A circuit proposal, not a hardware result
On September 4, 2026, researchers from Seoul National University and Samsung Advanced Institute of Technology submitted the first version of a preprint proposing a new coupler for superconducting qubits. The design exists as a circuit model, numerical layout and simulated control protocol. The paper does not report a fabricated chip, a cryogenic measurement or an experimentally observed gate. Its contribution is therefore architectural: it identifies a circuit that simulations say could separate two jobs that become harder to reconcile as processors grow, keeping neighboring qubits quiet while idle and coupling them strongly during a two-qubit operation.
The proposed double-resonator coupler places two resonators between a pair of transmon qubits. A transmon is a superconducting circuit whose energy levels encode quantum information. The resonators are connected by one Josephson junction and a capacitor. Instead of relying on a single intermediary, the design creates two hybridized modes, called common and differential modes. Each mode provides a virtual route through which one qubit can influence the other. Their coupling signs differ, so the two routes can interfere destructively. In everyday terms, one unwanted interaction can be arranged to counteract the other.
Why idle interactions matter
That cancellation targets two related forms of idle crosstalk. Transverse coupling can exchange excitations and mix the qubits’ states. Static ZZ coupling produces a state-dependent frequency shift, causing unwanted conditional phase to accumulate even when no entangling gate was requested. A useful tunable coupler needs both effects to become small at idle. It must then produce a strong, controlled interaction on demand. This is not merely an on-off switch. The circuit’s energy levels, fabrication tolerances and control pulse all influence whether the quiet and active operating points remain accessible.
For the paper’s nominal model, the qubits operate at 4.99 and 5.58 gigahertz, a separation of 590 megahertz, while each has an anharmonicity of minus 200 megahertz. That places the pair outside the “straddling” regime in which conventional cancellation is often easier. Numerical diagonalization finds zero crossings of the simulated ZZ interaction at reduced external-flux values of 0.328 and 0.346. The simulated transverse-decoupling point lies nearby at 0.339, where the remaining transverse coupling is below 1 megahertz. These values describe one modeled parameter set, not measured calibration targets from hardware.
A wider frequency window, with boundaries
The broader idea is to relax frequency-allocation pressure. Large superconducting processors cannot give every qubit an arbitrary frequency because gates, readout, unwanted resonances and fabrication variation must coexist. A coupler that cancels residual interaction across a wider detuning range could give designers more room to place frequencies and physical connections. In the simulation, complete ZZ cancellation remains available only over a finite range. For the modeled circuit, the paper identifies a critical detuning near 747 megahertz, beyond which the cancellation disappears. The proposal widens a design window rather than removing frequency constraints.
The authors also test sensitivity to modeled component changes. Their parameter sweeps indicate that the existence of a zero-ZZ point is relatively insensitive to changes in the Josephson energy of the coupler’s single junction, while depending more strongly on the asymmetry between the two resonator inductances. Removing direct inter-qubit capacitance in the model does not eliminate the cancellation mechanism because indirect paths through the resonator network remain. These are useful signs for eventual engineering, but a parameter sweep cannot reproduce every defect, loss mechanism, parasitic mode or packaging interaction that a fabricated chip may introduce.
Turning the simulated interaction back on
To activate a gate, the simulated controller moves the circuit away from its idle bias toward half a magnetic flux quantum. At the nominal inductive-energy difference of 16 gigahertz, the model reaches a ZZ interaction magnitude of 69.2 megahertz at a reduced flux of 0.464. The paper summarizes this as approximately 70 megahertz. A different modeled asymmetry raises the maximum to 79.2 megahertz. A stronger ZZ interaction accumulates the conditional phase for a controlled-Z gate faster, but moving too quickly through the spectrum can push population into unwanted states.
The paper therefore builds a shaped round-trip flux pulse. It slows in parts of the trajectory where the simulated eigenstates are most vulnerable to unintended transitions and moves faster elsewhere. The theoretical peak interaction implies a 7.2-nanosecond lower bound based only on phase accumulation. That is not a predicted high-fidelity operating time. When the authors optimize the pulse using time-dependent simulations, a 20-nanosecond gate reaches coherent infidelity of approximately 1 × 10^-5. At 16 nanoseconds, the simulated infidelity rises toward 1 × 10^-3, showing the cost of pushing closer to the theoretical speed limit.
What the error figures include
Even the 20-nanosecond figure needs careful labeling. The coherent-error simulation neglects decoherence and uses a truncated set of dressed energy states, with corrections for excluded states. At that duration, the paper attributes about 74 percent of the simulated coherent error to leakage outside the computational subspace and the remaining 26 percent mainly to population exchange between the single-excitation qubit states. The authors report that a corrected 20-state propagation reproduces the conditional phase of a 40-state calculation within 0.1 degree. This is a numerical consistency check, not experimental validation.
The researchers separately estimate what relaxation and flux noise could add. Assuming 50-microsecond relaxation times for both qubits and both resonator modes, their model gives relaxation-induced infidelity of 3 × 10^-4 for the 20-nanosecond pulse. About 28 percent of that contribution comes from the resonator channels. With the assumed 1/f flux-noise spectrum and amplitude used in the paper, simulated dephasing infidelity is 1.6 × 10^-5. On those assumptions, the authors estimate total fidelity above 99.9 percent. No device produced that result.
Hardware must supply the next evidence
The relaxation estimate exposes the proposal’s most important physical challenge. Near the active bias, the qubit states become partly mixed with the coupler modes. Loss in either resonator can then damage the computational state directly. For coupler modes around 8 gigahertz, the paper says a 50-microsecond relaxation time corresponds to a quality factor near one million. Fabricating compact resonators that retain that low-loss performance beside qubits, wiring and flux controls is a demanding requirement. A real implementation would also need to test flux calibration, pulse distortion, thermal population, crosstalk from neighboring structures and repeatability across a wafer.
If those tests succeed, the double-resonator idea could become a useful option for connecting superconducting qubits without tying cancellation to a narrowly arranged frequency pattern or a deliberate direct capacitance. It may also reduce junction count compared with some multi-path couplers because the proposed network uses one Josephson junction. The next evidence must come from hardware: spectroscopy that locates both idle cancellations, measured on-off ratios, randomized or cycle benchmarking of the controlled-Z gate, coherence measurements of the resonator modes and operation inside a larger array. Until then, the reported speed, error rates, robustness and frequency flexibility are simulated properties of a proposed coupler, not capabilities of a working quantum processor.