Research map / Learning theory and algorithms / Combinatorial and search optimization
AI-Powered Algorithm-Centric Quantum Processor Topology Design
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Qtailor uses PPO-based reinforcement learning with a reward-replay mechanism to design quantum processor topologies tailored to individual circuits (degree ≤4), then uses force-directed grid layout to make them manufacturable, reducing mapped circuit depth versus Qiskit/SABRE.
Problem
Qubit mapping for quantum circuits is constrained to fixed processor topologies, and the topology itself strongly affects mapped circuit depth; the paper asks whether tailoring the topology to each circuit reduces depth on noisy processors.
Approach
Treat the qubit coupling graph as the RL environment: an agent (PPO) incrementally adds edges between qubit pairs under a degree-4 constraint, rewarded by mapped-circuit depth change via Qiskit; a reward-replay memory caches action-level rewards to cut costly compilation calls, and a force-directed grid layout makes the resulting topology grid-like with fewer crossings.
Key results
- Qtailor reduces mapped circuit depth by 5% to 46% relative to Qiskit/SABRE, with over 60% of samples showing at least 20% reduction.
- On 6-qubit circuits scaling from 60 to 210 gates, depth reduction grows from 31.03% to 40.74%.
- Reward-replay reduces the sampling cost and PPO convergence is reported 24% faster; the paper also reports gate reduction of 4.78% to 36.39%.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Qtailor's circuit-specific topology instead of Qiskit's fixed grid topology | Quantum circuit mapping · MQT Bench circuits | depth: 5% to 46% | Helped |
| Use Qtailor's circuit-specific topology instead of Qiskit's fixed grid topology | Quantum circuit mapping · MQT Bench circuits | total gates: 4.78% to 36.39% | Helped |
| Optimize the topology to reduce gate count with Qtailor | Quantum circuit execution · 10-qubit circuits | fidelity: 10.91% | Helped |
| Use Qtailor's suggested topology instead of the fixed grid | Quantum circuit mapping with Tket · Circuits evaluated with Tket | depth: 39% | Helped |
| Increase circuit gate count while using Qtailor's topology | Qtailor · 6-qubit circuits, 60 to 210 gates | depth reduction versus Qiskit: 31.03%, 31.52%, 33.87% , 39.11%, 39.65% and 40.74% | Helped |
| Add reward replay to PPO (ablation) | PPO · Topology-design training | training convergence time: 24% | Helped |
| Add reward replay to PPO (ablation) | PPO · Topology-design training | loss | No clear effect |
Surprising findings
- Minor edge differences between non-isomorphic graphs change circuit depth substantially (e.g., 211 vs 137 for QFT with one added edge).
Limitations
- Only connectivity constraints are modeled; other physical constraints (different fabrication, quantum systems) are not considered.
- Reward replay approximates r(s,a) by r(a), which may introduce inaccuracy.
- Evaluation uses noiseless or simulated depth and fidelity rather than real hardware results.
Experimental setup
Datasets: MQT Bench
Benchmarks: MQT Bench circuits (qnn, vqe, ae, ansatz, su2, QFT)
Metrics: circuit depth idle ratio total gate count fidelity
Baselines: Qiskit (SABRE) Tket
Builds on: PPO SABRE force-directed layout MQT Bench
Closest papers by meaning
- TensorRL-QAS: Reinforcement learning with tensor networks for improved quantum architecture search (NeurIPS 2025 · similarity 0.89)
- Curriculum reinforcement learning for quantum architecture search under hardware errors (ICLR 2024 · similarity 0.88)
- QCircuitBench: A Large-Scale Dataset for Benchmarking Quantum Algorithm Design (NeurIPS 2025 · similarity 0.87)
- Equivariant Reinforcement Learning for Clifford Quantum Circuit Synthesis (NeurIPS 2026 · similarity 0.86)
- Replay-buffer engineering for noise-robust quantum circuit optimization (NeurIPS 2026 · similarity 0.86)
- MG-Net: Learn to Customize QAOA with Circuit Depth Awareness (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.