Measuring the multilayer silicon based microstructure using differential reflectance spectroscopy.

The yield of a large-area ultra-thin display panel depends on the realization of designed thickness of multilayer films of all pixels. Measuring the thicknesses of multilayer films of a single pixel is crucial to the accurate manufacture. However, the thinnest layer is reaching the sub-20nm level, and different layers feature remarkable divergence in thickness with similar optical constants. This turns to a key obstruction to the thickness characterization by optical spectroscopy. Based on the tiny differences in absorptivity, a fast method for measuring the film thickness in a single pixel is proposed which combines the layer number reducing model and micro-area differential reflectance spectroscopy. The lower layers can be considered as semi-infinite in the corresponding spectral range whose thickness is infinite in the fitting algorithm. Hence, the thickness of the upper layer is fitted in a simplified layer structure. For demonstration, a multilayer silicon microstructure in a single pixel, p-Si/a-Si/n-Si (10nm/950nm/50nm) on complex substrate, is measured. The light spot diameter is about 60 microns with measuring-time in 2 seconds. The measurement deviation is 3% compared by a commercial ellipsometer. To conclude, the proposed method realizes the layer number reduction for fitting multilayer thickness with large thickness difference and similar optical constants, which provides a powerful approach for multilayer microstructure characterizations.

Deep learning for post-processing ensemble weather forecasts.

Quantifying uncertainty in weather forecasts is critical, especially for predicting extreme weather events. This is typically accomplished with ensemble prediction systems, which consist of many perturbed numerical weather simulations, or trajectories, run in parallel. These systems are associated with a high computational cost and often involve statistical post-processing steps to inexpensively improve their raw prediction qualities. We propose a mixed model that uses only a subset of the original weather trajectories combined with a post-processing step using deep neural networks. These enable the model to account for non-linear relationships that are not captured by current numerical models or post-processing methods. Applied to the global data, our mixed models achieve a relative improvement in ensemble forecast skill (CRPS) of over 14%. Furthermore, we demonstrate that the improvement is larger for extreme weather events on select case studies. We also show that our post-processing can use fewer trajectories to achieve comparable results to the full ensemble. By using fewer trajectories, the computational costs of an ensemble prediction system can be reduced, allowing it to run at higher resolution and produce more accurate forecasts. This article is part of the theme issue 'Machine learning for weather and climate modelling'.