An image recognition method, apparatus, and system for wafer bonding
By acquiring and processing multiple template images to create mean and sum images, and utilizing geometric recognition algorithms and template matching methods, the impact of camera signal-to-noise ratio on wafer bonding positioning accuracy was resolved, achieving high-precision image positioning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PIOTECH (HAINING) SEMICON EQUIP CO LTD
- Filing Date
- 2021-12-28
- Publication Date
- 2026-07-31
AI Technical Summary
In existing technologies, the impact of camera signal-to-noise ratio on wafer bonding image positioning accuracy has not been fully considered, resulting in low positioning accuracy.
By acquiring multiple template images, mean and sum images are created. Then, geometric recognition algorithms and template matching methods are used to identify template regions, thereby improving the signal-to-noise ratio and positioning accuracy.
It effectively reduces image recognition errors caused by noise and improves image positioning accuracy to the sub-pixel level.
Smart Images
Figure CN116403101B_ABST
Abstract
Description
Technical Field
[0001] This application generally relates to the field of semiconductor technology, and more specifically, to an image recognition method, apparatus, and system for wafer bonding. Background Technology
[0002] In the manufacturing or process inspection of large-scale integrated circuits in semiconductors, semiconductor wafers undergo multiple manufacturing stages and are processed or inspected by various different devices. Wafer bonding technology has become a key technology for the integrated development and practical application of semiconductor manufacturing technology. Wafer bonding refers to the process of bonding two flat wafers face to face and applying certain external conditions such as pressure, temperature, and voltage to generate atomic or molecular bonding forces at the interface between the two wafers, such as covalent bonds, metallic bonds, and molecular bonds. This allows the bonding strength between the two surfaces to reach a certain level, thus making the two wafers a single unit.
[0003] In wafer bonding processes, two wafers need to be aligned and bonded. This is typically achieved using a template image. However, existing methods for wafer positioning using template images focus on improving optical resolution to achieve sub-pixel level positioning, while neglecting the impact of camera signal-to-noise ratio on positioning accuracy.
[0004] Therefore, it is necessary to improve the existing solutions in order to solve the problems existing in the existing technology. Summary of the Invention
[0005] One of the objectives of this application is to provide an image recognition method, apparatus, and system for wafer bonding that solves the problem of low image positioning accuracy caused by camera signal-to-noise ratio.
[0006] According to one aspect of this application, this application provides an image recognition method for wafer bonding, comprising: acquiring multiple first template images, each of the multiple first template images including a first template region; creating a mean image based on the multiple first template images; acquiring multiple second template images, each of the multiple second template images including a second template region; creating a sum image based on the multiple second template images; and recognizing the mean image and the sum image to determine the position of the second template region in the sum image.
[0007] According to one aspect of this application, an image recognition apparatus for wafer bonding is provided, comprising: a first acquisition module configured to acquire a plurality of first template images, each of the plurality of first template images including a first template region; a first processing module configured to generate an average image based on the plurality of first template images; a second acquisition module configured to acquire a plurality of second template images, each of the plurality of second template images including a second template region; a second processing module configured to generate a sum image based on the plurality of second template images; and a recognition module that recognizes the average image and the sum image to determine the position of the second template region in the sum image.
[0008] According to one aspect of this application, this application also provides a non-volatile computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the image recognition method for wafer bonding as described in this application.
[0009] According to one aspect of this application, an image recognition system for wafer bonding is also provided, comprising: a processor; a non-volatile computer-readable storage medium storing computer-executable instructions coupled to the processor; and a stage for supporting a wafer; wherein the processor is configured to execute the computer-executable instructions to implement the image recognition method for wafer bonding described in this application on the wafer. The image recognition method, apparatus, and system for wafer bonding provided in this application can effectively overcome the influence of camera signal-to-noise ratio on image positioning accuracy, improve image positioning accuracy, and reduce image recognition accuracy errors. Attached Figure Description
[0010] The accompanying drawings, necessary for describing embodiments of this application or the prior art, will be briefly described below to facilitate the depiction of embodiments of this application. It is obvious that the drawings described below represent only a portion of the embodiments in this application. Those skilled in the art will be able to derive other embodiments from the illustrations in these drawings without requiring inventive effort.
[0011] Figure 1 This is a flowchart of an image recognition method for wafer bonding according to some embodiments of this application.
[0012] Figure 2 This is a flowchart illustrating the specific process of acquiring multiple first template images according to some embodiments of this application.
[0013] Figure 3 This is a flowchart illustrating the specific process of acquiring multiple second template images according to some embodiments of this application.
[0014] Figure 4 This is a flowchart of an image recognition method for wafer bonding according to other embodiments of this application.
[0015] Figure 5 This is a structural block diagram of an image recognition device for wafer bonding according to some embodiments of this application. Detailed Implementation
[0016] To better understand the spirit of this application, the following description is based on some preferred embodiments of this application.
[0017] Various embodiments of this application are discussed in detail below. Although specific embodiments are discussed, it should be understood that these embodiments are for illustrative purposes only. Those skilled in the art will recognize that other components and configurations can be used without departing from the spirit and scope of this application.
[0018] Figure 1 This is a flowchart of an image recognition method for wafer bonding according to some embodiments of this application. The following is in conjunction with... Figures 2 to 3 To describe in detail, such as Figure 1 The image recognition method for wafer bonding shown in some embodiments of this application is illustrated.
[0019] In some embodiments, multiple first template images are acquired (step S10). Each of the multiple first template images has a first template region. Figure 2 This is a flowchart illustrating the specific process of acquiring multiple first template images according to some embodiments of this application.
[0020] In some embodiments, a first template image is acquired (step S102).
[0021] In some embodiments, a first template region is searched in the first template image (step S104). The first template region can be searched by any known method, such as an automatic search method or a manual search method.
[0022] In some embodiments, the first template region is positioned at the center of the first template image (step S106). Various methods can be used to position the first template region at the center of the first template image in this step. For example, the center of the first template region can be moved to the center of the first template image; or the boundary of the first template image can be cropped so that the distance between the edge of the first template region and the boundary of the cropped first template image is the same, thereby positioning the first template region at the center of the first template image. However, in some embodiments, step S106 is not mandatory and can be omitted.
[0023] In some embodiments, the image sensor is focused on the first template region (step S108). In some embodiments, the image sensor may be a camera. By setting the position and size of the camera's ROI (Region of Interest), the ROI of the camera is made to include the first template region. The main purpose of focusing the image sensor on the first template region is to filter out regions of no interest, thereby improving the processor's operating speed. However, in some embodiments, step S108 may be omitted if the processor's operating speed is not a consideration.
[0024] In some embodiments, the first template region is exposed multiple times (step S110). In this step, an image sensor (e.g., a camera) is used to expose the first template region of the first template image multiple times to obtain multiple first template images.
[0025] See Figure 1 and Figure 2 In some embodiments, an average image is created based on multiple first template images (step S20). In this step, the pixel values at the same location in the multiple first template images are summed up, and their average value is calculated to obtain the average image. In some embodiments, the first template region is exposed 16 times in step S110 to obtain 16 first template images. The average image of the 16 first template images is obtained by summing up the pixel values at the same location in each of these 16 first template images, dividing by 16, and calculating the average value. It should be understood that 16 first template images are merely a preferred embodiment of this application; in some embodiments, the number of first template images can be any other, and this is not limited thereto. By creating an average image, noise in the template images can be reduced, and the signal-to-noise ratio can be improved.
[0026] In some embodiments, multiple second template images are acquired (step S30). Each of the multiple second template images has a second template region. Figure 3 This is a flowchart illustrating the specific process of acquiring multiple second template images according to some embodiments of this application.
[0027] In some embodiments, a second template image is acquired (step S302).
[0028] In some embodiments, the second template image is exposed multiple times (step S304). In this step, the second template image is exposed multiple times using an image sensor (e.g., a camera) to obtain multiple second template images.
[0029] See back Figure 1In some embodiments, a sum-value image is created based on multiple second template images (step S40). In this step, the pixel values at the same location in the multiple second template images acquired in step S30 are summed to create a sum-value image. In some embodiments, when creating the sum-value image, it is necessary to consider preventing data overflow due to insufficient image memory. In some embodiments, the second template image is exposed 16 times in step S304 to obtain 16 second template images. However, it should be understood that the number 16 is merely a preferred embodiment of this application, and the number of second template images can be any other number. By creating a sum-value image based on multiple second template images, the low edge contrast and blurred edges of the template image caused by low illumination intensity can be compensated for, thereby improving the signal-to-noise ratio of the image and accurately identifying the results.
[0030] In some embodiments, the mean image and the sum image are identified (step S50). Specifically, the mean image and the sum image are identified to determine the position of the second template region in the sum image. In this application, various methods can be used to identify the mean image and the sum image. For example, in some embodiments, a geometric recognition algorithm can be used to identify the mean image and the sum image, or a template matching method can be used.
[0031] Figure 4 This is a flowchart of an image recognition method for wafer bonding according to some embodiments of this application.
[0032] In some embodiments, multiple sum images are created (in step S402). Specifically, this is repeated. Figure 1 Steps S30 and S40 are performed to obtain multiple sum value images. In some embodiments, this process is repeated. Figure 1 Steps S30 and S40 are performed a total of 50 times to obtain 50 sum value images. However, it should be understood that 50 times is only a preferred embodiment of this application, and in some embodiments, other numbers of times may be repeated. Figure 1 Steps S30 and S40 are not specifically limited here.
[0033] In some embodiments, each of the mean image and multiple sum images is identified (step S502). Specifically, this is repeated. Figure 1 Step S50 in the process is used to obtain multiple sets of recognition results.
[0034] In some embodiments, the variance of the recognition results is calculated (step S60). Specifically, the variance of multiple sets of recognition results is calculated to determine the degree of dispersion of the multiple sets of recognition results.
[0035] The following uses geometric recognition algorithms to analyze... Figure 4The image recognition method for wafer bonding is illustrated in the flowchart below. It should be understood that the following geometric recognition algorithm steps are merely one specific embodiment of this application. In some embodiments, the geometric recognition algorithm may have other variations or parameter adjustments, which are not specifically limited here. The geometric recognition algorithm steps of a specific embodiment of this application are as follows:
[0036] First, define the first template region as the area within the first template image. Second, allocate a cache for the calculated score and angle deflection result of the first template region; the size of the cache is the same as the size of the first template region. Third, control the recognition accuracy, using high-precision recognition to ensure the error is within 0.05 pixels (sub-pixel level). Fourth, load the second template image and then use the first template image for recognition, calculating the recognition result. Fifth, load different second template images and repeat step four consecutively, summarizing the multiple sets of recognition results. Sixth, calculate the variance of the summed multiple sets of recognition results by a factor of three. Using a factor of three here allows for a clearer understanding of the dispersion of the multiple sets of recognition results.
[0037] Table 1 shows the comparison results of three times the variance obtained after exposing the second template image to different magnifications and number of exposures.
[0038] Table 1
[0039] Magnification Exposure times 3*σ calculation results (unit: nm) ×3.6 1 43.87754 ×3.6 16 14.99986 ×7.2 1 39.56873 ×7.2 16 11.21481
[0040] As shown in Table 1, at a magnification of 3.6x: when the second template image is exposed only once and this step is performed multiple times, the variance of the multiple recognition results is relatively large, at 43.87754 nm, indicating a high degree of dispersion in these recognition results. However, when the second template image is exposed 16 times and a summation image is created, and this step is performed multiple times, the variance of the multiple recognition results decreases significantly, to only 14.99986 nm. Similarly, at a magnification of 7.2x: when the second template image is exposed only once and this step is performed multiple times, the variance of the multiple recognition results is also relatively large, at 39.56873, indicating a high degree of dispersion in these recognition results. However, when the second template image is exposed 16 times and a summation image is created, and this step is performed multiple times, the variance of the multiple recognition results decreases further significantly, to only 11.21481. In other words, regardless of whether the magnification is 3.6x or 7.2x, after exposing the second template image 16 times to create the sum image and repeating this step multiple times, the variance of the multiple recognition results obtained is all below 15nm. That is to say, the error of the image recognition result is reduced to approximately 15nm.
[0041] Figure 5This is a structural block diagram of an image recognition apparatus for wafer bonding according to some embodiments of this application. Figure 5 As shown, the image recognition device 50 for wafer bonding includes a first acquisition module 502, a first processing module 504, a second acquisition module 506, a second processing module 508, an identification module 510, and a determination module 512. The image recognition method for wafer bonding of this application can be implemented by the image recognition device 50 for wafer bonding of this application. Specifically, step S10 can be implemented by the first acquisition module 502; step S20 can be implemented by the first processing module 504; step S30 can be implemented by the second acquisition module 506; step S40 can be implemented by the second processing module 508; step S50 can be implemented by the identification module 510; and step S60 can be implemented by the determination module 512. That is, the image recognition device 50 for wafer bonding can implement any step of any method described herein via the first acquisition module 502, the first processing module 504, the second acquisition module 506, the second processing module 508, the identification module 510, and the determination module 512. Figures 1 to 4 The method steps are described below.
[0042] Other embodiments of this application relate to a non-volatile computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement any step of any method described herein, including Figures 1 to 4 The method steps described herein. The program can be implemented in any of a variety of ways, including program-based technologies, component-based technologies, and / or object-oriented technologies. For example, the program may be implemented using ActiveX controls, C++ objects, JavaBeans, Microsoft Foundation Classes (MFC), Streaming SIMD Extensions (SSE), or other technologies or methods as needed.
[0043] Additionally, some embodiments of this application provide an image recognition system for wafer bonding. The system includes a processor, a non-volatile computer-readable storage medium storing computer-executable instructions, and a stage. The non-volatile computer-readable storage medium storing the computer-executable instructions is coupled to the processor. The stage can be used to support a wafer. The processor is configured to execute the computer-executable instructions to implement any step of any of the methods described herein on the wafer, including... Figures 1 to 4 The method steps are described below.
[0044] The image recognition method, apparatus and system for wafer bonding provided in this application have at least the following advantages: (1) by making an average image, the noise in the first template image is reduced, which effectively reduces the image recognition error caused by noise; and (2) by accumulating the second template image, the signal-to-noise ratio of the image is improved, the recognition result is accurate, and the recognition accuracy is improved, so that the image recognition accuracy can reach the sub-pixel level standard.
[0045] It should be noted that throughout this specification, the reference to "some embodiments of this application" or similar terms means that a particular feature, structure, or characteristic described together with other embodiments is included in at least one embodiment and may not necessarily be presented in all embodiments. Therefore, the corresponding appearance of the phrase "some embodiments of this application" or similar terms throughout this specification does not necessarily refer to the same few embodiments. Furthermore, the particular features, structures, or characteristics of any particular embodiment may be combined with one or more other embodiments in any suitable manner.
[0046] The technical content and features of this invention have been disclosed above. However, those skilled in the art may still make various substitutions and modifications based on the teachings and disclosures of this invention without departing from the spirit of this invention. Therefore, the scope of protection of this invention should not be limited to the content disclosed in the embodiments, but should include various substitutions and modifications that do not depart from this invention, and should be covered by the claims of this patent application.
Claims
1. An image recognition method for wafer bonding, characterized in that, The image recognition method for wafer bonding includes: Acquire multiple first template images, each of which includes a first template region; A mean image is generated based on the multiple first template images; Multiple second template images are repeatedly acquired, and multiple sum images are repeatedly created based on the multiple second template images acquired each time, wherein each of the multiple second template images includes a second template region; Each of the mean image and the plurality of sum images is identified to obtain multiple sets of identification results; and The variance of the multiple sets of recognition results is calculated to determine the position of the second template region in the sum image.
2. The image recognition method for wafer bonding according to claim 1, wherein, The collection of multiple first template images includes: Focus the image sensor on the first template region of the first template image; and The first template region of the first template image is exposed multiple times to obtain the multiple first template images.
3. The image recognition method for wafer bonding according to claim 2, wherein, Acquiring multiple first template images further includes: before focusing the image sensor on the first template region of the first template image, positioning the first template region of the first template image at the center of the first template image.
4. The image recognition method for wafer bonding according to claim 1, wherein, The collection of multiple second template images includes: The second template image is exposed multiple times to obtain the multiple second template images.
5. The image recognition method for wafer bonding according to claim 1, wherein, The mean image and the sum image are identified based on a geometric recognition algorithm.
6. The image recognition method for wafer bonding according to claim 1, wherein, The shape and size of the first template area are the same as those of the second template area.
7. An image recognition apparatus for wafer bonding, characterized by, The image recognition device for wafer bonding includes: A first acquisition module is configured to acquire multiple first template images, each of which includes a first template region; A first processing module is configured to generate a mean image based on the plurality of first template images; The second acquisition module is configured to repeatedly acquire multiple second template images, each of which includes a second template region; The second processing module is configured to repeatedly generate multiple sum images based on the multiple second template images acquired each time; A recognition module configured to recognize each of the mean image and the plurality of sum images to obtain multiple sets of recognition results; and A determination module, configured to calculate the variance of the multiple sets of recognition results, to determine the location of the second template region in the sum image.
8. The image recognition apparatus for wafer bonding according to claim 7, wherein, The first processing module is further configured to: Focus the image sensor on the first template region of the first template image; and The first template region of the first template image is exposed multiple times to obtain the multiple first template images.
9. The image recognition apparatus for wafer bonding according to claim 8, wherein, The first processing module is further configured such that the first template region of the first template image is located at the center of the first template image.
10. The image recognition apparatus for wafer bonding according to claim 7, wherein, The second acquisition module is further configured to expose the second template image multiple times to obtain the multiple second template images.
11. The image recognition device for wafer bonding according to claim 7, characterized in that, The mean image and the sum image are identified based on a geometric recognition algorithm.
12. The image recognition device for wafer bonding according to claim 7, characterized in that, The shape and size of the first template area are the same as those of the second template area.
13. A non-volatile computer-readable storage medium, characterized in that, The non-volatile computer-readable storage medium stores one or more programs that can be executed by one or more processors to implement the image recognition method for wafer bonding as described in any one of claims 1-6.
14. An image recognition system for wafer bonding, comprising: processor; A non-volatile computer-readable storage medium storing computer-executable instructions, coupled to the processor; as well as A stage, used to support a wafer; The processor is configured to execute the computer-executable instructions to implement the image recognition method for wafer bonding according to any one of claims 1-6 on the wafer.