Overview
This project was to integrate my quantum-circuit-optimizer with QubitPulseOpt to analyze how gate-level optimization affects pulse-level fidelity across the full compilation stack. Unlike existing work that evaluates optimization passes in isolation, this framework instead measures end-to-end fidelity from OpenQASM input to simulated pulse execution. Obviously simulated pulse execution is a big caveat here, but the pipeline structure remains regardless.
Key Research Question: Do gate-level optimization metrics (gate count, circuit depth) reliably predict pulse-level fidelity? Our analysis revealed that pulse duration rather than gate count is the strongest fidelity indicator (r=-0.74). (Somewhat expected but also interesting)
By the numbers
| Metric | Value |
|---|---|
| Circuits analyzed | 371 |
| Mean gate reduction | 23.1% |
| Max gate reduction | 96.2% |
| Mean process fidelity | 0.680 ± 0.224 |
| Hardware validation | 20 Qubits (IQM Garnet) |
Optimization Pass Effectiveness
| Pass | Gates Removed | % Improved |
|---|---|---|
| Gate Cancellation | 14,024 | 68% |
| Rotation Merging | 6,512 | 29% |
| Identity Elimination | 55 | 9% |
| Commutation | 0 (enables others) | 0% |
Fidelity Correlations
| Parameter | Pearson r | R² |
|---|---|---|
| Pulse duration | -0.743 | 0.553 |
| Input gates | -0.606 | 0.368 |
| Input depth | -0.585 | 0.342 |
| Input qubits | -0.569 | 0.324 |
Architecture
- Parse & Validate: Extract initial metrics (gates, depth, qubits)
- Optimize (C++): Apply configurable pass sequence, track per-pass gate changes
- SABRE Routing: Map to hardware topology, insert SWAP gates
- Pulse Compilation: Native gate decomposition, pulse schedule generation
- T₁/T₂ decoherence modeling, process and state fidelity computation
Hardware Validation
Validated simulation results on the IQM Resonance Garnet 20-qubit superconducting processor:
| Circuit | Gates (Orig → Opt) | Reduction | Fidelity (Orig) | Fidelity (Opt) |
|---|---|---|---|---|
| GHZ 4q | 4 → 4 | 0% | 0.494 | 0.469 |
| GHZ 8q | 8 → 8 | 0% | 0.406 | 0.375 |
| GHZ 12q | 12 → 12 | 0% | 0.256 | 0.288 (+12%) |
| QFT 4q | 30 → 9 | 70% | 0.100 | 0.088 |
Key Findings:
- Optimizer correctly identifies GHZ circuits as already minimal (0% reduction) (a good negative result!)
- QFT circuits achieve 70% gate reduction with significant rotation merging
- 12-qubit GHZ showed 12% fidelity improvement after optimization
Benchmark Compilation
| Circuit Type | Qubit Range | Description |
|---|---|---|
| GHZ states | 2–12 | Entanglement preparation |
| QFT | 2–8 | Quantum Fourier Transform |
| QAOA | configurable | MaxCut optimization |
| Random | 4–8 qubits, depth 5–30 | Stress testing |
Research Impact
This work provides actionable guidance for quantum compiler design:
- Prioritize cancellation: Gate cancellation provides the largest fidelity gains (68% improvement rate)
- Commutation enables cancellation: While commutation provides no direct reduction, it creates opportunities for subsequent passes
- Minimize pulse duration The strong correlation (r=-0.74) emphasizes decoherence-aware optimization over pure gate count reduction
- Optimal pass sequence:
cancel → commute → rotate
While none of these are groundbreaking results, and possibly even seem trivial to some, it can be easy to miss the forest for the trees when writing complex compiler tools. They are just grounded metrics backed by my results.