A team of researchers has used machine learning approaches to effectively demonstrate automatic charge state detection in quantum dot devices. This is a major step towards automating the setup and tuning of quantum bits (qubits) for quantum information processing. The journal APL Machine Learning published this study recently.
Semiconductor qubits are quantum bits made of semiconductor materials. Because these materials can be integrated with regular semiconductor technology, they are often used in traditional electronics. Because of their compatibility, researchers believe they are promising candidates for qubits of the future in the effort 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 adjust a number of factors, including gate voltage, to produce these qubit states.
However, because there are too many parameters, tuning gets increasingly difficult as the number of qubits increases. This becomes troublesome when it comes to realizing large-scale computers.
To overcome this, we developed a means of automating the estimation of charge states in double quantum dots, crucial for creating spin qubits where each quantum dot houses one electron.
Tomohiro Otsuka, Associate Professor, Advanced Institute for Materials Research, Tohoku University
Otsuka and colleagues used charge sensors to generate charge stability diagrams that showed the best gate voltage combinations to guarantee exactly one electron per dot. To automate this tuning procedure, an estimator that could categorize charge states according to differences in charge transition lines in the stability diagram was necessary.
Image Credit: atdigit/Shutterstock.com
The Constant Interaction model (CI model), a lightweight simulation model, was used to prepare the data that the convolutional neural network (CNN) trained on to create this estimator. Pre-processing methods improved the data's robustness to noise and simplicity, which maximized the CNN's capacity to categorize charge states properly.
When tested using experimental data, the estimator performed well in estimating the majority of charge states; nevertheless, certain states showed greater error rates. To solve this, the researchers used Grad-CAM visualization to find decision-making patterns in the estimator.
They discovered that coincidental-connected noise mistakenly believed to be charge transition lines frequently caused errors. By modifying the training data and optimizing the estimator's structure, the researchers increased the accuracy for previously error-prone charge states while maintaining high performance for other charge states.
Utilizing this estimator means that parameters for semiconductor spin qubits can be automatically tuned, something necessary if we are to scale up quantum computers. Additionally, by visualizing the previously black-boxed decision basis, we have demonstrated that it can serve as a guideline for improving the estimator's performance.
Tomohiro Otsuka, Associate Professor, Advanced Institute for Materials Research, Tohoku University
Journal Reference:
Muto, Y., et al. (2024) Visual explanations of machine learning model estimating charge states in quantum dots. APL Machine Learning. doi.org/10.1063/5.0193621.