Data-Processing and Machine-Learning Tools for Gate Virtualization of Quantum Dot Arrays

Crosstalk between gates remains a challenging obstacle to the scalable operation of quantum dot devices. This can be addressed by defining virtual gates, which employ corrections to compensate for capacitive coupling between gates. We present several computational tools to aid in the process of determining the proper virtualization coefficients by analyzing charge stability diagrams using both classical and machine-learning and computer-vision techniques. We use an ensemble of convolutional neural network pixel classifiers trained on simulated data to identify horizontal, vertical, and interdot transitions in charge stability diagrams. Taking Hough transforms of the charge-state transitions allows us to calculate their slopes and subsequently determine the required crosstalk compensation coefficients. We demonstrate these tools by virtualizing a state-of-the-art 10-dot, 2-dimensional quantum dot array in germanium.