Processing system, program, non-transitory recording medium, and method for generating learned model

Abstract

To provide a processing system capable of obtaining information relating to the position of a processing schedule region even when a key pattern is not set.SOLUTION: The processing system for processing an object to be processed includes a chuck table for holding the object to be processed, a processing unit for processing the object to be processed held by the chuck table, an imaging unit for imaging the object to be processed held by the chuck table, and a controller having a processing device and a storage device. The storage device stores a learned model configured by machine learning, so as to output the information relating to the position of a linear processing schedule region set in the object to be processed, when a first image including the object to be processed is inputted, and a program.SELECTED DRAWING: Figure 7

Description

[Technical field]

[0001] The present invention relates to a processing system for processing a workpiece, a program for obtaining information regarding the position of a planned processing area of ​​the workpiece, a non-transitory recording medium storing the program, and a method for generating a trained model for generating a trained model that outputs information regarding the position of a planned processing area of ​​the workpiece. [Background technology]

[0002] When processing a plate-shaped workpiece such as a semiconductor wafer in a linear processing area, the position of the processing area in the workpiece is calculated by a processing device using a method called pattern matching, which extracts a characteristic pattern from an image obtained by imaging the workpiece (see, for example, Patent Document 1). The processing device processes the workpiece with high precision based on the calculated position of the processing area.

[0003] In this method, a characteristic pattern (optically identifiable pattern) of a device or the like provided on the workpiece is selected as a key pattern by the operator of the processing device and set in advance in the processing device. Since the distance from the key pattern to the area to be processed is known, the processing device can calculate the position of the area to be processed by extracting a pattern similar to the set key pattern from the image. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2005-166991 A Summary of the Invention [Problem to be solved by the invention]

[0005] In the above-mentioned pattern matching, a characteristic pattern suitable for the key pattern needs to be selected by an operator and set in the processing device. Therefore, for example, in a situation where an operator has little experience and does not set an appropriate key pattern, the processing device cannot calculate the position of the intended processing area and cannot process the workpiece.

[0006] Therefore, an object of the present invention is to provide a processing system etc. that can obtain information regarding the position of a region to be processed even if a key pattern is not set. [Means for solving the problem]

[0007] According to one aspect of the present invention, there is provided a processing system for processing a workpiece, comprising: a chuck table for holding the workpiece; a processing unit for processing the workpiece held on the chuck table; an imaging unit for imaging the workpiece held on the chuck table; and a controller having a processing device and a memory device, wherein the memory device stores a trained model constructed by machine learning so as to output information relating to the position of a linear processing area set on the workpiece when a first image showing the workpiece is input; and a program, and the controller performs the following steps in accordance with the program and the trained model: causing the imaging unit to image the workpiece held on the chuck table to obtain the first image; obtaining the information relating to the position of the processing area from the first image; and having the processing unit process the processing area based on the information.

[0008] Preferably, the trained model is a neural network including an input layer to which the first image is input, and an output layer to output the information regarding the position of the area to be processed.

[0009] In addition, preferably, the processing system further includes a processing feed mechanism for moving the processing unit and the chuck table relatively in the processing feed direction, and a rotation mechanism for rotating the chuck table, and the controller further executes a procedure for rotating the chuck table based on the information so that the direction of the intended processing area is parallel to the processing feed direction in accordance with the program.

[0010] In addition, preferably, the information is a second image indicating the approximate position of the area to be processed, and the controller further executes, in accordance with the program, a step of applying image processing to the second image to detect the outer edge of the area to be processed, a step of calculating the center position in the width direction of the area to be processed from the detected outer edge, and a step of setting the center position as the position of the area to be processed to be processed by the processing unit.

[0011] According to another aspect of the present invention, a processing system for processing a workpiece includes a chuck table that holds the workpiece, a processing unit that processes the workpiece held on the chuck table, an imaging unit that images the workpiece held on the chuck table, and a first controller and a second controller, each having a processing device and a memory device, wherein a first program is stored in the memory device of the first controller, and a second program is stored in the memory device of the second controller, and a trained model configured by machine learning to output information regarding a position of a linear processing region set on the workpiece when a first image of the workpiece is input, and the first controller controls the second program. A processing system is provided that executes the following steps according to one program: having the imaging unit capture an image of the workpiece held on the chuck table to obtain the first image; sending the first image to the second controller; receiving the information regarding the position of the planned processing area or information based thereon from the second controller; and having the processing unit process the planned processing area based on the information received from the second controller, and the second controller executes the following steps according to the second program and the learned model: receiving the first image from the first controller, acquiring the information regarding the position of the planned processing area from the first image, and sending the information regarding the position of the planned processing area or information based thereon to the first controller.

[0012] Preferably, the trained model is a neural network including an input layer to which the first image is input, and an output layer to output the information regarding the position of the area to be processed.

[0013] In addition, preferably, the processing system further includes a processing feed mechanism for relatively moving the processing unit and the chuck table in a processing feed direction, and a rotation mechanism for rotating the chuck table, and the first controller further executes a procedure for rotating the chuck table based on the information so that the direction of the intended processing area is parallel to the processing feed direction in accordance with the first program.

[0014] In addition, preferably, the information is a second image indicating an approximate position of the area to be processed, and the first controller further executes, in accordance with the first program, a step of applying image processing to the second image to detect the outer edge of the area to be processed, a step of calculating the center position in the width direction of the area to be processed from the detected outer edge, and a step of setting the center position as the position of the area to be processed to be processed by the processing unit.

[0015] According to a further aspect of the present invention, there is provided a program for obtaining information regarding the position of a linear processing area set on a workpiece, the program including a trained model constructed by machine learning so as to output information regarding the position of the linear processing area set on the workpiece when a first image showing the workpiece is input, the program causing a computer to execute the steps of inputting the first image into the trained model and executing a calculation of the trained model.

[0016] Preferably, the trained model is a neural network including an input layer to which the first image is input, and an output layer to output the information regarding the position of the area to be processed.

[0017] According to yet another aspect of the present invention, there is provided a non-transitory recording medium on which the program is stored.

[0018] According to yet another aspect of the present invention, there is provided a method for generating a trained model for generating a trained model that outputs information regarding the position of a linear planned processing area set on a workpiece when an image of the workpiece is input, the method including: a first training image acquisition step of acquiring a first training image including a sample corresponding to the workpiece by imaging the sample; a second training image acquisition step of acquiring a second training image in which the position of the planned processing area is specified by specifying the position of the planned processing area in the first training image; and a learning step of generating a trained model that outputs information regarding the position of the linear planned processing area set on the workpiece when an image of the workpiece is input, by machine learning using the first training image and the second training image. Effect of the Invention

[0019] The processing system and the program according to each aspect of the present invention include a trained model configured by machine learning to output information about the position of a linear processing area set on the workpiece when a first image showing the workpiece is input. Therefore, according to the processing system and the program according to each aspect of the present invention, information about the position of the processing area can be obtained even if a key pattern is not set. [Brief description of the drawings]

[0020] [Figure 1] FIG. 1 is a perspective view showing a cutting device as a machining system. [Diagram 2] FIG. 2 is a perspective view showing the workpiece. [Diagram 3] FIG. 3 is an enlarged plan view showing a part of the surface of the workpiece. [Figure 4] FIG. 4 is a perspective view showing the structure of the cutting unit. [Diagram 5] FIG. 5 is a perspective view showing the workpiece after cutting. [Figure 6] FIG. 6 is an enlarged plan view showing a part of the workpiece after cutting. [Figure 7] FIG. 7 is a functional block diagram showing the functional structure of the controller. [Figure 8] FIG. 8 is a diagram showing an example of the first image. [Figure 9] FIG. 9 is a diagram showing an example of the second image. [Figure 10] FIG. 10 is a diagram showing an example of two lines indicating the outer edges of a region to be processed and one line indicating the center position in the width direction of the region to be processed. [Figure 11] FIG. 11 is a diagram showing a trained model providing system that provides a trained model to a cutting device. [Figure 12] FIG. 12 is a functional block diagram showing the functional structure of the controller when generating a trained model. [Figure 13] 13(A), 13(B), 13(C), and 13(D) are diagrams showing examples of the first learning image. [Figure 14] 14(A), 14(B), 14(C), and 14(D) are diagrams showing examples of the second learning image. [Figure 15] FIG. 15 is a diagram showing a machining system according to a modified example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings. Fig. 1 is a perspective view showing a cutting device 2 as a processing system. In Fig. 1, some components are expressed by functional blocks. In addition, the X-axis (front-back axis), Y-axis (left-right axis), and Z-axis (up-down axis) used in the following description are perpendicular to each other.

[0022] As shown in Fig. 1, the cutting device (machining system) 2 includes a base 4 that supports various components. A recess 4a with an open upper end is provided on the upper surface of the base 4 at one side (the front left side in Fig. 1) along the Y axis of the base 4. A cassette table 6 that is raised and lowered by a lifting mechanism (not shown) including, for example, a ball screw is disposed within this recess 4a.

[0023] A cassette 8 capable of accommodating, for example, a plurality of workpieces 11 is placed on the upper surface of the cassette table 6. A plurality of shelves (not shown), each capable of supporting one workpiece 11, are provided inside the cassette 8 so as to be stacked one on top of the other. For ease of explanation, the outline of the cassette 8 is shown by a two-dot chain line in FIG. 1.

[0024] 2 is a perspective view showing a workpiece 11. The workpiece 11 is, for example, a disk-shaped wafer made of a semiconductor such as single crystal silicon, and has a generally flat front surface (first surface) 11a and a back surface (second surface) 11b facing the opposite side to the front surface 11a. The front surface 11a side of the workpiece 11 is partitioned into a plurality of rectangular regions by a plurality of linear processing regions (streets) 13 that are set to intersect with each other.

[0025] Devices 15 such as ICs (Integrated Circuits), LEDs (Light Emitting Diodes), MEMS (Micro Electro Mechanical Systems), etc. are formed in the multiple regions partitioned by the multiple processing regions 13. By cutting this workpiece 11 at the processing regions 13, multiple device chips each having a device 15 are obtained.

[0026] 3 is an enlarged plan view showing a part of the surface 11a of the workpiece 11. For example, structures such as electrodes, terminals, wiring, etc. constituting a device 15 are provided on the surface 11a side of the workpiece 11, and an optically identifiable pattern (design) is formed by the arrangement of these structures. Note that, for example, structures such as electrodes, terminals, wiring, etc. constituting a TEG (Test Element Group) may be provided on the intended processing region 13 of the workpiece 11.

[0027] When such a workpiece 11 is machined by the cutting device 2, the workpiece 11 is supported by an annular frame 17 made of a metal such as SUS (stainless steel) in order to make the workpiece 11 easier to handle. A circular opening 17a is present in the center of the frame 17, penetrating the frame 17 in the thickness direction. The diameter of the opening 17a is larger than the diameter of the workpiece 11, and the workpiece 11 is placed inside the opening 17a.

[0028] With the workpiece 11 placed inside the opening 17a, a circular tape 19 is attached to the workpiece 11 and the frame 17. The tape 19 includes a substrate formed into a circular film shape from a resin such as polyolefin, polyvinyl chloride, or polyethylene terephthalate, and an adhesive layer (glue layer) formed on the substrate from an epoxy-, acrylic-, or rubber-based adhesive. The adhesive layer may be made of an ultraviolet-curable resin that cures when irradiated with ultraviolet light.

[0029] The center of the tape 19 is attached to the back surface 11b of the workpiece 11, and the outer periphery of the tape 19 is attached to the frame 17, so that the workpiece 11 is supported by the frame 17 via the tape 19. The workpiece 11 is accommodated in the cassette 8 in a state where it is supported by the frame 17 via the tape 19.

[0030] The type, material, shape, structure, size, etc. of the workpiece 11 are not limited to the above-mentioned aspects. For example, the workpiece 11 may be a substrate made of other semiconductors (GaAs, InP, GaN, SiC, etc.), sapphire, glass (quartz glass, borosilicate glass, etc.), ceramics, resin, metal, etc. In addition, the type, number, shape, structure, size, arrangement, etc. of the device 15 are not limited to the above-mentioned aspects.

[0031] Furthermore, the workpiece 11 may be a package substrate such as a CSP (Chip Size Package) substrate or a QFN (Quad Flat Non-leaded package) substrate. Such a package substrate is obtained, for example, by mounting a plurality of device chips on a substrate and then sealing the substrate with a resin (mold resin). Then, the package substrate is divided into individual pieces for each device chip, thereby completing a plurality of package devices, each of which includes a device chip.

[0032] 1, at a position adjacent to recess 4a along the Y axis, recess 4b is provided which is long in the direction along the X axis and has an open upper end on the upper surface of base 4. A chuck table moving mechanism (processing feed mechanism) 10 is disposed within recess 4b. Chuck table moving mechanism 10 includes a rotational drive source (not shown) such as a motor connected to an end of a ball screw, and an X-axis moving table (not shown) having a nut portion coupled to the ball screw, and moves the X-axis moving table along the X axis.

[0033] The upper part of the X-axis moving table is covered by a table cover 12. In addition, bellows-shaped dust-proof and drip-proof covers 14 that expand and contract in accordance with the movements of the X-axis moving table and the table cover 12 are attached to both ends of the table cover 12 in the direction along the X-axis. In other words, the chuck table moving mechanism 10 has a structure in which the X-axis moving table and the like are covered by the table cover 12 and the dust-proof and drip-proof cover 14.

[0034] A chuck table 16 for holding a workpiece 11 is disposed above the X-axis moving table in a manner that exposes it from the table cover 12. The chuck table 16 is connected to a rotary drive source (rotation mechanism) (not shown) such as a motor fixed to the X-axis moving table, and rotates around a rotation axis that is roughly parallel to the Z-axis. The chuck table 16 is moved along the X-axis together with the X-axis moving table by the above-mentioned chuck table moving mechanism 10 (processing feed).

[0035] The chuck table 16 includes a disk-shaped frame 18 made of a metal such as stainless steel. A recess with a circular opening at the top end is formed on the upper surface of the frame 18. A suction source (not shown) is connected to the bottom of the recess via a flow path (not shown) provided inside the frame 18 or a valve (not shown) arranged outside the frame 18. A disk-shaped, porous holding plate 20 that matches the shape of the recess is fitted into the recess.

[0036] Therefore, when the valve is opened while the suction source is in operation, the negative pressure of the suction source acts on the upper surface 16a of the chuck table 16, which includes the upper surface of the holding plate 20, through a flow path or the like. As the suction source, for example, a vacuum pump combining an air supply source and an ejector is used. However, a rotary pump or the like may also be used as the suction source. Four clamps 22 are arranged around the chuck table 16 to fix the annular frame 17 that supports the workpiece 11.

[0037] Moreover, one or more transport mechanisms (not shown) capable of transporting the workpiece 11 (frame 17) are disposed above the recess 4b. The workpiece 11 is, for example, carried out of a cassette 8 or the like by this transport mechanism, and placed on the upper surface 16a of the chuck table 16 via a tape 19 so that the surface 11a side is exposed upward.

[0038] When the workpiece 11 is placed on the upper surface 16a of the chuck table 16 and a negative pressure from a suction source is applied to this upper surface 16a, the workpiece 11 is held by the chuck table 16 via the tape 19. The upper surface 16a of the chuck table 16 is configured to be approximately parallel to the X-axis and Y-axis so as to properly hold the workpiece 11. The frame 17 that supports the workpiece 11 is fixed by four clamps 22.

[0039] The base 4 is provided with a gate-shaped support structure 24 that straddles the recess 4b along the Y axis. A pair of cutting unit moving mechanisms (indexing feed mechanism, incision feed mechanism) 26 are disposed on the upper portion of the support structure 24. The pair of cutting unit moving mechanisms 26 are substantially the same in structure, except that they are configured to be symmetrical (mirror image) with respect to a plane parallel to the X axis and Z axis. Therefore, corresponding components in the pair of cutting unit moving mechanisms 26 are given the same reference numerals, and duplicated explanations will be omitted.

[0040] The pair of cutting unit moving mechanisms 26 share a pair of Y-axis guide rails 28 that are fixed to the front (surface) of the support structure 24 and are long in a direction generally parallel to the Y-axis. A Y-axis moving plate 30 constituting each cutting unit moving mechanism 26 is attached to the pair of Y-axis guide rails 28 in a manner that allows it to slide along the Y-axis.

[0041] A nut portion (not shown) constituting a ball screw is provided on the rear side (back side) of each Y-axis moving plate 30, and a screw shaft 32 that is long in a direction generally parallel to the Y-axis guide rail 28 is rotatably connected to each nut portion. A rotation drive source 34 such as a motor is connected to one end of each screw shaft 32. By rotating the screw shaft 32 by each rotation drive source 34, the target Y-axis moving plate 30 moves along the Y-axis guide rail 28.

[0042] A pair of Z-axis guide rails 36 that are long in a direction generally parallel to the Z-axis are fixed to the front (surface) of each Y-axis moving plate 30. A Z-axis moving plate 38 is attached to the pair of Z-axis guide rails 36 fixed to each Y-axis moving plate 30 in such a manner that it can slide along the Z-axis.

[0043] A nut portion (not shown) constituting a ball screw is provided on the rear side (back side) of each Z-axis moving plate 38, and a screw shaft 40 that is long in a direction generally parallel to the Z-axis guide rail 36 is rotatably connected to each nut portion. A rotation drive source 42 such as a motor is connected to one end of each screw shaft 40. By rotating the screw shaft 40 by each rotation drive source 42, the target Z-axis moving plate 38 moves along the Z-axis guide rail 36.

[0044] A cutting unit (processing unit) 44 is fixed to the lower part of each Z-axis moving plate 38. That is, the cutting device 2 has a pair of cutting units 44. The structure of the pair of cutting units 44 is substantially the same except that they are configured to be symmetrical (mirror image) with respect to a plane parallel to the X-axis and Z-axis. Therefore, corresponding components in the pair of cutting units 44 are given the same reference numerals, and duplicated explanations will be omitted.

[0045] Fig. 4 is an exploded perspective view showing a schematic structure of the cutting unit 44. As shown in Fig. 4, the cutting unit 44 includes a cylindrical spindle housing 46. The spindle housing 46 accommodates a spindle 48 having an axis generally parallel to the Y axis.

[0046] The tip (one end) of the spindle 48 is exposed to the outside of the spindle housing 46. A screw hole 48a is provided in the tip surface of the tip of the spindle 48. A rotation drive source (not shown), such as a motor, is connected to the base end (the other end) of the spindle 48.

[0047] An annular cutting blade 52 is attached to the tip of the spindle 48 via a blade mounter 50. The blade mounter 50 includes a mount flange 54 that is attached to the tip of the spindle 48. The mount flange 54 is composed of a fixed mount 56 that is fixed to the tip of the spindle 48, and a press flange 58 that presses the cutting blade 52 attached to the fixed mount 56.

[0048] The fixed mount 56 includes a disk-shaped flange portion 60 that supports the cutting blade 52, and a cylindrical boss portion 62 that protrudes from the center of a circular front surface (one surface) 60a of the flange portion 60. The fixed mount 56 has a through hole 56a that penetrates the center of the flange portion 60 from the front surface 60a side to the back surface (the other surface) 60b side, and penetrates the center of the boss portion 62 from the tip end 62a side to the base end side (flange portion 60 side).

[0049] The fixed mount 56 of the mount flange 54 is attached to the spindle 48 from the back surface 60b side of the flange portion 60 by inserting the tip portion of the spindle 48 into the through hole 56a from the back surface 60b side of the flange portion 60. The through hole 56a is provided with an annular receiving portion 56b that receives a washer 64. The washer 64 is placed in the through hole 56a, and a screw 66 is fastened into the screw hole 48a of the spindle 48 via the washer 64, whereby the fixed mount 56 is fixed to the tip portion of the spindle 48.

[0050] The flange portion 60 has an annular protrusion 60c on the outer periphery on the surface 60a side, protruding slightly in a direction intersecting the surface 60a (perpendicular to the surface 60a). The tip surface 60d of the protrusion 60c is formed to be generally flat. The outer periphery 62b of the boss portion 62 has a screw thread in an area on the tip 62a side.

[0051] The cutting blade 52 is an annular member having a structure in which abrasive grains made of diamond, cubic boron nitride, or the like are dispersed and fixed by a binder such as metal, resin, or ceramic. That is, a circular opening 52a is provided in the center of the cutting blade 52, penetrating the cutting blade 52 in the thickness direction. When attaching the cutting blade 52 to the fixed mount 56, the boss portion 62 is inserted into the opening 52a so that one surface of the cutting blade 52 contacts the tip surface 60d of the flange portion 60.

[0052] With the cutting blade 52 attached to the fixed mount 56, a pressing flange 58 is attached to the fixed mount 56. The pressing flange 58 is made of a material such as stainless steel and has a truncated cone shape, and has a curved surface 58a which corresponds to the side surface of the truncated cone, and a flat surface 58b which corresponds to the bottom surface of the truncated cone and is located approximately opposite to the curved surface 58a.

[0053] An opening 58c is provided in the center of the pressing flange 58, penetrating the pressing flange 58 in the thickness direction. When the pressing flange 58 is attached to the fixed mount 56, the boss portion 62 is inserted into this opening 58c. When the pressing flange 58 is attached to the fixed mount 56 with the cutting blade 52 attached to the fixed mount 56, the flat surface 58b of the pressing flange 58 comes into contact with the other surface of the cutting blade 52.

[0054] The blade mounter 50 further includes a blade fastener 68 used to fasten the annular cutting blade 52 to the mount flange 54. The blade fastener 68 is typically an annular nut (round nut) having an opening 68a in its center into which the boss portion 62 of the fixed mount 56 is inserted. A screw thread corresponding to the screw thread on the outer circumferential surface 62b of the boss portion 62 is formed on the inner circumferential surface of the opening 68a.

[0055] When the cutting blade 52 is fixed to the mount flange 54, the boss portion 62 is inserted into the opening 52a of the cutting blade 52 and the opening 58c of the pressing flange 58, and then the boss portion 62 is inserted into the opening 68a of the blade fixing device 68. Thereafter, the boss portion 62 and the blade fixing device 68 are rotated relative to each other, whereby the boss portion 62 is fastened to the blade fixing device 68, and the cutting blade 52 is clamped between the fixed mount 56 and the pressing flange 58.

[0056] 1, a blade cover 70 capable of partially covering the cutting blade 52 attached to the spindle 48 is provided at the end of the spindle housing 46 constituting the cutting unit 44 on the side of the recess 4b. A pair of nozzles 72 capable of supplying a processing liquid (working fluid) such as water to the cutting blade 52 and the like are arranged at the bottom of the blade cover 70 so as to sandwich the cutting blade 52 along the Y axis.

[0057] A camera (imaging unit) 74 capable of capturing images of the workpiece 11 held on the chuck table 16 is disposed adjacent to each cutting unit 44 along the X-axis. Each camera 74 is fixed to the lower part of the corresponding Z-axis moving plate 38, similar to the cutting unit 44.

[0058] The camera 74 includes a two-dimensional optical sensor such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor sensitive to visible light or infrared light, and is used to capture images of the workpiece 11 held by the chuck table 16.

[0059] The images captured by the camera 74 are used for adjusting (aligning) the positions and orientations of the workpiece 11 held by the chuck table 16 and each cutting unit 44. In addition, based on the images of the workpiece 11 captured by the camera 74, a task for checking the quality of the processing (kerf check) may be performed.

[0060] When the target Y-axis moving plate 30 is moved along the Y-axis by the above-mentioned cutting unit moving mechanism 26, the cutting unit 44 and camera 74 supported by the target Y-axis moving plate 30 also move along the Y-axis (indexing feed). Also, when the target Z-axis moving plate 38 is moved along the Z-axis by the cutting unit moving mechanism 26, the cutting unit 44 and camera 74 supported by the target Z-axis moving plate 38 also move along the Z-axis (cutting feed).

[0061] 1, a recess 4c with an open upper end is provided on the upper surface of the base 4 at a position opposite to the recess 4a with respect to the recess 4b. A cleaning unit 76 for cleaning the workpiece 11 after processing is disposed in the recess 4c. The cleaning unit 76 includes, for example, a spinner table 78 and a nozzle 80, and sprays a cleaning fluid from the nozzle 80 onto the workpiece 11 held and rotated by the spinner table 78.

[0062] The upper part of the base 4 is covered by a device cover 82 capable of housing each of the components. In Fig. 1, for the sake of convenience, the outline of the device cover 82 is shown by a two-dot chain line. A touch screen (input device, output device) 84 serving as a user interface is disposed on one side (the left side in Fig. 1) of the device cover 82 along the Y axis. In addition, an indicator light (output device) 86 capable of conveying information by the color of light and the state of light emission (light emitting, blinking, off, etc.) is provided on the upper part of the device cover 82.

[0063] As an input device for the operator to input information, a keyboard, a mouse, or the like may be used instead of the touch screen 84. Similarly, as an output device for presenting information to the operator, a display device such as a liquid crystal display that does not have an input function, or a speaker that can transmit information by voice may be used.

[0064] A controller (control unit) 88 is connected to the various components (such as the chuck table moving mechanism 10, the transport mechanism, the cutting unit moving mechanism 26, the cutting unit 44, the camera 74, the cleaning unit 76, the touch screen 84, and the indicator light 86) of the cutting device 2 described above. The controller 88 is configured, for example, by a computer including a processing device 90 and a storage device 92, and the operation of each component is controlled by this controller 88.

[0065] The processing device 90 includes, for example, a processor such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The storage device 92 includes, for example, a main storage device such as a DRAM (Dynamic Random Access Memory) and an auxiliary storage device such as a hard disk drive or a flash memory. The processing device 90 operates in accordance with a program (software) stored in the storage device 92, thereby realizing the functions of the controller 88.

[0066] When cutting the workpiece 11 with the cutting device 2 configured as above, first, the workpiece 11 is held by the chuck table 16. For example, after the workpiece 11 is carried out of the cassette 8 by the transport mechanism, it is placed on the chuck table 16 so that the front surface 11a side is exposed upward and the back surface 11b side (the tape 19 side) faces the upper surface 16a of the chuck table 16.

[0067] In this state, when a suction force (negative pressure) of a suction source is applied to the upper surface 16a of the chuck table 16, the workpiece 11 is held by the chuck table 16 via the tape 19. The frame 17 supporting the workpiece 11 via the tape 19 is fixed by a plurality of clamps 22.

[0068] Next, the workpiece 11 is cut in the intended processing region 13 by the cutting blade 52. Specifically, first, the rotary drive source (rotation mechanism) rotates the chuck table 16 so that the length direction of the target intended processing region 13 becomes parallel to the X-axis. In addition, the position of the cutting unit 44 in the direction along the Y-axis is adjusted by the cutting unit moving mechanism 26 so that the cutting blade 52 is positioned above the extension line of the target intended processing region 13.

[0069] Furthermore, the position (height) of the cutting unit 44 in the direction along the Z axis is adjusted by the cutting unit moving mechanism 26 so that the lower end of the cutting blade 52 is located below the front surface 11a of the workpiece 11. When cutting the workpiece 11 to form a plurality of device chips, the position of the cutting unit 44 in the direction along the Z axis is adjusted by the cutting unit moving mechanism 26 so that the lower end of the cutting blade 52 is located below the back surface 11b of the workpiece 11 (the upper surface of the tape 19).

[0070] Thereafter, while the cutting blade 52 is rotating, the chuck table moving mechanism 10 moves the chuck table 16 along the X-axis. As a result, the chuck table 16 and the cutting blade 52 move relatively along the X-axis (processing feed), and the cutting blade 52 cuts into the intended processing region 13 of the workpiece 11. That is, the workpiece 11 is cut in the intended processing region 13.

[0071] The same procedure is repeated to cut the workpiece 11 in all of the intended processing regions 13. After cutting, the workpiece 11 is transported by the transport mechanism to the cleaning unit 76 and cleaned by the cleaning unit 76. After cleaning by the cleaning unit 76, the workpiece 11 is transported by the transport mechanism to the cassette 8 and stored therein.

[0072] Fig. 5 is a perspective view showing the workpiece 11 after cutting. Fig. 5 shows the workpiece 11 cut at the intended processing regions 13. When the workpiece 11 is cut by the cutting blade 52 as described above, a processing mark (kerf) 11c extending from the front surface 11a to the back surface 11b of the workpiece 11 is formed in the intended processing region 13. When the workpiece 11 is cut at all the intended processing regions 13, a plurality of device chips each having a device 15 are obtained.

[0073] Fig. 6 is an enlarged plan view of a part of the workpiece 11 after cutting. As shown in Fig. 6, the processing mark 11c is formed linearly along the length direction of the processing target region 13 so as to pass through the center position in the width direction of the processing target region 13. If a structure such as a TEG exists on the processing target region 13, this structure is also cut.

[0074] Incidentally, when cutting the workpiece 11 by the above-mentioned method, a method called pattern matching is generally used to adjust (align) the positions and orientations of the workpiece 11 held by the chuck table 16 and each cutting unit 44. Specifically, the controller 88 images the front surface 11a side of the workpiece 11 with the camera 74, and extracts a characteristic pattern consisting of electrodes, wiring, terminals, etc. from the obtained image.

[0075] When this method is adopted, a characteristic pattern provided on the workpiece 11 is selected in advance by the operator as a key pattern, and set in the cutting device 2. On the other hand, in this pattern matching, a characteristic pattern suitable for the key pattern needs to be selected by the operator and set in the cutting device 2. Therefore, for example, in a situation where the operator is inexperienced and does not set an appropriate key pattern, the controller 88 cannot calculate the position of the to-be-processed region 13, and the workpiece 11 cannot be processed.

[0076] Therefore, in the cutting device 2 of this embodiment, the controller 88 acquires information regarding the position, orientation, etc. (hereinafter, position, etc.) of the linear to-be-processed region 13 without using a key pattern, from an image obtained by capturing an image of the surface 11a of the workpiece 11 with the camera 74. Specifically, the controller 88 acquires information regarding the position, etc. of the to-be-processed region 13 by using a trained model configured by machine learning so as to output information regarding the position, etc. of the to-be-processed region 13 when an image of the workpiece 11 is input.

[0077] This eliminates the need for the operator to select a characteristic pattern suited to the key pattern and set it in the controller 88, so that even in a situation where the operator is inexperienced and is unable to set an appropriate key pattern, the controller 88 can calculate the position of the intended processing area 13, etc., and appropriately cut the workpiece 11.

[0078] The trained model is typically stored in the storage device 92 in a state where it is incorporated into a program for implementing the functions of the controller 88. In other words, the program includes the trained model. However, the trained model may also be stored in the storage device 92 in a manner independent of the program.

[0079] Fig. 7 is a functional block diagram showing the functional structure of a controller 88 realized by a program and a trained model. For convenience of explanation, Fig. 7 shows the chuck table 16, the camera 74, the touch screen 84, and the like in addition to the blocks showing the functional structure of the controller 88. As shown in Fig. 7, the controller 88 includes a first image acquisition unit 88a, a second image acquisition unit 88b, a position determination unit 88c, a display control unit 88d, a positional relationship adjustment unit 88e, and a processing control unit 88f.

[0080] The first image acquisition unit 88a causes the camera 74 to capture the workpiece 11 and acquires a first image 21 including the workpiece 11. The second image acquisition unit 88b acquires a second image 23 including information on the position of the to-be-processed region 13, etc., based on the first image 21. The position determination unit 88c performs image processing, etc. on the second image 23 to determine the position and orientation of the to-be-processed region 13.

[0081] The display control unit 88d displays the first image 21, the second image 23, the outer edge of the determined area to be processed 13, a line (center line) passing through the center of the area to be processed 13, and the like on the touch screen 84. The positional relationship adjustment unit 88e adjusts the relationship of the position and orientation between the chuck table 16 that holds the workpiece 11 and each cutting unit 44, based on the position and orientation of the area to be processed 13 determined by the position determination unit 88c. The machining control unit 88f causes the cutting units 44 to cut the area to be processed 13 of the workpiece 11.

[0082] Next, a detailed description will be given of the functions of each component of the controller 88 and the procedures executed by the controller 88. When cutting the workpiece 11, the controller 88, in accordance with a program stored in the storage device 92, transports the workpiece 11 contained in the cassette 8 to the chuck table 16 by the transport mechanism and causes the workpiece 11 to be held on the chuck table 16.

[0083] Next, the first image acquisition unit 88a of the controller 88 executes a procedure of having the camera 74 capture an image of the workpiece 11 held on the chuck table 16 to acquire a first image 21. Specifically, the controller 88 adjusts the positional relationship between the chuck table 16 and the camera 74 in accordance with a program stored in the storage device 92, and has the camera 74 capture an image of a partial area of ​​the surface 11a of the workpiece 11.

[0084] Fig. 8 is a diagram showing an example of a first image 21 obtained by imaging the workpiece 11 with the camera 74. For ease of explanation, the same reference numerals as those of the components of the workpiece 11 are attached to the parts of the first image 21 in which the components of the workpiece 11 are captured. The first image 21 shown in Fig. 8 shows the region to be processed 13 and the device 15. This first image 21 is obtained, for example, by enlarging and imaging a part of the front surface 11a side of the workpiece 11.

[0085] If the outer edges of the to-be-processed region 13 located at both ends in the width direction of the to-be-processed region 13 are not captured in the first image 21, there is a high possibility that information regarding the position, etc. of the to-be-processed region 13 cannot be properly acquired. Therefore, in such a case, the controller 88 moves the chuck table 16 and the camera 74 relatively to each other, and causes the camera 74 to capture an image of another part of the region of the surface 11a of the workpiece 11.

[0086] When the first image 21 is acquired by the first image acquisition unit 88a, the second image acquisition unit 88b of the controller 88 executes a procedure for acquiring information on the position of the to-be-processed region 13 and the like from the first image 21. Specifically, the controller 88 inputs the first image 21 to the trained model 94 in accordance with the program stored in the storage device 92, and executes processing related to the calculation. As a result, the second image acquisition unit 88b obtains the second image 23 including information on the position of the to-be-processed region 13 and the like.

[0087] Here, the trained model 94 is configured by machine learning so as to output a second image 23 including information on the position of the to-be-processed region 13 set on the workpiece 11 when a first image 21 including the workpiece 11 is input. In this embodiment, a neural network NN is used as the trained model 94. However, there is no limitation on the type of the trained model 94. For example, a support vector machine (SVM) or the like may be used as the trained model 94.

[0088] The neural network NN is a hierarchical neural network, and includes an input layer 94a to which data is input, an output layer 94b to which data is output, and a plurality of hidden layers (intermediate layers) 94c provided between the input layer 94a and the output layer 94b. Each of the input layer 94a, the output layer 94b, and the hidden layer 94c includes a plurality of neurons (units, nodes).

[0089] The neurons in the input layer 94a are connected to the neurons in the first hidden layer 94c, and the neurons in the output layer 94b are connected to the neurons in the last hidden layer 94c. The neurons in the hidden layer 94c are also connected to the neurons in the input layer 94a or the hidden layer 94c of the previous layer, and to the neurons in the output layer 94b or the hidden layer 94c of the next layer.

[0090] There are no limitations on the number of neurons included in the input layer 94a, the output layer 94b, and the hidden layer 94c, the activation function of each neuron, the connection relationship between each neuron, etc. Furthermore, there are no limitations on the number of layers in the hidden layer 94c. Note that a neural network NN including two or more hidden layers 94c is sometimes called a deep neural network (DNN). Furthermore, learning of a deep neural network is sometimes called deep learning.

[0091] As described above, this neural network NN is configured by machine learning so that when a first image 21 including the workpiece 11 is input, the neural network NN outputs a second image 23 including information on the position of the to-be-processed region 13 set on the workpiece 11. There is no limitation on the machine learning method of the neural network NN. A specific example of the machine learning method of the neural network NN will be described later.

[0092] When a first image 21 showing the workpiece 11 is input to the neural network NN, a second image 23 including information on the position of the to-be-processed region 13 and the like is generated by inference of this neural network NN. For example, a calculation using information on the pixels constituting the first image 21 as input data is performed in the input layer 94a, hidden layer 94c, and output layer 94b in this order, and information on the pixels constituting the second image 23 is output as output data from the output layer 94b.

[0093] Fig. 9 is a diagram showing an example of the second image 23 obtained by the second image acquisition unit 88b. As shown in Fig. 9, the second image 23 is, for example, a binary image, and includes a first region 23a represented by pixels of a first luminance and a second region 23b represented by pixels of a second luminance different from the first luminance.

[0094] In the second image 23, the first region 23a is disposed at a position generally corresponding to the to-be-processed region 13 in the first image 21, and the second region 23b is disposed at a position generally corresponding to the device 15 in the first image 21. That is, the second image 23 is generated so as to include information on the position of the to-be-processed region 13, etc. Therefore, the controller 88 can calculate the position and orientation of the to-be-processed region 13 based on this second image 23.

[0095] The second image 23 does not necessarily have to be a binary image as long as it contains information about the position of the to-be-processed region 13 extracted from the first image 21. In addition, both the first image 21 and the second image 23 contain information about the position of the to-be-processed region 13, and there is a clear correlation between them.

[0096] When the second image 23 is acquired by the second image acquisition unit 88b, for example, a position determination unit 88c of the controller 88 executes a procedure for determining the position and orientation of the to-be-processed region 13 from the second image 23. While the second image 23 acquired by the second image acquisition unit 88b can indicate the approximate position and orientation of the to-be-processed region 13, it may not necessarily indicate the position and orientation of the to-be-processed region 13 with sufficiently high accuracy.

[0097] Therefore, the position determination unit 88c executes a procedure of detecting the outer edge of the to-be-processed region 13 by performing image processing on the second image 23, and a procedure of calculating the center position in the width direction of the to-be-processed region 13 from the detected outer edge, thereby determining the position and orientation of the to-be-processed region 13. Fig. 10 is a diagram showing an example of two lines (solid lines) 23c indicating the outer edge of the to-be-processed region 13, and one line (dashed line) 23d indicating the center position in the width direction of the to-be-processed region 13. Note that in Fig. 10, the lines 23c and 23d are superimposed on the second image 23 for convenience of explanation.

[0098] Specifically, the controller 88 performs preprocessing such as smoothing on the second image 23 in accordance with a program stored in the memory device 92, and then uses any edge detection method to extract the position of the outer edge of the area to be processed 13, i.e., two lines 23c indicating the outer edge of the area to be processed 13.

[0099] As the edge detection method, for example, a Sobel filter that detects the contour with a gradient detection filter by first derivative, a Laplacian filter that detects the contour with a gradient detection filter by second derivative, etc. are used. In addition, the controller 88 calculates the center position in the width direction of the processing target region 13, that is, a line 23d that is equidistant from the two lines 23c, according to a program stored in the storage device 92.

[0100] Thereafter, the controller 88, in accordance with the program stored in the storage device 92, sets the center position in the width direction of the to-be-processed area 13 represented by the calculated line 23d as the position of the to-be-processed area 13 to be cut by the cutting unit 44. Specifically, one or more coordinates representing the position are stored in the storage device 92 of the controller 88.

[0101] Moreover, the controller 88 calculates the orientation of the to-be-processed region 13 (the direction of the length of the to-be-processed region 13) according to a program stored in the storage device 92. Specifically, the controller 88 calculates the orientation of the above-mentioned line 23d as the orientation of the to-be-processed region 13, and stores it in the storage device 92. The controller 88 can also obtain coordinates representing the position of the target to-be-processed region 13 in each of a plurality of separate regions on the workpiece 11, and calculate the orientation of the to-be-processed region 13 based on these plurality of coordinates.

[0102] In this way, the position determination unit 88c executes the procedure for determining the position and orientation of the to-be-processed region 13, thereby making it possible to adjust with high precision the relationship between the position and orientation of the chuck table 16 and each cutting unit 44. However, in cases where the accuracy of the trained model 94 is extremely high and the position and orientation of the to-be-processed region 13 are indicated with sufficiently high precision by the second image 23, part or all of the procedure by the position determination unit 88c may be omitted.

[0103] That is, the center position in the width direction of the to-be-processed region 13 may be determined directly from the second image 23 without image processing being applied to the second image 23. Furthermore, the trained model 94 may be configured to output coordinates or the like indicating the position and orientation of the to-be-processed region 13, instead of the second image 23 including information on the position and the like of the to-be-processed region 13. In this case as well, the procedure by the position determination unit 88c may be omitted.

[0104] When the position and orientation of the to-be-processed region 13 are determined, for example, the display control unit 88d executes a procedure for displaying information on the position and orientation of the to-be-processed region 13 on the touch screen 84. Specifically, the controller 88, in accordance with a program stored in the storage device 92, causes the touch screen 84 to display an image in which two lines 23c and one line 23d are superimposed on the first image 21.

[0105] However, there is no limit to the type of image displayed on the touch screen 84. For example, only the first image 21 shown in Fig. 8 may be displayed on the touch screen 84, or only the second image 23 shown in Fig. 9 may be displayed. Furthermore, any two or more of these three types of images may be displayed in a switching manner at a predetermined time, any two or more of these three types of images may be displayed side by side at the same time, or any two or more of these three types of images may be displayed overlapping each other.

[0106] Furthermore, when the position and orientation of the planned processing area 13 are determined by the position determination unit 88c, the positional relationship adjustment unit 88e executes a procedure for adjusting the relationship between the position and orientation of the chuck table 16 that holds the workpiece 11 and each cutting unit 44. Specifically, the controller 88 rotates the chuck table 16 by the rotary drive source (rotation mechanism) (alignment) in accordance with a program stored in the storage device 92 so that the orientation (length direction) of the planned processing area 13 determined by the position determination unit 88c becomes parallel to the X-axis (processing feed direction). This allows the cutting device 2 to appropriately cut the workpiece 11 along the planned processing area 13 with each cutting unit 44.

[0107] When the positional relationship adjustment unit 88e adjusts the positional and directional relationship between the chuck table 16 and each cutting unit 44, the machining control unit 88f executes a procedure for causing the cutting units 44 to perform cutting work on the position of the to-be-machined region 13 set by the position determination unit 88c. Specifically, the controller 88 first adjusts the position of the chuck table 16 in the direction along the X-axis by the chuck table moving mechanism 10 in accordance with the position of the to-be-machined region 13 in accordance with the program stored in the storage device 92.

[0108] Furthermore, the controller 88 adjusts the position of the cutting unit 44 in the direction along the Y axis and the position (height) in the direction along the Z axis by the cutting unit moving mechanism 26 in accordance with the position of the to-be-processed region 13 in accordance with the program stored in the storage device 92. Furthermore, the controller 88 rotates the cutting blade 52 in accordance with the program stored in the storage device 92.

[0109] Thereafter, the controller 88 moves the chuck table 16 along the X-axis by the chuck table moving mechanism 10 while rotating the cutting blade 52 in accordance with the program stored in the storage device 92. As a result, the chuck table 16 and the cutting blade 52 move relatively along the X-axis (machining feed), and the workpiece 11 is machined in the intended machining area 13.

[0110] As described above, the cutting device (machining system) 2 according to this embodiment includes a trained model 94 configured by machine learning so as to output information relating to the position of the linear to-be-machined region 13 set on the workpiece 11 when the first image 21 showing the workpiece 11 is input. Therefore, according to the cutting device 2 according to this embodiment, information relating to the position of the to-be-machined region 13 and the like can be obtained even if a key pattern is not set.

[0111] There is no limitation on the method of mounting the trained model 94 on the cutting device 2. For example, the manufacturer of the cutting device 2 can generate the trained model 94, incorporate it into the cutting device 2, and then provide the cutting device 2 to a user. In addition, the manufacturer of the cutting device 2 can provide the trained model 94 by a method such as distribution to an existing cutting device (a cutting device not mounted with the trained model 94) used by the user, thereby providing the cutting device 2 mounted with the trained model 94 to the user after the fact.

[0112] 11 is a diagram showing a trained model providing system 100 that provides a trained model 94 to an existing cutting device 2A. The trained model providing system 100 includes a distribution unit 102 that distributes the trained model 94, and a network 104 that connects the distribution unit 102 to multiple cutting devices 2A used by users.

[0113] The distribution unit 102 corresponds to a server or the like managed by the manufacturer of the cutting device 2, and is connected to the network 104 via a wired or wireless connection. The cutting device 2A is an existing cutting device used by a user, and does not have a built-in trained model 94. The structure of the cutting device 2A is the same as that of the cutting device 2 (see FIG. 1), except for the presence or absence of the trained model 94. The controller 88 of the cutting device 2A is connected to the network 104 via a wired or wireless connection.

[0114] The manufacturer of the cutting device 2 manages the distribution unit 102 and distributes the trained model 94 to the cutting device 2A from the distribution unit 102 via the network 104. Specifically, the manufacturer of the cutting device 2 first generates the trained model 94 by machine learning and creates a program 106 including the trained model 94. This program 106 is configured to allow a computer to execute, for example, a process of inputting a first image 21 including the workpiece 11 into the trained model 94 and a process related to the calculation of the trained model 94.

[0115] Then, in response to a user's request, the manufacturer of the cutting device 2 supplies the program 106 to the cutting device 2A via the network 104. The program 106 distributed by the distribution unit 102 is received by the cutting device 2A and stored in the storage device 92 provided in the controller 88 of the cutting device 2A. This completes the cutting device 2 equipped with the trained model 94. Thereafter, the controller 88 reads out and executes the program 106 at a predetermined timing to obtain information regarding the position of the to-be-processed region 13, etc.

[0116] The manufacturer of the cutting device 2 can also provide the program 106 using the recording medium 108. Specifically, the manufacturer of the cutting device 2 provides the recording medium 108 in which the program 106 is stored to the user of the cutting device 2A. The recording medium 108 is a non-transitory recording medium readable by a computer, such as an optical disk or a flash memory. The user of the cutting device 2A then causes the controller 88 to read the program 106 stored in the recording medium 108 and store it in the storage device 92.

[0117] As described above, by distributing the program 106 using the network 104 or the recording medium 108, the trained model 94 can be introduced afterwards into the cutting device 2A used by the user. Furthermore, when the trained model 94 is updated, the manufacturer of the cutting device 2 can distribute the updated trained model 94, a program including the updated trained model 94, or data equivalent to the parameters of the updated trained model 94, using the network 104 or the recording medium 108. This provides the cutting device 2 equipped with the latest trained model 94 to the user.

[0118] Next, a specific example of a method for generating a trained model according to this embodiment will be described. The trained model 94 is generated, for example, by performing supervised learning using a first training image obtained by imaging the workpiece 11 and a second training image obtained by processing the first training image. In this embodiment, the trained model 94 is generated using the above-mentioned cutting device 2, but the trained model 94 may be generated using other devices, computers, etc.

[0119] 12 is a functional block diagram showing the functional structure of the controller 88 when generating the trained model 94. When generating the trained model 94, first, a sample (image acquisition workpiece, learning workpiece) 31 corresponding to the workpiece 11 is imaged to acquire a first learning image 41 including the sample corresponding to the workpiece 11 (first learning image acquisition step).

[0120] The sample 31 is a workpiece having the same or similar structure as the workpiece 11 to be processed by the cutting device 2. For example, the sample 31 is a disk-shaped wafer made of the same material as the workpiece 11, and has a generally flat front surface (first surface) 31a and a back surface (second surface) 31b facing the opposite side to the front surface 31a.

[0121] A front surface 31a side of the sample 31 is partitioned into a plurality of rectangular regions by a plurality of linear processing regions (streets) 33 (see FIG. 13) that are set to intersect with each other. Devices 35 (see FIG. 13), such as ICs (Integrated Circuits), LEDs (Light Emitting Diodes), and MEMS (Micro Electro Mechanical Systems), are formed in each of the regions partitioned by the processing regions 33.

[0122] The structure, dimensions, arrangement, etc. of the device 35 are similar to those of the device 15 of the workpiece 11. That is, structures such as electrodes, terminals, wiring, etc. that constitute the device 35 are provided on the front surface 31a side of the sample 31, and an optically identifiable pattern (design) is formed by the arrangement of these structures. However, the structure, dimensions, arrangement, etc. of the device 35 may be different from those of the device 15 of the workpiece 11.

[0123] The sample 31, like the workpiece 11, is placed inside the opening of an annular frame 37 and is supported by the frame 37 via a tape 39. The frame 37 and tape 39 used for the sample 31 may be the same as the frame 17 and tape 19 used for the workpiece 11. The sample 31 is accommodated in the cassette 8 while being supported by the frame 37.

[0124] When acquiring the first learning image 41, first, the controller 88, in accordance with the program stored in the storage device 92, transports the sample 31 contained in the cassette 8 to the chuck table 16 by the transport mechanism, and holds the sample 31 on the chuck table 16. Note that the operation of the cutting device 2 when the sample 31 is transported and held is similar to the operation of the cutting device 2 when the workpiece 11 is transported and held.

[0125] Next, the first learning image acquisition unit 88g of the controller 88 images the front surface 31a side of the sample 31 with the camera 74, and acquires the first learning image 41. Specifically, the controller 88 performs imaging multiple times under different conditions in accordance with a program stored in the storage device 92, and acquires multiple first learning images 41 in which the sample 31 appears. For example, the controller 88 performs imaging multiple times with the camera 74 while moving the chuck table 16 along the X-axis with the chuck table moving mechanism 10.

[0126] Moreover, the controller 88 captures images multiple times with the camera 74 while, for example, moving the camera 74 along the Y-axis with the cutting unit moving mechanism 26. Furthermore, the controller 88 captures images multiple times with the camera 74 while, for example, rotating the chuck table 16 with a rotary drive source (rotation mechanism).

[0127] The controller 88 may capture images multiple times using the camera 74 while moving the camera 74 along the Z-axis using the cutting unit moving mechanism 26. In this case, multiple first learning images 41 are obtained that have different conditions, such as the range captured by the camera 74 and the position of the focal point. Similarly, the controller 88 may capture images multiple times using the camera 74 while illuminating the sample 31. In this case, the controller 88 may capture images multiple times using the camera 74 while changing the lighting conditions.

[0128] By such a procedure, different regions in the sample 31 are imaged, and a plurality of different first training images 41 are acquired. Note that there is no limit to the number of first training images 41 acquired by such a procedure, but it is considered that the accuracy of estimation of the trained model 94 increases as the number of appropriate first training images 41 increases. Therefore, for example, it is desirable to acquire 1000 or more first training images 41. The acquired first training images 41 are stored in the storage device 92 of the controller 88.

[0129] 13(A), 13(B), 13(C), and 13(D) are diagrams showing examples of the first learning image 41. For ease of explanation, the same reference numerals as those of the sample 31 are given to the portions showing the components of the sample 31 in each of the first learning images 41. Both the region to be processed 33 and the device 35 are shown in the first learning images 41 shown in FIG. 13(A), 13(B), and 13(C).

[0130] Specifically, the first learning image 41 shown in Fig. 13(A) shows one planned processing area 33 extending in a lateral direction. The first learning image 41 shown in Fig. 13(B) shows two planned processing areas 33 that intersect with each other. The first learning image 41 shown in Fig. 13(C) shows one planned processing area 33 extending in a direction tilted from the lateral direction. On the other hand, the first learning image 41 shown in Fig. 13(d) does not show a planned processing area 33.

[0131] After the first learning image 41 is obtained, a second learning image 43 in which the position, etc. of the to-be-processed region 13 is specified is obtained by specifying the position, etc. of the to-be-processed region 13 in the first learning image 41 (second learning image obtaining step). That is, the first learning image 41 is processed so as to specify the position, etc. of the to-be-processed region 13, and the second learning image 43 is obtained.

[0132] The second learning image 43 is obtained, for example, by replacing pixels in a portion of the first learning image 41 corresponding to the intended processing region 13 with pixels of a first luminance, and by replacing pixels in a portion of the first learning image 41 corresponding to the device 15 with pixels of a second luminance different from the first luminance. In this case, the second learning image 43 is a binary image.

[0133] Specifically, for example, when the operator designates a portion in the first learning image 41 that corresponds to the to-be-processed region 13, the controller 88 operates according to a program stored in the storage device 92 to process the first learning image 41 into the second learning image 43. There is no limitation on the specific method for processing the first learning image 41 into the second learning image 43. For example, the controller 88 may be responsible for part or all of the work related to the operator's designation by operating according to a program stored in the storage device 92.

[0134] All of the first learning images 41 are processed in the same manner, thereby acquiring the same number of second learning images 43 as the first learning images 41. The acquired second learning images 43 are associated with the first learning images 41 before processing, and then stored in the storage device 92 of the controller 88. That is, one first learning image 41 and one second learning image 43 obtained from this first learning image 41 are stored in the storage device 92 as a pair.

[0135] Figures 14(A), 14(B), 14(C), and 14(D) are all diagrams showing examples of the second learning image 43. In Figures 14(A), 14(B), 14(C), and 14(D), a region made up of pixels of the first luminance is represented as a first region 23a, and a region made up of pixels of the second luminance is represented as a second region 23b.

[0136] The second learning image 43 shown in Fig. 14(A) is obtained by processing the first learning image 41 shown in Fig. 13(A) and has a first region 23a extending in the lateral direction. The second learning image 43 shown in Fig. 14(B) is obtained by processing the first learning image 41 shown in Fig. 13(B) and has a cross-shaped first region 23a.

[0137] Furthermore, the second learning image 43 shown in Fig. 14(C) is obtained by processing the first learning image 41 shown in Fig. 13(C) and has a first region 23a extending in a direction tilted from the straight-ahead direction. The second learning image 43 shown in Fig. 14(D) is obtained by processing the first learning image 41 shown in Fig. 13(D) and does not have the first region 23a.

[0138] After the pair of the first learning image 41 and the second learning image 43 is obtained, machine learning is performed using the first learning image 41 and the second learning image 43. As a result, when the first image 21 including the workpiece 11 is input, a trained model is generated that outputs the second image 23 including information regarding the position of the linear planned processing region 13 set on the workpiece 11 (training step).

[0139] In the learning step, for example, supervised learning is performed using a pair of the first learning image 41 and the second learning image 43 as teacher data. Specifically, the first learning image 41 is input as input image data, and the second learning image 43 is input as a correct label (correct output image data) to the neural network NN.

[0140] Then, the parameters (neuron weights, biases, etc.) of the neural network NN are updated so as to reduce the difference between the output of the neural network NN and the second learning image 43. As a result, the neural network NN is reconfigured so as to output, from the output layer 94b, the second image 23 including information on the position of the linear to-be-processed region 13 set on the workpiece 11 when the first image 21 including the workpiece 11 is input to the input layer 94a. As the learning algorithm, for example, the error backpropagation method is used.

[0141] The trained neural network NN is stored in the storage device 92 of the controller 88. As a result, a cutting device 2 equipped with a trained model 94 is obtained. The calculation for training the neural network NN may be executed by another device, a computer, or the like. In this case, the trained neural network NN, a program including the trained neural network NN, or data corresponding to parameters of the trained neural network NN are input to the controller 88 from the outside and stored in the storage device 92.

[0142] The present invention is not limited to the above-described embodiment, and may be modified in various ways. For example, in the above-described embodiment, the position determination unit 88c of the controller 88 performs image processing or the like on the second image 23 to determine the position and orientation of the processing target area 13. However, the operator may determine the position and orientation of the processing target area 13 from the second image 23. Information on the position and orientation of the processing target area 13 determined by the operator is stored in the storage device 92 of the controller 88.

[0143] In this case, the display control unit 88d of the controller 88 may display the second image 23 superimposed on the first image 21. This makes it easier for the operator to determine the position and orientation of the intended processing region 13 from the second image 23. However, there is no limitation on the type of image displayed on the touch screen 84.

[0144] For example, as in the above-described embodiment, any multiple images may be displayed so as to be switched at a predetermined time, any multiple images may be displayed side by side at the same time, or any multiple images may be displayed overlapping each other. When any multiple images are displayed overlapping each other, for example, all images except one of the images are processed to be semi-transparent and overlapped. Furthermore, lines or the like indicating the position or orientation of the processing target area 13 determined by the operator may be displayed overlapping these images.

[0145] In the above-described embodiment, the cutting device 2 is exemplified as a processing system equipped with the trained model 94, but the trained model 94 may be mounted on another processing device (processing system). For example, the trained model 94 may be mounted on a laser processing device (processing system) that processes the workpiece 11 by irradiating it with a laser beam.

[0146] The laser processing device includes, for example, a laser irradiation unit (processing unit) for irradiating a laser beam instead of the cutting unit 44 of the cutting device 2. This laser irradiation unit includes a laser oscillator including a laser medium such as Nd:YAG suitable for laser oscillation, and an optical system for guiding the laser beam generated by the laser oscillator to the workpiece 11. The optical system is composed of optical elements such as lenses and mirrors.

[0147] The controller of the laser processing device is configured, for example, similarly to the controller 88 of the cutting device 2, and the learned model 94 is stored in the memory device 92 of this controller. This provides a laser processing device that can obtain information regarding the position of the intended processing region 13 even if a key pattern is not set.

[0148] Furthermore, the processing system equipped with the trained model 94 may be configured to acquire information on the position of the linear planned processing area 13, etc., outside a processing device such as the cutting device 2. FIG. 15 is a diagram showing a processing system 200 according to a modified example. The processing system 200 according to the modified example includes, in addition to one or more cutting devices 2B, a controller (second controller) 202 that provides information on the position of the planned processing area 13, etc., to each cutting device 2B. The cutting devices 2B and the controller 202 are connected via a network 204.

[0149] Here, unlike the cutting device 2 according to the embodiment described above, each cutting device 2B does not have a built-in trained model 94. However, the structure of the cutting device 2B is similar to that of the cutting device 2 (see FIG. 1) except for the presence or absence of the trained model 94. In this modification, the controller (first controller) 206 included in the cutting device 2B corresponds to the controller 88 of the cutting device 2.

[0150] The controllers 202 and 206 have the same structure as the controller 88 of the cutting device 2. Specifically, the controllers 202 and 206 are each configured by a computer including a processing device and a storage device, and the necessary functions are realized by the processing device operating in accordance with a program (software) stored in the storage device. As shown in Fig. 15, the controllers 202 and 206 are connected to a network 204 by wire or wirelessly.

[0151] The controller 202 for providing each cutting device 2B with information on the position of the to-be-processed region 13 and the like has at least the function of the second image acquisition unit 88b associated with the controller 88 of the cutting device 2 and a function for transmitting and receiving information. That is, the controller 202 has a trained model 94 built in, together with a program (second program) for realizing the function of the controller 202.

[0152] On the other hand, the controller 206 of the cutting device 2B has at least the functions of a first image acquisition unit 88a, a display control unit 88d, a positional relationship adjustment unit 88e, and a processing control unit 88f associated with the controller 88 of the cutting device 2, and a function for transmitting and receiving information. That is, the controller 206 has a program (first program) for realizing the functions of the controller 206 built in.

[0153] In the processing system 200, the controller 202 has the function of the position determination unit 88c related to the controller 88 of the cutting device 2. Meanwhile, the controller 206 may have the function of the position determination unit 88c instead of the controller 202. In other words, it is sufficient that one or both of the controller 202 and the controller 206 have the function of the position determination unit 88c.

[0154] When cutting the workpiece 11, the controller 206 executes a procedure for causing the camera 74 to capture an image of the workpiece 11 held on the chuck table 16 to obtain a first image 21. Specifically, the controller 206 adjusts the positional relationship between the chuck table 16 and the camera 74 in accordance with a program stored in its storage device, and causes the camera 74 to capture an image of a partial area of ​​the surface 11a of the workpiece 11.

[0155] When the first image 21 is acquired by the first image acquisition unit 88a, the controller 206 executes a procedure for sending the first image 21 to the controller 202 in accordance with the program stored in the storage device. Then, the controller 202 executes a procedure for receiving the first image 21 from the controller 206 in accordance with the program stored in the storage device.

[0156] Thereafter, the controller 202 executes a procedure for acquiring information on the position of the to-be-processed region 13, etc., from the first image 21. Specifically, the controller 202 inputs the first image 21 to the trained model 94 in accordance with a program stored in the storage device, and executes processing related to the calculation of the trained model 94. As a result, the controller 202 obtains the second image 23 including information on the position of the to-be-processed region 13, etc.

[0157] When the second image 23 is acquired, for example, the controller 202 executes a procedure for determining the position, etc. of the to-be-processed region 13 from the second image 23 in accordance with a program stored in the storage device. Note that, as described above, the procedure for determining the position, etc. of the to-be-processed region 13 may be performed by the controller 206. Also, in cases where the position or orientation of the to-be-processed region 13 is indicated with sufficiently high accuracy by the second image 23, part or all of this procedure may be omitted.

[0158] Next, the controller 202 executes a procedure for sending information on the position of the to-be-processed region 13, etc. to the controller 206, according to the program stored in the storage device. Then, the controller 206 executes a procedure for receiving information on the position of the to-be-processed region 13, etc. from the controller 202, according to the program stored in the storage device.

[0159] The controller 206, which has received the information regarding the position and the like, executes a procedure for displaying, for example, information regarding the position and orientation of the to-be-processed region 13 on the touch screen 84, according to the program stored in the storage device. The controller 206 also executes a procedure for adjusting the relationship between the position and orientation of the chuck table 16 that holds the workpiece 11 and each cutting unit 44, according to the program stored in the storage device. The controller 206 then executes a procedure for causing the cutting units 44 to cut the position of the to-be-processed region 13, according to the program stored in the storage device.

[0160] In addition, the structures, methods, and the like according to the above-described embodiments and modifications may be implemented without departing from the scope of the present invention. [Explanation of symbols]

[0161] 2: Cutting equipment (processing system) 2A:Cutting device 2B:Cutting device 4: Base 4a: Recess 4b: Recess 4c: Recess 6: Cassette table 8: Cassette 10: Chuck table movement mechanism (processing feed mechanism) 12: Table cover 14: Dust-proof and water-resistant cover 16: Chuck table 16a:Top surface 18:Frame body 20: Holding plate 22: Clamp 24:Support structure 26: Cutting unit movement mechanism (indexing feed mechanism, cutting feed mechanism) 28: Y-axis guide rail 30: Y-axis moving plate 32: Screw shaft 34: Rotation drive source 36: Z-axis guide rail 38: Z-axis moving plate 40: Screw shaft 42: Rotation drive source 44: Cutting unit (processing unit) 46: Spindle housing 48: Spindle 48a: Screw hole 50: Blade mounter 52: Cutting blade 52a: Opening 54: Mount flange 56: Fixed mount 56a: Through hole 56b: Receiving part 58: Pressing flange 58a: Curved surface 58b:Flat surface 58c: opening 60: Flange part 60a: Surface 60b: Back side 60c: Convex 60d: Tip surface 62: Boss section 62a: Tip 62b: Outer surface 64: Washer 66: Screw 68: Blade fixing device 68a: Opening 70: Blade cover 72: Nozzle 74: Camera (imaging unit) 76: Cleaning unit 78: Spinner table 80: Nozzle 82: Equipment cover 84: Touch screen (input device, output device) 86: Indicator light (output device) 88: Controller (control unit) 88a: First image acquisition unit 88b: Second image acquisition unit 88c:Positioning section 88d: Display control section 88e: Position adjustment section 88f: Processing control section 88g: First learning image acquisition unit 90: Processing equipment 92: Storage device 94: Model 94a: Input layer 94b: Output layer 94c: Hidden layer 100: Trained model provision system 102: Distribution Department 104: Network 106: Program 108: Recording media 200: Processing system 202: Controller (Second controller) 204: Network 206: Controller (first controller) 11: Workpiece 11a: Surface (first side) 11b: Back side (second side) 11c: Processing marks (kerf) 13: Area to be processed (street) 15: Device 17: Frame 17a:Aperture 19: Tape 21: First image 23: 2nd image 23a: 1st area 23b:Second area 31: Sample (workpiece for image acquisition, workpiece for learning) 31a: Surface (first side) 31b: Back side (second side) 33: Area to be processed (street) 35 :Device 37: Frame 39: Tape 41: First learning image 43: Second learning image

Claims

1. 1. A processing system for processing a workpiece, comprising: A chuck table for holding the workpiece; a processing unit that processes the workpiece held on the chuck table; an imaging unit for imaging the workpiece held on the chuck table; a controller having a processing unit and a storage unit; The storage device stores a trained model configured by machine learning so as to output information regarding the position of a linear planned processing area set on the workpiece when a first image showing the workpiece is input, and a program; The controller performs the following steps according to the program and the trained model: a step of imaging the workpiece held on the chuck table with the imaging unit to obtain the first image; obtaining information on a position of the intended processing area from the first image; A processing system that executes a procedure for causing the processing unit to process the intended processing area based on the information.

2. The processing system of claim 1 , wherein the trained model is a neural network including an input layer to which the first image is input and an output layer to which the information regarding the position of the area to be processed is output.

3. The processing system comprises: a processing feed mechanism that relatively moves the processing unit and the chuck table in a processing feed direction; a rotation mechanism for rotating the chuck table, The controller, according to the program, 2. The machining system according to claim 1, further comprising the step of rotating the chuck table based on the information so that the direction of the intended machining area becomes parallel to the machining feed direction.

4. The information is a second image showing an approximate location of the intended processing area, The controller, according to the program, A step of detecting an outer edge of the intended processing area by performing image processing on the second image; A step of calculating a center position in a width direction of the intended processing area from the detected outer edge; The machining system according to claim 1 , further comprising a step of setting the central position as a position of the intended machining area to be machined by the machining unit.

5. 1. A processing system for processing a workpiece, comprising: A chuck table for holding the workpiece; a processing unit that processes the workpiece held on the chuck table; an imaging unit for imaging the workpiece held on the chuck table; a first controller and a second controller, each having a processing unit and a memory unit; A first program is stored in the storage device of the first controller, The storage device of the second controller stores a trained model configured by machine learning so as to output information regarding a position of a linear planned processing area set on the workpiece when a first image showing the workpiece is input, and a second program; The first controller, according to the first program, a step of imaging the workpiece held on the chuck table with the imaging unit to obtain the first image; sending the first image to the second controller; receiving the information on the position of the intended processing region or information based thereon from the second controller; and executing a procedure of causing the machining unit to machine the intended machining area based on information received from the second controller; The second controller performs the following steps according to the second program and the trained model: receiving the first image from the first controller; obtaining information on a position of the intended processing area from the first image; and sending the information relating to the position of the to-be-machined region or information based thereon to the first controller.

6. The processing system of claim 5 , wherein the trained model is a neural network including an input layer to which the first image is input and an output layer to which the information regarding the position of the area to be processed is output.

7. The processing system comprises: a processing feed mechanism that relatively moves the processing unit and the chuck table in a processing feed direction; a rotation mechanism for rotating the chuck table, The first controller, according to the first program, 6. The machining system according to claim 5, further comprising the step of rotating the chuck table based on the information so that the direction of the intended machining area becomes parallel to the machining feed direction.

8. The information is a second image showing an approximate location of the intended processing area, The first controller, according to the first program, A step of detecting an outer edge of the intended processing area by performing image processing on the second image; A step of calculating a center position in a width direction of the intended processing area from the detected outer edge; The machining system according to claim 5 , further comprising a step of setting the central position as a position of the intended machining area to be machined by the machining unit.

9. A program for obtaining information regarding a position of a linear processing area set on a workpiece, A trained model configured by machine learning is configured to output information regarding the position of the linear processing area set on the workpiece when a first image showing the workpiece is input, inputting the first image into the trained model; A program for causing a computer to execute the steps of executing the calculations of the trained model.

10. The program according to claim 9 , wherein the trained model is a neural network including an input layer to which the first image is input and an output layer to which the information regarding the position of the area to be processed is output.

11. A non-transitory recording medium storing the program according to claim 9 or 10.

12. A method for generating a trained model for generating a trained model that outputs information regarding a position of a linear planned processing area set on a workpiece when an image of the workpiece is input, comprising: a first learning image acquisition step of acquiring a first learning image including an image of a sample corresponding to the workpiece by imaging the sample; a second learning image acquisition step of acquiring a second learning image in which the position of the planned processing area is designated by designating the position of the planned processing area in the first learning image; a learning step of generating a trained model by machine learning using the first learning image and the second learning image, which generates a trained model that outputs information regarding the position of the linear processing area set on the workpiece when an image of the workpiece is input.

Images