Quantum Compiling with Reinforcement Learning on a Superconducting Processor

Effectively implementing quantum algorithms on noisy intermediate-scale quantum (NISQ) processors is a central task in modern quantum technology. NISQ processors feature tens to a few hundreds of noisy qubits with limited coherence times and gate operations with errors, so NISQ algorithms naturally require employing circuits of short lengths via quantum compilation. Here, we evaluate a reinforcement learning (RL)-based quantum compiler on a superconducting processor. Our experiments reveal that for two-qubit circuits, the RL-based compiler surpasses conventional methods, demonstrating its ability to discover hardware-amenable circuits with near-optimal lengths. However, for three-qubit circuits, the RL-based compiler does not achieve unity theoretical fidelity. To address this limitation, we integrate a variational strategy with the RL-based compiler, highlighting their complementary strengths. Systematic experiments show that this variational RL-based compiler consistently identifies near-optimal circuits, even under stringent hardware constraints, outperforming conventional techniques. Furthermore, we analyze the impact of decoherence and gate errors, providing critical insights into the practical performance of RL-based compilers on quantum hardware. These findings exemplify the codesign of the software with hardware for efficient quantum compilation, offering valuable insights for the advancement of RL-based compilers.