DemoQPU17 surface-17

A 17-qubit surface-code demo: 17 star transmons on a 5 × 5 grid (9 data, 8 auxiliary), 24 fixed couplers, a hairpin readout with its own Purcell filter per qubit on four feedlines, and an XY and a Z line per qubit. It is DeviceLayout.jl’s DemoQPU17 example (AWS, MIT), shaped after the ETH Zurich surface-17 device, rebuilt on the Quantcore PDK (GPL-3.0).

This is a plan-driven chip: the shared surface-17 generator first plans every instance and line, then the notebook places them kind by kind. It is a large build (117 instances, 154 routes).

The whole demoqpu17 chip as the v2 notebook builds it

At a glance

Notebook

examples/designs/devicelayout/demoqpu17.ipynb (an older one-cell version is kept as demoqpu17.v1.ipynb)

Source

DeviceLayout.jl examples/DemoQPU17 (MIT, © Amazon.com, Inc. or its affiliates)

Chip

15 × 15 mm Quantcore die, sapphire 650 µm

On the chip

17 StarQubits · 24 couplers · 17 readouts with 17 Purcell filters · 4 feedlines · 34 control lines

Builds today

yes (layout built on 2026-11-19 in 31 s)

How the notebook builds it

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

The seven layout segments of demoqpu17 (the whole chip is the last picture)
  1. Die. The 15 × 15 mm die with the QCT frame, the lattice centre at the die centre.

  2. Bond pads. The generator’s pads, moved to the 100 µm pad-edge rule.

  3. Feedlines. The capacitive tees, the input capacitors and the four feedlines.

  4. Qubits. The 17 star transmons, with junctions sized per qubit.

  5. Readouts. The readout and Purcell λ/4 hairpins.

  6. Couplers. The 24 nearest-neighbour couplers.

  7. Control lines. The Z and XY lines, with airbridges at crossings.

  8. Whole chip. Targets, sim groups, centring and the save.

LOM estimates

Numbers are from the saved output of the .v1 notebook (this chip was not re-run for these docs): each qubit is solved with its readout chain, Purcell filter and the whole feedline. The readout columns are the readout resonator; the Purcell filter sits about 40 MHz above it with κ of 25 to 38 MHz.

qubit

f_q (GHz)

α (MHz)

resonator

f_r (GHz)

κ (MHz)

χ (MHz)

Q31

4.814

-219

RO_Q31

7.138

4.97

-9.88

Q32

5.774

-208

RO_Q32

7.588

5.12

-11.3

Q33

4.809

-217

RO_Q33

6.766

4.54

-11.9

Q22

4.671

-224

RO_Q22

7.32

4.33

-8.46

Q12

5.536

-209

RO_Q12

7.02

3.16

-13.5

Q25

5.647

-202

RO_Q25

6.932

1.96

-17.9

Q24

4.672

-224

RO_Q24

7.413

4.42

-8.2

Q23

5.652

-206

RO_Q23

7.118

2.95

-14.7

Q13

4.671

-223

RO_Q13

7.228

4.31

-8.74

Q35

4.815

-220

RO_Q35

7.323

4.95

-9.19

Q34

5.774

-208

RO_Q34

7.588

5.1

-11.3

Q44

4.951

-216

RO_Q44

7.141

5.98

-10.8

Q54

5.864

-193

RO_Q54

7.037

1.38

-22.9

Q41

5.755

-196

RO_Q41

6.939

1.45

-21.5

Q42

4.952

-216

RO_Q42

7.233

5.83

-10.4

Q43

5.88

-203

RO_Q43

7.499

3.73

-14

Q53

4.95

-215

RO_Q53

7.048

6.19

-11.2

group

pair

ZZ (kHz)

neighbours

hybridised

Q23+Q33

Q23-Q33

-2.06e+04

yes

no

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.

The notebook examples/designs/devicelayout/demoqpu17.ipynb (with its .json and PNGs) is a Quantcore PDK rebuild whose geometry comes from the shared generator examples/designs/devicelayout/_surface17.py. See “Rebuild (Quantcore PDK)” at the end. The source description below is unchanged.

Source: examples/DemoQPU17 (DemoQPU17.jl, params.jl, routing.jl) and docs docs/src/examples/qpu17.md (rendered) · components in src/schematics/ExamplePDK (ExampleStarTransmon, ExampleFilteredHairpinReadout, ExampleSeriesClawCapacitor, ExampleChip) · licence MIT (Amazon, verified) · Read on 2026-11-03 from a depth-1 clone in build/external/DeviceLayout.jl (Julia, not run). What it is: a demo after ETH Zurich (Krinner et al., Nature 2022). 17 star transmons sit on a 5 × 5 grid with 1.65 mm pitch, 9 data + 8 ancilla in a checkerboard. Neighbours share 24 fixed CPW couplers. Each qubit has a λ/4 hairpin readout resonator with its own λ/4 Purcell-filter hairpin, coupled through a tap capacitor; the filter is claw-coupled onto one of 4 feedlines. Each feedline has a series claw input capacitor. Every qubit has an XY line and a Z line. Crossovers are automatic airbridges; airbridges on all paths; ground holes (r 5 µm, 100 µm grid). Frame: ExampleChip 15 × 15 mm centred at (0, 0), with qubit (3, 3) at the origin. Box (0, 0, 15000, 15000), Designer = source + (7500, 7500). CPW: feedlines and control feeds 10/6, couplers and resonators 10/10, XY/Z near the qubit 3.3/2. Readout lengths use effective index 2.5.

Lattice generator (instead of 17 rows)

Site (row r, col c), r = 1 is the top row: x = 7500 + (c − 3)·1650, y = 7500 − (r − 3)·1650. Active sites (1 = qubit), rows top→bottom: 01100 / 01111 / 11111 / 11110 / 00110. Data qubit if (r + c) is even (D), else ancilla (A). Mirroring: LR if c > 3 or (c = 3 and r < 3); UD if r > 3 or (r = 3 and c > 3). Mirroring flips which wedge faces which neighbour, so XY/Z always point outward.

site

role

f_RO (GHz)

RO line

site

role

f_RO

RO line

(1,2)

A

7.0

1

(3,3)

D

6.8

1

(1,3)

D

7.3

2

(3,4)

A

7.6

3

(2,2)

D

7.4

1

(3,5)

D

7.4

3

(2,3)

A

7.1

2

(4,1)

A

6.9

4

(2,4)

D

7.5

2

(4,2)

D

7.3

4

(2,5)

A

6.9

2

(4,3)

A

7.5

4

(3,1)

D

7.2

1

(4,4)

D

7.2

3

(3,2)

A

7.6

1

(5,3)

D

7.1

4

(5,4)

A

7.0

3

Components (source)

instance

source component

key params (units)

notes

Q_rc (17)

ExampleStarTransmon

5-tip star, island_outer_radius 135, inner radius 80 between tips (data qubits: 20 at the readout wedge, a big readout coupler), island_ground_gap 15, island_coupler_gap 15, star_tip_width 50, rounding 5. Wedges N/E/S/W go to neighbours, wedge 5 to readout. Tip 1 carries the SQUID (ExampleSimpleSQUID). XY hook at 162°, Z hook at the SQUID

couplers with no neighbour are grounded (GROUNDED_COUPLERS)

RO_rc (17)

ExampleFilteredHairpinReadout

readout_total_length = filter_total_length = c/(4·2.5·f) − 1.0 mm (e.g. 7.0 GHz → 3.283 mm). Hairpin straight 1.5 mm, bend r 50, tap at 0.77 mm (tap CPW 5/25 µm, tap cap 25/15 µm, filter–readout gap 15). Filter→feedline: ExampleShuntClawCapacitor (inner cap 20 × 200, claw trace 10, gaps 5/20, feedline segment 300 µm). Per-site extra filter lengths: (3,1) 400, (2,5) 700, (4,1) 700, (3,3) 550 + 710 at ±45° (tap 0.45 mm, straight 1.0 mm)

fused to Q.readout

CIN_1…4

ExampleSeriesClawCapacitor (input_length 700)

inner cap 20 × 200, claw trace 10, inner gap 5, outer gap 20

at each RO_IN launcher

ports 1…48

ExampleChip launchers

12 per side, 1 mm pitch

clockwise from top-left; 6 unused

Routes (source)

  • Couplers (24): every NN pair, about 1.3 mm. In the source each star’s coupler path ends at ±825 µm and the two are fused.

  • Feedlines (4): RO_IN → CIN → RO of each site in order → RO_OUT (LINE_SPECIFICATIONS, READOUT_GROUPS): 1: L44 → L3 via (3,1), (3,2), (3,3), (2,2), (1,2); 2: L15 → L8 via (2,5), (2,4), (2,3), (1,3); 3: L20 → L27 via (3,5), (3,4), (4,4), (5,4); 4: L39 → L32 via (4,1), (4,2), (4,3), (5,3). Meander waypoints add length so each filter tap falls near a feedline half-wavelength.

  • XY (17) / Z (17): launcher → Q.xy / Q.z, feed CPW 10/6 tapering to 3.3/2. XY ends 50 µm from the ground edge; Z ends 2 µm from the qubit with a 16 µm return cut by the SQUID. Port map (index: line): 1 Z22, 2 XY12, 4 Z12, 5 Z33, 7 Z13, 9 XY13, 10 Z23, 11 XY23, 12 Z24, 13 XY24, 14 Z25, 16 XY25, 19 XY35, 21 Z35, 22 XY34, 23 Z34, 24 XY44, 25 Z44, 26 XY54, 28 Z54, 31 Z53, 33 XY53, 34 Z43, 35 XY43, 36 Z42, 37 XY42, 38 Z41, 40 XY41, 42 XY33, 43 XY31, 45 Z31, 46 XY32, 47 Z32, 48 XY22. Unused: 6, 17, 18, 29, 30, 41.

LOM node structure

17 grounded star islands (SQUID to ground, flux-tunable through Z, driven through XY). 24 Q–Q capacitive couplers (CPW links). Each Q has a readout λ/4 (C_g at wedge 5) → tap capacitor → Purcell λ/4 filter → claw capacitor → feedline k. Each feedline is one conductor with a series input capacitor and two 50 Ω ports. 34 control lines.

Targets

Readout frequencies in the table (6.8–7.6 GHz, effective index 2.5). Qubit frequencies are not stated.

Rebuild (Quantcore PDK)

Deliverable: examples/designs/devicelayout/demoqpu17.ipynb / .json / demoqpu17{,_centre,_jj,_bridge}.png. It is built on the PDK die frame (15 × 15 mm box, device ID QPU17) and the PDK stack (sapphire). The shared generator examples/designs/devicelayout/_surface17.py (Config → plan_chip → build) also builds metal/m53_wallraff_17q.

part

rebuild

qubits

metal:StarQubit radius 150 (pocket = 135 µm star + 15 µm gap), centre radius 70, coupler gap 15, readout gap 8 (data) / 15 (aux). Pins: couplers E/N/W/S; readout on the qubit’s diagonal (DeviceLayout mirroring: NW / NE / SE / SW per feedline quadrant); SQUID at 112.5° (local frame), between the N coupler and the readout pad, like tip 1; XY end at 157.5°

junctions

Quantcore Squid (trilayer, registry.trilayer_junction(..., squid=True)), sized with size_junction(f_q, 89 fF, n=2). The source gives no f_q; the PDK plan is data 4.80–5.20, aux 5.80–6.15 GHz. R_n: aux 47–52 kΩ (5 of 8 in 40–50 kΩ), data 65–76 kΩ (outside: a 4.8–5.2 GHz SQUID transmon at C_Σ 89 fF needs E_J ≈ 15 GHz)

couplers

24 straight cpw routes 10/10 µm between facing pins (1275 µm)

filtered hairpin readout (was MISSING)

readout λ/4 = route Q.pin_rdout → shorted free end, 3 hairpin legs along the diagonal (legs at v = 0, −80, −160 µm; pitch 80, r 35); Purcell λ/4 = route T_<q>.second_end → shorted free end, 4 legs at v = 140…380 µm; lengths v_ph/4f × 0.96 (readout) / × 0.948 (filter), LOM-calibrated loading; shunt claw → metal:CapNInterdigitalTee (8 fingers 60 µm, gap 5) at the module tip; tap capacitor → d.couple("K_<q>", "RO_<q>", "PF_<q>", C_fF=1.2) (J ≈ 12 MHz), not drawn

series claw capacitor (was MISSING)

KQC FingerCapacitorSquare (6 fingers 60 µm) 700 µm from each input pad

Q33 (closed plaquette)

readout (5 legs) and filter (4 legs) inside the plaquette; tee at (5550, 8790) in the channel west of the Q22–Q32 coupler; filter lead bridged over that coupler (AB_PF_Q33_0); Z_Q33 from the top through the Q12–Q23 plaquette, XY_Q33 from the left under row 3 (2 + 3 coupler bridges)

feedlines

qcc cpw routes through the tees in DeviceLayout’s order; stubs of 520 µm along each tee so its own XY/Z lines cross perpendicularly; no half-wave meanders

XY / Z

cpw 10/6 routes from the DeviceLayout ports to free ends (XY open at r = 250 µm, Z shorted at r = 265 µm beside the SQUID); the strip entry point is chosen per line (fewest, cleanest crossings)

crossings

30 AirbridgeConnections on the upper line (couplers < resonators < feedlines < Z < XY), found on the drawn polylines; plan problems 0

pads

42 BondPads (300 × 300 µm, taper 300) at the DeviceLayout port positions, ports 900 µm from the edge

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

  • fast DRC 0 errors, 54 dangling_port warnings (34 squid_a/b, 20 free edge coupler pads);

  • LOM 138 nodes / 351 branches (C 250, R 42, TLres 42, SQUID 17);

  • per-qubit groups: readouts within 1 % of params.jl (6.77–7.59 GHz), filters resonant 40–50 MHz above, κ_filter 27–38 MHz, κ_readout 1.4–5.8 MHz, χ −8…−22 MHz, dressed f_q 2–5 % below the plan;

  • Q23–Q33 ZZ −20.6 MHz: too large, because the LOM treats the coupler route between two pads as a wire without its own line C;

  • the fast test_examples -k demoqpu17 test passes.

Simplifications:

  • no ground-plane holes and no line bridges;

  • no feedline half-wave meanders;

  • control lines 10/6 µm all the way;

  • tap capacitor declared, not drawn;

  • grounded couplers left as free pads;

  • the qubit is the Metal StarQubit, not DeviceLayout’s wedge star (C_Σ from the rough StarQubit LOM; FEM job Q23_Q33).