JPH04318912A - Pattern position detecting method - Google Patents

Abstract

PURPOSE:To perform pattern matching at high speed and with high accuracy. CONSTITUTION:By taking in a mark, whose shape is known, with a light view optical system, and adding it in the direction perpendicular to the cross section and differentiating it, and taking the absolute value, a measured pattern being divided into blocks is gotten, and based on the known mark shape, the operation with the presumed pattern being divided in blocks is performed, and based on it, the evaluation function concerning the position of the matching degree of the mark at large is set to detect the position.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention relates to a pattern matching method for position detection, and in particular, a pattern position detection method suitable for pattern position detection during alignment of semiconductor exposure equipment, which requires high accuracy. Regarding.

0002

[Prior Art] Generally, in semiconductor manufacturing equipment, in order to achieve high precision alignment between a wafer and a reticle, an ITV camera or CCD sensor captures a bright field image or a dark field image of a mark on an object. Alternatively, marks provided on an object are scanned with a laser beam, and reflected light, diffracted light, etc. are detected to perform alignment. especially,
When detecting a position based on image data, various improvements have been made to image processing to improve detection accuracy.

For example, the position where the degree of similarity between two-dimensional image data or one-dimensional data obtained by adding the image data in a predetermined one-axis direction and the estimated pattern estimated from the mark shape is maximum is determined. A template matching method is known in which the position is determined and the position is used as the mark position in the axial direction. With this method, once the estimated pattern is created, there is no need to take into account characteristics such as the size and shape of the mark when detecting the position, so there is no need to judge the mark position using conditional branching during processing. The content can be simplified.

An example of calculation of this template matching will be shown with reference to FIGS. 5A to 5D. Here, Figure 5
5A is a top view of the two-dimensional mark, and FIG. 5B is a sectional view thereof. Moreover, (C) of FIG. 5 shows an actually measured pattern obtained by adding data when such a two-dimensional mark is viewed with a bright field optical system in a direction perpendicular to the cross section. Here, the graph of this measured pattern is y=M
Let it be (x). Moreover, (D) of FIG. 5 shows an estimated pattern determined based on (A) and (B) of FIG. Let the graph of this estimated pattern be y=A(x). Note that in (D) of FIG. 5, s represents the start of the calculation process, and e represents the end of the calculation process.

[0005] There are various concrete forms of evaluation functions representing the degree of similarity in template matching. Here, it is assumed that a correlation coefficient used in statistical processing is used. Estimated pattern y=A(x) and measured pattern y=M
The correlation coefficient γ of (x) is γ=SAM/(SA ・SM) However,

[0006]

[Math 1] becomes.

[0007] The estimated pattern is scanned in the x direction (horizontal direction in the drawing), and the position where the correlation coefficient γ is maximum is determined as the position of the actually measured pattern in the x direction. The y direction can also be determined in the same manner.

[0008] Furthermore, Japanese Patent Laid-Open No. 61-278136 discloses a pattern position detection method in which a line-symmetric target pattern is used and a symmetry evaluation function is set to detect the center position of the line-symmetric pattern. .

That is, by taking the position in the horizontal axis x direction and the signal strength in the vertical axis y direction, the actually measured pattern at position I is expressed as y=X(I
), and the matching width is (2N+1), then the symmetry evaluation function Y(I) is

0010

[Math 2]

It is expressed as follows. However, f1 is a weighting function, which is determined to improve detection accuracy. I
The value of is changed, that is, the symmetry confirmation position is scanned in the x direction, and the position where this function is minimized is determined as the target line-symmetrical position.

[0012] Furthermore, the above publication discloses an example of emphasizing a peak portion with relatively little change in shape in an actually measured pattern, which is formed from the edge portion of a mark, in order to make the weighting function less susceptible to noise. There is.

[0013]

Problems to be Solved by the Invention However, the following problems occur in the actually obtained measured pattern of marks, which may result in deterioration of position detection accuracy or detection failure. (1) Random noise is added. (2) Mistaking another mark's pattern or a part thereof as the target pattern. (3) A pattern of another mark intersects a part of the target pattern, or another pattern hides the target pattern. (4) Pattern enlargement, contraction, shift, and pattern brightness changes occur due to process conditions, optical systems, illumination systems, etc.

[0014] Furthermore, in normal template matching as described above, when the estimated pattern size becomes large,
A problem arises in that front scanning for matching on captured images takes time. Furthermore, JP-A-61-2
The pattern position detection method disclosed in Japanese Patent No. 78136 has the following problems. (1) The target pattern or at least its detection portion must be line symmetrical.

(2) Even if it is not the target position, if a point with good symmetry is found, such as the center of another mark or a flat part away from the mark, it will be mistaken as the target position. Therefore, a means for roughly grasping the target position is required, or it is necessary to assume that the deviation between the detection site and the target position is slight.

(3) An example has been disclosed in which the peak part formed from the edge of the mark is emphasized in order to make it less susceptible to changes in the measured pattern due to noise, process conditions, etc., but if the measured pattern consists of two or more When a signal has a peak, it is necessary that the signal level at both ends of the peak and the shape of the peak are approximately equal; if this is not true, there can be only one peak.

The present invention has been made in view of the above points, and by setting an evaluation function representing a simplified degree of matching that is less susceptible to the influence of problems that occur, it is possible to achieve high accuracy and high speed. It is an object of the present invention to provide a pattern position detection method that can perform pattern matching and ease restrictions on position detection marks and their detection conditions, such as similar marks, mark spacing/arrangement, and mark intersections.

[0018]

[Means for Solving the Problems] In order to achieve the above object, in the pattern position detection method according to the present invention, for a mark with a known shape, an image-processed actual pattern is estimated based on the known mark shape. In pattern matching to identify and detect the position by comparing with the pattern, the estimated pattern is divided into blocks based on the pattern shape, calculations are performed on the divided parts with the actual pattern, and marks are made based on the results. Position detection is performed by setting an evaluation function regarding the position of the overall matching degree.

[0019]

[Operation] According to the pattern position detection method according to the present invention,
The estimated pattern is divided into blocks, and a number of specific combinations of blocks that are less affected by the problem are selected and processed using a matching degree evaluation function that is weighted to increase their weight. .

[0020] Therefore, by setting an evaluation function representing the degree of matching that is less susceptible to the problem that occurs and is simplified, pattern matching can be performed with high accuracy and at high speed. Restrictions regarding position detection marks and their detection conditions, such as placement and intersection of marks, can be relaxed.

[0021]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Generally, in semiconductor manufacturing equipment, not only precision but also high-speed processing is required, so even a two-dimensional pattern is compressed into two one-dimensional patterns related to two predetermined axes for processing. Therefore, although the embodiments described below will also be described in accordance with this example, it is possible to directly process data of any dimension by performing similarity calculation using data of a target dimension. Embodiments of the present invention will be described below with reference to the drawings. First, with reference to FIGS. 1A to 1F, the principle of the present invention will be explained, and two examples of the general form of the evaluation function will be explained.

FIG. 1A is a top view of the two-dimensional mark, and FIG. 1B is a cross-sectional view of the two-dimensional mark. Figure 1 (C) shows an actual measurement pattern obtained by capturing such two-dimensional marks with a bright field optical system (not shown), adding them in the direction perpendicular to the cross section, differentiating them, and taking the absolute value. It shows. This can also be obtained by setting the lowest level of shading of data captured by the dark field optical system to "0". By performing such processing, only the data in the edge portion has a value, and the rest becomes "0". This reduces the amount of data to be processed and further reduces the amount of data required per edge to 1
This image capture method is effective for the present invention because it can simultaneously divide the image into two blocks. However, with patterns in the process, an image similar to the one shown in the figure may be obtained even in bright field due to the effects of laminated layers and resist, but differential processing is not necessarily necessary for this. Here, the graph of this actually measured pattern is assumed to be y=M(x). Further, 1, 2, 3, and 4 in the figure represent peaks corresponding to the edges in (B) of FIG.

FIG. 1(D) is an estimated pattern calculated based on FIG. 1(A). Let the graph of this estimated pattern be y=A(x). In the figure, peak 1 with the same shape
, 2, 3, and 4, and the value is "0" in other parts. Also, s1, s2, s3, and s4 represent the beginning of each peak, and e1, e2, e3, and e4 represent the end of each peak, and the width d from the beginning to the end of each is the same length. . This peak is shown in Figure 1.
This corresponds to the mark edge part in (A), and the step difference in the mark appears as a peak pattern.

The specific shape of the estimated pattern in FIG. 1(D) is a normal distribution graph in the range of ±3σ shown in FIG. 1(E) with a width corresponding to the measured pattern at a position corresponding to each edge portion. , and the remaining part is created so that the value is "0".

The shape of this estimated pattern may be a simplified one such as the triangular peak shown in FIG. You can also use

In (D) of FIG. 1, a block division is adopted in which each of the remaining four peak portions, excluding the portion where the value is “0”, constitutes one block. As in this example, if you exclude the part where the signal strength value is "0" and are able to divide it into blocks, there is no problem, but
When there are many cases where "0" is not included in the mark, such as in a bright field image, or when you want to divide the mark into even smaller blocks, for example, divide the marks into blocks so that the distance between two peaks is divided by a certain ratio. can be taken.

When using the similarity between the measured pattern and the estimated pattern at each peak, the similarity for each peak is expressed as r1
, r2 , r3 , r4 , the evaluation function F1 is, for example,

[0028]

[Math 3] Totoru.

[0029] When using the similarity of peak combinations,
If the similarity is determined as rij for the combination of peaks i and j (1≦i<j≦4), then the evaluation function F
2 is, for example,

[0030]

[Mathematical 4] Also, by combining F1 and F2, for example,

[0031]

[Formula 5] can also be used. A similar development is possible when the number of peaks changes using different patterns. Next, an embodiment of the present invention to which the general form of the above evaluation function is applied will be described.

FIG. 2A shows how a mark 11 on a wafer seen through a reduction lens (not shown) is replaced by a mark 12 on a reticle during chip alignment of a semiconductor exposure apparatus.
It shows how it is adjusted relatively to. Here, the mark 12 on the reticle has four square transparent windows 13, through which the light that hits the mark 11 on the wafer is reflected back, and the mark 11 on the wafer can be seen through the windows 13. is now possible.

FIG. 2B is a diagram showing an actually measured pattern obtained by vertically or horizontally adding the images of FIG. 2A captured by a two-dimensional CCD (not shown). In the figure, in patterns 14 and 15 of marks on the reticle and wafer, the interval l between every other peak is equal, such as peaks (1, 3) and (2, 4). Furthermore, the signal intensity of the reticle pattern 14 is greater than that of the wafer pattern 15.

[0034] When detecting the position of each mark by pattern matching and detecting a shift, the two types of mark patterns may interfere with position detection. For example, one edge may be mistaken for the edge of another mark, or
If the relative positions of the two marks are slightly shifted, the edge of the mark 11 on the wafer goes behind the window 13 of the reticle and becomes invisible, and the peak of the measured pattern of the mark 11 on the wafer decreases accordingly.

Here, when template matching is performed using an evaluation function such as the following equation 6 using the estimated pattern of the wafer as shown in FIG. The maximum point of the evaluation function may occur at two different points. This state is shown in FIG. 2(C).

[0036]

[Math 6]

To deal with this, the peaks (1, 3), (2
, 4). That is, the simplest method is to calculate only the combination of peaks (1, 4) and (2, 3). F1,F
2 application examples are respectively F1 = r1 ・r4 + r2 ・r3
…
(1) F2 = r14 + r23

...(2). In our adaptation,
For equation (1), substitute "0" for the coefficients a, b, and d of equation 3 above, and equation (2) for equation 4 above, and for the weighting coefficient of c, c14=1, c22=1, other c=0

[0038] has been substituted. Regarding the above formula (1) 2
Since we are only dealing with the product of two similarities, we do not use the 1/2 power to simplify the equation. If a better weighting coefficient value is found through experiments or the like, it will be changed. Here, r
As an example of the basic calculation method for determining 1.

[0039]

[Math 7]

[0040]

[Math. 8]

[0041]

[Formula 9] can be mentioned. For equations 7 and 8 above, the position where the evaluation function is minimum is the target position, and for equation 9 above, the position where it is maximum is the target position. Next, r14, r
As an example of how to determine 23,

[0042]

[Math. 10] In addition, as a calculation method that takes advantage of symmetry,

[0043]

[Formula 11] can be mentioned. Regarding equations 10 and 11 above, the position where the evaluation function is maximum is the target position.

(A) and (B) in FIG. 3 are similar to (B) in FIG.
An example is shown in which the peak position of the measured pattern corresponding to the pattern is shifted. That is, FIGS. 3A and 3B are two measurement examples of actually measured patterns of marks having the same shape. The distance between peaks (1, 2) and (3, 4) is l1 (=11) in Figure 3 (A) and l in Figure 3 (B).
2 (=12). In this example, peak (1,
It is assumed that the center position of the combinations 2) and (3, 4) does not shift relative to changes in conditions. Further, regarding the combination of peaks (1, 3) and (2, 4), it is assumed that the shifts are in opposite directions and the amounts are equal. The exposure amount changes,
Such changes can occur due to changes in etching conditions during the process.

In this example, it is assumed that the positions O of the centroids of four peaks are determined. Let Cij be the coordinates of the center of gravity O obtained from the positional relationship between the two peaks i, j (1≦i<j≦4) in the estimated pattern and the center of gravity O and the positions of the peaks i, j in the measured pattern. , when combined so as to cancel the shift, the coordinate x of point O in the actual measurement pattern becomes x=(C12+C34)/2 or x=(C13+C24)/2 or x=(C14+C23)/2. If we weight and combine these
X={α・(C12+C34)+β・(C13+C24
)+γ・(C14+C23)} /
2 However, α+β+γ=1

[0046] The values of α, β, and γ shall be predicted or determined based on the measurement results so as to have the best accuracy. For example, in the template matching method,
Due to the difference in peak interval caused by the shift, C12, C
34, C14, and C23, the values of α and γ are set to “0”. In this example, to find Cij, F1 = ri ・rj or F2 = rij

The position where F1 or F2 is maximized is determined using the following equation. With this method, aberrations of the optical system,
It is possible to achieve alignment that is less affected by changes in mark width due to magnification errors, process changes, etc.

FIGS. 4A and 4B show an example in which another pattern or a part thereof enters the target pattern, or the target pattern enters the shadow of the window and the peak disappears. That is, (A) in FIG. 4 corresponds to the former case when there is an extra number of peaks, and (B) in FIG. 4 corresponds to the latter case when there is an insufficient number of peaks. In normal pattern matching, the possibility of erroneously recognizing relatively similar patterns increases as unrelated peaks are included in the detection, and the possibility of missing peaks increases the possibility of not recognizing the desired pattern. On the other hand, in this method, for example, in (A) of FIG. 4, the peak (1
, 2, 3, 4), and (1, 2) in Figure 4(B).
,3),(1,2,4),(1,3,4),(2,3,
This can be improved by setting an evaluation function that emphasizes the combination of 4).

As described above, in the present invention, the estimated pattern is divided into blocks, and some specific combinations of the blocks that are less susceptible to the effects are selected according to the problem that occurs, and weighted to increase their relative weight. Processing is performed using a matching degree evaluation function. Therefore, even if there is a change in the shape of the pattern to be detected, the detection ability can be improved if the change can be predicted in advance. Further, speeding up can be achieved by using only the minimum necessary data instead of all the data of the detected pattern. Therefore, when the present invention is applied to alignment patterns of semiconductor manufacturing equipment, it is useful for improving accuracy and productivity.

[0050]

[Effects of the Invention] As detailed above, according to the present invention,
By setting an evaluation function that represents the degree of matching that is less susceptible to the problem that occurs and is simplified, it is possible to perform pattern matching with high precision and high speed, and also to identify similar marks, mark spacing/arrangement, and mark intersections. It is possible to provide a pattern position detection method that makes it possible to relax restrictions regarding position detection marks and their detection conditions.

[Brief explanation of the drawing]

FIG. 1 is a diagram for explaining the principle of the present invention, in which (A) is a plan view of a two-dimensional mark, (B) is a cross-sectional view of the two-dimensional mark in (A), and (C) is a cross-sectional view of the two-dimensional mark in (A). (D) is a diagram showing the measured pattern obtained from the two-dimensional mark, (D) is a diagram showing the estimated pattern calculated based on the two-dimensional pattern in (A), (E) and (
F) is a diagram showing an example of a specific shape of the estimated pattern of (D).

FIG. 2 is a diagram for explaining one embodiment of the present invention;
) is a diagram showing how the mark on the wafer is aligned relative to the mark on the reticle, (B) is a diagram showing (A)
(C) is a diagram showing the measured pattern obtained from the image of (B
) is a graph showing the relationship between the position and the evaluation function when an estimated pattern such as that shown in FIG. 1 (D) is used for the actually measured pattern.

FIGS. 3A and 3B are diagrams showing actually measured patterns when the peak position is shifted, respectively.

FIG. 4A is a diagram showing an actually measured pattern when there is an extra number of peaks, and FIG. 4B is a diagram showing an actually measured pattern when the number of peaks is insufficient.

[Figure 5] (A) is a plan view of the two-dimensional mark, (B) is (A
), (C) is a diagram showing an actually measured pattern using a bright field optical system, and (D) is a diagram showing an estimated pattern using a bright field optical system.

[Explanation of symbols]

11... Mark on wafer, 12... Mark on reticle, 13
...Window, 14...Reticle pattern, 15...Wafer pattern.

Claims (7)

[Claims] 1. In pattern matching for identifying and position detecting a mark having a known shape by comparing an image-processed measured pattern with an estimated pattern based on the known mark shape, the estimated pattern is is divided into blocks based on the pattern shape, calculations are performed on the divided parts with the actually measured pattern, and based on the calculation, an evaluation function regarding the position of the matching degree of the entire mark is set, and position detection is performed. pattern position detection method. 2. The pattern according to claim 1, wherein the evaluation function is obtained by weighting and adding similarities of arbitrary blocks or combinations of blocks among the divided parts. Location detection method. 3. The evaluation function calculates the degree of similarity between the measured pattern and the estimated pattern for each block divided into blocks or a combination of arbitrary blocks, and performs an operation between each degree of similarity. 2. The pattern position detection method according to claim 1, wherein the pattern position detection method is an evaluation function of the matching degree of the entire mark. 4. The pattern position detection according to claim 3, wherein the evaluation function is a product of any of the similarities or a product obtained by adding an operation to the product, weighted and added. Method. 5. The pattern position detection method according to claim 1, wherein the block division is performed on the remaining portion of the estimated pattern after excluding a portion where the gradation or binary gradation is 0. . 6. The pattern position detection method according to claim 5, wherein the dividing into blocks is performed using a dark field optical system. 7. The pattern position detection method according to claim 5, wherein the dividing into blocks is performed by image processing using a bright field optical system and differentiation.