on-chip crosstalk
On-chip crosstalk is the quantum-processor version of an old, ordinary annoyance: you try to talk to one thing and accidentally nudge its neighbors too. A quantum chip is a crowded little city of qubits, control lines, and readout wires packed micrometers apart, and the laws of electricity do not respect the labels you put on them. A microwave pulse you send to flip qubit A leaks a faint copy onto qubit B; a flux current meant to tune one qubit bends the frequency of the one beside it; and even when you send nothing at all, neighbors can keep quietly tugging on each other. Each stray nudge is small, but quantum gates need to be precise to a fraction of a percent, so small errors that pile up across many qubits are exactly what corrupts a computation.
Crosstalk shows up through several physical channels, and good chip design fights each one. Capacitive and inductive coupling let a signal jump across a gap or through a shared magnetic field, so layout spreads things out, adds grounded shielding, and uses airbridges to stop ground planes from splitting into noisy islands. Classical drive leakage, where a control pulse simply spills onto the wrong qubit, is tamed by filtering, careful routing, and spacing qubit frequencies apart so a tone meant for one is off-key for the others. The subtlest channel is the always-on ZZ interaction: two fixed qubits sitting near each other shift one another's frequency a little all the time, even when idle, slowly smearing the phase of any computation. Tunable couplers exist largely to switch that ZZ term off between gates.
The honest picture is that crosstalk is never fully eliminated, only pushed down and budgeted for. You measure it, model it, and correct for what you can in software by tweaking pulses to pre-compensate, but every wire you add to control more qubits is another path for signals to leak. This is one reason today's processors stay small and noisy: as you pack qubits tighter to scale up, crosstalk and frequency crowding get worse together, and managing them is a central, unglamorous part of why building a bigger quantum chip is so hard.
A pulse of amplitude A aimed at qubit j leaks onto neighbor k scaled by a crosstalk coefficient c_jk; chip design and pulse pre-compensation aim to keep that coefficient as close to zero as possible.
Crosstalk is managed, not cured: you trade it down with layout, shielding, filtering, frequency planning, and tunable couplers, then characterize the residue and correct for it in software. Every extra control wire you add to scale up reopens a path for it.