Spin qubit tuning using machine learning

自旋量子比特是基于电子自旋的量子比特，有望实现基于当前集成电路技术的大型量子电路。要实现这些量子比特的全部潜力，必须找到最佳的操作参数。我们现在正在达到我们手动完成这一任务的能力极限。该项目可能允许全自动搜索最佳量子比特操作参数，谈论量子技术可伸缩性的瓶颈。目标是实现自旋量子比特操作参数的全自动优化。这将使我们能够同时优化多个量子比特，从而允许实现复杂的量子电路。为此，学生将开发机器学习技术，如贝叶斯优化和推理。通过这种方式，我们预计该算法将比人类在量子比特优化方面的专家更快、更好。单个量子比特的参数空间是多维的。为了有效地探索整个参数空间，将建立考虑开采与勘探权衡的算法。量子比特将在低温下使用专门的低噪音电子设备进行测量。量子比特将在半导体中以静电方式定义，如硅和砷化镓。机器学习在最近的突破中得到了应用，例如一种算法战胜了围棋世界冠军，以及超人人脸识别。显然，需要机器学习方法来释放量子比特技术的潜力。机器学习控制量子比特实验的全部潜力尚未被探索，而使用机器学习技术来推进量子实验已经被证明是制造和表征量子设备的关键。我们在澳大利亚IST的合作者为我们提供了所需的量子比特，与牛津大学工程系的合作使我们能够获得机器学习的最新技术，特别是贝叶斯优化和其他无法获得大量数据的方法。该项目通过允许快速和自动优化量子比特参数，从而使优化复杂的量子电路成为可能，对推动量子技术的进步非常重要。过去几年出现的新的机器学习方法可能会让我们解决这一挑战。该项目属于EPSRC量子技术研究领域。

Spin qubits, quantum bits based on electron spins, are promising for the realization of large quantum circuits based on current integrated circuit technologies. To realise the full potential of these qubits, the optimal operation parameters have to be found. We are now reaching the limits of our ability to do this manually. The project might allow for the fully automated search of optimal qubit operation parameters, talking the bottleneck of scalability in quantum technologies.The objective is to achieve fully-automatic optimisation of the operation parameters of a spin qubit. This would enable us to optimise many qubits at once and thus allow for the realization of complex quantum circuits. For this, the student will develop machine learning techniques such as Bayesian optimisation and inference. In this way, we expect that algorithm would be faster and better than human experts at qubit optimisation. The parameter space of a single qubit is multidimensional. To efficiently explore the entire parameter space, algorithms that consider exploitation versus exploration trade-offs will be established. The qubits will be measured at cryogenic temperatures with dedicated low-noise electronics. The qubits will be electrostatically defined in semiconductors, such as silicon and gallium arsenide. Machine learning has been used in recent breakthroughs, such as the victory of an algorithm over a Go world champion and super-human face recognition. There is a clear need for machine learning approaches to unleash the potential of qubit technologies. The full potential of machine learning to control qubit experiments has not yet been explored, while the use of machine learning techniques to advance quantum experiments is already proving essential in the fabrication and characterisation of quantum devices. Our collaborators in IST Austia provides us with the qubits we require, and a collaboration with the Department of Engineering at University of Oxford allows us to have access to the latest techniques in machine learning, in particular Bayesian optimisation and other approaches for which large amounts of data are not available. This project is important to advance quantum technologies by allowing fast and automatic optimisation of qubit parameters, thus enabling the possibility to optimise complex quantum circuits. New machine learning approaches that have become available in the last few years might allow us to tackle this challenge. This project falls within the EPSRC Quantum technologies research area.