Dataset Viewer
Auto-converted to Parquet Duplicate
contributor_id
stringclasses
2 values
name
stringclasses
2 values
affiliation
stringclasses
1 value
role
stringclasses
2 values
collections
listlengths
8
18
notes
stringclasses
2 values
saikat_das
Saikat Das
Levenson-Falk Lab, USC
generator, simulator
[ "exp6", "exp7", "exp8", "exp9", "exp10", "exp11", "exp12", "margin_paddle" ]
exp6-12 designs and Q3D solves, exp12 full re-export; paddle margin designs and solves
sadman_shanto
Sadman Ahmed Shanto
Levenson-Falk Lab, USC
generator, simulator, curator
[ "atlas", "campaign450", "exp13", "exp14", "exp15", "grid", "invxmon_pool", "largen", "margin_shielded", "multistack", "p3biased", "p3rand36", "pocket3_pool", "qm116", "random_v1", "synth_q3d", "twonet", "wonky_v1" ]
all other designs and Q3D solves, the lattice layouts (grid), priors, embeddings, curation
SQuADDS logo

SQuADDS_PlanarCapacitance: Q3D capacitance matrices of planar superconducting circuit layouts

Layouts of superconducting circuit components and their Maxwell capacitance matrices from Ansys Q3D, the data behind our paper QCap: a physics-corrected graph neural network for the capacitance matrix of any planar superconducting device layout. Each design comes with its GDS file, the Q3D matrix of every conductor in it, its geometric embedding, and two physics priors from a thin-film method of moments, the coarse prior and the extrapolated prior that our model corrects.

What's inside

Each config is one table, keyed by design.

  • designs: one row per design, with its component family, generator options and split
  • layouts: one row per GDS file, with its sha256 and conductor count N
  • simulations: one row per Q3D simulation, with its setup, stack, convergence and run time
  • labels: the Maxwell matrix of the last complete adaptive pass
  • label_passes: the matrix after every adaptive pass
  • geometric_embedding_nodes, geometric_embedding_edges and geometric_embedding_globals: the conductor vectors, pair vectors and global vector
  • physics_prior: the physics prior, one coarse method-of-moments solve per design
  • optimized_physics_prior: the extrapolated prior, from two coarse solves
  • model_input: the packed input that our models read
  • splits, test_sets, contributors, licences and registry: bookkeeping

Component families

The counts are those of Table I of the paper.

group component families train validation test
capacitor families interdigital, paddle, concentric, star, multipad star 32,073 4,071 4,009
qubit families Xmon, TransmonPocket, shielded concentric 735 93 926
generated layouts single conductor in a ground cut, random shape placements, mixed-shape layouts, local edge effects, third-body couplings 2,546 295 364
held-out sets crowded pockets, freeform geometries, quantum-metal QComponents, large N, unseen qubits and couplers, held-out TransmonPockets 0 149 600
multi-qubit grids one-, five- and nine-qubit grids (15, 35 and 67 conductors) 0 0 3
total 35,354 4,608 5,902

The designs table also lists 18 test designs that have no Q3D matrix. They are not scored.

The Q3D matrices of the test designs are held out and used only for evaluation. The exception is the 1,904 interdigital test designs, whose matrices were already public in SQuADDS_DB and are included here. Their layouts, embeddings and priors are here like every other design's.

Conventions

  • Capacitance is in fF with Maxwell signs: diagonal entries are positive and off-diagonal entries are negative.
  • Ground is conductor 0. Nets are never merged, so N is read from the layout and counts the ground.
  • Every simulation uses a 350 µm silicon substrate (eps_r 11.45) and zero-thickness metal. In the GDS files the ground is on layer 1/0 and the conductors are on layer 1/10.

Loading

from squadds.ml.capacitance import PlanarCapacitanceDB

db = PlanarCapacitanceDB()
val = db.designs(split="val")
matrices = db.labels(val.design_id[:5])   # Q3D Maxwell matrices in fF

The tables are Parquet files, so datasets.load_dataset("SQuADDS/SQuADDS_PlanarCapacitance", "designs") and pandas also work.

Contributors

Saikat Das designed and simulated the interdigital, paddle, concentric and star capacitors and part of the multipad star capacitors. Sadman Ahmed Shanto designed and simulated every other family and the multi-qubit grids, and computed the priors and embeddings.

Citation

If you use these data or models, please cite our paper and SQuADDS.

@misc{Shanto2026planarcapacitance,
  title   = {{QCap}: a physics-corrected graph neural network for the capacitance matrix of any planar superconducting device layout},
  author  = {Shanto, Sadman Ahmed and Das, Saikat and Levenson-Falk, Eli},
  year    = {2026},
  url     = {https://lfl-lab.github.io/SQuADDS/ml-paper/paper/}
}

@article{Shanto2024squaddsvalidated,
  doi = {10.22331/q-2024-09-09-1465},
  url = {https://doi.org/10.22331/q-2024-09-09-1465},
  title = {{SQ}u{ADDS}: {A} validated design database and simulation workflow for superconducting qubit design},
  author = {Shanto, Sadman and Kuo, Andre and Miyamoto, Clark and Zhang, Haimeng and Maurya, Vivek and Vlachos, Evangelos and Hecht, Malida and Shum, Chung Wa and Levenson-Falk, Eli},
  journal = {{Quantum}},
  issn = {2521-327X},
  publisher = {{Verein zur F{\"{o}}rderung des Open Access Publizierens in den Quantenwissenschaften}},
  volume = {8},
  pages = {1465},
  month = sep,
  year = {2024}
}

Licence

CC-BY-4.0.

Downloads last month
2,682

Models trained or fine-tuned on SQuADDS/SQuADDS_PlanarCapacitance

Space using SQuADDS/SQuADDS_PlanarCapacitance 1

Collection including SQuADDS/SQuADDS_PlanarCapacitance