JPH0774787B2 - Multi-layer pattern defect detection method and apparatus - Google Patents

Description

The present invention relates to a technique for binarizing image information with high accuracy and good reproducibility, for example, a multilayer pattern formed on a semiconductor element such as an LSI wafer. TECHNICAL FIELD The present invention relates to a multilayer pattern defect detection method and apparatus for automatically detecting the appearance of the pattern.

[Conventional technology]

Integrated circuits such as LSI tend to be highly integrated and miniaturized.
In forming such a fine wiring pattern, the detection of defects is important in determining the quality of the formation.

Detecting defects is no longer at the stage of visually locating a large number of personnel, but it is urgent to automate defect detection because it is difficult to visually detect.

Therefore, the image information of the semiconductor element surface obtained from an optical microscope or an electron microscope is converted into electrical information by an image pickup tube, an image pickup element, or the like, and then predetermined signal processing is performed to detect a defect. The device and method are open to the public. For example, Semiconductor World (6 June 1984
Mon.) 112 to 119 (Semiconductor World), or JP-A-59-192943.

By referring to the conventional device shown in FIG. 15 for the common and essential components common to these techniques, the pattern formed on the semiconductor chip 2 is converted into an electric signal with good reproducibility as a two-dimensional pattern. That is.

The operation will be briefly described using a conventional device.

The linear image sensors 5a and 5b have a self-scanning function and detect a pattern one-dimensionally. Then, the two-dimensional pattern of the chip 2 is detected by moving the LSI wafer 1 by the XY table 7 in the direction perpendicular to the scanning of the linear image sensor. 4a and 4b are illumination lights 3a and 3b
Is an objective lens that focuses light on a chip and enlarges the pattern of the chip to form an image on the image sensor. An electric signal from the image sensor is converted into a digital signal by A / D converters 11a and 11b. Further, the digital signal is corrected by the alignment circuit 12 for positional deviation between the input patterns caused by the accuracy of the XY table, chip arrangement error, etc., and the difference signal detection circuit 13 corrects the difference signal (absolute value) of these aligned digital signals. ) Is converted to. And
The difference signal is binarized by the binarization circuit 14, and if the difference signal is greater than or equal to the binarization threshold value Vth, it is determined as a defect.

The above example is an example of comparing two chips, but there is also a method of comparing adjacent cells (repetitive patterns) in the same chip.

[Problems to be solved by the invention]

The conventional technology has the following problems. That is, as shown in FIG. 16, when a minute projection (grain 50) is present on a part of the surfaces of two patterns to be compared, it was impossible to detect a defect by binarization by the conventional technique.

That is, although the grains present on the surface of the pattern shown in FIG. 16 (a) are not defects, they are detected dark as shown in FIG. 16 (b). FIG. 16 (c) shows the detection pattern signal waveform of AA 'part. In the figure, reference numeral 50 represents a change in shade due to grains. Similarly, an example of a detected image having a defect 51 in the pattern is shown in FIG.
It is shown in FIG. The difference signal (absolute value) between the above two detection patterns is as shown in FIG. 16 (f), and the defect is missed at the threshold Vth ₁ , and the grain is erroneously detected at the threshold Vth ₂ and the defect is caused by the grain. It is not possible to distinguish and detect only.

Grain cannot be avoided depending on the pattern manufacturing process, and cannot be distinguished from defects with ordinary optical microscopes, so it is covered by grains that perform inconsistency detection by the conventional method, resulting in fine defects. Detection is impossible.

Furthermore, even if the film thickness of the pattern fluctuates partly (in a normal portion), a difference signal appears between the two patterns, so that the film thickness fluctuation of the pattern also acts in the same manner as the above grain, and the defect Made it difficult to detect.

An object of the present invention is to solve the above-mentioned problems of the prior art.
A multi-layer pattern defect detection method and apparatus capable of accurately detecting defects even if there is a local film thickness variation in a wiring pattern without mistakenly recognizing a brain that is unavoidable in a manufacturing process as a defect. To provide.

[Means for solving problems]

The present invention statistically processes image signals at a plurality of locations obtained by imaging a pattern to be inspected consisting of a multilayer pattern with an image pickup means to generate a reference image signal serving as a reference, and the reference image signal and the pattern to be inspected. A multilayer pattern defect detection method characterized by comparing a pattern image signal to be inspected obtained by imaging with an imaging means and detecting as a defect based on a difference of a predetermined value or more. Further, the present invention is characterized in that, in the comparison in the method for detecting a multilayer pattern defect, a defect is detected when there is a difference in brightness equal to or larger than a variation in brightness for each one or a plurality of pixels. Further, the present invention statistically analyzes the image signals at a plurality of positions among the most image means for obtaining the inspected pattern image signal by picking up the inspected pattern consisting of the multilayer pattern and the inspected pattern image signal obtained by the imaging means. Reference image signal generating means for processing and generating a reference image signal to be a reference, and comparing the reference image signal generated by the reference image signal generating means with the pattern image signal to be inspected obtained from the image capturing means to determine a predetermined value. A multi-layer pattern defect detection apparatus comprising: a defect detection unit that detects a defect based on the above difference. That is, the present invention is achieved by combining the following technical constitutional requirements. That is, in the defect detection method for a multilayer pattern and the apparatus therefor, (1) the detected image and the reference image are compared with each other by using the multilayer pattern having a small contrast as a multi-valued shade.

(2) In order to allow a film thickness difference due to a grain on the pattern surface, a variation in the film thickness of the pattern, and the like, the difference image between the detection image and the two multilayer reference images is binarized with the following threshold image to detect defects. To detect.

That is, in the area surrounded by the pattern edges, a high threshold value (Vth ₁ ) in a large grain area, an intermediate threshold value (Vth ₂ ) in a small grain area, and a low threshold value (Vth ₂ ) in a non-grain area. Vth ₃ ) is set, and a low threshold value (Vth _e ) is set at the pattern edge, which are used for binarization.

Here, Vth ₁ Vth ₂ Vth ₃₀ .

Specifically, the threshold image format procedure is shown in FIG. A plurality of repetitive patterns are cut out from the detected image, these images are differentiated, and the pattern edges are brightly emphasized. Grain is also brightly emphasized due to the difference in shade from the surroundings. In these secondary differential images, the minimum value of brightness in the corresponding pixels is detected, and the randomly generated grains are removed. Then, the brightness is inverted for later processing. On the other hand, a difference image between a plurality of repeating patterns is detected from the detected image. As a result, changes in shade such as grains appear as a difference. Next, the maximum value of the brightness is detected for each pixel between the plurality of difference images, and the limit value of the lightness change that should be allowed is obtained. If a value equal to or greater than the above limit value is used as the binarization threshold value, it is possible to allow a change in shading of a normal portion such as grain. In the area surrounded by the pattern edges, a uniform threshold value equal to the limit value of the density difference is set. Therefore, maximum value propagation processing, that is, the maximum value in the area of 3 × 3 pixels is detected, and this is detected as the central pixel of the area. The maximum value is propagated to the surroundings by setting the brightness to. At this time, the procedure of taking the product of the inverted image obtained above and the maximum value and performing the maximum value propagation processing again is repeated, for example, five times. As a result, the area surrounded by the pattern edges is uniform, but a different threshold value is set for each area.

[Action]

FIG. 18 shows an example of a single layer pattern. From the image of the object to be inspected having a pattern in which grains are generated as shown in (a), according to the above means, pattern edges as shown in (b) and (c) are obtained. A threshold image having a different threshold value is obtained for each region surrounded by, corresponding to the gradation change due to the grain. By using this threshold image, as shown in (d) and (e), it becomes possible to perform binarization in which only the defect is discriminated and detected from the difference signal between the two detection patterns.

Therefore, the isolated defects existing in the area surrounded by the pattern edges, such as pinholes and pattern residuals, can be detected by discriminating only those grains or more present on the surface of each pattern. In particular, since this is a pattern in which grains are conventionally generated in a multilayer pattern, it is necessary to reduce the defect detection sensitivity to a pattern in which no grains are generated. It becomes possible to perform valuation.

Further, the shape defect appearing on the pattern edge can be detected without fail even in a manufacturing process in which even a fine grain, no matter how large a grain occurs, is detected.

With the binarized threshold image, a uniform optimum value can be automatically obtained for each region from the object to be inspected by the above-mentioned means without manual labor. This is because the pattern edge acts as a stopper for the propagation while propagating the grayscale difference corresponding to the grain to the surroundings, and different values corresponding to the grain in different regions,
This is because a uniform value is automatically set in the same area.

Further, the pattern edge not only acts as a propagation stopper, but also acts to automatically set a low threshold at the pattern edge to facilitate detection of shape defects.

Further, although FIG. 18 shows an example of the one-layer pattern, the above-mentioned means makes it possible to set the optimum threshold value for each of the complicated multilayer patterns.

Further, the maximum value propagation processing in the above means is effective in reducing the image area of the inspection target necessary for creating the threshold image. In a memory device or the like, if there are several repetitive patterns, a typical threshold value capable of removing grains generated in all regions (patterns, layers) can be set.

Note that this means can not only allow the change in density due to the grains, but also allow all the change in density caused by some cause in the normal part, such as the change in density due to the variation in the film thickness of the pattern.

〔Example〕

An embodiment of the present invention will be described below with reference to FIG. As a photoelectric converter for converting an optical image of a multilayer pattern into an electric signal, any of a linear image sensor, a TV camera and the like can be used, but in the present embodiment, a linear image sensor is used and the linear image sensor is self-contained. A two-dimensional pattern of an LSI wafer is detected by scanning and an XY table that moves in the direction perpendicular to it. FIG. 1 is a block diagram of a pattern visual inspection device. The output of the linear image sensor 5 is converted into a digital signal by the A / D converter 11 and input to the edge detection circuit 15a. The A / D converter output is also input to the image memory 16, and at the same time as being input, the corresponding pattern of the adjacent chip or the adjacent cell stored in the image memory is read from the image memory and input to the edge detection circuit 15b. By using the image memory, the pattern of the adjacent chip or the adjacent cell can be compared and inspected by one image sensor. The edge detection circuits 15a and 15b detect the edges of the pattern.

The misregistration detection circuit 18 shifts the binary pattern output from the edge detection circuits 15a and 15b, counts the number of mismatched pixels at the shifted position, and detects the amount of mismatch between the two binary patterns. Two shift amounts that minimize the mismatch amount in two orthogonal directions are obtained.

The A / D-converted digital signal output of the linear image sensor 5 and the output of the image memory are delayed by delay circuits 19a and 19b. The delay time is determined by the number M (for example, 1024) of pixels of the linear image sensor and the number N (for example, 256) of image sensor self-scanning required for alignment, and each of the delay circuits 19a and 19b is composed of a shift register having a bit number of M × N. To be done.

The alignment circuit 20 performs alignment by shifting the outputs of the delay circuits 19a and 19b by the alignment circuit 20 so that the optimal alignment state determined by the displacement detection circuit 18, that is, the amount of mismatch is minimized. Then, the difference in brightness (absolute value) of the detected images aligned by the difference image detection circuit 21 is calculated, and the binarization circuit is obtained.
At 22, a defect larger than the threshold value Vth is detected as a defect. With this method, it is possible to detect defects (for example, short circuit, disconnection, foreign matter, etc.) in the multilayer pattern without overlooking.

The binarization threshold Vth is a switch that cuts out the output of the A / D converter 11.
At 23, the image signal of the normal portion is cut out, and the threshold value generating circuit 24 obtains this. The purpose of this is to determine the threshold image, and only the small area may be cut out by the cutout switch 23.

The binarization threshold value thus obtained is an optimum value for each pattern of the multilayer pattern. Further, the threshold image thus obtained is stored in the memory 25 for a repeating unit,
This is repeated during inspection. The threshold image is read from the memory 25 in the binarization circuit 22 in synchronization with the detection pattern (output of the alignment circuit 20) to align the positions.

Next, a detailed block diagram of the threshold value generation circuit 24 is shown in FIG. 2 and the process of threshold value image generation will be described.

In the secondary differentiation circuit 241, each detected image is secondarily differentiated to brighten the pattern edge. Grain is not a defect, but the grain is brightly emphasized because there is a difference in shade from the surroundings. Next, the minimum value detection circuit 242 detects the minimum value of the secondary differential values in the corresponding pixels for these plurality of images, and removes the randomly generated grains. Then, the inverting circuit 24 operates so that the detected pattern edge operates as a stopper for the maximum value propagation processing.
Invert the above minimum value in 3. On the other hand, the difference image detection circuit 2
At 44, a difference image between the plurality of repeating patterns is detected from the detected image. As a result, a change in shade such as grain is detected as a difference. Next, the maximum value detection circuit 245 detects the maximum value of the brightness for each pixel among the plurality of difference images, and obtains the limit value of the grayscale change to be allowed. Since a uniform threshold value equal to the above limit value is set in the area surrounded by the pattern edges, the maximum value propagation processing circuit 246a detects the maximum brightness value in a local area, for example, 3 × 3 pixels, and determines the maximum brightness value in the local area. By setting the brightness of the central pixel of the area, the maximum value is propagated to the surroundings. And the inverting circuit 24
Propagation is advanced to the pattern edge by calculating the product of the output of 3 and the maximum value in the product circuit 247a. This maximum value propagation processing and processing consisting of products are repeated, for example, five times,
Although the area surrounded by the pattern edges is uniform, a different threshold value is set for each area. If the above-mentioned repetition is repeated 5 to 6 times for products such as memory devices, a steady state in which the value does not change is reached.

Next, the details of each part of the threshold generation circuit 24 will be described in more detail.

A configuration example that can be used as the secondary differentiating circuit 241 will be described with reference to FIG. Reference numeral in the figure
Reference numeral 30 is a shift register of three stages which receives, for example, an 8-bit digital video signal from the A / D converter 11, and outputs of the first and third stages are to the adder 31 and outputs of the second stage are amplifiers of gain 2.
Is supplied to each. Output of adder 31 and amplifier 32
Is output to the subtractor 33, and the pattern edge is differentiated. The shift register 30, the adder 31, the amplifier 32, and the subtracter 33 constitute a "1-21" operator.

FIG. 3 (b) is a secondary differentiating circuit for emphasizing edges in four directions of vertical, horizontal and diagonal, and outputs the output of the A / D converter 11 by 3 ×.
In addition to the 3-cutout circuit 34, edge operations are performed by four edge operators OP1 to OP4. Each edge operator may be the same as illustrated in Figure 3 (a). The outputs of the operators OP1 to OP4 are all supplied to the maximum value detection circuit 35.

Fourth Configuration Example Used as Minimum Value Detection Circuit 242 of FIG.
Shown in the figure. The output of the secondary differentiation circuit 241 is temporarily latched by the latch circuit 36.
Latch with. At the same time, read the data from the memory 37,
These are compared by the maximum value selection circuit 38, and if the output of the latch circuit is smaller than the memory content, WE (write enable)
Write to memory by signal. The memory 37 has a capacity for a repeating unit. The data whose magnitude is compared by the minimum value selection circuit 38 is the secondary differential value of the corresponding pixel of the repeating pattern. The initial value of the memory 37 is set to a value larger than the value that the second-order differential value of the video signal can take, for example, 255.

The maximum value detection circuit 245 is basically the same as the configuration example of the minimum value detection circuit 242 shown in FIG. However, the minimum value selection circuit 38 replaces the maximum value selection circuit, and if the latch circuit output is larger than the memory contents, it is written into the memory by the WE signal. The initial value of the memory is 0. Other basic operations are the same.

FIG. 5 shows a configuration example used as the inverting circuit 243.
The output of the minimum value detection circuit 242 and a constant, for example, 255 is the subtractor 39.
And the difference between them is taken.

FIG. 6 shows a configuration example used as the difference image detection circuit 244. The digital video signal from the A / D converter 11 is subtracted by the subtractor 4
0 and memory 41. A video signal delayed by a repeating unit is obtained from the memory 41 and input to the subtractor 40. The subtracter 40 detects the difference (absolute value) between these video signals and outputs it as a difference image.

FIG. 7 shows a configuration example used as maximum value propagation processing circuits 246a, 246b, 246c ... First-stage maximum value propagation processing circuit 24
6a inputs the output of the maximum value detection circuit 245 to the shift register 42, and the local area is 3 × 3 pixels (may be 5 × 5 pixels).
Cut out. Then, the maximum value (maximum value of brightness) of the pixelized signal in the 3 × 3 area is detected by the maximum value detector 43 and output. Further, the output of the product circuits 247a, b ... Is input to the maximum value propagation processing circuits 246b, c.

FIG. 8 shows a configuration example of the product circuit 247. The second differential image in which the pattern edge is emphasized and the difference image representing the difference in density such as grain are input to the multiplier 44 and multiplied. Then, the multiplier 45 multiplies a constant, for example 1/255,
The contrast is changed.

The configuration example of each unit of the threshold value generation circuit 24 has been described above. Of these, the secondary differentiating circuit 241 does not necessarily have to have the second-order differential of the above configuration, and may be changed depending on the inspection target. For example, the operator "10-201" can be used, and the first derivative can also be used.

The obtained threshold image for the repeating unit is stored in the memory 25 and output to the binarization circuit 22.

Next, main constituent blocks other than the threshold value generating circuit 24 in FIG. 1 will be described in detail.

FIG. 9 (a) shows the operators OP1 to OP shown in FIG. 9 (b).
It shows P4. This structure is basically the same as that shown in FIG.

Similarly, FIG. 9B is basically the same as FIG. 3B, but the outputs of the operators OP1 to OP4 are binarized circuits 154-1.
˜154-4 is binarized and supplied to the OR circuit 27. The output of the circuit 27 is applied to the shift registers 181a, 180a or the shift register 182a (Fig. 10) of the displacement detection circuit 18.

A tenth configuration example used as the positional deviation detection circuit 18 in FIG.
Shown in the figure. 7 × 7 pixels (shift registers 180a to 180f for delaying the output of the A / D converter of the linear image sensor 5 by one scanning from the output of the binarization circuit 154 and shift registers 181a to 181g of serial-in / parallel-out ( Other example: 7x with 2D local memory of 9x9 pixels)
Cut out 7 pixels. On the other hand, the output of the other binarization circuit 154 is delayed by using the same shift registers 182a to 182c and 183 to synchronize the output with the central position of the local memory.

The output of the shift register 183 and the local memory bit output are E
The XOR circuits 184a to 184n take an exclusive OR to detect the number of mismatched pixels. The counters 185a to 185n count the number of unmatched pixels. If the counters 185a to 185n are cleared to zero every N scans of the linear image sensor and the value is read immediately before that, the number of mismatched pixels in the area of M pixels × N scans can be known. Each bit output of the local memory is ± in the XY direction (two orthogonal directions) with respect to the output of the shift register 183.
In the range of 3 pixels, each pixel is shifted.
The counters 185a to 185n count the number of mismatched pixels in each shift amount when the pixel input pattern in the XY directions is shifted. Therefore, by checking which counter has the minimum value, the shift amount in the X and Y directions that minimizes the number of mismatched pixels can be found.

The minimum value detecting circuit 186 reads the values of the counters 185a to 185n, selects the counter having the minimum value, and shifts the shift amount 188 in the linear image sensor scanning direction (Y direction) and the shift amount 187 in the direction (X direction) perpendicular thereto. Is output.

FIG. 11 shows a configuration example used as the alignment circuit 20 of FIG. In the selection circuit 201, the delay circuit 19a and shift registers 200a to 2 for delaying one scanning by the shift amount 187.
The optimum shift position is selected from the output of 00f and input to the shift register 202. In the selection circuit 203, the shift amount is 18
Use 8 to select the optimum shift position in the scanning direction. Therefore, the local memory at the shift position where the mismatch amount is minimized is extracted from the output of the selection circuit 203.

On the other hand, by using the shift registers 204a to 204c and the shift register 205 that delay the output of the delay circuit 19b by one scan, the pixel of the local memory at the position delayed by the same amount as the output of the shift register 183 in FIG. Extract. In this state, the pixel output of the local memory output from the selection circuit 203 is at the optimum shift position without displacement with respect to the pixel output of the local memory output from the shift register 205.

Next, the operation of the threshold generation circuit 24 will be described with reference to FIG. The target is the IC multilayer pattern shown in FIGS. 19 (a) and 19 (b). In the second differentiation circuit 241, as shown in FIG. 17, repeated patterns A, B,
Each of the detected images 60 consisting of ... Is quadratic differentiated, the pattern edges are brightly emphasized, and a differential image 61 is obtained. That is, the cut-out circuit 34 cuts out data of 3 × 3 pixels from each of the detected images 60, and the second-order differential is detected by the four-direction differential operators OP1 to OP4. Since the pattern is oriented in all directions, the maximum value detection circuit 35
The maximum value is detected by and the differential image 61 in which the pattern edge in an arbitrary direction is brightly emphasized is obtained. These differential images
Reference numeral 61 corresponds to the repeated patterns A, B, C ... Of the detected image 60.
Although a grain generated at an arbitrary position is not a defect, the grain is also emphasized brightly because it has a gray level difference with the surroundings, and is detected as a differential image 61. Next, for this differential image 61,
In the minimum value detection circuit 242, the minimum value of the secondary differential values in the pixels corresponding to the repeated patterns A, B, C ... Is detected, and the minimum value differential image 62 consisting of the pattern edges from which the randomly generated grains are removed is obtained. can get. That is, the differential image A'of the repeated pattern A is temporarily latched by the latch 36 and stored in the memory 37. Next, the differential image B ′ of the repeating pattern B
Is input, it is latched by the latch 36, and at the same time, the memory 37
The data corresponding to the differential image B'of the differential image A'is read out from it, inputted to the minimum value selection circuit 38, and the smaller one is selected and then written in the memory 37. By repeating this, the grain generated at any position is erased and the repeated pattern A, B, C consisting of only the pattern edges is erased.
The minimum value differential image 62 consisting of the minimum value of the corresponding pixels of the differential images A ′, B ′, C ′ of ... Is obtained in the memory 37. In the thus obtained minimum value differential image 62, the inverted image 63 is formed by inverting the brightness in the inversion circuit 243 so that the pattern edge detected as the maximum value operates as a stopper for the maximum value propagation processing described later. I'll do it.

On the other hand, the difference image detection circuit 244 detects a difference image 64 between a plurality of repetitive patterns (unit patterns) A, B, C, ... (Between cells) in a normal part in the same chip. That is, the subtractor 40 is a memory
A difference image 64 between the detection images A, B, C ... Sequentially stored in 41 and the detection images B, C ,.

As a result, the pattern edge is erased, and only the change in shade such as grain is detected as the difference image 64. Next, in the maximum value detection circuit 245, the maximum value of brightness is detected for each corresponding pixel among the plurality of difference images 64 obtained from the difference image detection circuit 244, and the maximum value difference image 65 is obtained. That is, the repeating pattern A
The difference image A ″ between B and B is temporarily latched by the latch 36 and stored in the memory.
Remember in 37. Next, when the difference image B ″ between the repeated patterns B and C is input, it is latched by the latch 36, and at the same time, the data corresponding to the difference image B ″ of the difference image A ″ is read out from the memory 37, and these are set to the maximum value. It is input to the selection circuit 38, and the larger one is selected and then written in the memory 37. By repeating these, the brightest one of the randomly generated grains is detected as the maximum value difference image 65. The maximum value V _R max detected at the upper limit is the limit value of the light and shade change allowed as a grain, and the region R surrounded by the pattern edges has a uniform threshold value equal to the above limit value (maximum) value V _R max. , It is possible to detect a defect having a gradation change of not less than the grain.

Therefore, in the maximum value propagation processing circuit 246a, the local region 42
For example, the maximum value detection circuit 43 detects the maximum value of brightness in 3 × 3 pixels, and this is detected as the pixel E _{00 at} the center of the local area.
The maximum value Vmax is propagated to the surroundings to obtain the first maximum value propagation image 66. And the inverting circuit 24
The product of the inverted image 63 which is the output of 3 and the first maximum value propagation image 66 is calculated in the multiplier 44 of the product circuit 247a, and the multiplier 45
Is multiplied by 1/255 to obtain a first product image 67a. The maximum value propagation processing circuits 246b, 246c, ..., Product circuits 247b, 247c, ... Delay circuits 248b, 2
For example, 5 times are repeated by 48c, and a threshold image 68 having a uniform threshold in the region surrounded by the pattern edges but having a different threshold for each region is obtained. In particular, since the width of the pattern edge is set to be larger than the width of 3 × 3 pixels from the imaging magnification, the maximum value detected by the maximum value detector 43 always includes the pattern edge and is surrounded by the pattern edge. The maximum brightness of the grain is not propagated out of the shaded area.
In this way, different thresholds corresponding to the grains can be set for each region surrounded by the pattern edges, and defects having a gradation change equal to or larger than the grains can be detected. Moreover, since a low threshold is set at the pattern edge, it is possible to detect fine shape defects such as shorts and disconnections in the multilayer pattern without overlooking.

The main part of the embodiment has been described above in detail. According to this embodiment, the optimum threshold value for defect detection can be set for each pattern. The hardware configuration can also be realized by existing technology.

Next, another embodiment will be described with reference to FIGS. 12, 13 and 14. That is, another embodiment will be described in which a threshold image is not created in advance as in the above embodiment, but a threshold image is created from a detected image and used as a binarization threshold. In FIG. 12, one output of the alignment circuit 20 is constantly input to the threshold generation circuit 28, threshold images for repeating units are continuously generated, and input to the binarization circuit 22. The output of the alignment circuit 20 is not always the video signal of the normal part and may include a defect, which must be removed. Therefore, as shown in FIG. 14, a large number of threshold images are created according to the procedure shown in FIG. 2, and a defect-free image is selected from them. When there is a defect, the obtained threshold image has a larger value than the threshold image obtained from the normal portion, and therefore clears a larger value in the clear circuits 49a, 49b, .... However, when a small threshold value is selected, a false alarm occurs, so the median value is good as the threshold value. For example, in the case of 8 inputs, the fourth largest value is selected. A circuit block diagram for doing this is shown in FIG. However, the threshold generation circuit 28 shown in FIG.
60 is operated so as to continuously generate threshold images. That is, in the threshold value generation circuit 24 of the threshold value generation circuit 28, with respect to the minimum value detection circuit 242 and the maximum value detection circuit 245 shown in FIGS. 2 and 4, at the threshold image generation cycle (an integral multiple of the cycle of the repeating pattern). It is necessary to initialize the memory 37. In FIG. 13, a large number of threshold images successively obtained from the threshold generation circuit 24 are input to the delay circuit 47, the maximum value detection circuit 46a, and the latch 48a. The delay circuit 47 delays the threshold image by the time required to create the threshold image, and the maximum value detection circuit 46a detects the maximum value of the brightness of the threshold image between the corresponding pixels of the threshold image. Only the maximum value of the output of the latch 48a is cleared to zero in the clear circuit 49a by the output of the maximum value detection circuit 46a. The maximum value is detected and then cleared to zero repeatedly. For example, for 8 inputs (threshold image is 8 planes), when the maximum value is detected with the configuration shown in FIG. 13, the 4th largest data is detected. To be done. If this is performed over the entire threshold image, a set of threshold images from which defects have been removed can be obtained. The threshold image thus obtained is equivalent to the threshold image created from a video signal having no defect.

According to this embodiment, the detection image for which the threshold image is to be created does not have to be normal and may include defects, and the created threshold image is always aligned with the detection image. Therefore, there is an advantage that no ingenuity is required for the memory 25 shown in FIG. 1 and its reading.

〔The invention's effect〕

As described above, according to the present invention, even if there is a grain on the surface of the multilayer pattern or a local film thickness variation of the pattern,
It is possible to detect defects with high accuracy. Therefore, it can contribute to the automation of the pattern inspection.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a multi-layer pattern defect detecting device of the present invention, FIG. 2 is a block diagram showing a threshold value generating circuit shown in FIG. 1, and FIG. 3 is a second derivative shown in FIG. FIG. 4 is a diagram showing a configuration example of a circuit, FIG. 4 is a diagram showing a configuration example of a minimum value or maximum value detection circuit shown in FIG. 2, and FIG. 5 is a configuration example of an inverting circuit shown in FIG. FIG. 6 is a diagram showing a configuration example of the inter-cell difference image detection circuit shown in FIG. 2, FIG. 7 is a diagram showing a configuration example of the maximum value propagation processing circuit shown in FIG. 2, and FIG. Is a diagram showing an example of the configuration of the product circuit shown in FIG.
9 is a diagram showing a configuration example of the edge detection circuit shown in FIGS. 1 and 12, and FIG. 10 is a diagram showing a configuration example of the position shift detection circuit shown in FIGS. 1 and 12. Fig. 11 shows Fig. 1 and
FIG. 12 is a diagram showing an example of the configuration of the alignment circuit shown in FIG. 12, FIG. 12 is a block diagram showing another embodiment of the present invention different from FIG. 1, and FIG. 13 is a threshold generation circuit shown in FIG. FIG. 14 is a block diagram showing FIG. 14, FIG. 14 is a view for explaining a threshold image created by the threshold creating circuit shown in FIG. 13, FIG. 15 is a view showing a schematic configuration of a conventional device, and FIG. FIG. 17 is a diagram for explaining a problem of the conventional apparatus shown in FIG. 17, FIG. 17 is a diagram showing a threshold value creating procedure according to the present invention, FIG. 18 is a diagram showing a binarization result obtained by the present invention, FIG. 19 is a cross-sectional view and a plan view showing a multilayer pattern according to the present invention. Explanation of symbols 1 ... Wafer, 4 ... Objective lens 5 ... Image sensor, 11 ... A / D converter 15 ... Edge detection, 16 ... Image memory 18 ... Position deviation detection, 19 ... Delay 20 ...... Alignment, 21 …… Difference image detection 22 …… Binarization, 24 …… Threshold creation 25 …… Memory, 241 …… Second derivative 242 …… Minimum detection circuit, 243 …… Inversion circuit 244 …… (Between cells) difference image detection circuit 245 ...... maximum value detection circuit 246a, 246b ...... maximum value propagation processing circuit 347 ...... product circuit, 28 ...... threshold value creation circuit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-60-73310 (JP, A) JP-A-62-43505 (JP, A) JP-A-60-124833 (JP, A)

Claims (5)

[Claims] 1. A reference image signal serving as a reference is generated by statistically processing image signals at a plurality of positions obtained by capturing an image of a pattern to be inspected having a multilayer pattern with an image pickup means, and the reference image signal and the pattern to be inspected. Is compared with a pattern image signal to be inspected obtained by picking up an image by an image pickup means, and is detected as a defect based on a difference of a predetermined value or more. 2. The comparison according to claim 1, wherein when there is a difference in brightness equal to or larger than a variation in brightness for each one or a plurality of pixels, it is detected as a defect.
A method for detecting a multilayer pattern defect according to the item. 3. A threshold value obtained by the brightest brightness of the grain in a certain range for each area surrounded by the pattern edges with respect to the multilayer pattern and at the edge brightness of the pattern edge section. A defect existing in the multi-layer pattern is detected by forming an image, detecting a difference image between a reference image and a detection image obtained by picking up the multi-layer pattern by an image pickup means, and binarizing the difference image with the threshold image. A multi-layer pattern defect detection method, comprising: 4. An image pickup means for picking up an image of a pattern to be inspected comprising a multilayer pattern to obtain an image signal of the pattern to be inspected, and statistically processing image signals at a plurality of positions among the image signals of the pattern to be inspected obtained by the image pickup means. A reference image signal generating means for generating a reference image signal serving as a reference, and comparing the reference image signal generated by the reference image signal generating means with the inspected pattern image signal obtained from the image pickup means A multi-layer pattern defect detection device comprising: a defect detection unit that detects a defect based on a difference. 5. An image pickup means for picking up a detection image and a reference image by picking up an image of a multilayer pattern by an image pickup device, a differentiating means for differentiating the detected image or the reference image obtained from the image pickup means, and a differentiating means. Between the cells of the detection image or the reference image obtained from the image pickup means, and the minimum value detecting means for detecting and storing the edge image showing the minimum in which the grain is erased and the edge is emphasized from the differential image for the plurality of cells Regarding the inter-cell difference image detecting means for detecting the difference image consisting of only the grains by erasing the edge part, and the maximum showing the brightest brightness of the grain among the plurality of difference images obtained from the inter-cell difference image detecting means A maximum value detecting means for detecting and storing the brightness grain image, and a maximum brightness grain image obtained by the maximum value detecting means for propagating the grain to the surroundings. By repeating stopping for an edge image obtained by the minimum value detecting means a predetermined number of times, the area surrounded by the pattern edges has a certain brightness, and the pattern edge portion is based on the edge image. Threshold image forming means for forming a threshold image having a bright edge portion, difference image detecting means for detecting a difference image between the detection image obtained from the image pickup means and the reference image, and the difference image detecting means A multi-layer pattern defect detection apparatus comprising: a defect detection unit that binarizes a difference image with a threshold image obtained from the threshold image forming unit to detect a defect.