Candidate: Jimmy Hung
Title: Quantum Computation with Superconducting Parametric Cavity
Date: November 25, 2020
Time: 2:00 PM
Supervisor(s): Wilson, Chris
Multimode superconducting parametric cavity is a exible platform that has been used to study a variety of topics in microwave quantum optics ranging from parametric amplification, entanglement generation to higher order spontaneous parametric downconversion (SPDC). Leveraging the extensive toolbox of interactions available in this system, we can look to explore exciting applications in quantum computation and simulation. In this thesis, we study the use of the parametric cavity to realize continuous variable (CV) quantum computation.
We propose and examine in detail the scheme to compute with the microwave photons in the orthogonal frequency modes of the cavity via successive application of parametric pump pulses or cavity drives. The family of all Gaussian transformations can be accomplished easily with interactions already demonstrated in this system. From recent results and proposals involving higher order SPDC, there are also clear pathways towards realizing the non-Gaussian resources necessary for universal computation. Common measurements on the system are accomplished with standard measurement techniques on the output state of the cavity and additional useful measurements may be implemented using available parametric interactions or new device designs involving a qubit as a nonlinear probe.
Using the parametric cavity, we experimentally implemented a hybrid quantum-classical machine learning algorithm called the Quantum Kitchen Sinks (QKS) as the first step towards developing this platform for quantum computation. The algorithm is studied over two sets of experiments starting from partial experimental implementation of the quantum variational circuits to fully experimental implementation using multiple simultaneous continuous wave (CW) pumps. In both cases, we find that the quantum part of the algorithm implemented in the parametric cavity improved the classification accuracy on a difficult synthetic data set up to 90.1% and 99.5% respectively when compared to a classical linear machine learning algorithm.