dSABRE

A SABRE-style router for multi-core distributed quantum computers (DQCs). Given a quantum circuit and a multi-core architecture, dSABRE inserts intra-core SWAPs and inter-core teleports to make every two-qubit gate executable while minimising EPR-pair consumption — the dominant cost on near-term photonic-linked modular hardware.

This repository hosts the reference implementation, benchmark harnesses, per-circuit result JSONs, the standalone appendices, and the compiled paper (paper/dsabre.pdf).

What's in the box

code/                  reference implementation + benchmark scripts
  router.py            General_dSABRE_Router (multi-core routing loop)
  dsabre_ext.py        dSABRE_BurstExt — BFS extended-set variant ("dSABRE" in the paper)
  architecture.py      DistributedArchitecture, B-grid and H-grid builders
  layout.py            locality_aware_layout + run_passes helpers
  config.py            HardwareConfig
  benchmark.py         main benchmark runner (25q / 36q / 64q suites)
  bench_large.py       100q / 200q / 360q QFT scalability sweep
  bench_pytket_*.py    pytket-dqc baselines (main + layout-controlled)
  ablate_*.py          ablation drivers (extended-set decay, hop gain, ...)
  verify.py            structural-equivalence verifier for routed output
  verify_diff.py       cross-run diff for verification
  results/             per-experiment JSON results (matches paper tables)
  results_verify/      re-run verification artefacts

paper/
  dsabre.pdf           compiled manuscript (submitted to IEEE TCAD)
  appendices_standalone.pdf   online appendices referenced from the paper
  build_tables_xlsx.py        helper: builds paper_tables.xlsx from results/
  paper_tables.xlsx           consolidated table data

The paper's LaTeX sources are intentionally not in this repository.

Quick start

dSABRE is pure Python (Qiskit + NetworkX).

pip install qiskit networkx numpy
cd code
python benchmark.py --suite 25         # 25-qubit B-grid suite
python benchmark.py --suite 36         # 36-qubit B-grid suite
python benchmark.py --suite 64         # 64-qubit H-grid suite
python benchmark.py --suite all        # all three

Each run reads MQT-Bench circuits, sets up the matching architecture, runs SabreLayout (best of 3 seeds) followed by dSABRE forward → backward → forward routing, and writes results/results_<suite>.json.

To reproduce the scalability sweep (QFT at 100, 200, 360 qubits):

python bench_large.py

To re-verify that the routed circuits are structurally equivalent to the originals:

bash verify_run.sh

Using the router programmatically

from qiskit.converters import circuit_to_dag
from architecture import build_h_grid_architecture
from config import HardwareConfig
from layout import locality_aware_layout, run_passes
from dsabre_ext import dSABRE_BurstExt
import random

arch    = build_h_grid_architecture(rows=2, cols=3, m=4)
config  = HardwareConfig()
router  = dSABRE_BurstExt(arch, config)

dag     = circuit_to_dag(my_circuit)
layout  = locality_aware_layout(dag, arch, rng=random.Random(0))
routed, metrics = run_passes(router, dag, layout, n=3)   # fwd-bwd-fwd
print(metrics["eprs"], metrics["ls"])                    # EPR count, local SWAPs

route() accepts a Qiskit DAGCircuit (not a QuantumCircuit — call circuit_to_dag first). locality_aware_layout requires a random.Random instance, not a bare seed.

Headline numbers

Across 18 MQT-Bench circuits at 25 / 36 / 64 logical qubits, dSABRE reduces geometric-mean EPR consumption by

vs.	25q	36q	64q
TeleSABRE (best of 3 seeds)	−41 %	−44 %	−44 %
pytket-dqc (best of 5 seeds)	−68 %	−29 %	−16 %

with parity on the QFT scalability sweep up to 360 qubits on a 486-physical-qubit H-grid. Full per-circuit tables, the ablation study, and the cost-ratio sensitivity sweep are in the paper.

Citation

@misc{li2026dsabre,
      title={dSABRE: A SABRE-Style Router for Multi-Core Distributed Quantum Computers}, 
      author={Sanjiang Li},
      year={2026},
      eprint={2605.21960},
      archivePrefix={arXiv},
      primaryClass={quant-ph},
      url={https://arxiv.org/abs/2605.21960}, 
}

License

Released for academic use. Please cite the paper if you use the router, the benchmark scripts, or the result JSONs in derivative work.

Contact

Sanjiang Li — sanjiang.li@uts.edu.au — ORCID 0000-0002-3332-2546 Centre for Quantum Software and Information, University of Technology Sydney.