Embedded Files
Experimental realization of new generation of quantum technologies requires precise control of many experimental parameters. In practice these are often adjusted manually, which is not compatible with the scalability goals we have in mind when designing quantum devices. I believe machine learning is a great tool to achieve this much needed automation. In collaboration with Thomas Ihn lab we made an algorithm that automatically tunes quantum dots to desired states (Phys. Rev. Applied 13, 054019 (2020)) and another one that searches for best experimental samples of graphene, graphite and hBN (Phys. Rev. Applied 13, 064017 (2020) - short live demo here).
Finding correct Hamiltonians for bilayer graphene quantum dots
Finding correct Hamiltonians for bilayer graphene quantum dots
Quantum dots are one of the amazing platforms for quantum computing chips: they are natural two-level systems, scalable and controllable. There is many different ways to engineer them and one of the newest ones is to use bilayer graphene! The tricky part comes when we want to determine which specific bound states have be excited in the quantum dot - it happens to be a complex optimization problem. In arXiv: 2203.00697 we develop a new optimization method that combines global and local methods in just the right way to tractably fit complex optimization landscape of this model. Precise characterization of the newest generation of quantum devices is really crucial and both experimentally and theoretically challenging - we are thinking about all the newest devices and work with experimental groups to help describe them.
Machine learning for tuning of quantum dots
Machine learning for tuning of quantum dots
Quantum dots make great quantum devices: they are very compact, scalable and highly controllable. This control, however, can be very demanding as large number of experimental parameters come into play.
We designed and trained a convolutional neural network to control the key experimental voltages and therefore tune the quantum dots fully automatically into desired quantum states. An additional advantage is that the neural network seems to require way less data than a typical human operator. We believe this is a first step toward fully AI-controlled quantum experiments.
Read more here: Phys. Rev. Applied 13, 054019 (2020)
Machine learning for 2D materials
Machine learning for 2D materials
Thin nano-materials are essential to modern quantum technologies. What started with the isolation of graphene in 2004 is now a rich field of complex quantum devices for everything from quantum computing to investigation of completely new condensed matter physics phenomena.
Finding the right piece of the nano-material for a particular technological application is however an extremely lengthy and difficult task. We build an ML driven, freely accessible microscope API that automatically performs scanning of the sample and selection of best flakes!
Read more here: Phys. Rev. Applied 13, 064017 (2020)
Page updated
Google Sites
Report abuse