Regensburg 2013 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DS: Fachverband Dünne Schichten

DS 13: Poster Session III: Layer properties: electrical, optical and mechanical properties; Thin film characterization: structure analysis and composition (XRD, TEM; XPS, SIMS, RBS..)

DS 13.14: Poster

Monday, March 11, 2013, 17:00–20:00, Poster B1

Bayesian Fitting of Neutron and X-Ray Reflectivity Data — •Jean-Francois Moulin, Martin Haese-Seiller, and Andreas Schreyer — Helmholtz-Zentrum Geesthacht Boltzmann Strasse 1 21502 Geesthacht Germany

Deducing the structure of thin films by analysis of the X-ray or neutron reflectivity curves often proves to be challenging. The inversion problem does not have a unique solution and the χ2 landscape which has to be searched for a global minimum is often very unfavorable to the use of standard techniques such as the Levenberg-Marquardt algorithm. It is well known that human guidance is often required in order to start the search with a parameter set which is already very close to the real solution. In the quest for better fitting methodologies many efforts have been devoted to techniques which garantee that the algorithm does not get stuck in local minima. Notable examples are genetic algorithms, particle swarm optimisation and Bayesian analysis. Bayesian optimization makes use of a Monte Carlo exploration of the χ2 landscape and eventually leads to a description of the parameters corelations and the distributions of their values. In the Bayesian framework one encodes all a priori knowledge about the experiment and then extracts from the posterior data a quantitative information about what we can learn from the experimental observation. In this paper we will show how one can take advantage of the Bayesian optimization methods to charachterize thin films and we will demonstrate the use of readily available open source libraries to build an efficient and reconfigurable toolbox which can tackle a variety of problems.