This week on the 632nm podcast, we speak with physicist John Martinis, recipient of the 2025 Nobel Prize in Physics. Martinis discusses the physics and engineering behind transmon qubits, why controlling interactions between qubits became such a crucial problem, and why building a useful quantum computer ultimately requires much more than making individual qubits better.
He also takes us inside some of the less visible challenges of scaling quantum hardware: fabrication yield, wiring, cryogenic power, materials defects, and the enormous complexity of integrating thousands or millions of components into a single machine.
Martinis reflects on his work at UC Santa Barbara and Google, the origins of the experiment that earned him the Nobel Prize, what went wrong in some of his earlier projects, and why he is now trying to rethink how superconducting quantum computers are fabricated through his company Qolab.
The 632nm podcast features in-depth conversations with the world’s leading scientists and engineers.
I. The Hard Part of a Quantum Gate Is Turning It Off
Quantum computers are often described in terms of how accurately they can perform operations on qubits. But for superconducting quantum computers, Martinis argues that one of the most important breakthroughs was realizing that controlling an interaction is just as important as creating one.
A two-qubit gate can be relatively straightforward to construct. The harder problem is preventing qubits from interacting when they are supposed to be independent.
As Martinis puts it:
“What’s hard is to turn off the gate.”
That insight shaped much of his work at UC Santa Barbara. Early superconducting qubits could be coupled together, but residual interactions between neighboring qubits created errors. Martinis and his collaborators worked on tunable qubits and couplers that could turn interactions on and off much more effectively.
This also illustrates a broader theme of the conversation: the best individual component is not necessarily the best component for a large system. A qubit with extremely long coherence time is useful, but if it cannot be coupled efficiently to other qubits, that performance may not translate into a better computer.
Martinis describes this as a problem of system engineering: the architecture has to balance coherence, gate speed, connectivity, control, and many other constraints simultaneously.
II. Why Faster Qubits Can Matter More Than Longer Coherence
A transmon is, at its core, a remarkably simple device: a Josephson junction combined with a capacitor, forming a nonlinear LC circuit.
The nonlinearity is what allows the circuit to behave as a qubit rather than an ordinary harmonic oscillator. The ground state becomes the qubit’s zero state, while the first excited state becomes its one state.
But the engineering challenge is not simply preserving the quantum state for as long as possible. It is performing useful operations before that state decoheres.
Martinis emphasizes this distinction throughout the conversation. Superconducting qubits can have coherence times on the order of hundreds of microseconds, while gates can take only tens of nanoseconds. That ratio means a relatively modest coherence time can still support a large number of operations.
The tradeoff becomes particularly clear when comparing superconducting qubits with other platforms. Neutral atoms, for example, can offer very clean ways of controlling interactions by physically moving atoms, but moving them can be much slower.
For Martinis, the relevant question is therefore not simply “How long does the qubit remain coherent?” It is “How much useful computation can the system perform during that time?”
III. The Million-Qubit Problem Is a Fabrication Problem
Today’s superconducting quantum processors already contain hundreds of qubits. But going from hundreds to tens of thousands or millions introduces a fundamentally different set of problems.
One of the biggest is fabrication.
Current superconducting qubits rely on fabrication techniques that work extremely well for relatively small experimental devices. Martinis argues that those processes may not provide the reliability and uniformity required for machines containing millions of Josephson junctions.
His goal is therefore not simply to make a slightly better version of today’s chips, but to rethink the fabrication process using technologies developed by the semiconductor industry.
As he explains:
“I want to get 100% yield on 10 million, 100 million.”
That is an extraordinary manufacturing requirement. A quantum computer with millions of qubits cannot depend on individually tuning or discarding large numbers of defective components.
Martinis’s approach is to borrow from the manufacturing infrastructure that already produces conventional semiconductor devices containing billions of transistors. The challenge is adapting those processes to superconducting quantum circuits while preserving the properties that make Josephson junctions useful as qubits.
The problem is therefore as much about manufacturing engineering as quantum physics.
IV. From a Spaghetti of Wires to a Quantum Computer
Another major obstacle to scaling superconducting quantum computers is control.
Today’s systems require enormous numbers of microwave and DC connections running into cryogenic refrigerators. As the number of qubits increases, simply adding another wire for every new component quickly becomes impractical.
Martinis helped build much of this infrastructure during the early development of superconducting quantum computing. But after decades of watching the systems grow more complicated, he now sees the wiring itself as a warning sign.
Reflecting on photographs of early computers filled with cables, he says:
“I hate that... I kinda look at this in horror. It’s like, okay, come on, this is not gonna scale.”
His proposed solution is to move toward wafer-scale integration, bringing much more of the control infrastructure directly onto the chip rather than relying on increasingly large bundles of external cables.
This is another example of the difference between demonstrating a quantum effect and engineering a useful quantum computer. A laboratory prototype can tolerate enormous amounts of manual infrastructure. A machine intended to operate with thousands or millions of qubits cannot.
The same principle applies to cryogenic power, control electronics, packaging, and readout. Every subsystem that works at the scale of hundreds of qubits has to be reconsidered when the target is millions.
V. Why Quantum Computing Is an Orchard, Not a Single Low-Hanging Fruit
One of the most interesting parts of the conversation is Martinis’s perspective on technological progress.
It is tempting to look for the one bottleneck that, once solved, will suddenly make quantum computers dramatically better. But Martinis argues that this is the wrong mental model.
“The real problem with a quantum computer is it’s a system engineering problem. There are 10 or 100 things you have to worry about.”
Improving one component can create problems elsewhere. A design that improves coherence may make coupling more difficult. Faster control can increase wiring and power requirements. More connectivity can complicate fabrication. Better individual qubits do not necessarily produce a better large-scale machine.
Martinis describes his alternative philosophy with a memorable metaphor:
“I really want to plant the orchard.”
Rather than searching for one isolated improvement, the goal is to build an entire ecosystem of technologies that can work together: materials, fabrication, cryogenics, control, architecture, error correction, and manufacturing.
This way of thinking also explains his emphasis on measuring the entire system rather than highlighting a single exceptional device. In a complex machine, the average—and especially the worst-performing components—can matter more than the best qubit on the chip.
VI. The Experiment That Put Superconducting Circuits on a Firm Foundation
Near the end of the conversation, Martinis goes back to the work he did as a graduate student that helped establish macroscopic quantum behavior in superconducting circuits.
The experiment required building much of the infrastructure from scratch. The team needed custom coaxial cables, filters, low-noise electronics, a dilution refrigerator, and microwave components capable of operating at extremely low temperatures.
Their first attempts did not work.
Martinis describes an early test in which the data simply looked wrong. Rather than continuing blindly, the team treated the failure as information and began systematically eliminating sources of noise.
“We failed fast, right? That’s what you’re supposed to do, fail fast.”
They developed copper-powder filters, improved the microwave environment, and eventually obtained a remarkably clear signal. In one experiment, Martinis remembers looking at an oscilloscope and seeing three distinct points corresponding to the quantum energy levels they were trying to observe.
That moment was important not only because it demonstrated the phenomenon. The experiment was designed to be unusually careful and quantitative, establishing a foundation that other researchers could build upon.
As Martinis explains, the goal was not merely to obtain a striking result. It was to understand the system well enough that the technology could be reproduced and improved.
That philosophy has remained remarkably consistent throughout his career: build the experiment carefully, understand what is limiting it, and then figure out how to make the entire system work.
VII. Doing Something New
The conversation ultimately returns to the question of why Martinis continues to work on difficult experimental problems.
His answer is less about a particular technology than about the excitement of entering territory where the solution is not already known.
Looking back on his career, he describes the appeal of working on problems where there is little established history and where researchers have to build the tools themselves.
For young scientists, his advice is straightforward:
“Always keep the lookout for what’s new and exciting out there.”
And he emphasizes that genuinely foundational experiments do not come around very often. When they do, they require researchers willing to take risks, build unfamiliar technologies, and pursue questions that may not have obvious answers.
For Martinis, that is still what makes experimental physics compelling.
“It’s amazing that they pay you to learn and think.”
After decades spent building quantum hardware from the ground up, he is still pursuing the same basic impulse: understand something that does not yet make sense, build whatever is necessary to test it, and see where the experiment leads.
Full episode available now on YouTube and all major podcast platforms.
Listen to our full conversation with John Martinis on your favorite podcast platform:
Or search for “632nm” on your preferred podcast player
If you know someone who would enjoy this, please share it with them!