IBM 27-qubit Falcon

IBM’s heavy-hex lattice cut as the 27-qubit Falcon (ibmq_montreal): fixed-frequency transmons on the vertices and edges of a hexagonal lattice, each pair coupled by a resonant bus. It is two rows of ten qubits, three bridge qubits and four spur qubits. No layout is public; IBM’s device data give the graph and targets, and the layout is ours.

This notebook generates a lattice from a graph: a heavy_hex(rows) helper makes the 27 qubits and 28 edges, then the notebook places pockets, hangers on three feedlines and one λ/2 bus per edge. It is a large build (33 instances, 58 routes).

The whole ibm_falcon_27q chip as the v2 notebook builds it

At a glance

Notebook

examples/designs/ibm/ibm_falcon_27q.ipynb (an older one-cell version is kept as ibm_falcon_27q.v1.ipynb)

Source

IBM published device data (ibmq_montreal coupling map); Chamberland et al., PRX 10, 011022 (2020); no public layout

Chip

27 × 15.2 mm Quantcore die, sapphire 650 µm (εr 10.25)

On the chip

27 pocket transmons · 27 λ/4 hanger readouts · 28 λ/2 buses · 3 feedlines · 6 bond pads

Builds today

yes (layout built on 2026-11-19 in 28 s)

How the notebook builds it

Each Design segment is one cell with a picture of the chip so far (see the notebook template).

The layout segments of ibm_falcon_27q
  1. Die. A die sized to the lattice (2.5 mm margins round ten 2.2 mm columns), with the QCT frame and the device ID HHX27.

  2. Bond pads. The three feedlines’ pads on the west and east edges.

  3. Feedlines. Three straight feedlines; the middle one crosses the three bridge buses on airbridges.

  4. Qubits. 27 TransmonPocket6 pockets with trilayer junctions, and their bus and readout pads.

  5. Readouts. 27 λ/4 readouts, each ending in a 300 µm hanger.

  6. Coupling. 28 λ/2 meander buses, one per edge.

  7. Control lines. None: the planar die drives qubits through the feedlines.

  8. Whole chip. The centring check, targets, sim groups and the save.

LOM estimates

Numbers are from the saved output of the .v1 notebook (this chip was not re-run for these docs): QuCat normal modes with QuTiP dressed values, each qubit solved with its readout and the feedline. IBM’s measured values are left out.

group

qubit

f_q (GHz)

α (MHz)

resonator

f_r (GHz)

κ (MHz)

χ (MHz)

Q0+R0

Q0

4.929

-151

R0

7.16

0.317

-0.422

Q1+R1

Q1

4.826

-144

R1

7.271

0.331

-0.347

Q2+R2

Q2

4.966

-146

R2

7.152

0.316

-0.47

Q3+R3

Q3

5.101

-145

R3

7.286

0.333

-0.603

Q4+R4

Q4

5.027

-147

R4

7.215

0.324

-0.385

Q5+R5

Q5

5.01

-146

R5

7.229

0.325

-0.577

Q6+R6

Q6

4.968

-150

R6

7.328

0.339

-0.504

Q7+R7

Q7

4.914

-143

R7

7.082

0.307

-0.6

Q8+R8

Q8

4.92

-143

R8

7.088

0.307

-0.406

Q9+R9

Q9

5.063

-149

R9

7.33

0.339

-0.541

Q10+R10

Q10

5.074

-146

R10

7.159

0.317

-0.412

Q11+R11

Q11

5.01

-146

R11

7.225

0.325

-0.577

Q12+R12

Q12

4.967

-143

R12

7.321

0.338

-0.377

Q13+R13

Q13

4.871

-147

R13

7.218

0.324

-0.423

Q14+R14

Q14

4.961

-142

R14

7.274

0.331

-0.568

Q15+R15

Q15

5.028

-147

R15

7.285

0.333

-0.372

Q16+R16

Q16

5.055

-145

R16

7.166

0.317

-0.618

Q17+R17

Q17

5.063

-149

R17

7.333

0.339

-0.54

Q18+R18

Q18

4.961

-142

R18

7.084

0.307

-0.623

Q19+R19

Q19

4.966

-143

R19

7.093

0.308

-0.421

Q20+R20

Q20

5.063

-149

R20

7.34

0.34

-0.539

Q21+R21

Q21

5.074

-146

R21

7.235

0.326

-0.396

Q22+R22

Q22

5.056

-145

R22

7.224

0.325

-0.6

Q23+R23

Q23

4.966

-146

R23

7.3

0.335

-0.438

Q24+R24

Q24

5.059

-145

R24

7.17

0.318

-0.503

Q25+R25

Q25

4.915

-143

R25

7.279

0.332

-0.547

Q26+R26

Q26

5.015

-149

R26

7.17

0.318

-0.564

pair

f_a (GHz)

f_b (GHz)

J (MHz)

target J (MHz)

ZZ (kHz)

Q3-Q5

5.104

5.012

1.46

1.49

70.7

Q5-Q8

5.012

4.92

1.52

1.41

74.5

Q8-Q11

4.92

5.012

1.11

1.47

51.1

Q11-Q14

5.012

4.966

1.52

1.4

51.9

Q14-Q16

4.966

5.058

1.17

1.55

56.4

Q16-Q19

5.058

4.966

1.62

1.47

84.4

Q19-Q22

4.966

5.058

1.17

1.49

57

Q22-Q25

5.058

4.92

1.62

1.36

476

Q25-Q26

4.92

5.015

1.18

1.43

56.4

Q0-Q1

4.928

4.825

1.3

1.31

58.3

Q1-Q4

4.825

5.027

0.895

1.32

-25.3

Q4-Q7

5.027

4.92

1.39

1.44

76.9

Q7-Q10

4.92

5.073

0.994

1.42

-261

Q10-Q12

5.073

4.966

1.49

1.58

88.6

Q12-Q15

4.966

5.027

1.05

1.49

31.4

Q15-Q18

5.027

4.966

1.39

1.7

45.9

Q18-Q21

4.966

5.073

1.05

1.49

58.8

Q21-Q23

5.073

4.965

1.58

1.42

101

Q2-Q3

4.962

5.104

1.5

1.36

1.08e+03

Q1-Q2

4.819

4.965

1.31

1.4

-1.81e+03

Q13-Q14

4.867

4.966

1.26

1.36

78.4

Q12-Q13

4.96

4.87

1.55

1.36

81.1

Q24-Q25

5.055

4.92

1.59

1.51

336

Q23-Q24

4.962

5.058

1.5

1.3

100

Q8-Q9

4.913

5.062

1.55

1.42

2.02e+03

Q19-Q20

4.96

5.062

1.64

1.46

126

Q6-Q7

4.967

4.92

1.37

2.01

42.1

Q17-Q18

5.062

4.966

1.54

1.52

77

The pair table lists the exchange coupling J and the static ZZ for the 29 qubit-bus-qubit edges. The edges Q22-Q25 (476 kHz), Q2-Q3 (1.08 MHz), Q1-Q2 (-1.81 MHz) and Q8-Q9 (2.02 MHz) are frequency collisions: the PDK’s α of about -145 MHz puts IBM’s neighbour detunings near |α|.

These are lumped-model estimates (component formulas and calibration), not FEM results. The LOM page says how far to trust them.

Build spec

The rest of this page is the detailed spec the notebook was built from, with every component, port, route and length.

Sources.

  • Device data (primary): Qiskit FakeMontreal backend files conf_montreal.json / props_montreal.json from qiskit-ibm-runtime (Apache-2.0; ibmq_montreal, Falcon r4, snapshot 2021-03-15; local copy build/external/qiskit_fake_backends/): the 28-edge coupling map, f_q 4.835–5.105 GHz, α −308 … −390 MHz, J (jq) 1.31–2.01 MHz on every edge, measurement LO centres 7.19–7.46 GHz (all values in the notebook’s cell 1).

  • Lattice: Chamberland, Zhu, Yoder, Hertzberg, Cross, “Topological and subsystem codes on low-degree graphs with flag qubits”, PRX 10, 011022 (2020) (arXiv:1907.09528): heavy-hexagon lattice, degree ≤ 3, cross-resonance gates on “fixed-frequency transmon qubits via microwave-resonant buses”, typical f01 ≈ 5 GHz, α ≈ −330 … −340 MHz; IBM Research blog “The IBM Quantum heavy hex lattice” (2021).

  • Device: Jurcevic et al., “Demonstration of quantum volume 64 on a superconducting quantum computing system”, Quantum Sci. Technol. 6, 025020 (2021) (arXiv:2008.08571): ibmq_montreal, qubits on a die bump-bonded to an interposer for signal delivery and readout.

Licence. Apache-2.0 data, papers cited; no heavy-hex layout is public (searched GitHub, Qiskit Metal, SQuADDS); layout and code ours (GPL).

What it is. 27 fixed-frequency floating transmons on the Falcon heavy-hex cut: two long rows of ten (bottom columns 1–10, top 0–9), three bridge qubits (columns 1, 5, 9), four spurs (columns 3, 7 above and below); one λ/2 meander bus per edge (28); a λ/4 readout per qubit on three horizontal feedlines (top: above the top spurs; middle: through the band between the rows, with airbridges over the three upper bridge buses; bottom: below the bottom spurs).

Generator. heavy_hex(rows, offset=1) (notebook cell 1): long rows → bridges at c ≡ offset + 2r (mod 4) → spurs. Falcon = heavy_hex([(1, 10), (0, 9)]); heavy_hex([(0, 13)] + [(0, 14)] * 5 + [(1, 14)], offset=0, spurs=False) gives the 127-qubit, 144-edge Eagle graph (a layout of it needs more feedlines).

Chip. 27.0 × 15.2 mm, Quantcore die frame (QCT), device ID HHX27; sapphire 650 µm, εr 10.25; 10/6 µm CPW; column pitch 2200 µm, level pitch 2400 µm. Junctions: Quantcore trilayer, L_J per qubit 6.49–7.17 nH (R_n 43–48 kΩ).

Notebook. examples/designs/ibm/ibm_falcon_27q.ipynb.

Components and routes

item

ref / type

notes

Q0 … Q26

metal:TransmonPocket6

pads 750 × 160, gap 20, pocket 950; bus pads 125 (corners ±x at +190 µm for row edges, middles ±y for vertical edges), readout pad 120

B_a_b (28)

meander route

λ/2; row edges 6.2 / 6.5 GHz alternating, vertical edges 6.35 GHz (assumed); length 60 400 / f µm (calibrated)

R0 … R26

meander → hanger

λ/4 3975–4121 µm, hanger 300 µm, ground gap 4, short; bridges’ readouts leave a corner pad and run +x 1 mm first

FL_top, FL_mid, FL_bot

cpw

BondPads on the side edges; FL_mid has airbridges at columns 1, 5, 9

Targets and LOM values

quantity

IBM

rebuild (LOM)

f_q

4.835–5.105 GHz

4.826–5.101 GHz, every qubit within 30 MHz (per-qubit L_J solved)

α

−308 … −390 MHz

≈ −145 MHz (PDK window)

readout

7.19–7.46 GHz

within 3 MHz; κ 0.32–0.35 MHz, χ −0.33 … −0.57 MHz

buses

–

6.23 / 6.51 / 6.37 GHz (row even / row odd / vertical)

J

1.31–2.01 MHz

0.84–1.51 MHz (bus pads 125 µm; the 6.5 GHz buses give the low end); ZZ 27–107 kHz on 22 edges

Simplifications / adaptations

Planar single die (Falcon: qubit die + interposer): readout on three feedlines with three airbridges, no drive lines (drive through the feedlines), no Purcell filters. α about half of IBM’s (PDK junction window). Bus frequencies are not published: assumed plan.

Frequency collisions. IBM’s frequency pattern assumes α ≈ −330 MHz. With the PDK’s α ≈ −145 MHz, neighbour detunings of 0.1–0.2 GHz sit near |Δ| ≈ |α| and the static ZZ becomes large on six edges: Q8–Q9 1.83 MHz, Q1–Q2 −1.71 MHz, Q2–Q3 0.93 MHz, Q22–Q25 0.40 MHz, Q24–Q25 0.29 MHz, Q7–Q10 −0.23 MHz. A Quantcore heavy-hex chip needs Chamberland §II C’s allocation redone for its α, or a junction window that reaches larger α.