JP6065342B2 - Edge detection device - Google Patents

Description

  The present invention relates to an edge detection apparatus for performing a kerf check with respect to an edge detection apparatus.

  A structure in which a low dielectric constant insulating film (Low-k film) using a low dielectric constant (Low-k) material is formed as an interlayer insulating film on the surface of a semiconductor substrate in order to increase the processing capability of a semiconductor device such as an IC or LSI. There are known semiconductor wafers. However, since the low dielectric constant insulating film has low mechanical strength, there is a problem that the low dielectric constant insulating film peels off when the semiconductor wafer is diced with a grinding blade.

  In order to solve this problem, the low dielectric constant insulating film is removed by irradiating the laser beam to the division line without using the grinding blade, and then the grinding blade is positioned in the laser groove formed by the laser light. A method of cutting and dividing a wafer is employed.

  Even in the method of dividing a semiconductor wafer with a laser beam and a grinding blade, a kerf check is required to confirm the situation such as a positional deviation between a groove (kerf) cut by the grinding blade and a division planned line. Proposals have been made to check the kerf in the laser groove.

  In Patent Document 1, the amount of light is set so that the kerf portion formed by the grinding blade is white and the surrounding laser grooving portion is black, and the division line after dicing by the grinding blade is imaged by the imaging means, and the captured image The edge position of the laser grooving area and the edge position of the kerf area are extracted by edge recognition processing for the image, and a histogram is created regarding the luminance distribution of the image data of each pixel in the laser grooving area of the predetermined range where the edge position is extracted. The brightest first peak region is extracted from the laser grooving region in a predetermined range based on the luminance distribution in the created histogram, and the portion that continues to the edge position of the kerf region in the extracted first peak region is chipped A chipping detection device that is recognized as a region is described.

  In Patent Document 2, the amount of light is set so that the kerf portion formed by the grinding blade is white and the surrounding laser grooving portion is black, and the division planned line after dicing by the grinding blade is imaged by the imaging means, and the captured image A candidate line as a grooving edge line or a kerf edge line is extracted by edge recognition processing for, and the nature of each extracted candidate line is determined based on information of the original image, and a plurality of candidate lines are determined as grooving edge lines according to the determination result, Cases are divided into combination candidate lines that form a set of kerf edge lines, kerf edge lines, and grooving edge lines, and the difference in physical characteristics between the laser grooving part and the kerf part is determined as a judgment factor from among the combination candidate lines that are divided into cases. Probable one based on the information of the original image Determining the combination candidate line describes an edge detection device to identify the kerf edge line from the candidate line.

JP 2009-021375 A JP 2010-010445 A

  In the techniques described in Patent Documents 1 and 2, the amount of light is set so that the kerf region formed by the grinding blade is white and the surrounding laser grooving region is black.

  However, the state of the bottom surface of the kerf formed on the semiconductor wafer differs depending on the use situation of the grinding blade. Therefore, even if the light amount is adjusted, the kerf region may not be uniform white, and the contrast at the boundary between the kerf region and the laser grooving region may be unclear. Even if image processing is performed using unclear image data, an appropriate detection result cannot be obtained.

  In order to solve such a problem, an object of the present invention is to provide an edge detection device capable of clearly recognizing a boundary portion between a kerf region and a laser groove region.

  According to an aspect of the present invention, an edge detection apparatus includes: a laser groove formed by irradiating a laser beam along a predetermined division line; and the laser formed by grinding with a grinding blade along the laser groove. With respect to a semiconductor wafer having a kerf deeper than the groove, the dividing line, the laser groove, and an imaging unit that images the kerf from above are arranged between the dividing line and the imaging unit. An objective lens whose depth of field is shallower than the depth of the kerf, and an illuminating device that irradiates the objective lens with light from a direction coaxial with an observation optical axis connecting the objective lens and the imaging means.

  Preferably, in the edge detection device, the objective lens is arranged so that a central axis of the objective lens is located at a center in a width direction of the kerf.

  Preferably, the edge detection device is obtained by the light amount control unit that adjusts the illumination device so that the kerf region is bright, the laser groove region is dark, and the division line region is bright, and the imaging unit. And an image processing unit that detects the kerf region, the laser groove region, and the division line region.

  Preferably, in the edge detection device, the light amount control unit adjusts the illumination device so that the kerf region and the laser groove region are dark and the division line region is bright, and the image processing unit is The laser groove area and the division line area are detected from the image obtained by the imaging means.

  It should be noted that the light / dark recognition of the kerf, the groove, and the line to be divided can be detected by using an image recognition means by CCD or an image processing means.

  The edge detection device of the present invention can clearly recognize the boundary between the kerf region and the laser groove region.

The perspective view of a semiconductor wafer. The fragmentary sectional view of a semiconductor wafer. The fragmentary sectional view of the semiconductor wafer in which the laser groove was formed. The fragmentary sectional view of the semiconductor wafer half-cut. The fragmentary sectional view of the semiconductor wafer cut full. Schematic which shows the relationship between the depth of field of an objective lens, and the depth of a laser groove and a kerf. Explanatory drawing which shows the result of having imaged the imaging area by the imaging means. Schematic which shows the relationship between the depth of field of an objective lens, and the depth of a laser groove and a kerf. Explanatory drawing which shows the result of having imaged the imaging area by the imaging means. The schematic block diagram of an edge detection apparatus. The schematic block diagram of the dicing apparatus with which the edge detection apparatus was attached. The flowchart of an edge detection method. The flowchart of another edge detection method. The flowchart of another edge detection method. The fragmentary sectional view of the semiconductor wafer half-cut. The schematic block diagram of the optical system of an edge detection apparatus.

  Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the accompanying drawings. Here, in the drawing, portions indicated by the same symbols are similar elements having similar functions.

  Although laser grooving will be described as an example, the present invention can be applied even when equivalent visibility is obtained, such as a semiconductor wafer in which a planned division line (street) appears black.

  The basic concept of the edge detection apparatus of this embodiment will be described with reference to the drawings. FIG. 1 is a perspective view showing the appearance of the semiconductor wafer 10. A semiconductor wafer 10 shown in FIG. 1 includes a substrate 12 made of Si, an element region 14 formed two-dimensionally on the substrate 12, and a division line 16 formed to cut the element region 14 into chips. Consists of. A dicing tape 18 is attached to the back surface of the semiconductor wafer 10.

  FIG. 2 is a partially enlarged view of the semiconductor wafer 10. As shown in FIG. 2, a low dielectric constant insulating film (Low-k film) 20 is formed on the surface of the substrate 12. In the element region 14, metal wiring is formed using the semiconductor element and the low dielectric constant insulating film 20 as an interlayer insulating film. The semiconductor wafer 10 shown in FIG. 2 is irradiated with laser light along the planned division line 16.

  As shown in FIG. 3, the low dielectric constant insulating film 20 is dissolved and vaporized by the heat of the laser beam, and the low dielectric constant insulating film 20 in the region of the division planned line 16 is removed. By irradiating the laser beam, a laser groove 22 is formed on the surface of the division line 16 of the substrate 12. Since the surface of the laser groove 22 is melted by the heat of the laser light, the surface of the laser groove 22 is in a rough state compared to the surface of the scheduled division line 16. The depth of the laser groove 22 is about 5 to 15 μm from the surface of the substrate 12. The semiconductor wafer 10 is irradiated with laser light by a laser processing machine or the like.

  As shown in FIG. 4, the semiconductor wafer 10 on which the laser groove 22 is formed is half-cut (step cut) by the grinding blade SP1 along the scheduled division line 16. The grinding blade SP1 is aligned with the region of the laser groove 22, and the grinding blade SP1 grinds the substrate 12 from the surface of the substrate 12 to a predetermined depth deeper than the depth of the laser groove 22. A kerf 24 is formed on the substrate 12. The depth of the kerf 24 is generally about 40 to 50 μm or about 1/2 to 1/3 of the thickness of the substrate 12.

  As shown in FIG. 5, the semiconductor wafer 10 on which the kerf 24 is formed is fully cut by the grinding blade SP2 having a thickness smaller than the grinding blade SP1 along the kerf 24 of the division line 16. The grinding blade SP2 is aligned with the region of the kerf 24, and the grinding blade SP2 grinds the substrate 12 until it reaches the dicing tape 18 on the back surface. A kerf 29 is further formed on the substrate 12.

  Next, detection of the kerf edge will be described. A semiconductor having a laser groove 22 formed by irradiating laser light along the division line 16 and a kerf 24 deeper than the laser groove 22 formed by grinding along the laser groove 22 with the grinding blade SP1. The kerf edge is detected for the wafer 10.

  The inventor has intensively studied the detection of the edge of the kerf 24 formed in the laser groove 22. As a result, even if the amount of light is adjusted depending on the state of the bottom surface of the kerf 24, the region of the kerf 24 does not become a uniform white color, and the contrast at the boundary between the region of the kerf 24 and the region of the laser groove 22 is unclear. I found out that it might be.

  A general objective lens used for kerf check and arranged in an imaging region will be described. FIG. 6 shows the relationship between the depth of field of the objective lens 130 and the depths of the laser groove 22 and the kerf 24. In order to clearly observe the laser groove 22 and the kerf 24, the objective lens 130 has a deep depth of field DOF1. That is, the depth of field DOF1 of the objective lens 130 is set in a range including the laser groove 22 and the bottom surface of the kerf 24.

  Light is applied to the imaging region from the illumination device via the objective lens 130, and the reflected light is observed by the imaging device. The surface of the laser groove 22 is rough. Therefore, the irradiated light is irregularly reflected, so that the region of the laser groove 22 is observed as black. On the other hand, the surface (bottom surface) of the region of the kerf 24 is a smooth surface 26. For this reason, the irradiated light is uniformly reflected, and the area of the kerf 24 is observed as white.

  However, in the region of the kerf 24 formed by grinding of the grinding blade, a non-smooth surface, that is, a so-called saw mark 28 may be formed on the bottom surface of the kerf 24 depending on the use situation of the grinding blade. The saw mark 28 is a cutting mark formed on the bottom surface of the kerf 24 along the longitudinal direction of the kerf 24, and is a horizontally long protrusion or groove. At the saw mark 28, the irradiated light is irregularly reflected.

  When the objective lens 130 described above is used, light is emitted from the illumination device to the imaging region, and a part of the planned division line 16 is imaged by the imaging means, the result shown in FIG. 7 is obtained. Even when focusing on the edge of the kerf 24, the objective lens 130 has the deep depth of field DOF1, so the saw mark 28 on the bottom surface of the kerf 24 is also within the range of the depth of field DOF1. As a result, as shown in FIG. 7, the kerf 24 including the laser groove 22, the smooth surface 26, and the saw mark 28 is clearly observed.

  In FIG. 7, the laser groove 22 and the saw mark 28 are illustrated in different colors in order to explain. However, since the region of the laser groove 22 and the saw mark 28 in the kerf 24 become black of the same color, it is difficult to specify the image by processing the boundary between the laser groove 22 and the kerf 24.

  Therefore, the inventor has found that the above-described problem can be solved by using the objective lens 30 having the depth of field DOF2 shallower than the depth of the kerf 24. FIG. 8 shows the relationship between the depth of field of the objective lens 30 and the depths of the laser groove 22 and the kerf 24. The depth of field DOF2 of the objective lens 30 is set so as to be a range in which the laser groove 22 is included and the bottom surface of the kerf 24 is not included.

  When the objective lens 30 described above is used, light is emitted from the illumination device to the imaging region, and a part of the planned division line 16 is imaged by the imaging means, the result shown in FIG. 9 is obtained. The saw mark 28 and the smooth surface 26 in the kerf 24 are outside the range of the depth of field DOF2 of the objective lens 30. Therefore, a focus shift occurs on the bottom surface of the kerf 24. That is, the bottom surface of the kerf 24 is not clearly imaged. Therefore, the area of the saw mark 28 that looks black and the area of the smooth surface 26 that looks white are averaged and visually recognized. As a result, the entire area of the kerf 24 is visible with a uniform light gray. Since the area of the black laser groove 22 is dark and the area of the gray kerf 24 is viewed brightly, the boundary between the area of the laser groove 22 and the area of the kerf 24 can be clearly viewed. Furthermore, by increasing the light quantity of the illumination device, the luminance value of the kerf 24 region itself is shifted to the white side as compared with the laser groove 22. Therefore, the boundary between the laser groove 22 and the kerf 24 can be made clearer.

  The objective lens 30 preferably has a depth of field DOF2 that is about 1/10 of the depth of cut into the semiconductor wafer 10, that is, about 1/10 of the depth of the kerf 24. Generally, about 1/10 of the depth of the kerf 24 is about 2 to 15 μm. The reason for setting the depth of field DOF2 in the above-described range is that when the depth of field DOF2 is too shallow, unevenness of several microns is present microscopically on the table surface on which the workpiece is mounted. Although these irregularities are subjected to surface correction, in the case of DOF2 less than the correction capability (generally about 2 μm), focus blurring occurs due to variations in correction capability, and the effect of the correction itself cannot be obtained, and the depth of field DOF2 This is because a predetermined contrast cannot be obtained if the depth is increased.

  Next, the edge detection apparatus according to the present embodiment will be described with reference to FIG. The edge detection apparatus 1 illuminates an imaging area, an objective lens 30 that is disposed opposite to the division line 16 of the semiconductor wafer 10 that is an imaging area, a CCD 40 that is an imaging means for imaging the imaging area via the objective lens 30, and the imaging area. And an illuminating device 50.

  The observation optical axis indicated by a two-dot difference line connecting the objective lens 30 and the CCD 40 is substantially perpendicular to the planned division line 16 that is the imaging region. That is, the division line 16 is observed from directly above.

  The light emitted from the illumination device 50 passes through the illumination lens 52 and is reflected by the mirror 54. Further, the light is reflected toward the objective lens 30 by the half mirror 56 disposed at an inclination of about 45 ° with respect to the observation optical axis. That is, the light from the illumination device 50 is irradiated to the objective lens 30 from the coaxial direction with respect to the observation optical axis connecting the objective lens 30 and the CCD 40.

  The light is irradiated onto the imaging region through the objective lens 30, and the light reflected by the imaging region passes through the objective lens 30 again, then passes through the half mirror 56, and is imaged by the CCD 40. The image picked up by the CCD 40 is output to the monitor 60. The operator can observe the state of the captured division line 16 from the image output to the monitor 60.

  Since the edge detection apparatus 1 of the present embodiment uses the objective lens 30 having a depth of field DOF 2 shallower than the depth of the kerf 24, the boundary between the region of the kerf 24 and the region of the laser groove 22 is observed. Can do.

  The edge detection apparatus 1 further includes a control unit 80, and the control unit 80 includes an image processing unit 82 and a light amount control unit 84. The image processing unit 82 performs kerf edge detection processing from the captured image data. The light quantity control unit 84 controls the intensity of light emitted from the illumination device 50. By controlling the light from the illumination device 50, the area of the laser groove 22 is adjusted to be black, and the area of the kerf 24 is adjusted to be light gray.

  In the edge detection apparatus 1 of the present embodiment, a diaphragm 32 is further installed between the half mirror 56 and the objective lens 30. The depth of field can be changed by changing the aperture 32. By opening the aperture 32, the depth of field can be reduced.

  Next, a dicing apparatus in which the edge detection apparatus 1 is incorporated will be described. FIG. 11 is a perspective view showing an external configuration of a dicing apparatus in which the edge detection apparatus 1 is incorporated. As shown in the figure, the dicing apparatus 110 of the present embodiment includes a supply / recovery unit 112 for supplying / recovering the semiconductor wafer 10 on which the laser groove 22 is formed, a processing unit 114 for processing the semiconductor wafer 10, A cleaning unit 116 that cleans the processed semiconductor wafer 10, a transport unit 118 that transports the semiconductor wafer 10, an operation panel 120 that performs various operations, the edge detection device 1, and a control unit that controls the overall operation (FIG. Not shown).

  The supply / recovery unit 112 for supplying / recovering the semiconductor wafer 10 includes a load port 122, and a cassette (not shown) in which a large number of semiconductor wafers 10 are stored is set in the load port 122. The semiconductor wafer 10 to be processed is stored in a cassette in a state of being mounted on a predetermined frame F via a dicing tape 18.

  The processing unit 114 that processes the semiconductor wafer 10 includes a work table 124 that holds the semiconductor wafer 10 by suction, a pair of grinding blades SP1 and SP2 that cut the semiconductor wafer 10 held on the work table 124, and a grinding blade SP1. The spindles 128A and 128B to which the SP2 is attached and the edge detection device 1 that detects the kerf edge of the surface of the semiconductor wafer 10 held on the work table 124.

  The work table 124 is provided on an X-axis table (not shown) and a θ-axis table (not shown) installed horizontally, and is driven by the θ-axis table to be centered (θ-axis, not shown). ) Rotate around. As the X-axis table slides on an X-axis guide (not shown), the work table 124 is moved horizontally in the X direction in the figure.

  The grinding blades SP1 and SP2 are constituted by a diamond blade or CBN abrasive electrode formed in a thin disc shape, an electrodeposition blade obtained by electrodeposition with nickel, a resin blade bonded by a resin, or the like. The grinding blade SP1 half-cuts the semiconductor wafer 10, and the grinding blade SP1 fully cuts the semiconductor wafer 10. The grinding blades SP1 and SP2 are attached to the tips of the spindles 128A and 128B so that the cutting direction thereof is parallel to the movement direction of the work table 124 (X direction in the figure). The spindles 128A and 128B are driven to rotate. A cutting nozzle (not shown) is provided in the vicinity of the grinding blades SP1 and SP2, and cutting water is supplied from the nozzle to the processing point.

  The spindles 128 </ b> A and 128 </ b> B are arranged opposite to each other above the work table 124 so that the rotation axis is orthogonal to the moving direction of the work table 124, and are rotated at a high speed of 30,000 rpm to 80,000 rpm.

  The spindles 128A and 128B are each attached to a blade Z-axis table (not shown) installed vertically. The blade Z-axis table slides on the blade Z-axis guide (not shown) in the Z direction (direction perpendicular to the XY plane), so that the grinding blades SP1 and SP2 move vertically in the Z direction. Then, it moves forward and backward vertically with respect to the work table 124.

  The cleaning unit 116 includes a spin cleaning device 116A, and the processed semiconductor wafer 10 is spin cleaned by the spin cleaning device 116A.

  The transfer unit 118 includes a handling robot 118A, and the handling robot 118A transfers the semiconductor wafer 10 between the units. That is, by the handling robot 118A, the semiconductor wafer 10 is taken out from the cassette of the supply / recovery unit 112 and transferred to the processing unit 114, and the semiconductor wafer 10 processed by the processing unit 114 is recovered from the processing unit 114. Transport to cleaning unit 116. Further, the semiconductor wafer 10 cleaned by the cleaning unit 116 is recovered from the cleaning unit 116, transported to the supply / recovery unit 112, and stored in the cassette of the supply / recovery unit 112.

  The edge detection device 1 detects the position of the kerf 24 formed by the grinding blade SP1 in the region of the laser groove 22. A deviation between the planned processing position and the actual position of the kerf 24 is obtained from this information data, and the semiconductor wafer 10 is half-cut by the grinding blade SP1 while correcting the positional deviation.

  An edge detection method using the edge detection apparatus 1 will be described with reference to FIG. In this embodiment, image processing means is applied. First, the semiconductor wafer 10 on which the laser groove 22 is formed is supplied to the dicing apparatus 110, and the semiconductor wafer 10 is half-cut by the grinding blade SP1. A kerf 24 is formed on the substrate 12 (S1).

  Next, the edge detection device 1 and the imaging region of the semiconductor wafer 10 are aligned. That is, the edge detection apparatus 1 is positioned in an imaging region including the kerf 24, the laser groove 22, and the division planned line 16 of the semiconductor wafer 10. In particular, the objective lens 30 is preferably arranged so that the central axis of the objective lens 30 is positioned at the center of the kerf 24 in the width direction. The reason is to eliminate information similar to the kerf 24 inside the semiconductor wafer 10 as much as possible (S2).

  Next, the light amount of the illumination device 50 is adjusted by the light amount control unit 84 so that the region of the laser groove 22 is visually recognized as black (dark) and the region of the kerf 24 is viewed as bright gray (bright) (S3).

  Next, an imaging area including the kerf 24, the laser groove 22 and the planned division line 16 is imaged by the CCD 40 (S4).

  Next, the image processing unit 82 detects the area of the planned division line 16, the area of the laser groove 22, and the area of the kerf 24 from the captured image data.

<Detection of kerf>
The kerf 24 is detected by binarization processing or edge detection. When the color of the kerf 24 is visually recognized in light gray (white), it is processed as a white object detection algorithm by data setting. Further, since the width of the kerf 24 can be estimated with reference to the width of the grinding blade SP1, the region of the kerf 24 can be roughly specified. The distribution tendency (for example, the average value) of the luminance of the specified kerf 24 region is white, and automatic detection of the color to be detected using this characteristic is also possible.

(Binarization processing)
When white object detection is performed by binarization processing, use a data setting or a known automatic threshold setting algorithm (eg, Otsu's automatic threshold determination method) to extract and extract white objects. The resulting contour can be kerf 24.

(Edge detection)
When performing edge detection, a change point that changes from white to black toward the outside of the kerf 24 region is detected with reference to the center of the whitest region in the inspection window. The edge of the kerf 24 (boundary with the laser groove 22) is calculated in sub-pixel coordinates by adopting a second-order differential zero cross obtained by further differentiating the edge intensity that is the changing point.

  Further, the processing speed can be improved by calculating a rough contour position by binarization processing as preprocessing of the edge detection processing and performing edge detection in the vicinity of the contour position.

<Laser groove detection>
The laser groove 22 is detected by binarization processing or edge detection. When the color of the laser groove 22 is visually recognized as black, it is processed as a black object detection algorithm by data setting. Further, since the width of the laser groove 22 can be estimated from the data setting and the width of the kerf 24 region can be estimated with reference to the width of the grinding blade SP1, the luminance distribution tendency of the region excluding the width of the kerf 24 from the width of the laser groove 22 For example, the average value) is black, and automatic determination using this characteristic is also possible.

(Binarization processing)
When black object detection is performed by binarization, data setting or an existing automatic threshold setting algorithm (eg, Otsu's automatic threshold determination method) is used to threshold the black object from the image. Extract by value processing. The contour coordinates of the extraction result can be grasped by accessing the pixel, and this contour can be recognized as the laser groove 22. Here, since the laser groove 22 may be divided by the white kerf 24, it is excluded at the time of detection. Extraction is performed separately in the upper and lower parts of the detection window, or the black area in the detection window is extracted by painting the center part black or by not accessing the pixels with the kerf 24 area outside the detection target range.

(Edge detection)
When performing edge detection, from the center of the objective lens 30, from the outside of the approximate kerf 24 region, or from the outside when there is already a detection result of the kerf 24, the laser groove 22 is directed to the outside. Change point that changes from white to white is detected. The edge of the kerf 24 region is calculated in sub-pixel coordinates by adopting a zero-order of the second derivative of luminance obtained by further differentiating the edge intensity at the changing point.

  Alternatively, the processing speed can be improved by calculating an approximate contour position by binarization processing as preprocessing of the edge detection processing and performing edge detection in the vicinity of the contour position.

  Since the color of the planned division line 16 is visually recognized as white, it is possible to detect the planned division line 16 by the same detection method as that for the area of the kerf 24 (S5).

  Finally, the amount of deviation between the edge of the kerf 24 and the planned half-cut processing position is obtained from the edge data of the kerf 24 detected at S5 and the data of the planned half-cut processing position (S6). Based on the amount of deviation, position correction is applied to the grinding blade SP1, and the semiconductor wafer 10 is half-cut by the position-corrected grinding blade SP1.

  Another edge detection method using the edge detection apparatus 1 will be described with reference to FIG. First, the semiconductor wafer 10 on which the laser groove 22 is formed is supplied to the dicing apparatus 110, and the semiconductor wafer 10 is half-cut by the grinding blade SP1. A kerf 24 is formed on the substrate 12 (S11).

  Next, the edge detection device 1 and the imaging region of the semiconductor wafer 10 are aligned. That is, the edge detection apparatus 1 is positioned in an imaging region including the kerf 24, the laser groove 22, and the division planned line 16 of the semiconductor wafer 10. In particular, the objective lens 30 is preferably arranged so that the central axis of the objective lens 30 is positioned at the center of the kerf 24 in the width direction. (S12).

  Next, the light quantity control unit 84 adjusts the light quantity of the illumination device 50 so that the area of the laser groove 22 is visually recognized as dark (dark) and the area of the kerf 24 is bright gray (bright) (S13).

  Next, an imaging area including the kerf 24, the laser groove 22, and the division line 16 is imaged by the CCD 40 (S14).

  Next, the image processing unit 82 detects the area of the planned division line 16, the area of the laser groove 22, and the area of the kerf 24 from the captured image data. Detection of the area of the planned division line 16, the area of the laser groove 22, and the area of the kerf 24 is performed by the same process as S5 in FIG. 12 (S15).

  In S15, since the laser groove 22 and the kerf 24 can be visually recognized at the same time, the illumination condition can be individually changed in the case where the detection is unstable.

  Next, the light amount of the illumination device 50 is adjusted by the light amount control unit 84 so that the region of the laser groove 22 including the region of the kerf 24 is black (dark) and the scheduled division line 16 is viewed brightly (S16).

  Next, the imaging region including the laser groove 22 and the division planned line 16 is imaged by the CCD 40 (S17).

  In S <b> 18, the area of the planned division line 16 and the area of the laser groove 22 are detected by the image processing unit 82 from the captured image data. The laser groove 22 is detected by binarization processing or edge detection. When the color of the laser groove 22 is visually recognized as black, it is processed as a black object detection algorithm by data setting. Further, from the data setting of the width of the laser groove 22, the luminance distribution tendency (for example, average value) of the region of the laser groove 22 becomes black, and automatic determination using this characteristic is also possible.

  In addition, since the light quantity of the illuminating device 50 is adjusted, the area | region of the white kerf 24 does not exist in the area | region of the laser groove 22. FIG. Therefore, it is not necessary to consider the area of the kerf 24. The method for detecting the laser groove 22 in S5 of FIG. 12 can be applied (S18).

  Finally, the edge data of the kerf 24 and the edge of the laser groove 22 detected in S15, or the edge data of the kerf 24 detected in S15 and the edge data of the laser groove 22 detected in S18 are combined. .

  Thus, the user can grasp the position of the kerf 24 in the laser groove 22 by drawing the apparatus result, and the apparatus can obtain the amount of deviation of the machining position by grasping the laser groove position and the machining position. .

  Specifically, the contour information of the kerf 24 detected in S15 is stored as coordinate data or image information in a memory or a storage medium. Further, the contour information of the laser groove 22 detected in S18 is stored as coordinate data or image information in a memory or a storage medium. The coordinate data of the kerf 24 is drawn in the image memory, or the pixel indicating the contour color of the kerf 24 is drawn on the memory from the image information, and the coordinate data of the laser groove 22 is drawn in the image memory, or the laser groove 22 is drawn from the image information. A pixel indicating the contour color is drawn on the memory. The drawing on the memory generated by the detection of the laser groove 22 or the kerf 24 is overlaid on the image when the laser groove 22 is detected. Alternatively, an overlay display is performed on the image when the kerf 24 is detected. The optical condition of the image used for display can be changed by data setting.

  By obtaining the edge of the laser groove 22 and the edge of the kerf 24 from the composite image obtained by combining the edge data of the kerf 24 detected in S15 and the edge data of the laser groove 22 detected in S18, A more accurate edge of the kerf 24 can be detected. Then, the amount of deviation between the edge of the kerf 24 and the planned half-cut machining position is obtained from the edge data of the kerf 24 and the data of the planned half-cut machining position. Based on the amount of deviation, position correction is applied to the grinding blade SP1, and the semiconductor wafer 10 is half cut by the position corrected grinding blade SP1 (S19).

  Another edge detection method using the edge detection apparatus 1 will be described with reference to FIG. First, the semiconductor wafer 10 on which the laser groove 22 is formed is supplied to the dicing apparatus 110, and the edge detection apparatus 1 and the imaging region of the semiconductor wafer 10 are aligned. That is, the edge detection apparatus 1 is positioned in an imaging region including the laser groove 22 and the division planned line 16 of the semiconductor wafer 10. In particular, the objective lens 30 is preferably arranged so that the central axis of the objective lens 30 is positioned at the center in the width direction of the laser groove 22 (S21).

  Next, the light quantity of the illumination device 50 is adjusted by the light quantity control unit 84 so that the area of the laser groove 22 is black and the scheduled division line 16 is viewed brighter than the laser groove (S22).

  Next, the CCD 40 captures an imaging area including the laser groove 22 and the division planned line 16. At this time, the kerf 24 is not formed (S23).

  Next, the image processing unit 82 detects the region of the planned division line 16 and the region of the laser groove 22 from the captured image data.

  The laser groove 22 is detected by binarization processing or edge detection. When the color of the laser groove 22 is visually recognized as black, it is processed as a black object detection algorithm by data setting. Further, from the data setting of the width of the laser groove 22, the luminance distribution tendency (for example, average value) of the region of the laser groove 22 becomes black, and automatic determination using this characteristic is also possible. The detection method of the laser groove 22 in S15 of FIG. 13 can be applied (S24).

  Next, a processing target position is calculated from the laser groove edge coordinates detected by the laser groove 22, and a processing position correction amount is calculated. Based on this result, the formed semiconductor wafer 10 is half-cut by the grinding blade SP1. . A kerf 24 is formed on the substrate 12 (S25).

  Next, the semiconductor wafer 10 on which the laser groove 22 is formed is half-cut by the grinding blade SP1. A kerf 24 is formed on the substrate 12 (S25).

  Next, the edge detection device 1 and the imaging region of the semiconductor wafer 10 are aligned. That is, the edge detection apparatus 1 is positioned in an imaging region including the kerf 24, the laser groove 22, and the division planned line 16 of the semiconductor wafer 10. In particular, the objective lens 30 is preferably arranged so that the central axis of the objective lens 30 is positioned at the center of the kerf 24 in the width direction.

  Next, the light amount of the illumination device 50 is adjusted by the light amount control unit 84 so that the region of the laser groove 22 is black and the region of the kerf 24 is visually recognized as a lighter gray than the laser groove 22 (S27).

  Next, an imaging area including the kerf 24, the laser groove 22, and the division line 16 is imaged by the CCD 40 (S28).

  Next, the image processing unit 82 detects the area of the planned division line 16, the area of the laser groove 22, and the area of the kerf 24 from the captured image data. Detection of the area of the planned division line 16, the area of the laser groove 22, and the area of the kerf 24 is performed in the same process as S5 in FIG. 12 (S29).

  Finally, the edge data of the kerf 24 detected in S24 and the edge data of the laser groove 22 detected in S29 are combined. Alternatively, it is possible to detect a positional deviation after processing by further superimposing the processing of S18 which is a detection processing after processing.

  Accordingly, the user can grasp the position of the kerf in the laser groove by drawing the apparatus result, and the apparatus can obtain the amount of deviation of the machining position by grasping the laser groove position and the machining position.

  This amount of deviation is performed by the same process as S19 in FIG. Based on the amount of deviation, position correction is applied to the grinding blade SP1, and the semiconductor wafer 10 is half-cut by the position-corrected grinding blade SP1 (S30).

  Next, a method for detecting the edge of the kerf 29 formed by the grinding blade SP2 will be described. As shown in FIG. 5, the kerf 29 is formed in the kerf 24. For this reason, it may be difficult to detect the boundary between the kerf 24 and the kerf 29.

  Therefore, as shown in FIG. 15, the semiconductor wafer 10 in which the laser groove 22 is formed and the kerf 24 is not formed is half-cut by the grinding blade SP2 to form an edge detecting kerf 29A. The amount of deviation between the data of the edge of the kerf 29A and the data of the planned cutting position for full cut is obtained, and position correction is applied to the grinding blade SP2. After the kerf 24 is formed by the grinding blade SP1, the semiconductor wafer 10 is fully cut by the grinding blade SP2.

  The detection of the edge of the kerf 29A can be performed based on the processing of the flowcharts of FIGS. In FIG. 12 to FIG. 14, the edge of the kerf 29 </ b> A can be detected by replacing the “half-cutting with the grinding blade SP <b> 1” process with the “half-cutting with the grinding blade SP <b> 2” process.

  However, it is necessary to detect the edge of the kerf 29A before the kerf 24 is formed by the grinding blade SP1.

  The case has been described where the image processing means is used to detect the recognition of the light and darkness of the kerf, the groove, and the division planned line. However, the present invention is not limited to this, and detection can be performed using an image recognition means using a CCD. An image recognition means using a three-dimensional sensor, an area sensor, a line sensor can be used by using an image recognition means using a CCD.

  In the present embodiment, it is generally preferable that the illumination lens 52 has a depth of field of about 15 to 30 μm. As shown in FIG. 16, the illumination lens 52 having a deep depth of field is used, and the mounting position of the illumination device 50 is adjusted, so that light is captured from a wide range of the illumination device 50. As a result, the illumination range and the illumination angle to the inside of the kerf 24 are expanded by the objective lens 30 having a shallow depth of field and the illumination lens 52 having a deep depth of field. As a result, a large amount of light diffusely reflected by the saw mark 28 inside the kerf 24 can be secured, and shading due to insufficient light quantity at the edge portion of the kerf 24 can be suppressed.

  The objective lens 30 of the present embodiment has a large radius of curvature on the surface side facing the CCD 40. Internal reflection can be suppressed as much as possible by increasing the radius of curvature.

  DESCRIPTION OF SYMBOLS 1 ... Edge detection apparatus, 10 ... Semiconductor wafer, 12 ... Board | substrate, 14 ... Element area | region, 16 ... Dividing line, 18 ... Dicing tape, 20 ... Low dielectric constant insulating film, 22 ... Laser groove, 24 ... Calf, 26 ... Smoothing unit, 28 ... saw mark, 30 ... objective lens, 40 ... CCD, 50 ... illumination device, 52 ... illumination lens, 60 ... monitor, 82 ... image processing unit, 84 ... light quantity control unit, 110 ... dicing device

Claims (4)

For a semiconductor wafer having a laser groove formed by irradiating a laser beam along a division line and a kerf deeper than the laser groove formed by grinding with a grinding blade along the laser groove, Imaging means for imaging the division planned line, the laser groove, and the kerf from above;
An objective lens disposed between the planned division line and the imaging means, and a depth of field shallower than a depth of the kerf;
An illumination device that irradiates the objective lens with light from a coaxial direction with respect to an observation optical axis connecting the objective lens and the imaging means;
An edge detection device comprising:   The edge detection device according to claim 1, wherein the objective lens is disposed such that a central axis of the objective lens is positioned at a center in a width direction of the kerf. A light amount control unit that adjusts the illumination device so that the area of the kerf is bright, the area of the laser groove is dark, and the area of the planned dividing line is bright;
The edge detection apparatus according to claim 1, further comprising: an image processing unit configured to detect the kerf region, the laser groove region, and the division-scheduled line region from the image obtained by the imaging unit. . The light amount control unit adjusts the illumination device so that the kerf region and the laser groove region are dark and the region of the planned division line is bright,
The edge detection apparatus according to claim 3, wherein the image processing unit detects the region of the laser groove and the region of the scheduled division line from an image obtained by the imaging unit.

Images