Master's thesis presentation, David Hambraeus

Examiner: Raphaël Van Laer  
Supervisor: Raphaël Van Laer
Co-advisors: Paul Burger, Johan Kolvik
Opponent: Ida Ekmark

Abstract: Phononic devices could enable and improve a broad range of functions in the realm of classical and quantum information processing. However, such devices are currently often designed by hand, combined with brute-force parameter sweeps, which severely limits the designs that can be investigated. This work presents a method for inverse-design of phononic devices, allowing a vastly larger design space to be explored. At the heart of the method lies a fast calculation of gradients using the adjoint method, whereby the gradient computation costs no more than a single normal simulation. I show that this method is theoretically applicable to phononic devices, and demonstrate that it works in simulation. As a proof of concept, I attempt to design a phononic beamsplitter. The designed devices perform very well, splitting the input power 50/50 with almost nothing being scattered into other modes or reflected. However, there are still some complications when transitioning from the first part of the optimization, with continuously varying material parameters, to the second part where level-set methods were used to make the design binary. If these problems can be solved, this method looks promising for use in the design of future phononic devices.