Learning the states of quantum dot systems: The ray-based approach

Given the progress in the construction of multi-quantum dot (QD) arrays in both 1D and 2D [1,2], it is imperative to replace the current practice of manual tuning to a desirable electronic configuration with a standardized and automated method. Recently, we have experimentally realized an auto-tuning paradigm proposed by Kalantre et al. [3] that combines machine learning (ML) and optimization routines, with ConvNets used to characterize the state and charge configuration of single and double QD states from measurements via the conductance of a nearby charge sensor [4]. Now we expand on this work and propose a novel approach where we use 1D traces (“rays”) measured in multiple directions in the gate voltage space to describe the position of the features characterizing each state (i.e., to “fingerprint” the state space). Using these “fingerprints” instead of 2D scans we train an ML algorithm to differentiate between various state configurations. Here, we report the performance of the ray-based learning on experimental data and compare it with our image-based approach.

[1] Zajac et al., Phys. Rev. Appl. 6, 054013 (2016).
[2] Mukhopadhyay et al. Appl. Phys. Lett. 112, 183505 (2018).
[3] Kalantre et al. npj Quantum Inf. 5, 6 (2019).
[4] Zwolak et al., Phys. Rev. Appl. 13, 034075 (2020).