The Signal
Jülich Team Shapes Transmon Pulses for 6.8-Nanosecond Gates
Researchers at Forschungszentrum Jülich report single-qubit gates with 6.8-nanosecond pulses and fidelities above 99.98%. Their pulse-shaping method reduces leakage from unwanted transitions in a superconducting transmon.
Quantum.world Team · Written with AI · fact-checked ·
“We demonstrate single qubit gates with a total leakage error consistently below 2.0 × 10−5 on a transmon qubit, and obtain fidelities above 99.98% at 6.8 ns pulse duration in both X and X/2 gates.”
Yuan Gao and co-authors, authors of the study
A team led by researchers at Forschungszentrum Jülich reports that pulse shaping can suppress unwanted transitions in a superconducting transmon while driving single-qubit gates in 6.8 nanoseconds. The results, published in Nature Communications on October 23, 2026, pair those short pulses with measured fidelities above 99.98% for both X and X/2 gates. The experiment addresses a trade-off in transmon control: stronger, faster microwave drives can push population out of the two states used to encode a qubit.
Strong drives expose transitions beyond the usual correction
A transmon is not a perfect two-level system. Its higher energy levels can be populated during a gate, creating leakage that may persist and disrupt later operations. Standard Derivative Removal by Adiabatic Gate, or DRAG, pulse shaping addresses unwanted transitions between neighboring levels, but the authors say that higher-order transitions become important when the drive approaches the qubit’s anharmonicity.
The study describes these effects in a dressed-state picture, in which the applied microwave drive changes the qubit’s energy states and opens pathways involving more than one photon. The team’s target is therefore not only the familiar transition from the first excited state to the second, but also transitions to higher levels. Those pathways become a growing source of error as the pulse gets shorter and stronger.
R2D adds recursive corrections to DRAG
The researchers call their approach R2D, for a double recursive application of DRAG. Starting from a chosen pulse envelope, the method modifies the squared real component of the drive to suppress frequency components associated with two-photon transitions; a further DRAG correction addresses the usual single-photon transition. Three adjustable parameters let the team tune the spectral notches during calibration.
To measure small leakage rates, the team also introduces an amplified leakage error sequence, or ALE. It repeats gates with phase adjustments chosen to make coherent leakage accumulate rather than cancel, allowing the researchers to quantify leakage below 10−6. The paper reports that R2D reduces leakage by more than a factor of 20 compared with standard single-photon DRAG in its comparison.
At 6.8 nanoseconds, leakage stays below 2 × 10−5
The group benchmarked X and X/2 gates over pulse durations from 6.6 to 9 nanoseconds. At 6.8 nanoseconds, the measured total gate errors were (1.9 ± 0.1) × 10−4 for X and (1.1 ± 0.1) × 10−4 for X/2; the corresponding fidelities exceed 99.98%. Total leakage stayed below 2 × 10−5 down to that pulse duration.
The performance does not extend unchanged to every pulse length tested. At 6.6 nanoseconds, leakage to the third excited state begins to dominate, while the team reports that optimization becomes unstable below 6.6 nanoseconds. At the other end, the control electronics’ waveform memory limits the benchmarking sequences above 9 nanoseconds.
Decoherence, not leakage, limits the measured gates
The authors’ error budget points to decoherence as the main limit for both gate types, at roughly 1.04 × 10−4. They attribute the remaining leakage floor mainly to voltage noise in room-temperature microwave electronics, and link rising coherent errors in shorter X gates to microwave distortion at higher amplitudes. For X/2 gates, the paper reports coherent error below 1.0 × 10−4.
That distinction matters: R2D addresses leakage from the driven qubit’s higher levels, but it does not remove the decoherence and control-chain errors that remain in the measured gate budget. The results come from a transmon experiment, not a demonstration across a multi-qubit processor. At the reported 6.8-nanosecond operating point, X and X/2 gates retain leakage below 2 × 10−5 while their total errors remain different by 8 × 10−5.