A self-consistent hybrid model connects empirical and optical models for fast, non-destructive inline characterization of thin, porous silicon layers

Epitaxially-grown wafers on top of sintered porous silicon are a material-efficient wafer production process, that is now being launched into mass production. This process makes the material-expensive sawing procedure obsolete since can be easily detached from its seed substrate. With high-throughput inline processes, fast and reliable evaluation processes crucial. The quality layers plays an important role regarding successful detachment. Therefore, we present non-destructive investigation algorithm thin, layers. We predict layer parameters directly reflectance data by using convolutional neural network (CNN), which inspired comprehensive optical modelling approach literature. There, numerical fitting reflection curves calculated with physical model performed. By adding to CNN, create hybrid model, not only predicts parameters, but also recalculates curves. allows consistency check for self-supervised optimization. Evaluation experimental shows high similarity Scanning Electron Microscopy (SEM) measurements. Since parallel computation possible 30.000 samples evaluated in roughly 100 ms.