JP2002195958A - Surface inspecting method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a surface inspecting method which can accurately inspect without improper determination even if intensity values of objects to be inspected have a wide variation. SOLUTION: The surface inspecting method measures the intensity values of surfaces of a plurality of the objects previously determined as conforming items in all inspected units, and has: a step for calculating the variation of the intensity values per the inspected unit; a step for deciding criteria of determining whether the objects to be inspected conform per the inspected unit based on a degree of the calculated variation; and a step for measuring the intensity values of the surfaces of the objects to be inspected in all inspected units and determining whether the objects to be inspected conform based on the decided criteria.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a defect to be inspected such as a circuit pattern formed on the surface of a fine chip such as a sensor chip (shape defect such as crack, peeling, deformation, presence of foreign matter, etc.). The present invention relates to a surface inspection method for automatically detecting a surface.

[0002]

2. Description of the Related Art In the case of a microchip such as a sensor chip formed by applying a micromachining technique, several thousand chips are formed on a single silicon wafer in a square of about 1 to 2 mm. Various circuit patterns are formed on each chip, and the patterns tend to be highly integrated and miniaturized. Detecting such a pattern defect is an important factor in determining the quality of a chip.

[0003] Defects are diversified, such as foreign substances such as dust attached on the chip, cracks, deformation and peeling of the pattern, and the number of chips to be inspected is very small and many.
It is difficult to visually detect defects. Therefore, a technology for automatically detecting such a defect has been developed.

For example, a circuit pattern on a chip surface to be inspected is picked up by a CCD camera or the like, and a difference image between the picked-up image (inspection image) and a non-defective image prepared in advance is generated. There is a method using an image processing technique of detecting. Here, the “good image” is an image generated by using all pixels as a certain luminance value (threshold), and the “difference image” is a black and white image generated by comparing the luminance values of the inspection image and the good image. It is a binary image. As a specific generation method of the difference image, a method is used in which all pixels of the inspection image are compared with the luminance value of the non-defective image, pixels that exceed the luminance value are set to white, and other pixels are set to black. I have.

[0005] In the difference image thus obtained, regions in which white color is continuous are grouped together, and a pass / fail judgment is made based on whether or not the area (number of pixels) of the region with the single frame exceeds a predetermined reference.

[0006]

In the conventional surface inspection method as described above, the threshold value (luminance value on a non-defective image) as a reference when generating a difference image is a constant value over the entire surface of the chip. . For example, the threshold is determined by an average value of luminance values of all pixels of a plurality of non-defective images. However, the brightness value of the chip surface is not uniform over the entire surface, but varies (irregularly distributed). For example, the brightness value on the chip surface has a level (unevenness) depending on the location. In this case, since the threshold value is a constant value in the conventional method, it is difficult to determine a non-defective product in consideration of the variation in the brightness value of the chip surface.

[0007] For example, if the luminance value is extremely high in a certain area on the chip surface, depending on the set threshold value, a situation may occur in which the product is determined as a good product but as a defective product. As described above, in the conventional method, it is difficult to determine the acceptability with high accuracy because of the influence of the variation of the luminance value.

An example of an inspection object having a variation in luminance value is, for example, a sensor chip used for a flow sensor. In this sensor chip, there are many variations in the brightness value of the portion (diaphragm portion) where the flow of air is detected. However, if a threshold value is set without considering such variations, that is, if a pass / fail judgment criterion is determined, an erroneous judgment may occur.

An object of the present invention is to provide a surface inspection method capable of performing a high-precision inspection without erroneous determination even for an inspection object having a large variation in luminance value.

[0010]

According to a first aspect of the present invention, there is provided:
The luminance values of the surfaces of a plurality of objects determined in advance as non-defective products are measured for all inspection units (for example, individual pixels constituting an image obtained by photographing with a camera), and the luminance value of each inspection unit is measured. A step of calculating the variation; a step of determining a quality judgment criterion of the inspection target for each inspection unit according to the calculated degree of variation; and measuring and determining the luminance value of the surface of the inspection target for all inspection units. Performing a pass / fail judgment on the inspection target based on the pass / fail judgment criteria.

According to a second aspect of the present invention, the luminance values of the surfaces of a plurality of objects determined in advance as non-defective products are measured for all inspection units, and the average value and standard deviation of the luminance values for each inspection unit are calculated. Determining the upper limit luminance value and the lower limit luminance value for each inspection unit based on the calculated average value and standard deviation; measuring the luminance value of the surface to be inspected for all inspection units; Extracting a location exceeding the value or a location not reaching the lower limit luminance value as an out-of-range location, performing an inspection on the extracted out-of-range location based on a predetermined inspection criterion, and determining an inspection target based on the inspection result. Performing a pass / fail determination.

According to a third aspect of the present invention, a luminance value is measured for all pixels of an image obtained by photographing the surface of a plurality of objects previously determined to be good, and an average value of the luminance values for each pixel is measured. Calculating an upper limit luminance value and a lower limit luminance value for each pixel based on the calculated average value and standard deviation, and an upper limit learning in which each pixel is configured with the determined upper limit luminance value. Generating an image, and a lower limit learning image in which each pixel is composed of the determined lower limit luminance value, and all pixels of an image obtained by photographing the surface of the inspection target and all pixels of an upper limit learning image and a lower limit learning image Comparing the pixel having a higher luminance value than the upper limit learning image and the pixel having a lower luminance value than the lower limit learning image as the out-of-range pixels; and Two distinct values A surface inspection method comprising: an image generation step; an inspection performed on the generated binary image based on a predetermined inspection standard; and a pass / fail determination of the inspection target based on the inspection result. It is.

[0013]

According to the first aspect of the present invention, the quality criterion of the inspection object is determined for each inspection unit in accordance with the degree of variation of the luminance value, and the inspection is performed based on the determined quality criterion. Since the quality of the target is determined, the quality can be determined without being affected by the variation in the brightness value. For example, even if there is a portion on the chip surface where the luminance value is extremely high compared to other portions, the pass / fail judgment is performed based on the criterion excluding such a luminance value. It is possible to avoid the over-determination of being determined.

According to the second aspect of the present invention, the upper limit luminance value and the lower limit luminance value for each inspection unit are determined using the average value and the standard deviation as indices indicating the variations. The value calculation process can be easily understood,
It is extremely easy to create and modify a processing program for executing the surface inspection method.

According to the third aspect of the present invention, the inspection image is compared with the two learning images, the upper limit learning image and the lower limit learning image, to generate a binary image which is used as a criterion for the pass / fail judgment. It can be directly applied to a surface inspection method using image processing.

[0016]

FIG. 1A shows a basic configuration of a chip inspection apparatus 1 for implementing a surface inspection method according to the present invention. The chip inspection apparatus 1 includes an inspection table 8 in which an inspection table 4 on which an inspection target such as a sensor chip is placed is disposed at the center;
A robot arm 3 that can be driven in three XYZ directions. The chip inspection device 1 is electrically connected to the computer 10 by a cable 1a, and controls the driving of the robot arm 3 according to a command from the computer 10. With this drive control, the imaging device 2 attached to the robot arm 3 can move to a position where the surface of the inspection target can be imaged.

Image data obtained by photographing the surface of the object to be inspected by the imaging device 2 is sent to the computer 10, where the surface inspection of the object to be inspected is performed based on the image processing. Thereby, the quality of the inspection object is determined.

FIG. 1B shows the imaging device 2 in an enlarged manner. The imaging device 2 includes a CCD camera 5, a microscope 6, and a lighting device 7. Dark field illumination is used for the illumination device 7, and an image in which unevenness on the surface of the inspection target is emphasized can be obtained by reflected light obtained when the light of the illumination is applied.

FIG. 2 is a block diagram showing main components in the computer. The captured image of the inspection object obtained by the imaging device 2 is A / D converted into image data, and stored in the inspection image storage unit 16 as storage means.
CPU1 as an arithmetic processing means for the image data
2 executes predetermined image processing according to a processing program stored in the ROM 13 and performs surface inspection processing based on the image processing. The inspection image obtained by the imaging device 2, the result image obtained by the surface inspection processing, and the like are displayed on the display 11 by the image display control circuit 20, and the state of the inspection target can be visually checked.

In the surface inspection method according to the present invention, the quality of the inspection is determined by performing image processing for comparing an inspection image obtained by imaging the inspection object with a learning image (upper limit learning image and lower limit learning image, which will be described later) based on non-defective products. Make a decision.

As a procedure for generating a learning image, first, a plurality of picked-up images (hereinafter, referred to as "learning image generation images") of inspection objects selected as non-defective products are prepared. Find the value. Here, the learning image generation image is stored in the learning image generation image storage unit 18 shown in FIG. Next, an upper limit luminance value and a lower limit luminance value are calculated for each pixel constituting the captured image of the inspection target based on the detected luminance value data. Then, two learning images are generated, an "upper limit learning image" in which all pixels are composed only of the upper limit luminance value and a "lower limit learning image" in which all pixels are composed only of the lower limit luminance value.

The number of images for generating a learning image necessary for generating a learning image is, for example, about 20. This is because no particular change is observed in the average value and the standard deviation of the brightness values obtained for each pixel even if 20 or more learning image generation images are collected.

The flowchart of FIG. 3 shows a specific procedure of the "learning image generation process" for generating a learning image. This “learning image generation processing” is executed by the CPU 12 in accordance with the processing program stored in the ROM 13.

ST11: Image storage unit 18 for generating a learning image
Fetches a plurality of learning image generation images stored in
Stored in AM14.

ST12: A luminance value is detected for each pixel for each learning image generation image, and an average value of the luminance values for each pixel is calculated according to the following equation.

[0026]

(Equation 1) ST13: The standard deviation of the luminance value for each pixel is calculated according to the following equation.

[0027]

(Equation 2) ST14: Create a normal distribution based on the obtained average value and standard deviation.

ST15: An upper limit luminance value and a lower limit luminance value are calculated for each pixel from the created normal distribution.

ST16: Generate a learning image in which all pixels are composed of the upper limit luminance value (hereinafter referred to as "upper limit learning image").

ST17: Generate a learning image in which all pixels are constituted by the lower limit luminance value (hereinafter, referred to as "lower limit learning image").

According to the above procedure, two learning images (upper limit learning image and lower limit learning image) are generated, and the learning images are stored in the learning image storage unit 19 shown in FIG.

The upper limit luminance value and the lower limit luminance value calculated in ST15 are obtained by the following formulas.

[0033]

(Equation 3)

Next, based on the learning image obtained in the above-mentioned "learning image generation process", a "surface inspection process" for determining a non-defective product of the inspection object is executed.

FIG. 4 is a flowchart showing the procedure of the "surface inspection process". This processing is also executed by the CPU 12 according to the processing program stored in the ROM 13.

ST21: Image the surface of the inspection object,
The image (hereinafter referred to as “inspection image”) is stored in the inspection image storage unit 16.

FIG. 5 shows an example of the "inspection image". This is an image when the surface of the sensor chip is imaged as the inspection target.

ST22: Perform "pre-processing" such as noise removal.

ST23: Generate a difference image between the upper limit learning image and the lower limit learning image stored in the learning image storage unit 19 and the inspection image. This difference image is a binary image composed of two colors, white and black. To generate the difference image, specifically,
All the pixels of the upper limit learning image and the lower limit learning image are compared with all the pixels of the corresponding inspection image.

First, all the pixels of the upper limit learning image are compared with all the pixels of the inspection image corresponding to the upper limit learning image, and a pixel having a higher luminance value than the upper limit learning image is set to white. Further, all the pixels of the lower limit learning image are compared with all the pixels of the inspection image corresponding to the lower limit learning image, and a pixel having a luminance value lower than the lower limit learning image is also set to white. On the other hand, for pixels that do not correspond to any of them, that is, for pixels having a luminance value between the upper limit learning image and the lower limit learning image, a binary image is generated as black.

FIG. 6 shows an example of the difference image. In FIG. 6, a portion represented by white indicates a pixel having a higher luminance value than the upper-limit learning image or a lower luminance value than the lower-limit learning image (referred to as “out-of-range pixel”).

ST24: "Post-processing" is performed to convert pixels other than a region where pixels represented by white (hereinafter, referred to as "white pixels") in the obtained difference image are continuous into black.

ST25: In the image after "post-processing",
A “labeling process” is performed in which regions in which white pixels are continuous are grouped together for each region, and a number is assigned to each of the regions so as to be individually distinguished.

FIG. 7 shows an image after the labeling process. As shown in FIG. 6, the regions (1, 2, 3,...)
Is appended.

ST26: The area of a rectangular area, that is, the number of pixels in the area, is measured with the maximum number of pixels in the vertical and horizontal directions of each area numbered in the labeling process.

ST27: The measurement result of ST26 is illuminated on the basis of the judgment criterion, and a judgment is made as to whether there is an area exceeding the judgment criterion. If there is an area that exceeds the criteria,
The inspection target is determined as a defective product, and if all the areas are within the determination standard, the inspection target is determined as a non-defective product.

ST28: A result image is generated and displayed on the display 11.

FIG. 8 shows an example of the result image. here,
This is a case where a defective portion of the result image is enlarged and displayed. As shown in FIG. 8, the defective spots are displayed by enclosing them with a solid line circle so that the defective spots on the surface of the inspection object (in this case, the sensor chip) can be easily grasped visually.

FIG. 9 is a graph showing a change in luminance value at an arbitrary cross section of the upper limit learning image and the lower limit learning image. In this graph, a line located on the upper side indicates a change in the luminance value of the upper limit learning image, and a line located on the lower side indicates a change in the luminance value of the lower limit learning image. Therefore, when the luminance value in the relevant section of the inspection image exceeds a line indicating a change in the luminance value of the upper limit learning image (hereinafter referred to as an “upper limit luminance value line”), or a line indicating a change in the luminance value of the lower limit learning image. (Hereinafter, referred to as “lower limit luminance value line”), the difference image is shown in white in the difference image.

As is clear from the above formula,
For a portion where the luminance value does not change much even if the inspection object changes, that is, a portion where the variation of the luminance value is small, the width of change between the upper limit luminance value line and the lower limit luminance value line is inevitably narrow. In the case of the cross section shown in FIG. 9, the change width of the upper limit luminance value line and the lower limit luminance value line is narrow in a part α of the graph. As described above, in the portion where the change width between the upper limit luminance value line and the lower limit luminance value line is narrow, for example, when the inspection target is a sensor chip, the variation of the luminance value for each inspection target such as the outer peripheral portion (see FIG. 5). It is a small part.

On the other hand, in a place where the luminance value largely changes depending on the inspection object, that is, in a place where there is a large variation in the luminance value, the change width of the upper limit luminance value line and the lower limit luminance value line necessarily increases. In the case of the cross section shown in FIG. 9, in a part β of the graph, the range of change between the upper limit luminance value line and the lower limit luminance value line is wide. As described above, when the inspection object is a sensor chip, for example, the variation of the luminance value for each inspection object such as a diaphragm (see FIG. 5) is large in the portion where the variation width between the upper luminance value line and the lower luminance value line is wide. It is a big part.

FIG. 10 is a graph showing the difference between the luminance value of the upper limit learning image and the luminance value of the lower limit learning image in absolute value in the same arbitrary section as in FIG. . In FIG. 10, the portion corresponding to the portion α in FIG. 9 has a small absolute value of the difference in luminance value, and the portion corresponding to the portion β in FIG.
The absolute value of the difference between the luminance values is large.

As described above, by generating a learning image based on a plurality of non-defective images, the inspection accuracy is determined in accordance with the magnitude of the variation in the luminance value for each pixel. That is, it is possible to detect a defect in consideration of the variation of the luminance value for each pixel. As described above, even when inspecting a portion where the difference between the brightness values of the upper limit learning image and the lower limit learning image is small, that is, a portion where the variation of the brightness value on the non-defective image is small, it is not affected by the portion where the variation is large. High-accuracy defect detection becomes possible.

[Brief description of the drawings]

FIG. 1 is a diagram showing a basic configuration of a chip inspection apparatus to which a surface inspection method according to the present invention is applied.

FIG. 2 is an exemplary block diagram showing main components of the computer.

FIG. 3 is a flowchart illustrating a procedure of a learning image generation process.

FIG. 4 is a flowchart illustrating a procedure of a surface inspection process.

FIG. 5 is a diagram showing an example of an inspection image.

FIG. 6 is a diagram illustrating an example of a difference image.

FIG. 7 is a diagram showing an image after a labeling process.

FIG. 8 is a diagram showing an example of a result image.

FIG. 9 is a graph showing a change in luminance value at an arbitrary cross section of the upper limit learning image and the lower limit learning image.

FIG. 10 is a graph showing a change in an absolute value of a difference between luminance values of an upper limit learning image and a lower limit learning image.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Chip inspection device, 2 ... Imaging device, 3 ... Robot arm, 4 ... Inspection table, 5 ... CCD camera, 6 ... Microscope, 7 ... Lighting device, 8 ... Inspection table, 10 ... Computer
11 display, 12 CPU, 13 ROM, 1
4 RAM, 16 inspection image storage unit, 17 non-defective image storage unit, 18 learning image generation image storage unit, 19 learning image storage unit.

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 2F065 AA00 AA49 AA61 CC00 DD04 FF04 FF26 FF41 HH16 JJ03 JJ26 PP24 QQ04 QQ24 QQ42 QQ43 SS04 UU05 2G051 AA51 AA65 AB07 AC21 CA04 EA08 EA11 BA03 EC02 BA02 EC03 DB02 DB09 DC14 DC23 DC33 DC40

Claims (3)

[Claims] A step of measuring the luminance value of the surface of a plurality of objects which have been previously determined to be non-defective for all inspection units and calculating a variation in the luminance value for each inspection unit; Determining a pass / fail criterion of the inspection target for each inspection unit in accordance with the step; measuring the luminance value of the surface of the inspection target for all the inspection units and performing pass / fail determination of the inspection target based on the determined pass / fail determination criterion And a step for inspecting the surface. 2. A step of measuring the luminance values of the surfaces of a plurality of objects determined in advance as non-defective products for all inspection units, and calculating an average value and a standard deviation of the luminance values for each inspection unit. Determining an upper limit luminance value and a lower limit luminance value for each inspection unit based on the average value and the standard deviation; andmeasuring the luminance value of the surface to be inspected for all inspection units, and a portion exceeding the upper limit luminance value or Extracting a location that does not reach the lower limit luminance value as an out-of-range location, performing an inspection on the extracted out-of-range location based on a predetermined inspection criterion, and performing a pass / fail determination on the inspection target based on the inspection result A surface inspection method comprising: 3. A step of measuring luminance values of all pixels of an image obtained by photographing the surface of a plurality of objects determined in advance as non-defective, and calculating an average value and a standard deviation of the luminance values for each pixel. Determining an upper limit luminance value and a lower limit luminance value for each pixel based on the calculated average value and standard deviation; an upper limit learning image in which each pixel is configured with the determined upper limit luminance value; and a determined lower limit. Generating a lower limit learning image in which each pixel is composed of a luminance value, and comparing all pixels of an image obtained by photographing the surface of the inspection target with the upper limit learning image and all pixels of the lower limit learning image, Determining a pixel having a higher brightness value than the upper limit learning image and a pixel having a lower brightness value than the lower limit learning image as an out-of-range pixel; and discriminating a portion determined as the out-of-range pixel from other portions. Value image And inspecting the generated binary image based on a predetermined inspection criterion, and determining the quality of the inspection target based on the inspection result.