Daniel Rodan Legrain: Hello everyone, and hello Michael.
Michael Hatridge: Hey.
Daniel Rodan Legrain: I am Daniel Rodan Legrein, the Quantum Science and Engineering Lead for North America at Qblogs. And it's a pleasure to welcome you all to today's Quantum Builders FISA Chat. This session is also a little bit special for us because we're actually celebrating one year of Quantum Builders. We started this series around this time last year with Will O'River. And since then, we've had the opportunity to really host a remarkable group of invited speakers from across the whole quantum ecosystem. including Berde Jong, Peter Shore, Steven Giver, Gretchen Campbell, Chuck Black, David Afshalom, Kyle Cerniak, Leven van der Sijpen, Irfan Siddiqui, Christoph Josak and Laura Schulz, among others. The idea behind this series has always been to bring together leading voices that are shaping the quantum science and engineering field and to have open conversations about where the field is going, all the way from foundation science to quantum hardware and scaling, systems engineering, and the broader quantum ecosystem. So today I'm especially excited to welcome Professor Michael Hatridge. Michael is an Associate Professor of Applied Physics at Yale and he's also the Deputy Director of the Yale Quantum Institute. His group, the Hat Lab, as many of you may know, builds superconducting quantum circuits and they also have a particular focus on quantum measurement, parametric amplification, parametric control, engineering path. and modular superconducting quantum architectures. His pass through the field is also a fascinating one. Michael is from Texas. He received his bachelor in science from Texas A and then his PhD from UC Berkeley working with John Clark. He then did his postdoc at Yale in Michel de Vohette group where he focused on quantum measurement and high fidelity superconducting qubit readout. And then he later built his own group at the University of Pittsburgh, where he also served as the director of the Pittsburgh Quantum Institute before returning to Yale, where he is now. What they really found very interesting about Michael's work is the way that it really connects several layers of superconducting quantum computing that are very often discussed separately. mean, the physics of quantum measurement, the hardware required to amplify very weak signals, the use of nonlinear and parametric interactions for control. And ultimately, I would even say the architectural question of how we build systems that can really scale. So Michael, thank you so much for joining us today.
Michael Hatridge: Yeah, thanks for having me.
Daniel Rodan Legrain: For everyone watching live, as per usual, please feel free to post your questions in the chat as we go and I will make sure to bring a few of them into the conversation towards the end. So Michael, let's start with your own journey through the field. You've worked across several major centers of superconducting quantum research. mean, doing your research, your PhD at Berkeley with John Clark and your postdoc at Yale with Michel Devore. Then you built the HATLAB. lab at Pittsburgh and now returning to Yale. So what would you say were the key inflection points along that path and how has your way of approaching quantum hardware evolved throughout these years?
Michael Hatridge: Yeah, that's a good question. I don't things have worked out very well for me. I I cannot claim that, you know, I had a very simple or or carefully curated way through the field. So in my PhD, for instance, John Clark's lab had a qubit portion of the laboratory. I was actually working on on low-field MRI and we studied prostate cancer, we studied breast cancer, we studied the reaction of your brain to sort of thinking, you know, people like functional MRI try to see processes in the brain. The through line I have is sort of that I was always an instrument builder. and that's kind of really I think the underlying DMA. We you talk about seeing amplifier, seeing measurement always in there. I was always always building some kind of measurement device. but I didn't measure a qubit in my PhD. I didn't measure a single qubit until my postdoc. We didn't do a two qubit gate, like I personally didn't do a two qubit gate until I was a professor. So for me it's it's just been a lot of serendipity and sort of following these strange instruments that we were building into this quantum field. The the first
Daniel Rodan Legrain: Very interesting.
Michael Hatridge: the first paramp I built was definitely on accident. We were trying to build a better magnetometer in the same style that we had used for detecting like MRI signals. And and so we tried to for for people who are old. The DC squid used to be an object that had resistive shunts and you would current bias it and measure a DC voltage. That was like the DC and a DC squid. And we we wanted to take those away and use these new circuit QED type techniques to build a dispersive magnetometer. And it worked really, really well. It worked in fact too well for us to explain until we figured out that like accidentally the magnetometer had a paramp hidden inside it. And so that's kind of like a very unintentional link. The magnetometer sort of worked too well. And and basically very quickly we learned to stop calling it a magnetometer, but to use the same circuit to amplify qubit signals. So this is what I mean when I say like a lot of it wasn't intentional.
Daniel Rodan Legrain: Well, I guess that's how a lot of the research goes as well, Michael.
Michael Hatridge: Yeah.
Daniel Rodan Legrain: So Michael, both your PhD advisor, John Clark, and your postdoctoral advisor, Michel Devoguet, received the 2025 Nobel Prize in Physics. I don't think many people can say that. Having trained closely with two of the three laureates, what perspective did the Nobel announcement give you on their contributions? And what did you learn from them that may be harder to appreciate just by looking at the published science?
Michael Hatridge: Yeah, that's a good question. I guess, amusingly, when I moved out to Berkeley, I got married shortly thereafter, and my wife was trying to figure out kind of what she'd gotten herself into. and she looked up an article that said maybe John Clark, this was like two thousand five, would win the Nobel Prize for this MQT experiment. And I said, Yeah, wouldn't that be great? so I guess that dream was always sort of baked into to to my interactions with John and Michelle. and you know, it was pretty exciting what 20 years later when it actually happened. you know they both are very, they're both very good mentors. They taught me a lot about how to do science, how to, you know, just sort of everything about the the skill set that I use these days. But they are also very concerned with like not fooling yourself with careful calibration, careful control experiments, saying what you know, saying what you don't know. I think the MQT experiment has a lot of those flavors in it, that they, you know, then carried forward for the forty years of research they did after that. and certainly I was part of that legacy in learning those skills in their laboratories.
Daniel Rodan Legrain: Wonderful. So following up a little bit on the story you explained at the beginning, when did you realize that quantum measurements and superconducting qubits would become the direction that you would want to pursue?
Michael Hatridge: Yeah, I think for me it's particularly easy to date. There was a 2011 Le Zouche sort of school on quantum machines, and and Sasha Karatkov was there talking about the back action of weak quantum measurements. And not to go like too off the rails into the physics, but when for instance an atomic person measures a single neutral atom or ion floating in space, they scatter many, many photons off it. And the first one pretty much carries away the quantum information, whether the s the atom is in the ground or excited state. But you almost never catch that photon. And so you get this picture in your mind, which a lot of us in like the microwave superconnecting world inherited, that the state function very, very quickly collapses when you measure. And then you sort of build your knowledge back up as you slowly connect the few photons that scatter actually into your detector. And that's certainly what I thought about quantum measurement, that I would not really see. many quantum features. They're kind of destroyed because you kind of first lose the quantum information, then gain the classical information. And Sasha Krotkov's theory on this was that that's not the case. If you do a really weak measurement so that not enough information transit the system for either you or any other observer, right? If anybody had the total information, there still is not enough. Then you would see these unique quantum features. You would see that the qubit doesn't just sort of collapse up or down. It does some really complicated sort of continuous quantum walk. And so that, like Michelle and I agreed that that's the experiment I should try to do when we went back from Lazouche. And there were a lot of arguments. We worked closely with Steve Gervin and Mazir Mirahimi, two theorists, and the theory and the data didn't match up. We had a lot of sort of struggles and back and forth, but basically what was technically new that allows to do the experiment is these new parametric amplifiers allowed the quantum efficiency to be something like twenty to thirty percent. So like a third of the total information we were collecting, which now is like completely passe. You know, people are like 70 to 80 quantum percent percent quantum efficiency is sort of minimum to be impressive these days. But at that time this this different aspect that most folks had not had in their experiment before allow us to do the experiment. And we basically saw that, you know, you think the qubit collapses upwards or downwards. In fact, it does this smooth quantum walk across the across the block sphere and we could very nicely track it. So this I learned a lot of quantum mechanics from this. Previously I had mostly built detectors. So the detectors have various quantum limits, but often they look at very classical things or the information they get looks very classical. So you say, Okay, the noise will stop getting better because of some quantum principle. But this sort of weird quantum back action We set out to find it we found it and it really sort of taught me a lot of what I now know about how to view quantum measurement as a really interesting quantum game we can play and not sort of where quantum mechanics goes to die.
Daniel Rodan Legrain: Fascinating, Michael, thank you. And then if you look back at your career, would you say there was a particular point when you thought, okay, superconducting qubits are not really feeling like beautiful physics experiment anymore, and they start feeling like a real engineering platform for quantum computing?
Michael Hatridge: that's a good question. In my laboratory we like to live right on that line, so I'm pretty happy that they have aspects of both. I guess No, I I yeah, I think that's kind of my answer. We were always engineering on the practical side. We're always trying to do crazy physics with these tools. and i myself, I guess, I find the most fun is along that boundary where sometimes it's crazy physics, sometimes it's engineering, and we sort of inform each other back and forth across that boundary.
Daniel Rodan Legrain: Let's keep talking a little bit more about the measurement. mean a lot of your early work focused on quantum measurement and measurement back action. What did those experiments teach us about what actually happens when we measure a quantum system?
Michael Hatridge: They taught us again that we can see the unique quantum properties of of a qubit or of a couple qubits even when they are being measured. And if you can sort of keep that quantum physics alive all the way to your detector, if your detector itself is an efficient amplifier, an efficient detector, what it means is sort of you can play games. So The neat thing about quantum measurements, if you ask a question like, is your qubit blue or not, you force it to make a choice. And there are certain natural choices that are very easy for your detector to do. In the framework of supernatural circuits, sir, circuit QED, it's very easy to measure the Z-axis of the qubit. Sort of the interactions are set up that if you take a regular amplifier and a regular microwave pulse, and you don't really work hard on their funny quantum properties, you'll get a Z basis measurement. But if you work hard on the quantum basis, you use squeezing as a resource, you use interesting kind of interactions that are other than the circuit QED interaction, and you change the properties of the behavior, sorry, the behavior of your detector. There's this really neat thing where if they're all three quantum, then you can basically build a machine that asks a different question, that measures the qubit along a weird axis, that measures two qubits in a way that what you get is their parity measurement. And classically, This doesn't really make sense. It doesn't really have an analog. Like there's this this question, right? If the tree falls in the forest, does it make a sound? You know it makes a sound. Like everyone no one is really asking this question. There's some massive boom out there and it doesn't require an observer. But in quantum mechanics, if you are what we call the omniscient observer, you're the only one who collects the information, and you collect it in such a way that you are deliberately blind, you don't learn certain things. That's actually good. And that's really, I think, funky and sort of unique quantum property. And it's kind of, for instance, what error correction uses. Error correction very carefully controls the system in such a way that you never ask, like, what state are you in? You ask, did you have an error? And if so, what type? And and classically, I think we don't have the intuition for these kind of measurements because we don't have this business of the observer, the act of measurement being really I mean We have some examples like bleaching and stuff like this, but we don't have this idea of of really interesting and exciting quantum collapse. So that's sort of been what we work on. These experiments are typically very hard. And so what you'll actually find, I think in my lab, sort of in the engineering picture you're asking before, is there's sort of the wackiest thing we can do, the most complicated measurement we can make work once. And then a few steps back from that are the things that we can make work regularly, and then we work hard to engineer those into quantum systems.
Daniel Rodan Legrain: Wonderful. Let's follow up on what you mentioned about amplifiers. And you have worked and your group has spent a lot of time on quantum-limited parametric amplifiers. So maybe for a broader audience that may be joining us today, what makes an amplifier quantum-limited and what makes a good amplifier in practice for high fidelity
Michael Hatridge: Mm-hmm.
Daniel Rodan Legrain: superconducting qubit readout?
Michael Hatridge: okay. let me answer the second one first. So if you go shopping for a microwave amplifier, it will have a data sheet. most of the properties of the quantum amplifiers we we we build would be considered mediocre or sub-mediocre compared to like the amplifiers that you buy. So the bandwidth that we can build, the saturation power that we can build, the power draw that the amplifier has, all those classical properties. Translate very well into these amplifiers, and the usual answer is that you want more of all of them. the one sort of magic thing about the amplifiers is that again, quantum mechanically, are you guys hearing some kind of feedback or somebody else's clicking? Maybe it's just me. Okay. Okay.
Daniel Rodan Legrain: I can hear you great, Michael.
Michael Hatridge: so The the unique quantum feature is that the amplifier itself, and and this was sort of a debate we had many times in our post in my postdoc with Michelle that we built experiments to try to test, the amplifier itself is a quantum object. That has weird consequences. That for means, for instance, you don't think that like classical light comes in the amplifier and a class sorry, quantum light comes in and a classical answer comes out. It does a quantum Translation. And so for instance, you can put amplifiers back to back and they can undo each other even at a quantum mechanical level. Or the properties of the amplifier will transform the light in a way that's interesting and quantum. The quantum limit here basically says that if you're gonna use this amplifier to transform the light in a way you'd like to digitize, for instance, at room temperature, you have a choice to make. And basically it comes back to Heisenberg's uncertainty principle. You can learn one quantity of a microwave field perfectly if you destroy the other. So if you think of a microwave tone as having an amplitude and a phase, kind of it's a little bit imprecise, but you could learn either the amplitude or the phase, but not both. And that's what we call a phase sensitive amplifier. So that one really has no quantum limit. The the the sacrifice you make is that you know if your information is in both quadratures of the light, you can't record it all. So of course you need to like use that amplifier in scenarios where all your information lies in one quadrature. Let's say only a small phase difference or only a small amplitude difference. sometimes you don't want to make that choice. And so you ask what's the best thing I can do that doesn't favor a particular kind of information. There it turns out you cannot learn both properties of a field with infinite precision. And so you you work it all out, and we find that just like in in Heisenberg's various like microscope and these various pictures of this experiment. There's a minimum uncertainty, which basically is is is the vacuum noise itself. So a quantum limited amplifier is either like one of these two types in in our microwave world, and we've pushed these properties to the ultimate limit. You either learn one thing perfectly by destroying the other, and again in this weird quantum way, destroying it so that no other observer could get it. It's not that like you ignore it and just don't write the answer down. You really have to squeeze it away, hide it away. Or you you don't choose. And then you get that there's basically this other source of fluctuations in your system, which is unavoidable. Sorry, that's maybe not a very helpful disquisition, but
Daniel Rodan Legrain: That was a great answer, Michael. But let me now follow up with one more question, maybe a more practical one now. Would you say readout is still one of the central bottlenecks of superconducting quantum computers? And maybe even going one step forward, what is still missing from today's qubit measurement toolbox?
Michael Hatridge: Yeah. I do think readout, if you look at various papers papers from groups like Google, readout is often cited as like the worst thing in their system. And part of the problem there is that like a fidelity of a qubit, you make that better by sealing it up in a box and protecting it from everything. You know, often fidelity of a gate is somewhat the same. If you look at readout, there's the qubit interaction with the readout interaction. Apparatus. There's the moving the information across your cryostat, and then there's the how you process that information in the quantum regime. So readout kind of has a problem that, you know, for instance, the properties of my readout are not specified by the quantum processor. They're not specified by the amplifier alone. It's the whole system bolted together with all of its n you know warts and defects that come together to make the quantum efficiency. So for instance, as an amplifier designer, We think we can get seventy to eighty percent quantum efficiency of the amplifier. we do give these out to folks who who work on some of the same government grants as us, but if you were to put a lossy cable in front of it, your readout won't work for beans. My amplifier cannot perform. Right? Also, if your qubits are very decoupled for the environment so that you can only extract the information very slowly, then the qubit will decay before the information can get out. So I think Most of the properties of the quantum computer are what we would call like closed system properties. They they involve how we behave when you isolate it as well as you can from the environment. Readout is the one that says how do you behave when you're interacting in a delicate and controlled way, but as strongly as you can with like the right environment. So I think it has a natural handicap that way. the other sort of general question that that we continue to learn also is that because we're in a hurry, we need to read out the qubit. 10,000, 100,000 times faster than its lifetime. I actually think it shares this property with Gates that both Gates and Radout want to push you to your edge. And you know, there's still, believe it or not, complications, wrinkles, and even the simple Transmon qubit that we're still sort of carefully documenting these days in terms of not what happens when you're not driving very hard. I think that's pretty well understood. But what, you know, what is the limit of just how hard you can push these systems to push these fidelities up?
Daniel Rodan Legrain: Sounds great. Lots of work to do in the years to come. So let me switch gears a little bit, Michael. I mean, we've talked about amplification, but a lot of your work uses parametric drives, not just for that, but also as a much broader tool for quantum control. For maybe those in the audience that may not be directly working with superconducting qubits, what does parametric control mean and why is it so powerful?
Michael Hatridge: Yeah, that's a good question. And and the the term is it's well baked into the field, but it I it's pretty misleading. So a real parametric control, the the typical example folks will give is somebody on a swing. So, you know, typical driving, resonant driving is where s your person A is on a swing, person B is pushing them. So they're resonantly driving you. If you push the person at the frequency of the swing, then you'll drive oscillations and they'll oscillate up more and more. parametric, the reason that we call it a parameter is if instead the the person on the swing, right, because like sometimes you don't have somebody there to drive you, but you can still swing just fine by basically pumping yourself. So what you'll typically do, don't do it this way at home, but it's like easier to think about it, is like stand up on the swing and then squat and then stand and then squat and then stand. And what's
Daniel Rodan Legrain: do it at home.
Michael Hatridge: yeah, don't do it at home. But what you're actually doing is you're driving. This parameter of the system, which is your moment of inertia. And by doing that, like most kids don't do it by standing, but they do it by making their body flat and then sitting up. You can drive yourself just fine. That is the essence of parametric. What it means is there's a parameter, something like the capacitance of something, the inductance of something, the moment of inertia of something that you are driving. It's now instead of a constant, it's some dynamical part of the circuit. There are ways that we drive superducting circuits that are explicitly parametric. You can think of the Joseph injunctions we use typically as inductors. And so if you were to modulate, for instance, the flux through the squid, that is a time-varying inductance that is explicitly parametric. what we really do most of the time in my lab is to just say, well, that's an interesting way to use nonlinearities in the circuit. So the swing, for instance. Is essentially nonlinear. The pendulum itself has a in basic physics you'll learn that the period of the pendulum depends on how hard you drive it. That essential nonlinearity is what we would call a fourth order nonlinearity. And folks in my lab know that that means four wave mixing is inherent to that system. And then we just like use that as a design tool for interactions. So the way we actually talk about in the lab is to say if you have a nonlinear system, You tell me that you have three-wave mixing, four-wave mixing, five-wave mixing, six-wave mixing, like whatever you want. Parametric driving is a way to turn those tools into lower order interactions that I can turn on and off with microwave driving. So it's really in that sense very abstract sounding. It's just like parametric driving, though, is the best way I can explain it. If you don't ask me too hard what parameter of the system I'm modulating. It's still the idea of a kid on a swing pumping themselves. We're still gonna drive away from the resonant frequency of the circuit and use its fundamental nonlinearity to generate some behavior we want. Except in this sort of more Hamiltonian engineering way of thinking, we're very abstract about what the parameter is. It's not something like capacitance, inductance second point two. It's you know, G six. And and that part gets very technical sounding, but but but the overall idea still is is the kid on a swing.
Daniel Rodan Legrain: I mean, your group has really used these ideas for gates, couplers, engineering, interaction, engineering paths. You already started kind of answering this question, but maybe from a more general perspective, what can parametric approaches let you do that would be otherwise difficult or maybe less flexible with, let's say, conventional
Michael Hatridge: Yeah.
Daniel Rodan Legrain: microwave control?
Michael Hatridge: It's a really interesting question. I told you in the very beginning that I came at a lot of this stuff sort of let's say people with a different mindset than me could have maybe gone through my own career in a very forward-looking way. I tended to back my way into things. we learned about parametric interactions because they're very common for parametric amplifiers. Like that's why they have the name in them, right? As I tried to learn about how other people were putting together their quantum computers and how, for instance, they're building gates, there's this period this this this moment in many of these publications, many of these talks, where they say, and here's a nonlinearity, and therefore I drive at this frequency, and everything is fine. I get the interaction that I want. And and I guess in my own personal journey, it's sort of been realizing that, You don't have to do parametric amplifiers first, but if you've lived that lifestyle, you're like, okay, now I understand. It's like a parametric amplifier, except we've changed a few of the terms in the Hamiltonian. And and that's really obtuse. That's maybe me specific. It's a you know, my poor suff suffering students are infected with this mindset by me. but we really came at it backwards. And and then what we found was once you say, you know, here are all these parametric susceptibilities. There's really nothing you can't do with them. sorry, yeah, cannot do with them. Like they can do everything you want. The gate can be parametric. And several of the projects in the laboratory, sort of at some point it kind of became a badge of honor, you know, like that's not parametric, but I was like, I thought we can make it be parametric. I'm not saying that that's the way you you you should do it moving forward, but that's definitely the way we come to view this stuff. And I guess it has a special name, it has a certain way of thinking about it, but it's really endemic. Let's say if you never want to use a parametric interaction in your laboratory, fine. But you're gonna first have to make sure that every frequency you drive the system at doesn't accidentally hit one of these parametric susceptibilities. So it really kind of works both ways. We use them to do all of our controls and compare them against sort of more conventional schemes. But even if you want to ignore them, for instance, the famous readout problem in Transmon qubits, you can pretty easily explain by saying that the cavity drive you're applying is accidentally hitting a parametric resonance of the transmon. So even if you don't believe in them, you know, they don't know that. And so you will turn them on on accident.
Daniel Rodan Legrain: Fascinating, Michael. So I would say that connects naturally to the problem of scalability.
Michael Hatridge: Yeah.
Daniel Rodan Legrain: And your group has really been developing modular superconducting architectures, including microwave, quantum state routers, and detachable quantum modules. So why is modularity, rather than, let's say, simply making one larger monolithic chip, such an attractive path for scaling superconducting quantum computers?
Michael Hatridge: good question. if you look at that famous Google result on their what was it fifty-three-bit machine, one of those qubits was dead. Right? so monolithic scale up looks pretty intimidating because transistors apparently you can make mm, you know, with a vanishingly small failure rate. Supernight qubits are just not like that, and I'm not sure what the prospects are for them becoming sort of ultimately fabbable in a repeatable high precision way. So if you say that your components are gonna be flawed, then you need a way to sort of limit your risk, limit the fact that your fab fails. So in that sense, modularity, I think, is directly tied to the sort of synthetic nature of superconducting circuits. Superconducting qubits are all fabbed, unlike atoms, you know, which have exact properties, are all fabbed very synthetically, which means they're all gonna be a little bit different. They're all gonna have their own character. And if that's fundamental to your building block, You need to truncate the size your system and figure out a way to build it out of pieces. Otherwise, you're gonna have some dead pieces, you know, just contaminating what you're trying to do. So although we have been looking at it in in my laboratory specifically, I don't think it's anything unique to us. I think it's you know the nature of supernecting qubits that are pushing us that way. But even in, for instance, the neutral atom systems, they have other constraints about the optics that they use, about the atomic arrays that they're gonna build. And again, there's something there when when you keep trying to push it sort of monolithically huge, the system is gonna push back on increasing cost and you're eventually gonna then compromise on how to build it out of pieces. So I actually think it's it's really easy to build an argument for modularity. It's it's much harder to imagine pretty much in any quantum system why a monolithic thing is not going to eat your lunch.
Daniel Rodan Legrain: Fair. Okay, so once you decide to go modular, what becomes hardest in practice? Is it the loss? Is it the coherence? Is it the connectivity? Is it the packaging, the fabrication, the calibration? Or is it something else?
Michael Hatridge: yeah, that is a good question. I guess we have been thinking about it a a little bit different way. I'm not so sure that the supernatural circuits again are completely synthetic. Everything is sort of built out of Joseph's injunctions, capacitors, inductors, transmission lines. The way we build connections between modules and the way we build things inside modules are pretty similar. In other physical systems, so for instance, some of the early work is by folks Like Kim and Monroe in in in in not neutrals, but you know, actual ions. there, for instance, it's very easy to or they're very good at manipulating atoms. But atoms, it's very hard to efficiently collect the light. So there's a just a big physics mismatch problem between the macroscopic object that collects their photons and the size of the atom. So they were then pushed to a modularity scheme where you draw a very strong line there to say things inside the module can be deterministic. Things outside the module need to be measurement based, need to use tricks like Hango Mandel to build back up entanglement among these systems that I can't interact with very easily directly. That may not still be true. Maybe these days they're much better at catching photons efficiently and routing them over fibers to another array. but for superducting circuits, it's just absolutely not true. Our atoms are sort of gigantic. our ability to make an atom talk to a transmission line is sort of so big that it's not even a parameter we talk about. So in our system, I don't know that they need to be different. and that's kind of some of the strategies that we have been looking at. So in our system, in particular, like with the quantum router, for instance, that particular experiment pushed us to think about the fact. So in that system, we really did sort of think about the other sort of method at first of of Hango Mandel type experiments. But then we realized actually we can move these photons around pretty efficiently. Maybe there is a cost, but maybe for instance, if my s if my error correction code knows where the line of not so good connections are in my system, maybe I can even tolerate that. Maybe I can plan for it. But I don't know that there's some huge difference between things inside a module and things across it. Or at least I'm trying to pretend that there isn't and see if I can still get away with building my circuit.
Daniel Rodan Legrain: So how should we think about that detachable quantum module?
Michael Hatridge: I really it's a good question. I think it really depends on how you fabricate it. If you're going to print chiplets, let's say, as many companies are, an Indian bump bond them down, then maybe that's like a one time irreversible thing. we in the laboratory write work in in 3D circuits, which are maybe not the most manufacturable to millions, but that's okay. I'm not gonna build millions in my laboratory. We really can test them in pieces, unhook them, pull out bad qubits. so in the academic laboratory we're gonna focus on everything being sort of build upable in stages that I can test components, fuse them apart, pull them back together. Because in a university laboratory, you know, we don't have the billion dollar fab, or or multi-billion dollar fab, I guess, that that some of these companies are building. And so we're gonna ruthlessly exploit this to build small systems, you know, literally up in stages. Maybe if you are a company, that's not that interesting to you, you're gonna ship these things off to a customer anyway. So you just need to somehow pre-qualify, but then one time fuse them together. So it it's it's kind of hard to say something too general without some of these sort of boundary conditions of of how you're making them and what your ultimate objective is.
Daniel Rodan Legrain: Sounds good. Listen, one thing I find especially interesting about this is that it isn't simply about an attempt to reproduce, I would say, a large industrial road map in Southern University. I mean, it really raises the broader question, as you very well pointed out, of what should academic laboratories be doing as the field industrializes. And you've described the HATLAB in terms of building circuits that are big and strange. enough to push the field in new directions. So what does big and strange mean to you? And what kind of questions does that philosophy lead an academic life like yours to pursue?
Michael Hatridge: Yeah, it's a good question. It's sort of like a exhortation internally to keep us looking ambitiously at like the biggest systems we think we can put together. that is my motto, but for instance, the paper that we're just about to put out focuses on individual transmon qubits because in our business of building big machines we found some details of the transmon that were still really exciting to study at the one qubit level. But I don't think one qubit level research is gonna teach you how to build a quantum error corrected machine. So big enough and strange enough is us trying to say, okay, if we have a new idea, what's the minimum size circuit we can put together that gives a fair test of that idea? If it's five qubits, we should do that. If it's like 500, probably not, right? But it's not to just sort of sit with these smaller systems only. We definitely do a lot of that research, but we we hope to, for instance, build systems that have four and eight, you know. I would love to build circuits with up to 20 qubits in them if my lab can master it in this modular fashion. I don't know exactly where that boundary is gonna be, but I suspect that if I was an industrial partner reading, you know, our papers or talking to our students when they go in the job market, that effort is gonna help them link up with you know the companies which absolutely are gonna build those size of circuits. You know,
Daniel Rodan Legrain: Sounds great.
Michael Hatridge: and we're not gonna I mean and maybe another way to say it is like we're not gonna scale the surface code. Like that that You know, folks with a lot of money have locked that down pretty well. but maybe that's not the right thing to build. Well, if I want to make that statement to you in a believable way, I probably should build some little piece of it to prove it out. And and so in my lab we we're trying to build both be good at building big enough systems that we can ask those questions and answer them, and also sort of keep an eye on what those opportunities are for systems that are not too big.
Daniel Rodan Legrain: Sounds fantastic, Michael, and makes perfect sense. So since we're talking about universities as well, maybe as a whole, I mean, they play a major role in building the quantum workforce. And as quantum hardware becomes increasingly multidisciplinary and increasingly engineered, would say, how should universities be training the next generation of scientists and quantum engineers? And what are the skills? that you think the field is gonna be needing the most.
Michael Hatridge: It's a really interesting question. I would say that probably when I was a graduate student, the background of the folks many of us was like really purely physics. So these were like often condensed matter folks in in superdonctric circuits that were making this transition to this new area. If I look at the young students, they know a lot of computer science compared to what we used to know. They, however, are not necessarily like physicists or condensed matter or low temperature physicists in the same way that they used to be. So I think what you see in the laboratories is that the actual sort of students who come out of them have changed like the focus of their education to a certain extent, what they know and and what they they they focus on. maybe they don't build their own dill unit anymore. Like someone in, you know, John Clark's lab could have done 30 years ago, they've they've replaced that knowledge with sort of a different orientation. So I think that's kind of happening organically in the laboratories. one of the things, so for instance, the University of Pittsburgh that I I I worked on was to try to ask, well, educationally, like how does that change the classes that you take? How does that change the structure of the the the the major itself? And so at at at at Pittsburgh, right now they have a physics and quantum computing major for undergraduates. That is trying to sort of make this turn, that is sort of trying to blur what were previously some well-established boundaries to create folks who have the right tool set to work on these machines. In the PhD program, PhDs have always been far more individualized, right? Individual laboratories, individual courses of study, individual projects within that laboratory. We don't have as much of a maybe a need for. How to make that adaptation, but I do definitely see it growing. So for instance, we I have advised one joint CS physics student. I imagine that that will happen more in the future. And eventually, as that sort of pathway grows, we'll probably try to figure out the the the sort of the class, certificate, what have you kind of supports to make that easy on folks.
Daniel Rodan Legrain: Sounds great. So, I mean, as universities compete with companies and national labs for talent, how do you keep a university lab and its students thriving, Michael?
Michael Hatridge: well at the graduate student level, the fact that there are all these companies and national labs helps, right? So we we we've probably never had more graduate student applicants at this moment than you know than this is probably the most we've we've we've had because there is actually a very clear pathway out of the laboratories. again, calming back to my own time as a graduate student, you know, there was not an industry that was gonna hire me to do ultra low field MRI. That just wasn't a thing. so I think at the graduate level it's actually helpful. What it makes harder are postdocs and and junior faculty members, because you know, once they've been trained in a lab like mine, then they're very attractive to the company and so then the company can pay them a lot of money, can offer them sort of a more direct pathway than sort of the long set of trainings for a postdoc or or a faculty member. That that is more challenging. you know, in terms of what we can do is we continue to to hunt out and find the excellent folks who who have that natural bent to want to work in the more academic environment. It's maybe not
Daniel Rodan Legrain: that.
Michael Hatridge: the the easiest way to solve the problem, but that you know, that's that's where we where we are at this moment. And and fortunately there are enough of those folks still that want the let's say the broader set of freedoms of an academic lab, the more curiosity driven thing, not having to say, Is it on my roadmap for the next decade? that we still sort of make that work. But it's definitely a lot a lot a lot harder than it used to be a decade ago.
Daniel Rodan Legrain: I guess it goes both ways. But I think that brings us nicely to the conversation of the broader future of the field. mean, Michael, if you had, say, a five-year wish list for superconducting quantum hardware, either measurement, control, architecture, instrumentation, anything that you would most like the community to solve, what would be at the very top?
Michael Hatridge: I guess I would like to see experiments that really attack fault tolerance directly. So we have, you know, individual error corrected qubits sort of at the breakeven level or a little bit past breakeven level. It would be really nice to see some kind of multi-logical qubit processor that really like checks all the boxes of of fault tolerance just to show to put down a marker to show that we can achieve that level and then sort of push past Push past.
Daniel Rodan Legrain: I mean, you've really seen the field evolve from foundational circuit QED and quantum measurement experiment to what's now increasingly complex processes and full systems engineering. What would you say has changed most in the way we think about building a quantum computer?
Michael Hatridge: It's an interesting question. I I think everything is hard the first time you do it. The Transmon, you know, was a mm I want to say it was more mysterious back in the day. Maybe it's just as mysterious these days, but we're understanding it at a much finer level. So but what I have found over the past man, I don't want to say twenty years, but that's increasingly a true number. is what happens is you have the accumulation of all these ideas and concepts that the first time you really struggle to understand. And then you sort of put them in your repertoire, in your tool set, and use them to understand the next thing and you understand the next thing. And what's really amazing after twenty years is sort of how much about quantum mechanics, quantum engineering as a real experimental field we now just sort of take for granted. I do think that's kind of how the field evolves. Like we just collectively know these systems a lot better than we used to. And now, you know, we worry about things. not that you couldn't have predicted twenty years ago, but that you had so many glaring problems right in your face that they they they kind of weren't on your radar, you know? But we've cleaned up so much, we know so much that now we really push on the boundaries of for instance, how hard, how fast, how much, how coherent, you know, the We were pretty excited to have a nine hundred nanosecond transmod when I was a postdoc. now if you walk up and tell me you have a nine hundred microsecond transmod, I'm like, probably. Like that's amazing. That's like a thousand fold on just this one aspect. But there's a lot of control systems, cryo, you know, fab that go into these machines that we can build these days. It's kind of amazing how much we can keep in our heads.
Daniel Rodan Legrain: And what would you say is the thing that people in the field currently underestimate? And at the same time, what do you think is overhyper? That's a tough one.
Michael Hatridge: Yeah, I I think I'm gonna decline on the overhyped one. I'm I'm happy to be an academic researcher, let's say, and and to not be answerable to the stock market. that's a different set of problems. In terms of things we are underestimating, I yeah, I don't know if it's a matter of underestimating. Again, I think it's this question of how many complicated things we can stitch together at once. we're very good at controlling our superconducting quantum circuits these days. You know, our our our our colleagues in the other areas like neutral and ions have also made big steps forward. but it's not clear that we have that we've mastered, you know, the full stairway of complexity up to fault tolerance. And it's not clear to me. whether it's getting easier or whether each step is just a little bit harder than the one before it. So I don't think it's like a particular topic we're underestimating. It's like the question of how much complexity can we really master.
Daniel Rodan Legrain: So that brings me to another question, Michael. What milestone would convince you that the superconducting quantum computing field has really entered a new, genuine phase?
Michael Hatridge: I well I guess it comes back to this question. I would like to see a multi-logical bit processor, you know, running the full the full sort of a game plan of fault tolerance. And I think we will get there, but I'm not sure that that's where we are right now.
Daniel Rodan Legrain: Sounds good. One last question from me and then we open it to the audience, Michael. What gives you the most optimism about the next decade of quantum research?
Michael Hatridge: I actually really like working with graduate students and postdocs because they come every year fresh to it. You know, so I I keep getting older and maybe more set in my ways, but I'm pushed out of that comfort zone by a bunch of students who are not that interested I not the how not the right way to say it, are not that beholden to the things that happened thirty years ago and are intensely curious and intensely driven and also looking at the problem for the first time. So like my favorite thing, the thing that gives me the most optimism in our university based research is that we always gonna have this stream of of of created new members of the field that are going to push us forward.
Daniel Rodan Legrain: Sounds great. I think that's a great place to bring in the audience. So, you know, let's take some questions from everyone watching live. And I think you may have convinced quite a few people with what you explained and in particular about getting new PhD students because I already see a question about whether you're seeking PhD students at this time, Michael.
Michael Hatridge: Yes, we are always hiring.
Daniel Rodan Legrain: And what would you recommend to a new student that is interested in entering the field and eventually working with you?
Michael Hatridge: You know, I I don't know. There are a lot of different tr trajectories folks take into the laboratory. I think you should develop the skills that make you sort of interested in in working with someone like me and that make it easier for us to identify you as someone to talk to. But those can be very varied. Some of my students love nanofabrication. Some of my students hate it with a passion. You know, some of my students love Deep programming. Some students really want to get their hands onto the the hardware. So it's not like I have a very set checklist I'm looking for in the students. We're looking for folks who are curious, who are driven, and can show some evidence of doing pretty hard, random looking tasks. And we're pretty generous in trying to figure out what that might look like because the lab itself keeps changing. You know, like definitely, for instance, if I picked my starting skill set twenty years ago. It would not be the one that, you know, we would use in the lab these days.
Daniel Rodan Legrain: I can hardly think of better places to do a PhD such as Yale and your own group, Michael. Here's another question from the audience. Could you elaborate a little bit more on some of the challenges that are related to research problems versus the ones industries have with respect to mindset and with respect to ways of attacking things or any other thing that you can think of? How do you compare the challenges between academia and industries?
Michael Hatridge: Yeah, that's a good question. So the Industry's job in some sense is to take all the beautiful physics and turn it into well behaved machines that, you know, work every day of the week. The academic's job is to pursue where things don't add up, where things don't make sense, to open up those new things and to ask like maybe should we should throw it all away. So they're very complimentary, but they just have a different mindset. And, you know, take it with a grain of salt, anything I say about industry, right? I mostly work in academic labs, though I do have Former students and and folks I engage with in industry. But I think the essential difference is a quantum engineer's job is to like squeeze all the exciting physics out of their qubits so that they behave. So they can make a million of them and they're all on board. In some sense, an academic physicist is to always work on the edge of the ones that are, you know, completely out of control, to really push on the boundaries. And to hopefully discover the next fundamental technology, the next fundamental approach, but not to keep to not to Yeah. Not n to always live on that boundary, maybe then behind them there's a a lot that still needs to be figured out by folks who are gonna make them sort of more regular and more repeatable.
Daniel Rodan Legrain: Sounds great, thank you. Here goes another one. Could you explain a bit more about error correction, the measurements? How do you measure the error and how do you think about the syndrome qubit and the data qubit?
Michael Hatridge: Yeah, that's a good question. depending on the error correction scheme, you may or may not have explicit syndrome or what we call encilla qubits. The overall idea, which comes all the way back to the to like a three-bit code. In a three-bit code, classically, if you have three ones as your logical one and three zeros as your logical zero, you can have one flip. and then you can say, well, there's still two ones left, so that's a one still. Quantum mechanically, we're not allowed to look at the individual bits, but we can still extract that same idea. So for instance, we don't say what's qubit one, what's qubit two, and what's qubit three and add them up. That's what you'd do classically. Instead, we get at it sort of indirectly. We say are qubit one and two the same or different? Are qubit two and three the same or different? And what we do is we design the states that we use, the way the system is built, is so that if there's no error, asking these questions is completely boring. But if there is an error, it still kind of leaves the state more or less alone. It doesn't do something drastic to it, but it gives us the information to infer which state we have. And so building an extra concode, you it's very complicated. Many of my theory colleagues spend nothing their their whole career on nothing but this. But if you were to like allow me to simplify, we know that errors can take your good state and move it to another part of state space. So error corrected machines need to be big. They need to have enough states possible within them so that all the allowed errors uniquely map to new states. As a measurement expert, what we want to then do is to build the measurement system in such a way that we extract which error has happened from the properties of states without ever measuring the state directly. you know, the the easy way to think about this is that we make a lot of use of this idea, for instance, are you the same or different for two qubits? That's what we call parity. And then you can extend this idea to ask like what's the joint parity of four qubits? That's what the surface code uses. Or, you know, in the various codes like the short code, there might be like what are called weight f is it five parodies, weight three parodies. So once you kind of say, Okay, I'm gonna use indirect measurements, learn the property of did the error correction happen sorry, error happen or not. But never measure the state directly, never in the computational basis. that's in a sense what we exploit you know, in in in of course much more complicated ways in some of these quantum area correction codes.
Daniel Rodan Legrain: Thank you, Michael. We have time for one more. People are wondering whether you also research tupa amplifiers and if it's something you're interested in as well.
Michael Hatridge: okay, wow, that's a pretty granular question. we have not historically researched TUPA amplifiers, partly because you know my laboratory was not in a place where we could build them for a long time, but partly also because I guess of my stylistic choices. So in fact our amplifiers these days are sort of dead simple. They have like 10 or 15 junctions in them. And so although the tupa is very successful, many of my colleagues build them, you know, they're definitely a very excellent amplifier these days. they have thousands of junctions. And so part of me being who I am is to say, okay, but like how about 15? Can I get most of these features? And in fact, we do really well, for instance, on properties like quantum efficiency. I think you know we still have the edge there. but even saturation power, we have learned how to build amplifiers. And I think if you show me a tupa, I can build you a few junction device with much, much fewer junctions that will match that property. I cannot match the directivity. I'm still gonna use microwave circulators. In my own research, that's been fine. but in some sense, we we don't build them ourselves, but obviously we're influenced by the fact that they exist and what properties it can and can't have. We just sort of use it as fuel to drive our development of the cavity-based amplifiers.
Daniel Rodan Legrain: Michael, we're reaching the top of the hour, so let me stop here and say thank you so much for sharing your time and insights with us today. I mean, it's really been a fascinating conversation from your journey through Berkeley, Yale, Pittsburgh, the quantum measurement, back action, parametric amplifying and control, modular architectures, and the role of academic labs in exploring new ideas. as well as training the next generation. I think we've covered a lot of topics today. At QBlocks, our motto is building quantum together. And today's conversation has really captured that spirit, mean, the idea of that project, progress in quantum really requires deep physics, careful engineering, creative architectures, and a community basically willing to explore a new path. As a small token of appreciation, we've sent one of our QBlocks mugs your way. Hopefully it becomes a good companion for the next round of HasLab experiments. For everyone watching, the conversation will also be available on Quantum Builders on Spotify and Apple Podcasts. And since this session marks roughly one year since we started this series, thank you as well for everyone who has joined us over the last year. We still have many conversations ahead, so please stay tuned for more upcoming Five Star Chats and Feather Guests. Michael, thank you again.
Michael Hatridge: Great. Thanks for inviting me and thanks everybody for listening.
Daniel Rodan Legrain: And thank you everyone for joining us today.
Michael Hatridge: Hey.
Daniel Rodan Legrain: I am Daniel Rodan Legrein, the Quantum Science and Engineering Lead for North America at Qblogs. And it's a pleasure to welcome you all to today's Quantum Builders FISA Chat. This session is also a little bit special for us because we're actually celebrating one year of Quantum Builders. We started this series around this time last year with Will O'River. And since then, we've had the opportunity to really host a remarkable group of invited speakers from across the whole quantum ecosystem. including Berde Jong, Peter Shore, Steven Giver, Gretchen Campbell, Chuck Black, David Afshalom, Kyle Cerniak, Leven van der Sijpen, Irfan Siddiqui, Christoph Josak and Laura Schulz, among others. The idea behind this series has always been to bring together leading voices that are shaping the quantum science and engineering field and to have open conversations about where the field is going, all the way from foundation science to quantum hardware and scaling, systems engineering, and the broader quantum ecosystem. So today I'm especially excited to welcome Professor Michael Hatridge. Michael is an Associate Professor of Applied Physics at Yale and he's also the Deputy Director of the Yale Quantum Institute. His group, the Hat Lab, as many of you may know, builds superconducting quantum circuits and they also have a particular focus on quantum measurement, parametric amplification, parametric control, engineering path. and modular superconducting quantum architectures. His pass through the field is also a fascinating one. Michael is from Texas. He received his bachelor in science from Texas A and then his PhD from UC Berkeley working with John Clark. He then did his postdoc at Yale in Michel de Vohette group where he focused on quantum measurement and high fidelity superconducting qubit readout. And then he later built his own group at the University of Pittsburgh, where he also served as the director of the Pittsburgh Quantum Institute before returning to Yale, where he is now. What they really found very interesting about Michael's work is the way that it really connects several layers of superconducting quantum computing that are very often discussed separately. mean, the physics of quantum measurement, the hardware required to amplify very weak signals, the use of nonlinear and parametric interactions for control. And ultimately, I would even say the architectural question of how we build systems that can really scale. So Michael, thank you so much for joining us today.
Michael Hatridge: Yeah, thanks for having me.
Daniel Rodan Legrain: For everyone watching live, as per usual, please feel free to post your questions in the chat as we go and I will make sure to bring a few of them into the conversation towards the end. So Michael, let's start with your own journey through the field. You've worked across several major centers of superconducting quantum research. mean, doing your research, your PhD at Berkeley with John Clark and your postdoc at Yale with Michel Devore. Then you built the HATLAB. lab at Pittsburgh and now returning to Yale. So what would you say were the key inflection points along that path and how has your way of approaching quantum hardware evolved throughout these years?
Michael Hatridge: Yeah, that's a good question. I don't things have worked out very well for me. I I cannot claim that, you know, I had a very simple or or carefully curated way through the field. So in my PhD, for instance, John Clark's lab had a qubit portion of the laboratory. I was actually working on on low-field MRI and we studied prostate cancer, we studied breast cancer, we studied the reaction of your brain to sort of thinking, you know, people like functional MRI try to see processes in the brain. The through line I have is sort of that I was always an instrument builder. and that's kind of really I think the underlying DMA. We you talk about seeing amplifier, seeing measurement always in there. I was always always building some kind of measurement device. but I didn't measure a qubit in my PhD. I didn't measure a single qubit until my postdoc. We didn't do a two qubit gate, like I personally didn't do a two qubit gate until I was a professor. So for me it's it's just been a lot of serendipity and sort of following these strange instruments that we were building into this quantum field. The the first
Daniel Rodan Legrain: Very interesting.
Michael Hatridge: the first paramp I built was definitely on accident. We were trying to build a better magnetometer in the same style that we had used for detecting like MRI signals. And and so we tried to for for people who are old. The DC squid used to be an object that had resistive shunts and you would current bias it and measure a DC voltage. That was like the DC and a DC squid. And we we wanted to take those away and use these new circuit QED type techniques to build a dispersive magnetometer. And it worked really, really well. It worked in fact too well for us to explain until we figured out that like accidentally the magnetometer had a paramp hidden inside it. And so that's kind of like a very unintentional link. The magnetometer sort of worked too well. And and basically very quickly we learned to stop calling it a magnetometer, but to use the same circuit to amplify qubit signals. So this is what I mean when I say like a lot of it wasn't intentional.
Daniel Rodan Legrain: Well, I guess that's how a lot of the research goes as well, Michael.
Michael Hatridge: Yeah.
Daniel Rodan Legrain: So Michael, both your PhD advisor, John Clark, and your postdoctoral advisor, Michel Devoguet, received the 2025 Nobel Prize in Physics. I don't think many people can say that. Having trained closely with two of the three laureates, what perspective did the Nobel announcement give you on their contributions? And what did you learn from them that may be harder to appreciate just by looking at the published science?
Michael Hatridge: Yeah, that's a good question. I guess, amusingly, when I moved out to Berkeley, I got married shortly thereafter, and my wife was trying to figure out kind of what she'd gotten herself into. and she looked up an article that said maybe John Clark, this was like two thousand five, would win the Nobel Prize for this MQT experiment. And I said, Yeah, wouldn't that be great? so I guess that dream was always sort of baked into to to my interactions with John and Michelle. and you know, it was pretty exciting what 20 years later when it actually happened. you know they both are very, they're both very good mentors. They taught me a lot about how to do science, how to, you know, just sort of everything about the the skill set that I use these days. But they are also very concerned with like not fooling yourself with careful calibration, careful control experiments, saying what you know, saying what you don't know. I think the MQT experiment has a lot of those flavors in it, that they, you know, then carried forward for the forty years of research they did after that. and certainly I was part of that legacy in learning those skills in their laboratories.
Daniel Rodan Legrain: Wonderful. So following up a little bit on the story you explained at the beginning, when did you realize that quantum measurements and superconducting qubits would become the direction that you would want to pursue?
Michael Hatridge: Yeah, I think for me it's particularly easy to date. There was a 2011 Le Zouche sort of school on quantum machines, and and Sasha Karatkov was there talking about the back action of weak quantum measurements. And not to go like too off the rails into the physics, but when for instance an atomic person measures a single neutral atom or ion floating in space, they scatter many, many photons off it. And the first one pretty much carries away the quantum information, whether the s the atom is in the ground or excited state. But you almost never catch that photon. And so you get this picture in your mind, which a lot of us in like the microwave superconnecting world inherited, that the state function very, very quickly collapses when you measure. And then you sort of build your knowledge back up as you slowly connect the few photons that scatter actually into your detector. And that's certainly what I thought about quantum measurement, that I would not really see. many quantum features. They're kind of destroyed because you kind of first lose the quantum information, then gain the classical information. And Sasha Krotkov's theory on this was that that's not the case. If you do a really weak measurement so that not enough information transit the system for either you or any other observer, right? If anybody had the total information, there still is not enough. Then you would see these unique quantum features. You would see that the qubit doesn't just sort of collapse up or down. It does some really complicated sort of continuous quantum walk. And so that, like Michelle and I agreed that that's the experiment I should try to do when we went back from Lazouche. And there were a lot of arguments. We worked closely with Steve Gervin and Mazir Mirahimi, two theorists, and the theory and the data didn't match up. We had a lot of sort of struggles and back and forth, but basically what was technically new that allows to do the experiment is these new parametric amplifiers allowed the quantum efficiency to be something like twenty to thirty percent. So like a third of the total information we were collecting, which now is like completely passe. You know, people are like 70 to 80 quantum percent percent quantum efficiency is sort of minimum to be impressive these days. But at that time this this different aspect that most folks had not had in their experiment before allow us to do the experiment. And we basically saw that, you know, you think the qubit collapses upwards or downwards. In fact, it does this smooth quantum walk across the across the block sphere and we could very nicely track it. So this I learned a lot of quantum mechanics from this. Previously I had mostly built detectors. So the detectors have various quantum limits, but often they look at very classical things or the information they get looks very classical. So you say, Okay, the noise will stop getting better because of some quantum principle. But this sort of weird quantum back action We set out to find it we found it and it really sort of taught me a lot of what I now know about how to view quantum measurement as a really interesting quantum game we can play and not sort of where quantum mechanics goes to die.
Daniel Rodan Legrain: Fascinating, Michael, thank you. And then if you look back at your career, would you say there was a particular point when you thought, okay, superconducting qubits are not really feeling like beautiful physics experiment anymore, and they start feeling like a real engineering platform for quantum computing?
Michael Hatridge: that's a good question. In my laboratory we like to live right on that line, so I'm pretty happy that they have aspects of both. I guess No, I I yeah, I think that's kind of my answer. We were always engineering on the practical side. We're always trying to do crazy physics with these tools. and i myself, I guess, I find the most fun is along that boundary where sometimes it's crazy physics, sometimes it's engineering, and we sort of inform each other back and forth across that boundary.
Daniel Rodan Legrain: Let's keep talking a little bit more about the measurement. mean a lot of your early work focused on quantum measurement and measurement back action. What did those experiments teach us about what actually happens when we measure a quantum system?
Michael Hatridge: They taught us again that we can see the unique quantum properties of of a qubit or of a couple qubits even when they are being measured. And if you can sort of keep that quantum physics alive all the way to your detector, if your detector itself is an efficient amplifier, an efficient detector, what it means is sort of you can play games. So The neat thing about quantum measurements, if you ask a question like, is your qubit blue or not, you force it to make a choice. And there are certain natural choices that are very easy for your detector to do. In the framework of supernatural circuits, sir, circuit QED, it's very easy to measure the Z-axis of the qubit. Sort of the interactions are set up that if you take a regular amplifier and a regular microwave pulse, and you don't really work hard on their funny quantum properties, you'll get a Z basis measurement. But if you work hard on the quantum basis, you use squeezing as a resource, you use interesting kind of interactions that are other than the circuit QED interaction, and you change the properties of the behavior, sorry, the behavior of your detector. There's this really neat thing where if they're all three quantum, then you can basically build a machine that asks a different question, that measures the qubit along a weird axis, that measures two qubits in a way that what you get is their parity measurement. And classically, This doesn't really make sense. It doesn't really have an analog. Like there's this this question, right? If the tree falls in the forest, does it make a sound? You know it makes a sound. Like everyone no one is really asking this question. There's some massive boom out there and it doesn't require an observer. But in quantum mechanics, if you are what we call the omniscient observer, you're the only one who collects the information, and you collect it in such a way that you are deliberately blind, you don't learn certain things. That's actually good. And that's really, I think, funky and sort of unique quantum property. And it's kind of, for instance, what error correction uses. Error correction very carefully controls the system in such a way that you never ask, like, what state are you in? You ask, did you have an error? And if so, what type? And and classically, I think we don't have the intuition for these kind of measurements because we don't have this business of the observer, the act of measurement being really I mean We have some examples like bleaching and stuff like this, but we don't have this idea of of really interesting and exciting quantum collapse. So that's sort of been what we work on. These experiments are typically very hard. And so what you'll actually find, I think in my lab, sort of in the engineering picture you're asking before, is there's sort of the wackiest thing we can do, the most complicated measurement we can make work once. And then a few steps back from that are the things that we can make work regularly, and then we work hard to engineer those into quantum systems.
Daniel Rodan Legrain: Wonderful. Let's follow up on what you mentioned about amplifiers. And you have worked and your group has spent a lot of time on quantum-limited parametric amplifiers. So maybe for a broader audience that may be joining us today, what makes an amplifier quantum-limited and what makes a good amplifier in practice for high fidelity
Michael Hatridge: Mm-hmm.
Daniel Rodan Legrain: superconducting qubit readout?
Michael Hatridge: okay. let me answer the second one first. So if you go shopping for a microwave amplifier, it will have a data sheet. most of the properties of the quantum amplifiers we we we build would be considered mediocre or sub-mediocre compared to like the amplifiers that you buy. So the bandwidth that we can build, the saturation power that we can build, the power draw that the amplifier has, all those classical properties. Translate very well into these amplifiers, and the usual answer is that you want more of all of them. the one sort of magic thing about the amplifiers is that again, quantum mechanically, are you guys hearing some kind of feedback or somebody else's clicking? Maybe it's just me. Okay. Okay.
Daniel Rodan Legrain: I can hear you great, Michael.
Michael Hatridge: so The the unique quantum feature is that the amplifier itself, and and this was sort of a debate we had many times in our post in my postdoc with Michelle that we built experiments to try to test, the amplifier itself is a quantum object. That has weird consequences. That for means, for instance, you don't think that like classical light comes in the amplifier and a class sorry, quantum light comes in and a classical answer comes out. It does a quantum Translation. And so for instance, you can put amplifiers back to back and they can undo each other even at a quantum mechanical level. Or the properties of the amplifier will transform the light in a way that's interesting and quantum. The quantum limit here basically says that if you're gonna use this amplifier to transform the light in a way you'd like to digitize, for instance, at room temperature, you have a choice to make. And basically it comes back to Heisenberg's uncertainty principle. You can learn one quantity of a microwave field perfectly if you destroy the other. So if you think of a microwave tone as having an amplitude and a phase, kind of it's a little bit imprecise, but you could learn either the amplitude or the phase, but not both. And that's what we call a phase sensitive amplifier. So that one really has no quantum limit. The the the sacrifice you make is that you know if your information is in both quadratures of the light, you can't record it all. So of course you need to like use that amplifier in scenarios where all your information lies in one quadrature. Let's say only a small phase difference or only a small amplitude difference. sometimes you don't want to make that choice. And so you ask what's the best thing I can do that doesn't favor a particular kind of information. There it turns out you cannot learn both properties of a field with infinite precision. And so you you work it all out, and we find that just like in in Heisenberg's various like microscope and these various pictures of this experiment. There's a minimum uncertainty, which basically is is is the vacuum noise itself. So a quantum limited amplifier is either like one of these two types in in our microwave world, and we've pushed these properties to the ultimate limit. You either learn one thing perfectly by destroying the other, and again in this weird quantum way, destroying it so that no other observer could get it. It's not that like you ignore it and just don't write the answer down. You really have to squeeze it away, hide it away. Or you you don't choose. And then you get that there's basically this other source of fluctuations in your system, which is unavoidable. Sorry, that's maybe not a very helpful disquisition, but
Daniel Rodan Legrain: That was a great answer, Michael. But let me now follow up with one more question, maybe a more practical one now. Would you say readout is still one of the central bottlenecks of superconducting quantum computers? And maybe even going one step forward, what is still missing from today's qubit measurement toolbox?
Michael Hatridge: Yeah. I do think readout, if you look at various papers papers from groups like Google, readout is often cited as like the worst thing in their system. And part of the problem there is that like a fidelity of a qubit, you make that better by sealing it up in a box and protecting it from everything. You know, often fidelity of a gate is somewhat the same. If you look at readout, there's the qubit interaction with the readout interaction. Apparatus. There's the moving the information across your cryostat, and then there's the how you process that information in the quantum regime. So readout kind of has a problem that, you know, for instance, the properties of my readout are not specified by the quantum processor. They're not specified by the amplifier alone. It's the whole system bolted together with all of its n you know warts and defects that come together to make the quantum efficiency. So for instance, as an amplifier designer, We think we can get seventy to eighty percent quantum efficiency of the amplifier. we do give these out to folks who who work on some of the same government grants as us, but if you were to put a lossy cable in front of it, your readout won't work for beans. My amplifier cannot perform. Right? Also, if your qubits are very decoupled for the environment so that you can only extract the information very slowly, then the qubit will decay before the information can get out. So I think Most of the properties of the quantum computer are what we would call like closed system properties. They they involve how we behave when you isolate it as well as you can from the environment. Readout is the one that says how do you behave when you're interacting in a delicate and controlled way, but as strongly as you can with like the right environment. So I think it has a natural handicap that way. the other sort of general question that that we continue to learn also is that because we're in a hurry, we need to read out the qubit. 10,000, 100,000 times faster than its lifetime. I actually think it shares this property with Gates that both Gates and Radout want to push you to your edge. And you know, there's still, believe it or not, complications, wrinkles, and even the simple Transmon qubit that we're still sort of carefully documenting these days in terms of not what happens when you're not driving very hard. I think that's pretty well understood. But what, you know, what is the limit of just how hard you can push these systems to push these fidelities up?
Daniel Rodan Legrain: Sounds great. Lots of work to do in the years to come. So let me switch gears a little bit, Michael. I mean, we've talked about amplification, but a lot of your work uses parametric drives, not just for that, but also as a much broader tool for quantum control. For maybe those in the audience that may not be directly working with superconducting qubits, what does parametric control mean and why is it so powerful?
Michael Hatridge: Yeah, that's a good question. And and the the term is it's well baked into the field, but it I it's pretty misleading. So a real parametric control, the the typical example folks will give is somebody on a swing. So, you know, typical driving, resonant driving is where s your person A is on a swing, person B is pushing them. So they're resonantly driving you. If you push the person at the frequency of the swing, then you'll drive oscillations and they'll oscillate up more and more. parametric, the reason that we call it a parameter is if instead the the person on the swing, right, because like sometimes you don't have somebody there to drive you, but you can still swing just fine by basically pumping yourself. So what you'll typically do, don't do it this way at home, but it's like easier to think about it, is like stand up on the swing and then squat and then stand and then squat and then stand. And what's
Daniel Rodan Legrain: do it at home.
Michael Hatridge: yeah, don't do it at home. But what you're actually doing is you're driving. This parameter of the system, which is your moment of inertia. And by doing that, like most kids don't do it by standing, but they do it by making their body flat and then sitting up. You can drive yourself just fine. That is the essence of parametric. What it means is there's a parameter, something like the capacitance of something, the inductance of something, the moment of inertia of something that you are driving. It's now instead of a constant, it's some dynamical part of the circuit. There are ways that we drive superducting circuits that are explicitly parametric. You can think of the Joseph injunctions we use typically as inductors. And so if you were to modulate, for instance, the flux through the squid, that is a time-varying inductance that is explicitly parametric. what we really do most of the time in my lab is to just say, well, that's an interesting way to use nonlinearities in the circuit. So the swing, for instance. Is essentially nonlinear. The pendulum itself has a in basic physics you'll learn that the period of the pendulum depends on how hard you drive it. That essential nonlinearity is what we would call a fourth order nonlinearity. And folks in my lab know that that means four wave mixing is inherent to that system. And then we just like use that as a design tool for interactions. So the way we actually talk about in the lab is to say if you have a nonlinear system, You tell me that you have three-wave mixing, four-wave mixing, five-wave mixing, six-wave mixing, like whatever you want. Parametric driving is a way to turn those tools into lower order interactions that I can turn on and off with microwave driving. So it's really in that sense very abstract sounding. It's just like parametric driving, though, is the best way I can explain it. If you don't ask me too hard what parameter of the system I'm modulating. It's still the idea of a kid on a swing pumping themselves. We're still gonna drive away from the resonant frequency of the circuit and use its fundamental nonlinearity to generate some behavior we want. Except in this sort of more Hamiltonian engineering way of thinking, we're very abstract about what the parameter is. It's not something like capacitance, inductance second point two. It's you know, G six. And and that part gets very technical sounding, but but but the overall idea still is is the kid on a swing.
Daniel Rodan Legrain: I mean, your group has really used these ideas for gates, couplers, engineering, interaction, engineering paths. You already started kind of answering this question, but maybe from a more general perspective, what can parametric approaches let you do that would be otherwise difficult or maybe less flexible with, let's say, conventional
Michael Hatridge: Yeah.
Daniel Rodan Legrain: microwave control?
Michael Hatridge: It's a really interesting question. I told you in the very beginning that I came at a lot of this stuff sort of let's say people with a different mindset than me could have maybe gone through my own career in a very forward-looking way. I tended to back my way into things. we learned about parametric interactions because they're very common for parametric amplifiers. Like that's why they have the name in them, right? As I tried to learn about how other people were putting together their quantum computers and how, for instance, they're building gates, there's this period this this this moment in many of these publications, many of these talks, where they say, and here's a nonlinearity, and therefore I drive at this frequency, and everything is fine. I get the interaction that I want. And and I guess in my own personal journey, it's sort of been realizing that, You don't have to do parametric amplifiers first, but if you've lived that lifestyle, you're like, okay, now I understand. It's like a parametric amplifier, except we've changed a few of the terms in the Hamiltonian. And and that's really obtuse. That's maybe me specific. It's a you know, my poor suff suffering students are infected with this mindset by me. but we really came at it backwards. And and then what we found was once you say, you know, here are all these parametric susceptibilities. There's really nothing you can't do with them. sorry, yeah, cannot do with them. Like they can do everything you want. The gate can be parametric. And several of the projects in the laboratory, sort of at some point it kind of became a badge of honor, you know, like that's not parametric, but I was like, I thought we can make it be parametric. I'm not saying that that's the way you you you should do it moving forward, but that's definitely the way we come to view this stuff. And I guess it has a special name, it has a certain way of thinking about it, but it's really endemic. Let's say if you never want to use a parametric interaction in your laboratory, fine. But you're gonna first have to make sure that every frequency you drive the system at doesn't accidentally hit one of these parametric susceptibilities. So it really kind of works both ways. We use them to do all of our controls and compare them against sort of more conventional schemes. But even if you want to ignore them, for instance, the famous readout problem in Transmon qubits, you can pretty easily explain by saying that the cavity drive you're applying is accidentally hitting a parametric resonance of the transmon. So even if you don't believe in them, you know, they don't know that. And so you will turn them on on accident.
Daniel Rodan Legrain: Fascinating, Michael. So I would say that connects naturally to the problem of scalability.
Michael Hatridge: Yeah.
Daniel Rodan Legrain: And your group has really been developing modular superconducting architectures, including microwave, quantum state routers, and detachable quantum modules. So why is modularity, rather than, let's say, simply making one larger monolithic chip, such an attractive path for scaling superconducting quantum computers?
Michael Hatridge: good question. if you look at that famous Google result on their what was it fifty-three-bit machine, one of those qubits was dead. Right? so monolithic scale up looks pretty intimidating because transistors apparently you can make mm, you know, with a vanishingly small failure rate. Supernight qubits are just not like that, and I'm not sure what the prospects are for them becoming sort of ultimately fabbable in a repeatable high precision way. So if you say that your components are gonna be flawed, then you need a way to sort of limit your risk, limit the fact that your fab fails. So in that sense, modularity, I think, is directly tied to the sort of synthetic nature of superconducting circuits. Superconducting qubits are all fabbed, unlike atoms, you know, which have exact properties, are all fabbed very synthetically, which means they're all gonna be a little bit different. They're all gonna have their own character. And if that's fundamental to your building block, You need to truncate the size your system and figure out a way to build it out of pieces. Otherwise, you're gonna have some dead pieces, you know, just contaminating what you're trying to do. So although we have been looking at it in in my laboratory specifically, I don't think it's anything unique to us. I think it's you know the nature of supernecting qubits that are pushing us that way. But even in, for instance, the neutral atom systems, they have other constraints about the optics that they use, about the atomic arrays that they're gonna build. And again, there's something there when when you keep trying to push it sort of monolithically huge, the system is gonna push back on increasing cost and you're eventually gonna then compromise on how to build it out of pieces. So I actually think it's it's really easy to build an argument for modularity. It's it's much harder to imagine pretty much in any quantum system why a monolithic thing is not going to eat your lunch.
Daniel Rodan Legrain: Fair. Okay, so once you decide to go modular, what becomes hardest in practice? Is it the loss? Is it the coherence? Is it the connectivity? Is it the packaging, the fabrication, the calibration? Or is it something else?
Michael Hatridge: yeah, that is a good question. I guess we have been thinking about it a a little bit different way. I'm not so sure that the supernatural circuits again are completely synthetic. Everything is sort of built out of Joseph's injunctions, capacitors, inductors, transmission lines. The way we build connections between modules and the way we build things inside modules are pretty similar. In other physical systems, so for instance, some of the early work is by folks Like Kim and Monroe in in in in not neutrals, but you know, actual ions. there, for instance, it's very easy to or they're very good at manipulating atoms. But atoms, it's very hard to efficiently collect the light. So there's a just a big physics mismatch problem between the macroscopic object that collects their photons and the size of the atom. So they were then pushed to a modularity scheme where you draw a very strong line there to say things inside the module can be deterministic. Things outside the module need to be measurement based, need to use tricks like Hango Mandel to build back up entanglement among these systems that I can't interact with very easily directly. That may not still be true. Maybe these days they're much better at catching photons efficiently and routing them over fibers to another array. but for superducting circuits, it's just absolutely not true. Our atoms are sort of gigantic. our ability to make an atom talk to a transmission line is sort of so big that it's not even a parameter we talk about. So in our system, I don't know that they need to be different. and that's kind of some of the strategies that we have been looking at. So in our system, in particular, like with the quantum router, for instance, that particular experiment pushed us to think about the fact. So in that system, we really did sort of think about the other sort of method at first of of Hango Mandel type experiments. But then we realized actually we can move these photons around pretty efficiently. Maybe there is a cost, but maybe for instance, if my s if my error correction code knows where the line of not so good connections are in my system, maybe I can even tolerate that. Maybe I can plan for it. But I don't know that there's some huge difference between things inside a module and things across it. Or at least I'm trying to pretend that there isn't and see if I can still get away with building my circuit.
Daniel Rodan Legrain: So how should we think about that detachable quantum module?
Michael Hatridge: I really it's a good question. I think it really depends on how you fabricate it. If you're going to print chiplets, let's say, as many companies are, an Indian bump bond them down, then maybe that's like a one time irreversible thing. we in the laboratory write work in in 3D circuits, which are maybe not the most manufacturable to millions, but that's okay. I'm not gonna build millions in my laboratory. We really can test them in pieces, unhook them, pull out bad qubits. so in the academic laboratory we're gonna focus on everything being sort of build upable in stages that I can test components, fuse them apart, pull them back together. Because in a university laboratory, you know, we don't have the billion dollar fab, or or multi-billion dollar fab, I guess, that that some of these companies are building. And so we're gonna ruthlessly exploit this to build small systems, you know, literally up in stages. Maybe if you are a company, that's not that interesting to you, you're gonna ship these things off to a customer anyway. So you just need to somehow pre-qualify, but then one time fuse them together. So it it's it's kind of hard to say something too general without some of these sort of boundary conditions of of how you're making them and what your ultimate objective is.
Daniel Rodan Legrain: Sounds good. Listen, one thing I find especially interesting about this is that it isn't simply about an attempt to reproduce, I would say, a large industrial road map in Southern University. I mean, it really raises the broader question, as you very well pointed out, of what should academic laboratories be doing as the field industrializes. And you've described the HATLAB in terms of building circuits that are big and strange. enough to push the field in new directions. So what does big and strange mean to you? And what kind of questions does that philosophy lead an academic life like yours to pursue?
Michael Hatridge: Yeah, it's a good question. It's sort of like a exhortation internally to keep us looking ambitiously at like the biggest systems we think we can put together. that is my motto, but for instance, the paper that we're just about to put out focuses on individual transmon qubits because in our business of building big machines we found some details of the transmon that were still really exciting to study at the one qubit level. But I don't think one qubit level research is gonna teach you how to build a quantum error corrected machine. So big enough and strange enough is us trying to say, okay, if we have a new idea, what's the minimum size circuit we can put together that gives a fair test of that idea? If it's five qubits, we should do that. If it's like 500, probably not, right? But it's not to just sort of sit with these smaller systems only. We definitely do a lot of that research, but we we hope to, for instance, build systems that have four and eight, you know. I would love to build circuits with up to 20 qubits in them if my lab can master it in this modular fashion. I don't know exactly where that boundary is gonna be, but I suspect that if I was an industrial partner reading, you know, our papers or talking to our students when they go in the job market, that effort is gonna help them link up with you know the companies which absolutely are gonna build those size of circuits. You know,
Daniel Rodan Legrain: Sounds great.
Michael Hatridge: and we're not gonna I mean and maybe another way to say it is like we're not gonna scale the surface code. Like that that You know, folks with a lot of money have locked that down pretty well. but maybe that's not the right thing to build. Well, if I want to make that statement to you in a believable way, I probably should build some little piece of it to prove it out. And and so in my lab we we're trying to build both be good at building big enough systems that we can ask those questions and answer them, and also sort of keep an eye on what those opportunities are for systems that are not too big.
Daniel Rodan Legrain: Sounds fantastic, Michael, and makes perfect sense. So since we're talking about universities as well, maybe as a whole, I mean, they play a major role in building the quantum workforce. And as quantum hardware becomes increasingly multidisciplinary and increasingly engineered, would say, how should universities be training the next generation of scientists and quantum engineers? And what are the skills? that you think the field is gonna be needing the most.
Michael Hatridge: It's a really interesting question. I would say that probably when I was a graduate student, the background of the folks many of us was like really purely physics. So these were like often condensed matter folks in in superdonctric circuits that were making this transition to this new area. If I look at the young students, they know a lot of computer science compared to what we used to know. They, however, are not necessarily like physicists or condensed matter or low temperature physicists in the same way that they used to be. So I think what you see in the laboratories is that the actual sort of students who come out of them have changed like the focus of their education to a certain extent, what they know and and what they they they focus on. maybe they don't build their own dill unit anymore. Like someone in, you know, John Clark's lab could have done 30 years ago, they've they've replaced that knowledge with sort of a different orientation. So I think that's kind of happening organically in the laboratories. one of the things, so for instance, the University of Pittsburgh that I I I worked on was to try to ask, well, educationally, like how does that change the classes that you take? How does that change the structure of the the the the major itself? And so at at at at Pittsburgh, right now they have a physics and quantum computing major for undergraduates. That is trying to sort of make this turn, that is sort of trying to blur what were previously some well-established boundaries to create folks who have the right tool set to work on these machines. In the PhD program, PhDs have always been far more individualized, right? Individual laboratories, individual courses of study, individual projects within that laboratory. We don't have as much of a maybe a need for. How to make that adaptation, but I do definitely see it growing. So for instance, we I have advised one joint CS physics student. I imagine that that will happen more in the future. And eventually, as that sort of pathway grows, we'll probably try to figure out the the the sort of the class, certificate, what have you kind of supports to make that easy on folks.
Daniel Rodan Legrain: Sounds great. So, I mean, as universities compete with companies and national labs for talent, how do you keep a university lab and its students thriving, Michael?
Michael Hatridge: well at the graduate student level, the fact that there are all these companies and national labs helps, right? So we we we've probably never had more graduate student applicants at this moment than you know than this is probably the most we've we've we've had because there is actually a very clear pathway out of the laboratories. again, calming back to my own time as a graduate student, you know, there was not an industry that was gonna hire me to do ultra low field MRI. That just wasn't a thing. so I think at the graduate level it's actually helpful. What it makes harder are postdocs and and junior faculty members, because you know, once they've been trained in a lab like mine, then they're very attractive to the company and so then the company can pay them a lot of money, can offer them sort of a more direct pathway than sort of the long set of trainings for a postdoc or or a faculty member. That that is more challenging. you know, in terms of what we can do is we continue to to hunt out and find the excellent folks who who have that natural bent to want to work in the more academic environment. It's maybe not
Daniel Rodan Legrain: that.
Michael Hatridge: the the easiest way to solve the problem, but that you know, that's that's where we where we are at this moment. And and fortunately there are enough of those folks still that want the let's say the broader set of freedoms of an academic lab, the more curiosity driven thing, not having to say, Is it on my roadmap for the next decade? that we still sort of make that work. But it's definitely a lot a lot a lot harder than it used to be a decade ago.
Daniel Rodan Legrain: I guess it goes both ways. But I think that brings us nicely to the conversation of the broader future of the field. mean, Michael, if you had, say, a five-year wish list for superconducting quantum hardware, either measurement, control, architecture, instrumentation, anything that you would most like the community to solve, what would be at the very top?
Michael Hatridge: I guess I would like to see experiments that really attack fault tolerance directly. So we have, you know, individual error corrected qubits sort of at the breakeven level or a little bit past breakeven level. It would be really nice to see some kind of multi-logical qubit processor that really like checks all the boxes of of fault tolerance just to show to put down a marker to show that we can achieve that level and then sort of push past Push past.
Daniel Rodan Legrain: I mean, you've really seen the field evolve from foundational circuit QED and quantum measurement experiment to what's now increasingly complex processes and full systems engineering. What would you say has changed most in the way we think about building a quantum computer?
Michael Hatridge: It's an interesting question. I I think everything is hard the first time you do it. The Transmon, you know, was a mm I want to say it was more mysterious back in the day. Maybe it's just as mysterious these days, but we're understanding it at a much finer level. So but what I have found over the past man, I don't want to say twenty years, but that's increasingly a true number. is what happens is you have the accumulation of all these ideas and concepts that the first time you really struggle to understand. And then you sort of put them in your repertoire, in your tool set, and use them to understand the next thing and you understand the next thing. And what's really amazing after twenty years is sort of how much about quantum mechanics, quantum engineering as a real experimental field we now just sort of take for granted. I do think that's kind of how the field evolves. Like we just collectively know these systems a lot better than we used to. And now, you know, we worry about things. not that you couldn't have predicted twenty years ago, but that you had so many glaring problems right in your face that they they they kind of weren't on your radar, you know? But we've cleaned up so much, we know so much that now we really push on the boundaries of for instance, how hard, how fast, how much, how coherent, you know, the We were pretty excited to have a nine hundred nanosecond transmod when I was a postdoc. now if you walk up and tell me you have a nine hundred microsecond transmod, I'm like, probably. Like that's amazing. That's like a thousand fold on just this one aspect. But there's a lot of control systems, cryo, you know, fab that go into these machines that we can build these days. It's kind of amazing how much we can keep in our heads.
Daniel Rodan Legrain: And what would you say is the thing that people in the field currently underestimate? And at the same time, what do you think is overhyper? That's a tough one.
Michael Hatridge: Yeah, I I think I'm gonna decline on the overhyped one. I'm I'm happy to be an academic researcher, let's say, and and to not be answerable to the stock market. that's a different set of problems. In terms of things we are underestimating, I yeah, I don't know if it's a matter of underestimating. Again, I think it's this question of how many complicated things we can stitch together at once. we're very good at controlling our superconducting quantum circuits these days. You know, our our our our colleagues in the other areas like neutral and ions have also made big steps forward. but it's not clear that we have that we've mastered, you know, the full stairway of complexity up to fault tolerance. And it's not clear to me. whether it's getting easier or whether each step is just a little bit harder than the one before it. So I don't think it's like a particular topic we're underestimating. It's like the question of how much complexity can we really master.
Daniel Rodan Legrain: So that brings me to another question, Michael. What milestone would convince you that the superconducting quantum computing field has really entered a new, genuine phase?
Michael Hatridge: I well I guess it comes back to this question. I would like to see a multi-logical bit processor, you know, running the full the full sort of a game plan of fault tolerance. And I think we will get there, but I'm not sure that that's where we are right now.
Daniel Rodan Legrain: Sounds good. One last question from me and then we open it to the audience, Michael. What gives you the most optimism about the next decade of quantum research?
Michael Hatridge: I actually really like working with graduate students and postdocs because they come every year fresh to it. You know, so I I keep getting older and maybe more set in my ways, but I'm pushed out of that comfort zone by a bunch of students who are not that interested I not the how not the right way to say it, are not that beholden to the things that happened thirty years ago and are intensely curious and intensely driven and also looking at the problem for the first time. So like my favorite thing, the thing that gives me the most optimism in our university based research is that we always gonna have this stream of of of created new members of the field that are going to push us forward.
Daniel Rodan Legrain: Sounds great. I think that's a great place to bring in the audience. So, you know, let's take some questions from everyone watching live. And I think you may have convinced quite a few people with what you explained and in particular about getting new PhD students because I already see a question about whether you're seeking PhD students at this time, Michael.
Michael Hatridge: Yes, we are always hiring.
Daniel Rodan Legrain: And what would you recommend to a new student that is interested in entering the field and eventually working with you?
Michael Hatridge: You know, I I don't know. There are a lot of different tr trajectories folks take into the laboratory. I think you should develop the skills that make you sort of interested in in working with someone like me and that make it easier for us to identify you as someone to talk to. But those can be very varied. Some of my students love nanofabrication. Some of my students hate it with a passion. You know, some of my students love Deep programming. Some students really want to get their hands onto the the hardware. So it's not like I have a very set checklist I'm looking for in the students. We're looking for folks who are curious, who are driven, and can show some evidence of doing pretty hard, random looking tasks. And we're pretty generous in trying to figure out what that might look like because the lab itself keeps changing. You know, like definitely, for instance, if I picked my starting skill set twenty years ago. It would not be the one that, you know, we would use in the lab these days.
Daniel Rodan Legrain: I can hardly think of better places to do a PhD such as Yale and your own group, Michael. Here's another question from the audience. Could you elaborate a little bit more on some of the challenges that are related to research problems versus the ones industries have with respect to mindset and with respect to ways of attacking things or any other thing that you can think of? How do you compare the challenges between academia and industries?
Michael Hatridge: Yeah, that's a good question. So the Industry's job in some sense is to take all the beautiful physics and turn it into well behaved machines that, you know, work every day of the week. The academic's job is to pursue where things don't add up, where things don't make sense, to open up those new things and to ask like maybe should we should throw it all away. So they're very complimentary, but they just have a different mindset. And, you know, take it with a grain of salt, anything I say about industry, right? I mostly work in academic labs, though I do have Former students and and folks I engage with in industry. But I think the essential difference is a quantum engineer's job is to like squeeze all the exciting physics out of their qubits so that they behave. So they can make a million of them and they're all on board. In some sense, an academic physicist is to always work on the edge of the ones that are, you know, completely out of control, to really push on the boundaries. And to hopefully discover the next fundamental technology, the next fundamental approach, but not to keep to not to Yeah. Not n to always live on that boundary, maybe then behind them there's a a lot that still needs to be figured out by folks who are gonna make them sort of more regular and more repeatable.
Daniel Rodan Legrain: Sounds great, thank you. Here goes another one. Could you explain a bit more about error correction, the measurements? How do you measure the error and how do you think about the syndrome qubit and the data qubit?
Michael Hatridge: Yeah, that's a good question. depending on the error correction scheme, you may or may not have explicit syndrome or what we call encilla qubits. The overall idea, which comes all the way back to the to like a three-bit code. In a three-bit code, classically, if you have three ones as your logical one and three zeros as your logical zero, you can have one flip. and then you can say, well, there's still two ones left, so that's a one still. Quantum mechanically, we're not allowed to look at the individual bits, but we can still extract that same idea. So for instance, we don't say what's qubit one, what's qubit two, and what's qubit three and add them up. That's what you'd do classically. Instead, we get at it sort of indirectly. We say are qubit one and two the same or different? Are qubit two and three the same or different? And what we do is we design the states that we use, the way the system is built, is so that if there's no error, asking these questions is completely boring. But if there is an error, it still kind of leaves the state more or less alone. It doesn't do something drastic to it, but it gives us the information to infer which state we have. And so building an extra concode, you it's very complicated. Many of my theory colleagues spend nothing their their whole career on nothing but this. But if you were to like allow me to simplify, we know that errors can take your good state and move it to another part of state space. So error corrected machines need to be big. They need to have enough states possible within them so that all the allowed errors uniquely map to new states. As a measurement expert, what we want to then do is to build the measurement system in such a way that we extract which error has happened from the properties of states without ever measuring the state directly. you know, the the easy way to think about this is that we make a lot of use of this idea, for instance, are you the same or different for two qubits? That's what we call parity. And then you can extend this idea to ask like what's the joint parity of four qubits? That's what the surface code uses. Or, you know, in the various codes like the short code, there might be like what are called weight f is it five parodies, weight three parodies. So once you kind of say, Okay, I'm gonna use indirect measurements, learn the property of did the error correction happen sorry, error happen or not. But never measure the state directly, never in the computational basis. that's in a sense what we exploit you know, in in in of course much more complicated ways in some of these quantum area correction codes.
Daniel Rodan Legrain: Thank you, Michael. We have time for one more. People are wondering whether you also research tupa amplifiers and if it's something you're interested in as well.
Michael Hatridge: okay, wow, that's a pretty granular question. we have not historically researched TUPA amplifiers, partly because you know my laboratory was not in a place where we could build them for a long time, but partly also because I guess of my stylistic choices. So in fact our amplifiers these days are sort of dead simple. They have like 10 or 15 junctions in them. And so although the tupa is very successful, many of my colleagues build them, you know, they're definitely a very excellent amplifier these days. they have thousands of junctions. And so part of me being who I am is to say, okay, but like how about 15? Can I get most of these features? And in fact, we do really well, for instance, on properties like quantum efficiency. I think you know we still have the edge there. but even saturation power, we have learned how to build amplifiers. And I think if you show me a tupa, I can build you a few junction device with much, much fewer junctions that will match that property. I cannot match the directivity. I'm still gonna use microwave circulators. In my own research, that's been fine. but in some sense, we we don't build them ourselves, but obviously we're influenced by the fact that they exist and what properties it can and can't have. We just sort of use it as fuel to drive our development of the cavity-based amplifiers.
Daniel Rodan Legrain: Michael, we're reaching the top of the hour, so let me stop here and say thank you so much for sharing your time and insights with us today. I mean, it's really been a fascinating conversation from your journey through Berkeley, Yale, Pittsburgh, the quantum measurement, back action, parametric amplifying and control, modular architectures, and the role of academic labs in exploring new ideas. as well as training the next generation. I think we've covered a lot of topics today. At QBlocks, our motto is building quantum together. And today's conversation has really captured that spirit, mean, the idea of that project, progress in quantum really requires deep physics, careful engineering, creative architectures, and a community basically willing to explore a new path. As a small token of appreciation, we've sent one of our QBlocks mugs your way. Hopefully it becomes a good companion for the next round of HasLab experiments. For everyone watching, the conversation will also be available on Quantum Builders on Spotify and Apple Podcasts. And since this session marks roughly one year since we started this series, thank you as well for everyone who has joined us over the last year. We still have many conversations ahead, so please stay tuned for more upcoming Five Star Chats and Feather Guests. Michael, thank you again.
Michael Hatridge: Great. Thanks for inviting me and thanks everybody for listening.
Daniel Rodan Legrain: And thank you everyone for joining us today.
