SWAP Overhead and Routing

qc/hardware

When two logical qubits that need to interact are mapped onto physical qubits that aren’t directly connected (per the coupling map), the transpiler inserts SWAP gates to physically move quantum states around until the interaction becomes possible on real hardware.

This has real costs:

• More circuit depth — more sequential layers, so more elapsed time and more exposure to Thermal Relaxation.
• More total gates — and since a SWAP is typically built from three CNOTs, each one adds three lots of two-qubit gate error, not just one.

This is exactly why layout (Physical vs Logical Qubits and Layout) and topology (Coupling Map and Topology) matter so much: a poor layout on a sparse topology can multiply your effective error rate through SWAP overhead alone, even with perfect individual gate fidelities.

Counting SWAPs in a transpiled circuit

def count_swap_gates(circuit: QuantumCircuit) -> int:
    swap_count = 0
    for instruction in circuit.data:
        if instruction.operation.name == "swap":
            swap_count += 1
    return swap_count

This uses circuit.data, a list of CircuitInstruction objects — one of the general tools for introspecting exactly what’s inside a circuit. See Circuit Introspection Cheat Sheet for the fuller toolkit.

Related

• Coupling Map and Topology
• Physical vs Logical Qubits and Layout
• Circuit Introspection Cheat Sheet

Self-Check

• Why is a SWAP gate actually three CNOTs, and what does that mean for error accumulation?
• Could you explain why a poor layout can hurt fidelity even with perfect individual gate quality?
• How would you count SWAP gates in a transpiled circuit using circuit.data?