JPH04194702A - Pattern recognizing method - Google Patents

Abstract

PURPOSE:To make a pattern alignment accuracy smaller than the size of a detected pixel and reduce the effect of sampling errors by extracting and comparing the errors of two patterns whose positions are aligned with each other with the accuracy not larger than the size of one pixel. CONSTITUTION:The pattern of a wafer 1 illuminated by a light source 3 and an illumination optical system 4 in synchronism with the scanning of an X-Y stage under a command from a total controller 21 is photoelectrically converted by a one- dimensional image sensor 6 via an objective lens 5 to detect a two-dimensional pattern and a binary-coded 7 two-dimensional picture image 10 is obtained to be stored 8. A matching unit 13 calculates a difference between the detected picture image 10 and a comparison picture image 11 and outputs the picture image 14 wherein pixel unit position correction is terminated. Then, a sub-pixel matching unit 15 calculates the position alignment quantity of not larger than the size of a pixel. A position aligning unit 16 conducts position alignment on the basis of a position alignment quantity. A picture image extracting unit 18 extracts a difference picture image from the picture image wherein a position correction is terminated. The difference image 17 is binary-coded at a threshold value in a defect decision unit 19, a variety of characteristic quantities are extracted and defect decision is performed.

Description

[Detailed Description of the Invention] [Industrial Application Field 1] The present invention generally relates to a method for recognizing defects by comparing patterns on LSI wafers and TPT, and is used for highly accurate positioning and alignment. This invention relates to a pattern recognition method for highly accurate position detection of marks, etc.

[Prior art] A conventional pattern recognition method is disclosed in Japanese Patent Application Laid-Open No. 57-196377.
As described in the publication, the target pattern was detected, the detected pattern was memorized, and the previously stored pattern and the detected pattern were aligned pixel by pixel. Defects in patterns can be recognized by extracting and comparing errors between two patterns.

This recognition target is a pattern of a semiconductor wafer such as a memory LSI, as illustrated in FIGS. 2(a), 2(b), and 2(c).
AnsisLer) patterns, printed wiring board patterns, ceramic substrate patterns, and patterns of masks, reticles, etc. used in the process of manufacturing them. Here, a semiconductor wafer pattern will be described as an example, but the same holds true for other patterns. The pattern of a semiconductor wafer is that dozens of chips are mounted on each wafer, and each chip has the same pattern. The principle of recognizing such pattern defects is shown in Figure 2(a)~
Explain 4 using (C).

2(a) to 2(C) are explanatory diagrams of the principle of the conventional general pattern comparison method. FIG. 2(a) shows the memory pattern,
Figure (b) shows the detection pattern, and Figure 2 (C) shows the pattern difference. Focusing on the fact that each chip has exactly the same pattern, detect and memorize the pattern in Figure 2(a), and then create another pattern that should be the same as that in Figure 2(b). Next, the two patterns are detected and aligned pixel by pixel, and the error between the two aligned patterns shown in FIG. 2(c) is extracted and compared. If there is no defect in any of the patterns, there is almost no difference between the patterns, but if there is a defect in any of the patterns, for example, the detection pattern in FIG. ), there are differences in the pattern at the defective part, so pattern defects can be recognized by comparing the patterns and detecting the location where the error occurs. Note that if there is a difference in this comparison, it can be said that one of the patterns has a defect, but it is not possible to determine which pattern has a defect.

[Problem to be solved by the invention]

In the above-mentioned conventional technology, since the image is digitized and input, only information on sampling points can be obtained, and the occurrence of sampling errors is unavoidable, but there is a problem in that it is difficult to recognize small defects due to the influence of these sampling errors. This will be explained with reference to FIGS. 3(a) to (C).

Figures 3(a) to (c) are x- in Figures 2(a) to (c).
FIG. 3(a) is a waveform diagram of the pattern on the x' edge line, and FIG. 3(a) is a detection signal waveform diagram of the memory pattern of FIG. 2(a), and FIG. 3(b)
is the detection signal waveform diagram of the detection pattern in Fig. 2(b),
Figure (C) is a detection signal waveform diagram when there is no sampling error in the detection pattern of Figure 2 (b), and Figure 3 (d) is a diagram of the detection signal waveform when there is no sampling error in the detection pattern of Figure 2 (b).
Difference signal waveform diagrams between Fig. 3(a) and Fig. 3(b) with sampling error, Fig. 3(e) are diagrams of Fig. 3(a) and Fig. 3(C).
) is a difference signal waveform diagram with no sampling error. In the figure, . indicates a detection signal at a sampling point. As shown in Figure 3 (b) and (c), it is not possible to set the same point at the same point for the exact same pattern, so the detected waveform at the sampling point will be different, resulting in errors in the detected waveform and pattern. This error is called sampling error. If there is no sampling error in the detection pattern to be compared, the defective part difference signal is sufficiently larger than the normal part difference signal as shown in Fig. 3(e), and the defect can be easily recognized. However, if the detection pattern has a sampling error, Figure 3(d)
As shown in the figure, the defective part difference signal is of the same level as the normal part difference signal, making it difficult to recognize the defect. If the defect size to be recognized is sufficiently large compared to the pixel size at the time of detection, sampling errors and defects can be identified by using the difference in the area of the place where the difference is large, but if it is small, the area of the place where the difference is large caused by the defect. The error is about the same as that due to sampling error, making it difficult to identify defects.

SUMMARY OF THE INVENTION An object of the present invention is to provide a pattern recognition method that can reduce the influence of sampling errors by making the accuracy of pattern alignment less than or equal to the detection pixel size.

[Means for Solving the Problems] In order to achieve the above object, the pattern recognition method of the present invention aligns patterns in pixel units, and then performs alignment with an accuracy less than that in pixel units. . In other words, a target pattern is detected, the detected pattern is memorized, the previously stored pattern and the detected pattern are aligned pixel by pixel, and the pattern that was previously aligned pixel by pixel is This method identifies defects in the patterns by extracting and comparing the errors between the two patterns that have been aligned with sub-pixel accuracy.

In order to perform positioning with precision less than this pixel unit, for example, the following least squares method is used. The two patterns are f(
xty), g(x, y), (1
) The position (dxo,
dyo) in pixel units so that there is a position (δxO, JyO) where the difference between the patterns of the detected image and the stored image is minimized between values O and 1 for both x and y coordinates.

E 2 (dx, dy) = ε (dx, dy) + ε (dx+
I, dy)+E (dx, dy+1)+ε(dx+l,
dy+I) HHHH(1)ε(dx,dy)・ΣΣI
f(x,y)-g(x+dx,y+ay)1- , ,
, , , (2) Here, X, y are the coordinates of the pattern in pixel units, dx.

dy is the alignment amount of the two patterns in pixel units, cl
xo, dyo are the alignment amounts below the pixel that minimize C2 dx, dy, Jx, δy are the alignment amounts below the pixel, δxO2δyO are the alignment amounts below the pixel that minimize the pattern difference δX, δy, ΣΣ Each shows the sum of the x and y coordinates of the range to be aligned.

To perform alignment in pixel units, gl, which is obtained by shifting g shown in the following equation (3), is used.

gl (x, y)□g(x+muty+dy)・・・・・・
・・・・・・・・・(3) The intermediate value between pixels (4)
, defined by equation (5).

fd(XIδx+y,dy)=f(x,i+δx books(f
(x+l,y)-f(x,y))+Jyhonge('+y"
1)-f (x, i) (4) gld(x, δX+ yy
δy)=gl (x, i+δχ book(gl(x-Ly)-
gl (xty)) + δy books (gl (x, y-]) - gl
(x, y)) (5) The squared error can be defined by equation (6).

εd(Jx, δ=ΣΣ(fd(x, δx, y, δy)
-gld(x, δX, yTδ) Book 12
(6) (6) Expression 0 as δX, '
Partially differentiate with respect to V and set this as 0 to organize (7)
, we obtain equation (8).

(ΣΣCO book Cj) book (ΣΣC1 zero Cj) - (ΣΣCO
Book Ci) (ΣΣCj 小Cj) δxO= −−−−−−
−−−−−−−−1−−−−−−−(7) (ΣΣCO 京Cj) Wealth (ΣΣCi book Cj) - (ΣΣCO book Ci) book (
ΣΣCj end Cj) (ΣΣCO book Ci) book (ΣΣCi book C
j) - (ΣΣCO*Cj) zero (ΣΣCj zero Cj) δxO
= −−−−−−−−−−−−−−−−−−−−−−(
8) (OCO Ning Cj)
Ci) Book (OCj<j) Here, C0=f(x,y)-gl(x,y)Ci=ge(x+
l,y)-f(x,y))-(gl(x-1,y)-g
l(x,y)) (9)Cj=ge(X,y+1)
-f(X,y))-(gl(X,y-1)-gl(X,
y)) Pattern f2 after alignment based on the alignment amount δxo and JyO below the pixel. g2 as the following (10), (
II) Calculate using the formula.

f2(xty)=fd(x, δxo, y, 8yo)
(10) g2(x,y)=gld(
x, δxo, yJyo) (II)
[Operation] The operation of this pattern recognition method is shown in Fig. 4 (a) to (d'1).
This will be explained with reference to FIG.

4(a) to 4(d) are waveform diagrams of an operation example of sub-pixel alignment in which alignment is performed at a pitch less than the pixel of the pattern of FIG. 2 according to the present invention; FIG. 4(a) is a stored waveform;
Figure 4(b) is a detected waveform with no defects, Figure 4(C) is a simple difference waveform with only pixel-by-pixel alignment, and Figure 4(d) is a detected waveform with no defects.
) is the difference waveform after subpixel alignment. FIG. 5 is a numerical table diagram of FIGS. 4(a) to 4(d), including the stored waveforms, detected waveforms, simple difference waveforms, and f2. g2. The numerical values of lf2-g21 are shown respectively. For example, if the stored waveform and the detected waveform are
Assume that the inspection waveform in Figure 4 (b) is as shown in Figures (a), (b), and Figure 5, and the inspection waveform in Figure 4 (b) is the average of the two pixels before and after the memory waveform in Figure 4 (a). This is equivalent to a waveform shifted by approximately 0.5 pixel.

Applying the least squares method to these waveforms to find δXO and actually calculating it, δxO=0.2, and from this value (t
o), pattern f after alignment using equation (11)
2. Find g2. At this time, the difference signal waveforms with and without subpixel alignment are shown in Figure 4 (C).
(d) and FIG. 5, and the residual error has been reduced by half. As a result, the value of the pattern difference due to the sampling error becomes sufficiently smaller than the value of the defect, so that the defect can be easily identified.

[Example] Hereinafter, a first example of the present invention will be described with reference to FIGS. 1 and 6. FIG. 1 is a block diagram of an LSI wafer pattern recognition device. This pattern recognition apparatus includes an XY stage 2 that scans a wafer 1, a light source 3 that illuminates the wafer, an illumination optical system 4, an objective lens 5 that detects an optical image of the illuminated wafer, and a detection section that includes a one-dimensional image sensor 6. ,
- An image input section 9 consisting of an A/D converter 7 for digitizing and storing the signal of the dimensional image sensor 6 and an image memory section 8; The image extraction unit 12 retrieves the image from the image memory unit 8, calculates the difference between the detected image 10 and the comparison image 11 using equation (2), and moves the comparison image as shown in equation (3) for alignment. From the pixel unit matching unit 13 and the image 14 after the pixel unit position correction from the pixel unit matching unit 13 and the detected image lO, (7)
Alignment amount δxO1δ below the pixel expressed by equation (8)
Based on the alignment amount from the sub-vixel matching unit 15 and sub-pixel matching unit that calculates yO (10)
, a difference image 17. between the alignment unit 16 that performs position correction expressed by equation (11) and the aligned image 17. An image processing unit 2o that includes a defect determination unit 19 that binarizes the difference image 17 and extracts various feature amounts of the location where the difference exists to determine defects; and an image processing unit 20 that controls the XY stage 2.
The control unit 21 is comprised of a computer that stores and displays defect information output from the controller and manages the overall sequence.

The operation of detecting pattern defects using this configuration will be described next. First, after initializing each part according to a command from the overall control unit 21, in synchronization with the scanning of the xY stage 2, the pattern of the wafer 1 illuminated by the light source 3 and the illumination optical system 4 is imaged as a -dimensional image via the objective lens 5. A two-dimensional pattern is detected by photoelectric conversion with the sensor 6, and digitized with the A/D converter 7 as a two-dimensional detected image 1o.
The obtained detected image is stored in the image memory section 8.

The image retrieval unit 12 retrieves the comparison image 11 by referring to a fixed address in the image memory unit 8. Here, the operation of the pixel unit matching section 13 will be explained with reference to FIG. FIG. 6 is an explanatory diagram of the operating principle of the matching section 13 shown in FIG. 1. From the detected image lo and the comparison image 11, the comparison image is determined by ±δ pixels of the positional deviation tolerance in the △X and △Y directions (in this example, δ = 1, but depending on the dimensional accuracy of the detection target and the defect detection device) Calculate the difference between the detected image lO and the comparison image 11 when the image is shifted by a value determined by positioning accuracy, and the necessary value shall be set appropriately) using equation (2),
dxo that minimizes ε2 in equation (1) (Sl in FIG. 6).

Calculate ayo (△X=-1, △y=o in Figure 6),
The position of the comparison image 11 is corrected using equation (3), and the image 14 after the pixel-by-pixel position correction is output.

The sub-pixel matching unit 15 calculates the sub-pixel alignment amount δxO1δyO expressed by equations (7) and (8) from the image 14 after the pixel unit position correction from the pixel unit matching unit 13 and the detected image lO. . Positioning section 1
6 performs position correction expressed by equation (11) based on the alignment amount from the sub-pixel matching unit I5 (to). The difference image extracting unit I8 extracts a difference image 17 from the image whose position has been corrected using the following equation (12).

S (i, j)・1f2(i, j)−g2(i, j)
(12) The defect determination unit 19 binarizes the difference image 17 using the defect determination threshold Vth, extracts various feature quantities such as area 1 width and projected length of the location where the difference exists, and determines a defect. .

According to this embodiment, since the two patterns of the detected image and the stored image are both moved in opposite directions by the same amount to create a sub-pixel aligned image, the image smoothing effect (for example, δxo= 0.5. f2(x,
y) □ (equivalent to applying an average value filter by f(x+1.y)+f(x,y))/2) is the same for the two patterns, and the error in the difference image caused by this is minimized. It has the effect of

The first modification of this example is (to), replacing equation (11) with (1
3).

Formula (14) is used.

f2(x,y)・f(x,y)
(13) g2 (x, y): f (x, y) + δχ Honge (x-1, y) - f (X + i)
+2 δF(f(xpy-1)-f(xpy))if
O, 0≦Ax<0.25, 0.0≦δy<o, 2sf
(x+1.y) + (1-δ) 0 (f (xty) i (x-1, y)) + 2 δyt (x-1, y-1) -
f(x-1,y))if 0.25≦δx<0.5,0
．． 0≦δy〈0.25f(X, y+1)+δX spill(x
−1,y−1)−f(x,y−1))”(1−2dδy
door(f(x,y)-f(x,y-1))if O,0
≦δx<0.25, 0.25≦δy<o, 5f(x+l
, y+1) + (1-Jx) Honge (x p y-1) −
f (x-1r y-1) ) + (1-21δy) (x-1, y) - f (x-1, y-1)) if 0.2
According to the five-line modification, f2.
The value of g2 can be made more continuous. That is, for the data example of g shown in FIG. 7, dx·0. In the case of δX・0.49 and dx=1. δx=0. FIG. 7 shows the values of g2 with and without this modification in the case of O1. If this transformation is not carried out, the values will be significantly different, but if this transformation is carried out, the values will be almost the same.

In addition, in a second modification of this embodiment, instead of detecting a two-dimensional pattern by photoelectric conversion using the one-dimensional image sensor 14 in synchronization with the scanning of the XY stage IO,
A two-dimensional pattern is detected by moving the stage IO in steps and photoelectrically converting it with a TV camera. Alternatively, instead of the -dimensional image sensor 14, various types of sensors can be used, such as using a point type sensor such as a photomultiplier and a scanning mechanism. Furthermore, the image processing section does not have to be entirely hardware, but can also be composed of hardware and software.

In addition, the third modification of this embodiment calculates the difference between the detected image and the stored image using equation (2), and instead of outputting the difference between the images corresponding to each shift amount as a matching value, the difference between the detected image and the comparison image is calculated. Extract edges etc. by applying a filter to each, and calculate the image difference between the filtered images (2)
Calculate using the formula, f2. g2 is calculated for the image before applying the filter.

In addition, the c back detection image and the comparison image are each filtered and binarized, and the difference between the images is calculated (
2) Calculate by formula, f2. g2 is calculated for the image before applying the filter. According to this modification, since a filter is used, there is an effect of making the detection image less susceptible to the difference in unnecessary information between the detected image and the comparison image.

Moreover, the fourth modification of this example is (1), (4), (
5).

(15-1), (15-4), (15-
5), replace with (15-9).

ε2 (dx, dy) = ΣΣE (dx+nx, dy+n
y) (15-1) fd(x, Jx, y,
δy)=f(x,y)+δ)F<x+ntV)-f(
x, y)) + δy books (f(x, y+n) - f(x, y)
) (15-4) gld(x, δ
X1yt 'y) = gt (xty) + δx books (gl
(x-n, y)-gl (xty) + δy books (gl(
x, y-n)-gl(x, y)) (1
5-5) Co=f(x,y)-gl(x,y) Ci=(f(x+n,y)-f(x,y))-(gl(
x-n,y)-gl(x,y)) (15-9)Cj
−(f(x,y+n)−f(x,i)−(gl(x,
y-n)-gl (x, y)) Here, n is the pitch of calculation of intermediate values between pixels, n-1, 2, 3,... ΣΣ in equation (1) is nx, It means the sum of ny from 0 to n.

ΣΣ in equations (6) and (7) is L' of the alignment range
Indicates all sums for l coordinates or sums for every n.

Minimize t 2 (dx, dy) in equation (1) (dx,
dy) pixel by pixel, and find the position ((
δX, δy).

This modification has the feature that the range that is aligned in advance by the pixel unit matching section can be rough.

Furthermore, the fifth modification of this embodiment is the pixel unit matching unit 1.
3 is not used. Depending on the conditions of the object, the position (δX, δy) that minimizes ε2(dx, dy) in equation (1) is between 0 and 1 in both x and y coordinates even without performing pixel-by-pixel matching. In this case, pixel-by-pixel alignment is not necessary.

In particular, if the fourth modification of this embodiment is adopted, there is a high possibility that this will become unnecessary. This modification is characterized by a simple configuration.

Further, in the sixth modification of this embodiment, the same method as the sub-vixel matching section 15 is used for the pixel-by-pixel matching section 13. At this time, the third transformation and the fourth transformation are generally added to the pixel-by-pixel matching section 13. That is, after shifting the stored image g by n pixels using equation (16), calculations are performed on f and go at the pitch n of pixel calculation by applying the average value filter of equation (17). Also, equation (20), (
11) uses equations (18) and (19).

According to this modification, since there are only two stages of sub-vixel matching, there are two sets of circuits with the same configuration when implemented in hardware, so it is characterized by good efficiency.

gS(Xt'/)"g(x+Q,y"n)
(1
6) fo (X, y) = ΣΣf (x+i, y+j) go (
x, y) Σgs(x+i, y+j)
(17) ΣΣ represents the sum of i, j=-m−m.

f2(x,y)・f(x,δXO+y+δyO)
, (18) g2(x+y)=gs
(x++, JxO, y+1.δyO)
(+9) Also, the seventh modification of this example is (4),
Using an arbitrary formula instead of a linear formula for formula (5), (7),
If equation (8) can be obtained analytically or numerically, these equations can be used. According to this modification, any formula can be used, so it is characterized by high versatility.

Next, a second embodiment of the present invention will be explained with reference to FIG.

FIG. 8 is a block diagram of a pattern recognition device that detects pattern position errors. This pattern recognition apparatus includes an XY stage 2 for positioning a wafer 1, a light source 3 for illuminating the wafer, an illumination optical system 4, an objective lens 5 for detecting an optical image of the illuminated wafer, a detection section consisting of a TV camera 22, and a TV camera 22. A/D for digitizing and storing the signal from the camera 22
An image input unit 9 consisting of a converter 7 and an image memory unit 8, a detected image IO input to the image input unit, and a stored image 11 in the image memory, calculate the following pixels expressed by equations (7) and (8). It consists of an image processing unit 20 consisting of a sub-vixel matching unit 15 that calculates the alignment amount δxO9δya, and a computer that controls the XY stage 2, stores and displays information output from the image processing unit 20, and manages the entire sequence. It is composed of an overall control section 21.

The operation of detecting pattern errors using this configuration will be described next. First, after initializing each part according to a command from the overall control unit 21, the XY stage 2 is driven and positioned.
The pattern of the wafer 1 illuminated by the light source 3 and the illumination optical system 4 is captured by the TV camera 22 through the objective lens 5. A two-dimensional pattern is detected by performing f conversion, and a two-dimensional detected image 10 is digitized by the A/D converter 7, and the obtained detected image is stored in the image memory section 8. The stored image 11 in the image memory 8 that was detected and stored the previous time is (
After shifting by one pixel using equation (20), the position error δXO9δyo is calculated using equations (7) and (8). Pass the position error to the overall control section.

gl(x,y)=g(x+l,y+l)
(20) According to the present invention, pattern positioning information is used as it is and outputted as a pattern position error, and is characterized in that highly accurate position error detection is possible.

The first modification of the present invention is that when calculating δxO9δyO from the detected image 10 and the stored image) 1 using equations (7) and (8),
There is a method for masking images. According to this modification,
It is possible to calculate the positional error δXO9δyo only from accurate information, which is a feature of high accuracy.

The second modification of the present invention is used to detect alignment marks, and by feeding back position error information,
Perform alignment. This modification has the feature that highly accurate alignment can be achieved.

A third variant of the invention is used for angle detection. The angle is calculated based on the error information of the positions of two or more points.

This modification has the feature of being able to detect angles with high accuracy.

[Effects of the Invention] According to the present invention, the influence of sampling errors can be reduced by setting the pattern alignment accuracy to be less than or equal to the detection pixel size,
Defects with a size comparable to the pixel size can be easily detected.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of an embodiment of the present invention, Fig. 2(a)
~(C) is a diagram explaining the principle of a pattern defect detection method using a conventional pattern comparison method; Figures 3(a) to (e) are pattern waveform diagrams of Figures 2(a) to (c); Figure 4 (a)
~(cl) is a waveform diagram of an example of the sub-vixel matching operation of the pattern of FIG. 2 according to the present invention, and FIG. 5 is a waveform diagram of FIG. 4(a).
6 is an explanatory diagram of the operation of the pixel unit matching section of FIG. 1, FIG. 7 is an explanatory diagram of a modification of the first embodiment, and FIG. FIG. 2 is a block diagram of a device showing a second embodiment. ■...Wafer, 2...XY stage, 3...Light source, 4...Illumination optical system, 5...Objective lens, 6...
- Dimensional image sensor, 7... A/D converter, 8...
- Image memory unit, 9... Image input unit, 10... Detected image, 11... Comparison image, 12... Image extraction unit, 13... Matching unit, 14... Pixel unit position correction 15... Sub-vixel matching section, 16... Registration section, 17... Difference image, 18
... difference image extraction section, 19 ... defect determination section, 20 ...
- Image processing unit, 21... Overall control unit, 22... Sub-vixel calculation unit, 23... TV camera. 1st closed 2nd figure 3 4th (2) S mouth'! ! -) Figure 6 yb7 Figure 8th section

Claims (1)

[Claims] 1. Detect a target pattern, align the pattern with a pre-stored pattern or a separately detected pattern pixel by pixel, and compare the pixel with respect to the pattern aligned pixel by pixel. Align with the following accuracy,
A pattern recognition method characterized in that defects in the patterns are recognized by extracting and comparing errors between two patterns aligned with an accuracy below the pixel. 2. Detect the target pattern, align the detected pattern with a pre-stored pattern or a separately detected pattern pixel by pixel, and compare the pattern aligned in pixel units with sub-pixel accuracy. A pattern recognition method characterized by aligning and recognizing a pattern position error based on the amount of alignment. 3. The pattern recognition method according to claim 1 or 2, wherein the pixel-by-pixel alignment is not performed. 4. The pattern recognition method according to claim 1 or 2, wherein the pixel-by-pixel alignment is performed by pixel or smaller alignment, considering a plurality of pixels as one pixel. 5. The pattern recognition method according to claim 1 or 2, wherein the sub-pixel alignment is performed by considering a plurality of pixels as one pixel. 6. The pattern recognition method according to claim 1 or 2, wherein the alignment below the pixel is performed on a masked image. 7. The pattern recognition method according to claim 1 or 2, in which the least squares method is used for alignment with an accuracy of less than the pixel. 8. The pattern recognition method according to claim 7, wherein the least squares method includes positions in the x and y directions as parameters. 9. The pattern recognition method according to claim 2, wherein the alignment is performed by correcting the position based on the position error of the detected alignment mark. 10. The pattern recognition method according to claim 2, wherein angle detection is performed based on position error information of two or more points.