In semiconductor electronics manufacturing, device performance often depends upon size. For example, microprocessor speed is linked to the width of transistor gates. Accurate measurement of feature width is an important but challenging problem. When a scanning electron microscope (SEM) forms an image of a silicon line, areas within a few tens of nanometers of the line edges are characteristically brighter than the rest of the top-down secondary electron image. In general, the shape of the secondary electron signal within such edge regions depends upon the energy and spatial distribution of the electron beam and the sample composition, and it is sensitive to small variations in sample geometry. Image formation can be modeled using Monte Carlo techniques that model the electron/sample interaction to follow representative electrons through the process of scattering inside of the material before being captured or escaping and being counted. In practice it is the inverse of this process (determination of sample shape given the image) that is required. There is presently no straightforward way to calculate this inverse. Indeed, in a strict mathematical sense, the inverse usually does not exist because many sample geometries produce the same image. Accordingly, assigning edge locations is done by finding a model sample for which the forward calculation produces an image equal to the one actually observed. Edge locations, and consequently linewidths, are assigned based upon this model sample. compared to the preferentially etched single crystal silicon samples we studied in previous years polysilicon line shapes are less constrained a priori, a larger set of possible shapes must be modeled and tested for a match to the observed image profile, and the possibility of encountering multiple acceptable matches is increased. We address this by the use of a precomputed library of modeled shapes in a further development of a method first employed for the SEM by Davidson and Vlad?r. The concept is shown in Fig. 1. Matching to the library is performed using a least squares method. We introduce interpolation of the library and the ?independent edges? approximation to improve accuracy and speed. We have tested this procedure on polysilicon test patterns on a thin gate oxide. For the library we varied sidewall angle and radius of the upper corner. The procedure produced good fits to the measured data. Scatter in the measured edge shape and line width parameters was about as expected based upon the observed image noise level. Uncertainties are greater for parameters (e.g., corner radius) which have only subtle effects on the observed image than for parameters (e.g., edge locations and corresponding overall linewidth) that have more dramatic effects.
Citation: Proceedings of SCANNING 2001 May 5-7, 2001 New York, New York, USA
Issue: No. 2
Pub Type: Journals
device performance, feature width, linewidth measurement, microprocessor speed, modeled shapes, scanning electron microscope, semiconductor manufacturing, transistor gates