Quantifying Charge Stability Drift in Quantum Dots using Machine Learning

The different sources of noise present in quantum dot devices present a key challenge to the operation of gate-defined spin qubits. For spin qubits defined in double quantum dots, drifts in the gate voltages may have an impact on the dots' charge states. This necessitates frequent retuning and complicates long-term, automated qubit operation. We develop a machine-learning approach to quantify this drift through extraction of triple points from charge stability diagrams over time. A convolutional neural network trained on 200,000+ simulated charge stability diagrams identifies triple points and extracts their coordinates in voltage space from experimental data. This allows us to visualise triple point drift trajectories and extract 2D time series of triple point voltages. We apply this approach to double quantum dot devices and observe both charge noise and abrupt shifts in the triple point voltages, consistent with complex charge dynamics in the device environment. The work presented here has direct implications for automated tracking and statistical characterisation of charge stability dynamics in gate-defined quantum devices, informing both feedback-based tuning and studies of charge noise mechanisms.