Wallraff 17-qubit (Metal 53)

The ETH Zurich distance-3 surface-code device: 17 flux-tunable star transmons (9 data, 8 auxiliary) on a lattice, 24 nearest-neighbour couplers, and per qubit a λ/4 readout resonator with its own Purcell filter, read out on four feedlines. The layout is the Qiskit Metal “53” mock-up (Apache-2.0, © Qiskit Metal contributors), traced from micrographs of Krinner et al. (Nature 605, 669, 2022): a mock-up, not a mask. The Quantcore rebuild is distributed under GPL-3.0.

Build this one to see a full 17-qubit chip built from a generator, with every part drawn through the same Designer steps as the small chips. It takes about half a minute to build.

The whole m53 chip as the v2 notebook builds it

At a glance

Notebook

examples/designs/metal/m53_wallraff_17q.ipynb (an older one-cell version is kept as m53_wallraff_17q.v1.ipynb)

Source

Qiskit Metal 53-Wallraff_17Qubit_SurfaceCode.ipynb and its builder resources/wallraff_17q/, © Qiskit Metal contributors, Apache-2.0; targets from Krinner et al. (2022)

Chip

14.4 × 14.4 mm

On the chip

17 star transmons · 24 couplers · 17 readout and 17 Purcell λ/4 resonators · 4 feedlines · 17 Z and 17 XY lines · bond pads on the used slots of 48

Builds today

yes (layout built on 2026-11-19 in 33 s)

How the notebook builds it

The layout segments of m53, one picture each
  1. Die. An empty 14.4 mm square with the QCT frame, the lattice centre at the die centre.

  2. Bond pads. The generator’s pads, slid to the 100 µm pad-edge rule (the picture zooms on one pad).

  3. Feedlines. The four feedlines with their interdigital tees and input capacitors.

  4. Qubits. Seventeen StarQubit transmons, each with a PDK SQUID (the picture zooms on the centre of the lattice).

  5. Readout resonators. A readout λ/4 and a Purcell λ/4 per qubit (zoom).

  6. Couplers. The 24 nearest-neighbour coupler lines (zoom).

  7. Z and XY lines. The flux and drive lines from the pads to every qubit.

  8. Whole chip. The qubit targets, sim groups for the FEM jobs, the content centred in the die, the design saved.

LOM estimates

Numbers are from the saved output of the .v1 notebook (the 17-qubit chip was not re-run for these docs). Each qubit is solved with its readout chain, Purcell filter and the whole feedline. The resonator column is the readout resonator.

qubit

f_q (GHz)

α (MHz)

resonator

f_r (GHz)

κ (MHz)

χ (MHz)

Z1

6.22

-135

RO_Z1

7.348

3.15

-13.8

D1

3.803

-171

RO_D1

6.766

7.42

-2.54

D4

3.804

-171

RO_D4

7.097

8.2

-2.35

X2

5.856

-148

RO_X2

7.557

9.67

-5.75

D2

3.804

-171

RO_D2

6.964

7.83

-2.42

X1

6.006

-142

RO_X1

7.391

5.86

-8.72

D7

3.944

-169

RO_D7

6.9

8.33

-2.62

Z2

6.083

-142

RO_Z2

7.514

6.18

-8.58

D5

3.804

-171

RO_D5

7.154

6.64

-2.48

Z3

5.927

-142

RO_Z3

7.227

5.11

-9.29

D3

3.804

-171

RO_D3

6.871

7.69

-2.48

X4

6.008

-143

RO_X4

7.474

6.68

-7.93

D8

3.804

-171

RO_D8

7.055

8.1

-2.38

X3

5.928

-143

RO_X3

7.282

5.66

-8.68

D6

3.657

-173

RO_D6

7.013

7.53

-2.25

D9

3.944

-169

RO_D9

6.86

8.25

-2.65

Z4

6.007

-143

RO_Z4

7.428

6.21

-8.35

group

pair

ZZ (kHz)

neighbours

hybridised

D5+Z3

D5-Z3

-2110

yes

no

The qubit-readout coupling g computed from the simulated χ is 313 to 392 MHz for the 17 qubits (the paper’s Table I gives 167 to 269 MHz), and the static ZZ of the D5-Z3 pair through its coupler is -2.11 MHz.

These are lumped-model estimates from 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: every component, port, route and length.

The metal: references in this spec build as written, through the metal: wrapper (ports are the Metal pin names, parameters the Metal options). metal:StarQubit and metal:CapFingerInFrame both resolve. The Quantcore and KQC substitutions are optional.

The notebook examples/designs/metal/m53_wallraff_17q.ipynb (id m53_wallraff_17q) is a Quantcore PDK rebuild with the paper’s values. It is generated by the surface-17 generator examples/designs/devicelayout/_surface17.py; see “Rebuild (Quantcore PDK)” at the end. The source’s own wallraff_17q.build() (348 metal: cells and 328 wires, readouts at 8.40 GHz) is the notebook’s optional reference section 7 (m53_wallraff_17q_reference.png, not saved).

Source: Quantum Metal docs/circuit-examples/F.Small-quantum-chips/53-Wallraff_17Qubit_SurfaceCode.ipynb. The builder is resources/wallraff_17q/wallraff_17q.py and the measured data is resources/wallraff_17q/geometry.json. Licence Apache-2.0. Device: Krinner et al., Nature 605, 669 (2022), traced from the Wallraff lecture micrographs at about 7.5 µm/px. The notebook says it is “a mockup, not a mask”. The chip has 17 star transmons (9 data D1–D9 and 8 auxiliary X1–X4, Z1–Z4) and 24 lattice coupler lines. Each qubit has a λ/4 readout resonator, a λ/4 Purcell filter, a flux line and a drive line. There are 4 multiplexed feedlines, each with a series input capacitor, and 48 launchpads. Metal builds 348 components. Die 14 377 × 14 377 µm, Metal origin at qubit D5 = chip centre, so Designer = Metal + (7188.5, 7188.5). All geometry.json coordinates are mm in that centred frame. CPW 8.5/5.5 µm (chosen for 50 Ω on Si; not resolvable from the images). Metal layer 1, with airbridges on the bridge layer. The build needs metal:StarQubit (5 pads at 72°; no Quantcore/KQC qubit has 4 couplers plus a readout pad) and metal:CapFingerInFrame (closest: KQC FingerCapacitorSquare).

Components

Every qubit is metal:StarQubit(radius=160, center_radius=53, gap_couplers=13, gap_readout=6, connector_length=40, trap_offset=11, junc_h=53, number_of_connectors=4, resolution=16). The cell is a 0.32 mm etched pocket. Its pins are pin_cpl1…4 and pin_rdout, and they sit 0.20–0.21 mm from the centre. Lattice: centre = (col · 1609.2, row · 1753.9) + 7188.5. Pad angles are compass degrees. The junction lies on the flux axis A. The readout pad is at A + 36·s. For a neighbour in direction φ, take rel = s·(φ − A) mod 360; its coupler pad is at A + s·T[rel], with T = {0: −36, 90: 108, 180: 180, 270: 252}. Example: D5 (A 0, s +1) has E 324, N 108, W 180, S 252. Metal rotation_* = angle + 90 (StarQubit puts each pin 90° behind its option). Pin roles: pin_cpl1 goes to the E neighbour, cpl2 to N, cpl3 to W and cpl4 to S. Bond a→b uses a.pin_cpl1 → b.pin_cpl3 (horizontal) or a.pin_cpl2 → b.pin_cpl4 (vertical).

qubit

(col,row)

at (x, y)

A, s

group

RO / PU length (µm)

flux pad · drive pad

Z1

(−1, 2)

(5579.3, 10696.3)

180, −1

NW

3682 / 3301

N−4.5 · N−2.5

D1

(0, 2)

(7188.5, 10696.3)

0, +1

NE

4045 / 3464

N+2.5 · N+0.5

D4

(−1, 1)

(5579.3, 8942.4)

180, −1

NW

3787 / 3150

W+5.5 · N−5.5

X2

(0, 1)

(7188.5, 8942.4)

0, +1

NE

3577 / 3125

N+4.5 · N+3.5

D2

(1, 1)

(8797.7, 8942.4)

0, +1

NE

3874 / 3221

E+5.5 · N+5.5

X1

(2, 1)

(10406.9, 8942.4)

0, +1

NE

3664 / 3238

E+2.5 · E+4.5

D7

(−2, 0)

(3970.1, 7188.5)

180, −1

NW

3944 / 3451

W+0.5 · W+2.5

Z2

(−1, 0)

(5579.3, 7188.5)

180, −1

NW

3641 / 3178

W+3.5 · W+4.5

D5

(0, 0)

(7188.5, 7188.5)

0, +1

NE

3746 / 3366

E+1.5 · N−0.5

Z3

(1, 0)

(8797.7, 7188.5)

0, −1

SE

3626 / 3225

E−3.5 · E−4.5

D3

(2, 0)

(10406.9, 7188.5)

0, −1

SE

3977 / 3450

E−0.5 · E−2.5

X4

(−2, −1)

(3970.1, 5434.6)

180, +1

SW

3676 / 3170

W−2.5 · W−4.5

D8

(−1, −1)

(5579.3, 5434.6)

180, +1

SW

3823 / 3221

W−5.5 · S−5.5

X3

(0, −1)

(7188.5, 5434.6)

180, +1

SW

3647 / 3231

S−4.5 · S−3.5

D6

(1, −1)

(8797.7, 5434.6)

0, −1

SE

3860 / 3231

E−5.5 · S+5.5

D9

(0, −2)

(7188.5, 3680.7)

180, +1

SW

3969 / 3398

S−2.5 · S−0.5

Z4

(1, −2)

(8797.7, 3680.7)

0, −1

SE

3698 / 3184

S+4.5 · S+2.5

Pads use the edge letter plus the slot offset in mm from the edge centre (1 mm pitch).

  • LaunchpadWirebond(pad_width 170, pad_height 230, pad_gap 70, taper_height 240) → BondPad with those sizes. 12 per edge at slots ±0.5…±5.5 mm. Each pad’s necking point sits 6634 µm from the centre, so for example N+2.5 is at (9688.5, 13822.5) with rot 180.

  • Feed pads: NW N−3.5/W+1.5; NE N+1.5/E+3.5; SE E−1.5/S+3.5; SW W−3.5/S−1.5.

  • Spare pads (no line): N−1.5, E+0.5, S+0.5, S+1.5, W−1.5, W−0.5.

  • Capacitors: FEEDCAP_g and PUCAP_q are CapFingerInFrame: frame_width 72.4, frame_trace 16.3, cap_gap 10, ground_gap 10. frame_length is 244.7 for the input caps; for PUCAP it is the per-position outer_length_um_at_mm − 10 µm, spanning 138.6–277.

  • Input-cap centres (mm): NW (−3.18, 5.76), NE (5.435, 3.35), SE (3.2, −5.76), SW (−5.47, −3.34).

  • RPCAP_q is CapNInterdigital(finger_count 2, finger_length 0, cap_width 25, cap_gap 10, cap_gap_ground 8). It joins taps on RO_q and PU_q; the line spacing at the tap is 148–172 µm (readout_purcell_coupler).

Routes

All routes are traced polylines (PolylineCPW). Load them from geometry.json["lines"], convert mm→µm, add 7188.5, and pass them as nodes (after Douglas–Peucker 5 µm, resampling 45 µm and fillet 15 µm, as the Metal builder does).

family

count

from → to

length (µm)

termination / notes

CPL_a_b

24

qubit coupler pin → neighbour coupler pin

1453–1538

lines.coupling; the line leaves each pin straight for 40 µm. The bond toward the flux-side neighbour leaves 36° off and jogs back

FLUX_q

17

LP_flux_q → free_end 215 µm from the qubit centre

3658–7498

short (term=0) beside the SQUID; bridges over CPL_

DRIVE_q

17

LP_drive_q → free_end 280 µm from the centre

3295–7786

open (term=6); bridges over CPL_, FLUX_

RO_q

17

q.pin_rdout → free_end (short)

3577–4045

λ/4, open at the qubit; tap “rp” → RPCAP; bridges over earlier lines

PURCELL_q

17

free_end (short) → PUCAP_q.frame

3125–3464

λ/4; tap “rp” → RPCAP

FEED_g

4

LP_feed (edge 1) → FEEDCAP_g → LP_feed (edge 2)

NW 10 659, NE 18 840, SE 10 241, SW 10 599

taps pu_q → stub PUSTUB_q → PUCAP_q.finger

Crossings: where a line crosses an earlier one, Metal cuts it and inserts an Airbridge (pins a/b). The stage order is couplers → flux → drive → readout → Purcell → feed, so every line bridges over the earlier ones. In KQC, use an AirbridgeConnection instance per crossing (route → AB.a, AB.b → route). Taps: a 3-port KQC WaveguideCoplanarSplitter per tap, with port c to the capacitor or stub.

Launchers, pads, test structures, markers, labels

48 pads (8 feed, 17 flux, 17 drive, 6 spare). There are no markers or labels. The airbridges that tie ground islands inside the plaquettes are not drawn; the DRC warns that the ground is split.

LOM node structure

  • 17 transmon islands. The source uses a single JJ standing in for the device’s SQUID; the flux line is a short beside the junction.

  • Each of the 24 bonds is a short CPW coupler line. Model it as C_g to both islands (a non-resonant coupling network).

  • Per qubit, the chain is: island → C_g → RO λ/4 (short end) → RPCAP C_c → PU λ/4 → PUCAP C_κ → feedline g → FEEDCAP series C → 50 Ω ports.

  • Feedline groups: NW {D4, D7, Z1, Z2}, NE {D1, D2, D5, X1, X2}, SE {D3, D6, Z3, Z4}, SW {D8, D9, X3, X4}.

  • Control: flux line = flux bias on the SQUID; drive = C_d.

Targets (if stated)

Readout frequencies are 6.77–7.55 GHz (paper). The shorter resonators belong to the higher-frequency auxiliary qubits. No qubit frequencies, Lj or Cj are given.

Simplifications for a first rebuild

  1. Build 54_wallraff_cell (D5–Z3) first, then stage this chip the same way the source does: pads → qubits → couplers → flux/drive → readout/Purcell → feeds. Run the DRC after each stage.

  2. Build the qubits with metal:StarQubit. Do not substitute 3-coupler KQC qubits, because D5 needs 5 pads.

  3. Use FingerCapacitorSquare for CapFingerInFrame and CapNInterdigital. Replace the tap stubs with splitters.

  4. Leave the crossings out at first (draw the earlier line only) and add AirbridgeConnection crossings last.

Rebuild (Quantcore PDK)

Paper values used (Krinner et al., Nature 605, 669 (2022), arXiv:2112.03708, Table I and Methods; the text was extracted from the arXiv PDF):

qubit

f_q idle (GHz)

α (MHz)

f_RO (GHz)

g (MHz)

κ_eff (MHz)

χ (MHz)

D1 … D9

3.885, 3.994, 3.952, 3.878, 3.895, 3.74, 4.056, 3.993, 4.143

−184, −184, −183, −184, −186, −184, −181, −183, −181

6.769, 6.979, 6.88, 7.12, 7.18, 7.032, 6.91, 7.075, 6.868

244, 269, 241, 241, 238, 244, 267, 265, 260

6.3, 7.2, 8.3, 8.7, 3.5, 9.0, 8.6, 8.2, 8.9

−2.0, −2.2, −1.9, −1.6, −1.5, −1.6, −2.3, −2.0, −2.4

X1 … X4, Z1 … Z4

6.097, 5.885, 6.022, 6.049, 6.328, 6.192, 5.956, 6.037

−170, −174, −170, −170, −163, −168, −171, −170

7.372, 7.554, 7.258, 7.461, 7.316, 7.502, 7.2, 7.412

167, 168, 167, 168, 171, 170, 167, 171

15.1, 20.1, 13.0, 11.0, 11.2, 12.2, 14.3, 10.0

−2.8, −1.9, −3.2, −2.6, −4.7, −2.9, −3.2, −2.8

Other device facts from Methods:

  • asymmetric SQUIDs; the data qubits idle at their minimum, the auxiliaries at their maximum;

  • flux lines are shorted near the SQUID;

  • J/2π ≈ 7 MHz at the CZ interaction point;

  • each readout is paired with its own resonant λ/4 Purcell filter;

  • 4 feedlines, with readouts about 200 MHz apart on each and a band of 6.8–7.6 GHz;

  • 150 nm Nb on high-resistivity Si, Al/AlOx junctions with Al bandages, Al-Ti-Al trilayer airbridges.

Files:

  • examples/designs/metal/m53_wallraff_17q.ipynb, .json;

  • m53_wallraff_17q{,_centre,_jj,_bridge,_reference}.png;

  • generator examples/designs/devicelayout/_surface17.py, shared with devicelayout/demoqpu17.

part

rebuild

frame / stack

PDK die frame, 14.4 × 14.4 mm box, device ID ETH17; PDK sapphire stack (the device is Nb on Si)

lattice

the mock-up’s 5 × 5 sites, pitch 1609.2 × 1753.9 µm, D5 at the die centre

qubits

metal:StarQubit, radius 180, centre radius 100, coupler gap 13, connector length 60 (keeps the SQUID leads inside the pocket), readout gap 20 (data) / 40 (aux). The size is chosen so the LOM α of a bare qubit is −164…−192 MHz. Couplers point E/N/W/S, the readout pad along the qubit’s feedline diagonal, the SQUID in the 90° E–N gap

junctions

Quantcore Squid (registry.trilayer_junction(..., squid=True)), sized with size_junction(f_idle, C_Σ, n=2), LOM-calibrated C_Σ of 117 fF (data) / 132 fF (aux). R_n: data 78–95 kΩ, aux 31–35 kΩ, all outside the PDK’s 40–50 kΩ, because the paper’s data band lies below the PDK’s 4–8 GHz and its α needs a large C_Σ. The 0.01 µm finger grid quantises the drawn data f_q (3.76 / 3.92 / 4.08 GHz)

readout chain

readout λ/4 hairpin from the readout pad, then a 1.2 fF tap (K_<q>, J ≈ 12 MHz, declared), then the Purcell λ/4 hairpin, then a metal:CapNInterdigitalTee (8 fingers) on the feedline. Lengths are v_ph/4f × 0.95 / 0.958 (data / aux readouts) and × 0.945 (filters), LOM-calibrated so the dressed f_r lands on Table I and each filter is resonant with its readout

feedlines

NW {Z1, D4, Z2, D7}, NE {X1, D2, D5, X2, D1}, SE {Z4, D6, Z3, D3}, SW {X4, D8, X3, D9}, each from its mock-up input pad through a series FingerCapacitorSquare

D5

It is in the closed plaquette X2–D2–D5–Z3. Both hairpins stay inside; the tee is at (8450, 9555) in the channel north of the X2–D2 coupler, with the filter lead bridged over that coupler. The flux line comes from N−0.5, through the empty plaquettes west of X2; the drive line comes from E+1.5 through plaquette D2–X1–Z3–D3 (D5’s two pads are exchanged)

control lines

cpw routes from the mock-up’s flux / drive pad slots. The generator exchanges a qubit’s two pads where our module layout crosses less. Z is shorted beside the SQUID after passing over the qubit’s N coupler; XY is open

crossings

37 AirbridgeConnections, automatic, on the drawn polylines; plan problems 0

Quick check (a script run of the notebook’s cells, not nbclient):

  • DRC: 0 errors, 54 dangling_port (34 squid_a/b, 20 free edge coupler pads).

  • LOM: 138 nodes; the four feedline nodes carry 4 / 5 / 4 / 4 tees.

  • LOM vs paper:

    • f_r within 0.5 % of Table I;

    • filters 30–110 MHz below their readouts (data ≈ 30–40, aux ≈ 35–110);

    • dressed f_q 2–5 % below the idle frequencies (the finger grid sets the data steps);

    • dressed α −169…−173 (data) and −135…−151 MHz (aux), against −181…−186 and −163…−174;

    • κ_r 6–8 MHz for data (paper 3.5–9) and 3–10 MHz for aux (paper 10–20);

    • χ −2.2…−2.6 MHz for data (paper −1.5…−2.4) and −5.5…−13.5 for aux (paper −1.9…−4.7);

    • g from χ ≈ 350–390 MHz for data (paper 238–269) and 310–335 for aux (167–171). The StarQubit LOM gives all five pads the same arc, so a narrower readout gap barely changes it.

  • D5–Z3 ZZ −2.1 MHz, against the paper’s < 8 kHz at idle: the LOM has no line C on a pad-to-pad coupler route .

  • FEM jobs: q3d D5_Z3 (nets D5.island, Z3.island) and eigenmode D5_readout.

The notebook is a 20-step staged build, following tutorial 03 Part A and m52. It has these features:

  • Pads: all 42 BondPads sit at the PDK’s 100 µm pad edge (edge="auto"), with D5’s pf_u_far set to 1575 µm.

  • Cheese: the source exported its GDS with Quantum Metal’s default renderer cheese, so the original wins: metal_cheese_rule() draws it in the build, 12 398 holes over the free ground, including the plaquettes.

  • Airbridges: D5_readout exports with airbridges.

  • Checks: content shift (0, 0). DRC as above. 24 PDK design warnings: pdk_rn × 17 (data 80–94 kΩ, aux 30–35 kΩ) and pdk_fq × 7 (D1–D6 and D8 idle below 4 GHz).

  • Iterate step: widening D5’s readout gap from 20 to 30 µm gives χ −2.48 → −1.91 MHz, g 392 → 346 MHz, f_r +50 MHz and f_q +15 MHz.

  • Box ports: the builder puts the port of a straight line that crosses the box edge at an angle on the line’s own axis, with the cut square to the line, so all 11 ports of D5_readout have a polygon (the box top moved out by 4 µm, so Z_X2_1 leaves the box on a straight section). Cutting the 50 Ω port along the box normal used to leave 4 of the 11 (FLNE_2, FLNE_3, XY_X2_1, Z_X2_1) without a polygon, because the box edge crosses those lines at about 45°. The layout is unchanged.

docs/images/designs/metal/m53_wallraff_17q{,_centre}.png are from this run.