Codesign of Quantum Software and Hardware: Towards Scalable and Robust Quantum Computing Systems

Quantum computing is becoming a hot topic and is considered one of the most promising new computing paradigms, especially with its enormous potential in fields such as machine learning, finance, cryptography, and chemistry. However, current quantum computing is still in a very immature state, with significant challenges at every level from software to hardware. This dissertation explores how to help build scalable and robust quantum computing systems from the perspective of software and hardware co-design. Specifically, the dissertation is divided into two main directions: 1) Paradigm shift from gate-level to pulse-level and cross-layer co-design; and 2) Establishing a deeper level of classical-quantum cooperation.
In the first direction, this dissertation discussed certain sophisticated quantum operations that may derive substantial benefits from circumventing the conventional decomposition into basic gates at the circuit level. Instead, these operations can be more effectively implemented directly at the physical layer. The advent of applications such as quantum simulations and quantum machine learning indicate that the classic "gate-to-circuit-to-program" paradigm may no longer serve as the most efficient or intuitive approach for quantum design. Exploring designs at the pulse level, as opposed to the gate level, could offer significant advantages. Utilizing quantum pulses over quantum gates has the potential to provide enhanced flexibility, superior fidelity, and greater scalability, along with the capacity for real-time adjustments.
In the second direction, a key focus is how to deeply promote the integration of classical machine learning or optimization algorithms with quantum algorithms. Given the current scarcity and high cost of quantum computing resources, it's challenging to conduct large-scale experiments. Therefore, we discuss two hybrid classical-quantum computing frameworks to solve this challenge: one involves sacrificing some classical computing resources to preheat quantum algorithms, and the other divide problems into parts solved by classical computers and parts solved by quantum computers.

量子计算（quantum computing）已成为当前研究热点，被视为最具前景的新型计算范式之一，尤其在机器学习、金融、密码学与化学等领域展现出巨大应用潜力。然而，当前量子计算仍处于极不成熟的阶段，从软件到硬件的各个层级均面临诸多严峻挑战。
本论文从软硬件协同设计（software and hardware co-design）的视角出发，探索如何助力构建可扩展、高鲁棒性的量子计算系统。具体而言，本文主要围绕两大研究方向展开：其一，实现从门级（gate-level）到脉冲级（pulse-level）的范式转变与跨层级协同设计（cross-layer co-design）；其二，建立更深层次的经典-量子协同（classical-quantum cooperation）机制。
在第一个研究方向中，本文针对部分复杂量子操作展开探讨——相较于传统电路级分解为基础门电路的实现方式，直接在物理层执行此类操作可获得显著性能增益。随着量子模拟、量子机器学习等应用的涌现，经典的"gate-to-circuit-to-program"范式已不再是量子设计中最高效、最直观的实现路径。相较于门级设计，脉冲级的量子电路设计能够带来诸多显著优势：采用量子脉冲而非量子门实现操作，可具备更强的灵活性、更高的保真度（fidelity）与可扩展性，同时支持实时调优。
在第二个研究方向中，本文的核心关注点在于如何深度推动经典机器学习或优化算法与量子算法的融合。鉴于当前量子计算资源稀缺且成本高昂，开展大规模实验面临诸多困难。为此，本文提出两种经典-量子混合计算框架以应对该挑战：其一，通过适度牺牲经典计算资源对量子算法进行预热；其二，将待求解问题拆解为经典计算机可处理的子任务与量子计算机可加速的子任务。