JPH045940B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a pattern defect inspection system that automatically identifies defective portions by comparing a pattern to be inspected input from a video input device with a reference pattern.

Traditionally, pattern inspection of semiconductor patterns and printed circuit boards was carried out visually by skilled workers, but due to the miniaturization of patterns and the increase in the number of producers due to higher integration, it has become extremely difficult for humans to inspect the patterns. This has become increasingly difficult, and attempts have therefore been made to develop automatic pattern defect inspection equipment.

FIG. 1 is an example of a conventional automatic pattern defect inspection apparatus. The image of the pattern to be inspected 1 is inputted by the image input device 2, and is converted into 2 images by the binarization circuit 3.
converted into a value signal. On the other hand, a reference pattern based on the design data is generated from the dictionary generation circuit 4 in synchronization with the video input device 2, and is compared for each pixel by the comparison circuit 5. In the comparator circuit 5, if there is a pixel in the input video that does not match the reference pattern,
That would be a defective area. Next, the determination circuit 6 determines the size of the defective area, and if it is larger than a certain level, it can be determined to be a fatal defect and output. In this way, defect inspection of the pattern to be inspected can be performed at high speed.

However, such defect inspection equipment can give good results if the input is a high-quality white or nearly black image, but if the pattern has various random patterns, such as the aluminum pattern of a complete IC or LSI wafer, When I try to apply it to pattern inspection,
For example, there is a problem in that binarized video data is unstable, which is difficult to deal with with such a device system.

FIG. 2 is an explanatory diagram of the above-mentioned problem. first,
Figure A shows the problem of threshold processing when inspecting patterns with complex patterns.

is a reference pattern and represents a boundary portion of a pattern on one scanning line. That is, a value of "1" indicates an inner area A of a particular pattern, and a value of "0" indicates an outer area B of that pattern.
is a video signal of a pattern to be inspected corresponding to the dictionary, and the vertical axis represents brightness. Here, both the patterns in areas A and B have shadows due to unevenness, etc., and even though there are no defects, the brightness levels are distributed in a normal distribution, and some of the distributions are partially Assume that they overlap. In such a case, it will no longer be possible to clearly separate areas A and B with threshold processing alone, and if a comparison is made with the standard pattern using exclusive OR, they will be extracted as defective areas. become. Although this result is erroneous, it is unavoidable in separation processing using a single threshold value.

FIG. 2B is an explanatory diagram of another problem with the conventional method. Area A is an area where something like a black spot D is visible due to surface irregularities, such as an aluminum pattern area of a complete IC wafer, for example. Region B is a region of silicon oxide and has relatively uniform brightness values. At this time, if a black spot C is observed in area B, it is a defective part, but
In area A, even if the spots are of the same size, they are not defects if they are similar to the originally existing spots as described above. If there is a spot E that is larger than the originally existing spot, it is a defect. However, since conventional devices only compare the input image with a reference pattern, the number of defects that are determined as defects remains the same regardless of the area. For example, the pattern in the aluminum region is full of defects, and if an attempt is made to correctly determine the defects in the aluminum region, there is a problem in that defects in the silicon oxide region will be overlooked.

SUMMARY OF THE INVENTION An object of the present invention is to provide a pattern defect inspection apparatus that overcomes the problems of the conventional apparatus and enables defect inspection of patterns that are more complex and have poor image quality.

In order to achieve the above object, the present invention includes a first means for scanning a target pattern to be inspected and inputting pixel data constituting the pattern to be inspected in time series; , a second means for generating pattern data in synchronization with the pixel data by obtaining pattern data indicating which region of the pattern to be inspected consisting of a plurality of regions is included from the design data of the object to be inspected; each region has a plurality of sets of at least one of a range and a determination means;
and third means for selecting one of the plurality of sets for each area and changing the content of the process for comparing and determining the pattern to be inspected and the reference pattern, in accordance with the second means. The place has its characteristics. That is, ideal pattern data that serves as a reference is created from the design data used when the object to be inspected was manufactured, and generated in complete synchronization with the video signal of the object to be inspected. Normally, this ideal pattern is used as a reference pattern for matching input video data, but in the present invention, dictionary data is used to inform the device for each pixel whether the input video data is, for example, within an aluminum region or a silicon oxide region. The aim is to differentiate between processing for aluminum patterns and processing for silicon oxide.

It is clear that the above-mentioned problems can be solved by doing so. For example, the problem described in Figure 2A can be solved by setting thresholds θ _A and θ _B for areas A and B, respectively, as shown in Figure 3, and using different thresholds depending on the dictionary pattern value. Even if there is a common part, it will not be determined as a defective area. However, in this case, even if there is a defect in which area A replaces area B, for example, it cannot be detected as a defect if the brightness of the defective portion is at a common level. However, the brightness level in both areas A and B changes in a normal distribution manner, and if the defect size is large to some extent, it is rare that all of them have a common level of brightness.
In reality, it's not a big problem. In fact, in the case of actual defect inspection equipment, most of the parts are normal, so preventing the occurrence of false alarms that judge normality as abnormal is a bigger problem than discovering small defects, and in that sense, it is a real problem. Inventions can have enormous practical effects.

It is clear that the problem in FIG. 2B can also be solved by, for example, changing the determination parameter regarding the defect size for each region.

The present invention will be explained below using examples. Fourth
The figure is an overall configuration diagram of a pattern defect inspection apparatus using the present invention. Reference numeral 18 denotes a computer that controls the whole, 1 is an object to be inspected, and 11 is a stage on which the object to be inspected 1 is placed.The computer can move the object to be inspected to any position on a plane.
Reference numeral 2 denotes a raster scanning type video input device such as a television camera, which scans an image of a certain area on the object to be inspected in synchronization with the horizontal synchronization signal HD and vertical synchronization signal VD generated from the synchronization signal generator 17. , the bright/dark signal can be converted into an electrical signal 20 and output. The electrical signal 20 is synchronized with the clock signal 25 generated from the synchronization signal generator 17.
It is converted into a discrete digital signal 21 by a converter 12. That is, the video signal of the object to be inspected is divided into a grid-like data array by the AD converter 12, and the brightness and darkness values of each point are generated as, for example, 8-dit grayscale data in synchronization with the clock signal 25. Become.

On the other hand, ideal pattern data for the object to be inspected 1 is calculated in advance as array data from the design data for the object to be inspected 1, and is transferred and stored in the storage circuit 16 from the computer 18 via the signal line 28. Reference numeral 15 denotes a readout control circuit for the memory circuit 16, which reads out ideal pattern data using the X, Y coordinates 27 outputted from the synchronization signal generator 17 as a readout address of the memory circuit 16. At this time, in the synchronization signal generation circuit 17, the horizontal synchronization signal HD is cleared to "0", and the clock signal 25 is cleared to "0".
The output of the counter that counts and the vertical synchronization signal VD
If the X and Y coordinates 27 are created from the output of the counter that counts the number of horizontal synchronizing signals HD that are cleared to "0" by
The positional relationship can be perfectly matched with the ideal pattern output data 23 read out by the readout control circuit 15.

If it can be perfectly matched, it will be possible to know which region of the object to be inspected is scanned by the electric signal 21 at each time point from the value of the ideal pattern data.
This can be done for 1. Reference numeral 13 denotes a defective area extraction circuit which selects the optimum process for each area based on ideal pattern data and processes the input signal 21 in synchronization with the clock 25. If there is a part that is determined to be defective as a result of processing, that part will be marked "1".
A defective area signal 22, which is "0" in normal areas, is sequentially output in synchronization with the input signal 21 for each clock. One specific embodiment of the defective area extraction circuit 13 will be explained with reference to FIG.

14 inputs the defective area signal 22 output from the defective area extraction circuit 13, and receives the ideal pattern data 2.
3 is a fatality determination circuit that selects the optimal process for each area and determines the fatality of the defect.
One specific embodiment of this fatality determination circuit 14 will be explained with reference to FIG.

When the defect is determined to be fatal by the fatality determination circuit 14, the X and Y coordinates output from the synchronization signal generation circuit 17 at that time are stored in a storage circuit built into the fatality determination circuit 14. The data is stored and input to the computer 18 as defect determination data after all video scanning is completed. If necessary, the results can be organized and displayed on the display device 19 for easy identification by the operator. The input/output signal lines of the computer 18 include a stage drive signal 34, a stage position signal 33, a starting signal for the inspection circuit system, 31 an inspection end signal, and 30 defect position data stored in the defect critical judgment circuit 14. This is the read data line.

FIG. 5 shows a more detailed embodiment of the defect area extraction circuit 13. In FIG. 5, 41 is a comparison circuit;
42 is a selection circuit that selects and outputs one of the input signals 46 and 47; 44 and 45 are read-only memories (ROM); and 43 is a register. In this embodiment, the ideal pattern data 23 is used as a dictionary pattern data signal indicating in which region the AD-converted input video data 21 is video data, and the dictionary pattern data value is used as a dictionary pattern data signal in the read-only memory 44. The optimum threshold value 49 for each area, which has been recorded in advance, is read out from. Input video data 21 is compared with an optimal threshold value 49 in a comparison circuit 41 . Comparison circuit 41 compares the value of signal 21 with signal 49
If the value of the signal 21 is greater than the value of the signal 49, the output signal 46 is set to "1" and the output signal 47 is set to "0", and vice versa.
shall be. 42 is a selection circuit, which uses the dictionary pattern data 23 as an address to read-only memory 45;
The signal 46 or 47 is selected based on the selection signal 50 read from the select signal 50 and outputted as the signal 48. In other words, depending on the area, there are cases where the input video data is defective when it is larger than the threshold value, and cases where it is defective when it is smaller than the threshold value. Whether it should be a defect signal is stored in advance in the read-only memory 45, and selected using dictionary pattern data indicating the area. Switching the threshold value or defect determination method using such dictionary pattern data corresponds to switching a plurality of processes, which is the gist of the present invention. 43 is a register for timing adjustment;
A signal 48 is set by the clock signal 25, and its contents are outputted to the outside as the output 22 of the defect extraction circuit 13. The output signal 22 is a defective video signal that is output with a delay of one clock after the input video data 21.

In addition, in this example, there are only two types of areas A and B in the pattern to be inspected, and only the case where the defect density is expressed as greater or less than the threshold value was handled, but when there are three or more types of areas, This can be achieved by increasing the number of bits in the dictionary pattern data and storing as many optimum threshold values as the number of regions in a read-only memory.If there is a range of defect density levels, a similar circuit can be built. By combining
It is clear that this can be easily achieved.

FIG. 6 shows a more detailed embodiment of the fatality determination circuit 15. In this example, the vertical, horizontal,
The criterion is that if the maximum value of the width of the diagonal method is larger than a threshold value determined for each region, it is determined that there is a fatal defect. Furthermore, there are only two types of pattern areas, A and B, and the dictionary pattern data signal 23 in FIG. 6 is 1 bit, and is "0".
When it is "1", it indicates area A, and when it is "1", it indicates area B. 22 is a defective video signal, 62a and 62b are AND gates, and 65 is an inverter. As a result, 63a is a defective image in which only the defective pixels in the A area are "1", and 63b is a defective image in which only the defective pixels in the B area are "1".

Circuits 61a and 61b surrounded by one-dot chain lines are exactly the same circuits, and each is a two-dimensional local image cutting circuit of 9×9 pixels. In the circuit of 61a, l ₁ , l ₂ ,
..., _l8 are shift registers each having a stage equal to the number of pixels for one horizontal scanning line of the video input device, and _d11 , _d12 , ..., _d18 , _d21 , _d22 , ..., _d99 is a shift register for one pixel. All of these shift registers are supplied with a clock signal 25 through a signal line (not shown), and are configured to shift in synchronization with pixel data. As a result, the past eight lines of defective video data are always stored in the shift register, and the pixel data of the 9×9 local areas on the defective video are read out in parallel from the output of the shift register. Since this local image is shifted by one pixel each time defective image data is input by one pixel, the output 9×9 local area image scans the entire image.

Now, if the 9×9 local area images read out at any time are P ₁₁ , P ₁₂ , ..., P ₉₉ , they are arranged as shown in Figure 7 on the two-dimensional video data, so the central pixel of the local area If we measure the distance from _P55 to the defect boundary in the radial direction, that is, the distance from _P55 to the pixel that becomes "0" for the first time in the radial direction, we can calculate the distance by adding the distances in the opposite direction. The defect size (width) in the direction can be measured. The signal flow from circuits 67 to 70 in FIG. 6 is a circuit for measuring the defect size for each region of the defect image in this manner. In FIG. 6, 67 is a selection circuit, 68a to 68h are dimension measurement circuits, 69a to 69d are addition circuits, 70 is a maximum value detection circuit, 76 is a read-only circuit, and 74 is a storage circuit.

The selection circuit 67 selects a local image 64a of the defective image on area A and a local image 64b of the defective image on area B as a signal indicating whether _P55 , the center pixel of the local image, is included in area A or B. 75. The signal 75 is the dictionary pattern data 23
_P55 signal circuit 6 by shift register 66
This is obtained by delaying by the delay due to 1a, that is, by 4 lines and 4 pixels.

The dimension measurement circuits 68a, 68b to 68h are the seventh
This circuit measures the size of a defect in each radial direction with the pixel _P55 as the center in the local area shown in the figure.
For example, taking 68a as an example, 68a has P ₅₅
Pixel data arranged to the right of the local area centered on
P ₅₄ , P ₅₃ , P ₅₂ , and P ₅₁ are connected, and the number of pixels until "0" appears for the first time in that order is counted. That is, if P ₅₅ = 0, “0”, P ₅₅ = 1, P ₅₄ =
1, if P ₅₃ =0, it is “2”. In the image of the local area, only the pixels of the defective part are "1", so the output value of 68a is the measurement of the dimension from _P55 to the right boundary. 68b similarly connects the pixel data arranged on the left side of _P55 , and measures the dimension from _P55 to the left boundary. Therefore, by adding the two dimensions using the adder 69a, the width of the defect can be calculated. Pixels 68c and 68d in the diagonal direction upward to the right are connected to each other, and the defect width in the diagonal direction upward right is calculated by the addition circuit 69b. In this way, the defect widths in the vertical direction and diagonal direction downward to the left are also determined, and the defect width in the four directions is determined by the maximum value detection circuit 70.
is input. The maximum value detection circuit 70 is a circuit that selects the maximum value of four widths, and as a result, the signal 7
1 approximately represents the maximum width of the defect. In this example, a 9x9 local area image is taken as the local image area, so the maximum defect width is up to 9 pixels, but if it is larger than a certain level, it will become a fatal defect.
If the size threshold is 9 pixels or less, there is no practical problem. Furthermore, by changing the number of pixels in the local image area, it is of course possible to arbitrarily change the range of detectable defect width.

On the other hand, 76 is a read-only memory, which is optimized for the area where P ₅₅ exists, by a dictionary pattern data signal 75 indicating whether the central pixel P ₅₅ of the local area is included in area A or area B. A threshold value 77 of the defect size is read out and compared with the maximum defect width 71 by a comparison circuit 72. As a result, if the defect maximum width is larger than the threshold value, it is a fatal defect and the signal 73 becomes "1". When the signal 73 returns from "1" to "0", the memory circuit 74 stores the current X,
The Y coordinate signal 27 is stored. If you do this,
The coordinate values of fatal defects on the object to be inspected are successively stored in the storage circuit in response to image scanning. The memorized defect coordinates are sent to the signal 30 after the inspection is completed.
is read out from the computer as .

In the embodiment shown in FIG. 6, the signals 64a and 64b are switched according to the dictionary pattern data 75, and the optimum threshold value 77 for each region is set by the read-only circuit 76.
It corresponds to the gist of the present invention that the data is read from the data source and inputted to the comparison circuit 72. By doing so, it becomes possible to accurately determine the fatality even if the pattern differs depending on the area, which was difficult to determine in the prior art.

As explained above, according to the present invention, it is possible to select and execute the optimal defect extraction/judgment process for each area with different properties of the pattern to be inspected, so it is possible to select and execute the optimal defect extraction/judgment process for each area with different properties of the pattern to be inspected. It becomes possible to inspect patterns with patterns, such as automatic inspection of completed IC and LSI wafers.

[Brief explanation of drawings]

Fig. 1 is a configuration diagram of a conventional pattern defect inspection device, Fig. 2 is an explanatory diagram of problems with the conventional device, Fig. 3 is an explanatory diagram of a method for solving the problems according to the present invention, and Fig. 4 is an illustration of a method of solving the problems according to the present invention. FIG. 5 is a diagram showing an embodiment of the present invention, FIG. 6 is a diagram showing another embodiment, and FIG. 7 is a diagram to help explain FIG. 6. be. DESCRIPTION OF SYMBOLS 1... Pattern or object to be inspected, 2... Video input device, 3... Binarization circuit, 4... Dictionary generation circuit, 5... Comparison circuit, 6... Judgment circuit, 12...
...AD converter, 13...Defect area extraction circuit, 14
... Fatality judgment circuit, 15 ... Read control circuit, 1
6... Memory circuit, 17... Synchronization signal generator, 18
...computer, 19...display device,
21...digitized input video signal, 22
. . . Video signal indicating defective area, 23 . ... Read-only memory (ROM), 49 ... Optimal threshold, 6
1a, b...Two-dimensional local image cutting circuit, l ₁ to _{l 8}
...A shift register with the number of pixels for one scanning line,
d ₁₁ to _{d 99} ...1 pixel shift register, 67...selection circuit, 68a to 68h...dimension measurement circuit, 69
a to 69d...Addition circuit, 70...Maximum value detection circuit, 71...Defect maximum width signal, 72...Comparison circuit, 74...Storage circuit for storing defect position, 75
...Delayed dictionary pattern data, 76...Read-only memory.

Claims (1)

[Claims] 1. A first means for scanning a target pattern to be inspected and inputting pixel data constituting the pattern to be inspected in a time-series manner; and each of the pixel data consists of a plurality of regions. a second means for generating pattern data indicating which area of the pattern to be inspected is obtained from the design data of the object to be inspected and generated in synchronization with the pixel data; and a threshold, an extraction range, and a determining means. A process of having a plurality of sets of at least one of them for each region, selecting one of the plurality of sets for each region according to the second means, and comparing and determining the pattern to be inspected and a reference pattern. A pattern defect inspection device comprising: a third means for changing the content of; 2 The third means is a patent that sets a different threshold value for each region and determines whether a pixel constitutes a defect by comparing each of the above pixel data with the threshold value of the region in which the pixel is included. A pattern defect inspection device according to claim 1. 3. The pattern according to claim 1, wherein the third means sets a threshold regarding dimensions for each region, and determines that each defective region is a fatal defect when its dimension is larger than the threshold. Defect inspection equipment.