Automated Tuning and Optimal Control of Spin Qubits in Quantum Dot Devices

Abstract

Silicon quantum dots present a promising foundation for realizing scalable quantum processors, leveraging the advantages of a mature semiconductor industry. Two significant challenges hinder their development: the laborious tuning of these devices and the coherent control of their spin qubits. This thesis presents contributions towards addressing these challenges by harnessing physics-informed machine learning.

Tuning these devices involves navigating complex parameter spaces, plagued with variability and fabrication imperfections, to identify optimal operating conditions. This process demands extensive time and resources by a researcher to perform large amounts of data collection and analysis. My work takes steps towards on achieving fully autonomous tuning of these devices, with the automated formation of a single quantum dot. This work involves the application of data analysis and computer vision techniques to extract relevant features from data, guiding the tuning process in real-time. This tool allows single quantum dots to be formed autonomously, freeing researchers to focus on investigating the physics of the device. Progress in multi-dot systems was also made by developing a data segmentation model that successfully identifies and segments charge and dot configurations in charge stability diagram data. This enables rapid data analysis to determine optimal voltage settings for achieving the desired device state.

Optimal control is crucial for guiding quantum systems through unitary operations while minimizing decoherence. Using a simulated open quantum system Hamiltonian for spin qubits, I developed a protocol to optimize experimental control signals, allowing for the implementation of unitary gate operations with arbitrary fidelity. The protocol designed experimental pulses for single-qubit rotations and entangling gates in a two-qubit system, achieving fidelities above the error correction threshold. Additionally, it utilizes modern machine learning frameworks, making it scalable to multi-qubit systems.

The work presented in this thesis serves as an important foundation for future advancements in our research group.