QFlow lite dataset: A machine-learning approach to the charge states in quantum dot experiments

Area of Science:

• Quantum Computing
• Materials Science
• Machine Learning

Background:

• Semiconductor quantum dots (QDs) are crucial for quantum computing, but their tuning is experimentally challenging.
• Current heuristic methods for tuning QD devices do not scale with increasing array sizes.
• A reliable, non-heuristic protocol for QD device tuning is essential for advancing quantum computing.

Purpose of the Study:

• To develop and validate a machine learning-based approach for tuning semiconductor quantum dot devices.
• To create a comprehensive dataset of simulated QD device characteristics for training machine learning models.
• To provide a new tool, QFlow lite, for researchers to utilize this machine learning approach.

Main Methods:

• Generated a dataset of simulated QD device characteristics (conductance, charge sensor response) against electrostatic gate voltages.
• Trained convolutional neural networks using this dataset to recognize QD device states.
• Developed QFlow lite, a Python-based software suite for training neural networks on the dataset.

Main Results:

• Machine learning models achieved ≈96.5% accuracy in recognizing QD device states.
• Accuracy spread was 0.5% for current-based and 1.8% for charge-sensor-based training data.
• The QFlow lite tool is available for researchers to apply and further develop machine learning in QD device tuning.

Conclusions:

• Machine learning offers a highly accurate and scalable method for tuning semiconductor quantum dot devices.
• The developed dataset and QFlow lite tool will accelerate research in quantum computing and related fields.
• This work establishes a definitive reference for a new dataset crucial for advancing machine learning applications in experimental science.