Purdue Engineering Distinguished Lecture Series: Will Oliver, Lecture

Hello uh welcome everyone uh to this distinguished lecture series. So beginning in 2018, the per engineering distinguished lecture series invites worldrenowned faculty and professionals to uh encourage thoughtprovoking conversations and ideas with faculty and students regarding the grand challenges and opportunities in their fields. And besides presenting a lecture to a broad audience of faculty, graduate and undergraduate students, they will engage in an interactive panel with Purdue faculty and students. And today we have Will for us uh talking about quantum. But to introduce Will, I will like to invite uh David Bar, our senior associate dean of faculty, college of engineering and professor of materials engineering. Don't clap for me, you're clapping for the next guy. Um, so thank you, Premy. Uh, it's my true pleasure to introduce our speaker for today. First, just a little bit of housekeeping. Um, it's always nice when it's really crowded. Um, so if there there's still a few chairs scattered around, so students in the back, if you are uh not scared of sitting next to somebody, please sneak on down. Um but so it's it's my pleasure to introduce uh Dr. William Oliver today for the Purdue Engineering Distinguished Lecture Series. Dr. Oliver received his PhD in electrical engineering from Stanford University, his mers in electrical engineering computer science from MIT and a BS in electrical engineering and a BA in Japanese from the University of Rochester in New York. He's jointly appointed the Henry Ellis Warren Professor of Electrical Engineering Computer Science and a professor of physics at MIT. He serves as the director of the center for quantum engineering, the associate director of the research laboratory of electronics, and he's a principal investigator with the engineering quantum systems group on the MIT campus. Will's research interests include materials growth, fabrication, design, measurement of superconducting cubits as well as development of cryogenic packaging and control electronics. from 2003 to 2023. He also worked at MIT's Lincoln Laboratory, most recently as a laboratory fellow from 2017 to 2023, where he was instrumental in growing the quantum computing group to its presence levels. He stepped down from that position to co-found Quantum, a quantum computing startup uh which was recently acquired by Google uh this past year. He's a fellow of the American Association of Advancement of Science, a fellow of the American Physical Society and a fellow of ILE E serves on National Quantum Initiative Advisory Committee, the US Committee for Superconducting Electronics and is an ILE E applied superconductivity lead editor. So we are really excited to have him come and talk with us today about exciting things in research in all things quantum and supercomputing. And we will turn it over to Will. So, >> thank you, David. >> Thank you very much. >> Good afternoon. It's really a pleasure to be here. Thanks for the kind introduction, David, and for the invitation to speak with you today. I'm actually not going to talk about um my own research. I I want to talk more generally about quantum computing, where we've come from, where we're going, what's hard, what's likely, what's hype, and what's not. I'm actually not seeing anything on the monitor here yet. Okay, it's up. I've given this talk a few times. Well, while we're getting that going, you may recognize this picture. Usually, it's in black and white. It's not in color. Oh, there it goes. Thank you. Um, this was a survey conference in 1927. And as many of you know, this past year um was the international year of quantum science and technology. And it celebrated a 100red years of the advent of quantum mechanics. Many of the people you see here won Nobel prizes for their work. Uh one woman here won it twice. And we look back and we say that was quantum 1.0. That was when we first understood that nature and the world around us is fundamentally quantum mechanical. This century we're starting a new endeavor and it's called quantum information science and technology. And you might think of it as quantum 2.0. And basically quantum 2.0 addresses the sensing, the communication, and the processing of information in ways that fundamentally use quantum mechanics to do it. And if you had to give it a definition, we would say that quantum 2.0 know does this sensing and communicating and processing in ways that are either practically prohibitive or even impossible to do with our conventional technologies. We'll we'll talk about that in more detail but that's a high level uh definition and I'm going to focus on computing today only but it's broader than just computing. And to have a touch point, I I like to go back and think about classical electronic computing because we're all familiar with classical computers. Mark Twain said, "History doesn't repeat itself, but it often rhymes." And I find this to be quite a useful uh rhyming exercise. Conventional electronic computing started over 100 years ago. The vacuum tube, three- terminal device. It was used for many years as a radio transceiver or in radio transceivers as an amplifier. It took four decades, 40 years to become used in a computer we all know called ENIAC. Now the following year the transistor was also invented and this is where quantum 1.0 becomes important because to make the transistor we needed to understand semiconductors the band structure of semiconductors and we needed quantum mechanics to do that. But once we understood how semiconductors work, we invented this device called a transistor which is manifestly classical. It's a faucet for current. You turn the current on, you turn the current off. Zero and one. Okay. With that transistor about 10 years later, we had a fully transistor-based computer called TX0 invented at MIT. Uh but it didn't look anything like what we're using today. To get closer to today's computers, we needed integrated circuits. That was the late 50s. But even with that development, it wasn't until the 70s that we started to see chips that you and I would look at and say, "Okay, yeah, that's an integrated circuit here. This one is called the Intel 404-bit processor." Another 25 years to get to the Pentium Pro. I remember using that computer. Fantastic. At the time, first large scale commercial uh computer that had millions of transistors on a chip. And then here we are, boy, who's counting? Already 30 years later, um, with GPUs, CPUs, TPUs, XPUs, more than 100 billion transistors on a single chip. Fantastic technology. Took over 100 years to develop. Okay, I say that because quantum computing is much more recent and nent compared to that. In the early 80s, people like Richard Fineman suggested that if you want to simulate quantum, use a quantum system to do that. Okay, good idea. What does that mean? It took us 15 years even to understand that. People like Peter Shaw, Nishimi Murray, Eddie Farhey, and many others, Lev Grover developed algorithms in the '9s um that told us that if and when we can build a quantum computer at scale, we'll be able to do amazing things that we simply cannot do with our conventional computers. And the last 30 years has been about trying to build those computers. And we have quantum computers today. They're just small 100 cubits or so. But the uh progress is accelerating rapidly. And that's what I would like to talk about. So you know takeaway message is this. Quantum computing is real. It's transitioning from laboratory curiosity to technical reality. That's happening right now. Um, but as engineers, we know that moving from a fundamental concept to a real system is going to take time. And of course, it's going to take engineering. And the last thing is you need to be in the game. You need to be in the game to play. It's a game we want to win. But also, you think about who's making these chips today, uh, or who's making and designing these GPUs. They didn't jump in five years ago once someone else figured it out. they've been in it for decades because there's so many trade secrets, so much knowledge that's aggregated in order to enable us to do that. And it will be the same for quantum computing. So with that, you know, two more topics I'd like to talk about cubits and algorithms, how they work, and then engineering quantum systems and a look ahead. So how is a quantum computer different? Um, I'd like to compare and contrast against classical. So a transistor is a bit of information, a zero or a one. It's discreet. Transistors are almost perfect. And if we set a transistor say in state one, we'll come back and measure it in state one. Set it in state zero, we'll measure it in state zero. Quantum computers, the logic element is called a Qbit, a quantum bit. And it's any coherent two-state system, quantum coherent. We'll talk more about what that means. And the way that we think about it is in the context of the planet Earth. Now, if you're at the North Pole, you're a classical state zero. You're at the South Pole, you're a classical state one. Transistors live at the north and the south pole. Cubits can be at the north and south pole too. Cubits can be classical. But when a cubit is anywhere other than the north or south pole, it's in what we call a quantum superposition of zero and one simultaneously. That's weird. For example, if you're standing on the equator, you're at one place. It's a single state, but you carry aspects of both zero and one at the same time. This has implications. We'll go into a little more detail on it, but one implication is that measurement becomes probabilistic. And so if you if you stand on the equator and you you measure, you will instantly snap to the north pole with a 50% chance or snap to the south pole with a 50% chance. Go back to the same place on the equator and 50/50 you're going to go to the north or south pole again. Okay? So quantum computers rely on encoding information in a fundamentally different way. One consequence of that is how we encode that information. Let's let's again consider a classical computer. If you have n transistors, a number n, you can actually represent an exponentially large number of states. Two to the nth power. Let's take a simple case. N equals 3. 2 to the 3 that's eight. There are eight ways that I can permute the zeros and ones. All zeros, all ones, everything in between. There are eight of those. And if I wanted to use those that exponentially large number of inputs to a classical computer and calculate an output, I basically have two ways I can do it. I can take one computer and I can feed in one at a time 00 get an output 001 get an output and do that many times. Or I could buy a large number of computers and try to do it in parallel. But either way, I have an exponentially large number of inputs. It's going to take me exponential time or exponential hardware. Now quantum mechanically you have n cubits and you they can represent an exponentially large 2 to the n number of classical states but the difference is we can put all of them together into a single superposition state all 2 to the n and you can figuratively think about using that single input state into the computer and processing on it. Now, we haven't measured anything yet. We'll come to that in a moment. But this is intuitively why quantum computers are powerful and they lead to something called quantum parallelism and quantum interference that I'd now like to give an intuitive look at visually. So, um before I do that, what what is a cubit? Well, basically comes in two flavors. One are the natural atoms that we get from the periodic table of elements. And depending on the number of protons, neutrons and electrons, we know these elements will have very different properties. And some of them are useful uh in quantum computing and in particular the the first two columns. But one feature they all have is they have discrete energy levels where the electrons can reside. And if we pick the outermost two shells, we can say one of those is going to be state zero and one is going to be the other will be state one. So those are the natural atoms. We can also build superconducting circuits which mimic those properties in some ways. So depending on the number of inductors, capacitors and something called a Joseph's injunction that we put together into a circuit, if we cool that circuit down to low temperatures, it will also exhibit discrete energy levels. It's a harmonic oscillator. And if we choose the lowest two to be our zero and our one, okay, we have a coherent two-le system. And in fact, you know, this circuit that I show here in the upper right, you could think of that as the simplest element. It's a hydrogen atom in some sense. It's our simplest electrical circuit. But you can certainly make it more complicated by adding additional circuit elements and thereby designing the properties that we want a mendlay of table of artificial elements. So the quantum technologies whether computing or sensing or communication are based on these types of two-level systems. Okay. So let's dig into quantum interference and quantum parallelism. To do that, we need to think about quantum bits and again the planet earth. So we're talking about a coherent two-level system. Let's imagine an electron in a magnetic field. Spin up and spin down and they have different energies due to the zeon effect. Now classically I can represent state zero as the north pole and state one as the south pole. Spins pointing up or spins pointing down. Okay. But again as we talked about we can put these states into quantum superposition where we're somewhere else on the planet earth. Um and you can represent it in this way in terms of phases should remind us of GPS or longitude and latitude on the earth. And one thing to keep in mind is that these are complex numbers amplitude and phase. And so if you're on the equator, one of the phases is pi /2 because you came down an angle 90°. But then you have the whole 360 uh that you could describe where on the equator am I residing. So let's take three such spins, three cubits. Uh we'll call them one, two, and three. And classically they represent 2 to the n classical states. I can write a state register of all of them pointed up, all of them pointed down, everything in between. And as we said, there are eight of them, two to the third power. This is just classical. We also said that they're quantum mechanical. So I can represent all of them in a single superposition state with a waiting coefficient C1, C2 through C8. Could change it to zeros and ones if we wanted to look digital. And it requires eight complex numbers now to represent this single superposition state. And by placing those coefficients C1 through C8 in my state register, I can now make this a quantum state register. Okay. And we're going to use this quantum state register to understand parallelism and interference. So let's take it one at a time. Quantum parallelism. A typical operation in a quantum algorithm is to apply a pulse or a gate. And what we're applying here is called a pi pulse because it's going to rotate the spin 180° value pi. I'm applying it to atom 1. So if in my state register it's pointing at the north pole, this pulse is going to rotate the spin and now it's pointed to the south pole. Rotates it by pi. If the spin were at the south pole, it's going to rotate it by pi as well. And now it's going to be pointed at the north pole. That's how this works. And so when we do this, what we're going to see is that this single pulse applied to one cubit is going to shuttle all of the coefficients. How does it do that? Well, let's take a look at before and after. So I apply this pulse to atom one. Let's just look at the spin up case. C1 through C4 will now become down and the C1 through C4 coefficients follow along. The other two spins are exactly the same if you look at it. However, for the portion of the superp position state where the spin is down, that's going to flip up and the coefficients follow along. So, let's think about what we just did. We applied one pulse. We applied it to one cubit and we changed 2 to the n coefficients. That's quantum parallelism. Very, very powerful. Okay, let's now consider quantum interference. We're going to apply a pulse now to the third cubit. You might notice that the amplitude is half as much. the area turns out to be half as much and it's only going to rotate half as far. Let's call it a pi over2 pulse before and after. And for a spin up, we're going to rotate that spin down to the equator. And I'm just going to arbitrarily call this the plus direction. Okay. Now, for the experts here, you know, there's a normalization 1 over of two. And I'm not going to worry about 1 over of two here just for simplicity. And it's not needed to see the interference happen. So let's just ch take the first one. Spin up will rotate into up plus down and that coefficient C5 will follow along. Now let's consider the case where we were pointed down. If we're pointed down and we rotate in the same direction, we'll be again on the equator but on the other side of the Earth. And so I'm going to say that's a minus sign. So down goes to up minus down. And when I do that, the coefficient C6 follows along, but the minus sign follows as well. And so you could say, well, what if C5 equals C6? Well, clearly the first one is going to double, and that's called constructive quantum interference. And the second one is going to subtract and eliminate and that is called a destructive interference. So this is quantum interference in the context of computing and it's also an example of quantum parallelism because it happens on the entire state space at the same time should say the quantum state register at the same time. So we applied one pulse we apply it to one cubit and all of this happened. This is what underlies the power of quantum computing. Now these gates that we apply are actually logic gates and we know from classical electronics classical boolean logic gates I'm not going to read them all here but there's a number of them here you'll find familiar the notgate the andgate the orgate and we know that with a combination of these gates you don't need all of them just a handful of them you can do what's known as universal boolean logic meaning from any point in the state space I can go to any other point in the state space hopefully the correct answer by applying this subset of gates. It's not unique. There are many such uh gates like this or combinations of gates. But this is called a universal gate set for boolean logic. It does require a two bit gate as well. Quantum computing completely rhymes with this. We have single cubit gates which are basically rotations around the planet earth. If I rotate around the x axis, I'll call that an x gate. If I rotate around the y-axis, I'll call it a y gate. Z-axis, Zgate, etc. And importantly, we also have two cubic gates. I'll show you an example in just a moment. The point is that with just a handful of these gates that I've shown, you can form a universal set of quantum logic gates. Run any quantum algorithm. In fact, you can run any boolean logic algorithm as well classically. But quantum computers often do no better and usually worse than classical computers when running boolean logic. They're actually pretty slow. But for certain problems they vastly outperform conventional computers. Okay. So let's take a look at what a single cubit operation is. The analog is going to be the classical notgate. Zero becomes one or one becomes zero. In this case our quantum notgate is called the Xgate. Takes quantum state zero to quantum state one or one to zero. This is what it looks like. On the left you see our planet Earth. The blue vector starts at the north pole and rotates to the south pole. That represents the state of the cubit. The red vector that's coming in and out is the envelope of the pulse that I'm applying. And you can see that it's being applied and to the x-axis, which is why it's an X gate. And on the right is the envelope. I'm not showing the the high frequency carrier inside which is resonant with the cubit transition. This is just the envelope. Now, in this case, I started at the north pole. I ended at the south pole. Classical states. is just a classical gate. But I don't need to start at the north and south pole. I can start anywhere on the surface of the planet Earth. And I apply the same gate and it's going to rotate my block vector as we call it the cubit state 180° around the x-axis. This is called uh the xgate. A two cubit gate. Um one example of the classical analog is the exclusive ore. And one way to look at exclusive ore is to say that you have a control bit and a target bit. And as you can see, when the control bit is zero, the target bit takes um y is just passed to the output. But if the control bit is one, I'm going to invert uh y. So zero becomes one and one becomes zero. And that's the truth table for exor. The quantum version of that is called the controlled notgate. As you might expect, um, it uses this funny symbology, but there's a control cubit and a target cubit, and it does exactly the same thing that I just described. The rotation of cubit Y is going to depend on the state of cubit X. And you can look at examples which are pretty interesting, such as this one here. My control cubit is in a superp position of 0 and one, and my target cubit is solidly zero. And quantum mechanics follows the rules of linear algebra. So you can just take these one at a time. So when the control bit is zero, uh nothing happens to the target and it just I get 0 0. Plus when the control bit is one, I need to flip the y bit. So that zero becomes a one. And now I have a state where I can no longer factor it into something that's just x-ike and something that's just y like. And this is called an entangled state. Of course, the experts know this doesn't prove that it's entangled. I'd have to measure it in different bases to prove that. But this is indeed an entangled state. My point here is I won't run through anymore, but but universal quantum computation is based on and achieved with a small set of these single and two cubic gates. And it looks a lot like conventional computing. So then what is an algorithm? Well, an algorithm is a sequence of these types of single and two cubic gates that gives me an answer to a problem. And the way that it works in quantum is to start with an initial state that is a superposition state of equally weighted all possibilities. And the way that I think of it is that these are possible answers to the question I'm posing that the algorithm is solving. They're all equally weighted at the beginning. And then the challenge of the algorithm designer is to create the steps that lead to the answer. So for example, a single cubit operation. One gate on one cubit impacts all of those coefficients. There's also quantum interference as we saw. And what happens is the coefficients alpha, beta, gamma, delta are no longer what they were before. They were equal. They've changed. They're also two cubit operations followed by quantum interference. And the goal at the end of this for the algorithm designer is to through this interference suppress to zero all of the wrong answers and enhance to one the correct answer. That's very important. And the reason it's important is because we know in quantum mechanics when we make a measurement the probability that we'll get any one of these states goes as the magnitude squared of that coefficient. So if the coefficient is one then with probability one I will get the answer here. It's 001. And every time I run the algorithm, I'll get the same thing. 0 0 1 0 0 1. And that encodes the answer to my problem. And all of the other states 2 to the n minus one, I will never measure because their probability amplitudes were suppressed to zero. It's very important. Okay. So these are the quantum algorithms that we know of today or you could say the algorithm primitives. Um I won't read through the whole chart but you can see on the left the category uh the algorithm type. Then the classical time is what it takes to run on a conventional computer and the quantum time is what it would take to run on a quantum computer. Then there's a speed up and research and limitation and and all the faces you see at the bottom are my colleagues at MIT. They're not the only ones working on quantum algorithms. Uh but I'm from MIT and I feel like I should show some of my colleagues uh because I don't actually work on quantum algorithms. although I should. Um, so what are the algorithms we know about? One of the most important is quantum simulation, scientific computing, simulating or emulating another quantum system like Richard Fineman said, uh, material science, quantum chemistry. Okay, this is where pharmaceutical drug development comes in. All right, and you can see that the speed up goes from something that looks like 2 to the n exponential to a polomial n to some constant. We'll we'll talk about that in a moment. Peter Shor's algorithm is factoring. It's a number theoretic algorithm, applied mathematics. Turns out this one's pretty important. It it cracks our current u public key crypto systems that are in common use. Uh it's why it's quite important. But there are a whole you know several algorithms that are focused on um numerical simulation. Linear systems of equations ax= b pervades all of engineering. quantum computers sample solutions to very large matrices. Um optimization optimization is ubiquitous. Every company is trying to optimize something. And here you'll see that there's a question mark in the speed up in the quantum time because it depends a lot on what question you're asking. You know, I might use ways or Google to find the quickest route home today and it may not be the optimal solution, but it's a good enough solution and I'm happy with it and conventional computers are pretty good at doing that. But if you needed the optimal solution, like your life depended on it, okay, then maybe a quantum computer has a role to play there. And then finally, search. Search is Grover's algorithm. This is a quadratic speed up. This is where um it's an unsorted search of an unsorted database. So, for example, if I give you a phone number, whose name does it correspond to? These things called phone books. We don't have them anymore, but we used to have them. And you would look up alphabetically ordered names, not by number. Okay. Okay. And so with that, let me now switch gears and talk a bit about engineering these systems. What is it going to take to realize commercial quantum advantage? This is very simple, but this is the way that I think about it. First off, there should be no known fast classical algorithm. If there is, use the fantastic computers we've already got. They're really, really good and they've been optimized for more than a hundred years. Second, you need to know the quantum algorithm. There has to be a quantum algorithm to solve your problem. There are many problems I'm aware of where I don't know how to do it efficiently on a conventional computer and I also don't know how to do it on a quantum computer. Okay, so you need that. And then the third thing is it better be a useful problem if it's going to be commercial. There are many demonstrations of what are sometimes called quantum advantage or quantum supremacy where we solve a math problem and it indeed is very efficient on the quantum computer and not so efficient on the classical computer but there's no commercial application and I think long term what we really seek is commercial quantum advantage so the overlap the v of this vin diagram is pretty small today and and I advocate to everyone here that we need to develop more quant quantum algorithms of the type that are useful for commercial application uh because we're working very hard to build quantum computers and we want to know that they're going to be useful in the future. This is very important. Now for such an algorithm such as quantum simulation there's two types of quantum advantage that we can gain and it's represented by this simple equation here and one is a polomial in the size of the problem n and the second is exponential in the size of the problem n and what we're talking about is the size of the um computer we need uh the time it's going to take to get a solution uh maybe the energy required to get that solution And we usually think of quantum computers and think oh they're going to be so much faster. But it's not the only axis to consider. Now if we improve the polomial this is called a polomial improvement. Uh Grover's algorithm is an example. It's a quadratic improvement. For example n would go to square root of n. Okay that that that that's decent. It's not decent if it's four and it becomes two. That's not a big deal. But if it's a million and it becomes a thousand. Okay that's that's respectable. But where quantum computing really shines is on the exponential side. This is what we really seek. Um if you can reduce the exponential to a polomial 2 to the n comes downstairs and it's now in a polinomial form that is an amazing improvement. We don't really experience exponential improvement in our daily lives. So let me let me give one example here of exponential growth. And if you wanted to simulate using a classical computer what's happening in a quantum computer in a brute force sense, you could ask, I've got this big superposition state. There's 2 to the n coefficients needed to represent it. How large a memory would I need to store all of those coefficients? And if you had a quantum computer with just 30 cubits, just 30, I could probably do it on the laptop that I'm using. two to the 30th power number of memory elements. Don't even double it. Just go to 50. Now you need a top 500 computer and the memory that's in that top 500 computer to store two to the 50 coefficients. Go to 80. Okay, now it's getting fictitious. You need all the computers in the worldworked together leveraging all of their memory. Double that to 160. You can go to Wikipedia and find that 2 to the 160th is about all of the silicon atoms on the planet Earth. Each one being used as a memory element now. And just 10x where we started with 300 cubits. Now I need all of the atoms in the known visible universe and each atom is being used as a memory element to store the coefficients in this massive superp position state of just 300 cubits. Okay, that's exponential growth. It's mind-blowing. Of course, the challenge is to leverage it and take advantage of it, right? And that's what we're trying to do with quantum computing. Now, why don't we have quantum computers right now? Uh, it comes down to two time scales. One is called the coherence time, and this is the cubits quantum mechanical lifetime. If you start with a massive superposition state, let's assume we can create that perfectly in the quantum computer. You then start running your algorithm. That algorithm is a series of steps taking me from the initial state to what is the final state in the answer. And while I'm doing that, I'm not the only one controlling the cubits. The environment around the cubits is also interfering or controlling it. Electromagnetic fluctuations, temperature, vibrations, even cosmic rays, everything. They're so sensitive. And so as over time, you can think of this state of the computer blurring and eventually disappearing. Now, the cubits didn't disappear. They're still there. But the point is, as I'm walking from point A to point B with my algorithm, along the way, the environment is taking me to another state. That is the wrong answer. And if I start again and do it, it's going to take me to a different wrong answer because it's stochastic than noise. The time frame over which this happens is called the coherence time. The second time scale is the gate time. What's the time required for a single cubit operation? What does it take to do one step? And the figure of merit then as you might imagine is the ratio of these two. How many steps can I take before my quantum mechanical coherence goes away? Something to note here is that long coherence times is not sufficient on its own. We often hear technology platforms say, "Well, my coherence time is an hour." Oh, yeah. Well, mine's a day. Well, that's great. Um, what you really care about is you should ask that person, okay, but what's your gate time? Because what really matters is how many gates you can do? And and it turns out that if you look at this metric, cubits platforms are basically in a dead heat at this point. So what I'm showing here on the left axis is this metric, left vertical axis, number of operations before an error. And on the right axis is what we in the jargon in the community would call gate fidelity. So if you have one error in a 100 steps, that's 99% good. One in a thousand, 99.9% good gate fidelity. And of course, higher fidelity is better. Now, the horizontal axis is just gate speed. How rapidly can I take those steps and that's the analog of the clock speed of our computers. So, you want to be in the upper right corner. Um, leading platforms today, a trapped ion is one of them. You hold an ion in free space and you use the energy levels. They have very high single cubit fidelities. Those are the blue squares. They have pretty good two cubit fidelities. Those are the red triangles, but they're pretty slow. They're trying to make them faster. I work on superconducting cubits. We're already pretty fast, so we're improving our fidelities. And uh neutral atoms are up and coming. Uh you can see how they fit in somewhere in between. Silicon quantum dots. NV stands for nitrogen vacancy. That's a defect in diamond. There's defects in silicon. Many candidates here. This is what they look like. Um on the left you see IBM's Osprey processor, Google's Willow processor. These are superconducting cubit processors. You can see that they're 3D integrated, heterogeneously integrated. Um and you know these are quite sophisticated systems at this point. In the upper middle is Quer's Aquila processor. Those are neutral atoms if you see the little dot inside the the magnet. Quantinum's H1 is on the upper right. That is a trapped ion uh quantum computer and you can see that it's built on a silicon chip integrated photonics. Lower right is D-wave a different type of computing called quantum analing. I won't talk about that today but it's superconducting and of course in the bottom middle that's obviously not a quantum computer that is the right flyer as you may recognize. Um it it's not the first time that humankind had flown. That's actually a misnomer. Um before that we had hot air balloons. Uh we had aer drromes, unmanned aer drones. Um but when we see the right flyer historically, we look back and we say, "Yeah, that's when it all came together." Um there's a person on board, self-propelled, took off, landed, and this was the dawn of commercial aviation, right? And and I put this picture here because two reasons. One is that I think that our quantum computers today are about this level of sophistication, but also it is the dawn of commercial quantum computing. I think that we have seen enough and the system is complete enough. You can access a quantum computer online, write a program and the answer comes back. The system is there. I wouldn't I wouldn't get on that plane, right? But I can look at it and say that yeah, the system is there. So to realize then the promise of quantum computing, we have to engineer quantum systems that are robust, reproducible, and extensible. Right? The way that we do that is through architecture. Of course, lots of technology improvements, but the architectural layers of a quantum processor look a lot like those of a conventional processor. You have the physical devices at the bottom. You have an application layer at the top, a logical controller to figure out what gates I need to apply to run uh my algorithm, interpret the program that I just input. And then you have two types of error suppression, passive and active. And I won't won't have time to go into the passive one today, but I will talk about the active one because it's so important. And on every clock cycle of a quantum computer, every time we take a step, we need to take the step in the algorithm and then we need to check for errors. We take the next step and we check for errors. We have to do that every clock cycle and that's called active error correction. That is the grand challenge today. Our physical cubits have error rates of one in a 100 to one in a thousand, one in 10,000. The yellow region that you see there. You could then ask what do we actually need to achieve commercial success, right? Run something at a commercially large scale. That's the blue region. Practical applications. You need to run a billion steps or even a trillion steps to be useful. There's a big chasm. By the way, CMOS is almost perfect. The error rate of CMOS is something like 1 in 10 the 20 or better. Um thankfully we don't have to be that good. We'd take it if it were. But we do need to be in the blue region. And the way that we're going to overcome this giant chasm is through something called quantum error correction. And the concept I think we're all familiar with. Rather than encoding the information in a single cubit, we're going to encode it into a team of cubits. Just like with people, we form teams of people. And if the individual people themselves are sufficiently performant, they cross some threshold, then grouping them together into a team, the team will outperform any of the individuals, right? Um my son is 16, but at one point in his life, he was 5 years old, and we would go to the playground, and there would be lots of kids playing on the playground, and it was heartwarming, but it was total chaos. Everybody's running all over the place. And if you added more kids to that playground, the chaos only increases, right? They are not yet beyond threshold to form a team. And our cubits 101 15 years ago were all kindergarteners basically. But in the intervening 15 years, we've improved the cubits to a point where adding more cubits can actually improve performance. Okay, you could ask how much? Well, the logical error per cycle we call that's the error of the team has two parts. One is C. That's a cost. Forming a team is not free, right? it you actually have to perform more gates. You have more people to take care of. There's an HR department. There's a security department, right? Like takes a lot to run a university. But you still do it because the win is in fact exponential as you see here. And and it's exponential in the size of the team, which in our jargon we call the distance. If your team is a 3x3 array, then that's distance three. If it's a 5x5, it's distance five. I didn't mention the blue cubits. forgot the blue cubits you can think of as referees. They're the ones that have a yellow card or a red card. In in fact, there are three types of errors. There's a green card. They're going to flag when an error occurs. And I measure and look at the referee cubits to infer errors, but I never look at the data cubits. The data cubits hold the quantum information. If I measure those, it's game over. So instead, I measure the blue ones every clock cycle. So this is what it looks like. We've got the logical error on the vertical left axis and the physical error of the teammates on the horizontal axis. And you'll see that the logical error is on a log scale. Today we're at about one error in a thousand for the physical. And what you see is if I go from a distance three to a distance five code, I improve by a factor lambda. My error goes down. If I go five to seven, I'll go down by that same factor lambda. And I can just keep doing that. I just keep making the team bigger and bigger and bigger until I hit the logical error rate that I desire. Something like one in a billion. I just stopped there. I could keep going. So what did it cost me to gain that? Well, in this case at D17, distance 17, I needed 577 cubits on my team, but I got an improvement down to, you know, 4* 10 theus 9. An improvement of almost a factor 1 million, right? That's the promise of quantum error correction. This is what we're going to have to do to get into that blue region. Now, we still have work to do at universities. My group, we focus on improving the performance of individual cubits. And if I can improve the cubit by a factor two, you can see at the logical level here, it improves by more than a factor 10. Right? That's also related to this exponential improvement. I bring this up because Google in 2025 published a paper which I think will be remembered in a field with a lot of hype and breakthroughs every day. This one actually is a breakthrough. Um what you're seeing is the logical error is a function of the number of times that we run the error correction algorithm. And in red is distance three and then in teal is distance five and in dark blue is distance seven. We're adding more cubits. The logical cubit is getting bigger but the error rate is going down. That's amazing, right? Conventional probability would not tell you that if I have an errorprone system and I add more errorprone systems, it's just going to be faster to error. But here we've created a code where by adding more cubits, the error went down. And this is what we need to do if we're going to build large scale quantum computers. So I'm at the end of the talk. Um, I just want to mention this almost exactly a year ago at GTC. This gentleman here, Jensen Hong, you probably recognize him, CEO of Nvidia. He was asked, "When do you think we're going to have a quantum computer?" I, you know, as an engineer, I really respect Jensen. And uh, he's he's smart, right? Very smart. And he he said, "Well, look, if you said 15 years, that's probably on the early side. 30 years, it's probably a little late. 20 years, I think it's we'll have a quantum computer in 20 years." And for those who were following it, um the stocks of the uh quantum computing companies all tanked because oh my god, Jensen says it's going to be 20 years before we have a quantum computer. I actually was quite excited. I was like, Jensen said we're going to have a quantum computer in just 20 years. That's amazing, right? We started with the history of classical electronic computing, which arguably started more than 100 years ago. Now 6 months after this he realized okay what I said was uh you know impacted those companies and when he answered this question he was thinking the quantum computer that's going to invent new pharmaceutical drugs. He was thinking the endgame but he qualified it. He said look you know um technology doesn't evolve according to a step function that there's going to be nothing for 20 years and then suddenly we're going to have the endgame quantum computer. That's not how it works. There will be improvements year by year by year by year and eventually something commercial will be useful. Um, if you used early classical computers in the 70s or 80s like I did, it was a very frustrating experience. It did something, but it never was quite enough. And the early quantum computers will be exactly like that, right? Make no mistake, they're not going to be perfect, but they will be useful and they will be commercial. And so this was this is my last slide. Um, if you think about technology readiness levels and maturity of quantum computing systems, I'd like us to just remember the TRL level of these two chips. The Intel 404 in 1971 by today's standard was pretty wimpy, but at that time it was TRL9. And of course today we have black black wall processor GPUs, right? And today that would be TRL 9. And so if you ask me and you can find it in this paper listed at the bottom, what is the TRL level of quantum computers today? I would say it's somewhere around 6 to 8 to be honest because you can literally log into your computer terminal at home, program a quantum algorithm, send it to a company like IBM, get the answer back, and that's the whole system. Right now, it needs to be much larger to be commercially powerful, but the system is there. Just like this Intel 404 processor did something. Okay. So, a few things I'd just like to leave you with. Quantum computers are not going to replace classical computers. First off, they only solve certain tasks efficiently, not everything. And second of all, we need a classical computer next door to help us with the error correction. So, they're always going to be run in tandem. We need more quantum algorithms. Please join hackathons. um help us develop tomorrow's commercial-grade quantum algorithms. We don't need to worry about encryption today. A quant shores algorithm when we have a quantum computer at scale will um compromise our current public key crypto system. But we do need to switch over to new crypto systems, classical ones um because we want to be forward secure. there is a notion of harvesting data today and storing it and then when we have a quantum computer coming back and looking at that data at that time right so NIST has released its new standards uh for for encryption just classical algorithms going by the name crystals kyber and things like that we need to switch over to that now so that we are also forward secure and to realize the promise of quantum computing we're almost certainly going to need error correction and with that um conclude the talk and I thank you for your attention. >> Thank you very much. Um should have shuttled around. So now is the time for questions and then there's the awkward pause. Okay, good. Right up front. Yeah, we'll get you. Don't worry. It's just coming. >> Yeah, thank you for a very nice talk and very nice overview. I'm also from MIT, so >> Oh, >> I got all my degrees from there. >> Um, your talk seems to be predicated on the factor that we're using to superconducting cubit to make the next generation quantum computer. But from the history of technology, we know that cryogenics does not fly. We have the measles and the lasers and then we have the transistor laser and also the transistor laser versus the uh lasers that does not need cryogenics. Somebody was clever enough to remove cryogenics from >> uh semiconductor lasers. It seems that you have a system that needs lots of cryogenics and it doesn't seem that the system will fly. Eventually we would move away from that in my >> Yeah. Yeah. I I completely agree and I tried not to talk about my own research. I tried to be you know fair across all the technologies. So what I talked about today is not solely about superconducting. It actually applies to all the different platforms. And you're absolutely right. Superconductors very likely will be the vacuum tube of conventional computing, but superconducting would be for quantum computing. They may be the first, but will they be the long-term winner? Okay, I don't have a crystal ball, but you're right. With cryogenics, probably not, to be honest, right? If we didn't have to cool them down to millichelvin temperatures, if we had an alternative that was just as good and didn't need the cryogenics commercially, we would switch to that. Um so superconducting cubits do require millich Kelvin temperatures. It's not a showstopper in in any sense but yeah it's inconvenient and when we have another technology platform that is as mature or more mature it's likely that we would switch over but today I think superconducting cubits are in the lead. What is the probability of using these um quantum computers for nefarious reasons such as like um when you said crypto um >> for the the wallet keys >> being able to decode that would would there be a probability of nefarious use >> almost certainly and that's why we really need to switch over to these new postquantum crypto schemes that we believe are going to be immune to attack by quantum computers. Look, quantum computers don't solve every problem efficiently. And you could say that we just were unlucky in the 80s when we decided that RSA type encryption is going to be our public key crypto system. We picked that before we knew about quantum computers and later realized, oh, uh, quantum computer could actually break that crypto system. So over the last 10 years, researchers have come up with a new crypto system just classical um based on lattice mathematics which we believe is also hard for quantum computers to break and so the task ahead of us is to switch over to that and on the software side it's already starting to happen. Um there's also a lot of embedded hardware that needs to change. So we're working on it. So there is a likelihood that quantum can be used for such a nefarious um application, but we're not defenseless. There are things we can do to minimize its impact. >> Thank you. >> So we're going to do one last question because you may have noticed this is the in between class time. >> Oh yeah. >> And the parade has started. So >> thank you for the very nice presentation. Uh so there have been concerns that quantum hardware have has been marching at a regular pace and uh they people have been delivering what they promised in terms of reliability and uh a number of cubits. Uh but on the other side the software and the algorithms have has been a bit behind. >> Yeah. >> Do you think do you agree with that concern? Do you think that maybe we should be working also more on the soft at the same time on the software to >> Absolutely. Yeah. Like I said, you know, we need to develop more algorithms and they need to be commercial grade ideally, not not just toy problems. It's very very important to do this, right? Um now I think they go hand in hand though that there's some bootstrapping. It's very hard. Peter Sher is a genius and he could in the abstract come up with an algorithm, but for me I'm a mortal, right? And um I actually need to try things and you know trial and error to come up with something new. And I think that with these quantum computers coming online, we will see a growth and like you know MIT has hackathons every year in quantum and I know a number of places do around the world. This is exactly what we need to do right and to develop new um algorithms. Most algorithms I believed are developed that way. Turbo codes for example in the early 90s nobody predicted that. Wasn't like we had a development of it. It was just people playing around with communication systems and signal processing and realized, oh wow, look what we can do. Um, so there is bootstrapping that's going to happen and I think that will accelerate now that we have at least these small quantum computers available. Um, we have a small presentation. >> Oh, I think I'll ask you to stand over this way. >> Yeah. Oh, maybe I should. There we go.