Automating quantum dot tuning for future quantum computers

Using a machine learning approach, the research team effectively demonstrated automated charge state detection in quantum dot devices, a major step towards automating the setup and tuning of quantum bits (qubits) for quantum information processing. APL Machine Learning This study was recently published.

Semiconductor qubits are quantum bits made from semiconductor materials. These materials are often used in traditional electronics because they can be integrated with regular semiconductor technology. Because of their compatibility, researchers believe they are promising candidates for future qubits to create quantum computers.

The fundamental unit of data in semiconductor spin qubits is the spin state of an electron trapped in a quantum dot, or qubit. Human experts must tune several factors, including gate voltages, to generate these qubit states.

However, because there are so many parameters, tuning becomes more difficult as the number of quantum bits increases, posing a problem in realizing large-scale computers.

To overcome this, we developed a means to automate the estimation of the charge state of double quantum dots, which are crucial for creating spin qubits, in which each quantum dot accommodates one electron..

Tomohiro Otsuka, Associate Professor, Advanced Institute for Materials Research, Tohoku University

Otsuka and colleagues used the charge sensors to generate a charge stability diagram showing the optimal combination of gate voltages to ensure exactly one electron per dot. To automate this tuning procedure, they needed an estimator that could classify charge states according to the differences in the charge transition lines in the stability diagram.

A lightweight simulation model, the Constant Interaction model (CI model), was used to prepare the data to train a convolutional neural network (CNN) to produce this estimate. The preprocessing method improved the simplicity and robustness of the data to noise, maximizing the ability of the CNN to properly classify charge states.

When tested with experimental data, the estimator estimated most charge states well, but had a high error rate for certain states. To address this, the researchers used Grad-CAM visualization to find patterns in the estimator's decision-making.

The researchers found that chance connections to noise were mistakenly believed to be charge transition lines, frequently causing errors. By modifying the training data and optimizing the estimator structure, the researchers improved accuracy for previously error-prone charge states while maintaining high performance for other charge states.

Using this estimator, we can automatically tune the parameters of semiconductor spin qubits, which is necessary to scale up quantum computers. Furthermore, we demonstrate that visualization of decision criteria, which have been black boxes until now, can serve as a guideline for improving the performance of the estimator..

Tomohiro Otsuka, Associate Professor, Advanced Institute for Materials Research, Tohoku University

Journal References:

Hiroyuki Muto others(2024) A visual illustration of a machine learning model for estimating quantum dot charge states. APL Machine Learning.doi.org/10.1063/5.0193621.