Tuningstages

The TuningStage abstraction represents a typical step of both manual and automated tuning, its workflow is illustrated in The tuningstage workflow implemented by the base class and used by all sub-classes.. This base class is meant to be sub-classed, where each sub-classes implements the methods called during the workflow.

tuningstage workflow

Fig. 12 The tuningstage workflow implemented by the base class and used by all sub-classes.

A TuningStage takes as input a list of parameters which are to be swept, as well as the parameters’ safety ranges and (trained) machine learning models to use for classification. Setpoints are calculated based on the ranges to sweep, also an input parameters and which have been determined during previous tuning. The current default method returns linearly spaced setpoints but is meant to be changed in the future.

Once the data is taken, it is fitted via a DataFit sub-class. Either the extracted feature vector or the entire measurement (normalized, filtered or Fourier frequencies) are passed to the classifier for a quality or regime assessment. If the desired quality/regime is found, a TuningResult instance is returned. If not, the signal strength is checked to determine whether sweep ranges need to be adjusted. If the detected current is too low/high, voltage ranges are adjusted towards more positive/negative values. If the new ranges do not exceed the gates’ safety ranges, the measurement loop resumes by calculating new setpoints. If safety ranges have been reached, the TuningStage returns an instance of a TuningStage indicating a unsuccessful run.

Currently implemented subclasses are GateCharacterization1D and ChargeDiagram. All methods used in these classes have been refactored so that they can be used without the TuningStage abstraction as well. They are in the *tasks.py files located close to the classes’ implementations.

Gate characterization

GateCharacterization1D implements pinchoff curves. See the notebook for an example.

Charge diagram

ChargeDiagram takes a 2D scan, most often plungers. See the notebook for an example.