Benchmarking single-electron transistor charge sensitivity using machine learning

As silicon quantum device manufacturing shifts from academic laboratories to industrial settings, metrics to assess fabrication processes and optimal geometries will need to be developed. Preferentially, these metrics should be extracted using automated protocols to facilitate the large-scale characterisation required to obtain statistical evidence of improvement. In this talk, we propose a device characterisation strategy that aims to compare device performance across different quantum dot technologies. We present this approach applied to charge sensing, where the objective is to sample from the volume in gate voltage space in which the device can be operated with a satisfactory charge sensitivity, this allows us to efficiently quantify this volume we denote the charge sensitivity volume. This volume is determined by a uniform sampling algorithm that searches gate voltage space to identify regions where the charge sensor’s Coulomb blockade oscillations are measured to have a charge sensitivity exceeding a user defined satisfactory threshold. With the increasing demand for fast automatic testing of entire wafers of devices, this characterisation strategy could provide a universal benchmark through which to compare single-electron transistor technologies.