Semiconductor manufacturing device

Abstract

This semiconductor manufacturing device (1) comprises: a cutting unit (2) that cuts a substrate having a scribing line by applying an external force to the scribing line of the substrate; an imaging unit (3) that images a cutting groove formed on the substrate; and a determination unit (47) that processes data (100) of a captured image of the cutting groove imaged by the imaging unit (3) and determines whether the captured image is a normal processing image or an abnormal processing image, the data (100) being processed by a trained program (300) trained on teacher data (200) including a normal processing image group, composed of a plurality of images of normal cutting grooves, and an abnormal processing image group, composed of a plurality of images of cutting grooves (C) having cutting failures.

Description

Semiconductor manufacturing equipment

[0001] The present invention relates to a semiconductor manufacturing apparatus, and more particularly to a breaking apparatus that cuts a substrate (semiconductor wafer) on which a scribe line is formed by applying an external force along the scribe line.

[0002] In the semiconductor device manufacturing process, a single substrate (semiconductor wafer) is divided into multiple chip regions, various circuits are formed in each region, and then the substrate is cut into individual chips. Generally, the substrate cutting process begins by forming grooves (cracks) called scribe lines extending in the thickness direction of the substrate along the intended cutting line on the substrate surface using a cutter wheel or laser (scribing process). The substrate is then cut by applying an external force along the scribe lines from the side opposite the scribe line (breaking process). In many cases, the substrate to be cut is attached and fixed to a substrate holding tape stretched over a ring-shaped holder.

[0003] For example, Patent Document 1 discloses a breaking device that breaks a substrate that has been attached and fixed to a substrate holding tape. In the breaking device disclosed in Patent Document 1, the substrate is placed with the surface opposite to the surface on which the scribe line is formed (the surface attached to the substrate holding tape) facing up, so that the scribe line is located between a pair of left and right receiving blades provided in the middle of the work stage. A breaking bar is then pressed down from above toward the scribe line, bending the substrate and cutting it. Because the substrate is fixed to the substrate holding tape, the individual chips do not separate, and a cutting groove is formed in the substrate along the scribe line.

[0004] Japanese Patent Publication No. 2020-181931

[0005] The kerfs formed by the above processes may have cutting defects such as cracks and chipping. In particular, silicon carbide (SiC) substrates, which have recently attracted attention as a material for next-generation power devices, are very hard and brittle, and therefore are prone to cracks and chipping when the substrate is cut. Since such cutting defects have a negative impact on the quality of semiconductor devices, after cutting the substrate, the substrate is sometimes removed from the semiconductor manufacturing equipment and checked for cutting defects. This checking process is usually performed by an operator using a microscope or the like to directly observe the kerfs formed in the substrate. Therefore, depending on the operator's level of skill, cutting defects may be overlooked.

[0006] The problem to be solved by the present invention is to stabilize the recognition of cutting defects that occur when cutting a substrate, and to further improve the recognition accuracy.

[0007] The semiconductor manufacturing apparatus of the present invention, which has been made to solve the above-mentioned problems, comprises a cutting processing unit that cuts a substrate by applying an external force to a scribe line formed on the substrate, an imaging unit that images a cut groove formed on the substrate, and a determination unit that processes data of the image of the cut groove imaged by the imaging unit using a trained program that has learned teacher data including a normally processed image group consisting of a plurality of images of normal cut grooves and an abnormal processed image group consisting of a plurality of images of cut grooves with cutting defects, and determines whether the image belongs to a normally processed image or an abnormal processed image.

[0008] In the semiconductor manufacturing apparatus having the above configuration, the determining unit determines whether the captured image of the kerf is a normal kerf (normally processed image) or a kerf with a cutting defect (abnormally processed image) using a trained program that has been trained in advance based on a group of normally processed images and a group of abnormally processed images. This allows a mechanical (objective) determination of whether the kerf formed in the substrate cut by the cutting unit is normal or abnormal, ensuring stable recognition of cutting defects regardless of the operator's level of proficiency. Furthermore, by increasing the number of images in the group of normally processed images and the group of abnormally processed images included in the training data, the accuracy of recognition of cutting defects can be improved.

[0009] In the semiconductor manufacturing apparatus according to the present invention, the abnormality processing image group may be classified into a plurality of image groups according to the abnormal state of the cutting groove, and the determination unit may be configured to determine to which of the plurality of abnormal states the captured image belongs.

[0010] This allows you to know not only the binary value of normal or abnormal, but also the details of the abnormality (degree and type of disconnection failure).

[0011] In the semiconductor manufacturing apparatus according to the present invention, the judgment unit may calculate the similarity between the captured image and the group of normally processed images and the group of abnormally processed images, respectively, and if the difference between the similarity between the captured image and the group of normally processed images and the similarity between the captured image and the group of abnormally processed images is equal to or less than a predetermined value, the judgment may be stopped and a process may be performed to allow a user to judge whether the captured image belongs to a normally processed image or an abnormally processed image.

[0012] This allows an experienced worker to determine the state of the cutting groove when it is difficult for the determining unit to determine whether the cutting groove is normal or abnormal.

[0013] In the semiconductor manufacturing apparatus, the trained program may be a program trained by adding the captured image, which the user has determined to be either the normally processed image or the abnormally processed image, to training data.

[0014] This allows the results of judgment made by a skilled worker on captured images that were difficult for the judgment unit to judge as normal / abnormal to be reflected in the trained program, thereby improving the accuracy of judgment made by the judgment unit.

[0015] The semiconductor manufacturing apparatus according to the present invention may further include a control unit that controls the cutting unit based on the determination result of the determination unit or the determination result of the user.

[0016] For example, when the control unit determines that the captured image belongs to the abnormal processing image, the control unit may control the cutting unit to stop the cutting process of the substrate. Also, the control unit may control the cutting unit to change the setting conditions for cutting the substrate depending on the abnormal state of the cutting groove.

[0017] According to the present invention, it is possible to stabilize and further improve the accuracy of recognizing processing defects that occur when cutting a substrate.

[0018] 1 is a schematic diagram showing the basic configuration of a breaking device, which is a semiconductor manufacturing device according to an embodiment of the present invention; FIG. 2 is a diagram showing an example of a substrate to be cut by the breaking device, in which (a) is a perspective view showing a state in which the substrate is fixed to a substrate holding tape, and (b) is a cross-sectional view showing a state in which the open top surface of the substrate shown in (a) is covered with a film; FIG. 3 is a diagram for explaining a substrate cutting process; FIG. 4 is a diagram for explaining an example of a trained program; A flowchart showing an example of a processing procedure related to generation of a trained program; A flowchart showing an example of a processing procedure related to determination of a cutting groove; A schematic diagram showing the overall configuration of an example of a semiconductor manufacturing system having the breaking device.

[0019] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0020] 2 is a diagram illustrating an example of a substrate to be cut by the semiconductor manufacturing apparatus according to this embodiment. As shown in FIG. 2, a wafer W to be cut in this embodiment is attached and fixed to a dicing tape T stretched over a ring-shaped wafer ring R. The dicing tape T is made of a stretchable resin, and its upper surface is formed as an adhesive surface. The lower surface W1 of the wafer W is attached to this adhesive surface, and the wafer W is fixed to the dicing tape T.

[0021] A plurality of scribe lines S arranged in a grid pattern are formed on the upper surface W2 of the wafer W in a preceding scribing process. The scribe lines are formed by rolling the cutting edge of a cutter wheel against the surface of the wafer W while pressing it against the surface, or by irradiating the surface of the wafer W with a laser beam. The scribe lines S may be formed on either the front surface (the surface on which the circuit is formed) or the back surface of the wafer W. The upper surface W2 of the wafer W is covered and protected by a film F, as shown in FIG. 2(b).

[0022] 1 is a schematic diagram showing the basic configuration of a semiconductor manufacturing apparatus according to this embodiment. The semiconductor manufacturing apparatus (breaking apparatus) 1 according to this embodiment includes a cutting unit 2, an imaging unit 3, a data processing unit 4, a control unit 5, an input unit 6, and a display unit 7.

[0023] The cutting unit 2 cuts the wafer W by applying an external force along a scribe line S formed on the wafer W. FIG. 3 is a diagram for explaining the cutting process of the wafer W by the cutting unit 2. The cutting unit 2 mainly comprises a table 21, a pair of receiving blades 22, 22 disposed midway between the table 21, and a break bar 23 disposed above the pair of receiving blades 22, 22. In the cutting process by the cutting unit 2, the wafer W is first placed on the table 21 with the surface W2 on which the scribe line S is formed facing downward, and the table 21 is moved to position the wafer W so that the scribe line S is located between the pair of receiving blades 22, 22. At this time, the receiving blades 22, 22 support the surface W2 of the wafer W via a film F from both sides of the scribe line S. Next, the break bar 23 is pressed down from above toward the scribe line S, thereby bending the wafer W between the receiving blades 22, 22 and causing the cracks in the scribe line S to penetrate in the thickness direction of the wafer W, thereby cutting the wafer W.

[0024] By performing the same operations as above while moving and rotating the table 21, the wafer W is cut along the scribe lines S arranged in a grid pattern. After cutting, the wafer W is divided into several cut pieces and held by the dicing tape T, and cut grooves C are formed between the cut pieces. Figure 3 shows the state in which the above cutting process is performed on the scribe line S located on the left side of the break bar 23, and cut grooves C are formed. A known mechanism can be used as the cutting processing unit 2.

[0025] The imaging unit 3 captures an image of the kerf C formed in the wafer W. For example, the imaging unit 3 is disposed next to the break bar 23 and captures an image of the kerf C formed in the wafer W from the surface W1 side via the dicing tape T. Alternatively, the imaging unit 3 is disposed between a pair of receiving blades 22 and captures an image of the kerf C formed in the wafer W from the surface W2 side via a film F. The imaging unit 3 is not particularly limited as long as it can clearly capture and distinguish the state of the line. For example, the imaging unit 3 may be a digital microscope equipped with an image sensor such as a CCD (Charge Coupled Device) or a CMOS (Complementary Metal-Oxide-Semiconductor). Alternatively, the imaging unit 3 may be a monochrome camera with a macro lens attached that can be connected to the control unit 5 via USB or the like.

[0026] The data processing unit 4 receives image data of the kerf C captured by the imaging unit 3 and processes it according to a predetermined program. The control unit 5 mainly performs overall control of the semiconductor manufacturing apparatus 1 and control of the user interface via the input unit 6 and display unit 7. Note that the data processing unit 4 and control unit 5 can be configured to realize their respective functions by using, for example, a personal computer as a hardware resource and executing dedicated processing and control software pre-installed on the computer. In this case, the input unit 6 is a keyboard and pointing device (such as a mouse) attached to the computer, and the display unit 7 is a display monitor of the computer.

[0027] The data processing unit 4 includes, as functional blocks, an image data acquisition unit 41, an image data storage unit 42, a teacher data creation unit 43, a teacher data storage unit 44, a machine learning unit 45, a learned program storage unit 46, and a judgment unit 47.

[0028] The image data acquisition unit 41 acquires image data of the cut groove C and stores it in the image data storage unit 42. The image data acquisition unit 41 acquires captured image data 100 of the cut groove C formed by the cutting processing unit 2 from the imaging unit 3 and stores it in the image data storage unit 42.

[0029] The teacher data creation unit 43 creates teacher data 200 for determining whether the state of the kerf C formed by the cutting unit 2 is normal or abnormal. The teacher data 200 is data having image data of the kerf C and information (label) that identifiably indicates whether the state of the kerf C is normal or abnormal. The image data of the kerf C may be captured image data 100 of the kerf C formed by the cutting unit 2 of the semiconductor manufacturing apparatus 1, or image data acquired by the image data acquisition unit 41 from outside the semiconductor manufacturing apparatus 1. The outside of the semiconductor manufacturing apparatus 1 includes, for example, a different semiconductor manufacturing apparatus 1 or a communicable device such as an external server.

[0030] The teacher data creation unit 43 labels multiple images of the kerf groove C as to whether the kerf groove C is normal or abnormal, and stores the image data to which the labeled information (labels) is added in the teacher data storage unit 44 as teacher data 200. For example, the teacher data creation unit 43 has an operator determine whether the kerf groove C shown in the image is normal or abnormal, and links information indicating the determination result to the image data. For example, the teacher data creation unit 43 determines whether the kerf groove C is normal or abnormal based on features of the image of the kerf groove C, determination criteria for the features, etc., and links information indicating the determination result to the image data.

[0031] The teacher data creation unit 43 can further hierarchically label the state of the kerf groove C and link the labeled information, etc. to the image data. For example, if the teacher data creation unit 43 labels the kerf groove C as "normal," it further hierarchically labels the quality level of the kerf groove C. For example, if the teacher data creation unit 43 labels the kerf groove C as "abnormal," it further hierarchically labels the details of the abnormal state of the kerf groove C.

[0032] The machine learning unit 45 generates a trained program 300 by performing machine learning using the teacher data 200, and stores the generated trained program 300 in the trained program storage unit 46. The trained program 300 is a program that has learned multiple pieces of teacher data 200, including normally processed images in which the cutting groove C is labeled as "normal" and abnormally processed images in which the cutting groove C is labeled as "abnormal."

[0033] FIG. 4 is a diagram illustrating an example of a trained program 300 in this embodiment. In FIG. 4, the trained program 300 is composed of a program (machine learning data) that has trained six pieces of training data 200 (201, 202, 203, 204, 205, 206). The training data 201, 202, and 203 show an example of determining that the state of the kerf groove C is normal, and are examples of normal processing images. The training data 204, 205, and 206 show an example of determining that the state of the kerf groove C is abnormal, and are examples of abnormal processing images. The normal state of the kerf groove C means that the state of the kerf groove C meets a predetermined judgment criterion. The abnormal state of the kerf groove C means that the state of the kerf groove C does not meet the predetermined judgment criterion.

[0034] In the example of FIG. 4 , the kerf groove C indicated by the training data 201 does not have any cutting defects such as chips or cracks. The kerf groove C indicated by the training data 202 has a chip, but the size of the chip is smaller than the judgment criterion. The kerf groove C indicated by the training data 203 has a crack, but the size of the crack is smaller than the judgment criterion. These training data 201, 202, and 203 are assigned a label of "normal" indicating that the condition of the kerf groove C is normal. Furthermore, the training data 201 is assigned a label of "high" indicating that the quality level of the kerf groove C is high, and the training data 202 and 203 are assigned a label of "low" indicating that the quality level of the kerf groove C is lower than "high."

[0035] On the other hand, the cutting groove C of the training data 204 has a chip that is larger than the judgment criterion. The cutting groove C of the training data 205 has a crack that is larger than the judgment criterion. In the training data 206, the position of the cutting groove C is shifted from the center line of the image. These training data 204, 205, and 206 are assigned a label of "abnormal" indicating that the state of the cutting groove C is abnormal. Furthermore, the training data 204, 205, and 206 are assigned labels of "chip," "crack," and "misalignment" according to the state of each abnormality.

[0036] In the example of FIG. 4 , the teacher data 201, 202, and 203 are included in a normal processed image group with a major classification labeled "normal." The teacher data 201, 202, and 203 are further classified into a condition-specific image group with subclassifications labeled "high" and "low" according to the quality of the cut groove C. Meanwhile, the teacher data 204, 205, and 206 are included in an abnormal processed image group with a major classification labeled "abnormal." The teacher data 204, 205, and 206 are further classified into condition-specific image groups with subclassifications labeled "chipped," "cracked," and "misaligned" according to the type of abnormal state of the cut groove C. The number of teacher data 200 constituting the trained program 300 varies depending on the number of teacher data 200 stored in the teacher data storage unit 44.

[0037] The trained program 300 is configured to machine-learn a model using a plurality of training data 200 (trainer data 201, 202, and 203 included in the normally processed image group and training data 204, 205, and 206 included in the abnormally processed image group) as a data set, and determine whether an input image belongs to the normally processed image group or the abnormally processed image group based on the machine-learned model. For example, the trained program 300 calculates an index (similarity) indicating the degree to which the input image is similar to each of the normally processed image group and the abnormally processed image group. The trained program 300 then determines that the input image belongs to the normally processed image group when the similarity between the input image and the normally processed image group is greater than the similarity between the input image and the abnormally processed image group. Furthermore, the trained program 300 determines that the input image belongs to the abnormally processed image group when the similarity between the input image and the normally processed image group is less than the similarity between the input image and the abnormally processed image group. The trained program 300 may be configured to calculate the similarity between the input image and each of the state-specific image groups included in the normally processed image group and the abnormally processed image group, and to determine which state-specific image the input image belongs to. The trained program 300 may be configured to, for example, stop the determination process when the difference between the similarity between the input image and the normally processed image group and the similarity between the input image and the abnormally processed image group is smaller than a predetermined value.

[0038] The determination unit 47 executes the learned program 300 on the captured image data 100 of the cut groove C formed by the cutting unit 2 to determine whether the captured image belongs to a normally processed image or an abnormally processed image. By executing the learned program 300 on the captured image data 100, the determination unit 47 may determine not only whether the captured image belongs to a normally processed image or an abnormally processed image, but also which state-specific image the captured image belongs to. The determination unit 47 outputs an instruction to the control unit 5 to execute processing according to the determination result. The processing according to the determination result includes, for example, processing to notify the control unit 5 of the determination result and processing to control the cutting process.

[0039] If the learned program 300 is configured to stop the judgment process when a specified condition is met, the judgment unit 47, based on the fact that the judgment process has been stopped, has the operator judge whether the captured image belongs to a normal processing image or an abnormal processing image (and to which state-specific image the captured image belongs), and outputs an instruction to the control unit 5 to execute processing to accept the judgment result.

[0040] The control unit 5 executes processing according to the normal / abnormal determination result based on instructions from the determination unit 47. For example, the control unit 5 executes processing to control the operation of the cutting unit 2. Specifically, if the control unit 5 determines that the captured image belongs to the normal processing image, it can execute processing for performing the next cut. If the control unit 5 determines that the captured image belongs to the abnormal processing image, it can execute processing such as canceling the cutting process or changing the cutting conditions. In addition, the control unit 5 can execute processing such as notifying the determination result, issuing a warning of a poor cut, calling the operator, and analyzing the frequency of occurrence of poor cuts.

[0041] Furthermore, the control unit 5 can execute a process for allowing the operator to determine whether a captured image is normal or abnormal based on an instruction from the determination unit 47. For example, the control unit 5 displays on the display unit 7 a screen on which the operator can select whether the captured image belongs to a normal processed image or an abnormal processed image (and to which state-specific image the captured image belongs), and outputs information on the selection result input by the operator via the input unit 6 to the determination unit 47. Then, the determination unit 47 instructs the control unit 5 to execute a process according to the above-mentioned determination result of normality or abnormality, based on information on the determination result by the operator output from the control unit 5. The captured image determined by the operator to belong to either a normal processed image or an abnormal processed image may be associated with the information on the determination result and stored in the teacher data storage unit 44.

[0042] Next, a processing procedure for generating the trained program 300 executed by the semiconductor manufacturing apparatus 1 according to this embodiment will be described. Fig. 5 is a flowchart showing an example of the processing procedure for generating the trained program 300 according to this embodiment. The processing procedure shown in Fig. 5 is executed by each function of the data processing unit 4. The processing procedure shown in Fig. 5 is executed when an image is registered as training data 200, when an image to be additionally registered as training data 200 is generated, etc.

[0043] First, the teacher data creating unit 43 reads out the captured image data 100 from the image data storage unit 42 (step S101).

[0044] Next, the teacher data creation unit 43 labels the captured image data 100 as "normal" or "abnormal" depending on the state of the kerf groove C indicated by the acquired captured image data 100, and creates teacher data 200 (step S102). Specifically, the teacher data creation unit 43 outputs to the control unit 5 an instruction to display on the display unit 7 the captured image and an image that allows the operator to select whether the kerf groove C indicated by the captured image is normal or abnormal, and labels the captured image data 100 based on information input by the operator via the input unit 6 regarding the selection result. At this time, the operator not only inputs the result of determining whether the kerf groove C is normal or abnormal, but also inputs information indicating the detailed state of the kerf groove C (e.g., the quality of the cutting process, the type of cutting defect, etc.). In other words, the teacher data creation unit 43 labels the captured image data 100 not only with the binary value of normal or abnormal, but also with information regarding the detailed state of the kerf groove C. Then, the teacher data creation unit 43 stores the labeled captured image data 100 as teacher data 200 in the teacher data storage unit 44 (step S103).

[0045] Next, the machine learning unit 45 performs machine learning using a plurality of training data 200 (a normally processed image group including a plurality of training data 200 (201, 202, and 203) labeled as "normal" and an abnormally processed image group including a plurality of training data 200 (204, 205, and 206) labeled as "abnormal") as a data set, and generates a trained program 300 (step S104). The machine learning unit 45 generates the trained program 300 that has learned a rule (model) for determining whether the input unknown kerf C is normal or abnormal, based on the features of the kerf C indicated by the training data 200 included in the normally processed image group and the features of the kerf C indicated by the training data 200 included in the abnormally processed image group. Specifically, the trained program 300 is configured to calculate the similarity between the input image and the normally processed image group and the abnormally processed image group, and to determine that the input image belongs to the normally processed images when the similarity between the input image and the normally processed image group is greater than the similarity between the input image and the abnormally processed image group, and to determine that the input image belongs to the abnormally processed images when the similarity between the input image and the normally processed image group is less than the similarity between the input image and the abnormally processed image group.The trained program 300 is also configured to stop the above-mentioned determination process when the difference between the similarity between the input image and the normally processed image group and the similarity between the input image and the abnormally processed image group is less than a predetermined value.Furthermore, the trained program 300 is configured to calculate the similarity between the input image and each condition-specific image group classified based on information regarding the detailed condition of the cutting groove C, and to determine to which condition-specific image the input image belongs based on the similarity.

[0046] The machine learning unit 45 stores the generated trained program 300 in the trained program storage unit 46 (step S105). If the trained program storage unit 46 already stores the trained program 300, the machine learning unit 45 overwrites the generated trained program 300 or stores it as a new program. Through the above processing procedure, the trained program 300 is generated and stored in the trained program storage unit 46.

[0047] Next, a process procedure for determining whether the kerf groove C is normal / abnormal, which is executed by the semiconductor manufacturing apparatus 1 according to this embodiment, will be described. FIG. 6 is a flowchart showing an example of the process procedure for determining the kerf groove C according to this embodiment. The process procedure shown in FIG. 6 is executed by the data processing unit 4 and the control unit 5. The process procedure shown in FIG. 6 is executed, for example, at any timing between the start and end of processing of one wafer W. Note that the process procedure shown in FIG. 6 uses the learned program 300 generated by the process procedure shown in FIG. 5 and stored in the learned program storage unit 46.

[0048] First, the image data acquisition unit 41 acquires the captured image data 100 of the cutting groove C formed by the cutting processing unit 2 from the imaging unit 3, and links the captured image data 100 to the identification information of the wafer W, the position information of the imaging location, etc., and stores it in the image data storage unit 42 (step S201).

[0049] Next, the determination unit 47 reads the captured image data 100 from the image data storage unit 42 and also reads the learned program 300 from the learned program storage unit 46 (step S202). Subsequently, the determination unit 47 executes the learned program 300 on the captured image data 100 (step S203), and determines whether the captured image belongs to a normal processed image showing a normal kerf C or an abnormal processed image showing an abnormal kerf C (step S204).

[0050] If the captured image is determined to be normal / abnormal (Yes in step S205) and is determined to belong to the normal processing image category ((i) in step S206), the determination unit 47 outputs an instruction to the control unit 5 to execute processing according to the determination result that the captured image is normal (step S207). The processing according to the determination result that the captured image is normal includes, for example, processing to display on the display unit 7 that the kerf groove C is normal and the quality level of the kerf groove C, and processing to continue the cutting process.

[0051] If the captured image is determined to be normal / abnormal (Yes in step S205) and the captured image is determined to belong to the abnormal processing image category ((ii) in step S206), the determination unit 47 outputs an instruction to the control unit 5 to execute processing according to the determination result that the image is abnormal (step S208). The processing according to the determination result that the image is abnormal includes, for example, processing to warn the operator that the cut groove C is abnormal, processing to display details of the abnormal state of the cut groove C (type of cutting defect, degree of cutting defect, etc.) on the display unit 7, processing to change the cutting conditions according to the abnormal state of the cut groove C (correction of the cutting position, change of the distance by which the break bar 23 is pressed down, etc.), and processing to stop the cutting process.

[0052] On the other hand, if the determination of normality / abnormality of the captured image is canceled (No in step S205), the determination unit 47 prompts the operator to determine whether the captured image is a normally processed image or an abnormally processed image, and outputs an instruction to the control unit 5 to execute processing to accept the determination result (step S209). This processing includes, for example, displaying the captured image and a screen on the display unit 7 that prompts the operator to select whether the captured image is a normally processed image or an abnormally processed image, and outputting information input by the operator via the input unit 6 regarding the selection result to the determination unit 47. The control unit 5 executes the above processing based on the instruction output in step S209 (step S210). In step S210, the determination unit 47 accepts the determination result by the operator. Thereafter, the determination unit 47 executes the processing of steps S206 to S208 based on the normality / abnormality determination result of the captured image by the operator.

[0053] The control unit 5 executes processing according to the normal / abnormal determination result based on the instructions output in steps S207 and S208 (step S211).

[0054] In step S210, the captured image determined by the operator as belonging to either a normally processed image or an abnormally processed image is linked to the determination result by the operator and the captured image data 100, and stored in the teacher data storage unit 44. Thereafter, the trained program 300 is updated by the processing procedure of steps S101 to S105.

[0055] According to the semiconductor manufacturing apparatus configured as described above, whether the kerf groove formed on the substrate cut by the cutting unit 2 is normal or abnormal is mechanically (objectively) determined, and cutting defects can be reliably recognized regardless of the operator's level of proficiency. Furthermore, by increasing the number of normal and abnormal processed images of the kerf groove C, the accuracy of kerf defect recognition can be improved. By using image data 100, which is an actual image of the kerf groove C formed on the wafer W by the cutting unit 2, as training data 200, it is possible to determine the kerf groove C according to the type of wafer W and the characteristics of the semiconductor manufacturing apparatus 1, thereby improving the accuracy of processing defect recognition. Furthermore, when the determination unit 47 has difficulty determining normality or abnormality, the results of a skilled operator's determination can be reflected in the trained program 300, thereby improving the accuracy of processing defect recognition.

[0056] 7 shows an example of a semiconductor manufacturing system including a semiconductor manufacturing apparatus (breaking apparatus) 1 according to this embodiment. In this semiconductor manufacturing system 500, a transfer robot (substrate transfer apparatus) 570 is arranged in the center of a rectangular parallelepiped storage container 580. In addition to one breaking apparatus 1 according to this embodiment, two scribe line forming apparatuses 510 (scribe line forming apparatuses 510A and 510B), one film bonding apparatus 520, one substrate inverting apparatus 530, one film peeling apparatus 550, and one substrate storage unit 560 are arranged around the transfer robot 570. The substrate inverting apparatus 530 and the film peeling apparatus 550 are arranged above and below each other at the same position on a plane.

[0057] The substrate accommodation unit 560 has a plurality of cassettes 561 that accommodate unprocessed and processed wafers W. The transport robot 570 is an apparatus that takes out the unprocessed substrate W from the cassette 561, transports it to one of the two scribe line forming devices 510A and 510B, and then transports the substrate W through the film bonding device 520, the substrate inverting device 530, the semiconductor manufacturing device (breaking device) 1, the substrate inverting device 530 (for the second time), the film peeling device 550, and the cassette 561 in this order.

[0058] The scribe line forming device 510 is a device that forms a scribe line on the surface of the wafer W. The film bonding device 520 is a device that bonds a film F to the surface of the wafer W on which a scribe line has been formed. The substrate inverting device 530 is a device that performs a process of inverting the wafer W transferred from the film bonding device 520 so that the surface of the wafer W faces downward, and a device that performs a process of inverting the wafer W that has been subjected to a breaking process in the semiconductor manufacturing device (breaking device) 1 so that the surface faces upward. The film peeling device 550 is a device that peels off the film F from the wafer W that has been subjected to the breaking process. Known devices can be used for the scribe line forming device 510, film bonding device 520, substrate inverting device 530, and film peeling device 550.

[0059] The scribe line formation process by the scribe line formation device 510 takes longer to process one wafer W than the film bonding process, substrate reversal process, breaking process, and film peeling process, which are processes performed by other devices in the semiconductor manufacturing system 500. Therefore, simply providing the same number of scribe line formation devices 1 as the other devices means that processing for one wafer in the other devices will finish before the scribe line formation process for one wafer in the scribe line formation device 510 finishes, and those other devices will be in a standby state until the processing by the scribe line formation device 1 finishes, causing a decrease in the operating efficiency of the semiconductor manufacturing system. Therefore, in the semiconductor manufacturing system 500, by providing more scribe line formation devices 510 than the other devices (in the example shown here, there are two scribe line formation devices 1 and one each of the other devices, but this example is not limited to this), the standby time of the other devices can be shortened and the operating efficiency of the semiconductor manufacturing system can be improved.

[0060] The above describes an embodiment of the semiconductor manufacturing apparatus (break apparatus) according to the present invention and an embodiment of the semiconductor manufacturing system having the apparatus. However, it goes without saying that the present invention is not limited to the above-described embodiment, and various modifications are possible.

[0061] Aspects It will be apparent to those skilled in the art that the exemplary embodiments described above are examples of the following aspects.

[0062] (Item 1) A semiconductor manufacturing apparatus according to one aspect of the present invention comprises: a cutting processing unit that cuts a substrate by applying an external force to a scribe line formed on the substrate; an imaging unit that images a cut groove formed on the substrate; and a determination unit that processes data of the image of the cut groove captured by the imaging unit using a trained program that has learned teacher data including a normal processing image group consisting of a plurality of images of normal cut grooves and an abnormal processing image group consisting of a plurality of images of cut grooves with cutting defects, and determines whether the image belongs to a normal processing image or an abnormal processing image.

[0063] (2) The semiconductor manufacturing apparatus according to the 2nd paragraph is the semiconductor manufacturing apparatus according to the 1st paragraph, wherein the abnormality processing image group is classified into a plurality of image groups by abnormality state according to the abnormal state of the cutting groove, and the judgment unit judges to which of the plurality of abnormal states the captured image belongs.

[0064] (Clause 3) The semiconductor manufacturing apparatus according to clause 3 is a semiconductor manufacturing apparatus according to clause 1 or 2, wherein the judgment unit calculates the similarity between the captured image and the group of normally processed images and the group of abnormally processed images, respectively, and if the difference between the similarity between the captured image and the group of normally processed images and the similarity between the captured image and the group of abnormally processed images is less than a predetermined value, the judgment is stopped and processing is performed to allow the user to judge whether the captured image belongs to a normally processed image or an abnormally processed image.

[0065] (4) The semiconductor manufacturing apparatus according to paragraph 4 is a program in the semiconductor manufacturing apparatus according to paragraph 3, in which the trained program is trained by adding the captured image, which the user has determined to be either the normally processed image or the abnormally processed image, to training data.

[0066] (Item 5) The semiconductor manufacturing apparatus according to item 5 is the semiconductor manufacturing apparatus according to any one of items 1 to 4, further comprising a control unit that controls the cutting processing unit based on the determination result of the determination unit.

[0067] (Item 6) A semiconductor manufacturing system according to one aspect of the present invention comprises: a semiconductor manufacturing apparatus according to any one of items 1 to 5; a scribe line forming apparatus that forms the scribe line on the substrate; and a substrate transport apparatus that transports the substrate into and out of the semiconductor manufacturing apparatus and the scribe line forming apparatus.

[0068] DESCRIPTION OF SYMBOLS 1...Semiconductor manufacturing apparatus (breaking apparatus) 2...Cutting processing section 3...Image capturing section 4...Data processing section 41...Image data acquisition section 42...Image data storage section 43...Teacher data creation section 44...Teacher data storage section 45...Machine learning section 46...Taught program storage section 47...Determination section 5...Control section 6...Input section 7...Display section 100...Captured image data 200...Teacher data 300...Taught program 500...Semiconductor manufacturing system 510, 510A, 510B...Scribe line forming device 520...Film bonding device 530...Substrate inversion device 550...Film peeling device 560...Substrate storage section 561...Cassette 570...Transport robot (substrate transport device) 580...Storage container W...Wafer C...Cutting groove

Claims

1. A semiconductor manufacturing device comprising: a cutting processing unit that cuts a substrate by applying an external force to a scribe line formed on the substrate; an imaging unit that images a kerf formed on the substrate; and a determination unit that processes image data of the kerf taken by the imaging unit using a trained program that has learned teacher data including a normally processed image group consisting of a plurality of images of normal kerfs and an abnormally processed image group consisting of a plurality of images of kerfs with cutting defects, and determines whether the image belongs to a normally processed image or an abnormally processed image.

2. The semiconductor manufacturing device of claim 1, wherein the abnormality processing image group is classified into a plurality of image groups by abnormality state according to the abnormal state of the cutting groove, and the judgment unit judges to which of the plurality of abnormal states the captured image belongs.

3. The semiconductor manufacturing apparatus of claim 1, wherein the judgment unit calculates the similarity between the captured image and the group of normally processed images and the group of abnormally processed images, and if the difference between the similarity between the captured image and the group of normally processed images and the similarity between the captured image and the group of abnormally processed images is less than a predetermined value, the judgment is stopped and processing is performed to allow a user to judge whether the captured image belongs to a normally processed image or an abnormally processed image.

4. A semiconductor manufacturing apparatus as described in claim 3, wherein the trained program is a program trained by adding the captured image, which the user has determined to belong to either the normally processed image or the abnormally processed image, to training data.

5. The semiconductor manufacturing apparatus according to claim 1, further comprising a control unit that controls the cutting unit based on the determination result of the determination unit.

6. A semiconductor manufacturing system comprising: the semiconductor manufacturing apparatus according to any one of claims 1 to 5; a scribe line forming device that forms the scribe line on the substrate; and a substrate transport device that transports the substrate into and out of the semiconductor manufacturing apparatus and the scribe line forming device.

Images