Quantum circuit compilation for nearest-neighbor architecture based on reinforcement learning

To compile quantum circuits on the target processor, quantum circuits need to satisfy the nearest neighbor architecture constraint first. Therefore, quantum circuit compilation for nearest neighbor architecture is necessary for the near-term quantum compiler. In this paper, a novel compilation method based on reinforcement learning is proposed for nearest neighbor architecture of quantum circuits. Initially, quantum circuits should be decomposed based on NCV. To minimize the nearest neighbor cost of decomposed quantum circuits, a methodology to select the order of auxiliary qubits is proposed. Then, the nearest neighbor architecture mapping problem is formulated by model-based reinforcement learning. An initial mapping method of quantum lines based on quantum weights is proposed for decomposed quantum circuits, and three state space search strategies are designed to reduce state space of the mapping process. The policy iteration and value iteration algorithms are applied in the improved state space to obtain the optimal mapping of quantum circuits for nearest neighbor architecture. The proposed compilation method is generic and can easily be configured for future architectures. Experiments of the proposed compilation method are performed on the benchmark dataset Revlib, and experimental results show that the proposed method outperforms other available state-of-the-art methods. For SWAP usage, the reduction of the proposed method is up to 62.5% and 66.7% compared to the state-of-the-art exact reordering and heuristic methods, respectively.