Limbo: A Multi-Chip-First Quantum Compiler Unifying Hypergraph Partitioning with Topological Cross-Chip Routing

We present Limbo, a quantum multi-chip-first compiler that combines graph-partitioning and routing-first approaches to compilation. On a 24-cell benchmark against TKET, Limbo outperforms in every cell. On the hardest cell, it reduces cross-chip 2-qubit operations by 78% and compile time by a factor of 2735; the median compile-time ratio across the matrix is 119. Under a calibrated noise model whose intra-chip and cross-chip error rates draw from published experimental characterizations, Limbo's output reaches higher Hellinger fidelity than TKET on every informative cell. The pipeline first analyzes the structural topology of the input interaction graph and selects a matching cloud vendor. It then partitions the qubits across chips using multi-level hypergraph partitioning, sequences the residual cross-chip operations through an event scheduler driven by topological sorting, and transpiles each per-chip sub-circuit against the chosen device's coupling map and native gate set. In testing, we executed the pipeline end-to-end on Rigetti's Cepheus superconducting QPU and exercised the noise-aware routing path against the device's live qubit calibration data. Direct hardware validation of the multi-chip architectural claim itself remains, at this moment, gated on publicly cloud-accessible multi-chip quantum hardware. The open-source SDK is available at github.com/Jminding/Limbo-Quantum, with the compiler API (planned to be released in Fall 2026), project information, and the full benchmark matrix at limbosys.dev.