Optical measurement of ultra fine linewidths using artificial neural networks
Measuring fine track widths with optical instruments has become increasingly difficult as the dimensions of the features of interest have become smaller than the traditional optical resolution limit. This has caused a move to non-optical methods such as scanning electron and atomic force microscopy techniques, or novel optical methods combined with signal processing techniques to provide measurements of these samples. This thesis presents one method to increase the measurement capabilities of an optical system. This is achieved by combining an optical profiler such as a scanning interferometer, with an artificial neural network (ANN). Once trained the ANN can calculate the object parameter for other tracks not contained in the training set. This process works extremely well; with experimental results showing that a 60nm track width can be calculated with a 2nm error using an optical system with a spot size of 2.6 microns. The technique can be extended to obtain other parameters such as height, sidewall slope and for other structures such as double tracks. Various aspects of the ANNs have been investigated, such as the training range, the size of network and the impact of noise etc. These studies show that the technique is extremely robust, and has huge potential for general usage.