Profile Reconstruction Methods In Optical Scatterometry Based Nanostructure Metrology

Optical scatterometry has been widely used in the nanostructure profile reconstruction in intergrated circuit (IC) manufacturing line based on its properties of rapidity, non-destruction and low cost. Optical scatterometry is inherent a model-based metrology technique, which searches for the optimal nanostructure profile parameters by continuously comparing the theoretical sisgnature with the measured one. Hence, the success of optical scatterometry not only relies on the forward optical modeling algorithms, but also relies on the solving of the related inverse problem. Commenly, the inverse problem is simplified as a least square regression problem under an ideal geometrical model without considering the effect of variations in manufacturing line and different error sources contained in the measured signature.Accordingly, we propose to perform the research of inverse nanostructure profile reconstruction in optical scatterometry. The research topics include the recognition of nanostructure profile, robust parameter extraction method and robust measurement uncertainty estimation. The main contents and innovations include:A support vector machine based nanostructure profile recognition method is proposed. This method avoids the destructive profile metrology by conventional scanning or transmission electron microscopies. By performing the theoretical and experimental analysis on a one-dimensional photoresist grating, the high recognition rate of the proposed method has been validated.The effect of abnormally distributed measurement errors on the conventional least square function based nonlinear regression and library search methods has been systematically investigated. The novel nonlinear regression algorithms based on data refinement and robust statistics have been proposed, respectively. The proposed methods present the capability to achieve the higher measurement accuracy for the first time in the field of optical scatterometry. To enable the application of data refinement and robust statistics in semiconductor industry, we further developed the data refinement based and robust statistics based library search methods, respectively. The proposed novel library search methods have the capability to achieve the faster and more accurate nanostructure reconstruction than that of the conventional library search methods.A data refinement based correction has been perfomed on the basis of the conventional parameter uncertainty calculation method. By rejecting those spectral data points corresponding to large measurement errors, the more robust and reliable uncertainty estimation can be obtained.An in-house developed dual-rotating compensator Mueller matrix ellipsometer as well as an algorithm software package have been used to perform the metrology experiments on typical semiconductor nanostructures such as the one-dimensional photoresist grating, the etched silicon graing, and the deep-etched multilayer grating. Experimental results have demonstrated the feasibility and effectiveness of the above proposed theories and methods.The proposed theories and methods in this dissertation will provide the theoretical foundations for the understanding of the mechanism of optical scatterometry, as well as providing the novel guidance for improving the measurement accuracy and precision in optics-based nanostructure metrology. It is also expected to have a promising application prospects in the on-line process monitoring and control in high-volume nanomanufacturing.