A method for determining a detection area, a wafer detection method and device

Abstract

This invention discloses a method for determining a detection area, a wafer inspection method, and an apparatus. The method for determining the detection area includes: determining a geometric image based on the design data of the target wafer, wherein the geometric image includes image features of the locations on the target wafer to be inspected; acquiring an optical image of the target wafer; performing mode conversion on the geometric image to obtain a template image, wherein the mode of the template image is the same as the mode of the optical image; and determining a region in the optical image that matches the template image to obtain the detection area. The solution provided by this invention is applicable to wafers with different design rules, and can automatically, quickly, and accurately determine the detection area on the wafer, providing a basis for subsequent defect detection and process optimization, thereby ensuring wafer yield and production efficiency.

Description

Technical Field

[0001] This invention relates to the field of semiconductor inspection technology, and in particular to a method for determining the inspection area, a wafer inspection method, and an apparatus. Background Technology

[0002] As semiconductor technology continues to advance, its manufacturing processes are becoming increasingly complex. For example, the continuous shrinking of design rules makes wafers more prone to defects during manufacturing, thus placing higher demands on wafer inspection.

[0003] Most existing wafer inspection methods are optimized with the aim of improving the overall performance of the inspection equipment. The main inspection methods include: first, manually drawing the inspection area (also known as the area of ​​interest (CA)); then, using methods such as region-based multi-thresholding (RBMT) to optimize the inspection area, thereby enhancing the defect identification capability of the inspection area and suppressing noise interference; finally, using higher-performance inspection equipment to perform defect inspection on the optimized inspection area, and continuously adjusting the inspection method based on the feedback of the inspection results.

[0004] However, manually drawing the inspection area is time-consuming and error-prone, directly impacting wafer production efficiency. Furthermore, as wafer design complexity increases, manually drawing the inspection area cannot quickly adapt to changes in design rules and the demands of rapid iteration. Summary of the Invention

[0005] This invention provides a method for determining the detection area, a wafer inspection method, and an apparatus applicable to wafers with different design rules. It can automatically, quickly, and accurately determine the detection area on the wafer, providing a basis for subsequent defect detection and process optimization, thereby ensuring wafer yield and production efficiency.

[0006] According to one aspect of the present invention, a method for determining a detection region is provided, comprising: determining a geometric image based on design data of a target wafer, wherein the geometric image includes image features of the location to be detected on the target wafer; acquiring an optical image of the target wafer; performing mode conversion on the geometric image to obtain a template image, wherein the mode of the template image is the same as the mode of the optical image; and determining a region in the optical image that matches the template image to obtain the detection region.

[0007] According to another aspect of the present invention, a wafer inspection method is provided, comprising: obtaining a detection area of ​​a target wafer using a detection area determination method according to any embodiment of the present invention; dividing the detection area into at least one detection set, wherein detection areas located in the same detection set correspond to the same detection label, the detection label being used to indicate the defect type and / or defect characteristics that may occur in the target wafer; and inspecting the detection set using a detection method corresponding to the detection label of the detection set to determine the defects of the target wafer.

[0008] According to another aspect of the present invention, a wafer inspection apparatus is provided, comprising: a data processing system and an image acquisition system; the image acquisition system is used to acquire optical images of a target wafer; the data processing system includes an image analysis module, an image conversion module, an image matching module, a calibration module, and a defect identification module; the image analysis module is used to determine a geometric image based on design data of the target wafer, wherein the geometric image includes image features of the location to be inspected on the target wafer; the image conversion module is used to perform mode conversion on the geometric image to obtain a template image, wherein the mode of the template image is the same as the mode of the optical image; the image matching module is used to determine a region in the optical image that matches the template image to obtain a detection region; the calibration module is used to divide the detection region into at least one detection set, wherein detection regions located in the same detection set correspond to the same detection label, the detection label being used to indicate the defect type and / or defect feature that may occur in the target wafer; and the defect identification module is used to inspect the detection set using a detection method corresponding to the detection label of the detection set to determine the defects of the target wafer.

[0009] The technical solution of this invention determines a geometric image based on the design data of a target wafer, including image features of the locations to be detected on the target wafer. Further, it performs mode conversion on the geometric image to obtain a template image, ensuring that the mode of the template image is the same as the mode of the acquired optical image of the target wafer. This allows for the determination of a region in the optical image that matches the template image, thus obtaining the detection region. Compared to traditional methods of manually drawing detection regions, firstly, by determining the geometric image based on the design data of the target wafer and then performing mode conversion to obtain the template image, a fully automatic conversion from design data to geometric image and then to template image is achieved, enabling this solution to adapt to wafers with different design rules. Secondly, since the mode of the template image is the same as the mode of the optical image, the problem of different modes weakening the matching accuracy during subsequent image matching can be avoided, thereby improving image matching accuracy and ultimately improving the accuracy of identifying the detection region in the optical image. Thirdly, matching optical images with template images to determine the detection area can automate the determination of the detection area, ensuring the efficiency of area determination, avoiding human errors that may occur when manually drawing the detection area, providing a basis for subsequent defect detection and process optimization, and thus ensuring the yield and production efficiency of the wafer.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating a method for determining a detection area provided in Embodiment 1 of the present invention;

[0013] Figure 2 This is a flowchart illustrating a method for determining a detection area according to Embodiment 2 of the present invention;

[0014] Figure 3 This is a schematic diagram of a geometric image provided in Embodiment 2 of the present invention;

[0015] Figure 4 This is a schematic diagram of a method for inputting a geometric image into an image conversion model to obtain a template image, as provided in Embodiment 2 of the present invention;

[0016] Figure 5 This is a region matching effect diagram provided in Embodiment 2 of the present invention;

[0017] Figure 6 This is a schematic flowchart of a wafer inspection method provided in Embodiment 3 of the present invention;

[0018] Figure 7 This is a schematic diagram of the structure of a wafer inspection device provided in Embodiment 4 of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] It should be noted that the terms "target," "training," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] Example 1

[0022] Figure 1 This is a flowchart illustrating a method for determining a detection area according to Embodiment 1 of the present invention. This embodiment is applicable to determining a detection area in an optical image of a target wafer. The method can be executed by a detection area determination device, which can be implemented in hardware and / or software and can be configured in electronic equipment (such as computer equipment, wafer manufacturing equipment, wafer inspection equipment, etc.). Figure 1 As shown, the method includes:

[0023] S110. Based on the design data of the target wafer, determine the geometric image, wherein the geometric image includes image features of the locations on the target wafer that need to be detected.

[0024] To determine whether a wafer has defects during manufacturing, it is typically inspected. In this invention, the target wafer refers to a specific type of wafer that needs to be inspected. When multiple types of wafers need to be inspected, each type can be used as the target wafer, and the method for determining the inspection area provided by this invention can be executed once for each type.

[0025] The design data for a target wafer reflects various parameters, layout design data, and dimensional characteristics of the target wafer's layer structure during the design process. Design data is typically stored in the form of Graphic Data System (GDS) files or Computer-Aided Design (CAD) files. GDS files are binary format files, while CAD files are files created using computer-aided design software. Both GDS and CAD files can describe the layout and circuit connections of the target wafer, such as various geometric shapes, hierarchical structures, circuit elements, and interconnections.

[0026] In one embodiment, target data corresponding to the location to be detected on the target wafer can be extracted from the design data, and then the target data can be parsed to obtain a geometric image.

[0027] In this embodiment, a location to be detected corresponds to a target data, and a target data corresponds to a geometric image.

[0028] For example, during the generation of design data, designers can annotate the locations that need to be detected, thereby extracting target data from the design data based on the annotation information.

[0029] In another embodiment, target data corresponding to the location to be detected on the target wafer can be extracted from the design data, the target data can be cropped according to a preset ratio, and finally the cropped target data can be parsed to obtain a geometric image.

[0030] In this embodiment, one location to be detected corresponds to one target data, and one target data corresponds to multiple geometric images.

[0031] Compared with the previous implementation method, cropping the target data according to a preset ratio divides the target data into multiple "small" data. The size of the geometric image will also be reduced accordingly, but the number will increase. This allows multiple geometric images to be processed in parallel in subsequent steps, improving the efficiency of determining the detection area.

[0032] Optionally, the preset ratio can be a fixed ratio that has been set, or it can be dynamically adjusted according to the target data.

[0033] S120: Acquire optical images of the target wafer.

[0034] Optical images can be acquired using wafer inspection equipment, such as optical microscopy or electron optical microscopy. This invention does not limit the format of the optical images; for example, the optical image can be an RGB image, a grayscale image, or a YUV image.

[0035] Optionally, after acquiring the optical image of the target wafer, the optical image can be preprocessed, such as by eliminating noise in the optical image, to improve the image matching accuracy in subsequent steps.

[0036] It should also be noted that the present invention does not restrict the execution order of steps S110 and S120: step S110 can be executed first, followed by step S120; step S120 can be executed first, followed by step S110; or steps S110 and S120 can be executed simultaneously.

[0037] S130. Perform mode transformation on the geometric image to obtain a template image, wherein the mode of the template image is the same as the mode of the optical image.

[0038] In this invention, the modality of an image includes, but is not limited to, at least one of the following: image color, image format, image resolution, image sharpness, spectrum, and shooting angle. Modality conversion of the geometric image to obtain a template image ensures that the modality of the template image is identical to that of the optical image, thereby avoiding the problem of different modalities weakening matching accuracy during subsequent image matching. After modality conversion, the modality of the template image and the optical image are completely identical.

[0039] For example, assume that the modality of an image includes image format, image resolution, and shooting angle. The converted template image 1 has a JPG format, an image resolution of 1600×1200, and a shooting angle of angle A. The acquired optical image 1 also has a JPG format, an image resolution of 1600×1200, and a shooting angle of angle A. It is evident that the modality of template image 1 and optical image 1 are completely identical, and the modality conversion is successful.

[0040] For example, suppose the modality of an image includes image color, image format, image resolution, and shooting angle. The converted template image 2 has a grayscale image color, a JPG format, a resolution of 1600×1200, and a shooting angle of angle B. The acquired optical image 2 has an RGB image color, a JPG format, a resolution of 1600×1200, and a shooting angle of angle A. It is clear that the modality of template image 2 and optical image 2 are not completely identical; the modality conversion fails, and a new modality conversion is required.

[0041] Specifically, the method for performing modal transformation on a geometric image to obtain a template image can be as follows: use a transformation tool to perform modal transformation on the geometric image to obtain a template image.

[0042] The conversion tool can be an open-source tool or the image conversion model established in this invention. The image conversion model is a model obtained by training a generative adversarial network with the goal of achieving image mapping.

[0043] Generative Adversarial Networks (GANs) are deep learning models that consist of a generator and a discriminator. The generator attempts to generate a template image similar to the optical image, while the discriminator tries to distinguish between the optical image and the template image. Through this adversarial process, the generator can continuously improve, eventually generating high-quality template images. Image mapping is achieved through the game-like learning between the generator and the discriminator. Compared to the modality transformation methods of ordinary open-source tools, GANs can better guarantee the success rate and quality of modality transformation.

[0044] The specific method for training a generative adversarial network to establish an image conversion model can be found in the description of Example 2 below, which will not be repeated here for the sake of brevity.

[0045] S140. Determine the region in the optical image that matches the template image to obtain the detection region.

[0046] In one embodiment, a region in the optical image with a similarity greater than a preset similarity to the template image can be identified, and this region can be used as the region that matches the template image to obtain the detection region.

[0047] In another embodiment, the template image can be slid across the optical image at a preset step size, and the correlation coefficient of the overlapping area between the template image and the optical image can be determined at each sliding; the overlapping area corresponding to the correlation parameter that satisfies the second preset condition is taken as the detection area.

[0048] Optionally, the preset step size can be set according to actual needs, such as 1 pixel, 2 pixels, 5 pixels, etc.

[0049] This invention provides a method for determining a detection region, comprising: determining a geometric image based on design data of a target wafer, wherein the geometric image includes image features of the location to be detected on the target wafer; acquiring an optical image of the target wafer; performing mode conversion on the geometric image to obtain a template image, wherein the mode of the template image is the same as the mode of the optical image; and determining a region in the optical image that matches the template image to obtain the detection region. The technical solution of this invention determines a geometric image including image features of the location to be detected on the target wafer based on the design data of the target wafer; further performs mode conversion on the geometric image to obtain a template image, making the mode of the template image the same as the mode of the acquired optical image of the target wafer; thereby determining a region in the optical image that matches the template image to obtain the detection region. Compared with the traditional method of manually drawing the detection region, firstly, by determining a geometric image based on the design data of the target wafer and then performing mode conversion on the geometric image to obtain a template image, a fully automatic conversion from design data to geometric image and then to template image is achieved, enabling this solution to adapt to wafers with different design rules. Secondly, since the template image has the same modality as the optical image, the problem of different modalities weakening the matching accuracy during subsequent image matching can be avoided, thereby improving image matching accuracy and thus improving the accuracy of identifying detection areas in optical images. Thirdly, matching optical images with template images to determine detection areas enables automated determination of detection areas, ensuring efficiency and avoiding human errors that may occur when manually drawing detection areas. This provides a foundation for subsequent defect detection and process optimization, thereby ensuring wafer yield and production efficiency.

[0050] Example 2

[0051] Figure 2 This is a flowchart illustrating a method for determining a detection region according to Embodiment 2 of the present invention. Based on Embodiment 1, this embodiment provides specific implementation methods for training a generative adversarial network to establish an image conversion model, determining the detection region, and subsequent defect detection. Figure 2 As shown, the method includes:

[0052] S210. Extract the target data corresponding to the location to be detected on the target wafer from the design data, wherein the design data is a GDS file or a CAD file.

[0053] The design data for the target wafer reflects various parameters, layout design data, and dimensional characteristics of the target wafer's layer structure during the design process. Design data is typically stored in the form of GDS files or CAD files.

[0054] Preferably, the design data is a GDS file.

[0055] During the design data generation process, designers can annotate the locations that need to be inspected, thereby extracting target data from the design data based on the annotation information. Locations requiring inspection may include: areas with complex layouts, locations prone to defects during manufacturing, etc.

[0056] The target data is in the same format as the design data. That is, if the design data is a GDS file, then the target data is also a GDS file; if the design data is a CAD file, then the target data is also a CAD file.

[0057] S220. The target data is cropped and analyzed according to a preset ratio to obtain a geometric image.

[0058] Since the target data corresponds to the location that needs to be detected on the target wafer, the resulting geometric image after cropping and parsing will also include the image features of the location that needs to be detected on the target wafer.

[0059] In one embodiment, the preset ratio can be a fixed, pre-defined ratio, or it can be dynamically adjusted based on the target data. After the target data is cropped and parsed, the resulting geometric image is multiple images in image format. This allows multiple geometric images to be processed in parallel in subsequent steps, improving the efficiency of detecting the region.

[0060] Optionally, the geometric image can be rectangular to facilitate data cropping and subsequent image matching. Figure 3 This is a schematic diagram of a geometric image provided in Embodiment 2 of the present invention.

[0061] S230: Acquire optical images of the target wafer.

[0062] Optical images can be acquired using a wafer inspection device, which can be an optical microscopy inspection device or an electron-optical microscopy inspection device. This invention does not limit the format of the optical images.

[0063] Optionally, after acquiring the optical image of the target wafer, the optical image can be preprocessed, such as by eliminating noise in the optical image, to improve the image matching accuracy in subsequent steps.

[0064] It should also be noted that the present invention does not restrict the execution order of steps S210-S220 and step S230: steps S210-S220 can be executed first, followed by step S230; steps S230 can be executed first, followed by steps S210-S220; or steps S210-S220 and step S230 can be executed simultaneously.

[0065] S240. Establish an image conversion model. The image conversion model is a model obtained by training a generative adversarial network with the goal of achieving image mapping.

[0066] Generative adversarial networks (GANs) consist of a generator and a discriminator. The generator attempts to produce a template image that resembles the optical image, while the discriminator tries to distinguish between the optical image and the template image. Through this adversarial process, the generator can continuously improve and eventually produce a high-quality template image.

[0067] The generator and discriminator employ nonlinear activation functions, such as any one of ReLU (Rectified Linear Unit), Leaky ReLU (Leaky Rectified Linear Unit), or ELU (Exponential Linear Unit).

[0068] ReLU is a widely used activation function in deep learning. Its purpose is to introduce non-linearity into the neuron output of a neural network, which is necessary for learning complex function mappings. ReLU is computationally simple and efficient, but its output is zero in the negative region, which can limit training accuracy. Leaky ReLU is a variant of ReLU that allows negative inputs to be multiplied by a small non-zero slope as they pass through the neural network, instead of setting them completely to zero like standard ReLU. This design helps solve the "dead ReLU" problem that can occur in ReLU, where a neuron's gradient will remain zero indefinitely when its input is consistently negative, preventing the parameters of that neuron and its subsequent networks from being updated. ELU improves upon the ReLU function, with the advantage that its mean negative output is closer to zero, reducing the bias effect, providing a more stable gradient, and promoting faster network convergence and improved performance.

[0069] When processing fine structures in semiconductor images, the ELU activation function can more effectively capture complex features, improving the accuracy and quality of image conversion. Therefore, this invention preferably uses ELU as the nonlinear activation function for both the generator and discriminator.

[0070] Specifically, the method for training a generative adversarial network to build an image translation model includes the following five steps.

[0071] Step 1: Obtain the dataset, which includes training geometric images of multiple regions on the target wafer and measured optical images of the corresponding regions.

[0072] The training geometric images and the corresponding measured optical images of the regions can be stored in two separate subsets for easier image management. For example, subset A can be used to store the training geometric images, and subset B can be used to store the corresponding measured optical images of the regions.

[0073] Step 2: Input multiple training geometric images from the dataset into the generator. The generator obtains the theoretical optical image corresponding to each training geometric image based on the semantic segmentation model.

[0074] Multiple training geometric images from subset A are input into the generator. The generator simulates the theoretical optical image corresponding to each training geometric image based on the semantic segmentation model, so that the modality of the theoretical optical image is the same as that of the measured optical image.

[0075] The generator can be based on the U-Net or V-Net network architecture, which is a classic semantic segmentation model that can effectively process the detailed information of images.

[0076] V-Net employs 3D convolutional operations, capturing the spatial features of volumetric data through a fully convolutional architecture, and replaces skip connections with residual connections to enhance training stability and gradient flow. U-Net features an encoder-decoder structure and skip connections, whose skip connections enable more efficient feature preservation and resolution recovery in the generator, making it more suitable for accurate image modality conversion. Therefore, the generator of this invention preferably uses the U-Net network architecture.

[0077] Furthermore, the number of encoder channels in the generator is greater than 1024, while the number of decoder channels is gradually reduced to 1024. This involves introducing larger convolutional kernels and a denser channel design. In the encoder part, the number of channels is increased from 1024 to 2048, and then gradually reduced back to 1024 in the decoder. This design allows the model to learn richer, finer-grained feature representations, capturing more detailed information. During upsampling, a denser channel fusion strategy, such as adding skip connections and multi-scale feature fusion, ensures effective recovery of high-resolution features. This structural optimization is particularly important for processing complex and subtle semiconductor images, not only enhancing the model's ability to process local and global information but also effectively recovering high-resolution images while maintaining feature accuracy, ensuring more precise and detailed image mode transformation.

[0078] Step 3: The discriminator distinguishes between the theoretical optical image and the measured optical image of the corresponding region, and updates the parameters of the generator and the discriminator to guide the generator to gradually generate candidate theoretical optical images that meet the initial conditions.

[0079] The discriminator can be based on a convolutional neural network, such as the PatchGAN network architecture, capable of identifying theoretical optical images and measured optical images of corresponding regions, determining which image is real and which is fake. Optionally, the discriminator can process the entire image at the granular level during the recognition process, or it can divide the image into several small regions (paths) and process them at the granular level of the small regions.

[0080] The discriminator updates the parameters of the generator and discriminator by distinguishing between theoretical optical images and measured optical images of the corresponding regions, thereby guiding the generator to gradually generate candidate theoretical optical images that meet the initial conditions. For example, the initial condition can be that the loss functions of the generator and discriminator no longer decrease or tend to stabilize during the parameter update process, or a preset threshold for the number of parameter updates can be used as the initial condition. The theoretical optical image obtained by the generator when the initial condition is met is then used as the candidate theoretical optical image for this round of training.

[0081] Furthermore, a random dropout layer is configured in the discriminator. This layer is used during training to randomly discard neurons and their connections with a preset probability. This effectively suppresses overfitting and improves the network's generalization ability. By randomly "dropping" neurons and their connections with a preset probability during training, the random dropout layer prevents the model from over-relying on specific neurons, prompting the network to learn more robust feature representations. This randomness disrupts the accidental correlations between neurons, making the model more adaptable to unseen data and increasing its sensitivity to differences in input data. In image modality conversion tasks, many structural details are highly similar, easily leading to overfitting of real data and ignoring generated data. By introducing a random dropout layer in the discriminator, the model can more accurately distinguish subtle differences between real and generated images, enhancing its image recognition ability and ensuring that the converted image is more realistic and consistent in visual effects and optical properties.

[0082] Step 4: Calculate the image evaluation value based on the candidate theoretical optical image and the measured optical image of the corresponding region.

[0083] In one implementation, the image evaluation value includes the peak signal-to-noise ratio (PSNR), which is calculated using the following formula:

[0084] Where MAX is the maximum pixel value that can be obtained in the optical image, determined by the hardware parameters of the imaging unit of the detection region determination device. For example, for an 8-bit image, MAX = 255. MSE represents the mean square error between the candidate theoretical optical image and the corresponding measured image. MSE is calculated using the following formula: M and N represent the length and width of the image, respectively, and I(i,j) and These are the pixel values ​​at position (i,j) of the candidate theoretical optical image and the measured optical image of the corresponding region, respectively.

[0085] Step 5: When the image evaluation value meets the first preset condition, stop training and use the generative adversarial network at this time as the image conversion model.

[0086] When the image evaluation value meets the first preset condition, it means that the generative adversarial network has been trained and training is stopped. The generator of the generative adversarial network at this time can be used as the image conversion model, and the candidate theoretical optical image generated at this time can be used as the template image.

[0087] For example, assuming the image evaluation value includes the image peak signal-to-noise ratio (PSNR), the first preset condition can be that the image PSNR is greater than or equal to a preset PSNR, such as 20.

[0088] Step 6: When the image evaluation value does not meet the first preset condition, perform image processing on the measured optical images in the dataset, and retrain the generative adversarial network based on the processed measured optical images and the training geometric images of the corresponding regions until the image evaluation value meets the first preset condition.

[0089] When the image evaluation value does not meet the first preset condition, it indicates that the candidate theoretical optical image obtained by the generative adversarial network still does not meet the requirements and training needs to continue. At this time, image processing is performed on the measured optical images in the dataset, and the loss functions (such as adversarial loss and / or L1 loss) of the generator and discriminator are determined. The loss function is used to guide the parameter update of the generator and discriminator. Then, the generative adversarial network is retrained based on the processed measured optical images and the training geometric images of the corresponding regions until the image evaluation value meets the first preset condition.

[0090] Optionally, the image processing includes at least one of the following processing steps: improving the brightness of the measured optical image, improving the contrast of the measured optical image, and improving the sharpness of the measured optical image.

[0091] S250. Input the geometric image into the image conversion model to obtain the template image.

[0092] After the image conversion model is trained, the geometric image is input into the model to complete the modality conversion of the image and obtain the template image. The modality of the template image is the same as that of the optical image. Figure 4 This is a schematic diagram of a geometric image input into an image conversion model to obtain a template image, as provided in Embodiment 2 of the present invention.

[0093] S260. Slide the template image on the optical image according to a preset step size, and determine the correlation coefficient of the overlapping area between the template image and the optical image at each slide.

[0094] Optionally, the preset step size can be set according to actual needs, such as 1 pixel, 2 pixels, 5 pixels, etc.

[0095] Preferably, the preset step size is 1 pixel.

[0096] The template image is slid across the optical image at preset step sizes, and the correlation coefficient of the overlapping area between the template image and the optical image is determined at each slide. The correlation coefficient at each slide is calculated using the following formula:

[0097] Wherein, NCC(x,y) is the normalized cross-correlation NCC value at position (x,y) in the overlapping region of template image T and optical image I, T i,j Let I be the pixel value of the template image T at position (i,j). x+i,y+i Let (x+1, y+1) be the pixel value of the sub-window in the optical image I relative to position (x, y) during sliding. The average pixel value of the template image T. This represents the average pixel value of the overlapping region of optical image I. m and n are the width and height of the template image, respectively.

[0098] In the above formula for calculating the correlation coefficient, The cross-correlation between the overlapping regions of the template image T and the optical image I is used to measure the similarity between the two images; and This is used to calculate the normalized cross-correlation value between the overlapping areas of the template image T and the optical image I, making it unaffected by brightness and contrast.

[0099] S270. The overlapping area corresponding to the relevant parameters that meet the second preset condition is taken as the detection area.

[0100] The second preset condition can be that the correlation coefficient is greater than or equal to the cross-correlation threshold. The value of the cross-correlation threshold can be set according to actual needs, such as 0.8 or 0.9.

[0101] When the correlation coefficient of the overlapping region between the template image and the optical image is greater than or equal to the cross-correlation threshold, it indicates that the overlapping region matches the template image and is a detection region. When the correlation coefficient of the overlapping region between the template image and the optical image is less than the cross-correlation threshold, it indicates that the overlapping region does not match the template image and is not a detection region.

[0102] It should be noted that the cross-correlation threshold should not be too small, because there may be cases where the correlation coefficient between a template image and two overlapping regions is greater than or equal to the cross-correlation threshold. If this occurs, the overlapping region with the highest correlation coefficient can be selected as the detection region.

[0103] To illustrate the effectiveness of the detection area determination method provided by the present invention Figure 5This is a region matching effect diagram provided in Embodiment 2 of the present invention. For example... Figure 5 As shown, you can see Figure 5 After modal transformation, the template image in the second column has the same modality as the optical image in the third column. The region matching effect in the fourth column also shows that, after calculating the correlation coefficient, the detection region was accurately found, and the time consumed per row was approximately 0.85 seconds, significantly less than the time consumed by manually drawing the detection region. Moreover, in different experiments, the accuracy of the detection region reached 100%.

[0104] This invention provides a method for determining a detection region, comprising: determining a geometric image based on design data of a target wafer, wherein the geometric image includes image features of the location to be detected on the target wafer; acquiring an optical image of the target wafer; performing mode conversion on the geometric image to obtain a template image, wherein the mode of the template image is the same as the mode of the optical image; and determining a region in the optical image that matches the template image to obtain the detection region. The technical solution of this invention determines a geometric image including image features of the location to be detected on the target wafer based on the design data of the target wafer; further performs mode conversion on the geometric image to obtain a template image, making the mode of the template image the same as the mode of the acquired optical image of the target wafer; thereby determining a region in the optical image that matches the template image to obtain the detection region. Compared with the traditional method of manually drawing the detection region, firstly, by determining a geometric image based on the design data of the target wafer and then performing mode conversion on the geometric image to obtain a template image, a fully automatic conversion from design data to geometric image and then to template image is achieved, enabling this solution to adapt to wafers with different design rules. Secondly, since the template image has the same modality as the optical image, the problem of different modalities weakening the matching accuracy during subsequent image matching can be avoided, thereby improving image matching accuracy and thus improving the accuracy of identifying detection areas in optical images. Thirdly, matching optical images with template images to determine detection areas enables automated determination of detection areas, ensuring efficiency and avoiding human errors that may occur when manually drawing detection areas. This provides a foundation for subsequent defect detection and process optimization, thereby ensuring wafer yield and production efficiency.

[0105] Example 3

[0106] Figure 6 This is a schematic flowchart of a wafer inspection method provided in Embodiment 3 of the present invention. This embodiment is applicable to the situation of defect detection of a target wafer. The method can be executed by a wafer inspection device, which includes an optical microscopy inspection device or an electron optical microscopy inspection device, etc. Figure 6 As shown, the method includes:

[0107] S310. The detection area of ​​the target wafer is obtained by using any of the detection area determination methods in the embodiments of the present invention.

[0108] For details on the method of obtaining the detection area of ​​the target wafer, please refer to the description of Example 1 and / or Example 2. For the sake of brevity, it will not be repeated here.

[0109] S320. Divide the detection area into at least one detection set, wherein detection areas located in the same detection set correspond to the same detection label, and the detection label is used to indicate the defect type and / or defect characteristics that may occur in the target wafer.

[0110] S330. Use the detection method corresponding to the detection label of the detection set to detect the defect of the target wafer.

[0111] As can be seen from steps S320 and S330, after obtaining the detection area, the detection area can be further detected to determine the defects of the target wafer.

[0112] Specifically, since different regions may have different defect types and / or defect characteristics, the detection area can be divided into at least one detection set. Detection areas within the same detection set correspond to the same detection label. The detection label is used to indicate the defect type and / or defect characteristics that may appear on the target wafer. Then, the detection set is detected using the detection method corresponding to the detection label of the detection set to determine the defects of the target wafer.

[0113] This can improve the efficiency of defect detection on the target wafer, avoid unnecessary inspections, and reduce waste of resources.

[0114] This invention provides a wafer inspection method, comprising: acquiring an inspection area of ​​a target wafer; dividing the inspection area into at least one inspection set, wherein inspection areas within the same inspection set correspond to the same inspection label, the inspection label indicating the possible defect type and / or defect characteristics of the target wafer; and inspecting the inspection set using a detection method corresponding to the inspection label of the inspection set to determine the defects of the target wafer. The technical solution of this invention, since the inspection area of ​​the target wafer is acquired using the inspection area determination method described in any one of the embodiments of this invention, the wafer inspection method has the beneficial effects achievable by the aforementioned inspection area determination method. Furthermore, the divided inspection areas are calibrated according to inspection requirements, and corresponding detection algorithms are used to identify defects in inspection areas with the same inspection set label, thereby significantly improving the defect detection efficiency and accuracy of the target wafer, reducing resource waste, and demonstrating good scalability.

[0115] Example 4

[0116] Figure 7 This is a schematic diagram of a wafer inspection device provided in Embodiment 4 of the present invention. Figure 7 As shown, the device includes a data processing system 701 and an image acquisition system 702.

[0117] Image acquisition system 702 is used to acquire optical images of a target wafer;

[0118] The data processing system 701 includes an image parsing module 701a, an image conversion module 701b, an image matching module 701c, a calibration module 701d, and a defect identification module 701e.

[0119] Image analysis module 701a is used to determine a geometric image based on the design data of the target wafer, wherein the geometric image includes image features of the location to be detected on the target wafer;

[0120] Image conversion module 701b is used to perform mode conversion on a geometric image to obtain a template image, wherein the mode of the template image is the same as the mode of the optical image;

[0121] Image matching module 701c is used to determine the region in an optical image that matches the template image in order to obtain the detection region;

[0122] The calibration module 701d is used to divide the detection area into at least one detection set, wherein the detection areas located in the same detection set correspond to the same detection label, and the detection label is used to indicate the defect type and / or defect characteristics that may occur in the target wafer.

[0123] The defect identification module 701e is used to detect the detection set using a detection method corresponding to the detection label of the detection set, and to determine the defects of the target wafer.

[0124] Optionally, the image acquisition system 702 may acquire optical images of the target wafer based on an optical system or an electro-optical system.

[0125] Optionally, the image analysis module 701a is specifically used to extract target data corresponding to the location to be detected on the target wafer from the design data, wherein the design data is a Graphical Data System (GDS) file or a Computer-Aided Design (CAD) file; the target data is cropped and analyzed according to a preset ratio to obtain a geometric image.

[0126] Optionally, the image conversion module 701b is specifically used to establish an image conversion model. The image conversion model is a model obtained by training a generative adversarial network with the goal of achieving image mapping. The geometric image is input into the image conversion model to obtain a template image.

[0127] Optionally, the generative adversarial network includes a generator and a discriminator; the image conversion module 701b is also used to train the generative adversarial network to establish an image conversion model. The method for training the generative adversarial network to establish an image conversion model includes: acquiring a dataset, which includes training geometric images of multiple regions on a target wafer and measured optical images of the corresponding regions; inputting multiple training geometric images from the dataset into a generator, which obtains a theoretical optical image corresponding to each training geometric image based on a semantic segmentation model; the discriminator identifying the theoretical optical image and the measured optical image of the corresponding region, updating the parameters of the generator and discriminator to guide the generator to gradually generate candidate theoretical optical images that meet initial conditions; calculating an image evaluation value based on the candidate theoretical optical images and the measured optical images of the corresponding regions; stopping training when the image evaluation value meets a first preset condition, and using the generative adversarial network at this point as the image conversion model; when the image evaluation value does not meet the first preset condition, performing image processing on the measured optical images in the dataset, and retraining the generative adversarial network based on the processed measured optical images and the training geometric images of the corresponding regions until the image evaluation value meets the first preset condition.

[0128] Optionally, the image evaluation values ​​include the peak signal-to-noise ratio (PSNR), which is calculated using the following formula:

[0129] Where MAX is the maximum pixel value that can be obtained in the optical image, and MSE represents the mean square error between the candidate theoretical optical image and the corresponding measured image.

[0130] Optionally, the image processing includes at least one of the following processing steps: improving the brightness of the measured optical image, improving the contrast of the measured optical image, and improving the sharpness of the measured optical image.

[0131] Optionally, the number of encoder channels in the generator is greater than 1024, and the number of decoder channels in the generator is gradually reduced to 1024.

[0132] Optionally, a random deactivation layer can be configured in the discriminator. The random deactivation layer is used during training to randomly discard neurons and their connections with a preset probability.

[0133] Optionally, the image matching module 701c is specifically used to slide the template image on the optical image according to a preset step size, and determine the correlation coefficient of the overlapping area between the template image and the optical image at each slide; and take the overlapping area corresponding to the correlation parameter that satisfies the second preset condition as the detection area.

[0134] The wafer inspection apparatus provided in the embodiments of the present invention can execute the wafer inspection method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the method.

[0135] In some embodiments, the method for determining the detection area / wafer inspection method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by a processor, one or more steps of the method for determining the detection area / wafer inspection method described above may be performed. Alternatively, in other embodiments, the processor may be configured to perform the method for determining the detection area / wafer inspection method by any other suitable means (e.g., by means of firmware).

[0136] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0137] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0138] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0139] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0140] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0141] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0142] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the detection area determination method / wafer inspection method as provided in any embodiment of this invention.

[0143] In implementing the computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0144] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0145] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for determining a detection area, characterized in that, include: Based on the design data of the target wafer, a geometric image is determined, wherein the geometric image includes image features of the locations on the target wafer that need to be detected; Acquire optical images of the target wafer; The geometric image is subjected to mode conversion to obtain a template image, wherein the mode of the template image is the same as the mode of the optical image; In the optical image, a region that matches the template image is determined to obtain the detection region.

2. The method for determining the detection area according to claim 1, characterized in that, The step of determining the geometric image based on the design data of the target wafer includes: Extract target data corresponding to the locations to be detected on the target wafer from the design data, wherein the design data is a Graphical Data System (GDS) file or a Computer-Aided Design (CAD) file; The target data is cropped and parsed according to a preset ratio to obtain the geometric image.

3. The method for determining the detection area according to claim 1, characterized in that, The modality transformation of the geometric image to obtain the template image includes: An image conversion model is established, which is a model obtained by training a generative adversarial network with the goal of achieving image mapping; The geometric image is input into the image conversion model to obtain the template image.

4. The method for determining the detection area according to claim 3, characterized in that, The generative adversarial network includes a generator and a discriminator; The method for training the generative adversarial network to establish the image transfer model includes: Acquire a dataset, which includes training geometric images of multiple regions on the target wafer and measured optical images of the corresponding regions. The generator inputs multiple training geometric images from the dataset into the generator, which obtains the theoretical optical image corresponding to each training geometric image based on a semantic segmentation model. The discriminator distinguishes between the theoretical optical image and the measured optical image of the corresponding region, and updates the parameters of the generator and the discriminator to guide the generator to gradually generate candidate theoretical optical images that meet the initial conditions. Based on the candidate theoretical optical image and the measured optical image of the corresponding region, calculate the image evaluation value; When the image evaluation value meets the first preset condition, training stops, and the generative adversarial network at this time is used as the image conversion model; When the image evaluation value does not meet the first preset condition, the measured optical image in the dataset is processed, and the generative adversarial network is retrained based on the processed measured optical image and the training geometric image of the corresponding region until the image evaluation value meets the first preset condition.

5. The method for determining the detection area according to claim 4, characterized in that, The image evaluation value includes the image peak signal-to-noise ratio (PSNR), which is calculated using the following formula: Where MAX is the maximum pixel value that can be obtained in the optical image, and MSE represents the mean square error between the candidate theoretical optical image and the corresponding measured image.

6. The method for determining the detection area according to claim 4, characterized in that, The image processing includes at least one of the following processing steps: increasing the brightness of the measured optical image, increasing the contrast of the measured optical image, and increasing the sharpness of the measured optical image.

7. The method for determining the detection area according to claim 4, characterized in that, The number of encoder channels in the generator is greater than 1024, and the number of decoder channels in the generator is gradually reduced to 1024.

8. The method for determining the detection area according to claim 4, characterized in that, The discriminator is configured with a random deactivation layer, which is used to randomly discard neurons and their connections with a preset probability during the training process.

9. The method for determining the detection area according to claim 1, characterized in that, The step of determining the region in the optical image that matches the template image to obtain the detection region includes: The template image is slid across the optical image at a preset step size, and the correlation coefficient of the overlapping area between the template image and the optical image is determined at each slid. The overlapping area corresponding to the relevant parameters that meet the second preset condition is taken as the detection area.

10. The method for determining the detection area according to claim 9, characterized in that, The correlation coefficient is calculated using the following formula for each slide: Wherein, NCC(x,y) is the normalized cross-correlation NCC value at position (x,y) in the overlapping region of the template image T and the optical image I, and T i,j Let I be the pixel value of the template image T at position (i,j). x+i,y+i Let be the pixel value of the sub-window (x+1, y+1) in the optical image I relative to position (x, y) during sliding, and T be the average pixel value of the template image T. The average pixel value of the overlapping region of the optical image I. m and n are the width and height of the template image, respectively.

11. A wafer inspection method, characterized in that, The method includes: The detection area of ​​the target wafer is obtained by using the method for determining the detection area as described in any one of claims 1-10; The detection area is divided into at least one detection set, wherein the detection areas located in the same detection set correspond to the same detection label, and the detection label is used to indicate the defect type and / or defect characteristics that may occur in the target wafer; The detection set is inspected using a detection method corresponding to the detection label of the detection set to determine the defects of the target wafer.

12. A wafer inspection device, characterized in that, include: Data processing system and image acquisition system; The image acquisition system is used to acquire optical images of the target wafer; The data processing system includes an image parsing module, an image conversion module, an image matching module, a calibration module, and a defect identification module; The image analysis module is used to determine a geometric image based on the design data of the target wafer, wherein the geometric image includes image features of the locations to be detected on the target wafer; The image conversion module is used to perform mode conversion on the geometric image to obtain a template image, wherein the mode of the template image is the same as the mode of the optical image; The image matching module is used to determine the region in the optical image that matches the template image, so as to obtain the detection region; The calibration module is used to divide the detection area into at least one detection set, wherein the detection areas located in the same detection set correspond to the same detection label, and the detection label is used to indicate the defect type and / or defect characteristics that may occur in the target wafer. The defect identification module is used to detect the detection set using a detection method corresponding to the detection label of the detection set, and to determine the defects of the target wafer.

Images