Benchmarking machine learning models for multi-class state recognition in double duantum dot data

2025-11-27

AI Summary

Automated classification of multi-class states in charge stability diagrams (CSDs) of semiconductor double quantum dots suffers from low automation and poor model generalizability. Method: We systematically benchmark four deep learning architectures—U-Net, Vision Transformer (ViT), Mixture Density Network (MDN), and CNN—on both synthetic and experimental CSD data, rigorously evaluating the impact of dataset size and normalization strategies. Results: A lightweight CNN with min-max normalization achieves the highest classification accuracy on experimental CSDs, reducing parameters by two orders of magnitude relative to competing models and significantly lowering mean squared error (MSE). This architecture delivers an optimal balance of accuracy, computational efficiency, and deployability. Crucially, this work provides the first empirical validation of a lightweight deep learning model for real-time tuning of physical quantum devices. It establishes an efficient, robust, and generalizable state identification paradigm for automated calibration of large-scale quantum dot arrays.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Semiconductor quantum dots (QDs) are a leading platform for scalable quantum processors. However, scaling to large arrays requires reliable, automated tuning strategies for devices' bootstrapping, calibration, and operation, with many tuning aspects depending on accurately identifying QD device states from charge-stability diagrams (CSDs). In this work, we present a comprehensive benchmarking study of four modern machine learning (ML) architectures for multi-class state recognition in double-QD CSDs. We evaluate their performance across different data budgets and normalization schemes using both synthetic and experimental data. We find that the more resource-intensive models -- U-Nets and visual transformers (ViTs) -- achieve the highest MSE score (defined as $1-mathrm{MSE}$) on synthetic data (over $0.98$) but fail to generalize to experimental data. MDNs are the most computationally efficient and exhibit highly stable training, but with substantially lower peak performance. CNNs offer the most favorable trade-off on experimental CSDs, achieving strong accuracy with two orders of magnitude fewer parameters than the U-Nets and ViTs. Normalization plays a nontrivial role: min-max scaling generally yields higher MSE scores but less stable convergence, whereas z-score normalization produces more predictable training dynamics but at reduced accuracy for most models. Overall, our study shows that CNNs with min-max normalization are a practical approach for QD CSDs.

Problem

• Benchmark ML models for multi-class state recognition in quantum dot data.
• Evaluate performance across data budgets and normalization schemes.
• Identify practical models for accurate quantum dot state identification.

Innovation

• CNNs with min-max normalization for QD state recognition
• Benchmarking ML models on synthetic and experimental data
• Evaluating trade-offs between accuracy and computational efficiency