|Auto-tuning of double dot devices in situ with machine learning
|Year of Publication
|Zwolak, JP, McJunkin, T, Kalantre, SS, Dodson, JP, MacQuarrie, ER, Savage, DE, Lagally, MG, Coppersmith, SN, Eriksson, MA, Taylor, JM
|Phys. Rev. Applied
There are myriad quantum computing approaches, each having its own set of challenges to understand and effectively control their operation. Electrons confined in arrays of semiconductor nanostructures, called quantum dots (QDs), is one such approach. The easy access to control parameters, fast measurements, long qubit lifetimes, and the potential for scalability make QDs especially attractive. However, as the size of the QD array grows, so does the number of parameters needed for control and thus the tuning complexity. The current practice of manually tuning the qubits is a relatively time-consuming procedure and is inherently impractical for scaling up and applications. In this work, we report on the in situ implementation of an auto-tuning protocol proposed by Kalantre et al. [arXiv:1712.04914]. In particular, we discuss how to establish a seamless communication protocol between a machine learning (ML)-based auto-tuner and the experimental apparatus. We then show that a ML algorithm trained exclusively on synthetic data coming from a physical model to quantitatively classify the state of the QD device, combined with an optimization routine, can be used to replace manual tuning of gate voltages in devices. A success rate of over 85 % is determined for tuning to a double quantum dot regime when at least one of the plunger gates is initiated sufficiently close to the desired state. Modifications to the training network, fitness function, and optimizer are discussed as a path towards further improvement in the success rate when starting both near and far detuned from the target double dot range.