Machine Learning-Assisted Precision Manufacturing of Atom Qubits in Silicon

Area of Science:

• Quantum computing
• Materials science
• Machine learning

Background:

• Silicon donor-based qubits offer a scalable platform for quantum computing.
• Precise donor atom placement is critical for qubit performance.
• Current manufacturing methods lack real-time feedback on donor atom configuration.

Purpose of the Study:

• To develop machine learning (ML) techniques for real-time prediction of donor atom numbers during qubit fabrication.
• To enable automated manufacturing of high-fidelity quantum bits (qubits).

Main Methods:

• Utilized machine learning image recognition on scanning tunneling microscope (STM) images.
• Developed convolutional neural networks (CNNs) to predict donor atom distribution.
• Implemented techniques to mitigate overfitting, including reduced model complexity, data preprocessing, and augmentation.

Main Results:

• Achieved over 90% accuracy in predicting the donor atom number at the qubit site.
• Demonstrated consistent performance across developed ML models.
• Validated the effectiveness of ML in real-time manufacturing feedback.

Conclusions:

• Machine learning provides an accurate and automated method for controlling donor atom configurations in silicon qubits.
• This work is a significant step towards the automated fabrication of qubits for quantum computation and sensing.
• The developed ML techniques can accelerate the scaling of quantum technologies.