In the integration of superconducting qubits, researchers have long grappled with a fundamental contradiction. To protect a quantum state from environmental noise and preserve it for as long as possible, interactions with the outside world must be thoroughly cut off. Yet to perform quantum error correction and logical operations, a qubit must be coupled strongly and quickly to neighboring qubits and readout circuitry.
Isolate a qubit from its surroundings, and information survives—but operations slow down. Strengthen the coupling, and operations speed up—but noise floods in and corrupts the data. This trade-off has been a physical barrier since the earliest days of superconducting circuits.
A research team at the Massachusetts Institute of Technology (MIT) has announced a new superconducting qubit architecture designed to resolve this dilemma. The paper was published in the peer-reviewed journal Physical Review Applied (DOI: 10.1103/3l3b-7jsm).
Rather than assigning every role to a single physical element, the team proposed an "arm qubit" that physically separates the element responsible for holding information from the element responsible for external coupling.
However, one crucial caveat must be stated up front: the striking performance figures reported in this study are predictions derived from numerical simulations that incorporate physical models and decoherence factors—no physical device has actually been fabricated using microfabrication techniques.
The paper's lead author is Jeremy B. Kline, a PhD student in MIT's Department of Electrical Engineering and Computer Science (EECS). Co-authors include Alec Yen, who earned his PhD in spring 2026, undergraduate student Stanley Chen, and Associate Professor Kevin P. O'Brien, a principal investigator at MIT's Research Laboratory of Electronics (RLE).
What the team presents is a detailed theoretical blueprint that seeks to overcome the limitations of existing qubit designs through a "qubit co-design" approach.
The physics of the "quarton coupler" that separates storage from communication
The backbone of the arm qubit lies in coupling two independent superconducting circuit modes with vastly different operating principles and frequency bands.
One is the "data mode," responsible for holding quantum data over long periods. For this, the team adopted a fluxonium-like design, which recent superconducting circuit research has shown to exhibit excellent coherence properties.
Fluxonium combines a Josephson junction with a large inductance and operates at a relatively low frequency band of 1–2 GHz. Its defining feature is large anharmonicity. This nonlinearity clearly separates the energy gap between the ground and first excited states from the gaps between higher levels, making it harder for external noise to induce unwanted transitions between levels. Assuming a dielectric quality factor of , the estimated echo phase-coherence time () of the data mode exceeds 380 (microseconds).
The other element is the "arm mode," which takes on all interactions with the outside world. This is built from a transmon-like circuit operating at a higher frequency band above 7 GHz.
The arm mode mediates coupling to other arm qubits and to the readout resonator used to measure computation results. To use a body analogy: the brain (data mode) holds memories, while an arm extending from it (arm mode) reaches out and exchanges information with the surroundings.
The challenge was how to connect these two modes. Naively linking different circuits risks letting high-frequency communication noise leak into the data-storage element, causing unwanted mixing between levels.
For this connection, the team employed a "quarton coupler," a device the same group had already experimentally demonstrated. Prior work on the quarton coupler was published in Nature Communications in 2025 (DOI: 10.1038/s41467-025-59152-z).
The quarton coupler, which arranges four Josephson junctions in a loop, can induce extremely strong nonlinear coupling by tuning the magnetic flux bias. It suppresses the unnecessary energy dissipation and mode hybridization typical of linear coupling, while still allowing strong interaction to be extracted only when needed for operations. This coupling strength—called near-ultrastrong nonlinear coupling—secures a fast information-transfer pathway via the arm mode while keeping the data mode highly isolated.
Previous research has explored superconducting circuits with multiple resonant modes within a single cell. However, most of that work targeted autonomous error correction using bosonic codes or erasure detection, which converts leakage errors into erasable errors.
By contrast, the arm qubit was specifically designed to satisfy the conflicting demands of shielding the data mode while enabling strong external coupling. Another design advantage is that connections between external arm qubits rely purely on capacitive coupling. Because there is no need to fill the space between qubits with complex tunable coupling elements, the design is inherently robust against parameter variation during chip fabrication.
Gate fidelity and readout speed derived from simulation
Kline and colleagues ran comprehensive numerical simulations incorporating realistic circuit loss models to quantify the theoretical limits of the arm qubit.
The resulting figures suggest performance that surpasses today's leading superconducting circuit experiments.
First, in simulations of the controlled-Z (CZ) gate—the fundamental two-qubit operation—the gate time was a mere 17 ns (nanoseconds), with an infidelity of . That corresponds to a fidelity of 99.9914%.
This calculation used a dielectric quality factor of , representative of standard current fabrication processes, as the baseline. A sensitivity analysis of loss factors was also performed: even under the pessimistic assumption of quality degrading to , the CZ error rate remained at . Conversely, applying —matching the cleanest fabrication processes reported in recent years—lowered the error rate to .
Single-qubit gate error rates were also extremely low. Thanks to the large anharmonicity of the fluxonium-like data mode, leakage into higher levels is suppressed, and the simulated single-qubit error rate fell below .
Improvements in measurement precision are also notable. Quantum computation requires a readout process at the end of a calculation to determine whether a qubit's state is "0" or "1." For the arm qubit, assuming a quantum efficiency of , the design achieved a state-assignment error rate of in just 27 ns. The quantum non-demolition (QND) error rate within the same timeframe came to .
A key theoretical strength of this design is that it achieves this fast, high-precision operation while thoroughly suppressing crosstalk during idle periods.
One problem that arises when integrating many qubits is always-on ZZ interaction, where idle qubits interfere with one another unintentionally. In the arm qubit simulation, parasitic ZZ interaction during idle periods was suppressed to below 0.4 kHz. Meanwhile, the cross-Kerr interaction between the data mode and the ancillary mode reaches several hundred MHz, allowing an extremely large switching ratio for necessary operations.
The design also showed strong resilience against the Purcell effect, in which photon leakage from readout circuitry causes data to decay. Because the arm mode acts as an intermediate buffer, the calculated Purcell-limited lifetime reached 167 ms (milliseconds) without requiring a dedicated Purcell filter on the chip. As for shot-noise dephasing caused by idle-period photon fluctuations, tuning knobs that appropriately detune the arm mode's frequency can extend this (expressed as ) to as long as 15.8 ms.
The gap between these figures and current experimental records
To properly understand the significance of these simulation results, they must be compared against the state-of-the-art experimental records actually measured with physical devices at research institutions worldwide.
Today, transmon qubits remain the mainstream choice for superconducting quantum computers. One of the best real-device benchmarks for a two-qubit gate using a tunable coupler between transmons was reported in September 2025 by a team centered at Finland's IQM Quantum Computers (arXiv:2508.16437). Over 40 hours of continuous operation, the team reported an average CZ gate fidelity of 99.93% (an error rate of roughly ).
Meanwhile, experiments with hybrid architectures combining fluxonium and transmon are also progressing. A study published in Physical Review X in 2023 (DOI: 10.1103/PhysRevX.13.031035) used a "fluxonium-transmon-fluxonium (FTF)" configuration, with a tunable transmon coupler placed between two fluxonium qubits, achieving a measured peak CZ fidelity of 99.85–99.9% (an error rate of roughly to ).
データを表で見る
| CZ gate error rate (Error rate (×10⁻⁴)) | |
|---|---|
| FTF experiment (2023) | 10 |
| Transmon real device (IQM 2025) | 7 |
| Arm qubit (low-Q simulation) | 1.5 |
| Arm qubit (baseline-Q simulation) | 0.86 |
| Arm qubit (high-Q simulation) | 0.58 |
As the chart shows, the error rate of 0.86 () that the arm qubit produced under baseline simulation conditions () is roughly an order of magnitude lower than the best current real-device measurements.
There is also a gap in readout speed. In typical dispersive readout, shortening the readout time causes the photon number inside the resonator to surge, which can ionize the qubit and destroy its state. As a result, real-device readout times are often no faster than about 50 ns, and when a safety margin is included, several hundred nanoseconds is not uncommon. The 27 ns readout time shown by the arm qubit is an attractive theoretical figure for shortening error-correction cycles.
The table below summarizes the arm qubit's simulated values alongside representative experimental demonstrations reported in recent years.
| Metric | Arm Qubit (this study, simulated) | State-of-the-art real-device measurements (2023–2026 experiments) | Experimental source |
|---|---|---|---|
| Two-qubit (CZ) gate error rate | (gate time 17 ns) | ~ (IQM transmon) | |
| ~ (FTF configuration) | arXiv:2508.16437 (2025) | ||
| Phys. Rev. X 13, 031035 (2023) | |||
| Single-qubit gate error rate | ~ to (standard transmon) | Various benchmark measurements | |
| Readout time and error rate | 27 ns (assignment error rate ) | ~50–100 ns (assignment error rate on the order of $10^{-3}$) | Conventional dispersive readout measurements |
| Physical device fabrication/measurement | Not yet fabricated (numerical analysis only) | Demonstrated continuous operation on real chips | Various real-device papers |
As the table shows, the arm qubit's figures remain confined to simulation. While the quarton coupler—the enabling technology behind the ultrastrong coupling—has itself been experimentally verified as a physical device since 2025, no physical device combining these components has been built in this study; the MIT team is now aiming toward that fabrication step.
What remains to be verified in moving from theoretical model to real chip
However precise a simulation may be, a massive hurdle awaits the transition to physical implementation.
The greatest uncertainty concerns the validity of the material parameters the simulation assumes. The coherence-time and gate-fidelity estimates in this study depend heavily on the dielectric quality factor $Q$. The team adopted as its baseline, but in complex circuits featuring multilayer wiring and closely spaced Josephson junctions, there is always a risk that interfacial oxide layers or substrate surface defects introduce unexpected dielectric losses. It cannot be ruled out that minute geometric deviations arising during actual chip fabrication could invite unmodeled parasitic resonances or two-level-system (TLS) defects, degrading coherence beyond what the model predicts.
The design freedom of the circuit parameters also needs experimental verification. The simulation sets a specific charge-matrix-element ratio of 8.1. Whether such parameters can be reliably reproduced using actual nanofabrication techniques such as thin-film deposition and electron-beam lithography remains to be tested in future experimental stages.
Professor O'Brien himself offered a candid reflection in a press release:
"Because the simulation results are so promising, there's a real sense of suspense to this work. What's needed next is to check whether we can actually fabricate it, and to see whether we missed anything in the modeling or design. If we can build this qubit, it could become a building block for future error-corrected quantum computers."
This research was funded by the U.S. Army Research Office (ARO), the Air Force Office of Scientific Research (AFOSR), the Doc Bedard Fellowship at MIT's Center for Quantum Engineering, and the Laboratory for Physical Sciences (LPS).
Another limitation that must not be overlooked is that the scope of the simulations in this paper is confined to a coupled circuit of at most two arm qubits and their associated readout resonator.
System-level questions—such as how crosstalk and heat influx from wiring routing would grow when dozens or hundreds of arm qubits are arranged in a two-dimensional grid, and whether the system as a whole could clear the error-correction threshold when running large-scale quantum error-correcting codes such as the surface code—remain entirely unaddressed and left for future research.
Can the team clear the hurdle of physical fabrication and actually reproduce, inside a dilution refrigerator at millikelvin temperatures, the ultrastrong coupling and high coherence the simulation depicts? The true value of the theoretical blueprint MIT's team has laid out will be tested by the waveform data that emerges from the first real chip.