Quokka: 基於檔案之量子計算模擬器

Abstract

隨著量子運算發展出解決各式應用的方法與軟硬體快速地發展, 許多具有高度平行化的量子模擬框架陸續被推出。 量子模擬需要大量的記憶體和計算能力，尤其是當需要模擬的qubit數目非常大時。為了解決這個問題，基於多節點的解決方案被推出以應對大qubit的案例。 然而多節點的模擬意味著使用者除了應付記憶體增加的費用外，還要額外對計算單元進行付費。 這樣的惡性循環使得量子電腦不能被普及到普羅大眾。 在此篇論文，我們提出一個新穎的解決方案，Quokka，緩和上述的額外成本。 Quokka在軟體層面以資料流的角度下去實作，在硬體層面可透過ＳＳＤ這個特化的硬體來達到最佳的運算效能。 由於資料串流需要更多的調參，我們定義了針對不同硬體做調整的格式並給出如何在特定硬體下做出各種適配的調整。 在我們的實驗中，我們能達到memory-based 版本的0.8倍，並在價格中省下了80 倍預算，這是之前的論文所媲美不了的全新研究。

With the growing demand for quantum computing in a variety of applications, numerous highly parallel quantum circuit simulation (QCS) frameworks have been introduced. These frameworks utilize multiple machines to provide the memory capacity required for large-scale QCS. Unfortunately, such approaches are expensive and inefficient as the performance is bounded by the bandwidth of the interconnection network. In this thesis, we develop a storage-based quantum circuit simulator is proposed to provide a cost-effective alternative to existing memory-based schemes. To efficiently utilize multicore CPUs and solid-state disk arrays, a qubit representation is developed, together with a threading model, to enable efficient simulation of quantum circuits of different qubit sizes with parameterized configurations. Moreover, several performance issues are addressed by proposed strategies. The experimental results show that the proposed storage-based simulation provides a platform-aware approach to increase the qubit size supported by the simulator. The size of quantum circuit is bounded by the volume sizes of the adopted SSDs and the resulted performance is limited by the maximum bandwidth of the storage system. Thanks to the combination of high capacity, good data bandwidth, and low cost of SSDs, as well as the expandability of I/O devices, our approach provides a cost-effective and flexible solution for large-scale quantum circuit simulation. As a result, the proposed approach reduced costs by 80 times compared to memory-based version.