Principled approach to automatically annotating charge stability diagrams

Machine learning methods for the calibration of quantum dot (QD) devices have gained popularity in recent years [1]. However, many of the proposed methods rely on large, labeled datasets typically generated using simulations or by manually labeling experimental data. Yet, while proper training and validation require representative datasets with reliable labels, simulated data may lack important features representing real-world noise and imperfections, and human labeling of experimental data is error-prone, can’t be standardized, and scales poorly. To help overcome these limitations, recently we introduced a large dataset of simulated double QD data that incorporates synthetic noise typical of QD experiments [2]. I will present a new, robust classical algorithm for fully automated labeling of two-dimensional (2D) charge stability diagrams of double-QD devices [3]. Our method combines ray-based measurements [4] with a variety of geometric and statistical techniques to effectively deal with noise typical of real-world data and can be generalized to 2D measurements of multi-QD devices and to higher dimensional data. I will also demonstrate the performance of the algorithm on simulated and experimental data.

[1] J. Zwolak and J. Taylor. Rev. Mod. Phys. 95, 011006 (2023).

[2] J. Ziegler et al. Phys. Rev. Applied 17, 024069 (2022).

[3] B. Weber and J. Zwolak. A principled approach to automatically annotating charge stability diagrams (in preparation).

[4] J. Zwolak et al. PRX Quantum 2, 020335 (2021).