Automation of Quantum Dot Measurement Analysis via Explainable Machine Learning

The rapid advancement of quantum dot (QD) devices for quantum computing has created a need for more efficient, automated device characterization and tuning methods. Many measurements are captured as images during the tuning process, which must be analyzed to guide subsequent steps. Here we present a synthetic modeling approach to simulate experimental data using an explainable boosting machine (EBM). We want to use an EBM since features in these images often reflect behaviors or states of the QD devices, that once carefully interpreted, can aid in their control and calibration. For example, a triangle plot, taken on a QD device, can reveal critical characteristics for its tuning, such as the necessary voltage ranges to form an isolated current channel. While a different approach, such as convolutional neural networks (CNNs), can validate a successful measurement, they do not offer insights into adjusting the device if a bad image is detected – CNNs often sacrifice model intelligibility for accuracy. Our approach enhances the explainability of predictions without compromising accuracy, making synthetic modeling a superior strategy for QD device tuning.