Machine Learning-Based control of 2D Arrays of Quantum Dots

Recent advances towards employing quantum dots (QD) as a platform for quantum computation
and simulation have shown promising results [1]. However, as the control parameters space
grows significantly with increasing number of QDs, working with large QD arrays is challenging.
Thus, finding a scalable and non-heuristic control approach to tune the electronic configuration
in QDs is necessary. Due to high-dimensional patterns defining dot states, machine learning
(ML) algorithms present a natural solution.
In this project, we extend a recent proposal [3,4] of employing a ML-based auto-tuner to linear
QD devices to the more general case of 2D arrays. We use a Thomas-Fermi solver to establish
an ensemble of simulated measurements for 2x2 QD arrays. This data set allows us to train and
evaluate an image-based classifier that maps the charge stability diagrams showing the
electronic configuration of the QD device into classes defining the number of dots formed in the
system. This work will set foundations for research on machine learning-based control of 2D QD
devices.
[1] Hensgens et al., Nature 548, 70 (2017).
[3] Kalantre et al., npj Quantum Inf. 5: 6 (2019).
[4] Zwolak et al., arXiv:1909.08030 (2019).