Condensed Matter Seminar Series

Stefanie Czischek

University of Waterloo

 Automated tuning of quantum dot devices with extremely small neural networks

Solid-state quantum dots form one of many promising candidates for qubits, the basic elements of quantum computing hardware, by using semiconductor devices to confine single charge carriers such as electrons. Dominant quantum effects can be observed in the few-carrier regime of a quantum dot, which can be reached via precise gate tuning. While this gate tuning requires accurate calibration and control of the semiconductor device, machine learning methods have recently been used to demonstrate successful automated tuning procedures for single and double quantum dots. With possible implementations on low-power memristor-based computing hardware going beyond von-Neumann architectures, artificial neural networks are further promising candidates for the development of miniaturized automated tuning elements with a possible integration inside the quantum dot cryostat. Driven by the restrictions of state-of-the-art memristor crossbar array devices, we explore extremely small feed-forward neural networks for the detection of charge transition lines in quantum dot stability diagrams. We use a transfer learning approach to demonstrate that neural networks trained on numerically created synthetic datasets are capable of detecting transition lines in small patches of experimentally measured pre-processed stability diagrams. By detecting transition lines in arrays of small patches, we develop an algorithm to robustly tune experimental devices into single-charge states, providing a first step towards an on-chip autotuning procedure.

Zoom-link: https://ucph-ku.zoom.us/j/66299519490?pwd=cHlLS0pjYVRSQit2SW1NNFhHb0VBQT09

Zoom-password: NielsBohr