practicalness of like not figuring out the next best substance for a qubit didn't sound like although maybe you can tell me i'm wrong and maybe not the like you know application layer stuff you know actually you know sort of writing those algorithms but but kind of laying in the middle do i kind of have the the picture right and like what does that work end up looking like yeah so when you think about so so when you think about a computer, there's a stack, and there are so many layers in the stack, and there are also several abstraction layers that are required until you reach something that you can program and write algorithms with.

And so quantum machines, we are in those middle layers, the control layers, both control hardware and control software. So the control hardware is the thing that's basically orchestrates during the runtime of the program, it orchestrates the algorithm. So that's the thing that actually runs the quantum algorithm in real time, right? Because it sequences the pulses that you send to the QPU. So you can kind of think about the qpu think about it as the alu in your cpu but you need to send it all the commands you need to do it in the right sequence and you need to and in fact with quantum because of the noise you need to do it with the right timing as well which is something that you don't think about in classical because there is no noise

if you push me more into this direction we can talk about the existence of time, but I don't know if you have time for that. That's awesome. So I just looked this up. I went on AWS Bracket and it sounds like basically you could do things that led to the disaster in the Half-Life video game for 68 cents.

That's basically what I'm getting. So you can go to AWS Bracket. It costs 30 cents to get the task up and running. And then it costs 0.0002 per shot. And in their example, that's in dollars. Their example is example two here. A researcher runs a quantum annealing program, which is I'm pretty sure what created the headcrabs in Half-Life 1, on a D-Wave Advantage quantum computer in Oregon, and it cost them 68 cents. So it's just absolutely amazing that you can just do this.

So folks at home, you know, get a quantum program and spend 68 cents doing this. If nothing else, you could tell your friends. It's pretty amazing. Yeah, yeah, it is. It is, I think, very cool and pretty amazing indeed. So yeah, so going back to that control area, so it really is the part that runs the entire quantum algorithm and sequences it. And on top of that, and so it has to have also a programming interface, which is kind of like when you think about it, the assembly level language of the quantum computer, right? And that's what we do.

So we make the control system, which is classical hardware in which we have actually designed a new type of processor, another quantum processor, a classical processor, that gets these instructions from us, these low-level quantum processor, a classical processor, that gets these instructions from us, these low-level quantum instructions, and then it runs the entire program on the QPU, on the quantum processor, against the quantum processor, right? So it's, yeah.

And then on top of that, then you have all the control software stuff, because, you know, this assembly-level language that I mentioned, it's talking the language of pulses. Send this pulse to this qubit, send that pulse, change the frequency of this pulse, change the amplitude of that pulse, but we want to get to the gates, right? We want to get to program gates and algorithms

so you need to have another layer which is called the control software layer in which you run a bunch of complicated calibrations it's a graph of complicated calibrations that run programs on the control hardware and the quantum processor in order to calibrate what are the parameters of the system or how do I perform gate and so on and once you calibrated those gates then you can create a compiler

that compiles from the gate level languages down to the pass level and down to the machine. So if I hazard a guess that the timing and such is pretty complicated, so QPU probably, are you guys using like FPGAs or are you like doing ASIC tapeouts?

Or if you can't share, that's fine. You can hand wave and say black box. That's acceptable as well. I imagine you mentioned microwaves before. These things, the timescales need to be very well timed, very precise. Is that roughly the kind of stuff you guys are talking about as opposed to DSPs? So right now we're using FPGAs to implement what they call the pulse processor. So again, that component that receives the instructions and runs them. And so currently

we're implementing it with FPGAs, the biggest FPGAs in the market. And it's a bunch of FPGAs if you want to control a relatively large QPU with tens of qubits, you would need many FPGAs. So at some point, yeah, I mean, the field will go to a dedicated ASIC to do the quantum control. Currently, it's still in the relatively early stages where you want to get the flexibility and you want to change stuff as you go. And so you do that with FPGAs.

Yeah, absolutely. Yeah, absolutely. Yeah, yeah. That's, I mean, it sounds fascinating. So Quantum Machines as a company, are you guys, so it sounds like you have some of this stuff. Is it something people can try out today? Like, could I just like go to your website

and like learn how to use your stuff? Or is it more like, no, no, no, you got to get in the field and it would serve some of these other things we're talking about? Yeah, no, definitely. So at Quantum Mach in the field and it would serve some of these other things we're talking about? Yeah, no, definitely. So at Quantum Machines, we have our first system. It's been in the market for three years now.

Our uniqueness, by the way, our uniqueness is that our pulse processor actually is not just running quantum operations. It's running also classical operations, and it's running. So it brings, it integrates classical processing into the heart of the quantum control layer. It allows you to program in a single code, which is our programming language, QUA, allows you to program the classical operations and the quantum operations, which are the pulses and measurements, in order to integrate between them. So for example, when you try to do quantum error correction, you need to measure stuff, you need to measure your qubits during the runtime, and you need to process them classically and respond to that again with the next gates or pulses. So to do that...

Oh, it's like path dependent. Yeah, I see. Yes, you need to have feedback. So you need to have very fast real-time feedback between sending quantum instructions, which are the pulses, measuring your qubits, which is measuring signals coming back from the QPU, processing them, doing classical processing on them, which again, you want the programmer to do from the programming environment. And then the algorithm needs to understand in real time that it has to respond with the right pulse

to correct an error, for example. So when you say like controls, you mean like even control loops. So you're like controlling the computer, but then also control, like it's controls all the way down. Okay, I think I see what you're meaning now. Yeah, we're running the entire sequence.

So for example, if I want to loop over a gate, I want to send gate 500 times, I write a for loop to run this gate 500 times. But what if I want to measure a qubit, and based on the measurement, I want to branch. So I want to say, so this is called mid-circuit measurements that affect the quantum circuit itself. So we are the first to introduce such features into a commercial control system that allows, for example, mid-circuit measurements and branching of the algorithm.

So if statement, like when you think about it, an if statement is such a, it's the basic thing right like if x equals zero do that if not do something else that did not exist and still doesn't exist in most quantum computers today that you cannot measure qubits and based on the results decide what you're doing next that has to be done in the classical control system right because that's the thing that measures and branches the the sequence decide the next pulses yeah right so that's what we do. That's the key stuff that we do. And we've been selling. We have over 200 customers around the world.

So we are over 130 people in the company. So yeah, it's great to be in a company that's really in the action of everything that's happening today. So we're not sitting developing something that's going to be used in 10 years. We're really working with our customers very intensively. Are you looking for interns, full-time people? How has Quantum Machines as a company to refer?

I mean, this technology, I could go on way, way longer. I have a lot more questions, but I tried to be considerate here. But are you guys enjoying the place to work? People are liking it. You're hiring. What is it kind of like as a company culture? So Quantum Machines, I think it's a fun place to come work at. We have people all around the world, actually.

We have people in the U.S Actually, we have people in the US, in Europe, and in Israel. So we're based in Israel, but we have people in US, Europe, Canada. We are looking for both interns as well as full-time employees. We're looking for great

software engineers. I know that this podcast is being listened by top-notch software people. So I know that this podcast is being listened by, you know, top notch software people. So I think that's, that's yeah. I mean, and I'm excited to talk here because I mean,

I think that we really need great people, software, also hardware people that are not necessarily coming from quantum, but actually bringing, you know, their expertise into the field. It's very important that we don't let the crazy physicists try to build a quantum computer because that's not going to work

without the the real engineers coming coming uh and helping out so so we're really looking for um great people to that that want to learn about this there's a lot of stuff to do in many layers in the stack. Yeah, I worked briefly for some people doing laser physics. And yeah, I can relate to you. I appreciate what they do. I can't do it. But yeah, that sounds fascinating.

Well, thank you so much for your time. I appreciate you coming on. This has been helpful for me. I learned a ton. I'm excited again. I want to go do a ton of Google searches, which I try not to do during the podcast. You hear the clacking on the keyboard. Jason's doing that for us. But you really stimulated me. So I know people out there are going to love it. Great topic. I feel like it's like, right. Like it's happening. I feel like it's one of those things, you know, it's not like an overnight thing, but like you can kind of see, see the changes, you know, setting up and we'll see, you know, we'll see what happens in the next few years, but like, it definitely feels like a really exciting field. Yeah. Thank you so much. I mean, this, this, this was a very fun to, to come and speak to you guys Jason, Patrick, I mean appreciate that, you guys are awesome and yeah hope to be here maybe a few years

from now when quantum computers are up and running solving simulated quantum systems and other stuff I thought you'd just break crypto and then you'll just be rich thank you so much for your time. And thank you everyone for listening in.

And as always, thanks to our listeners. We appreciate it. And I'll see you next time. Music by Eric Barneller. music by eric barnmeller programming throwdown is distributed under a creative commons attribution share alike 2.0 license you're free to share copy distribute transmit the work to remix adapt the work but you must provide attribution to patrick and i and share alike in kind