QDFlow: an open-source quantum-dot physics simulator

As progress is made towards the scalability of quantum dot (QD) devices, the increasing complexity of tuning and operating these devices becomes a major bottleneck. Machine-learning (ML) tools have shown promise in automating many of the time-consuming steps in the process of device preparation; however, ML models require large, diverse, and labeled datasets for training and validation. We present QDFlow, an open-source Python simulator that allows generating arbitrarily large amounts of realistic QD data, complete with experimentally-motivated noise. QDFlow moves beyond constant-capacitance approximations, solving the Thomas-Fermi equations to obtain self-consistent charge densities and capacitances and determine the charge density in the device. This allows QDFlow to accurately model full charge-stability diagrams, including regions with low barriers where adjacent dots merge together. QDFlow supports multi-QD topologies, broad parameter sweeps, and controllable noise models, producing reproducible datasets suited for training, benchmarking, and stress-testing ML-driven tuning frameworks. Datasets generated by QDFlow have already been used to train several different ML models, many of which have been tested and applied to state-of-the-art experimental devices.