JPH04107946A - Automatic visual inspector - Google Patents

Abstract

PURPOSE:To improve the availability factor of an apparatus so that a threshold value is determined based on the statistic of difference quantity data of plural points on an object to be inspected. CONSTITUTION:In a sample 3, many chips 3b are arranged in a rectangular lattice shape on one surface of a discoidal silicon substrate 3a. When one- dimensional line sensor such as CCD is used as an image pickup means 8 for such semiconductor wafer 3, the left side chip 3b out of adjacent two chips 3b is first taken in as picture signal A on an inspection width WT, processed by a signal-processing circuit 9 and thereafter housed in a picture storage 10. Then, the right side chip 3b is image-picked up as picture signal B on the same inspection width WD, processed by the signal-processing circuit 9 and thereafter sent to a difference detection circuit 11 without being housed in the picture storage 10. In the difference detection circuit 11, two picture signals are compared with each other and it is determined that there is a deficiency when the difference between the two signals exceeds a certain value. Thus, it is possible to set an optimum deficiency detection threshold value automatically and in a short time.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a technology for inspecting the appearance of the surface of an object such as a semiconductor wafer, and in particular, to a technology used to automatically perform an appearance inspection in the manufacturing process of semiconductor devices. It's about techniques that work.

[Conventional technology]

For example, when mass producing LSIs (Large Scale Integrated Circuits), the most important issue is improving the yield of the wafer processing process for forming semiconductor elements. Most of the causes of this decrease in yield are poor appearance, and reducing this has become an important issue. For this reason, it is necessary to automate the wafer visual inspection.

By the way, the present inventor has studied the problem of setting a threshold value when visual inspection of semiconductor Unino 1 substrates, masks, reticles, liquid crystals, etc. is performed using differential value processing.

The following are the techniques studied by the present inventor, and the outline thereof is as follows.

In image processing for this type of appearance inspection, the inspection target is imaged with a television camera, etc., the difference value obtained by comparing the image patterns between the two chips is used, and a threshold value is further set. Performing defect determination.

Methods for setting the threshold include, for example, determining the difference area of images and automatically setting the threshold according to the change in the maximum area, and setting the threshold based on the false detection rate of normal pattern areas. There are known methods to do this.

In addition, when the inspection target is a semiconductor wafer, there may be slight vibrations during the movement of the X-Y stage on which the semiconductor wafer is mounted, and dimensional differences in the pattern of the semiconductor wafer (product accuracy variations, for example, during the layer formation stage of the chip). (such as differences in the width of the patterns) can cause false detections. In this case, detection sensitivity is determined by mechanical precision and product precision, and the threshold value needs to be determined in consideration of mechanical precision and product precision.

[Problem to be solved by the invention]

However, in the method that uses the maximum area of the difference area as the criterion as described above, if the input image point contains a defect,
It is not possible to determine the threshold because the defective area is detected as the largest area, and the method of setting the threshold based on the false detection rate of the normal pattern area detects the normal pattern area that does not contain defects. The inventors of the present invention have found that there is a problem in that no consideration is given to identification, and defects must be visually confirmed.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a technique that can easily and accurately set a threshold value.

The objectives and novel features of the present invention will become apparent from the description of this specification and the accompanying drawings.

[Means to solve the problem]

A brief overview of typical inventions disclosed in this application is as follows.

In other words, it is an automatic visual inspection device that compares images of the same pattern part of an object to be inspected and determines defects based on the difference value and threshold value. The threshold value is determined based on statistics.

[Effect]

According to the above-mentioned means, the threshold determined based on the statistics of the difference amount data at multiple points on the object to be inspected can be set as an approximate determination value of the optimal threshold, or Can be used as a value. therefore,
It becomes possible to automatically and quickly set the optimal defect detection threshold.

[Embodiment 1] FIG. 1 is a block diagram showing an embodiment of an automatic visual inspection apparatus according to the present invention.

X-Y stage 1 that can freely move in the X and Y directions
A sample stage 2 is attached to the upper surface of the sample stage 2, and a sample (semiconductor wafer) 3 is set on this sample stage. On the other hand, a light source 4 is provided to illuminate the surface of a sample 3, which is an object to be inspected, and a condenser lens 5 is provided on the optical path thereof.

An objective lens 6 is disposed above the sample 3, and a half mirror 7 is disposed above this and on the output optical path of the condenser lens 5. Furthermore, an imaging means 8 is arranged at the focal position of the objective lens 6.

This imaging means 8 photoelectrically converts the reflected light from the sample 3, and is a -dimensional line sensor or a two-dimensional rTV.
(Industrial TV) Constructed using a camera. A signal processing circuit 9 is connected to the image pickup means 8 for amplifying the image signal, correcting distortion, A/D conversion, etc., and this signal processing circuit 9 has a circuit for storing the digitized image signal. An image storage unit 10 is connected thereto.

The signal processing circuit 9 is connected to a difference detection circuit 11 that detects the difference between the output signal of the image 8 self-memory section 10 and the signal between the signal processing circuit 9. A defect determination section 12 that determines defects is connected.

The difference detection circuit 11 is connected to a difference image storage section 13 that stores the detection results, and the defect determination section 12 is connected to a threshold register 14 that stores a threshold value. Further, a main control section 15 using a microcomputer or the like is provided to select the threshold value of the threshold register 14 based on the difference data of the difference image storage section 13.

In the above configuration, in order to conduct an external appearance inspection, first, the sample 3 is placed on the sample stage 2, and the light source 4 is turned on. The output light reaches the half mirror 7 through the condenser lens 5, and further reaches the sample 3 through the objective lens 6. The reflected light from the illuminated portion of the sample 3 passes through the half mirror 7 and forms a pattern on the imaging means 8. The image signal photoelectrically converted by the imaging means 8 is subjected to signal processing by the signal processing circuit 9, and then temporarily stored in the image storage section 10.

The image signal of the other chip stored in the image ε memory unit 10 and the image signal of the current chip directly output from the signal processing circuit 9 are compared by the difference detection circuit 11, and the difference between the two image signals is calculated. , the difference signal is stored in the difference image storage section 13
is memorized.

The main control unit 15 calculates an optimal threshold value based on the difference data stored in the difference image storage unit 13 and sets the threshold value in the threshold register 14 .

Next, a method for detecting defects will be described using a semiconductor wafer as an example.

FIG. 2 is a plan view showing the structure of a semiconductor wafer. In the sample 3, a large number of chips 3b are arranged in a grid on one side of a disk-shaped silicon substrate 3a.

When a -dimensional line sensor such as a CCD (charge-coupled device) is used as the imaging means 8 for such a semiconductor wafer 3, as shown in FIG. The left chip 3b receives an image signal A with an inspection width W7, and after being processed by the signal processing circuit 9, it is stored in the image storage unit 10. Next, the right chip 3b receives an image signal with the same inspection width W. B is imaged, processed by the signal processing circuit 9, and then sent to the difference detection circuit 11 without being stored in the image storage unit 10.The difference detection circuit 11 receives the two image signals (diagonal J1 part) in FIG. A defect is determined when the difference is greater than a certain value.

In addition, when the imaging means 8 is constituted by ITV, as shown in FIG. 4, among the two adjacent chips 3b, the left chip 3b first captures the image signal C with the inspection width WT, and the third r! As in the case of lJ, this is processed by the signal processing circuit 9
After processing, the image is stored in the image storage unit 10. Next, the right chip 3b is tested by the same test l1WIII.
is imaged as an image signal, processed by the signal processing circuit 9, and then sent to the difference detection circuit 11 without being stored in the image storage section 10. In the difference detection circuit 11, 2 in FIG.
The two image signals (shaded areas) are compared, and a defect is determined when the difference is greater than a certain value.

Figure 5 and '! Figure J6 explains how a differential signal appears when patterns at the same position on two chips are compared. As shown in FIG. 5, pattern 1 shows a normal part as a reference for comparison, and pattern 2 is a comparison target including a defect 16. Although the pattern widths (pattern dimensions) are slightly different between the two, other dimensional differences are not considered defects. When the absolute value of the difference between the image signals of such a pattern is taken, as shown in FIG. 6, a difference signal is produced in the defective part, but at the same time, a difference signal is also produced in the normal part due to the difference in pattern dimensions.

In order to correctly detect only defects and not detect normal parts as defects, the value of the difference signal in the normal part becomes a problem. Therefore, it is necessary to set the threshold value for defect detection to a value that does not determine a normal part as a defect.

Next, the principle of setting the optimum threshold value, which is a feature of the present invention, will be explained.

FIG. 7 is an explanatory diagram qualitatively showing the relationship between the threshold value, defect detection rate, and false detection.

Generally, as the threshold value is lowered, the number of false detections increases rapidly below the threshold value. On the other hand, the defect detection rate does not increase as rapidly as the false detection rate. Therefore, as shown in FIG. 8, if the optimal threshold value is set to a value that does not result in a large number of false positives, efficient testing becomes possible.

FIG. 8 is an explanatory diagram that compares two image signals in a certain range and shows the relationship between the value of the difference signal and the frequency. As is clear from the figure, the frequency of difference values decreases rapidly as the difference value increases. However, if a defect exists in the image, a difference value will occur at a point with a value higher than the optimal threshold. Therefore, based on images at multiple points, the frequency distribution of difference values shown in Figure 8 is taken, and the maximum difference value (a+>) at each image input point is taken, and the image has a relatively large maximum difference value. The human input point is regarded as an area where a defect may exist, and the optimal threshold value is calculated based on the maximum difference value at other image input points.

Next, details of the optimum threshold value determination process of the present invention will be explained with reference to FIGS. 9 to 13. FIG. 9 is a flowchart showing the initial threshold estimation process in the present invention, FIG. 10 is a flowchart showing the process for automatically setting the optimal threshold, and FIG. 11 is an explanatory diagram showing the relationship between the maximum difference value and frequency. FIG. 12 is an explanatory diagram showing the relationship between the threshold value and the number of detections, and FIG. 13 is an explanatory diagram showing the relationship between the difference value and cumulative frequency (ratio). In the following, processing from difference detection to subsequent steps will be explained.

After detecting the difference (step 91), the maximum value of the difference is detected (step 92). Next, a histogram of the maximum difference value is calculated (step 93). That is, j3
As shown in FIG. 11, when the image human effort points differ, the shape of the pattern also differs, and the maximum difference value varies.

Here, when the average value of the maximum difference values is SAY@, the standard deviation of the maximum difference values is σ3, and the optimal threshold value is TH, the optimal threshold value TH is expressed by the following equation.

TH=SAY@+nσ.

Here, n is a real number, and according to the inventors' experiments, n
The best results were obtained at approximately =2.

The histogram calculation continues until a preset number N is reached (step 94), and when the number N is reached, the average value of the histogram of the maximum difference value is calculated and this is set as the initial threshold value THO (step 95). ).

Next, specify the inspection area (specify which of the plurality of chips is to be inspected) (step 101), and step 95
The initial threshold value THO is stored in the threshold value register 14 as the threshold value THO (step 102). Next, the x-y stage 1 is driven to start automatic inspection (step 103), and the number of detected defect candidates is stored in a data table (step 104).

Then, centering on the initial threshold THO, find the threshold value at multiple points above and below (in this example, 5 points, M=5).
Steps 105, 106). After determining the threshold value at point M, a graph as shown in FIG. 12 is created (step 107).

In FIG. 12, the flat part of the curve indicates an actual defect detection area, and the slope part indicates an erroneous detection area.

Here, the difference in the number of detections (of defect candidates) between the upper and lower thresholds THI, TH2, TH3, and TH4 found around the initial threshold THO is calculated.
It is determined whether the point where the slope of the curve in the figure suddenly changes (THO in the figure) is appropriate as the optimal threshold (step 108).
.

If the difference between the curves near THO is small, the threshold value is too high, so the optimal threshold value is modified to be lower. On the other hand, if the slope of the curve is too steep, the threshold value is too low, so the optimal threshold value is corrected to be higher.

In the above process, instead of determining the initial threshold THO, a large number of thresholds are set at regular intervals to determine the number of detections, and the point where the slope of each threshold value changes suddenly is determined. The optimal threshold value can be determined.

[Embodiment 2] FIG. 13 is an explanatory diagram showing another embodiment of another method of determining the optimum threshold value according to the present invention.

In this example, all the cumulative frequency distributions of difference values at the image input points of the normal pattern portion of multiple points are accumulated, and the value at which the ratio S of the shaded area in the figure is equal to or less than a certain ratio is set as the optimal threshold value TH. . In this case, the value of the ratio S is experimentally about 10-s, and the number of image input points is set so that enough data can be obtained to determine the forward rotation ratio S with high reliability.

Although the invention made by the present invention has been specifically explained based on Examples above, it goes without saying that the present invention is not limited to the above-mentioned Examples and can be modified in various ways without departing from the gist thereof. .

For example, in the embodiments described above, when determining the optimal threshold value, any other method may be used as long as it performs statistical processing based on the distribution of difference values.

In the above description, the invention made by the present inventor is mainly applied to the visual inspection of semiconductor wafers, which is the field of application of the invention, but it is not limited to this. It is also possible to apply this method to the appearance inspection of masks, etc.

〔Effect of the invention〕

Among the inventions disclosed in this application, the effects obtained by typical ones are as follows.

In other words, it is an automatic visual inspection device that compares images of the same pattern part of an object to be inspected and determines defects based on the difference value and threshold value. Since the threshold value is determined based on statistics, it is possible to automatically and quickly set the optimal defect detection threshold value, improving equipment utilization rate and reducing the burden on workers. It can be reduced.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing an embodiment of an automatic visual inspection device according to the present invention, Fig. 2 is a plan view showing the configuration of a semiconductor wafer, and Fig. 3 is an explanatory diagram for comparative inspection when a one-dimensional line sensor is used. , Fig. 4 is an explanatory diagram of comparative inspection when using ITV, Fig. 5 is an explanatory diagram showing an example of a normal pattern and a defective pattern, Fig. 6 is a differential value output characteristic diagram corresponding to the pattern of Fig. 5, Figure 7 is an explanatory diagram that qualitatively shows the relationship between the threshold value, defect detection rate, and false detection. Figure 8 compares two image signals in a certain range and shows the relationship between the value of the difference signal and the frequency. 9 is a flowchart showing the initial threshold estimation process in the present invention, FIG. 10 is a flowchart showing the process for automatically setting the optimal threshold, and FIG. 11 is a flowchart showing the relationship between the maximum difference value and frequency. FIG. 12 is an explanatory diagram showing the relationship between the threshold value and the number of detections. FIG. 13 is an explanatory diagram showing another example of the relationship between the difference value and the cumulative frequency (ratio). DESCRIPTION OF SYMBOLS 1... X-Y stage, 2... Sample stand, 3... Sample, 3a... Silicon substrate, 3b... Chip, 4...
... light source, 5 ... condensing lens, 6 ... objective lens,
7... Half mirror 18... Imaging means, 9... Signal processing circuit, 10... Image storage section, 11... Difference detection circuit, 12... Defect determination section, 13... Difference Image storage unit, 14... Threshold register, 15... Master control unit, 16... Defect. Agent: Daiwa Tsutsui, Patent Attorney Figure 2, Figure 3, 6b: “”. Figure 4 Figure 5 Figure 6 Figure 7 - Threshold Figure 8 Figure 9 Figure 10 Figure 11

Claims (1)

[Claims] 1. An automatic visual inspection device that compares images of the same pattern portion of an object to be inspected and determines defects based on a difference value and a threshold value, the apparatus comprising: An automatic visual inspection device characterized in that a threshold value is determined based on statistics of point difference amount data. 2. The automatic visual inspection apparatus according to claim 1, wherein a portion of the difference value distribution where the difference value becomes abnormally large corresponding to a defective portion is removed from the processed data. 3. The feature is that the maximum value of the difference value of image signals at multiple points on the object to be inspected or the cumulative frequency distribution of the difference values is determined, and the difference value of a certain ratio from the maximum value of the difference value is set as a threshold value. The automatic visual inspection device according to claim 1. 4. An automatic visual inspection device that compares images of the same pattern part of the object to be inspected and determines defects based on the difference value and threshold value, and stores the threshold value for each type of object to be inspected. and a threshold setting means that determines the threshold values at predetermined intervals and sets a changing point at each of which the number of detected defects rapidly increases as an optimal threshold value. Automatic appearance inspection equipment. 5. The threshold setting means determines the maximum value of the difference values of the image signals at a plurality of points on the object to be inspected or the cumulative frequency distribution of the difference values, and calculates the difference value at a fixed ratio from the maximum value of the difference values. 5. The automatic visual inspection apparatus according to claim 4, further comprising means for setting a threshold value and determining a plurality of threshold values before and after the threshold value.