JPS5942905B2 - Pattern inspection device using inflection point comparison - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a pattern inspection device for integrated circuits, printed circuit boards, etc.

For example, by recognizing scratches on integrated circuits and printed circuit boards,
Through inspection, it is necessary to remove noise from the input pattern, modify the pattern, or extract minute portions in the pattern.

Traditionally, visual inspection was carried out using a microscope, but as the density of accumulation increases, it becomes less efficient and inaccurate.
A fully automated recognition device is required, and an inspection device that can identify flaws at low cost is desired.

Conventional flaw recognition methods include line thinning and enlarging processing methods such as those disclosed in JP-A-48-61030, JP-A-49-34385, and Journal of the Institute of Electronics and Communication Engineers, Paper 50-C8. be.

However, with this method, if the pattern has an acute angle less than a certain curvature, it will be mistakenly determined to be a defect. Therefore, in order to remove errors in cropping the video signal and areas with minute acute angles at the peripheral edge of the pattern, it is necessary to perform area correction that excludes the peripheral edge. Therefore, the circuit becomes complex and the cost increases. On the other hand, the "laser light diffraction distribution detection method" previously proposed by the present inventors had a problem in that the S/N ratio for determining meandering patterns and transparent scratches was low.

In order to solve these problems, the object of the present invention is to provide a pattern inspection device that can detect defects with high defect determination ability using a simple circuit without placing any restrictions on the patterns to be inspected on integrated circuits or printed circuit boards. Our goal is to provide the following.

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a flow chart of a pattern inspection method in an embodiment of the present invention. The present inventors previously proposed a method for inspecting patterns for defects by comparing the inflection points of a pattern with dictionary data.

In such a case, the present invention provides a method of extracting inflection points from inside a turn and a method of applying the same to defect inspection. That is, as shown in FIG. 1, the present invention applies a black-and-white binary pattern 9 as it is and after inverting it by 1, O, thinning and enlarging it, and by comparing the two, it is possible to obtain a pattern with a radius of curvature below a certain value. Detect the pattern edge area and use its inflection point data 17 and dictionary data 19
Defects are discriminated and detected by comparing them with

That is, a pattern material 1 is imaged by a television camera 2, and a binary signal 9 of the pattern is photoelectrically converted.
The first pattern thinning means directly performs pattern thinning 10, and the 1, 0 inverted 12 signal is also pattern thinned 13 by the second pattern thinning means.

The binary signal 9 of the pattern may be generated by scanning the material 1'' by passing the light from the laser 3 through the rotating mirror 4, receiving the light with the photodetector 8, and photoelectrically converting it, or by using a flying spot.
It can also be obtained by projecting the light from the scanner 5 onto the material 1'' through the fiber line 6, and receiving the reflected light with the photodetector 7 for photoelectric conversion.Next, after thinning 10 and 13, The patterns are enlarged 11 and 14 and restored by the first and second pattern enlargement and restoration means, respectively, and the two are compared 15.
The quantization error is compressed (16) and an inflection point is detected (17).

On the other hand, inflection point data having a radius of curvature r or less is extracted 19 from the defect-free drawing data 18 as dictionary data.

The two inflection points 17 and 19 are compared 20 and if they do not correspond, a defect is detected 21.

Figures 2a-g and 3a-e are explanatory diagrams of the optical inflection point detection method in Figure 1, and Figures 4a-d and 5
Figures a to c are explanatory diagrams of more specific methods of Figures 2 and 3, respectively.

All of these methods use known methods.

First, the pattern is divided into image areas and filled with black and white, and this pattern binary signal is stored in correspondence with one location in the 5x5 area of the shift register SRA, as shown in Figure 4a. go. When thinning the lines, even if there is only one x mark in the 5x5 area in Figure 4a, it will be in the "O" state (
white), replace the bit in the center position with “O゛”,
Let it be data A.

For example, if the pattern is as shown in Figure 4b, the shift ν
When stored in the register SRA, it becomes as shown in Figure 4c,
Since only two predetermined locations are "O", the center position is expressed as "°0". When the pattern is thinned, the outline of the pattern recedes to the position of the dotted line, as shown in FIG. 4d, so that the portions P and q below the radius of curvature r are removed. next,
When enlarging, as shown in Figure 5a, if the center position of the 5x5 area of the tide register SRB is in the "1" (black) state, all other "0" marks are "1".゛Display in the state.

Data B processed in this way is transferred to shift register SRC.
Store in. As shown in FIG. 5c, the data stored in the shift register SRC has a contour line that returns to a solid line.
However, the parts P and q are not restored. By the above-described method, the binary signal shown in FIG. 2a is directly thinned into lines as shown in FIG. 2b, and further enlarged and reconstructed to become as shown in FIG. 2c. On the other hand, the binary signal in Fig. 2a is inverted by 1 and 0 to become as shown in Fig. 2d, thinned to become as shown in Fig. 2e, and further enlarged and restored to be shown in Fig. 2f. become that way.

Next, when the restored pattern of FIG. 2C is compared with the pattern obtained by inverting FIG.

In this way, for example, if there are scratches on the lands 23 and leads 24 of a printed board as shown in FIG. 3a, if the scratches are made into thin lines as shown in FIG. The result is as shown in Figure 3c.

On the other hand, when the pattern in FIG. 3a is inverted by 1,0, thinned, and then enlarged and inverted, it becomes as shown in FIG. 3d.

Therefore, by comparing the pattern in FIG. 3c and the pattern in FIG. 3d, a portion where they do not match, that is, an inflection point as shown in FIG. 3e is detected.

FIG. 6 is a block diagram of a pattern inspection apparatus showing an embodiment of the present invention. In FIG. 6, inverters 31, 3
4 performs 1, 0 inversion, and shift registers 32, 33, 35, 36, 37, and 38 are the shift registers SRA, SRB, and SRB shown in FIGS.
This corresponds to SRC, and the operation is as described above, so a description thereof will be omitted.

As described above, according to the present invention, defects are discriminated by extracting inflection points of a pattern with a certain radius of curvature or less and comparing it with dictionary data, so it is not necessary to set restrictions on the pattern to be inspected. Defects on the order of several micrometers can be detected accurately and quickly.

That is, as described above, in the present invention, the binary pattern is thinned by the width r, the enlarged pattern and the binary pattern are inverted by 1, 0, and then thinned and enlarged by the width r, and then the pattern is thinned and enlarged by the width r. By comparing the pattern with the reversed pattern, information on the position of an inflection point such as an edge having a radius of curvature r or less can be obtained. On the other hand, dictionary data formed by extracting inflection point position information with radius r or less is obtained from defect-free design drawing data, and when the two are compared, positions that do not match become pattern defects. Therefore, only defects can be detected regardless of pattern shape.

[Brief explanation of drawings]

FIG. 1 is a flow chart of the pattern inspection method in the embodiment of the present invention, FIGS. 2a-g and 3a-e are explanatory diagrams of the inflection point detection method in FIG. 1, and FIG. -d
, FIGS. 5a to 5c are explanatory diagrams of a method of thinning and enlarging lines using a shift register, and FIG. 6 is a block diagram of a pattern inspection apparatus showing an embodiment of the present invention. 1: Pattern material, 2: Television camera, 3: Laser, 4: Rotating mirror, 5: Flying spot scanner, 6: Fiber line, 7, 8: Photodetector, 2
3: Padded part on printed board, 24: Lead part, 2
5: Inflection point, SRA, SRB, SRC: shift register.

Claims (1)

[Claims] 1 pattern binary signal forming means, a first pattern thinning means for the binary signal formed by the forming means, and a first pattern thinning means for the binary signal formed by the first pattern thinning means; an enlargement and restoration means; a means for inverting the binary signal formed by the binary signal forming means; a second pattern thinning means for thinning the inverted pattern outputted from the inverting means; a second enlargement and restoration means for the inverted pattern thus thinned; a comparison circuit for the outputs of the first and second pattern restoration means; and means for detecting an inflection point of the pattern from the output of the comparison circuit. , a means for extracting the following inflection point data in a radius from the dictionary data, and detecting defects by comparing an output signal from the extraction means with an output signal from the means for detecting the inflection point of the pattern. A pattern inspection device using inflection point comparison, characterized in that it has means.