Phys. Rev. Applied 17, 024069 (2022) - Toward Robust Autotuning of Noisy Quantum dot Devices

Toward Robust Autotuning of Noisy Quantum dot Devices

Joshua Ziegler, Thomas McJunkin, E.S. Joseph, Sandesh S. Kalantre, Benjamin Harpt, D.E. Savage, M.G. Lagally, M.A. Eriksson, Jacob M. Taylor, and Justyna P. Zwolak

Phys. Rev. Applied 17, 024069 – Published 25 February 2022

Abstract

The current autotuning approaches for quantum dot (QD) devices, while showing some success, lack an assessment of data reliability. This leads to unexpected failures when noisy or otherwise low-quality data is processed by an autonomous system. In this work, we propose a framework for robust autotuning of QD devices that combines a machine learning (ML) state classifier with a data quality control module. The data quality control module acts as a “gatekeeper” system, ensuring that only reliable data are processed by the state classifier. Lower data quality results in either device recalibration or termination. To train both ML systems, we enhance the QD simulation by incorporating synthetic noise typical of QD experiments. We confirm that the inclusion of synthetic noise in the training of the state classifier significantly improves the performance, resulting in an accuracy of $95.0 (9) %$ when tested on experimental data. We then validate the functionality of the data quality control module by showing that the state classifier performance deteriorates with decreasing data quality, as expected. Our results establish a robust and flexible ML framework for autonomous tuning of noisy QD devices.

Received 30 July 2021
Revised 6 November 2021
Accepted 24 January 2022
Corrected 21 September 2022

DOI:https://doi.org/10.1103/PhysRevApplied.17.024069

Physics Subject Headings (PhySH)

Machine learning Noise Quantum computation Quantum control Quantum information architectures & platforms Quantum information with solid state qubits

Double quantum dots Semiconductor compounds

Condensed Matter, Materials & Applied PhysicsQuantum Information, Science & Technology

Corrections

21 September 2022

Correction: The previously published Figure 4 contained incorrect labels along the color bars in parts (g) and (h) and has been replaced with the correct version.

Authors & Affiliations

Joshua Ziegler ^1,*, Thomas McJunkin ^1,2, E.S. Joseph ², Sandesh S. Kalantre^3,4, Benjamin Harpt², D.E. Savage ⁵, M.G. Lagally⁵, M.A. Eriksson², Jacob M. Taylor^1,3,4, and Justyna P. Zwolak ^1,†

¹National Institute of Standards and Technology, Gaithersburg, Maryland 20899, USA
²Department of Physics, University of Wisconsin-Madison, Wisconsin 53706, USA
³Joint Quantum Institute, University of Maryland, College Park, Maryland 20742, USA
⁴Joint Center for Quantum Information and Computer Science, University of Maryland, College Park, Maryland 20742, USA
⁵Department of Materials Science and Engineering, University of Wisconsin-Madison, Wisconsin 53706, USA

^*joshua.ziegler@nist.gov
^†jpzwolak@nist.gov

Article Text (Subscription Required)

Click to Expand

References (Subscription Required)

Click to Expand

Issue

Vol. 17, Iss. 2 — February 2022

Subject Areas

Reuse & Permissions

Access Options

Article Available via CHORUS

Download Accepted Manuscript

Author publication services for translation and copyediting assistance advertisement

Images

Figure 1
Framework for QDs autotuning with data quality assessment. A false-color scanning electron micrograph of a Si/Si_xGe_1-x quadruple QD device. The gates in the upper channel (barriers $S B_{1}$ , $S B_{2}$ , and a plunger $S P_{1}$ ) are used to form a charge sensor for the QDs formed in the lower channel (using barriers $B_{1}$ , $B_{2}$ , and $B_{3}$ , and plungers $P_{1}$ and $P_{2}$ ). There are two consecutive machine learning modules guiding the autotuning system: DQC is used to determine the quality of the measured scan and DSE is used to assess the state of the device. The autotuning loop begins with the QD device shown on the left. A two-dimensional voltage sweep of two plunger gates $(V_{P_{1}}, V_{P_{2}})$ is measured by a QD charge sensor in the upper left channel. The numerical gradients of the measurements are then fed into the DQC module to determine whether the scan is suitable for classification. Depending on the returned quality class, the scan is passed to the DSE module for state assessment and optimization (the high-quality class), the device is recalibrated to improve the data quality (the moderate-quality class), or the autotuning loop is terminated (the low-quality class). Before recalibration or termination, further data analysis could be performed to better guide the recalibration.
Reuse & Permissions
Figure 2
(a) Sample simulated charge stability diagrams as a function of plunger gates with different types of noise added. Top: simulated sensor ( $S$ ) output. Bottom: gradient of sensor in the $V_{P_{1}}$ direction, $d S / d V_{P_{1}}$ . Noise magnitudes in these plots match the optimized parameters except for dot jumps (B) and the Coulomb peak (C) which are exaggerated for visibility. (b) Box plot showing the performance of DSE classifiers on experimental holdout dataset for models trained on: simulated noiseless data without (A) and with preprocessing ( $A_{proc}$ ), simulated data with each noise type incorporated (one at a time; plots B through F), and the optimized combination of noises (dot jumps, sensor jumps, $1 / f$ , and white noise; plot G). Each box plot depicts the distribution of the performance from 20 models. While $1 / f$ noise (D), white noise (E), and sensor jumps (F) each lead to significant improvement over the model trained with noiseless simulated data (A), the optimized noise combination (G) provides a large reduction in variability as well as a significant boost in accuracy. Optimization of the DSE model architecture further improves the performance ( $G_{opt}$ ).
Reuse & Permissions
Figure 3
(a) Box plots of model accuracy for each assigned quality class for the experimental data. Inset: box plots of the mean absolute error (MAE) for each quality class. (b) Example data and predictions of both the simplistic (i.e., trained on noiseless simulated data) and robust (i.e., trained on noisy simulated data) models. Raw sensor data (left), gradient data (middle), and predictions (right). We show a high-quality DD example, a moderate-quality CD example, and a low-quality CD example. For the bar plot, we include the full prediction vector for the simplistic and robust models, as well as the assigned label for each image.
Reuse & Permissions
Figure 4
(a),(b) Full charge stability diagrams of two double QD devices. In (a), a few characteristic noises can be seen: minor $1 / f$ or white noise is seen in the speckling throughout, jumps in the transition due to slow tunnel rates at the bottom of the image, and smearing of the transitions near the top of the image due to fast tunnel rates. Visualization of the prediction of an average simplistic state classifier (i.e., model trained on noiseless simulated data) (c),(d), and the optimized robust state classifier (i.e., model trained on noise-augmented simulations) (e),(f). The color at each point is the average of the color of each state weighted by the model’s prediction. Hue is averaged by angle in hue space, e.g., blue and green are averaged to teal. (g),(h) Visualization of the predictions of the DQC module.
Reuse & Permissions
Figure 5
Top row: plots of the MAE of the DSE used to set noise thresholds versus the simulated noise level. The scatter plot is colored by the predicted state. Bottom row: the solid lines show the means of the MAE at each noise level. The dashed lines illustrate the $2.5 %$ and $50 %$ MAE levels used to set the thresholds for the DQC module.
Reuse & Permissions
Figure 6
(a) An image from the experimental dataset. (b) First layer outputs (after activation) of a robust ( $G_{opt}$ ) model. (c),(d) First layer outputs of two different simplistic (A) models. The robust model is nearly perfectly correct, one of the simplistic models is somewhat correct, and the other simplistic model is strongly incorrect.
Reuse & Permissions

Physical Review Applied

Toward Robust Autotuning of Noisy Quantum dot Devices

Joshua Ziegler, Thomas McJunkin, E.S. Joseph, Sandesh S. Kalantre, Benjamin Harpt, D.E. Savage, M.G. Lagally, M.A. Eriksson, Jacob M. Taylor, and Justyna P. Zwolak

Phys. Rev. Applied 17, 024069 – Published 25 February 2022

Abstract

Physics Subject Headings (PhySH)

Corrections

Authors & Affiliations

Article Text (Subscription Required)

References (Subscription Required)

Issue

Subject Areas

Access Options

Authorization Required

Other Options

Download & Share

Images

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Log In

Search

Article Lookup

Paste a citation or DOI

Enter a citation