Quantum Builders
Engineering Better Qubits: Materials, Noise, and Quantum Hardware at MIT Lincoln Laboratory
A fireside chat with:
Kyle Serniak
This article is based on a conversation between Kyle Serniak and Daniel Rodán Legrain as part of the Quantum Builders series, sponsored by Qblox. Watch the full webinar for more on what limits superconducting qubit performance and how shared fabrication infrastructure is accelerating the field.
If you want to know what's actually holding superconducting qubits back, the answer is rarely the part that gets the headlines.
The community talks about T1 and T2 times. The actual constraints sit underneath those numbers: unwanted excitations called quasiparticles, two-level system defects in the materials, and the wiring and shielding choices that affect everything else.
Kyle Serniak works on all of it. He's the assistant group leader of the Quantum-Enabled Computation Group at MIT Lincoln Laboratory, a research affiliate at MIT RLE and EQuS, and part of the MIT Center for Quantum Engineering. He did his PhD on non-equilibrium quasiparticles with Michel Devoret at Yale, work that's still central to qubit performance today.
The conversation walks through what's limiting the platform, what's improving, and how the SQUILL Foundry is changing who can build on superconducting hardware.
Quasiparticles are unwanted electronic excitations in the superconducting films that make up qubits. They tunnel across Josephson junctions and exchange energy with the qubit, which causes errors.
Serniak frames the problem at the foundation of the platform.
"The only reason that we can ever really resolve any of that quantum phenomena is if we're in the limit where we can just completely forget about all of the electronic degrees of freedom in these circuits," Serniak said.
That's the trade-off built into the hardware. Superconducting qubits exist because engineers can build them out of standard circuit components in a low-temperature regime. The cost is that those circuits contain a huge number of electrons, and even one stray quasiparticle in the wrong place can degrade performance.
"It's very foundational in the way that we choose to encode information in superconducting circuits that there's going to be some limitation at some level from quasiparticles," Serniak said.
Cooling helps. So does shielding. Neither makes quasiparticles disappear entirely.
A more recent finding is that quasiparticles aren't only generated thermally. Ionizing radiation, and even mechanical vibrations from the cryostat, can create them.
That matters for error correction.
"Those in particular are a little bit difficult to deal with because they typically result in errors that are correlated in both space on a processor and in time," Serniak said. "These spatial-temporally correlated errors pose some challenges for error correction specifically in decoding, where a lot of the assumptions that go into those protocols assume just completely random errors."
The parallel he draws is to classical electronics, where radiation-hardened CMOS exists because the same problem was solved decades ago in a different domain. Superconducting qubits are working through a similar engineering process.
Serniak's PhD work used a technique called charge parity measurements to identify which qubit errors were specifically caused by quasiparticles versus other mechanisms. The technique itself was developed by Leo DiCarlo's group a few years earlier, but Serniak's team used it to find something the field hadn't accounted for.
The measurements showed that quasiparticles were causing qubit excitation as well as relaxation. That was inconsistent with the picture of well-thermalized quasiparticles sitting near the superconducting gap edge.
The explanation turned out to be photon-assisted tunneling. High-energy photons coupling into the qubit at the Josephson junction were generating quasiparticles directly, producing errors that didn't correlate with the thermal background.
"It caused folks to really reevaluate the way that they're doing filtering and shielding in their setups," Serniak said. "Certainly, the average qubit is a little bit colder today than it was 10 years ago."
The deeper effect was methodological. The work demonstrated that the right experimental signature could disentangle one error mechanism from another, rather than treating coherence loss as a single number.
Quasiparticles aren't the only foundational constraint. The other one most of the field agrees on is dielectric loss from two-level system defects in the materials.
"It's not contentious to say that most of the field really views two-level system defects, dielectric loss from two-level system defects, as one of the primary challenges facing superconducting qubits," Serniak said.
The progress comes from the same kind of targeted measurement that worked for quasiparticles. Instead of treating loss as a single metric, the field is developing experiments designed to probe specific defect mechanisms.
Serniak points to the intersection of three variables that experimentalists can control: device design, materials, and metrology.
"It's that whole picture holistically that really folks are grabbing on to and recognizing there's a way to solve these problems," Serniak said. "It's the recognition that there is this intersectionality, I think, that people are really, that's going to really accelerate the work in the space over the next couple of years."
He's careful not to single out a primary bottleneck for scaling. Materials, information encoding, control, filtering, and shielding all push and pull on each other.
"I'm a believer that the quantum computer that we have one day really may not look so similar to what we're using right now," Serniak said. "It really depends on how you're going to try to encode that information to really understand the most critical pieces."
The SQUILL Foundry is a service operated out of Serniak's group at Lincoln Lab. It's funded by the Laboratory for Physical Sciences through their Qubit Collaboratory, and it's one of three foundry programs LPS set up roughly four years ago. The other two, at HRL and Intel, focus on quantum dot qubits. SQUILL handles superconducting qubits.
The model lets external research groups submit device designs and receive fabricated chips, removing the need to build a full fabrication facility in-house.
"You submit a project application, and as long as your project fits within some criteria, you get a design rule checker software package to make sure that the designs you're generating for your experiment are going to be able to be fabricated in our process," Serniak said.
Designs come in as GDS files, get aggregated onto a multi-user mask, fabricated by the Lincoln team, and shipped back to the submitter.
The program has grown to roughly 50 active users. Tens of published papers have come out of devices fabricated through it. Turnaround from design submission to shipped device runs roughly 8 to 10 weeks.
Serniak describes three usage patterns.
The first is new faculty starting labs. Building out fabrication infrastructure from scratch takes six to twelve months minimum. SQUILL lets a new group run experiments while their own facility comes online.
The second is established groups that need a specific advanced capability they haven't developed in-house. Examples include airbridge crossovers and flip-chip integration.
The third is risk mitigation. Even groups with mature fabrication processes use SQUILL as a parallel pathway when their own equipment is being serviced or replaced.
"Tools go down, right. So having a parallel pathway to have devices fabricated is generally useful for any group," Serniak said.
The model also reflects a structural advantage that Federally Funded Research and Development Centers like Lincoln Lab can offer.
"Lincoln Laboratory is a much larger institution than just superconducting qubit research," Serniak said. "But central to that mission is supporting the community and supporting the sponsor community in all the ways that make sense."
Serniak's role spans Lincoln Lab and MIT campus. The combination is deliberate, not incidental.
Lincoln Lab brings staff scientists with deep institutional knowledge. MIT campus brings students and postdocs with new ideas. The collaboration runs both directions.
"On the Lincoln Lab side, you have staff scientists, generally speaking, with a lot of institutional knowledge, people who love what they're doing and have been doing it for many years," Serniak said. "From the campus side, it's an academic group. So you've got students and postdocs with fresh ideas and just boundless enthusiasm coming through. Bringing those folks together can make for some really exciting research and development."
The day-to-day at Lincoln Lab is more structured than academic work, closer to industry. Serniak views the methodical pace as a feature, not a constraint.
The field used to require everyone working on superconducting qubits to know a little cryogenics, a little microwave engineering, a little fabrication, a little of everything. Serniak sees that changing.
"It's natural maturity of a field that the knowledge that was maybe 15 years ago condensed into everyone's head, a little bit of cryogenics, a little bit of microwave engineering, a little bit of fabrication engineering, it's very reasonable for it to be more distributed," Serniak said. "That allows for much more depth of understanding of all those individual components."
The implication is that people coming into the field from adjacent disciplines have real value to add. Process engineers from industrial foundries. Professional microwave engineers. Cryogenics specialists. Algorithm developers. The platform now demands enough specialization that no one needs to be a generalist to contribute.
His advice for students choosing between academia, national labs, and startups is grounded in self-knowledge rather than career strategy. Identify what you genuinely like and don't like about the day-to-day, then optimize for environments that maximize the first and minimize the second.
For PhD students mid-program, he names the transition that catches many people off guard: getting comfortable with research timelines that aren't measured in semesters.
"It can be the first time that you are in a situation where you're doing work and there's no longer the six-month or four-month or a semester timeline associated with a project," Serniak said. "That's something that people often get really bogged down with."
Three things stand out.
The biggest gains in superconducting qubit performance will come from materials. Better materials improve every other axis of the system at once, regardless of the cubit encoding or control approach layered on top.
The methodology for finding and fixing error mechanisms is maturing. Charge parity measurements worked for quasiparticles. Similar targeted experiments are now being designed for two-level system defects. The field is moving from coherence as a single number toward coherence as a decomposable signal.
Shared fabrication is changing who can participate. The SQUILL Foundry serves as a model for how academic groups, industry partners, and national labs can share the cost of producing high-quality quantum devices. That's a structural shift in how the field operates.
The work Serniak describes runs on shared infrastructure. Foundries, control electronics, characterization stacks, the equipment that academic and industrial teams rely on every day. Qblox builds control hardware for groups working at the leading edge of superconducting, spin, and other qubit platforms, with a focus on the kind of measurement chain Serniak singles out as critical to disentangling error mechanisms.
If you're working on superconducting qubits and want to talk about the control side of your stack, get in touch with the Qblox team.
