Photolithography allows large-scale fabrication of nanocomponents in the semiconductor industry. This technique consists of manufacturing a desired pattern on a photoresist film transferred onto the substrate during the etching process. Therefore, the mask quality is essential for reliable etching. For example, the presence of a residual layer of resist might be considered as a mask defect and can lead to the failure of the etching process. We propose the use of a Kohonen self-organizing map for automatic detection of a residual layer from an ellipsometric signature. The feasibility of the suggested inspection by the use of a classification technique is discussed and simulations are carried out on a 750-nm period grating.
Neural networks (NN) have received a great deal of interest over the last few years. They are being applied
accross a wide range of problems in pattern recognition, artificial intelligence, and classification as well as in
the inverse problem of scatterometry. Optical scatterometry is a non-direct characterization method that has
been widely employed in the semiconductor industry for critical dimensions control. It is based on the analysis
of the light scattered from periodic structures. This analysis consists of the resolution of an inverse problem
in order to determine the parameters defining the geometrical shape of the structure. In this work, we will
study the performances of the NN according to various internal parameters when it is applied to solve the
scattered problem. This will allow us to examine how a NN reacts and to select the optimal configuration of
these parameters leading to a rapid and accurate characterization.
Progress in microelectronics has allowed the fabrication of optical gratings with small period-to-wavelength ratios, which are very useful in several applications, such as telecommunication and optical sensors. A rapid and nondestructive characterization process is essential to check the agreement of the produced with the expected structure. The main difficulty is in assuring sufficient homogeneity all along the grating surface. We show that neural characterization coupled with neural selection can be efficiently applied to this quality measurement. Depicted results concern a 1-µm-period grating fabricated by reactive ion etching on a silicon substrate.
The characterization of sub-micrometer period gratings by resolution of the inverse problem has become a real need in the area of the microelectronic. We present an optical scatteromettric method based on the use of a Neural Network (NN). This one permits to learn the relationship linking the diffracted efficiencies to the geometrical parameters. The great advantage of this method is to reject the limitations in resolution that occur with classical microscopic characterization. Theoretical results are demonstrated in this paper. The characterization can be achieved with accuracy close to 5 nm. We also study the index influence on the results and the importance of the choice of the assumed profile shape. Experimental results concerning a silicon 1-μm-period grating are also demonstrated. Finally, a comparison with results obtained by a microscopic characterization permits the validation of the presented method.