Machine learning techniques for state recognition and auto-tuning in quantum dots
Scaling up NISQ-era quantum computers from individual qubits is a challenging control problem. Limitations arise due to imperfect fabrications leading to variability across devices as well as because most techniques for qubit initialization are semi-automated or rely on human heuristics. In the project, we seek to bridge this gap and use modern machine learning techniques to solve classical control problems with a specific qubit implementation - semiconductor quantum dots. Quantum dots are defined at the interface of semiconductor heterostructures by applying appropriate voltages to metallic gate electrodes. We investigate whether it is possible to identify quantum dot state using a convolution neural network applied to the current-gate voltage characteristics. Using such a trained network, we recast the control problem of finding a required dot state into an optimization problem and demonstrate it on simulated as well as experimental data. This work lays the groundwork for automated heuristics using machine learning for tuning quantum dot devices into appropriate regimes of operation.