Qdflow Simulator Generates Realistic Quantum Dot Data with Ground-Truth Labels for Machine Learning Applications

The increasing demand for sophisticated quantum dot devices drives innovation in machine learning calibration and control, but progress currently relies on access to extensive, accurately labelled datasets, which are difficult and time-consuming to create experimentally. To address this challenge, Donovan L. Buterakos, Sandesh S. Kalantre, and colleagues at the University of Maryland and the National Institute of Standards and Technology present QDFlow, a new open-source physics simulator. QDFlow generates realistic synthetic data for multi-quantum dot arrays, complete with ground-truth labels, by combining established physics models with flexible noise simulation. This tool enables the creation of large, diverse datasets, accelerating machine learning development, providing robust benchmarks, and supporting fundamental research into quantum dot device behaviour.

This approach combines a self-consistent Thomas-Fermi solver, a dynamic capacitance model, and flexible noise modules to produce charge stability diagrams and ray-based data closely resembling experimental measurements, while maintaining full access to ground truth labels. The core of QDFlow lies in its physics engine, which utilizes the Thomas-Fermi solver to determine stable charge configurations and sensor outputs for defined quantum dot devices. Researchers implemented a one-dimensional nanowire model, confining charges to a linear structure, and modeled the electrostatic potential created by gates to induce charge density.

The simulation incorporates over twenty PhysicsParameters, allowing precise control over device properties and enabling the generation of 2D charge stability diagrams and 1D rays that directly reflect real quantum dot tuning procedures. To emulate experimental conditions, the team engineered a flexible noise module, adding effects such as thermal broadening, charge offset drift, and voltage fluctuations to the simulated data. This module allows researchers to customize noise processes, creating datasets qualitatively comparable to experimental measurements while retaining access to ground truth labels. By refining and extending a previously introduced Thomas-Fermi solver, scientists improved the flexibility, physical relevance, and integration of QDFlow with downstream machine learning workflows, ultimately accelerating the development of automated control tools for quantum dot systems.

Realistic Quantum Dot Simulation via Charge Density

QDFlow represents a significant advance in simulating quantum dot systems, addressing a critical need for realistically labeled datasets to support the development and benchmarking of machine learning algorithms. Unlike existing simulations that often rely on simplified, static capacitance models, QDFlow fully simulates charge density, enabling the generation of charge stability diagrams and ray-based data that closely mimic experimental behaviour. This physics-informed approach captures dynamic effects, such as dot merging and variations in transition slopes, which are inaccessible using traditional methods. The team’s work delivers a versatile, open-source platform capable of generating diverse synthetic datasets with controllable noise, including thermal broadening and telegraph noise. Early applications of QDFlow have already accelerated machine learning model training for tasks including global state recognition, ray-based navigation, and data quality assessment. The authors acknowledge that future work could expand QDFlow’s capabilities through multi-dimensional modelling, integration with experimental feedback loops, and systematic studies of robustness under varying noise conditions, potentially establishing it as a central resource for bridging theoretical and experimental quantum dot research.