JPH061489B2 - Pattern inspection method - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern inspection method applied to a comparative inspection of a two-dimensional pattern such as a circuit pattern of a printed circuit board or an integrated circuit.

[Prior Art] Conventionally, in a comparison inspection method for comparing and inspecting a pattern to be inspected with a reference pattern, an image by reflected light or transmitted light of the pattern to be inspected obtained by an imaging device is divided into minute pixels, By binarizing the amount of reflected light or the amount of transmitted light or brightness (hereinafter referred to as density) of each of these pixels to 1,0, a binary pattern composed of a set of pixels having a value of 1,0 is obtained. Similarly, the reference pattern is generated in the same manner or from the design data to obtain the binarized pattern, and the binarized inspected pattern thus obtained and the 2
A binarization method for comparing and inspecting with a standardized reference pattern is often used.

[Problems to be solved by the invention]

However, the above-described conventional binarization method, that is, as shown in FIG. 8, the image 1 of the pattern to be inspected or the reference pattern is divided into minute pixels 2 ... And the density of each pixel 2 ... Is binarized. For the means for obtaining the binary pattern 3 by
It was inevitable to generate irregularities of about 1 pixel at the pattern boundary portion when the value was converted, and therefore, in principle, it was impossible to detect defects of about several pixels or less.

Further, even if the density of each pixel is binarized, even if the binarization can be performed accurately and the binarization can be performed accurately, this misalignment occurs when there is a registration error between the pattern to be inspected and the reference pattern. In order to prevent the minute from being determined as a defect, it is necessary to expand the range of the pixel of the reference pattern to be compared with the pixel of the pattern to be inspected to correct the alignment error. For example, assuming that the alignment error is one pixel, it is necessary to compare not only the corresponding pixel but also the pixel adjacent by one pixel and if there is a match with the reference pattern, it must be determined that there is no defect. However, in contrast to the binarized reference pattern 4 shown in FIG. 9 (a), the binary reference pattern 4 shown in FIG. 9 (b) is used.
In the case of the binarized inspected pattern 5, a defect 6 of 2 pixels has occurred in the binarized inspected pattern 5, but the alignment error has occurred in the minus 1 pixel in the X direction. , The adjacent pixel (X ₄ , Y ₄ ) in the X direction in FIG. 9 (b), which is compared with the pixel (X ₄ , Y ₄ ) in FIG. 9 (a).
Produces the same "1" output as the binarized reference pattern 4, so that the defect 6 for two pixels in FIG. 9 (b) is not detected.

In this way, if the pixel comparison range is expanded in consideration of the alignment error, defects up to twice the size of the alignment error may be overlooked. There was a problem in the case of a multi-valued system in which the noise was reduced.

The present invention has been made under the above circumstances, and a pattern inspection method capable of detecting a defect up to a size of about one pixel even if there is an alignment error between a pattern to be inspected and a reference pattern. The purpose is to provide.

[Means for solving problems]

In order to solve the above-mentioned problems, the present invention compares the respective gradient vectors obtained from the density distribution of the image of the pattern to be inspected with the density distribution of the image of the reference pattern.

[Action]

That is, according to the present invention, since the gradient vector represents the shape of the density distribution by the above means, the partial shapes of the two density distributions of the inspection pattern and the reference pattern can be compared by comparing the gradient vectors. As a result, when a shape that does not exist in the reference pattern appears in the pattern to be inspected, the difference can be detected even if the alignment error is considerably large.

〔Example〕

The present invention will be described below with reference to an embodiment shown in FIGS. 1 to 5.

FIG. 1 shows a part of the image 11 by the transmitted light of the pattern to be inspected by a multi-valued method. In the figure, x and y are spatial position coordinates and f (x _i , y _j ) is a pixel ( x _i , y _j ) density distribution (amount of transmitted light). In this case, the gradient vector A corresponding to gradf obtained from the multi-valued image 11 of the pattern to be inspected is calculated by the following equation; The map is shown in FIG.

Further, FIG. 3 is a diagram in which a part of the image 12 by the transmitted light of the reference pattern corresponding to the pattern to be inspected is created from the design data of the pattern to be inspected by a multi-valued method, where x and y are spaces. The target position coordinates, g (x _i , y _j ) respectively represent the density distribution of the pixel (x _i , g _j ). In this case, the gradient vector B corresponding to gradg obtained from the image of the multivalued reference pattern is calculated by the following formula; The map is shown in FIG.

Then, the density distribution f (x _i , y _j ) of the image 11 of the multi-valued inspected pattern and the density distribution of the image 12 of the multi-valued reference pattern described above.
Compare each gradient vector A and gradient vector B with g (x _i , y _j ). At this time, the gradient vector A is taken into consideration in consideration of the alignment error between the multivalued inspection pattern and the multivalued reference pattern.
(x _i , y _j ) is compared with the gradient vector B (x _i , y _j ) and its adjacent gradient vector B (x _k , y _l ), and the absolute value of their vector difference; | B (x _k , y _l ) −A (x _i , y _j ) | is the minimum value in the range of k = i-1, i, i + 1, l = j-1, j, j + 1 is C
(x _i , y _j ), the minimum value C (x _i , y _j ) of this gradient vector difference is obtained from FIGS. 2 and 4 and shown in FIG.

Next, a certain threshold value V is set for the minimum value C (x _i , y _j ) of the gradient vector difference described above, and C (x _i , y _j ) ≧ V If there is a pixel, the pattern to be inspected is "defective"
If V = 40, for example, the fifth
The shaded portion shown in the figure is determined to be a defect.

Therefore, as shown in FIG.
When there is a defect having an area corresponding to a pixel, such a defect cannot be detected by the conventional method of comparing only the densities in consideration of the alignment error of ± 1 pixel, but the present invention completely detects the defect. can do.

Further, FIG. 6 shows another embodiment of the present invention. In the case where an image 11 of a multi-valued pattern to be inspected has an isolated defect having an area corresponding to about 1 pixel, a corresponding reference pattern of The image has a total height of 0, and when the minimum value C (x _i , y _j ) of the gradient vector differences is obtained in the same manner as above, the value shown in FIG. 7 is obtained, and at this time, the threshold value V When 40 is set to 40 (V = 40), the pixels in the shaded area shown in FIG. 7 are detected as defects.

That is, in FIGS. 1 and 3, the density distributions of the image 11 of the multi-valued pattern to be inspected and the image 12 of the reference pattern are shown.
By comparing only f (x _i , y _j ) and g (x _i , y _j ), for example, the density of the pixel (x ₅ , y ₅ ) of the pattern to be inspected shown in FIG. Considering the error, the pixels (x ₄ , y ₄ ), (x ₄ , y ₅ ), (x ₄ , y ₆ ), (x ₅ , y ₄ ), (x ₅ ,, y ₅ ), (x
₅ , y ₆ ), (x ₆ , y ₄ ), (x ₆ , y ₅ ), (x ₆ , y ₆ ), and the density of 9 pixels is compared. In this case, the density of each pixel is isolated. The pixels of the pattern to be inspected (x
₍₅ , y ₅ ) pixels with almost the same density as the reference pattern
Since it is in (x ₄ , y ₄ ), (x ₄ , y ₅ ), (x ₄ , y ₆ ), the pixel (x ₅ , y ₅ ) of the pattern to be inspected is determined to be "not a defect",
Similarly, the surrounding pixels are also determined to be “no defect”. However, in the present invention, the gradient vector represents the magnitude and direction of the gradient of the surface, and represents how the density at that point is related to the density of the surrounding pixels, in other words, what shape the density distribution has. Since it is one amount, for example, the pixel (x ₅ , y ₅ ) of the pattern to be inspected shown in FIG.
Is judged to be so-called "jumping out", which means that the shape is judged by comparison with surrounding pixels, while the so-called "jumping out" is made to the pixels of the reference pattern shown in FIG. Therefore, it can be determined that they are different from each other, and such a determination is possible even if the alignment error is considerably large. That is, since the gradient vector is one quantity that represents the shape of the density distribution, comparing the gradient vectors means comparing the partial shapes of the density distributions of the two patterns. If a shape not present in the reference pattern appears in the pattern to be inspected, the difference can be found even if the alignment error is considerably large.

In the above embodiment, for example, as another formula for calculating the gradient vector A obtained from the density distribution f (x _i , y _j ) of the image of the multi-valued pattern to be inspected shown in FIG. According to the (Sobel) method, the following expression; A (x _i , y _j ) = [(f (x _{i + 1} , y _j-1 ) + 2f (x _{i + 1} , y _j ) + f (x _{i + 1} , y _{j + 1} ) -f (x _i-1 , y _i-1 ) -2f (x _i-1 , y _j ) -f (x _i-1 , y _{i + 1} )}, {f (x _{i -1} ,, y _{j + 1} ) + 2f (x _i , y _{j + 1} ) + f (x _{i + 1} , y _{j + 1} ) -f (x _i-1 , y _j-1 ) -2f (x _i , y _j-1 ) -f (x _{i + 1} , y _j-1 )}], the same is true for the gradient vector B, and the comparison method is limited to the directions of both vectors A and B. It is also possible to compare

Further, in the pattern inspection method according to the present invention, when the density change is gentle, even if a defect is large, it may not be detected. Therefore, if the conventional method is used in combination with a loose direct comparison inspection, a more complete inspection can be performed. It will be possible.

In addition, it goes without saying that the present invention can be variously modified and implemented without departing from the spirit of the present invention.

〔The invention's effect〕

As is apparent from the above description, according to the present invention, since the gradient vectors obtained from the density distributions of the images of the pattern to be inspected and the reference pattern are compared, even if there is a registration error, about 1 pixel It is possible to provide an excellent pattern inspection method capable of detecting even defects.

[Brief description of drawings]

FIG. 1 is an explanatory view showing an embodiment of a multi-valued inspected pattern in a pattern inspection method according to the present invention, and FIG. 2 is an explanatory view showing a map of a gradient vector similarly obtained from the multi-valued inspected pattern. FIG. 3 is an explanatory view showing a multi-valued reference pattern corresponding to the multi-valued inspected pattern, FIG. 4 is an explanatory view showing a map of a gradient vector similarly obtained from the multi-valued reference pattern, and FIG. 5 is a gradient. FIG. 6 is an explanatory view showing the map of the minimum value of the vector difference, FIG. 6 is an explanatory view of another multi-valued inspected pattern according to the present invention, and FIG. 7 is an explanatory view showing the map of the minimum value of the gradient vector difference, FIG. 8 is an explanatory view showing a state in which an image of a pattern obtained by an image pickup device is divided into minute pixels and binarized into the density of each pixel, FIG.
FIG. 9A is an explanatory diagram of the binarized reference pattern, and FIG. 9B is an explanatory diagram of the binarized inspected pattern. (x _i , y _j ) ... Pixel, 11 ... Image of multi-valued inspected pattern, f (x
_i , y _j ) ... density distribution, 12 ... multi-valued reference pattern image, g (x
_i , y _j ) ... concentration distribution.

Claims (2)

[Claims] 1. In a pattern inspection method for comparing and inspecting a pattern to be inspected and a reference pattern, respective gradient vectors obtained from the density distribution of the image of the pattern to be inspected and the density distribution of the image of the reference pattern are compared with each other. A pattern inspection method characterized by the above. 2. The pattern inspection method according to claim 1, wherein the image of the reference pattern is created from design data of the pattern to be inspected.