JPH0338783A - Image positioning method - Google Patents

Abstract

PURPOSE:To enhance positioning sensitivity by dividing the pattern element of a reference image into two areas with equal area in a lateral direction, and performing positioning by using the histogram of the area of a light-and- shade image whose positioning relation corresponds to respective area. CONSTITUTION:The areas 4 and 5 are formed by dividing the pattern part of the reference image 3 into two equally, and the light-and-shade image 1 is superimposed on the reference image 2, and the area that is the pattern area 2 of the light-and-shade image 1 and also, the pattern area 4 of the reference image 3 is designated as the area 2*4, and the area that is the pattern area 2 of the light-and-shade image 1 and also, the pattern area 5 of the reference image 3 as the area 2*5. And a position where the sum of the absolute values 8 of the difference between the histogram 6 of the light-and-shade image in the area 2*4 and the histogram 7 of the light-and-shade image in the area adding the pattern area 5 and the area 2*5 goes to a minimum is found. Thereby, the positioning can be executed with high sensitivity.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical field to which the invention pertains) The present invention relates to a method for aligning a grayscale image and a reference image thereof.

(Prior Art) One of the conventional methods for inspecting LSI patterns is to compare and match a grayscale image obtained by a scanning electron microscope (SEM) with a designed pattern.

This requires that the pattern observed in the gray scale image and the design pattern, which is usually given in binary, be aligned with high precision before comparison and verification.

Conventionally, the following two methods have been considered as methods for this alignment. One method is to binarize grayscale images and then align the binarized images with each other. However, S
If there is a large amount of noise, such as in an E'M image, the accuracy of binarization is low, resulting in a disadvantage that the alignment accuracy is low.

Another method is to find the relative positional relationship that maximizes the product-sum calculation value of the grayscale image and the reference image. However, this method has the disadvantage that if the contrast of the grayscale image is poor, the rate of change in the product-sum calculation value, which is calculated by dividing the amount of change in the product-sum calculation value due to changes in relative positional relationship by the product-sum calculation value at that position, is low. be.

For example, if the density per unit area of the pattern area is A, the density per unit area of the non-pattern area is B, and the matching ratio between the pattern area and the non-pattern area is α, then
The rate of change in the product-sum calculation value is (A-B)/'(αA+(1-α
)B)...(1).

Here, when α→1, it becomes (AB)/A.

Now, let us assume that the area change ratio during fine alignment adjustment is 1%, and A=
150. When B=100, the observed signal of the product-sum calculation change rate is only about 0.3%. As described above, the method using the product-sum calculation of the grayscale image and the reference image does not have sufficient alignment sensitivity and has low accuracy.

(Objective of the Invention) An object of the present invention is to provide an image alignment method that does not require the binarization method of grayscale images as in the above-mentioned prior art and has higher alignment sensitivity than the method using product-sum calculation values. do.

(Structure of the Invention) (Characteristics of the Invention and Differences from the Prior Art) In order to achieve the above-mentioned object, the present invention has the following objectives: one or more of the 3-pattern elements of the reference image, each having an approximately equal area in the horizontal and vertical directions. Two areas A1. The main feature is that alignment is performed using a histogram of regions of a gray scale image whose positional relationship corresponds to each region.

It is not necessary to binarize grayscale images as in conventional technology,
Another difference is that the alignment sensitivity is higher than the method of determining the relative positional relationship that maximizes the product-sum calculation value of the grayscale image and the reference image.

(Example) Fig. 1 is a diagram explaining the operating principle of the present invention, (a)
indicates a grayscale image, 〇 is the grayscale image, and 2 is the area of the pattern portion. 3B shows a reference image, 3 is the reference image, and 4 and 5 are areas A1 and A2 obtained by dividing the pattern portion into two equal areas.

Figure (c) is a superimposition of the above figure (,) and figure (b), and 2*4 in the figure is pattern area 2 of grayscale image 1,
and represents the pattern area 4 of the reference image 3.

Similarly, 2*5 represents pattern area 2 of grayscale image 1 and pattern area 5 of reference image 3.

4 In addition, the figure (d) is a histogram diagram, and the numerical symbol 6
represents the histogram of the grayscale image in the 2x4 area shown in the above figure (c), and the number symbol 7 represents the histogram of the grayscale image in the combined area of the pattern area 5 and 2x5 area, respectively. . FIG. 6(e) shows the absolute value 8 of the difference between the two histograms 6 and 7.

The vertical axes in FIGS. 1(d) and (e) above represent frequency and the horizontal axis, and the 2nd axis in FIG. 1(c) for the pattern area 5 divided into two equal parts shown in FIG. 1(b) *The proportion of 5 areas is S
shall be. Further, assuming that the grayscale image l is a non-pattern and the amount of distribution expansion of its histogram is γ, the absolute value 8 of the difference p in the histogram 6.7 in FIG. 1(e) is given by the following equation.

Therefore, the area of the shaded part with an absolute value of 8 is given by E in parentheses [ ] in equation (3). Changes in the area of the diagonal line in e).

The alignment sensitivity F divided by h is expressed by the following formula (4).

From the above equation (4), in S-+l, which requires fine alignment,
F→−ω. That is, the alignment sensitivity F is high. Also,
When S->1, h→0. Therefore, by using the area h, that is, the sum of the absolute values 8 of the histogram differences p as a parameter, and finding the position where this parameter is minimum, alignment can be performed with high sensitivity.

FIG. 2 is an explanatory diagram for creating a dividing line 9 that bisects the pattern areas 4 and 5 of the solid reference image 3 shown in FIG. 1(b). That is, in the reference image 3, point a is the start point of the run length on the same scanning line, point b is the end point of the run length on the same scanning line, and point C is the midpoint between the start point a and the end point. The broken line 9 is a dividing line obtained by tangent to the midpoint C on each scanning line, and the area 10.1 divided by this dividing line 9 is
1 corresponds to pattern areas 4 and 5 in FIG. 1(b). In this way, a dividing line can be created by connecting the start point a and the midpoint C of the end point of the run length.

FIG. 3 shows a block diagram of an image alignment device according to an embodiment of the present invention.

In the figure, 12 is a grayscale image input section, 13 is a grayscale image recording section, 14 is a reference image input section, 15 is a reference image recording section, and 16 is a pattern area selection section. This is an image selection unit that reduces the image area used for alignment to reduce the amount of processing when there is no need to do so, and is not necessarily a necessary element. 17 is a pattern dividing section;
The pattern portion of the reference image 3 shown in the figure is divided into two equal areas. Reference numeral 18 indicates a relative position changing unit for changing the relative position of the grayscale image 1 and the reference image 3; 19 indicates a histogram unit A in one area of the divided pattern; 19 indicates a histogram unit B in the other area of the divided pattern; 21 22 is an absolute value summation unit that divides the sum of the insulation values 8 of the histogram difference p; 23 is a determination unit;
record the output value, extract the relative position parameter in the relative position change unit 18 whose value is the minimum, and if the minimum value is not found, change the relative position parameter in the relative position change unit 18. . This results in
The relative positional relationship between the grayscale image 1 and the reference image 3 can be determined from the absolute value 8 of the difference between the histograms 6.7 shown in FIGS. 1(d) and 1(e).

FIG. 4 shows a block diagram of an embodiment of the solid pattern dividing section 17 shown in FIG. In the figure, 171 is a run-length circuit, and 172 is a starting point a shown in FIG. 2 for each scanning line obtained by the run-length circuit 171.

A table 173 for recording the end point is the starting point a.

This is a midpoint calculation circuit that calculates the midpoint C from the end point.

Reference numeral 174 is a starting point/end point changing circuit, which sets the starting point on one scanning line as the starting point of area 11 in FIG.
1 to the start point of area 10, or run length circuit 1
The starting point obtained in step 71 is used as the ending point of the area IO to create a new starting point/end point table. This table represents the divided areas 10.11 of the pattern area of the reference image 3. 1
75 is a circuit that outputs the divided images of the start point/end point changing circuit 174, and outputs them to the relative position changing section 18.

In the above description of the embodiment, the pattern area of the reference image 3 is divided (see FIG. 2) in the horizontal direction, that is, by the vertical dividing line 9, and the relative positional relationship between the grayscale image 1 and the reference image 3 is Positioning was performed by shifting in the lateral direction. On the other hand, alignment can also be performed by dividing the dividing line 9 in the horizontal direction, that is, in the vertical direction, and shifting the relative positional relationship between the grayscale image 1 and the reference image 3 in the vertical direction. By using them together, it is also possible to correct positional deviations in two-dimensional directions.

Next, a case where the pattern width of the reference image 3 and the pattern width of the grayscale image 1 are different will be described. For example, reference image 3
When the pattern width of the reference image is longer, the dividing line 9 of the pattern area of the reference image is close to the center of the pattern area of the grayscale image, and the deviation from the center is Δ, and the pattern width of the reference image d raf, Assuming that the pattern width in the grayscale image is d, the histogram of the pattern area 4 shown in FIG. 1(b) is expressed by the following equation.

However, t = 1 - (two negative ships - Δ) The histogram of the pattern area 5 shown in FIG. 1(b) is expressed by the following equation.

However, t'=i-(1"! [2A-10Δ)2 Therefore, the histogram difference p is given by the following equation.

Therefore.

The area of the histogram difference p is given by the following formula, and when the deviation Δ→O, h→0, and the alignment sensitivity F is 1, = length h/d mutually = work h Δ, so alignment can be performed with high precision. can be done.

However, if the pattern width of the reference image 3 is smaller than the pattern width of the grayscale image 1, contrary to the above, correct positioning cannot be performed. In this case, the pattern of the reference image may be expanded or thickened, and the result may be used as a new reference image.

(Effects of the Invention) 11- As explained above, according to the present invention, there is no need to binarize a grayscale image, and alignment sensitivity is high and alignment can be performed with high precision.

In the present invention, the reference image is a binarized image, but even if the reference image is given as a grayscale image, if the reference image can be binarized, it is converted to a binary reference image and then aligned. . Even if the reference image is difficult to binarize, the position of the equal area dividing line can be determined.

Alternatively, if sufficient prediction is possible, a reference pattern can be formed from the dividing line for alignment.

[Brief explanation of drawings]

The first drawing is a diagram explaining the operating principle of the present invention, and the second drawing is the first drawing.
An explanatory diagram for creating a dividing line that bisects the pattern area of the solid reference image 3 shown in FIG. 3(b), FIG. 4 is a block diagram of an embodiment of the pattern dividing section 17 in FIG. 3. FIG. Technique: Grayscale image, 2: Pattern area, 3:
... Reference image, 4, 5 ... Pattern region 12 Area A1. A2, 6, 7... Histogram of the grayscale image, 8... Absolute value of the difference between the histograms, 9 ・
... Parting line, 10.11... Divided area, 12... Grayscale image input section, 13.15... Recording section, 14... Reference image input section, 16... Pattern area selection section, 17.
...Pattern dividing section, 18...Relative position changing section, 19
... Histogram section A, 20... Histogram section B, 21... Histogram difference calculation section, 22...
Absolute value summation unit, 23... Judgment unit, 171... Run length circuit, 172... Start point end point table, 173.
...Middle point calculation circuit, 174...Start point end point change circuit, 1
75...Division side output circuit. Leaves of mourning

Claims (1)

[Claims] In aligning the grayscale image and the reference image, one or more of the reference image pattern elements is divided into two areas A_1 and A_2 of approximately equal area in the horizontal or vertical direction, respectively, and then the areas A_1 and A_2 are divided into areas A_1 and A_2 of the grayscale image. Obtain the histogram of the portion that overlaps with each of A_2, use the sum of the absolute values of the differences between the two histograms as a parameter representing the positional shift between the grayscale image and the reference image, and calculate the relative relationship between the grayscale image and the reference image so that this parameter is minimized. An image alignment method characterized by correcting a positional shift between a grayscale image and a reference image by changing their positions.