# Auto-tuning of double dot devices in situ with machine learning

**Authors:** Justyna P. Zwolak, Thomas McJunkin, Sandesh S. Kalantre, J. P. Dodson, E. R. MacQuarrie, D. E. Savage, M. G. Lagally, S. N. Coppersmith, Mark A. Eriksson, Jacob M. Taylor  
**arXiv:** [1909.08030](https://arxiv.org/abs/1909.08030)  
**Date:** 2019-09-17  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/1909.08030v1/

## Key points

- Quantum dots (QDs) are arrays of semiconductor nanostructures used in quantum computing
- QDs have easy access to control parameters, fast measurements, long qubit lifetimes, and potential scalability
- As the size of the QD array grows, so does the number of parameters required for control, leading to increased tuning complexity
- A team of researchers has implemented an auto-tuning protocol proposed by Kalantre et al. using machine learning (ML)
- The ML algorithm is trained exclusively on synthetic data from a physical model and can replace manual tuning of gate voltages in devices
- The approach achieves a success rate of over 85% when tuning to a double quantum dot regime if at least one plunger gate is initiated sufficiently close to the desired state
- Modifications to the training network, fitness function, and optimizer can lead to further improvement in success rates when starting both near and far detuned from the target double dot range
- This work presents a promising avenue for automating the tuning process in QD devices using ML algorithms trained on synthetic data from physical models.

## Summary

In the field of quantum computing, there are numerous approaches with their own unique challenges in understanding and controlling their operation. One such approach involves electrons confined in arrays of semiconductor nanostructures known as quantum dots (QDs). QDs are particularly attractive due to easy access to control parameters, fast measurements, long qubit lifetimes, and potential scalability. However, as the size of the QD array grows, so does the number of parameters required for control, leading to increased tuning complexity. The current practice of manually tuning qubits is time-consuming and impractical for scaling up and applications. To address this issue, a team of researchers has implemented an auto-tuning protocol proposed by Kalantre et al. [arXiv:1712.04914] in situ using machine learning (ML). They establish a seamless communication protocol between an ML-based auto-tuner and the experimental apparatus and demonstrate that an ML algorithm trained exclusively on synthetic data from a physical model can be used to replace manual tuning of gate voltages in devices. The team shows that this approach can achieve a success rate of over 85% when tuning to a double quantum dot regime if at least one plunger gate is initiated sufficiently close to the desired state. They also discuss modifications to the training network, fitness function, and optimizer as a path towards further improvement in success rates when starting both near and far detuned from the target double dot range. Overall, this work presents a promising avenue for automating the tuning process in QD devices using ML algorithms trained on synthetic data from physical models. This could significantly reduce tuning complexity and enable more efficient scaling up of QD-based quantum computing systems.