Semiconductor device detection method and device, electronic equipment and storage medium

Abstract

The invention provides a semiconductor device detection method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring two DR images with a vertical relation for a semiconductor device to be detected; obtaining the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with the vertical relation based on a preset three-dimensional volume reconstruction model; performing defect detection on the to-be-detected semiconductor device based on the three-dimensional volume; wherein the preset three-dimensional volume reconstruction model obtains the three-dimensional volume through large-scale structure recovery and small-scale detail reconstruction fusion. According to the method, the defect detection of the semiconductor device can be realized with high efficiency and high precision on the premise of saving manpower and material resources.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of detection technologies, and particularly to a semiconductor device detection method, apparatus, electronic device, and storage medium. Background Art

[0002] In recent years, with the increasing development of the semiconductor industry, in order to meet the needs of the current era, the design and manufacturing processes of semiconductor devices have become increasingly precise and complex. Therefore, it has become particularly important to detect internal defects at an early stage, which can not only reduce unnecessary losses but also improve efficiency, thus gaining a competitive advantage in terms of time.

[0003] Traditional non-destructive testing methods mainly rely on manual inspection, which is costly and accompanied by low inspection efficiency. Therefore, there is an urgent need for an efficient and intelligent non-destructive testing method for mass production. Although DR technology is widely used in medical examinations and has the advantages of real-time and low cost, and can detect device defects to a certain extent, when faced with small semiconductor devices with complex, dense, and occluded internal structures, it is still necessary to use computer tomography technology to reconstruct the three-dimensional structure of the device to achieve defect detection. However, the principle of computer tomography technology is based on circumferential projection data and cannot meet the real-time online reconstruction requirements.

[0004] Therefore, there is an urgent need for a semiconductor device detection method that can improve the speed and accuracy of defect detection while saving human and material resources. Summary of the Invention

[0005] In view of this, the present application provides a semiconductor device detection method, apparatus, electronic device, and storage medium, which can efficiently and accurately detect defects of semiconductor devices while saving human and material resources.

[0006] To solve the above technical problems, the technical solution of the present application is implemented as follows:

[0007] In one embodiment, a semiconductor device detection method is provided, and the method includes:

[0008] Collect two DR images with a vertical relationship for the semiconductor device to be detected;

[0009] Obtain the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship based on a preset three-dimensional volume reconstruction model;

[0010] Perform defect detection on the semiconductor device to be detected based on the three-dimensional volume;

[0011] Wherein, the preset three-dimensional volume reconstruction model includes: a channel number enhancement module, a structure restoration module, a detail enhancement module, and a fusion module;

[0012] The channel number increasing module is used to increase the channel numbers of the two DR images with a vertical relationship respectively while keeping the spatial resolution unchanged; and fuse the two DR images with increased channel numbers to obtain a first fusion result.

[0013] The structure restoration module is used to perform large-scale structure restoration on the first fusion result to obtain first feature information.

[0014] The detail enhancement module is used to perform small-scale detail reconstruction on the first fusion result to obtain second feature information.

[0015] The fusion module is used to fuse the first feature information and the second feature information to obtain a second fusion result; the second fusion result is the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship.

[0016] Among them, the generation of the preset three-dimensional volume reconstruction model includes:

[0017] Generate a three-dimensional volume reconstruction model to be trained.

[0018] Obtain a training data set; among them, the training data set includes multiple data pairs, and each data pair includes two DR images with a vertical relationship and a reference three-dimensional volume; the DR image data with a vertical relationship is obtained from the acquisition data set composed of the DR images of the reference semiconductor device collected axially, and the reference three-dimensional volume is obtained by performing FDK reconstruction based on the acquisition data set.

[0019] Train the three-dimensional volume reconstruction model to be trained based on the training data set to obtain a preset three-dimensional volume reconstruction model.

[0020] Among them, when training the three-dimensional volume reconstruction model to be trained based on the training data set, a preset loss function is introduced.

[0021] When the values of N consecutive preset loss functions are all less than a preset loss threshold, end the training of the three-dimensional volume reconstruction model to be trained to obtain a preset three-dimensional volume reconstruction model.

[0022] Among them, N is an integer greater than 1; the preset loss function is the weighted sum of the mean squared error loss function and the structural similarity index loss function; the value of the preset loss function is calculated using the preset loss function based on the reference three-dimensional volume and the three-dimensional volume obtained by the three-dimensional volume reconstruction model to be trained.

[0023] Among them,

[0024] The structure restoration module uses an encoder to perform upsampling learning on the first fusion result to obtain a potentially compact representation feature; uses a decoder to receive the feature information output by the encoder, and reconstructs the latent information through an upsampling operation to obtain the first feature information; and introduces skip connections at different scales during the encoding and decoding process.

[0025] Among them,

[0026] The detail enhancement module obtains feature maps with different receptive fields for the first fusion result through three parallel dilated convolutions with different dilation rates, connects the feature maps with different receptive fields, and uses a convolutional layer to restore the number of channels corresponding to the connection result to obtain a first processing result; obtains a second processing result for the first fusion result through two convolutional layers; and fuses the first processing result and the second processing result to obtain second feature information.

[0027] Among them,

[0028] The detail enhancement module fusing the first processing result and the second processing result to obtain second feature information includes:

[0029] Performing global pooling on the first processing result to obtain a global feature vector, and using a fully connected layer and a softmax function to generate a first attention weight corresponding to the first processing result;

[0030] Performing global pooling on the second processing result to obtain a global feature vector, and using a fully connected layer and a softmax function to generate a second attention weight corresponding to the second processing result;

[0031] Determining the second feature information based on the first processing result, the first attention weight, the second processing result, and the second attention weight.

[0032] In another embodiment, a semiconductor device detection apparatus is provided, and the apparatus includes:

[0033] An acquisition unit for acquiring two DR images with a vertical relationship for the semiconductor device to be detected;

[0034] An acquisition unit, configured to obtain a three-dimensional volume of a semiconductor device to be detected corresponding to the two DR images with a vertical relationship based on a preset three-dimensional volume reconstruction model; wherein, the preset three-dimensional volume reconstruction model includes: a channel number enhancement module, a structure restoration module, a detail enhancement module, and a fusion module; the channel number enhancement module is configured to enhance the channel numbers of the two DR images with a vertical relationship respectively while keeping the spatial resolution unchanged; and fuse the two DR images with enhanced channel numbers to obtain a first fusion result; the structure restoration module is configured to perform large-scale structure restoration on the first fusion result to obtain first feature information; the detail enhancement module is configured to perform small-scale detail reconstruction on the first fusion result to obtain second feature information; the fusion module is configured to fuse the first feature information and the second feature information to obtain a second fusion result; the second fusion result is the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship.

[0035] A detection unit, configured to perform defect detection on the semiconductor device to be detected based on the three-dimensional volume.

[0036] Wherein, the device further includes:

[0037] A training unit, configured to generate a three-dimensional volume reconstruction model to be trained; obtain a training data set; wherein, the training data set includes multiple data pairs, and each data pair includes two DR images with a vertical relationship and a three-dimensional volume; the DR image data with a vertical relationship is obtained from a data set composed of DR images of a reference semiconductor device acquired axially, and the three-dimensional volume is obtained by performing FDK reconstruction based on the data set; train the three-dimensional volume reconstruction model to be trained based on the training data set to obtain a preset three-dimensional volume reconstruction model.

[0038] In another embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, a semiconductor device detection method is implemented.

[0039] In another embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, a semiconductor device detection method is implemented.

[0040] As can be seen from the above technical solution, in the above embodiments, by collecting two DR images with a vertical relationship for the semiconductor device to be detected, the three-dimensional volume of the semiconductor device to be detected can be obtained using a preset three-dimensional volume reconstruction model, and then the semiconductor device can be defect-detected; the preset three-dimensional volume reconstruction model can perform large-scale information restoration and small-scale information reconstruction, and fuse the results of both to obtain a high-precision three-dimensional volume. Therefore, it is possible to achieve defect detection of semiconductor devices with high efficiency and high precision on the premise of saving human and material resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0042] Figure 1 Schematic diagram of the process of establishing a preset three-dimensional volume reconstruction model in an embodiment of the present application;

[0043] Figure 2 Schematic diagram of the structure of a preset three-dimensional volume reconstruction model in an embodiment of the present application;

[0044] Figure 3 Schematic diagram of the structure of a channel number enhancement module in an embodiment of the present application;

[0045] Figure 4 Schematic diagram of the structure of a structure restoration module in an embodiment of the present application;

[0046] Figure 5 Schematic diagram of the structure of a detail enhancement module in an embodiment of the present application;

[0047] Figure 6 Schematic diagram of the semiconductor device detection process in an embodiment of the present application;

[0048] Figure 7 Schematic diagram of the semiconductor device detection structure in an embodiment of the present application;

[0049] Figure 8 Schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0051] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above accompanying drawings are used to distinguish similar objects, and are not necessarily used to describe the order or sequence of the objects. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0052] The technical solutions of the present invention will be described in detail below with specific embodiments. The following several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0053] Traditional non-destructive testing methods mainly rely on manual inspection, which is costly and accompanied by low detection efficiency. Therefore, there is an urgent need for an efficient and intelligent non-destructive testing method for batch production methods. Although DR technology is widely used in medical examinations and has the advantages of real-time and low cost, and can detect device defects to a certain extent, when facing small semiconductor devices with complex, dense and occluded internal structures, it is still necessary to use computer tomography technology to reconstruct the three-dimensional structure of the device to achieve defect detection. However, the principle of computer tomography technology is based on circumferential projection data and cannot meet the real-time online reconstruction requirements.

[0054] For X-ray CT imaging, it is usually necessary to collect projection data of 360 degrees around the object, and the reconstruction algorithm is complex and time-consuming, which cannot meet the real-time detection requirements. To improve the detection speed, CT imaging based on sparse views is generally adopted, mainly including reconstruction algorithms based on compressive sensing, reconstruction methods based on optimization algorithms, reconstruction techniques based on model assistance, and reconstruction methods based on deep learning. However, these methods have their respective limitations, such as slow imaging speed, the imaging effect decreasing with the reduction of the angle, and the imaging result depending on model construction. Among them, for the problem of three-dimensional volume reconstruction based on deep learning methods, there is currently a single-projection three-dimensional reconstruction network, but the detailed structure in the reconstruction result of devices with rich internal information is not well restored. Although the dual-projection three-dimensional reconstruction network uses orthogonal projection information to complement each other, there is still a problem that small-scale details cannot be well restored, which limits the quality of the reconstructed volume.

[0055] Therefore, there is an urgent need for a semiconductor device detection method that can improve the defect detection speed and detection accuracy on the premise of saving human and material resources.

[0056] Based on the above problems, in the embodiments of the present application, a semiconductor device detection method is provided. By only collecting two DR images with a vertical relationship of the semiconductor device to be detected, the three-dimensional volume of the semiconductor device to be detected can be obtained using a preset three-dimensional volume reconstruction model, and then the semiconductor device can be defect-detected; the preset three-dimensional volume reconstruction model therein can perform large-scale information restoration and small-scale information reconstruction, and fuse the results of both to obtain a three-dimensional volume with high precision. Therefore, on the premise of saving human and material resources, the defect detection of semiconductor devices can be realized with high efficiency and high precision.

[0057] Before detecting the semiconductor device in the embodiments of the present application, it is necessary to first establish a preset three-dimensional volume reconstruction model, and the specific establishment process is as follows:

[0058] See Figure 1 , Figure 1 which is a schematic diagram of the process of establishing a preset three-dimensional volume reconstruction model in the embodiments of the present application. The specific steps are as follows:

[0059] Step 101, generate a three-dimensional volume reconstruction model to be trained.

[0060] Step 102, obtain a training data set.

[0061] Among them, the training data set includes multiple data pairs, and each data pair includes two DR images with a vertical relationship and a reference three-dimensional volume; the DR image data with a vertical relationship is obtained from the acquisition data set composed of DR images of the reference semiconductor device collected axially, and the reference three-dimensional volume is obtained by FDK reconstruction based on the acquisition data set.

[0062] The obtained training dataset can be a pre-stored dataset or a collected dataset.

[0063] The implementation for collecting and obtaining the training dataset is as follows:

[0064] Circumferential DR image acquisition is performed on semiconductor devices (reference objects) of multiple defect types; for each semiconductor device, 360 DR images in different directions are acquired.

[0065] During the acquisition process, relevant parameters such as tube voltage, tube current, exposure time, and position during detection are ensured to be unchanged.

[0066] The acquired DR images can be preprocessed first. The preprocessing includes:

[0067] Based on chip parameters and mechanical parameters, the acquired DR images are subjected to size cropping, taking logarithms, normalization, etc. When performing size cropping, the size is cropped to 256×256×360. The normalization process normalizes the image and voxel gray values to [0,1], then the reference three-dimensional volume of the reference object is obtained, and model training is performed.

[0068] Obtaining the reference three-dimensional volume of each reference object includes:

[0069] Mature algorithms such as FDK can be used to obtain the reference three-dimensional volume of the reference object based on 360 DR images of each reference object.

[0070] Step 103: Train the three-dimensional volume reconstruction model to be trained based on the training dataset to obtain a preset three-dimensional volume reconstruction model.

[0071] During the training process, the Adam optimizer is used, and the learning rate starts from 1e-3, and the decay value of the learning rate is set to 0.8.

[0072] During the training process, a preset loss function is also introduced in the embodiments of the present application; the preset loss function uses a combination of the mean squared error (MSE) loss function and the structural similarity index (SSIM) loss function to quantify the difference between the three-dimensional volume obtained through the three-dimensional volume reconstruction model to be trained and the reference three-dimensional volume. The preset loss function is specifically the weighted sum of the mean squared error loss function and the structural index loss function; the value of the preset loss function is calculated based on the reference three-dimensional volume and the three-dimensional volume obtained through the three-dimensional volume reconstruction model to be trained using this preset loss function. Specifically:

[0073] The value of the mean squared error loss function is obtained by calculating the sum of squared errors based on the reference three-dimensional volume and the three-dimensional volume obtained through the three-dimensional volume reconstruction model to be trained using the mean squared error loss function; this loss will achieve good numerical reconstruction of the three-dimensional volume. The specific expression is as follows:

[0074]

[0075] Among them, loss MSE is the value of the sum of squared errors function, and y 0 is the reference three-dimensional volume, is the three-dimensional volume obtained by the three-dimensional volume reconstruction model to be trained;

[0076] The value of the structural index loss function is determined using the structural index loss function based on the reference three-dimensional volume and the three-dimensional volume obtained by the three-dimensional volume reconstruction model to be trained. This loss will achieve good reconstruction of the three-dimensional volume that conforms to human vision. The specific expression is as follows:

[0077]

[0078] Among them, loss SSIM is the value of the structural index loss function, and y 0 is the reference three-dimensional volume, is the three-dimensional volume obtained by the three-dimensional volume reconstruction model to be trained;

[0079] The value of the preset loss function is the weighted sum of the value of the squared difference loss function and the value of the structural index loss function, and the specific representation is as follows:

[0080] loss = αloss MSE + βloss SSIM

[0081] Among them, loss is the value of the preset loss function, α and β are weight values, and the specific values of α and β are not limited and can be set according to actual needs.

[0082] When the values of the preset loss function for N consecutive times are all less than the preset loss threshold, the training of the three-dimensional volume reconstruction model to be trained is ended, and a preset three-dimensional volume reconstruction model is obtained; where N is an integer greater than 1.

[0083] In the embodiments of the present application, the structures of the three-dimensional volume reconstruction model to be trained and the preset three-dimensional volume reconstruction model are the same. Taking the structure of the preset three-dimensional volume reconstruction model as an example, a detailed description is given.

[0084] See Figure 2 , Figure 2 which is the structural schematic diagram of the preset three-dimensional volume reconstruction model in the embodiments of the present application. Figure 2 The preset three-dimensional volume reconstruction model in

[0085] Among them, the channel number increasing module is used to increase the channel numbers of two DR images with a vertical relationship respectively while keeping the spatial resolution unchanged; and fuse the two DR images with increased channel numbers to obtain a first fusion result.

[0086] See Figure 3 , Figure 3 which is a schematic structural diagram of the channel number increasing module in an embodiment of the present application. Figure 3 The channel number increasing module includes two dense residual modules and a fuser 313. The two dense residual modules are respectively denoted as a first dense residual module 311 and a second dense residual module 312.

[0087] The two DR images with a vertical relationship are respectively denoted as a first DR image 301 and a second DR image 302 and input into the channel number increasing module. Specifically, the first DR image 301 is input into the first dense residual module 311, and the second DR image 302 is input into the second dense residual module 312; the first dense residual module 311 and the second dense residual module 312 respectively increase the channel numbers of the input DR images while keeping the spatial resolution unchanged, and the fuser 313 fuses the results output by the first dense residual module 311 and the second dense residual module 312 and outputs a first fusion result.

[0088] The structure restoration module is used to perform large-scale structure restoration on the first fusion result to obtain first feature information.

[0089] The structure restoration module uses an encoder to perform upsampling learning on the first fusion result to obtain a potentially compact representation feature; uses a decoder to receive the feature information output by the encoder and reconstructs the potential information through an upsampling operation to obtain first feature information; and introduces skip connections at different scales during the encoding and decoding processes.

[0090] See Figure 4 , Figure 4 which is a schematic structural diagram of the structure restoration module in an embodiment of the present application. Figure 4 The structure restoration module in

[0091] Among them, the decoder consists of four downsampling blocks, which are used to learn potential compact representation features; the key features of the input information are captured through these four downsampling modules, while redundant information is removed; assuming that the information corresponding to the first fusion result input is: 8, 128, 256, 256, then the information after the first sampling block is 8, 128, 128, 128, the information after the second sampling block is 8, 128, 64, 64, the information after the third sampling block is 8, 128, 32, 32, and the information after the fourth sampling block is 8, 128, 16, 16. Here is just an example, and the specific implementation is not limited to the above information; among them, the first data 8 represents the batch size, the second data 128 represents the number of channels, and the third and fourth data represent the length and width, which are reduced by a multiple. And each downsampling block consists of two residual blocks and a max pooling layer.

[0092] The encoder consists of four upsampling blocks, and the reconstruction of potential information is realized through upsampling operations.

[0093] At the same time, in order to retain the original important information and restore the fine-grained spatial details lost during the encoding process, therefore, when obtaining the first fusion result through the above structure recovery module, that is, different-scale skip connections are introduced during the encoding and decoding process, as Figure 4 shown by the dotted line in

[0094] The detail enhancement module is used to perform small-scale detail reconstruction on the first fusion result to obtain the second feature information, specifically:

[0095] The detail enhancement module obtains feature maps with different receptive fields for the first fusion result through three parallel dilated convolutions with different dilation rates, connects the feature maps with different receptive fields, and uses a convolutional layer to restore the number of channels corresponding to the connection result to obtain the first processing result; the second processing result is obtained for the first fusion result through two convolutional layers; the first processing result and the second processing result are fused to obtain the second feature information.

[0096] Among them, fusing the first processing result and the second processing result to obtain the second feature information includes:

[0097] Performing global pooling on the first processing result to obtain a global feature vector, and using a fully connected layer and a softmax function to generate the first attention weight corresponding to the first processing result;

[0098] Performing global pooling on the second processing result to obtain a global feature vector, and using a fully connected layer and a softmax function to generate the second attention weight corresponding to the second processing result;

[0099] Determining the second feature information based on the first processing result, the first attention weight, the second processing result, and the second attention weight.

[0100] See Figure 5 , Figure 5 , which is a schematic diagram of the detailed enhancement module structure in the embodiment of the present application. Figure 5 The detailed enhancement module in includes: three parallel dilation convolutions 501 with different dilation rates, a connector 502, a first convolutional layer, a second convolutional layer, a third convolutional layer, a first global pooling, a second global pooling, a first fully connected layer and a softmax function, a second fully connected layer and a softmax function, a first weighting module, a second weighting module, and a summation module.

[0101] The first fusion result is input into three parallel dilation convolutions with different dilation rates. The dilation rates of the three parallel dilation convolutions with different dilation rates are 1, 3, and 5 respectively. Feature maps with different receptive fields are obtained through the three parallel dilation convolutions with different dilation rates; and the three feature maps with different receptive fields are connected through the connector, and then a convolution operation is performed using the first convolutional layer to obtain a first processing result. Specifically, the first convolutional layer can be a 1×1 convolutional layer, which is used to adjust the number of channels so that the number of channels is restored to the number of channels corresponding to the original image.

[0102] At the same time, the first fusion result is input into the second convolutional layer, and after processing, it is input into the third convolutional layer to obtain a second processing result. Here, the second convolutional layer and the third convolutional layer can be 3×3 convolutional layers.

[0103] The first processing result is input into the first global pooling for pooling to obtain a global feature vector, and the first attention weight corresponding to the first processing result is generated using the first fully connected layer and the softmax function;

[0104] The second processing result is input into the second global pooling for pooling to obtain a global feature vector, and the second attention weight corresponding to the second processing result is generated using the second fully connected layer and the softmax function;

[0105] Weighting is performed using the first weighting module based on the second attention weight and the second processing result; weighting is performed using the second weighting module based on the second attention weight and the second processing result;

[0106] The weighted results of the first weighting module and the second weighting module are summed through the summation module to obtain the second feature information.

[0107] The fusion module is used to fuse the first feature information and the second feature information to obtain a second fusion result; the second fusion result is the three-dimensional volume of the semiconductor device to be detected corresponding to two DR images with a vertical relationship.

[0108] The preset three-dimensional volume reconstruction model provided in the embodiments of the present application can obtain both large-scale structure restoration and small-scale detail reconstruction.

[0109] The following will combine the accompanying drawings to detail the semiconductor device detection process in the embodiments of the present application.

[0110] See Figure 6 , Figure 6 which is a schematic diagram of the semiconductor device detection process in the embodiments of the present application. The specific steps are as follows:

[0111] Step 601, collect two DR images with a vertical relationship for the semiconductor device to be detected.

[0112] In the specific embodiments of the present application, relevant parameters such as tube voltage, tube current, exposure time, and position during detection are ensured to be unchanged during the process of collecting two DR images with a vertical relationship.

[0113] The collected DR images can be processed in a manner similar to that during model training, such as size cropping, taking logarithms, normalization, etc., which can better ensure the acquisition of a three-dimensional volume with high precision.

[0114] Step 602, obtain the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship based on the preset three-dimensional volume reconstruction model.

[0115] Step 603, perform defect detection on the semiconductor device to be detected based on the three-dimensional volume.

[0116] In the embodiments of the present application, only by collecting two DR images with a vertical relationship for the semiconductor device to be detected, the three-dimensional volume of the semiconductor device to be detected can be obtained using the preset three-dimensional volume reconstruction model, and then defect detection can be performed on the semiconductor device; the preset three-dimensional volume reconstruction model can perform large-scale information restoration and small-scale information reconstruction, and fuse the results of both to obtain a three-dimensional volume with high precision. In the embodiments of the present application, only two DR images of the semiconductor device to be detected are collected, and the three-dimensional volume of the semiconductor device to be detected can be obtained, so that defect detection of the semiconductor device can be efficiently realized on the premise of saving human and material resources; when obtaining the three-dimensional volume of the semiconductor device to be detected, large-scale information restoration and small-scale information reconstruction are considered, so that defect detection of the semiconductor device can be realized with high precision on the premise of saving human and material resources.

[0117] All the above optional technical solutions can be combined arbitrarily to form optional embodiments of the present disclosure, which will not be elaborated here one by one.

[0118] Based on the same inventive concept, an embodiment of the present application also provides a semiconductor device detection device. SeeFigure 7 , Figure 7 This is a schematic structural diagram of a semiconductor device detection apparatus in an embodiment of the present application. The apparatus includes:

[0119] An acquisition unit 701, configured to acquire two DR images with a vertical relationship for a semiconductor device to be detected;

[0120] An obtaining unit 702, configured to obtain a three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship based on a preset three-dimensional volume reconstruction model; wherein, the preset three-dimensional volume reconstruction model includes: a channel number increasing module, a structure restoration module, a detail enhancement module, and a fusion module; the channel number increasing module is configured to increase the channel numbers of the two DR images with a vertical relationship respectively while keeping the spatial resolution unchanged; and fuse the two DR images with increased channel numbers to obtain a first fusion result; the structure restoration module is configured to perform large-scale structure restoration on the first fusion result to obtain first feature information; the detail enhancement module is configured to perform small-scale detail reconstruction on the first fusion result to obtain second feature information; the fusion module is configured to fuse the first feature information and the second feature information to obtain a second fusion result; the second fusion result is the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship;

[0121] A detection unit 703, configured to perform defect detection on the semiconductor device to be detected based on the three-dimensional volume;

[0122] In another embodiment, the apparatus further includes:

[0123] A training unit 704, configured to generate a three-dimensional volume reconstruction model to be trained; obtain a training data set; wherein, the training data set includes multiple data pairs, and each data pair includes two DR images with a vertical relationship and a three-dimensional volume; the DR image data with a vertical relationship is obtained from a data set composed of DR images of a reference semiconductor device acquired axially, and the three-dimensional volume is obtained by performing FDK reconstruction based on the data set; train the three-dimensional volume reconstruction model to be trained based on the training data set to obtain a preset three-dimensional volume reconstruction model.

[0124] In another embodiment,

[0125] The training unit 704 is further configured to introduce a preset loss function when training the three-dimensional volume reconstruction model to be trained based on a training data set; when the values of N consecutive preset loss functions are all less than a preset loss threshold, end the training of the three-dimensional volume reconstruction model to be trained to obtain a preset three-dimensional volume reconstruction model; where N is an integer greater than 1; the preset loss function is a weighted sum of a mean squared error loss and a structural similarity index loss; the mean squared error loss is obtained by calculating the sum of squared errors between a reference three-dimensional volume and the three-dimensional volume obtained by the three-dimensional volume reconstruction model to be trained; the structural similarity index loss is determined based on the reference three-dimensional volume and the three-dimensional volume obtained by the three-dimensional volume reconstruction model to be trained.

[0126] In another embodiment,

[0127] The structure recovery module uses an encoder to perform upsampling learning on the first fusion result to obtain a potentially compact representation feature; uses a decoder to receive the feature information output by the encoder and reconstruct the latent information through an upsampling operation to obtain first feature information; and introduces skip connections at different scales during the encoding and decoding process.

[0128] In another embodiment,

[0129] The detail enhancement module obtains feature maps with different receptive fields for the first fusion result through three parallel dilated convolutions with different dilation rates, connects the feature maps with different receptive fields, and uses a convolutional layer to restore the number of channels corresponding to the connection result to obtain a first processing result; obtains a second processing result for the first fusion result through two convolutional layers; and fuses the first processing result and the second processing result to obtain second feature information.

[0130] In another embodiment,

[0131] The detail enhancement module fusing the first processing result and the second processing result to obtain second feature information includes:

[0132] Performing global pooling on the first processing result to obtain a global feature vector, and using a fully connected layer and a softmax function to generate a first attention weight corresponding to the first processing result;

[0133] Performing global pooling on the second processing result to obtain a global feature vector, and using a fully connected layer and a softmax function to generate a second attention weight corresponding to the second processing result;

[0134] Determining the second feature information based on the first processing result, the first attention weight, the second processing result, and the second attention weight.

[0135] The units in the above embodiments can be integrated into one body or separately deployed; they can be combined into one unit or further split into multiple sub-units.

[0136] In another embodiment, an electronic device is further provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, a semiconductor device detection method is implemented.

[0137] In another embodiment, a computer-readable storage medium is further provided, on which computer instructions are stored. When the instructions are executed by a processor, a semiconductor device detection method is implemented.

[0138] Figure 8 It is a schematic diagram of the physical structure of the electronic device provided by the embodiment of the present invention. As Figure 8 shown, the electronic device may include: a processor (Processor) 810, a communication interface (Communications Interface) 820, a memory (Memory) 830, and a communication bus 840. Among them, the processor 810, the communication interface 820, and the memory 830 complete mutual communication through the communication bus 840. The processor 810 can call the logical instructions in the memory 830 to execute the following methods:

[0139] Collect two DR images with a vertical relationship for the semiconductor device to be detected;

[0140] Based on a preset three-dimensional volume reconstruction model, obtain the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship;

[0141] Perform defect detection on the semiconductor device to be detected based on the three-dimensional volume;

[0142] Among them, the preset three-dimensional volume reconstruction model includes: a channel number enhancement module, a structure restoration module, a detail enhancement module, and a fusion module;

[0143] The channel number enhancement module is used to separately enhance the channel numbers of the two DR images with a vertical relationship while keeping the spatial resolution unchanged; and fuse the two DR images with enhanced channel numbers to obtain a first fusion result;

[0144] The structure restoration module is used to perform large-scale structure restoration on the first fusion result to obtain first feature information;

[0145] The detail enhancement module is used to perform small-scale detail reconstruction on the first fusion result to obtain second feature information;

[0146] The fusion module is used to fuse the first feature information and the second feature information to obtain a second fusion result; the second fusion result is the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship.

[0147] In addition, when the logical instructions in the above-mentioned memory 830 can be implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0148] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution in this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0149] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disks, optical discs, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments.

[0150] The flowcharts and block diagrams in the accompanying drawings of this application illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments disclosed in this application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in the order marked in different drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, as well as the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0151] Those skilled in the art can understand that the features described in various embodiments and / or claims disclosed in this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, without departing from the spirit and teachings of this application, the features described in various embodiments and / or claims of this application can be combined and / or combined in various ways, and all such combinations and / or combinations fall within the scope of the disclosure of this application.

[0152] Specific embodiments are used herein to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention, and is not used to limit this application. For those skilled in the art, changes can be made in the specific implementation manners and application scopes according to the ideas, spirits, and principles of the present invention. Any modifications, equivalent replacements, improvements, etc. made by them should be included within the scope of protection of this application.

Claims

1. A semiconductor device detection method, characterized in that: The method comprises: For the semiconductor device to be inspected, two DR images with a vertical relationship are collected; Obtaining the three-dimensional volume of the semiconductor device to be inspected corresponding to the two DR images having a vertical relationship based on a preset three-dimensional volume reconstruction model; Performing defect detection on the semiconductor device to be inspected based on the three-dimensional volume; The preset three-dimensional volume reconstruction model includes: a channel number enhancement module, a structure recovery module, a detail enhancement module and a fusion module; The channel number enhancement module is used to enhance the channel numbers of the two DR images having a vertical relationship while ensuring that the spatial resolution remains unchanged; and fuse the two DR images after the channel numbers are enhanced to obtain a first fusion result; The structure recovery module is used to perform large-scale structure recovery on the first fusion result to obtain first feature information; The detail enhancement module is used to reconstruct small-scale details of the first fusion result to obtain second feature information; The fusion module is used to fuse the first feature information and the second feature information to obtain a second fusion result; wherein the second fusion result is the three-dimensional volume of the semiconductor device to be inspected corresponding to the two DR images having a vertical relationship.

2. The method according to claim 1, characterized in that The generation of the preset three-dimensional volume reconstruction model includes: Generate a three-dimensional volume reconstruction model to be trained; Acquire a training data set; wherein the training data set includes a plurality of data pairs, each of which includes two DR images with a vertical relationship and a reference three-dimensional volume; the DR image data with a vertical relationship is acquired from an acquisition data set consisting of DR images of a reference semiconductor device acquired axially, and the reference three-dimensional volume is obtained by FDK reconstruction based on the acquisition data set; The three-dimensional volume reconstruction model to be trained is trained based on the training data set to obtain a preset three-dimensional volume reconstruction model.

3. The method according to claim 2, characterized in that When training the three-dimensional volume reconstruction model to be trained based on the training data set, introducing a preset loss function; When the values ​​of N consecutive preset loss functions are all less than the preset loss threshold, the training of the to-be-trained three-dimensional volume reconstruction model is terminated to obtain a preset three-dimensional volume reconstruction model; Among them, N is an integer greater than 1; the preset loss function is a weighted sum of a square difference loss function and a structural exponential loss function; the value of the preset loss function is calculated based on the reference three-dimensional volume and the three-dimensional volume obtained by the three-dimensional volume reconstruction model to be trained using the preset loss function.

4. The method according to claim 1, characterized in that: The structure recovery module uses the encoder to perform upsampling learning on the first fusion result to obtain a potential compact representation feature; uses the decoder to receive the feature information output by the encoder, reconstructs the potential information through an upsampling operation, and obtains the first feature information; And introduce skip connections of different scales in the encoding and decoding process.

5. The method according to claim 1, characterized in that The detail enhancement module obtains feature maps with different receptive fields through three parallel dilated convolutions with different dilation rates on the first fusion result, connects the feature maps with different receptive fields, and uses a convolution layer to restore the number of channels corresponding to the connection result to obtain a first processing result; Obtaining a second processing result from the first fusion result through two convolutional layers; The first processing result and the second processing result are fused to obtain second feature information.

6. The method according to claim 5, characterized in that The detail enhancement module fuses the first processing result and the second processing result to obtain second feature information, including: Performing global pooling on the first processing result to obtain a global feature vector, and using a fully connected layer and a softmax function to generate a first attention weight corresponding to the first processing result; Performing global pooling on the second processing result to obtain a global feature vector, and generating a second attention weight corresponding to the second processing result using a fully connected layer and a softmax function; Determine second feature information based on the first processing result, the first attention weight, the second processing result and the second attention weight.

7. A semiconductor device detection device, characterized in that: The device comprises: An acquisition unit, used for acquiring two DR images having a vertical relationship with respect to the semiconductor device to be inspected; An acquisition unit, used for obtaining the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship based on a preset three-dimensional volume reconstruction model; wherein the preset three-dimensional volume reconstruction model includes: a channel number improvement module, a structure recovery module, a detail enhancement module and a fusion module; the channel number improvement module is used for respectively improving the channel number of the two DR images with a vertical relationship while ensuring that the spatial resolution remains unchanged; and fusing the two DR images after the channel number improvement to obtain a first fusion result; the structure recovery module is used for performing large-scale structure recovery on the first fusion result to obtain first feature information; the detail enhancement module is used for performing small-scale detail reconstruction on the first fusion result to obtain second feature information; the fusion module is used for fusing the first feature information and the second feature information to obtain a second fusion result; the second fusion result is the three-dimensional volume of the semiconductor device to be detected corresponding to the two DR images with a vertical relationship; A detection unit is used to perform defect detection on the semiconductor device to be detected based on the three-dimensional volume.

8. The device according to claim 7, characterized in that The device further comprises: A training unit is used to generate a three-dimensional volume reconstruction model to be trained; obtain a training data set; wherein the training data set includes multiple groups of data pairs, each group of the data pairs includes two DR images with a vertical relationship and a three-dimensional volume; the DR image data with a vertical relationship are obtained from a data set composed of DR images of a reference semiconductor device collected axially, and the three-dimensional volume is obtained by FDK reconstruction based on the data set; the three-dimensional volume reconstruction model to be trained is trained based on the training data set to obtain a preset three-dimensional volume reconstruction model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method described in any one of claims 1 to 6 is implemented.

Images