Wafer global alignment learning device and operating method therefor
The wafer GA learning device uses deep learning to address misalignment issues in semiconductor wafer inspection by generating a target image model from sample images, enhancing the alignment process's reliability and universality.
Patent Information
- Application Number
- PCT/KR2026/096880
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-12
- Filing Date
- 2026-05-12
- Publication Date
- 2026-11-19
AI Technical Summary
Existing semiconductor wafer measurement and inspection systems face challenges in ensuring consistent alignment between the wafer scan direction and die arrangement direction, leading to misalignment issues that affect the reliability and universality of the global alignment (GA) process.
A wafer GA learning device utilizing deep learning to improve the universality and reliability of the GA process by learning from target image samples, including successful and failed target images, and generating a target image model to automate the alignment process.
Enhances the efficiency and reliability of the GA process by automating the alignment process, improving the consistency of target image recognition and reducing the need for manual intervention.
Smart Images

Figure KR2026096880_19112026_PF_FP_ABST
Abstract
Description
Wafer global alignment learning device and method of operation thereof
[0001] The present invention relates to a wafer global alignment (GA) learning device and a method of operation thereof.
[0002] More specifically, the invention relates to a device for learning global alignment through deep learning for measurement and inspection of semiconductor wafers, and a method of operation thereof.
[0003] When measuring and inspecting semiconductor wafers, the wafer scan direction must be parallel to the wafer's die arrangement direction. To this end, a wafer Global Alignment (GA) process is performed prior to measurement and inspection to calculate the error (degree of misalignment) between the wafer scan direction and the wafer's die arrangement direction, and measurement and inspection are conducted after correcting the two directions to ensure they are parallel.
[0004] This will be explained in detail through Figures 1 to 4.
[0005] FIG. 1 is an exemplary drawing illustrating a case where the wafer scan direction and the die arrangement direction are parallel during the wafer measurement and inspection process. FIG. 2 is an exemplary drawing illustrating a case where the wafer scan direction and the die arrangement direction are not parallel during the wafer measurement and inspection process. FIG. 3 is an exemplary drawing illustrating a target image used in wafer GA. FIG. 4 is another exemplary drawing illustrating a target image used in wafer GA.
[0006] Referring to FIG. 1, a plurality of dies can be formed within a wafer (W). An example of FIG. 1 illustrates a case where the arrangement direction of the dies is parallel to the direction of movement of the semiconductor wafer (W) for measurement and inspection. At this time, a target image (10a) can be patterned at any location within the die for GA of the wafer (W).
[0007] The target image (10a) may be, for example, like FIGS. 3 and FIGS. 4.
[0008] For example, the first target images (10a to 10d) may have the same basic shape as in FIG. 3, but may vary in size and / or color. Of course, there may be various forms of target images not illustrated, while maintaining the basic shape as shown in FIG. 3. In the case of variation between dies within the same wafer, variation in the target image as shown in FIG. 3 may occur. The example in FIG. 3 is a variation in the target image that may occur between dies within the same wafer, and the example in FIG. 3 may be a variation in the target image that may occur between different wafers.
[0009] As another example, the second target images (20a to 20d) may have the same basic shape as in FIG. 4, but may vary in size and / or color. Of course, there may be various forms of target images not illustrated, while maintaining the basic shape as shown in FIG. 4. In the case of variation between dies within the same wafer, variation in the target image as in FIG. 4 may occur.
[0010] Referring further to FIG. 1, assuming that the wafer scan direction and the die alignment direction are aligned parallel, a wafer measurement and inspection performer or user can anticipate that the target image (10a) will move to a position (10a'). If the predicted position of the target image (10a') is located within a range of a certain offset defined by the user, the wafer can be determined to be correctly aligned for measurement and inspection.
[0011] However, as shown in FIG. 2, there may be cases where the arrangement direction of the die is not parallel to the direction of movement for measuring and inspecting the semiconductor wafer (W).
[0012] At this time, the wafer measurement and inspection performer or user can expect the target image (10a) to move to position (10a') under the premise that the wafer scan direction and the die arrangement direction are aligned parallel.
[0013] However, since the die arrangement direction of the wafer shown in FIG. 2 is not parallel to the wafer scan direction, the position of the actual moved target image (10a") and the position predicted by the user (10a') deviate by a certain offset. That is, it can be determined that the wafer is not properly aligned for measurement and inspection.
[0014] At this time, the target images patterned for GA on different wafers may be different. For example, a target image (10a) as shown in FIG. 3 may be patterned on the first wafer, and a target image (20a) as shown in FIG. 4 may be patterned on the second wafer, which is different from the first wafer.
[0015] At this time, the target image patterned on each wafer for GA may be subject to variation due to external conditions that occur when scanning the wafer.
[0016] For example, referring to FIGS. 1 and FIGS. 3, even if the target image patterned at position (10a) is the target image (10a') of FIGS. 3, it may become the changed target image (10b, 10c, or 10d) of FIGS. 3 at the position (10a') after moving in the wafer scan direction.
[0017] The description through FIGS. 1 and FIGS. 3 is exemplary, and it goes without saying that the description of the target images (10a, 10b, 10c, and 10d) of FIGS. 3 can be replaced with the description of the target images (20a, 20b, 20c, and 20d) of FIGS. 4.
[0018] Continuing, referring to FIGS. 1 and FIGS. 3, if the target image for verifying GA is within the variation range of 10a to 10d, it must be recognized as the same target image.
[0019] In other words, there is a need to be able to trust the consistency of the patterned target image for GA during the scanning process of the semiconductor wafer, even for variations in the target image.
[0020] The technical problem that the present disclosure aims to solve is to provide a wafer GA learning device and a method of operation thereof that improves the universality and reliability of GA by performing learning on a target image patterned on a wafer for GA using deep learning.
[0021] According to some embodiments, a wafer GA learning device includes a receiving unit that receives target image samples patterned on the wafer for wafer GA from an external device, wherein the target image samples include a successful target image in which the GA is successful and a failed target image in which the GA is failed, wherein the successful target image includes a success label and the failed target image includes a failure label; a deep learning unit that performs learning on the successful target image and the failed target image based on the target image samples and generates or updates a target image model according to the learning performance; and a transmitting unit that transmits the target image model to the external device where the GA is performed.
[0022] According to one embodiment, the wafer GA learning device further includes a sample count checking unit that checks whether the number of target image samples received from the external device is sufficient for the deep learning unit to learn to generate the target image model, wherein the sample count checking unit transmits the target image samples to the deep learning unit if it determines that the number of target image samples is sufficient for learning, and commands the receiving unit to receive additional target image samples from the external device if it determines that the number of target image samples is not sufficient for learning.
[0023] According to one embodiment, the wafer GA learning device further includes a quality verification unit for determining the performance of the target image model generated through the deep learning unit, wherein the quality verification unit compares whether the labeling of the target image through the target image model matches the labeling of the image used as the target image sample, and if it is determined that the number of successful cases exceeds a specified level, it commands the transmitting unit to transmit the target image model to the external device, and if it is determined that the number of successful cases does not exceed a specified level, it commands the receiving unit to additionally receive the target image samples from the external device.
[0024] According to one embodiment, the wafer GA learning device further includes a performance enhancement block for improving the performance of the target image model, wherein the performance enhancement block includes: a test unit that tests whether the target image model is operating normally by applying the target image model generated by the deep learning unit to the target image samples received from the external device; a first quality verification unit that determines whether the quality of the target image model is lower than a predetermined value according to the test, and if the quality of the target image model is determined to be high, does not perform additional functions, and if the quality of the target image model is determined to be low, issues an update command to update the target image model for the target image samples; and an update unit that generates an updated target image model by performing learning based on the target image samples according to the update command.
[0025] According to one embodiment, a wafer GA learning device further includes a second quality verification unit that verifies the quality of the updated target image model, wherein the second quality verification unit compares whether the labeling of the target image through the updated target image model matches the labeling of the image used as the target image sample, and if it is determined that the number of successful cases exceeds a specified level, it commands the updated target image model to be transmitted to the external device, and if it is determined that the number of successful cases does not exceed a specified level, it commands the target image samples to be additionally received from the external device.
[0026] A method of operating a wafer GA learning device according to some embodiments comprises the steps of: receiving, through a receiving unit, target image samples to be patterned on the wafer for wafer GA from an external device, wherein the target image samples include a successful target image in which the GA is successful and a failed target image in which the GA is failed, wherein the successful target image includes a success label and the failed target image includes a failure label; performing learning on the successful target image and the failed target image based on the target image samples through a deep learning unit and generating a target image model based on the learning performance; and transmitting, through a transmitting unit, the target image model to the external device where the GA is performed.
[0027] A method of operating a wafer GA learning device according to one embodiment further includes the step of checking whether the number of target image samples received from the external device is sufficient for the deep learning unit to learn to generate the target image model through a sample number checking unit, wherein the step of checking the number of samples through the sample number checking unit transmits the target image samples to the deep learning unit if it is determined that the number of target image samples is sufficient for learning, and commands the receiving unit to additionally receive the target image samples from the external device if it is determined that the number of target image samples is not sufficient for learning.
[0028] According to one embodiment, the method of operating a wafer GA learning device further includes a step of determining the performance of the target image model generated through the deep learning unit through a quality verification unit, wherein the step of determining the performance through the quality verification unit compares whether the labeling assigned to the target image through the target image model matches the labeling of the image used as the target image sample, and if it is determined that the number of successful attempts exceeds a specified level, commands the transmitting unit to transmit the target image model to the external device, and if it is determined that the number of successful attempts does not exceed a specified level, commands the receiving unit to additionally receive the target image samples from the external device.
[0029] A method of operating a wafer GA learning device according to one embodiment further includes a step of improving the performance of the target image model through a performance improvement block, wherein the step of improving the performance of the target image model through the performance improvement block includes: a step of testing whether the target image model is operating normally by applying the target image model generated by the deep learning unit to the target image samples received from the external device through a test unit; a step of determining whether the quality of the target image model is lower than a predetermined value according to the test through a first quality verification unit, wherein if the quality of the target image model is determined to be high, no additional function is performed, and if the quality of the target image model is determined to be low, an update command is issued so that the target image model for the target image samples is newly updated; and a step of generating an updated target image model by performing learning based on the target image samples according to the update command through an update unit.
[0030] A method of operating a wafer GA learning device according to one embodiment further includes the step of checking the quality of the updated target image model through a second quality verification unit, wherein the step of checking the quality of the updated target image model through the second quality verification unit includes comparing whether the labeling assigned to the target image through the updated target image model matches the labeling for the image used as the target image sample, and if it is determined that the number of successful cases exceeds a specified level, commanding the updated target image model to be transmitted to the external device, and if it is determined that the number of successful cases does not exceed a specified level, commanding the receiving of additional target image samples from the external device.
[0031] Through a wafer GA learning device and a method of operation thereof according to some embodiments, for a GA process using a target image of a semiconductor wafer measurement and inspection system, by performing deep learning on the target image, the universality and reliability of the GA process can be improved, and the efficiency of the GA process execution can be improved by implementing automation.
[0032] FIG. 1 is an exemplary drawing illustrating a case where the wafer scan direction and the die arrangement direction are parallel during the wafer measurement and inspection process.
[0033] FIG. 2 is an exemplary drawing illustrating a case where the wafer scan direction and the die arrangement direction are not parallel during the wafer measurement and inspection process.
[0034] Figure 3 is an exemplary drawing illustrating a target image used in wafer GA.
[0035] Figure 4 is another exemplary drawing illustrating a target image used in wafer GA.
[0036] FIG. 5 is an exemplary block diagram illustrating a semiconductor wafer inspection (equipment) system including a wafer GA learning device according to some embodiments.
[0037] FIG. 6 is an exemplary block diagram illustrating a wafer GA learning device according to some embodiments.
[0038] FIG. 7 is an exemplary flowchart illustrating the operation method of a wafer GA learning device according to some embodiments.
[0039] FIG. 8 is an exemplary block diagram illustrating different wafer GA learning devices according to some embodiments.
[0040] FIG. 9 is an exemplary flowchart illustrating the operation method of different wafer GA learning devices according to some embodiments.
[0041] Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. The advantages and features of the present disclosure and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the technical concept of the present disclosure is not limited to the following embodiments but can be implemented in various different forms. These embodiments are provided merely to make the present disclosure complete and to fully inform those skilled in the art of the scope of the present disclosure, and the technical concept of the present disclosure is defined only by the scope of the claims.
[0042] It should be noted that when assigning reference numerals to the components of each drawing, the same components are given the same reference numeral whenever possible, even if they are shown in different drawings. Furthermore, in describing the present disclosure, if it is determined that a detailed description of related known components or functions could obscure the essence of the present disclosure, such detailed description is omitted.
[0043] Unless otherwise defined, all terms used herein (including technical and scientific terms) may be used in a meaning commonly understood by those skilled in the art to which this disclosure pertains. Additionally, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise. The terms used herein are for describing the embodiments and are not intended to limit this disclosure. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text.
[0044] Spatially relative terms such as "below," "beneath," "lower," "above," and "upper" may be used to facilitate the description of the relationship between one component and other components as illustrated in the drawings. Spatially relative terms should be understood as encompassing different orientations of components during use or operation, in addition to the orientations depicted in the drawings. For example, if a component depicted in a drawing is inverted, a component described as "below" or "beneath" of another component may be placed "above" of that component. Therefore, the exemplary term "below" may encompass both the lower and upper directions. Components may also be oriented in other directions, and accordingly, spatially relative terms may be interpreted according to the orientation.
[0045] Additionally, terms such as first, second, A, B, (a), (b), etc., may be used to describe the components of the present disclosure. These terms are intended only to distinguish the components from other components and do not limit the nature, order, or sequence of the components. Where it is stated that a component is "connected," "coupled," or "joined" to another component, it should be understood that the component may be directly connected or joined to the other component, but that another component may also be "connected," "coupled," or "joined" between each component.
[0046] As used in the specification, "comprises" and / or "comprising" do not exclude the presence or addition of one or more other components, steps, actions, and / or elements to the mentioned components, steps, actions, and / or elements.
[0047] Before describing the present specification, we will clarify some terms used in the present specification.
[0048] Hereinafter, several embodiments of the present disclosure will be described in detail with reference to the attached drawings.
[0049] FIG. 5 is an exemplary block diagram illustrating a semiconductor wafer inspection system including a wafer GA learning device according to some embodiments. FIG. 6 is an exemplary block diagram for explaining a wafer GA learning device according to some embodiments. FIG. 7 is an exemplary flowchart for explaining a method of operation of a wafer GA learning device according to some embodiments.
[0050] Referring to FIGS. 5 to 7, a semiconductor wafer inspection system (1) including a wafer GA learning device (201) according to some embodiments includes a plurality of tools (Tool1, Tool2, and Tool3).
[0051] The number of multiple tools (202) (Tool1, Tool2, and Tool3) is exemplary, and the number of multiple tools (Tool1, Tool2, and Tool3) is not limited to the present drawing.
[0052] A wafer GA learning device (201) according to some embodiments includes a target image learning block (210).
[0053] A wafer GA learning device (201) according to some embodiments may be, for example, a server device capable of processing.
[0054] A target image learning block (210) included in a wafer GA learning device (201) according to some embodiments includes a receiving unit (212), a sample count checking unit (214), a deep learning unit (216), a quality checking unit (218), and a transmitting unit (219).
[0055] The configuration of the wafer GA learning device (201) according to some embodiments is not limited thereto, and it is obvious that the wafer GA learning device (201) according to some embodiments may be configured with other blocks as long as the operation is driven.
[0056] Multiple tools (202) (Tool1, Tool2, and Tool3) become blocks where wafer measurement and inspection processes are performed.
[0057] Each of the plurality of tools (202) (Tool1, Tool2, and Tool3) can receive models (204) related to the target images required to perform a GA process based on the target images patterned on the wafers (e.g., the target images described through FIG. 3 and FIG. 4) through a wafer GA learning device (201) according to some embodiments.
[0058] The input to the models (204) associated with the target images may be the target images, and the output may be a binary bit that is output as true / false. In this process, samples (203) of the target images may be transmitted and received between a plurality of tools (202) (Tool1, Tool2, and Tool3) and a wafer GA learning device (201) according to some embodiments.
[0059] A wafer GA learning device (201) according to some embodiments can improve the universality and reliability of the GA by learning models (204) with samples (203) of a target image through deep learning (e.g., a deep learning convolutional neural network (CNN)), even if variations in the target image occur. Additionally, the reliability and efficiency of a semiconductor wafer inspection system (1) including a wafer GA learning device (201) according to some embodiments can be improved by automating the learning of the target image so that the user's designation of the target image is not required or is minimized.
[0060] Below, the operation of a wafer GA learning device (201) according to some embodiments is examined in detail.
[0061] In the tool (202) (hereinafter collectively referred to as tool (202), but it is obvious that the description of tool (202) may also apply to Tool 1, Tool 2, and Tool 3), the user marks a GA image (target image) for learning a target image of a wafer GA learning device (201) according to some embodiments (S301).
[0062] Afterwards, the tool (202) performs a GA based on the target (S302).
[0063] The tool (202) determines whether the GA execution is successful (S303). At this time, if the tool (202) determines that the GA has failed (N), it does not perform additional communication with the wafer GA learning device (201) according to some embodiments. If the tool (202) determines that the GA has succeeded (Y), it transmits the wafer GA learning device (201) according to some embodiments with a success label (e.g., True label) on the target image used for success to the wafer GA learning device (201) according to some embodiments. At this time, the tool (202) transmits the wafer GA learning device (201) according to some embodiments with the success-labeled target image, along with a failure label (e.g., False label) on an image obtained randomly from an arbitrary location excluding the location of the target image used for success. That is, when the tool (202) determines that the GA is successful (Y), it transmits the target image labeled as successful and the randomly obtained images at locations excluding the location of the target image labeled as failure to the wafer GA learning device (201) according to some embodiments.
[0064] The steps following step S303 are described in terms of the operation of the wafer GA learning device (201) according to some embodiments.
[0065] A wafer GA learning device (201) according to some embodiments receives, through a receiver (212), a target image that has been successfully labeled and randomly obtained images at locations excluding the location of the target image that has been failed labeled as sample images (S304).
[0066] The sample count verification unit (214) determines whether the sample images received from the tool (202) are sufficient for training (S305). The criterion for determining whether the sample images are sufficient for training is exemplary and can be any natural number.
[0067] If the sample count verification unit (214) determines that the sample images received from the tool (202) are not sufficient for learning (N), the wafer GA learning device (201) according to some embodiments waits until additional samples are received from the tool (202) (S305b).
[0068] If the sample count verification unit (214) determines that the sample images received from the tool (202) are sufficient for training (Y), the deep learning unit (216) performs training based on the sample images and creates a deep learning model that performs GA (S306). The deep learning model used at this time may be, for example, a Convolutional Neural Network (CNN), but the deep learning model used by the deep learning unit (216) is not limited to this. The sample images trained by the deep learning unit (216) have success labels for target images that have succeeded in GA, and failure labels for images that are randomly collected from locations excluding the locations of target images that have succeeded in GA and have a very low probability of being target images. Therefore, if training is performed based on this, automated training for target images can be performed without user intervention.
[0069] Subsequently, the quality verification unit (218) verifies the quality of the model generated through the deep learning unit (216) (S307). More specifically, the quality verification unit (218) compares whether the labeling of the target image through the model generated through the deep learning unit (216) matches the labeling of the image used as a sample image. That is, for the same image, if the result performed through the model generated through the deep learning unit (216) is a successful labeling that is a successful labeling in the sample image, then the model of the deep learning unit (216) can be determined to be accurate. If the quality verification unit (218) determines that the accuracy of the model is higher than a certain level (THD) (Y), then it can be concluded that the model generated through the deep learning unit (216) is reliable. In this case, the model (204) generated through the deep learning unit (216) is transmitted to the tool (202) through the transmission unit (219) (S308).
[0070] The tool (202) updates the original target image model with the model received from the wafer GA learning device (201) according to some embodiments (S309).
[0071] If, in step S307, for the same image, the result of the model generated through the deep learning unit (216) is a failure label when the sample image is labeled as a success, it can be determined that the model of the deep learning unit (216) is inaccurate. If the quality verification unit (218) determines that the accuracy of the model is lower than a certain level (THD) (N), it waits to receive more sample images from the tool (202) to increase the accuracy of the model generated by the deep learning unit (216) (S305b).
[0072] FIG. 8 is an exemplary block diagram for illustrating a different wafer GA learning device according to some embodiments. FIG. 9 is an exemplary flowchart for illustrating a method of operation of a different wafer GA learning device according to some embodiments.
[0073] Referring to FIGS. 5, 8, and 9, other wafer GA learning devices (201) according to some embodiments may additionally include a performance enhancement block (220) in addition to the block (210) described above through FIGS. 5 and 6.
[0074] The performance enhancement block (220) includes a receiving unit (221), a sample count verification unit (222), a test unit (223), a first quality verification unit (224), an update unit (225), a second quality verification unit (226), and a transmitting unit (227).
[0075] The performance enhancement block (220) can further enhance the performance of the target image model generated through the block (210).
[0076] Performance enhancement of the target image model through the performance enhancement block (220) is first performed in the step where the tool (202) performs GA without a target image (S401). That is, in the performance enhancement block (220), GA is performed without the user of the tool (202) specifying a target image.
[0077] At this time, it is determined whether the GA execution through the block (210) was successful (S402). If it failed (N), the user specifies a target image in the tool (202) to perform the GA, and proceeds to ensure the GA is successful.
[0078] If the performance enhancement block (220) performs GA without the user of the tool (202) specifying a target image, and if the GA is successful (Y), the tool (202) sends the labeled image samples to a wafer GA learning device (201) according to some embodiments (S403).
[0079] The steps following step S403 are described in terms of the operation of the wafer GA learning device (201) according to some embodiments.
[0080] The receiver (221) receives labeled target image samples from the tool (202) (S404).
[0081] Afterward, the sample count verification unit (222) determines whether there is a sufficient number of samples for performance improvement (S404). Whether there is a sufficient number of samples for performance improvement can be a sufficient number of samples to run deep learning, and the right is not limited to the number of samples.
[0082] If it is determined that the number of target image samples confirmed through the sample count verification unit (222) is insufficient (N), wait until more samples are collected (S404b).
[0083] Otherwise, if it is determined that the number of target image samples confirmed through the sample count verification unit (222) is sufficient (Y), the test unit (223) performs a test on the target image samples received through the receiving unit (221) based on the model generated through the block (210) (S405).
[0084] And, when the first quality verification unit (224) confirms that the result tested through the test unit (223) is higher than the performance improvement threshold (i.e., the reliability of the performance of the model generated through the block (210) is guaranteed based on the result tested through the test unit (223)) (N), the operation of the performance improvement block (220) is terminated.
[0085] However, if the first quality verification unit (224) confirms that the result tested through the test unit (223) is lower than the performance improvement threshold (i.e., the reliability of the performance of the model generated through the block (210) is not guaranteed based on the result tested through the test unit (223)) (Y), it performs an update on the model generated through the block (210) through the update unit (225). More specifically, the update unit (225) may give a command to the block (210) to generate a model. Not limited to this, a deep learning unit may exist within the update unit (225) to generate a new model itself.
[0086] Afterwards, the second quality verification unit (226) determines whether the new model created through the update unit (225) or block (210) has a performance higher than the performance improvement threshold (S408).
[0087] If it is determined that performance improvement has failed (N), wait to receive additional samples (S404b). Otherwise, if it is determined that performance improvement has succeeded (Y), transmit the updated model to the tool (202) through the transmitter (227) (S409).
[0088] Although embodiments of the present disclosure have been described above with reference to the attached drawings, those skilled in the art will understand that the present disclosure may be implemented in other specific forms without altering its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. The scope of protection of the present disclosure shall be interpreted by the claims below, and all technical concepts within the equivalent scope shall be interpreted as being included within the scope of rights of the present disclosure.
Claims
1. A receiver receiving target image samples patterned on a wafer for wafer global alignment from an external device, wherein the target image samples include a successful target image in which the global alignment is successful and a failed target image in which the global alignment is failed, wherein the successful target image includes a success label and the failed target image includes a failure label; A deep learning unit that performs learning on successful target images and failed target images based on the above target image samples, and generates or updates a target image model according to the learning performance; and A wafer global alignment learning device comprising a transmitter that transmits the above target image model to the above external device where the above global alignment is performed.
2. In Paragraph 1, The method further includes a sample count checking unit that checks whether the number of target image samples received from the external device is sufficient for the deep learning unit to train to generate the target image model, The above sample count verification unit is, If it is determined that the number of the above target image samples is sufficient for training, the above target image samples are transferred to the deep learning unit, and A wafer global alignment learning device that commands the receiver to receive additional target image samples from the external device when it is determined that the number of target image samples is insufficient for learning.
3. In Paragraph 2, It further includes a quality verification unit that determines the performance of the target image model generated through the deep learning unit, The above quality verification unit is, Compare whether the labeling assigned to the target image through the above target image model matches the labeling for the image used as the target image sample, and If it is determined that the number of successful attempts exceeds a specified level, the transmitting unit is commanded to transmit the target image model to the external device, and A wafer global alignment learning device that, if it is determined that the number of successful attempts does not exceed a specified level, commands the receiver to additionally receive the target image samples from the external device.
4. In Paragraph 1, Further including a performance enhancement block that improves the performance of the above target image model, The above performance enhancement block is, A test unit that tests whether the target image model is functioning normally by applying the target image model generated by the deep learning unit to the target image samples received from the external device; A first quality verification unit that determines whether the quality of the target image model is lower than a predetermined value according to the above test, and if the quality of the target image model is determined to be high, does not perform additional functions, and if the quality of the target image model is determined to be low, issues an update command to update the target image model for the target image samples; and A wafer global alignment learning device comprising an update unit that generates an updated target image model by performing learning based on the target image samples according to the above update command.
5. In Paragraph 4, The above performance enhancement block is, It further includes a second quality verification unit that verifies the quality of the above-mentioned updated target image model, The second quality verification unit compares whether the labeling assigned to the target image through the updated target image model matches the labeling for the image used as the target image sample, and If it is determined that the number of successful attempts exceeds a specified level, a command is given to transmit the updated target image model to the external device, and A wafer global alignment learning device that commands additional target image samples to be received from the external device if it is determined that the number of successful attempts does not exceed a specified level.
6. A step of receiving target image samples to be patterned on the wafer for wafer global alignment from an external device through a receiving unit, wherein the target image samples include a successful target image in which the global alignment is successful and a failed target image in which the global alignment is failed, wherein the successful target image includes a success label and the failed target image includes a failure label; A step of performing learning on successful target images and failed target images based on the target image samples through a deep learning unit, and creating or updating a target image model according to the learning performance; and A method of operation of a wafer global alignment learning device comprising the step of transmitting the target image model through a transmitting unit to the external device where the global alignment is performed.
7. In Paragraph 6, The method further includes the step of checking, through a sample count verification unit, whether the number of target image samples received from the external device is sufficient for the deep learning unit to train in generating the target image model, The step of verifying the number of samples through the sample number verification unit is, If it is determined that the number of the above target image samples is sufficient for training, the above target image samples are transferred to the deep learning unit, and A method of operation of a wafer global alignment learning device that commands the receiver to receive additional target image samples from the external device when it is determined that the number of target image samples is insufficient for learning.
8. In Paragraph 7, The method further includes a step of determining the performance of the target image model generated through the deep learning unit through a quality verification unit, wherein The step of determining performance through the above-mentioned quality verification unit is, Through the above target image model, compare whether the labeling for the target image matches the labeling for the image used as the target image sample, and If it is determined that the number of successful attempts exceeds a specified level, the transmitting unit is commanded to transmit the target image model to the external device, and A method of operation of a wafer global alignment learning device that, if it is determined that the number of successful attempts does not exceed a specified level, commands the receiver to additionally receive the target image samples from the external device.
9. In Paragraph 6, The step of improving the performance of the target image model through a performance improvement block is further included, The step of improving the performance of the target image model through the above performance improvement block is: A step of testing whether the target image model is functioning normally by applying the target image model generated by the deep learning unit to the target image samples received from the external device through the test unit; A step of determining, through a first quality verification unit, whether the quality of the target image model is lower than a predetermined value according to the test, wherein if the quality of the target image model is determined to be high, no additional function is performed, and if the quality of the target image model is determined to be low, an update command is issued to update the target image model for the target image samples; and A method of operation of a wafer global alignment learning device comprising the step of generating an updated target image model by performing learning based on the target image samples according to the update command through an update unit.
10. In Paragraph 9, The performance enhancement block further includes a step of verifying the quality of the updated target image model through a second quality verification unit, wherein The step of verifying the quality of the updated target image model through the second quality verification unit comprises comparing whether the labeling assigned to the target image through the updated target image model matches the labeling for the image used as the target image sample, and If it is determined that the number of successful attempts exceeds a specified level, a command is given to transmit the updated target image model to the external device, and A method of operation of a wafer global alignment learning device that, if it is determined that the number of successful attempts does not exceed a specified level, commands the device to additionally receive the target image samples from the external device.
