Dot fit

The DotFit class aims to locate triple points in double dot charge diagrams. It does not cover single dots. In an ideal case, the final charge diagram of a coarse tuning algorithm shows triple points easily locatable so that a fine-tuning algorithm can continue and improve the regime. The current implementation uses a peak finding algorithm to locate voltage combinations showing a higher signal and tries to assign the type of triple point, electron or hole, to each peak. To detect the peaks, a binary structure of the diagram is computed and used as a mask to remove noise. The peaks are then etermined trough binary erosion. All three methods used, binary_erosion, generate_binary_structure and maximum_filter are implemented in scipy.ndimage.filter and scipy.ndimage.morphology.

Unfortunately, the current implementation only works with excellent/synthetic data, which is the reason why it has not been used for the autonomous tuning paper. An example of the fit is shown below.

Fig. 16 Example of a double dot fit.

Dot labels

Charge diagrams in nanotune come in four flavors: good single, poor single, good double and poor double dot. The idea behind this choice is to be able to predict not just the dot regime, but also its quality. Similar to pinchoff curves, labelling data imposes a bias to how new data is going to be classified. The range of regimes is far bigger here though.

In code, the four regimes are labelled by four integers, which defined in nanotune.configuration.conf.json under the dot_mapping key:

• 0: good single

• 1: poor single

• 2: good double

• 3: poor double.

When classifying the quality of a specific regime, regular labels of True/1, 0/False are used.