Machine Learning as an Enabler for Qubit Scalability, Quantum Computing

机器学习作为量子位可扩展性和量子计算的推动者

• 批准号: 2891528
• 金额: --
• 依托单位: University of Oxford
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2891528
• 关键词: Machine; Learning; Enabler; Qubit; Scalability

Quantum computing, a relatively contemporary field of study, boasts of abundant potential applications. These extend across broad sectors such as drug discovery for curative and preventative medical solutions, to cryptography for bolstering security in the digital age. The physical implementations supporting these applications are equally diverse, including systems utilising trapped ions, spin qubits situated inside quantum dots, and superconducting qubits that leverage circuit quantum electrodynamics.Despite the diverse nature of these implementations, they are uniformly affected by the same set of challenges, particularly in relation to coherence times, gate speeds and scalability in their operations. Optimal performance demands fast gate speeds partnered with long coherence times. Single qubit systems that display an ideal balance between gate speed and coherence time have been created in each of the aforementioned implementations. On the flip side, these parameters have often been realised through the manual fine-tuning and characterisation undertaken by human operators. This approach, while effective, is both resource-intensive and arduously unsuitable for large scale operations, particularly those demanding millions of qubits to delineate any profound quantum advantage.It is against this backdrop that this project proposes to automate the tuning and calibration process. This process, typically referred to as quantum control, will be guided by the principles of machine learning. The specific machine learning method that will be in focus for this project is Bayesian Optimisation. The project is poised to utilise this technique for the development of a broad-spectrum tuning algorithm for quantum devices - a domain severely underexplored in scientific literature.Bayesian optimisation operates within a probabilistic framework capable of detailing uncertainty. This feature can be used to inform decision-making algorithms. The method holds distinct advantages over alternative methods - it enables easier attribution of cause-and-effect relationships to experimental data, as compared to classical machine learning algorithms and neural networks. Additional benefits lie in its greater generalisability to devices that are yet to be seen or experienced and often results in quicker predictive capabilities.This project will specifically focus on semiconductor devices, like the complementary metal-oxide-semiconductor (CMOS) and superconducting circuits formed from Josephson Junctions. These devices possess significant physical similarity, making them suitable candidates for joint investigation. Access to these devices will be made possible via partnerships with IST Austria, University of Basel, University of Chalmers and the company QuantWare.Adding another layer of support to the project is our collaboration with QuantrolOx, a company specialising in the creation of control software for superconducting qubits via machine learning. They will lend their expertise in machine learning techniques and instrument integration, wherever necessary, providing requisite assistance. The principal goal is the development of Bayesian optimisation algorithms customised for tuning semiconductor devices. This will enable us to obtain state-of-the-art performance for these devices, while simultaneously reducing the time spent on fine-tuning each device. This project is well-aligned with the EPSRC Quantum technologies research area.

量子计算是一个相对现代的研究领域，拥有丰富的潜在应用。这些涉及广泛的领域，例如用于治疗和预防性医疗解决方案的药物发现，以及用于增强数字时代安全性的密码学。支持这些应用的物理实现同样多种多样，包括利用捕获离子、位于量子点内部的自旋量子位以及利用电路量子电动力学的超导量子位的系统。尽管这些实现的性质各不相同，但它们都受到相同挑战的影响，特别是与操作中的相干时间、门速度和可扩展性有关。最佳性能需要快速的门速度和较长的相干时间。在上述每个实现中都创建了在门速度和相干时间之间显示理想平衡的单量子位系统。另一方面，这些参数通常是通过人类操作员进行的手动微调和表征来实现的。这种方法虽然有效，但却是资源密集型的，并且非常不适合大规模操作，特别是那些需要数百万量子位来描绘任何深刻的量子优势的操作。正是在这种背景下，该项目提出自动化调谐和校准过程。这个过程通常被称为量子控制，将遵循机器学习原理。该项目重点关注的具体机器学习方法是贝叶斯优化。该项目准备利用这项技术来开发量子设备的广谱调谐算法，这是科学文献中尚未充分探索的领域。贝叶斯优化在能够详细描述不确定性的概率框架内运行。此功能可用于为决策算法提供信息。该方法比其他方法具有明显的优势——与经典的机器学习算法和神经网络相比，它可以更容易地将因果关系归因于实验数据。额外的好处在于它对尚未见过或经历过的设备具有更大的通用性，并且通常会带来更快的预测能力。该项目将特别关注半导体设备，例如互补金属氧化物半导体（CMOS）和由其形成的超导电路约瑟夫森枢纽。这些设备具有显着的物理相似性，使它们成为联合调查的合适候选者。通过与 IST Austria、巴塞尔大学、查尔姆斯大学和 QuantWare 公司的合作，可以访问这些设备。我们与 QuanttrolOx 的合作为该项目增添了另一层支持，QuanttrolOx 是一家专门为以下领域创建控制软件的公司：通过机器学习超导量子位。他们将在必要时提供机器学习技术和仪器集成方面的专业知识，提供必要的帮助。主要目标是开发为调整半导体器件而定制的贝叶斯优化算法。这将使我们能够获得这些设备最先进的性能，同时减少微调每个设备所花费的时间。该项目与 EPSRC 量子技术研究领域非常契合。