Since the industrial production of nanostructured surfaces such as absorption and anti-reflective coatings is a rapidly growing sector, there is an increasing need for measurement methods for rapid quality inspection in the process cycle. Defects in the nanostructure can significantly change the surface properties and must therefore be reliably identified. Scattered light measurement methods are predestined for in-process measurements because they are noninvasive and fast. However, in order to determine defects, it must be known what the relationship is between the type and rate of surface defects that occur and the measured scattered light distribution. In the case of purely periodically nanostructured surfaces, the inference from scattered light to defects has already been successful, since the resulting scattered light distributions can be modelled comparatively easily. For non-periodic nanostructured surfaces, on the other hand, there is still no universal procedure for the realisation of scattered light measurement methods, which is why their potential for fast, processrelated quality testing cannot yet be exploited. Therefore, the aim of this project is to develop a design method for scattered light measurement systems based on machine learning algorithms that can be applied universally to various surface nanostructures. To this end, it is to be clarified how scattered light bases can be determined automatically from simulated training data by machine learning, so that measured scattered light distributions are represented by linear combinations of these bases, and whether high correlations between surface defect properties and the coefficients of individual bases can be achieved in a targeted manner. In addition, it is to be clarified how the scattered light bases can be used to derive significant simplifications of the scattered light measurement setup. In order to prove the broad applicability of the design method, three fundamentally different nanostructure groups are investigated, which result from the different combinations of stochastic and deterministic nanostructure and defect distribution. The measurement setups and evaluation algorithms for defect classification derived from the design method are finally validated by angle-resolved scattered light measurements. In summary, the uniform design of scattered light measurement systems for the fast, process-oriented assessment of different classes of nanostructured surfaces is to be made possible by using or developing machine learning methods in combination with simulated scattered light training data for scattered light analysis. Therefore, the joint research project is based on a bundling of the expertise of both applicants from the areas of surface characterisation, optical scattered light measurement technology and measurement uncertainty analysis, as well as compressive sensing and classification using machine learning algorithms. 

Details