SciPost Submission Page
QDsim: A user-friendly toolbox for simulating large-scale quantum dot devices
by Valentina Gualtieri, Charles Renshaw-Whitman, Vinicius Hernandes, Eliska Greplova
Submission summary
| Authors (as registered SciPost users): | Valentina Gualtieri |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2404.02712v1 (pdf) |
| Code repository: | https://gitlab.com/QMAI/papers/qdsim |
| Date submitted: | 2024-04-04 10:40 |
| Submitted by: | Gualtieri, Valentina |
| Submitted to: | SciPost Physics Codebases |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approach: | Computational |
Abstract
We introduce QDsim, a python package tailored for the rapid generation of charge stability diagrams in large-scale quantum dot devices, extending beyond traditional double or triple dots. QDsim is founded on the constant interaction model from which we rephrase the task of finding the lowest energy charge configuration as a convex optimization problem. Therefore, we can leverage the existing package CVXPY, in combination with an appropriate powerful solver, for the convex optimization which streamlines the creation of stability diagrams and polytopes. Through multiple examples, we demonstrate how QDsim enables the generation of large-scale dataset that can serve a basis for the training of machine-learning models for automated tuning algorithms. While the package currently does not support quantum effects beyond the constant interaction model, QDsim is a tool that directly addresses the critical need for cost-effective and expeditious data acquisition for better tuning algorithms in order to accelerate the development of semiconductor quantum devices.
Current status:
Reports on this Submission
Strengths
1- The code allows to treat devices with many quantum dots
2- The computation of the stability diagram is rather fast
Weaknesses
1- The code is based on approximations as the constant-interaction model that limit the accuracy in certain regimes
2- When the capacitances are not user-provided, their determination from the sample geometry is unclear.
3- It is unclear how some quantities are computed, for example the "current", while important underlying parameters like the tunnel couplings and the temperature are not among the input data of the code.
Report
The authors present a software package to compute charge stability diagrams for devices containing a large number of coupled quantum dots. As a motivation, they mention the possible production of large datasets for the training of AI-based automatic tuning of quantum dot arrays for applications in quantum technology.
The presented software is based on the so-called constant-interaction model which approximates the effect of the electron-electron interactions in the device through capacitances. Within this approximation, the charge stability diagrams are found by identifying the charge distributions with integer electron numbers on each dot that minimize the free energy of the device which is expressed in terms of the capacitance matrix, the applied gate voltages, and the numbers of charges on the quantum dots.
The proposed package is described in the paper, with example applications, and it could be useful for the efficient simulation of the model devices within the approximations it is based on. The gitlab repository contains installation instructions and a tutorial section with examples that should allow users to use the software. However, I couldn't find a full-fledged detailed documentation of the code.
It seems to me that the paper thereby more or less meets the criteria of SciPost Physics Codebases. Before the paper is considered for publication, I would nevertheless suggest that the authors take into account the remarks in the list of requested changes.
Requested changes
1- It is not obvious that quantum dots, and in particular small ones with a low electron number, are well described by the constant-interaction model. Errors in the simulation may have fatal consequences in the proposed application of the software since the quality of an AI depends crucially on the quality of the training data. The authors should therefore discuss the limitations of accuracy induced by their approximations and describe the regimes where their software can be expected to provide acceptable results. A benchmarking demonstration of the accuracy through the comparison with more precise codes could be appropriate.
2- The quantum dot device is described by its capacitance matrix which includes the dot-to-dot capacitances as well as the dot-to-gate capacitances. Those capacitances can be provided by the user, but they can also be determined by the software based on the physical dot locations. The latter functionality seems crucial for users who want to simulate a given device. Unfortunately, the paper does not describe how exactly the capacitances are determined and how the detailed properties of the sample such as regions with different dielectric constants are taken into account. A precise simulation would however need a rather thorough treatment of the electrostatic environment (see for example Chatzikyriakou et al., Phys. Rev. Research 4, 043163 (2022)). The authors should describe how their software determines the capacitance matrix from the device geometry, and discuss the approximations involved. A demonstration of the accuracy through the comparison with more precise codes for the case of a small device or with an experimental device might be appropriate.
3- While some randomness in the capacitance matrix may be useful to describe a large ensemble of similar samples, it is not obvious to me why it is useful to add different kinds of noise to the output data. Maybe the authors want to comment on their motivation to have such a feature in the software.
4- The lower panels of Figs. 3, 5, 7, 9 show current data. However, it is not clear where this current flows, where the source and the drain are located, and what allows electrons to move in the sample. Capacitive coupling is not enough as an input for a current calculation, it is the tunnel couplings between dots that are needed to have current. Moreover, the current in quantum dot devices usually depends strongly on temperature, which is also not among the input data. It should be explained how the current is calculated within the described software. Moreover, in the captions of those Figures it is said that (b) shows the current with noise and (c) without noise. From looking at the plots, one has the opposite impression. Are the subfigures (b) and (c) inverted?
5- The definition of "e" below eq. (7) as the elementary charge with sign +1 for holes and -1 for electrons is confusing. It is not clear to me what convention is used exactly to describe the charges.
Recommendation
Ask for minor revision