What Is Qubit Connectivity?

Qubit connectivity describes which pairs of qubits in a processor can interact directly through a native two-qubit gate. It is usually drawn as a coupling graph: each node is a qubit and each edge is a pair on which a native two-qubit operation can be applied directly. In this sense, quantum connectivity is a property of the hardware, not of the algorithm running on it.

On fixed-connectivity gate-model hardware, a missing edge is usually handled by routing the logical states through intermediate qubits with operations such as SWAPs; on reconfigurable hardware, the cost may instead be motion or reconfiguration. Connectivity therefore helps determine the routing overhead of a circuit, and each extra imperfect operation creates another opportunity for error.

Three terms come up often:

• Nearest-neighbor: each qubit couples only to adjacent qubits on a line or grid.
• Degree: the number of partners a qubit has in the coupling graph.
• All-to-all connectivity: every pair of qubits has a direct edge.

Why Connectivity Changes the Cost of a Circuit

A quantum compiler must map the qubits of a circuit onto the coupling graph, a task called qubit routing. If two qubits that need to interact are not adjacent, the compiler inserts SWAP gates to bring them together. A SWAP exchanges the states of two qubits and is commonly built from three CNOT gates:

In this decomposition, each inserted SWAP adds three entangling gates; the cost can differ on hardware with a native SWAP-like gate or a different native two-qubit gate set., and the number of SWAPs needed grows with the distance between the qubits on the graph. Finding the cheapest routing is itself a hard optimization problem, and compilers rely on heuristics (see Cowtan et al., 2019).

This is why benchmarks respond to connectivity. Quantum Volume uses random circuits that pair arbitrary qubits, so its score depends on connectivity as well as on gate errors.

How Connectivity Differs Across Platforms

Each hardware platform has a different native coupling graph.

Linke and colleagues (2017) ran the same algorithms on a five-qubit ion device and a five-qubit superconducting device, and attributed part of the performance difference to the ion device's full connectivity. The comparison was small in scale, so it illustrates the effect and does not rank platforms.

Reconfigurable Connectivity With Neutral Atoms

In a neutral-atom processor, atoms are held in optical tweezers and entangled through the Rydberg blockade, whose useful blockade radius is set by the Rydberg state and drive strength. Direct blockade gates are normally applied to atoms brought within that radius, while more distant atoms retain weaker van der Waals interactions. The difference is that the tweezers can move atoms, so the coupling graph depends on where atoms are placed, not on wiring fixed at fabrication.

Bluvstein and colleagues (2022) demonstrated coherent transport of entangled atom arrays and used it to realize programmable connectivity between gate layers. The cost is time: moving atoms takes longer than a gate, so compilers must schedule transport carefully.

QuEra's Aquila, an analog system on Amazon Braket since 2022, uses geometry directly. Atom positions are set before each run, and the interaction between two atoms falls off with distance as

so the layout chosen by the user decides which atoms interact strongly. Gemini, QuEra's gate-based system on premises since 2025, builds on the same atom-array approach.

Connectivity and Error Correction

Error-correcting codes place their own demands on the coupling graph. Surface codes need only nearest-neighbor interactions on a 2D grid (Fowler et al., 2012), which suits fixed-lattice hardware.

Higher-rate qLDPC codes generally require parity-check graphs that are not geometrically local in a single 2D nearest-neighbor layout. Bravyi and colleagues (2024) described bivariate bicycle codes that encode 12 logical qubits in 144 data qubits, with each check touching six qubits and a connectivity graph that splits into two planar layers. Xu and colleagues (2024) proposed running high-rate qLDPC codes, including hypergraph-product and lifted-product examples, on reconfigurable atom arrays. Whether a platform can supply such connections at acceptable speed and error is an active engineering question.

FAQ

Is more qubit connectivity always better?

No. Richer connectivity reduces the number of SWAP gates a circuit needs, but it can come with slower gates, crosstalk or harder control. What matters is the connectivity together with the fidelity and speed of the gates on each edge.

What is all-to-all connectivity?

It means every pair of qubits can interact directly, with no intermediate steps. Trapped ions in a single chain are the usual example. Larger systems often give up some of this, for example by dividing ions into zones.

Does reconfigurable connectivity remove the need for SWAP gates?

It can replace many SWAPs with physical movement of atoms, but it does not make long-range interaction free. Transport takes time, and that time has to be balanced against decoherence and atom loss.

How does connectivity relate to qubit count?

They are separate properties. A processor with many qubits but sparse connectivity may need many extra gates for a given circuit. Connectivity, gate fidelity and circuit depth together determine what a device can run.

Key Takeaways

• Qubit connectivity is the graph of which qubit pairs can interact directly, and missing edges cost extra gates such as SWAPs.
• Superconducting devices typically have fixed nearest-neighbor couplings, while trapped ions offer all-to-all connectivity within a chain.
• Neutral-atom arrays can change the coupling graph by moving atoms, trading transport time for flexibility.
• Surface-code syndrome extraction uses local 2D connections; higher-rate qLDPC implementations generally need nonlocal connectivity, additional connectivity layers or qubit movement.