Image analysis – Applications – Manufacturing or product inspection
Reexamination Certificate
1999-04-15
2003-01-07
Werner, Brian (Department: 2621)
Image analysis
Applications
Manufacturing or product inspection
C382S144000, C382S149000
Reexamination Certificate
active
06504947
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates in general to pattern inspection, and relates in particular to a method and apparatus for performing rounding correction of reference images used for Die-To-DataBASE inspection in terms of design data.
2. Description of the Related Art
In general, it is necessary to inspect the reproduction accuracy of fine patterns produced on such objects as integrated semiconductor circuits fabricated on photomasks, reticules, wafers and liquid crystal panels, in terms of the dimensions and shapes of ideal patterns produced from design data, using a pattern inspection apparatus built for this purpose.
This type of pattern inspection apparatus is required to first perform conversion of design data, coordinates of rectangular or trapezoidal shapes and line data described in stroke characters, into binary digital data comprised by [0]s and [1]s.
Next, an object pattern to be inspected is scanned with a laser beam, and using an image of the object pattern obtained either by transmitting light through the object pattern and focusing on a photodetector, or an optical imaging system, the characteristics of the edge profile of the object image are defined, and optical point-spread functions for the pattern are generated. The functions and binary bit data are convoluted to compute multi-level data (multi-valued data) which are used to produce a reference image.
Then, an actual image produced from optical scanning or imaging system and the reference image obtained from the design data are read synchronously into the apparatus to compare both images at the corresponding pixel locations, thereby detecting any discrepancies or defects in the actual image of the object pattern compared to the reference image of the same, according to Die-To-DataBASE protocol (detection test of defects in the actual pattern).
In the actual image pattern, defects such as rounding of corners, dimensional defects such as line width variations are sometimes created due to the effects of optical conditions and manufacturing processes, so that the reference images obtained from design data do not match precisely with the actual image. Such pseudo-discrepancies are difficult to detect in the conventional inspection method and apparatus.
Therefore, an approach may be considered to examine the features in various inspection locations of the object pattern, so that the reference image can be corrected beforehand in appropriate ways to recognize corners or to properly detect edge positions.
In the conventional method of corner rounding, a corner shape in a reference image is corrected using a corner inspection process suitable to the actual image, leading to correcting the design data without changing their binary format and bit (pixel) units.
In this case, corner shape modifications in the reference image are performed by first making a modified bit pattern based on the design bit pattern, and then performing logic computations using the binary bit data of the real image of the object pattern to carry out corrections for the corner shape in the reference image.
In the meantime, by examining the corner shapes and detecting edge positions in the actual image, correction templates in bit-units are prepared so as to match specific corner angles (for example, 0, 45, 90, 135 degrees) of the corners which are found in the design data.
From these corner templates, those which are closest to the corner shapes in the actual image are selected, and the original design data are corrected using the selected corner templates, and the modified design data are again processed to produce a final multi-valued image in 4-16 tonal gradations. The final reference image so produced are compared with the real image so that the differences in the tonal gradations in the final reference image and the actual image can be determined using a suitable threshold value of defect detection algorithm, to determine whether the discrepancies are functional defects. This type of inspection approach is outlined in Japanese Patent Applications, First Publications, Hei 4-350776 and Hei 6-273915.
A problem inherent in the conventional method of pattern inspection is that the productivity is low. A reference image is produced first from the original design data, and using the templates to approximate the corners in the real image, the original design data are correct to produce another reference image. Although the final reference image closely duplicates the actual image produced by optical scanning, the overall process is cumbersome and suffers from low productivity. It also presents a problem of creating pseudo-defects during the processing, caused by the difference in the tone gradations between correction templates and the actual image. This will be explained in more detail.
Correction templates for the design data are selected from many templates for matching of pattern shapes by selecting individual bit string patterns representing templates that are most similar to the string patterns in the actual image. Such computed bit strings are convoluted with point spread functions computed from the light intensity values obtained by optical scanning and the corrected template images to produce the final multi-valued reference image for comparison purposes. This process becomes highly computation-intensive when the number of pixels increase in an image, and consequently inspection productivity suffers.
Another problem is that, because corner identify and gradation rounding adjustment processes are carried out separately for one expanded reference data, computation requirements are further increased.
To counter such problems of productivity, processing speed may be shortened by including corner information in the pre-expansion design information, but this method suffers from an enormous amount of design information, which leads to increase in memory resources and expansion time. The result is that the overall productivity of inspection is lowered.
Further, because the corner rounding step in the pattern correction process is carried out at each pixel location according to a bit string of a given template, the existing method suffers from a drawback that the gradation values of the corrected pattern are limited by the gradation distribution available in the template.
Further, gradation correction is carried out without regard to the shapes of surrounding patterns, edge sections in a pattern that do not coincide with pixel locations and rounded sections extending from an edge to a corner are problematic, because the differences in gradation, between the reference image and the actual image obtained by scanning, become large, making such areas vulnerable to be identified as pseudo-defects.
For these reasons, if a design pattern contains a corner section, produced by two slanting edges intersecting at some angle, that is not at a pixel boundary, correction of gradation difference by template at the pixel level does not allow interpolation of gradation between the pixels.
Especially, those defects in the object pattern, existing in a pixel boundary or in corners that are smaller than the resolution capability of the inspection apparatus and are poor in contrast relative to the surrounding, are difficult to detect by comparing such patterns with the reference image, because they exhibit inferior difference in gradation relative to the surrounding.
SUMMARY OF THE INVENTION
It is an object of the present invention to provide a pattern inspection apparatus to generate a reference image having corner sections closely matching those in the real image of an object pattern, and to significantly reduce the chances of finding pseudodefects in the corner sections.
The object has been achieved in a method for a multi-level corner rounding process: where reference data are generated by expanding the design data so as to produce multi-level gradation reference data at each pixel having a resolution capability that is finer than an inspection resolution capability; the corner sections are a
Nishii Satoshi
Nozaki Takeo
NEC Corporation
Werner Brian
Whitham Curtis & Christofferson, PC
LandOfFree
Method and apparatus for multi-level rounding and pattern... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for multi-level rounding and pattern..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for multi-level rounding and pattern... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3001307