| name | neural-inverse-design-srf-cavity |
| description | Deep neural network approaches for inverse design of superconducting radio-frequency (SRF) cavities and transmon qubits for bosonic quantum computation — mapping target device parameters to candidate geometries. |
| category | quantum-computing |
| trigger_words | ["inverse design quantum","SRF cavity","transmon qubit","bosonic quantum computation","neural network device design","electromagnetic optimization","qubit-cavity coupling"] |
Neural-Network Inverse Design of SRF Cavities for Bosonic Quantum Computation
Paper: arXiv:2607.02289v1
Authors: Joseph Yaker, Jovan Markovic, Alessandro Reineri, Doga Murat Kurkcuoglu, Silvia Zorzetti
Core Insight
Two deep neural network approaches solve the inverse design problem for SRF cavity-transmon systems: one proposes cavity geometries for target observables, another proposes transmon designs for target qubit-cavity parameters (g, ν_q, α).
Key Results
- 5% Accuracy: Recovered cavity designs match targets within ~5%
- 2% Accuracy: Transmon designs match targets within ~2%
- Fast Alternative: Maps desired behavior directly to candidate geometries
- One-to-Many: Addresses the inverse-design challenge where multiple geometries can produce same observables
Two-Level Design Stack
Level 1: SRF Cavity Geometry
- Input: Target cavity observables
- Output: Candidate cavity geometries
- Verified by end-to-end re-simulation
Level 2: Transmon Qubit Design
- Input: Target coupling rate (g), qubit frequency (ν_q), anharmonicity (α)
- Output: Transmon geometries + positions within cavity field
- Sensitively depends on both geometry and field position
Applications
- Bosonic Quantum Computing: Long-lived electromagnetic modes for quantum information
- Device Scaling: Fast design iteration for growing parameter spaces
- Quantum Architecture: Coupling nonlinear elements to cavity modes
