Tunable Versus Fixed Couplers
A coupler mediates interaction between two qubits; making it tunable lets designers switch the interaction on for a gate and off for idling.
The Coupling Problem
To run a two-qubit gate, two qubits must interact. To keep an idle qubit coherent and free of crosstalk, that same interaction should vanish. A fixed capacitive or inductive link cannot do both: it provides an always-on exchange coupling whose strength is set at fabrication. Fixed couplers keep the hardware simple but leave residual ZZ interaction and static crosstalk that must be handled in software or absorbed into the error budget.
Tunable Couplers
A tunable coupler inserts a third element, usually a frequency-tunable transmon or SQUID, between the two computational qubits. Two coupling paths then run in parallel: a direct capacitive path and an indirect path through the coupler. By tuning the coupler frequency, the indirect path is made to interfere with the direct path, and at a specific bias the net effective coupling passes through zero.
- Off setting: net coupling near zero, suppressing residual ZZ and idle crosstalk.
- On setting: strong effective coupling for a fast, high-fidelity two-qubit gate.
- Continuous control lets the gate strength be calibrated rather than fixed at fabrication.
Trade-offs
Tunable couplers add a flux line, a bias to calibrate, and an extra mode that can host loss or leakage. The coupler must be kept out of the computational subspace and its own coherence matters during the gate. Fixed couplers avoid all of that hardware but force the designer to live with static ZZ, which can be partly cancelled by echo sequences or by careful frequency allocation.
Modern large processors overwhelmingly favor tunable couplers because the on/off ratio they provide is the most direct route to low idle crosstalk in a dense two-dimensional array. The choice is one of the central architectural decisions in superconducting hardware.
Either way, the coupler is what turns isolated qubits into an interacting processor, and its behavior feeds directly into gate calibration and crosstalk mitigation.