CD Metrology Separates Shape From Scale in Pattern Transfer
Alexander E. Braun, Senior Editor -- Semiconductor International, 7/1/2000
 |
This ignores the effect of the measured feature's shape variation on the analyzed waveform. Deducing feature shape from this requires waveform signal interpretation. This procedure has worked well; but now that more complex processes become necessary — such as ashing photoresist, where it is important to look at the edge to decide how much to take off — this feed-forward information must be used more thoroughly. Some waveform portions are more strongly related to different parts of the cross-sectional profile. A single-parameter CD, based on a single edge detection algorithm, cannot account for a real object's shape and scale.
Small CD linewidth metrology requires more than a single parameter to be useful; a shape index is needed. Fully used, intensity waveform data are sufficient to monitor a process and detect changes before there is a metrology-related production failure. The edges of small linewidth features have become a substantial proportion of the total width and must be taken into account. A single CD width based on a specific single point on a waveform has value only if the shape of the object under measurement is nominal. A minimum of two measurements begins to correlate shape as it enables the approximation of a slope and an initial shape index. In an ideal metrology system, the photoresist sample under test might pass simple CD with metrology but be stopped for rework based on shape anomaly.
According to McIntosh, there is no reason to limit CD metrology to a single width measurement. The string of intensity vs. distance on which the modern CD-SEM operates to determine a width can just as easily make repeated and rapid calculations in memory, comparing all aspects of the waveform against a known template; logical operations based on this result in a multiple-parameter CD metrology system that takes shape as well as scale into account.
The system develops a database by capturing normal images. Then the intensity is used to analyze them, checking whether features have a small foot in the photoresist, a tiny rounding, etc. This provides a distribution of what is normal with shape versus scale. Once these data are available, it becomes possible to test each measurement against what is normal. If it fits within the normal distribution, then it is a "pass." This cannot be done with a single-edge detection measurement algorithm.
Once the metrology data can be filtered to ensure it is normal, it can be categorized as to what are the tolerances of a "normal" shape, and information given to etch to execute recipes that will selectively modify the area and bring it to spec.
By combining the CD-SEM with scatterometry, it is possible to get a reliable z versus x and y mapping. This is perfect metrology information. A SEM can only give intensity, not z, but its relationship to z can be interpreted. With modeling, scatterometry directly provides z versus x and y. There are some problems in deconvolution, but these are solved easily. •
