改進基於 ZX 圖之量子電路最佳化流程中之 雙量子位元邏輯閘萃取技術

Abstract

在將演算法轉化為實際運行量子電路的過程中量子電路最佳化扮演著不可或缺的角色。在此領域中，透過ZX圖是一種降低小角度旋轉閘的常見方法。我們的研究聚焦在ZX-calculus的萃取，亦即如何將ZX圖轉換回量子電路的過程中降低雙量子邏輯閘的數量。於降低CZ邏輯閘的部分，我們研發出一套預測系統以動態調整萃取流程，同時仰賴配對演算法挑選萃取對象。於降低CX邏輯閘，我們採用了窮舉搜索的方式尋找區域性最佳解。在QASMBench電路中研究成果降低了28%的雙量子邏輯閘數量，同時減少25%的電路深度。同時，我們在Qsyn中使用C++語言實現了所提出的方法，使得Qsyn的萃取時間僅為PyZX的1/10。本研究顯著提高了量子電路最佳化的效率和效果，使未來該領域有更好的發展空間。

Quantum circuit optimization plays an indispensable role in transforming algorithms into functional quantum circuits. Within this field, the ZX-diagram-based method is a prevalent approach for minimizing small-angle rotation gates. Our research reduces double-qubit gate count in the extractor component of ZX-calculus, which involves converting ZX-diagrams back into quantum circuits. For the CZ gates minimization, the research relies on a predictor to dynamically rearrange the steps during extraction and provides a heuristic to select candidates. With respect to CX gates, we adopt the exhaustive search method to solve the local optimal solution. Our proposed method achieves a 28% reduction in double-qubit gate count and a 25% reduction in circuit depth in QASMBench. Furthermore, we implement the proposed method in Qsyn using C++, enabling Qsyn to perform quantum circuit optimization with an extraction time that is only 1/10 that of PyZX. This work significantly advances the efficiency and effectiveness of quantum circuit optimization, paving the way for further innovations in the field.