This article is based on a conversation between Steve Girvin and Daniel Rodán Legrain as part of the Quantum Builders series, sponsored by Qblox. Watch the full webinar for more on the origins of circuit QED, the science behind the 2025 Nobel Prize, and what fault tolerance actually looks like in practice.
The theory that defined how superconducting quantum computers work started with a 45-minute conversation at a blackboard.
Steve Girvin was at Yale to give a colloquium on the fractional quantum Hall effect. Rob Schoelkopf was on his schedule. By the end of the meeting, the two of them had sketched out the idea that became circuit QED, the theoretical framework underneath nearly every superconducting qubit platform in use today.
Girvin is the Sterling Professor of Physics and Applied Physics at Yale, a member of the US National Academy of Sciences and the Royal Swedish Academy of Sciences, and a co-founder of circuit QED with Schoelkopf and Michel Devoret. The conversation covered how he ended up in the field, what the 2025 Nobel Prize recognized in Devoret's early work, and where quantum error correction stands right now.
Girvin came to circuit QED from somewhere else entirely. He was a condensed-matter theorist, John Hopfield's last student in the field, working on the fractional quantum Hall effect and mesoscopic superconductors. He didn't know quantum computers were being seriously pursued until 1999.
That year, Yasunobu Nakamura's group at NEC published the first time-domain experiment with a Cooper-pair box, demonstrating Rabi oscillations of a superconducting qubit.
"I happened to see that paper when it was published in Nature, and I was completely just stunned, and found out about this whole field that people were trying to build," Girvin said.
A year later, he met Rob Schoelkopf. In 2001, he moved to Yale. Devoret arrived six months after that.
The blackboard conversation that started circuit QED happened during Girvin's visit. Schoelkopf had invented the radio-frequency single-electron transistor, which could measure tiny fractions of an electron charge in microseconds. He was using it to measure the charge state of a Cooper-pair box and wanted to place the box inside a microwave resonator.
"Without completely understanding what I was saying, I said, well, why don't you do it like in cavity QED, where you measure the state of the photons coming out, rather than the charge state of the qubit," Girvin said.
That suggestion, refined over a couple of years of work, became the foundation of the architecture.
Girvin credits his amateur experimental background for making the framework possible.
"Even though I'm a theorist, that sort of amateur experimental background turned out to be extremely useful for understanding the quantum mechanics of electrical circuits," he said.
He had been a teenage ham radio operator running antennas between trees and matching impedances by hand. The intuition translated.
Michel Devoret shared the 2025 Nobel Prize in Physics with John Clarke and John Martinis for work done at Berkeley in the mid-1980s, when Devoret was a postdoc and Martinis a graduate student in Clarke's lab.
The work proved two things about a Josephson junction circuit. First, that the superconducting phase variable could tunnel through a potential barrier as a quantum mechanical particle. Second, the same variable had quantized energy levels that could be addressed with microwave spectroscopy.
Girvin walked through the framing.
"Tony Leggett had the idea that actually that classical particle could itself behave quantum mechanically, if we built a circuit where there was not too much dissipation," Girvin said.
The Clarke lab built that circuit and arranged the potential so the phase variable sat trapped in a small well, occasionally tunneling out to roll downhill. Tunneling showed up experimentally as a voltage appearing at a random moment, like radioactive decay.
The experimental case was hard to make. Temperature and noise could mimic the same signal.
"They had to do heroic filtering and cooling, and making sure the electrons were as cold as the actual refrigerator temperature, many heroic steps and many serious control experiments to make sure that what they were seeing was actually quantum mechanical tunneling of this phase particle, and not temperature or some noise from room temperature disturbing it and making it move," Girvin said.
The second experiment used microwave spectroscopy to confirm the energy levels were quantized. Quantized energy levels meant the circuit was acting as a synthetic atom. The qubit, though no one called it that yet.
"This was really the historic beginning of what laid the groundwork for all the subsequent developments in superconducting quantum circuits," Girvin said.
Girvin describes Devoret as a strong theorist who once scooped him on a Physical Review Letters paper by several months. The collaboration between them produced repeated cycles of theoretical disagreement that resolved just before experiments confirmed the answer.
The example he returned to was the back-action of a weak quantum measurement, work done by Devoret's postdoc Michael Hatridge.
"Michelle and I had a long series of arguments about whether what did that mean, if you were trying to measure these two incompatible things," Girvin said. "Was it going to produce decoherence, or was it going to just produce something else, some other kind of back action? We each managed to change the other person's mind at least twice. We went back and forth, back and forth, and we finally got the intuition right just before the experiment started working."
The structural choice that made those exchanges possible was physical. Schoelkopf and Devoret's labs had no walls between them.
"It was pretty unusual to have such tight coupling between two experimentalists and with theorists who joined the effort," Girvin said.
The group meeting started with four people in a hallway alcove. It now fills a dedicated room with around 50 attendees.
Girvin's group at Yale, along with Schoelkopf's, has been central to the development of bosonic codes, including cat codes and the Gottesman-Kitaev-Preskill (GKP) code. These encode quantum information in superpositions of microwave photon states inside a single resonator rather than entangling many physical qubits.
The advantage is the size of the Hilbert space.
"Oscillators have a bigger Hilbert space than a qubit. You can build in some redundancy and get an error correcting code just inside a single resonator, without having to take many physical qubits and entangle them in some clever way to produce the logical qubit," Girvin said.
The limitation is the gain ceiling. A bosonic code can extend the logical qubit's lifetime beyond the best physical component it's made of, but the multiplier won't reach the millions or billions that fault tolerance ultimately demands.
"In the end, you're not going to get error correction gains of millions or billions. You're going to get a few, maybe 10 someday," Girvin said.
The practical path is hybrid. Bosonic codes serve as better physical qubits, which then get assembled into a surface code or color code at a higher level.
Girvin compares the state of the art across two platforms.
Devoret's recent GKP bosonic error correction experiment achieved an error correction gain a little under 2.5. The recent Google surface code experiments achieved approximately the same gain. The Google result used hundreds of physical qubits. The GKP result used one resonator. The actual wall-clock lifetime of the GKP logical qubit is roughly twice that of the Google logical qubit.
"They're kind of comparable in results, but very different in hardware cost," Girvin said.
Different platforms have different failure modes. Cold atoms have coherence times near a second, which means error correction for memory matters less for them than it does for superconducting qubits. The trade-off is the same physics that gives atoms long coherence times makes their gates slower and lower fidelity, so they need error correction for gates instead.
"We're both struggling with error correction, but it's a slightly different thing," Girvin said.
Girvin pushes back on the framing that fault tolerance arrives as a moment. He cites Schoelkopf on this.
"It's not like one day suddenly you're fault-tolerant. It's a journey, and you're just making things more and more tolerant of more and more kinds of faults. It's a gradual journey, not a destination," Girvin said.
That changes how progress should be measured. Right now, error correction is barely improving things at all. The community is starting to learn what fault tolerance demands at each layer of the stack.
The work has to happen at multiple levels simultaneously. Better physical qubits. Better error correcting codes. Better high-speed control electronics for measuring and correcting errors. Better algorithms co-designed for the error modes that remain after correction.
Girvin draws an analogy to the history of CMOS, where hundreds of billions of dollars were eventually invested in industrial scaling, but academic research continued for decades alongside it. The same dynamic applies to quantum.
"If you're IBM or Google, a large company that's trying to focus on scaling up, you have to pick your hardware architecture and stick with it," Girvin said. "When you're investing a billion dollars in scaling up a particular architecture, you don't want somebody distracting you with, hey, I have a better way to do it that might work, but I'm not sure."
That's what academic labs are positioned to do. The high-dielectric-constant materials that eventually entered CMOS manufacturing came out of universities. Girvin expects the same pattern in quantum.
He founded the DOE's Co-design Center for Quantum Advantage at Brookhaven partly to enable that kind of cross-pollination. The center brought him into ongoing collaborations with computer scientists like Nathan Wiebe at Toronto, and connected Schoelkopf and Devoret's groups with materials scientists like Andrew Houck and Nathalie de Leon at Princeton.
"We had never thought about materials science before. We didn't need to, because microwave hygiene and other simple things, filtering, were limiting us for 15 years," Girvin said. "But now that those problems have been solved, we have to start thinking about materials science."
Girvin sat on an NSF panel examining how to design an undergraduate quantum engineering program. The premise behind the panel was direct: assembling quantum computers should not require six years of graduate training in quantum physics.
The teaching change he describes is concrete. In his own undergraduate quantum computing course, he skips the Schrödinger equation entirely. The course treats quantum mechanics as linear algebra and proceeds from there.
"We need computer scientists, mechanical engineers, chemists, physicists at the bachelor and master and PhD level to build out a robust workforce," Girvin said.
The harder requirement is systems thinking. Physics PhDs are usually not trained as systems engineers, and quantum computing now demands abstraction across layers in the way classical computing always has.
"You're the expert at maybe the hardware level, but you have to simplify the story as you pass it up to the transpiler designer or the compiler designer," Girvin said. "You have to simplify it, but it still has to convey the key information. What is the noise model, the error model, how long do gates take, the sort of resource requirements? You can't overwhelm them with all the dirty details of the physics that fascinate you."
For graduate students, Girvin's advice is about developing perspective. There are only so many hours in a day, and the value of a PhD comes partly from learning to pick problems that are both important and possibly solvable.
"There are some things that are important that we have no idea how to solve, and there are some things that have a chance of being solved in the near term. Learning how to do that is, I think, very important," Girvin said.
For entrepreneurs, his concern is sharper. He's watching the gap between hype and technical understanding widen.
"It's amazing to me how much entrepreneurs often know very little, actually, about how the technology works," Girvin said. "There's a very large amount of hype in our field now which I'm worried about. That, connected with a large amount of money flowing, is potentially bad for the field. I think it's incumbent on everybody to at least learn something about the basic physics of what's possible with this kind of technology, because it's not going to solve all possible computational problems."
When asked for resources, he points to a few responsible CEO-level primers from large consulting firms (Deloitte gets a nod), the Les Houches lecture notes on circuit QED, and Will Oliver's review article taking a quantum engineering approach to superconducting qubits.
Three things stand out. The rate of progress in the field has exceeded what most people working in it twenty years ago expected. Coherence times have improved by roughly a factor of a million since Nakamura's first experiment. Whether the trend continues for another several orders of magnitude is genuinely uncertain.
Multiple platforms will continue to advance in parallel, and the boundary between them is becoming more porous. Superconducting qubits, trapped ions, and Rydberg arrays each face different versions of the same problems. Each platform can borrow approaches from the others.
The fault-tolerance threshold won't arrive as a single moment. Error correction is now achieving small gains in carefully controlled experiments. The work over the next decade is improving every layer of the stack at once: physical error rates, codes, control electronics, software, and algorithms co-designed for the residual error modes.
Circuit QED and the experiments that followed depend on precise microwave control across hundreds or thousands of channels. As error correction starts to deliver real gains, the control stack becomes one of the constraints that decides how fast the next decade of work happens. Qblox builds control hardware for groups working at this layer of the problem, across superconducting, spin, and other qubit platforms.
To talk about the control side of your superconducting qubit stack, get in touch with the Qblox team.
