Researchers from Forschungszentrum Jülich, RWTH Aachen University and the University of Cologne have carried out basic operations on a superconducting qubit in 6.8 nanoseconds, with measured accuracies of 99.981% and 99.989%. Jülich’s press release, published on 23 October, calls it “a new speed record for quantum gates”. The work appears in a peer-reviewed paper in Nature Communications published the same day.
There is an important limit. The method was tested on a single qubit, and the institution says so plainly: “so far the method has only been tested on a single, isolated qubit.”
Faster gates mean more work before a qubit loses its state
A quantum gate is the basic step of a quantum computer, the equivalent of flipping or nudging a bit. Qubits hold their fragile quantum states for only a short time, so the faster each gate runs, the more of them fit into that window. Jülich says this matters for complex algorithms and for error correction, which consumes many operations.
For scale, the release notes that 6.8 nanoseconds is roughly how long light takes to travel two metres in a vacuum. It says such operations “typically take several tens of nanoseconds in superconducting quantum computers”.
Speed has a cost. Superconducting qubits are switched with microwave pulses, and shorter pulses must be stronger. A transmon, the common type of superconducting qubit used here, has more energy levels than the two used for computing. Push it too hard and it can jump to a higher level, an error known as leakage.
The release compares this to a lift meant to run only between the ground and first floors. Raise the load too briskly and it overshoots. The team found that strong, fast pulses can push the qubit not just one floor too far but two or three, through what are called multiphoton transitions, “an effect that had previously been overlooked”, according to Jülich. These errors “can persist across multiple computational steps and impair quantum error correction,” the release says.
A reshaped pulse called R2D suppresses the unwanted jumps
The fix is in the shape of the pulse. The team’s method, R2D, builds on DRAG (Derivative Removal by Adiabatic Gate), a correction technique Jülich describes as standard practice for more than a decade. The paper describes R2D as “double recursive applications” of the DRAG pulse shape, which filters out the frequencies that would drive the qubit beyond its intended levels. It needs “only three parameters”, the authors write.
Compared with standard DRAG, Jülich says, R2D cuts the probability of unwanted multiphoton transitions “by a factor of more than 20”. The paper puts it as a “leakage reduction exceeding a factor of 20”.
The tests ran on one transmon qubit made at Jülich’s Helmholtz Nano Facility, as part of QSOLID, a German national quantum computing project. Two kinds of single-qubit gate, each 6.8 nanoseconds long, reached accuracies of 99.981% and 99.989%. The chance of the qubit leaking to a higher energy level stayed below 0.002% per operation. The paper also introduces a measurement sequence that it says can detect leakage down to one in a million.
The experimental group of Yuan Gao and Rami Barends worked with the theory group of José da Costa Jesus, Francisco Cárdenas-López and Felix Motzoi.
Other groups have reported quantum milestones in recent days
The Jülich paper arrives in a busy stretch for quantum hardware on TSN. NVIDIA used Jülich’s JUPITER supercomputer to simulate a cheaper error-correction method, Quantinuum reported 99.9995% logical fidelity on its Helios machine, and a USTC team pushed a self-correcting qubit past break-even in a single ion. Earlier, RIKEN researchers ran five silicon spin qubits at once.
The “record” in the paper is narrower than the headline
The release’s title says “speed record”; the paper is more careful. It claims “a record value for single-qubit gates in superconducting qubits” for how hard the qubit is driven relative to its energy structure, a technical ratio, not for raw speed against every other platform. TSN has not compared 6.8 nanoseconds with the fastest gates reported elsewhere, which is why the record claim here is Jülich’s.
That distinction does not shrink the result. The achievement is doing two things at once that usually trade off: very short gates and very low leakage. The paper’s own reading is that the remaining errors are “mainly limited by decoherence”, meaning the qubit’s natural loss of its state rather than the pulse. That suggests the control technique is no longer the main bottleneck on this device, though better qubits would be needed to show more.
One qubit is the start, not the processor
Useful machines need many qubits operating together. Jülich itself says further experiments are needed to learn whether the method works as well in larger processors and in two-qubit operations. A pulse tuned on one isolated qubit has not yet faced the conditions of a full chip, with many neighbours driven side by side.
Like the other recent results, this one advances a single component of a future machine. The test for R2D, which the authors say “can be straightforwardly adapted by other platforms”, is whether it holds up on a multi-qubit chip.
Sources
- Forschungszentrum Jülich: Fast but accurate – a new speed record for quantum gates (23 October 2026)
- Gao, Y. et al. Ultrafast single-qubit gates through multi-photon transition removal. Nature Communications 17, 10874 (2026)
Related stories
- NVIDIA Simulates 104 Qubits on JUPITER to Test a Cheaper Quantum Error Fix
- Quantinuum Reports 99.9995% Logical Fidelity on Its Helios Machine
- USTC Pushes a Self-Correcting Qubit Past Break-Even in One Ion
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
