QFlow 2.0: Quantum dot data for machine learning

Using a modified Thomas-Fermi approximation, we model a reference semiconductor
system comprising a quasi-1D nanowire with a series of five depletion gates whose voltages
determine the number of dots and the charges on each of those dots, as well as the conductance
through the wire. The data set consists of 1001 generated gate configurations averaging over
different realizations of the same type of device. Each sample data is then stored as a
100x100 pixel map from voltages to (i) current through the device at infinitesimal bias and
(ii) the output of the charge sensor evaluated as the Coulomb potential at the sensor location
? the experimentally relevant parameters that can be measured. In addition, the output
includes also (iii) the label determining the state of the device, distinguishing between a
single dot, a double dot, a short circuit and a barrier, and (iv) the information about the
number of charges on each dot (with a default value 0 for short circuit and a barrier). Data
structure file and project description are included with the data. This data set was used to
train a convolutional neural network to automatically recognize a state of a subregion within
the plunger voltage space and to test the idea of ?auto-tuning? the experimental device (for
details see arXiv:1712.04914).