The increasing demand for sophisticated quantum dot devices drives innovation in machine learning calibration and control, but advances now rely on access to a wide range of accurate labeled datasets that are both difficult and time-consuming to create experimentally. To address this challenge, Donovan L. Buterakos, Sandesh S. Kalantre, and colleagues of the University of Maryland and the National Institute of Standards and Technology, introduce QDFlow, a new open-source physics simulator. QDFLOW combines established physical models with flexible noise simulations to generate realistic synthetic data for multi-Quantum dot arrays with terrestrial truth labels. This tool allows for the creation of large and diverse data sets, accelerate machine learning development, provide robust benchmarks, and support basic research into the operation of quantum dot devices.
This approach combines a self-integrated Thomas-Fermi Solver, a dynamic capacity model, and a flexible noise module to generate ray-based data that is very similar to experimental measurements with charge stability diagrams and RAY-based data, maintaining full access to ground truth labels. The QDFlow's core resides in the physics engine, which utilizes the Thomas-Fermi solver to determine the stable charge configuration and sensor output of a defined quantum dot device. The researchers implemented a one-dimensional nanowire model, restricting the charge to a linear structure and modeling the electrostatic potential created by the gate to induce charge density.
The simulation incorporates over 20 physics parameters, allowing precise control over device properties and the generation of 2D charging stability diagrams and 1D rays that directly reflect actual quantum dot tuning procedures. To emulate the experimental conditions, the team designed a flexible noise module, adding effects such as heat spread, charging offset drift, and voltage fluctuations to the simulated data. This module allows researchers to customize the noise process and create data sets that rival experimental measurements while maintaining access to ground truth labels. By improving and extending the previously introduced Thomas-Fermi solver, scientists have improved the flexibility, physical relevance, and downstream machine learning workflows and integration of QDFLOW, ultimately accelerating the development of automated control tools for quantum DOT systems.
Realistic quantum dot simulation with charge density
QDFLOW represents a major advance in quantum dot systems simulation and addresses the critical needs of realistically labeled data sets to support the development and benchmarking of machine learning algorithms. Unlike existing simulations, which often rely on simplified capacitance models, QDFLOW allows for the generation of charge stability diagrams and RAY-based data that perfectly simulates charge density and closely mimics experimental behavior. This physics-based approach uses traditional methods to capture dynamic effects such as inaccessible dot merging and transition gradient variations. The team's work offers a versatile, open source platform that can generate a variety of synthetic datasets with controllable noise, such as heat spread and telegraph noise. QDFLOW's early applications have already accelerated machine learning model training for tasks such as global state recognition, RAY-based navigation, and data quality assessment. The authors acknowledge that systematic study of multidimensional modeling, integration with experimental feedback loops, and robustness under various noise conditions could potentially establish future work as a central resource for expanding the functionality of QDFLOW and bridging theoretical and experimental quantum DOT research.
