Heavy-Hex Topology

qc/hardware

The physical connectivity pattern of IBM Heron processors (e.g. FakeTorino, 133 qubits — see fake backends):

  • Horizontal chains of qubits run across rows
  • Vertical bridge qubits connect rows at staggered positions
  • Most qubits: degree 2 (two neighbors)
  • Junction qubits: degree 3 (three neighbors) — branching points

Not all-to-all — gates between non-neighbors require SWAPs (expensive, see Transpilation). Efficient circuit design means working with this graph structure, not against it.

Building GHZ on a subgraph: start from the qubit that minimizes the maximum distance to any other qubit in the subgraph — the graph-theoretic “center” (minimum eccentricity). On a junction/T-shape, that’s the junction itself, even though it can only fire one CX per layer (so it “tells” its neighbors one at a time, staggered, while already-informed neighbors relay onward in parallel).

General case (many qubits): build a BFS spanning tree from a well-chosen center qubit, then entangle each qubit with its BFS parent, layer by layer. Depth ≈ longest root-to-leaf distance in the tree. This generalizes “start from the middle” from a line to an arbitrary graph.

Self-Check

  • Could you describe what a junction qubit is and why it’s structurally different from most qubits on the chip?
  • Why does building a GHZ state on this topology start from the graph-theoretic “center” rather than an arbitrary qubit?
  • What does a BFS spanning tree buy you here that a simple fan-out doesn’t?