Browsing by Author "Lahtinen, Jyri"

Vacuum arc electrical breakdowns cause problems in many appliances operating in high electric field, such as the Compact Linear Collider (CLIC), a proposed next-generation particle accelerator in CERN. The breakdown phenomenon is not well-understood despite decades of research. Diffusive mass transport in metallic surfaces under electric fields is hypothesised to play a role in the events leading to breakdowns. Kinetic Monte Carlo (KMC) is a well established simulation method for studying diffusion. The weakness of KMC is that it requires knowledge of the rates of all processes that can happen during simulation: in the case of diffusion, these are migration events of mobile objects. The rates can be found from migration barriers, which in turn can be calculated using various methods. In this thesis, the parametrisation scheme of an existing atomistic KMC model for studying Cu surface diffusion was improved. In this model, the migration barrier is a function of the local environment of the migrating atom. The barriers in different environments were calculated with the nudged elastic band (NEB) method. It is an accurate way of finding barriers, but too expensive to be used for calculating them all in the improved parametrisation scheme. This problem was treated with a multidisciplinary approach of training an artificial neural network (ANN) to predict the barriers, using a limited dataset calculated with the NEB method. Good prediction performance was achieved for the case of stable migration processes on smooth surfaces, and the predictor function was found to be sufficiently fast to be called during KMC runtime.