最佳化計算叢集中的量子電路分散式模擬

Abstract

隨著計算需求的增加，傳統的處理器或許有一天不再能滿足人們的需求。而量子電腦作為一種新的計算平台，擁有在處理複雜問題上的巨大潛力。然而，目前量子電腦的位元數受限，無法快速的發展演算法成為眼下的問題。模擬器在此時扮演著重要的角色，使得研究者可以方便 dubug 與評估量子電路，成為進行量子相關研究重要的工具。 本研究在以記憶體或檔案方式儲存完整量子狀態的方法之上，研發分散式的量子模擬器。我們使用多節點拓展更多可模擬之量子位元數量並降低所需模擬電路的時間。就我所知，這是目前僅有可同時支援這兩種方式的模擬器。而在分散式量子模擬器當中，如何減少節點之間資料交換的成本是最大的挑戰。為了降低溝通與本機操作所需時間，我們透過實作有效率的資料交換方式並且避免掉不必要的讀寫以及改善平行化。此外，透過 gate blocking (GB) 的方法與特殊的 Swap操作，我們可以大量減少 I/O 與溝通次數，整體相較於 QuEST可以達到接近三倍的速度提升。

As computational demands grow, traditional processors may one day struggle to meet people's needs. Quantum computers, as a new computing platform, hold tremendous potential for solving complex problems. However, the current limitation on the number of qubits in quantum computers hampers rapid algorithm development, presenting a significant challenge. Simulators play a vital role at this juncture, enabling researchers to conveniently debug and evaluate quantum circuits, serving as essential tools for quantum-related studies. In this study, we develop a distributed quantum simulator built upon methods for storing complete quantum states in memory or files. We employ multi-node architectures to expand the number of qubits that can be simulated and reduce the simulation time for complex circuits. To our knowledge, this simulator is the only one that can simultaneously support both storage methods. In distributed quantum simulation, minimizing the cost of data exchange between nodes is a major challenge. To reduce the time required for communication and local operations, we implement efficient data exchange methods, avoid unnecessary read-write operations, and improve parallelization. Additionally, through the use of gate blocking (GB) techniques and specialized Swap operations, we can significantly reduce I/O and communication overheads, achieving nearly a three-fold speed improvement compared to QuEST.