Wafer inspection method

Abstract

To provide a wafer inspection method capable of simplifying determination of the processing state of a wafer.SOLUTION: A wafer inspection method for inspecting a wafer in which a modified layer is formed along a dividing scheduled line includes a laser beam irradiation step of positioning, on or inside the surface of the wafer, a focal point of a laser beam whose power output does not exceed a processing threshold value of the wafer and which is transparent to the wafer, and applying a laser beam from the back surface side of the wafer so that the shape of an area to be irradiated with the laser beam on the back surface of the wafer is asymmetric with respect to the modified layer, an imaging step of capturing an image of reflected light of the laser beam by imaging the reflected light of the laser beam, and a determination step of determining the processing state of the wafer based on the image. In the determination step, the processing state of the wafer is determined with a learned model configured by machine learning so that the processing state of the wafer is output when the image is input.SELECTED DRAWING: Figure 9

Description

[Technical field]

[0001] The present invention relates to a wafer inspection method for inspecting a wafer having a modified layer formed therein. [Background technology]

[0002] In the device chip manufacturing process, a wafer is used in which devices are formed in multiple areas defined by multiple planned division lines (streets) arranged in a grid pattern. By dividing this wafer along the planned division lines, multiple device chips, each equipped with a device, are obtained. The device chips are incorporated into various electronic devices such as mobile phones and personal computers.

[0003] A cutting device that cuts the wafer with an annular cutting blade is used to divide the wafer. Meanwhile, in recent years, a process for dividing the wafer by laser processing has also been developed. For example, Patent Document 1 discloses a method in which a laser beam that is transparent to the wafer is focused inside the wafer to modify (alter) the inside of the wafer. When this method is used, a modified layer (altered layer) is formed inside the wafer along the planned division line, and a crack (crack) extends from the modified layer toward the front surface side of the wafer.

[0004] The modified layer or the cracked area of ​​the wafer becomes more fragile than other areas. Therefore, when an external force is applied to the wafer, the wafer breaks along the modified layer or the cracked area and is divided along the intended division line. In other words, the modified layer and the crack function as the division starting point (the trigger for division).

[0005] If the cracks do not extend properly from the modified layer toward the surface of the wafer when the modified layer is formed, the wafer may not be divided as intended even if an external force is applied to the wafer thereafter, resulting in processing defects. Therefore, after the modified layer is formed, an inspection may be performed to check whether the cracks are properly formed inside the wafer.

[0006] For example, Patent Document 2 discloses a method of acquiring an image of the reflected light of the laser beam (reflected light image) by irradiating a laser beam for observation from the back side of a wafer on which a modified layer is formed, and capturing the reflected light of the laser beam with an imaging unit. The laser beam incident on the wafer is affected by cracks formed inside the wafer. Therefore, by observing the pattern of reflected light appearing in the reflected light image, it is possible to determine whether or not cracks are properly formed inside the wafer. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] JP 2005-86161 A [Patent Document 2] JP 2020-68316 A Summary of the Invention [Problem to be solved by the invention]

[0008] As described above, the processed state of a wafer can be determined based on an image of reflected light (reflected light image) of a laser beam for observation irradiated onto the wafer. However, the image of reflected light appearing in the reflected light image varies in shape and shade due to various factors such as the irradiation conditions of the laser beam, the imaging conditions, and the state of the wafer. Therefore, in order to properly determine the processed state of a wafer, it is necessary to align the conditions for imaging the reflected light of the laser beam and to change the settings of the image processing for determining the processed state of the wafer according to the reflected light image, which makes the inspection of the wafer complicated.

[0009] The present invention has been made in view of the above problems, and has an object to provide a wafer inspection method capable of simplifying the determination of the processed state of the wafer. [Means for solving the problem]

[0010] According to one aspect of the present invention, there is provided a wafer inspection method for inspecting a wafer having a modified layer formed therein along a planned division line, the method including: a laser beam irradiation step in which a focal point of a laser beam having an output that does not exceed a processing threshold value of the wafer and that is transparent to the wafer is positioned on the front surface or inside of the wafer, and the laser beam is irradiated from the back surface side of the wafer such that a shape of a region on the back surface of the wafer that is irradiated with the laser beam is asymmetric with respect to the modified layer; an imaging step in which an image of the reflected light of the laser beam is obtained by imaging the reflected light; and a determination step in which a processed state of the wafer is determined based on the image, wherein in the determination step, the processed state of the wafer is determined using a trained model configured by machine learning so as to output the processed state of the wafer when the image is input. The trained model classifies the processing state of the wafer into a processing state in which an appropriate crack extending from the modified layer is formed on the wafer, a processing state in which no crack extending from the modified layer is formed on the wafer, and a processing state in which an inappropriate crack extending from the modified layer is formed on the wafer. A method for inspecting a wafer is provided.

[0011] Preferably, in the imaging step, the reflected light is imaged multiple times while the wafer and the laser beam are moved relatively along a direction parallel to the planned division lines. Also, preferably, in the imaging step, the reflected light is imaged multiple times while the wafer and the laser beam are moved relatively along a direction perpendicular to the planned division lines. Also, preferably, in the imaging step, the reflected light is imaged multiple times while the focal point of the laser beam is moved relatively along the thickness direction of the wafer.

[0012] Also, preferably, the trained model is a neural network including an input layer and an output layer, ApplicableWhen the image is input to the input layer, the neural network outputs the processing state of the wafer from the output layer. Preferably, the neural network is trained by supervised learning using a plurality of learning images including the reflected light image and classified according to the processing state of the wafer, and the learning images are classified into a first reflected light image corresponding to a case where a crack normally extends from the modified layer toward the front surface side of the wafer, a second reflected light image corresponding to a case where a crack does not extend from the modified layer toward the front surface side of the wafer, or a third reflected light image other than the first reflected light image and the second reflected light image. Preferably, the wafer inspection method further includes a visualization step of visualizing the features of the image extracted by the neural network. Also, preferably, the neural network is a convolutional neural network, and the visualization step includes obtaining a heat map representing the activation state of a convolutional layer when the image is input to the convolutional neural network. Effect of the Invention

[0013] In a wafer inspection method according to one aspect of the present invention, the processed state of the wafer is determined using a trained model configured by machine learning to output the processed state of the wafer when an image of reflected light of a laser beam is input. This makes it possible to use images of reflected light of a laser beam captured under various conditions to determine the processing conditions of the wafer, simplifying wafer inspection. [Brief description of the drawings]

[0014] [Figure 1] FIG. 2 is a perspective view showing a laser processing device. [Diagram 2] FIG. [Diagram 3] 1 is a partial cross-sectional front view showing a laser processing apparatus for forming a modified layer on a wafer. [Figure 4] FIG. 4(A) is an enlarged cross-sectional view showing a portion of a wafer in which a crack has been formed, FIG. 4(B) is an enlarged cross-sectional view showing a portion of a wafer in which a crack has not been formed, and FIG. 4(C) is an enlarged cross-sectional view showing a portion of a wafer in which a serpentine crack has been formed. [Diagram 5]1 is a partial cross-sectional front view showing a laser processing apparatus for inspecting a wafer. [Figure 6] FIG. 2 is a plan view showing a portion of a wafer irradiated with a laser beam. [Figure 7] FIG. 7(A) is a cross-sectional view showing how a laser beam is irradiated onto a wafer in which a crack has been formed, and FIG. 7(B) is a cross-sectional view showing how a laser beam is irradiated onto a wafer in which a crack has not been formed. [Figure 8] FIG. 8(A) is an image diagram showing a first reflected light image, FIG. 8(B) is an image diagram showing a second reflected light image, and FIG. 8(C) is an image diagram showing a third reflected light image. [Figure 9] FIG. 2 is a block diagram showing a control unit. [Figure 10] Figure 10(A) is an image diagram showing a learning image classified as a first reflected light image, Figure 10(B) is an image diagram showing a learning image classified as a second reflected light image, and Figure 10(C) is an image diagram showing a learning image classified as a third reflected light image. [Figure 11] FIG. 11(A) is a plan view showing the wafer and laser beam moving relatively along the X-axis direction, and FIG. 11(B) is a plan view showing the wafer and laser beam moving relatively along the Y-axis direction. [Figure 12] FIG. 12(A) is a cross-sectional view showing how the focal point of a laser beam is positioned in an area where a crack has been formed, and FIG. 12(B) is a cross-sectional view showing how the focal point of a laser beam is positioned in an area where a crack has not been formed. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, an embodiment according to one aspect of the present invention will be described with reference to the accompanying drawings. First, a configuration example of a laser processing device that can be used to implement a wafer inspection method according to this embodiment will be described. FIG. 1 is a perspective view showing a laser processing device 2. In FIG. 1, the X-axis direction (processing feed direction, first horizontal direction) and the Y-axis direction (indexing feed direction, second horizontal direction) are perpendicular to each other. Also, the Z-axis direction (height direction, vertical direction, up-down direction) is perpendicular to the X-axis direction and the Y-axis direction.

[0016] The laser processing apparatus 2 includes a base 4 that supports each of the components that make up the laser processing apparatus 2. The upper surface of the base 4 is a flat surface that is roughly parallel to the horizontal direction (XY plane direction), and a moving unit (moving mechanism) 6 is provided on the upper surface of the base 4. The moving unit 6 includes a Y-axis moving unit (Y-axis moving mechanism, indexing feed unit) 8, an X-axis moving unit (X-axis moving mechanism, processing feed unit) 18, and a Z-axis moving unit (Z-axis moving mechanism) 30.

[0017] The Y-axis movement unit 8 includes a pair of Y-axis guide rails 10 arranged along the Y-axis direction on the upper surface of the base 4. A flat Y-axis movement table 12 is mounted on the pair of Y-axis guide rails 10 so as to be slidable along the Y-axis guide rails 10.

[0018] A nut portion (not shown) is provided on the back surface (lower surface) side of the Y-axis moving table 12. A Y-axis ball screw 14 arranged along the Y-axis direction between a pair of Y-axis guide rails 10 is screwed into this nut portion. In addition, a Y-axis pulse motor 16 that rotates the Y-axis ball screw 14 is connected to an end of the Y-axis ball screw 14. When the Y-axis pulse motor 16 rotates the Y-axis ball screw 14, the Y-axis moving table 12 moves in the Y-axis direction along the Y-axis guide rails 10.

[0019] The X-axis movement unit 18 includes a pair of X-axis guide rails 20 arranged along the X-axis direction on the surface (upper surface) of the Y-axis movement table 12. A plate-shaped X-axis movement table 22 is mounted on the pair of X-axis guide rails 20 so as to be slidable along the X-axis guide rails 20.

[0020] A nut portion (not shown) is provided on the back surface (lower surface) side of the X-axis moving table 22. An X-axis ball screw 24 arranged along the X-axis direction between a pair of X-axis guide rails 20 is screwed into this nut portion. An X-axis pulse motor 26 that rotates the X-axis ball screw 24 is connected to an end of the X-axis ball screw 24. When the X-axis pulse motor 26 rotates the X-axis ball screw 24, the X-axis moving table 22 moves in the X-axis direction along the X-axis guide rails 20.

[0021] A chuck table (holding table) 28 for holding a wafer 11 (see FIG. 2) to be processed by the laser processing device 2 is provided on the surface (upper surface) of the X-axis moving table 22. The upper surface of the chuck table 28 is a flat surface that is approximately parallel to the horizontal direction (XY plane direction) and constitutes a holding surface 28a for holding the wafer 11. The holding surface 28a is connected to a suction source (not shown) such as an ejector via a flow path (not shown), a valve (not shown), and the like formed inside the chuck table 28.

[0022] When the Y-axis moving table 12 is moved along the Y-axis direction, the chuck table 28 moves along the Y-axis direction. Also, when the X-axis moving table 22 is moved along the X-axis direction, the chuck table 28 moves along the X-axis direction. That is, the Y-axis moving unit 8 and the X-axis moving unit 18 control the movements of the chuck table 28 in the X-axis direction and the Y-axis direction. Also, the chuck table 28 is connected to a rotation drive source (not shown) such as a motor that rotates the chuck table 28 around a rotation axis roughly parallel to the Z-axis direction.

[0023] A Z-axis moving unit 30 is provided at the rear end of the base 4 (rear of the Y-axis moving unit 8, the X-axis moving unit 18, and the chuck table 28). The Z-axis moving unit 30 has a support structure 32 arranged on the upper surface of the base 4. The support structure 32 includes a rectangular parallelepiped base 32a fixed to the base 4, and a columnar support 32b protruding upward from the end of the base 32a. The surface (side surface) of the support 32b is formed flat along the Z-axis direction.

[0024] A pair of Z-axis guide rails 34 are provided along the Z-axis direction on the surface of the support portion 32b. A flat Z-axis moving table 36 is mounted on the pair of Z-axis guide rails 34 so as to be slidable along the Z-axis guide rails 34.

[0025] A nut portion (not shown) is provided on the back side of the Z-axis moving table 36. A Z-axis ball screw (not shown) disposed along the Z-axis direction between a pair of Z-axis guide rails 34 is screwed into this nut portion. A Z-axis pulse motor 38 that rotates the Z-axis ball screw is connected to an end of the Z-axis ball screw. A support member 40 is fixed to the front side of the Z-axis moving table 36. When the Z-axis pulse motor 38 rotates the Z-axis ball screw, the Z-axis moving table 36 and the support member 40 move in the Z-axis direction along the Z-axis guide rails 34.

[0026] The laser processing device 2 is also equipped with a laser irradiation unit 42 that irradiates a laser beam onto the wafer 11 (see FIG. 2) held by the chuck table 28. At least some of the components of the laser irradiation unit 42 (such as a laser processing head 44) ​​are supported by a support member 40.

[0027] The laser irradiation unit 42 includes a processing laser irradiation unit 46A (see FIG. 3) that irradiates a laser beam (processing laser beam) for processing the wafer 11, and an observation laser irradiation unit 46B (see FIG. 5) that irradiates a laser beam (observation laser beam) for observing the inside of the wafer 11. The configurations, functions, uses, etc. of the processing laser irradiation unit 46A and the observation laser irradiation unit 46B will be described later.

[0028] An imaging unit (camera) 48 is provided at the tip of the laser irradiation unit 42. The imaging unit 48 includes an image sensor such as a CCD (Charged-Coupled Devices) sensor or a CMOS (Complementary Metal-Oxide-Semiconductor) sensor, and captures an image of the wafer 11 (see FIG. 2) held by the chuck table 28. For example, based on the image of the wafer 11 acquired by the imaging unit 48, the chuck table 28 and the laser processing head 44 are aligned.

[0029] When the Z-axis moving table 36 is moved along the Z-axis direction, the laser processing head 44 and the imaging unit 48 move (rise and fall) along the Z-axis direction. This allows adjustment of the height of the focal point of the laser beam emitted from the laser irradiation unit 42, focusing of the imaging unit 48, etc.

[0030] The laser processing apparatus 2 also includes a display unit (display section, display device) 50 that displays various information related to the laser processing apparatus 2. For example, a touch panel is used as the display unit 50. In this case, the operator can input information such as processing conditions to the laser processing apparatus 2 by touching the display unit 50. In other words, the display unit 50 also functions as an input unit (input section, input device) for inputting various information to the laser processing apparatus 2, and is used as a user interface. However, the input unit may be an operation panel, a mouse, a keyboard, etc. that are provided separately and independently from the display unit 50.

[0031] Furthermore, the laser processing apparatus 2 includes a control unit (control unit, control device) 52. The control unit 52 is connected to each component (movement unit 6, chuck table 28, laser irradiation unit 42, imaging unit 48, display unit 50, etc.) constituting the laser processing apparatus 2. The control unit 52 controls the operation of the laser processing apparatus 2 by outputting control signals to the components of the laser processing apparatus 2.

[0032] For example, the control unit 52 is configured by a computer. Specifically, the control unit 52 includes a processor such as a CPU (Central Processing Unit) that performs calculations necessary for the operation of the laser processing device 2, and a memory such as a ROM (Read Only Memory) or a RAM (Random Access Memory) that stores various information (data, programs, etc.) used for the operation of the laser processing device 2.

[0033] Laser processing is performed on a wafer 11 by the laser processing device 2. Fig. 2 is a perspective view showing the wafer 11. For example, the wafer 11 is a disk-shaped wafer made of a semiconductor material such as silicon, and has a front surface 11a and a back surface 11b that are generally parallel to each other. The wafer 11 is partitioned into a plurality of rectangular regions by a plurality of planned division lines (streets) 13 that are arranged in a lattice pattern so as to intersect with each other.

[0034] Devices 15, such as ICs (Integrated Circuits), LSIs (Large Scale Integrations), LEDs (Light Emitting Diodes), MEMS (Micro Electro Mechanical Systems) devices, etc., are formed on the surface 11a of the multiple regions partitioned by the division lines 13. By dividing the wafer 11 along the division lines 13, multiple device chips, each including a device 15, are obtained.

[0035] There are no limitations on the type, material, shape, structure, size, etc. of wafer 11. For example, wafer 11 may be a wafer of any shape and size made of a semiconductor other than silicon (GaAs, InP, GaN, SiC, etc.), sapphire, glass, ceramics, resin, metal, etc. Furthermore, there are no limitations on the type, number, shape, structure, size, arrangement, etc. of devices 15, and devices 15 do not necessarily have to be formed on wafer 11.

[0036] A dividing starting point that functions as a dividing trigger when dividing the wafer 11 is formed in the wafer 11. For example, a modified layer that functions as a dividing starting point is formed inside the wafer 11 by irradiating a laser beam along the dividing lines 13. The wafer 11 is then divided along the dividing lines 13 using the modified layer as a dividing starting point to manufacture device chips. A specific example of a wafer processing method (device chip manufacturing method) for dividing the wafer 11 into a plurality of device chips will be described below.

[0037] 3 is a partial cross-sectional front view showing the laser processing apparatus 2 that forms a modified layer (degraded layer) 17 on the wafer 11. When processing the wafer 11 with the laser processing apparatus 2, first, the wafer 11 is held by the chuck table 28 (holding step).

[0038] For example, the wafer 11 is placed on the chuck table 28 so that the front surface 11a faces the holding surface 28a and the back surface 11b is exposed upward. In this state, when the suction force (negative pressure) of the suction source is applied to the holding surface 28a, the wafer 11 is sucked and held by the chuck table 28.

[0039] A protective sheet for protecting the device 15 (see FIG. 2) may be attached to the front surface 11a of the wafer 11. For example, the protective sheet includes a circular base material and an adhesive layer (glue layer) provided on the base material. In this case, the wafer 11 is suction-held by the chuck table 28 via the protective sheet.

[0040] Next, a modified layer 17 is formed inside the wafer 11 (modified layer forming step). In the modified layer forming step, the modified layer 17 is formed on the wafer 11 by irradiating the wafer 11 with a laser beam from the processing laser irradiation unit 46A.

[0041] The processing laser irradiation unit 46A includes a laser oscillator 60 such as a YAG laser, a YVO4 laser, a YLF laser, or the like, and an optical system 62 that guides the laser beam emitted from the laser oscillator 60 to the wafer 11 held by the chuck table 28. The optical system 62 includes a plurality of optical elements (lenses, mirrors, etc.) and controls the traveling direction, shape, etc. of the laser beam.

[0042] For example, the optical system 62 includes a mirror 64 and a condenser lens 66 such as a convex lens. The laser beam emitted from the laser oscillator 60 is reflected by the mirror 64 and enters the condenser lens 66, and is condensed at a predetermined position by the condenser lens 66. Then, the laser beam irradiated from the processing laser irradiation unit 46A is used as a laser beam (processing laser beam, first laser beam) 68 for processing the wafer 11.

[0043] In the modified layer forming step, first, the chuck table 28 is rotated to align the length direction of a predetermined dividing line 13 (see FIG. 2) with the X-axis direction. The chuck table 28 is also moved along the Y-axis direction to align the positions of the dividing line 13 and the focal point 68a of the laser beam 68 in the Y-axis direction. Furthermore, the focal point 68a of the laser beam 68 is positioned below the upper surface (rear surface 11b) of the wafer 11 and above the lower surface (front surface 11a) of the wafer 11.

[0044] Thereafter, the chuck table 28 is moved along the X-axis direction while irradiating the laser beam 68 from the processing laser irradiation unit 46A. As a result, the wafer 11 and the laser beam 68 move relatively along the X-axis direction, and the laser beam 68 is irradiated along the planned division line 13.

[0045] The irradiation conditions of the laser beam 68 are set so that the area inside the wafer 11 where the focal point 68a is located is modified (altered) by multiphoton absorption. Specifically, the wavelength of the laser beam 68 is set so that at least a part of the laser beam 68 passes through the wafer 11. That is, the laser beam 68 is a laser beam that is transparent to the wafer 11. The irradiation conditions of the other laser beams 68 are also set so that the inside of the wafer 11 is appropriately modified. For example, when the wafer 11 is a silicon wafer, the irradiation conditions of the laser beam 68 can be set as follows. Wavelength: 1064nm Average power output: 1W Repetition frequency: 100kHz Machining feed speed: 800mm / s

[0046] When the laser beam 68 is irradiated along the dividing lines 13, the inside of the wafer 11 is modified by multiphoton absorption. As a result, modified layers 17 are formed along the dividing lines 13 inside the wafer 11. Thereafter, by repeating the same procedure, modified layers 17 are formed along all of the dividing lines 13.

[0047] 4(A) is an enlarged cross-sectional view showing a part of the wafer 11. When the wafer 11 is irradiated with a laser beam 68, a modified layer 17 is formed in the area of ​​the wafer 11 where the focal point 68a is positioned and in the area nearby. Also, a crack 19 (fissure) is formed extending from the modified layer 17. The crack 19 extends from the modified layer 17 along the thickness direction of the wafer 11 and reaches the surface 11a of the wafer 11.

[0048] Next, an external force is applied to the wafer 11 to divide the wafer 11 along the intended division lines 13 (division step). For example, in the division step, a circular tape (expandable tape) that can be expanded by application of an external force is attached to the front surface 11a or back surface 11b of the wafer 11. Then, the expandable tape attached to the wafer 11 is pulled radially outward to expand the wafer 11, thereby applying an external force to the wafer 11.

[0049] The expanding tape may be expanded manually by an operator or by a dedicated expanding device. Furthermore, the application of an external force to the wafer 11 may be performed by a method other than expanding the expanding tape.

[0050] Here, the area of ​​the wafer 11 where the modified layer 17 or the crack 19 is formed becomes more fragile than other areas of the wafer 11. Therefore, when an external force is applied to the wafer 11, the wafer 11 is divided along the intended division lines 13, starting from the modified layer 17 and the crack 19. That is, the modified layer 17 and the crack 19 function as division starting points. In this way, a plurality of device chips, each including a device 15 (see FIG. 2), are manufactured.

[0051] However, if the irradiation conditions (average output, etc.) of the laser beam 68 are not properly set in the modified layer forming step, the cracks 19 may not be properly formed even if the modified layer 17 is formed. For example, the cracks 19 may not be generated, or the cracks 19 may extend while meandering in an unintended direction. Fig. 4(B) is an enlarged cross-sectional view showing a part of the wafer 11 in which the cracks 19 are not formed, and Fig. 4(C) is an enlarged cross-sectional view showing a part of the wafer 11 in which the meandering cracks 19 are formed.

[0052] If the cracks 19 are not properly formed inside the wafer 11, the wafer 11 may not be divided as intended even if an external force is applied to the wafer 11, which may result in processing defects. Therefore, it is preferable to inspect the processed state of the wafer 11 after the modified layer forming step and before the dividing step, and to check whether the cracks 19 are properly formed inside the wafer 11. A specific example of a method for inspecting the wafer 11 will be described below.

[0053] 5 is a partial cross-sectional front view showing the laser processing apparatus 2 for inspecting the wafer 11. When inspecting the wafer 11, first, a laser beam for observing the wafer 11 is irradiated onto the wafer (laser beam irradiation step). In the laser beam irradiation step, a laser beam is irradiated onto the wafer 11 from the observation laser irradiation unit 46B.

[0054] The observation laser irradiation unit 46B includes a laser oscillator 70 such as a YAG laser, a YVO4 laser, or a YLF laser, and an optical system 72 that guides the laser beam emitted from the laser oscillator 70 to the wafer 11 held by the chuck table 28. The laser oscillator 60 (see FIG. 3) of the processing laser irradiation unit 46A can also be used as the laser oscillator 70. The optical system 72 includes a plurality of optical elements (lenses, mirrors, etc.) and controls the traveling direction, shape, etc. of the laser beam.

[0055] Specifically, the optical system 72 includes a beam shaping unit 74 that shapes the laser beam emitted from the laser oscillator 60. The beam shaping unit 74 may be a plate-like member (light shielding plate) including a transmission portion through which the laser beam passes and a light shielding portion that blocks the laser beam. When the laser beam passes through the beam shaping unit 74, the laser beam is shaped according to the shape of the transmission portion. The beam shaping unit 74 may be configured by a diffractive optical element (DOE) or a liquid crystal on silicon - spatial light modulator (LCOS-SLM).

[0056] The optical system 72 also includes a dichroic mirror 76 and a condenser lens 78 such as a convex lens. The laser beam shaped by the beam shaping unit 74 is reflected by the dichroic mirror 76 and enters the condenser lens 78, where it is condensed at a predetermined position by the condenser lens 78. Note that the condenser lens 66 (see FIG. 3) of the processing laser irradiation unit 46A can also be used as the condenser lens 78. The laser beam irradiated from the observation laser irradiation unit 46B is used as a laser beam (observation laser beam, second laser beam) 80 for observing the wafer 11.

[0057] Moreover, the observation laser irradiation unit 46B includes an imaging unit (camera) 82. The imaging unit 82 includes an image sensor such as a CCD sensor or a CMOS sensor, and captures an image of the reflected light of the laser beam 80.

[0058] The laser beam 80 emitted from the observation laser irradiation unit 46B is reflected by the surface 11a of the wafer 11 and enters the observation laser irradiation unit 46B. The reflected light of the laser beam 80 passes through the condenser lens 78 and the dichroic mirror 76 to reach the imaging unit 82, and is imaged by the imaging unit 82. As a result, an image of the reflected light of the laser beam 80 (reflected light image) is acquired.

[0059] In the laser beam irradiation step, first, the positional relationship between the chuck table 28 and the observation laser irradiation unit 46B is adjusted so that the focal point 80a of the laser beam 80 overlaps with the intended division line 13 (modified layer 17). In addition, the focal point 80a of the laser beam 80 is positioned on the front surface 11a or inside (between the front surface 11a and the back surface 11b) of the wafer 11. In this state, the laser beam 80 is irradiated from the observation laser irradiation unit 46B to the back surface 11b side of the wafer 11.

[0060] The irradiation conditions of the laser beam 80 are set so that the reflected light of the laser beam 80 is incident on the condenser lens 78. Specifically, the wavelength of the laser beam 80 is set so that at least a part of the laser beam 80 is transmitted through the wafer 11. In other words, the laser beam 80 is a laser beam that is transparent to the wafer 11.

[0061] In addition, the output of the laser beam 80 is set so as not to exceed the processing threshold of the wafer 11. Specifically, the output of the laser beam 80 is set so as not to form a modified layer, cracks, or the like that function as a splitting starting point in the area of ​​the wafer 11 irradiated with the laser beam 80. Therefore, even if the wafer 11 is irradiated with the laser beam 80, laser processing that affects the quality of the wafer 11 is not performed on the wafer 11. For example, the average output of the laser beam 80 can be set to 1 / 1000 or more and 1 / 10 or less of the average output of the laser beam 68 (see FIG. 3), and other irradiation conditions of the laser beam 80 (repetition frequency, processing feed speed, etc.) can be set similarly to the irradiation conditions of the laser beam 68.

[0062] 6 is a plan view showing a part of the wafer 11 to which the laser beam 80 is irradiated. The laser beam 80 is irradiated so that the shape of the region on the back surface 11b of the wafer 11 to which the laser beam 80 is irradiated becomes asymmetric with respect to the modified layer 17. Specifically, the region on the back surface 11b of the wafer 11 that overlaps with the planned division line 13 is divided into two regions 21A and 21B by the modified layer 17. The shape of the region 21A to which the laser beam 80 is irradiated and the shape of the region 21B to which the laser beam 80 is irradiated become asymmetric with respect to the modified layer 17.

[0063] For example, the laser beam 80 is shaped by the beam shaping unit 74 (see FIG. 5) so that the cross-sectional shape in a direction (XY plane direction) perpendicular to the traveling direction (Z-axis direction) of the laser beam 80 emitted from the condenser lens 78 (see FIG. 5) is semicircular. When the focusing point 80a of the laser beam 80 is positioned so as to overlap with the modified layer 17, the semicircular laser beam 80 is irradiated onto the region 21A, and the laser beam 80 is not irradiated onto the region 21B. However, there is no restriction on the cross-sectional shape of the laser beam 80. For example, the cross-sectional shape of the laser beam 80 may be a polygonal shape such as a triangle or a rectangle, or may be a fan shape.

[0064] 7(A) is a cross-sectional view showing a state in which a wafer 11 in which a crack 19 has been formed is irradiated with a laser beam 80. When a laser beam 80 is irradiated from the back surface 11b side of the wafer 11 in which a crack 19 has been formed, the laser beam 80 travels inside the wafer 11 and is reflected on the front surface 11a side of the wafer 11.

[0065] Here, if a crack 19 is formed inside the wafer 11, extending from the modified layer 17 to the front surface 11a, the region of the wafer 11 below the modified layer 17 is divided by a small space (air layer) inside the crack 19. The laser beam 80 is also reflected when it reaches the crack 19. As a result, the reflected light of the laser beam 80 travels inside the wafer 11 along substantially the same path as the incident light of the laser beam 80, and is emitted from the back surface 11b of the wafer 11.

[0066] 7(B) is a cross-sectional view showing a state in which a laser beam 80 is irradiated onto a wafer 11 on which no cracks 19 are formed. When no cracks 19 are formed on the wafer 11, the progress of the laser beam 80 is not impeded by the cracks 19. Therefore, the laser beam 80 incident on one side of the modified layer 17 passes through the region below the modified layer 17 while being reflected on the front surface 11a side of the wafer 11, and is emitted from the other side of the modified layer 17. As a result, the path of the incident light of the laser beam 80 and the path of the reflected light are approximately symmetrical with respect to the modified layer 17 as an axis.

[0067] As described above, the path of the reflected light of the laser beam 80 changes depending on the processing state (state of the cracks 19) of the wafer 11. Then, as shown in FIG. 5, the reflected light of the laser beam 80 emitted from the back surface 11b side of the wafer 11 passes through the condenser lens 78 and the dichroic mirror 76 and reaches the imaging unit 82.

[0068] Next, an image of the reflected light of the laser beam 80 is obtained by capturing an image of the reflected light of the laser beam 80 (imaging step). In the imaging step, the reflected light of the laser beam 80 is captured by the imaging unit 82. As a result, an image of the reflected light of the laser beam 80 (reflected light image) is obtained. Examples of reflected light images 90 captured by the imaging unit 82 are shown in Figs. 8(A) to 8(C).

[0069] 8(A) is an image diagram showing a reflected light image 90 (first reflected light image 90A) obtained by capturing an image of the reflected light of a laser beam 80 emitted from a wafer 11 on which a crack 19 has been formed. When a crack 19 has been formed in the wafer 11, the reflected light of the laser beam 80 is emitted from the incident light side of the laser beam 80 (see FIG. 7(A)). As a result, a semicircular pattern corresponding to the reflected light of the laser beam 80 is displayed on the upper side of the first reflected light image 90A.

[0070] 8(B) is an image diagram showing a reflected light image 90 (second reflected light image 90B) obtained by capturing an image of the reflected light of the laser beam 80 emitted from a wafer 11 on which no cracks 19 are formed. When no cracks 19 are formed on the wafer 11, the reflected light of the laser beam 80 is emitted from the side opposite to the incident light of the laser beam 80 (see FIG. 7(B)). As a result, a semicircular pattern corresponding to the reflected light of the laser beam 80 is displayed below the second reflected light image 90B.

[0071] Fig. 8(C) is an image diagram showing a reflected light image 90 (third reflected light image 90C) obtained by capturing the reflected light of the laser beam 80 emitted from the wafer 11 in which a meandering crack 19 is formed. When a meandering crack 19 is formed inside the wafer 11 (see Fig. 4(C)), the laser beam 80 that reaches the crack 19 is reflected irregularly, and the reflected light of the laser beam 80 is emitted from the incident light side of the laser beam 80. As a result, a pattern different from the first reflected light image 90A is displayed on the upper side of the third reflected light image 90C.

[0072] Next, the processed state of the wafer 11 is judged based on the reflected light image 90 (judging step). As described above, a pattern reflecting the processed state of the wafer 11 (presence or absence of cracks 19, shape of cracks 19, etc.) appears in the reflected light image 90. That is, there is a correlation between the reflected light image 90 and the processed state of the wafer 11. Therefore, the processed state of the wafer 11 can be judged based on the reflected light image 90.

[0073] However, the shape and shading of the reflected light image appearing in the reflected light image 90 vary depending on various factors such as the irradiation conditions of the laser beam 80, the imaging conditions, and the state of the wafer 11. Therefore, in order to properly determine the processed state of the wafer 11, it is necessary to standardize the conditions for imaging the reflected light of the laser beam 80 and change the settings of the image processing for determining the processed state of the wafer 11 according to the reflected light image 90, which makes the inspection of the wafer 11 complicated.

[0074] Therefore, in this embodiment, a trained model configured by machine learning is used so as to output the processing state of the wafer 11 when a reflected light image 90 is input. This makes it possible to extract features of the reflected light image 90 and classify the reflected light image 90. As a result, it becomes possible to use images of the reflected light of the laser beam 80 captured under various conditions to determine the processing conditions of the wafer 11, and the inspection of the wafer 11 is simplified.

[0075] Fig. 9 is a block diagram showing the control unit 52. In addition to a block showing the functional configuration of the control unit 52, Fig. 9 also shows the display unit 50 and the imaging unit 82 of the observation laser irradiation unit 46B (see Fig. 5). The processing state of the wafer 11 is judged based on the image of the reflected light of the laser beam 80 by the control unit 52.

[0076] The control unit 52 includes a determination section 100 that determines the processed state of the wafer 11 based on an image of reflected light of the laser beam 80. A reflected light image 90 acquired by the imaging unit 82 is input to the determination section 100. The determination section 100 then determines the processed state of the wafer 11 based on the reflected light image 90 and outputs the determination result. The control unit 52 also includes a storage section 102 that can store various information (data, programs, etc.), and a notification section 104 that notifies the result of the determination by the determination section 100.

[0077] The determination unit 100 includes a trained model 110 configured by machine learning so as to output the processing state of the wafer 11 when the reflected light image 90 is input. There is no limitation on the type of trained model 110, and for example, a support vector machine (SVM), a neural network, etc. can be used. In this embodiment, as an example, a case where the trained model 110 is a neural network NN will be described.

[0078] The neural network NN is a hierarchical neural network, and includes an input layer 112 to which data is input, an output layer 114 to which data is output, and a plurality of hidden layers (intermediate layers) 116 provided between the input layer 112 and the output layer 114. The input layer 112, the output layer 114, and the hidden layer 116 each include a plurality of neurons (units, nodes). The neurons in the input layer 112 are connected to neurons in the first hidden layer 116, and the neurons in the output layer 114 are connected to neurons in the hidden layer 116, which is the final layer. The neurons in the hidden layer 116 are connected to neurons in the input layer 112 or the hidden layer 116 of the previous layer, and neurons in the output layer 114 or the hidden layer 116 of the subsequent layer.

[0079] The number of neurons included in the input layer 112, output layer 114, and hidden layer 116, and the activation function of each neuron can be freely set. There is also no limit to the number of hidden layers 116. A neural network NN including two or more hidden layers 116 can be called a deep neural network (DNN). Learning of a deep neural network can be called deep learning.

[0080] The neural network NN is trained to output the processed state of the wafer 11 when the reflected light image 90 is input. There is no limitation on the training method of the neural network NN. For example, the neural network NN is trained by supervised learning using a plurality of training images (reflected light images) including an image of the reflected light of the laser beam 80.

[0081] The learning image is obtained by carrying out the above-mentioned laser beam irradiation step and imaging step using, for example, a wafer (test wafer) for collecting learning images. Specifically, first, a modified layer is formed on a test wafer configured similarly to the wafer 11 (see FIG. 3). Then, a laser beam 80 is irradiated onto the test wafer on which the modified layer is formed, and the reflected light of the laser beam 80 is imaged by an imaging unit 82 (see FIG. 5). This results in a reflected light image that can be used as a learning image. A learning image 120 is shown in FIG. 10(A) to FIG. 10(C).

[0082] A plurality of learning images can be obtained by capturing an image of the reflected light of the laser beam 80 by the imaging unit 82 multiple times while changing the irradiation conditions of the laser beam 80, the irradiation position of the laser beam 80, the imaging conditions, etc. Also, a plurality of learning images may be obtained by using a plurality of test wafers on which modified layers are formed under different processing conditions.

[0083] Next, the multiple learning images 120 are classified (labeled). For example, each of the multiple learning images 120 is classified into a first reflected light image 120A, a second reflected light image 120B, or a third reflected light image 120C. Fig. 10(A) is an image diagram showing a learning image 120 classified into a first reflected light image 120A, Fig. 10(B) is an image diagram showing a learning image 120 classified into a second reflected light image 120B, and Fig. 10(C) is an image diagram showing a learning image 120 classified into a third reflected light image 120C.

[0084] The first reflected light image 120A is a reflected light image acquired when a crack extends linearly from a modified layer formed on the test wafer along the thickness direction of the test wafer and reaches the surface of the test wafer (see FIG. 4(A)). The second reflected light image 120B is a reflected light image acquired when no crack is formed on the test wafer (see FIG. 4(B)).

[0085] The third reflected light image 120C is a reflected light image other than the first reflected light image 120A and the second reflected light image 120B. For example, a reflected light image acquired when a crack extends in a meandering manner (see FIG. 4(C)) is classified as the third reflected light image 120C.

[0086] The learning images 120 are classified by an operator who can distinguish the state of the cracks 19 based on, for example, the reflected light images. Alternatively, after acquiring the learning images 120, the test wafer may be cut and the modified layer and cracks present on the cross section may be observed to directly confirm the state of the cracks 19. In this case, the learning images 120 can be classified based on the actual state of the cracks 19.

[0087] Alternatively, three types of test wafers on which modified layers are formed under three different processing conditions (processing conditions under which cracks are normally formed, processing conditions under which cracks are not formed, and other processing conditions) may be prepared in advance, and reflected light images may be obtained using each test wafer. In this case, the work of checking and classifying the learning images 120 one by one can be omitted.

[0088] Next, the neural network NN shown in Fig. 9 is trained using the classified training images 120. Specifically, supervised training is performed using the training images 120 and the classification results of the training images 120 (processed states of the wafer) as training data. For example, the neural network NN is trained using 100 each of the first reflected light images 120A, the second reflected light images 120B, and the third reflected light images 120C, totaling 300 training images 120. For example, the error backpropagation method can be used as the training algorithm.

[0089] When the above learning is performed, the parameters (neuron weights and biases) of the neural network NN are updated so that a determination result of the processed state of the wafer 11 (state of the cracks 19) is output from the output layer 114 when the reflected light image 90 is input to the input layer 112. In this way, a neural network NN capable of determining the processed state of the wafer 11 based on the reflected light image 90 is generated.

[0090] As described above, the neural network NN handles the classification problem of the reflected light image 90. For this reason, it is preferable to use a convolutional neural network (CNN) as the neural network NN. In this case, a convolutional layer, a pooling layer, a local contrast normalization (LCN) layer, a fully connected layer, etc. are provided as the hidden layer 116. Then, the filter values ​​of the convolutional layer and the weights and biases of the neurons of the fully connected layer are updated by supervised learning using the learning images 120.

[0091] When the reflected light image 90 captured by the imaging unit 82 is input to the neural network NN configured as described above, the processed state of the wafer 11 is determined by inference of the neural network NN. Specifically, calculations using the reflected light image 90 as input data are performed in the input layer 112, hidden layer 116, and output layer 114 in this order, and data corresponding to the processed state of the wafer 11 is output from the output layer 114.

[0092] When the neural network NN is a convolutional neural network, features of the reflected light image 90 are extracted by a convolution operation (generation of a feature map) in the convolution layer and a pooling process in the pooling layer. In addition, an operation for classifying the reflected light image 90 is performed in the fully connected layer. Then, a numerical value corresponding to the classification result of the reflected light image 90 is output from the output layer 114.

[0093] For example, the output layer 114 includes three neurons to which a softmax function is applied as an activation function. Each neuron outputs a numerical value (first output value) corresponding to the probability that the reflected light image 90 belongs to the first reflected light image 120A (see FIG. 10(A)), a numerical value (second output value) corresponding to the probability that the reflected light image 90 belongs to the second reflected light image 120B (see FIG. 10(B)), and a numerical value (third output value) corresponding to the probability that the reflected light image 90 belongs to the third reflected light image 120C (see FIG. 10(C)).

[0094] The determination unit 100 also includes a determination result output unit 118 that outputs a result of the determination made by the trained model 110. For example, the determination result output unit 118 outputs to the outside, as the determination result of the determination unit 100, the processed state of the wafer 11 corresponding to the largest output value among the three output values ​​output from the output layer 114 of the neural network NN.

[0095] Specifically, when the first output value is the largest, the judgment result output unit 118 outputs a signal indicating that an appropriate crack 19 has been formed in the wafer 11. When the second output value is the largest, the judgment result output unit 118 outputs a signal indicating that a crack 19 has not been formed in the wafer 11. When the third output value is the largest, the judgment result output unit 118 outputs a signal indicating that an inappropriate crack 19 has been formed in the wafer 11. However, the judgment result output unit 118 may output the first output value, the second output value, and the third output value directly to the outside.

[0096] If the probability of the judgment result output from the neural network NN is equal to or lower than a predetermined threshold, the reflected light image 90 used for judgment may be acquired again. For example, if the probability of the judgment result is equal to or lower than 60%, preferably equal to or lower than 80%, the irradiation conditions of the laser beam 80 (see FIG. 5) are changed, and then the imaging unit 82 captures the reflected light of the laser beam 80 to acquire the reflected light image 90 again. After that, the processing state of the wafer 11 is judged based on the newly acquired reflected light image 90.

[0097] The result of the judgment by the judgment unit 100 is output to the memory unit 102 and the notification unit 104. Then, the memory unit 102 stores the judgment result of the judgment unit 100 together with the reflected light image 90. As a result, the judgment result of the processing state of the wafer 11 is accumulated in the memory unit 102. Also, the notification unit 104 notifies the operator of the judgment result of the judgment unit 100.

[0098] For example, the notification unit 104 generates a control signal for displaying the determination result of the determination unit 100 on the display unit 50, and outputs the control signal to the display unit 50. As a result, the determination result of the determination unit 100, i.e., a message indicating the processing state of the wafer 11, is displayed on the display unit 50.

[0099] The display unit 50 may display the output values ​​(first to third output values) of the neural network NN. The display unit 50 may also display the reflected light image 90 used for the judgment and the judgment result of the judgment unit 100 at the same time. In this case, the operator can compare the reflected light image 90 with the judgment result of the judgment unit 100 to consider whether the judgment by the judgment unit 100 is appropriate or not.

[0100] Furthermore, the notification unit 104 may cause the laser processing device 2 (see FIG. 1) to issue a warning depending on the judgment result of the judgment unit 100. For example, the laser processing device 2 is equipped with a warning light (not shown) and a speaker (not shown). Then, when the judgment unit 100 judges that the processing state of the wafer 11 is abnormal, the notification unit 104 outputs a control signal to the warning light and the speaker to light up the warning light in a predetermined color or pattern and to cause the speaker to emit a sound or voice notifying the occurrence of the abnormality. In this way, the abnormality in the processing state of the wafer 11 is reported to the operator.

[0101] The criteria for issuing the warning can be set appropriately. For example, a warning is issued when it is determined that no crack 19 is formed in the wafer 11, or when it is determined that an inappropriate crack 19 (such as a meandering crack 19) is formed in the wafer 11. Also, a warning may be issued when the probability that an appropriate crack 19 is formed in the wafer 11 falls below a predetermined threshold.

[0102] As described above, the processing state of the wafer 11 is judged by the judgment unit 100. Then, if the processing state of the wafer 11 is judged to be appropriate, the next process (division step, etc.) is performed on the wafer 11. On the other hand, if the processing state of the wafer 11 is judged to be inappropriate, the processing of the other wafers 11 by the laser processing device 2 is interrupted. Then, the irradiation conditions of the laser beam 68 (see FIG. 3), the state of the optical system 62 (see FIG. 3) of the processing laser irradiation unit 46A, the state of the processed wafer 11, etc. are confirmed. After that, the processing conditions are adjusted, parts are replaced, etc., so that the modified layer 17 and cracks 19 are appropriately formed on the wafer 11.

[0103] The function of the determination unit 100 may be realized by either software or hardware. For example, the operations in the input layer 112, the output layer 114, and the hidden layer 116 of the neural network NN are described by a program, and the program is stored in the storage unit 102. When inspecting the wafer 11, the program is read from the storage unit 102 and executed by the control unit 52.

[0104] During or after the determination of the processing state of the wafer 11, the features of the reflected light image 90 extracted by the neural network NN may be visualized (visualization step). For example, when the neural network NN is a convolutional neural network, a heat map showing the activation state of the convolutional layer when the reflected light image 90 is input to the convolutional neural network is obtained by applying Grad-CAM (Gradient-weighted Class Activation Mapping) to the convolutional neural network. Then, the notification unit 104 outputs a control signal to the display unit 50, and causes the display unit 50 to display the visualization result (heat map). This allows the operator to understand the basis of the determination made by the neural network NN.

[0105] As described above, in the wafer inspection method according to this embodiment, the processed state of the wafer 11 is determined by the trained model 110 configured by machine learning so as to output the processed state of the wafer 11 when an image of the reflected light of the laser beam 80 is input. This makes it possible to use images of the reflected light of the laser beam 80 captured under various conditions to determine the processing conditions of the wafer 11, simplifying the inspection of the wafer 11.

[0106] In the imaging step (see FIG. 5), the wafer 11 and the laser beam 80 may be moved relatively to each other while the imaging unit 82 captures an image of the reflected light of the laser beam 80 multiple times.

[0107] 11(A) is a plan view showing the wafer 11 and the laser beam 80 moving relatively along the X-axis direction. In the imaging step, when the chuck table 28 (see FIG. 5) is moved along the X-axis direction, the wafer 11 and the laser beam 80 move relatively along a direction (X-axis direction) parallel to the planned division line 13. In this state, when the reflected light of the laser beam 80 is imaged multiple times at predetermined time intervals by the imaging unit 82 (see FIG. 5), multiple images of the reflected light of the laser beam 80 irradiated to different regions in the length direction of the planned division line 13 are obtained.

[0108] When the multiple reflected light images thus obtained are sequentially input to the judgment unit 100 (see FIG. 9), it is continuously judged whether or not the cracks 19 (see FIG. 7(A) etc.) are properly formed in each of the multiple regions irradiated with the laser beam 80. This makes it possible to quickly judge whether or not the cracks 19 are properly formed along the intended division line 13.

[0109] 11(B) is a plan view showing the wafer 11 and the laser beam 80 moving relatively along the Y-axis direction. In the imaging step, when the chuck table 28 (see FIG. 5) is moved along the Y-axis direction, the wafer 11 and the laser beam 80 move relatively along a direction perpendicular to the planned dividing line 13 (the Y-axis direction) so that the laser beam 80 straddles the modified layer 17. In this state, when the reflected light of the laser beam 80 is imaged multiple times at predetermined time intervals by the imaging unit 82 (see FIG. 5), multiple images of the reflected light of the laser beam 80 irradiated to different regions in the width direction of the planned dividing line 13 are obtained.

[0110] When the multiple reflected light images thus obtained are sequentially input to the judgment unit 100 (see FIG. 9), it is continuously judged whether or not the crack 19 (see FIG. 7(A) etc.) is properly formed in each of the multiple regions irradiated with the laser beam 80. This makes it possible to quickly identify at which position in the width direction of the planned division line 13 the crack 19 is formed.

[0111] In addition, in the imaging step, instead of moving the chuck table 28 (see FIG. 5), the wafer 11 and the laser beam 80 may be moved relatively along the X-axis or Y-axis direction by changing the positions and angles of the optical elements included in the optical system 72 of the observation laser irradiation unit 46B (see FIG. 5).

[0112] In the imaging step, the focal point 80a (see FIG. 5) of the laser beam 80 may be relatively moved along the thickness direction (Z-axis direction) of the wafer 11 while imaging the reflected light of the laser beam 80 multiple times. This makes it possible to confirm how far the crack 19 extends from the modified layer 17.

[0113] 12(A) is a cross-sectional view showing a state where a focal point 80a of a laser beam 80 is positioned in an area where a crack 19 is formed. When the laser beam 80 is irradiated onto the wafer 11 with the focal point 80a positioned in the area where the crack 19 exists, the laser beam 80 is reflected by the crack 19, and the reflected light of the laser beam 80 passes through the incident light side of the laser beam 80 and is emitted from the back surface 11b of the wafer 11. As a result, a reflected light image 90 similar to a first reflected light image 90A (see FIG. 8(A)) corresponding to the case where the crack 19 has reached the front surface 11a of the wafer 11 (see FIG. 7(A)) is acquired.

[0114] 12B is a cross-sectional view showing a state where the focal point 80a of the laser beam 80 is positioned in an area where no cracks 19 are formed. When the laser beam 80 is irradiated onto the wafer 11 with the focal point 80a positioned in an area where no cracks 19 exist, the laser beam 80 passes through an area below the cracks 19 while being reflected on the front surface 11a side of the wafer 11. Then, the reflected light of the laser beam 80 passes through the opposite side to the incident light of the laser beam 80 and is emitted from the back surface 11b of the wafer 11. As a result, a reflected light image 90 similar to a second reflected light image 90B (see FIG. 8B) corresponding to the case where no cracks 19 are formed (see FIG. 7B) is acquired.

[0115] Therefore, when the focal point 80a of the laser beam 80 is moved along the thickness direction (Z-axis direction) of the wafer 11 while the imaging unit 82 (see FIG. 5) captures the reflected light of the laser beam 80 multiple times, multiple reflected light images 90 that reflect the extension state of the crack 19 are obtained. Then, the multiple reflected light images 90 are input sequentially to the determination unit 100 (see FIG. 9), which determines whether or not there is a crack 19. This makes it possible to confirm how far the crack 19 has extended from the modified layer 17.

[0116] Incidentally, when acquiring learning images 120 (see FIGS. 10(A) to 10(C)) used for training the neural network NN, the wafer 11 and the laser beam 80 may be moved relatively to each other as described above, while the reflected light of the laser beam 80 is captured multiple times by the imaging unit 82. This allows a large number of learning images 120 to be collected efficiently.

[0117] In addition, the structures, methods, etc. according to the above-described embodiments can be modified as appropriate without departing from the scope of the object of the present invention. [Explanation of symbols]

[0118] 11 Wafers 11a surface 11b Back side 13. Planned division line (street) 15 Devices 17 Modified layer (degenerated layer) 19 Crack 21A,21B area 2 Laser processing equipment 4 Foundation 6 Mobile unit (mobile mechanism) 8 Y-axis movement unit (Y-axis movement mechanism, indexing feed unit) 10 Y-axis guide rail 12 Y-axis moving table 14 Y-axis ball screw 16 Y-axis pulse motor 18 X-axis movement unit (X-axis movement mechanism, processing feed unit) 20 X-axis guide rail 22 X-axis moving table 24 X-axis ball screw 26 X-axis pulse motor 28 Chuck table (holding table) 28a Holding surface 30 Z-axis movement unit (Z-axis movement mechanism) 32 Support structure 32a base 32b Support part 34 Z-axis guide rail 36 Z-axis moving table 38 Z-axis pulse motor 40 Support member 42 Laser irradiation unit 44 Laser processing head 46A Laser irradiation unit for processing 46B Observation laser irradiation unit 48 Imaging unit (camera) 50 Display unit (display unit, display device) 52 Control unit (control unit, control device) 60 Laser Oscillator 62 Optical system 64 Mirror 66 Condenser Lens 68 Laser beam (processing laser beam, first laser beam) 68a Focus point 70 Laser Oscillator 72 Optical system 74 Beam shaping unit 76 Dichroic Mirror 78 Condenser Lens 80 Laser beam (observation laser beam, second laser beam) 80a Focus point 82 Imaging unit (camera) 90 Reflected Light Images 90A First reflected light image 90B Second reflected light image 90C 3rd reflected light image 100 Judgment section 102 Storage section 104 Notification Department 110 Pre-trained models 112 Input Layer 114 Output Layer 116 Hidden layer (intermediate layer) 118 Judgment result output unit 120 Learning Images 120A First reflected light image 120B Second reflected light image 120C 3rd reflected light image

Claims

1. A wafer inspection method for inspecting a wafer having a modified layer formed therein along a planned division line, comprising the steps of: a laser beam irradiation step of positioning a focal point of a laser beam, the focal point of which is not greater than the processing threshold of the wafer and is transparent to the wafer, on the surface or inside of the wafer, and irradiating the wafer from the back surface side of the wafer with the laser beam so that the shape of the area of ​​the back surface of the wafer irradiated with the laser beam becomes asymmetric with respect to the modified layer; an imaging step of capturing an image of the reflected light by capturing an image of the reflected light of the laser beam; A determining step of determining a processing state of the wafer based on the image, In the determination step, a processed state of the wafer is determined using a trained model configured by machine learning so as to output the processed state of the wafer when the image is input; A wafer inspection method characterized in that the trained model classifies the processing state of the wafer into a processing state in which appropriate cracks are formed on the wafer extending from the modified layer, a processing state in which no cracks are formed on the wafer extending from the modified layer, and a processing state in which inappropriate cracks are formed on the wafer extending from the modified layer.

2. 2. The wafer inspection method according to claim 1, wherein in the imaging step, the reflected light is imaged a plurality of times while the wafer and the laser beam are moved relatively along a direction parallel to the intended dividing lines.

3. 2. The wafer inspection method according to claim 1, wherein in the imaging step, the reflected light is imaged a plurality of times while the wafer and the laser beam are moved relatively along a direction perpendicular to the intended dividing lines.

4. 2. The wafer inspection method according to claim 1, wherein in the imaging step, an image of the reflected light is captured a plurality of times while a focal point of the laser beam is relatively moved along a thickness direction of the wafer.

5. The trained model is a neural network including an input layer and an output layer, 5. The wafer inspection method according to claim 1, wherein the neural network outputs a processed state of the wafer from the output layer when the image is input to the input layer.

6. the neural network is trained by supervised learning using a plurality of training images including the image of the reflected light and classified according to the processing state of the wafer; The wafer inspection method of claim 5, characterized in that the learning image is classified into either a first reflected light image corresponding to a case where a crack extends normally from the modified layer toward the surface side of the wafer, a second reflected light image corresponding to a case where a crack does not extend from the modified layer toward the surface side of the wafer, or a third reflected light image other than the first reflected light image and the second reflected light image.

7. 7. The method of claim 5, further comprising a visualization step of visualizing the image features extracted by the neural network.

8. The neural network is a convolutional neural network, 8. The wafer inspection method according to claim 7, wherein the visualization step includes obtaining a heat map representing activation of a convolutional layer when the image is input to the convolutional neural network.

Images