Training method for use in chip layout encoder, method, device, and program for screening chip layout
By geometrically transforming sample chip layouts and training an encoder to extract relevant features, the method addresses the challenge of extracting layout features in chip layouts, enhancing clustering accuracy and reducing redundancy in screening processes.
Patent Information
- Application Number
- JP2024532532
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2022-09-26
- Filing Date
- 2023-05-31
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2043-05-31
AI Technical Summary
The challenge of training a chip layout encoder to effectively extract layout features of chip layouts is exacerbated by the interference and diffraction effects caused by the mismatch between the characteristic dimensions of chip layouts and the wavelength of light sources used in photolithography, leading to distortion and reduced chip performance and yield.
A method involving geometric transformation of sample chip layouts to generate reference layouts, training an initial encoder using these layouts, and extracting features to obtain a chip layout encoder that can accurately distinguish between different types of chip layouts, thereby improving clustering and screening accuracy.
The proposed method enhances the accuracy of clustering results and reduces redundancy in screening processes by focusing on extracting layout features that can differentiate between chip layout types, ultimately improving chip performance and yield.
Smart Images

Figure 2024543202000001_ABST
Abstract
Description
[Technical field]
[0001] FIELD OF THE DISCLOSURE Embodiments of the present disclosure relate to the field of integrated circuit technology, and more particularly to training, chip layout screening methods and apparatus for use in chip layout encoders.
[0002] This application claims priority to a Chinese patent application filed on September 26, 2022, bearing application number 202211176018.8 and entitled "Method for training chip layout encoder, method and apparatus for screening chip layout," the entire contents of which are incorporated herein by reference. [Background technology]
[0003] With the continuous development of integrated circuit (IC) technology, the characteristic dimensions of IC layouts (also called chip layouts) are also continuously shrinking. Here, the characteristic dimensions of chip layouts refer to the minimum dimensions of semiconductor devices. At present, the characteristic dimensions of chip layouts have already reached the nanometer level.
[0004] In general, a photolithography device must first be operated to expose a chip layout onto a photoresist using a light source to obtain a mask layout, and then a photolithography device must be operated to expose the mask layout onto a wafer using a light source to obtain an imaged layout. Because the characteristic dimensions of the chip layout are smaller than the wavelength of the light source used in the photolithography process, interference and diffraction effects are enhanced, which results in distortion and ambiguity in the imaged layout obtained by exposing the mask layout onto the wafer, which further reduces the performance and yield of the chip. In this case, a light source mask optimization must be performed on the chip layout to obtain an optimized mask layout, thereby improving the performance and yield of the chip.
[0005] When optimizing the light source mask for the chip layout, it is necessary to extract the layout features of the chip layout. Therefore, how to train and obtain a chip layout encoder that can extract the layout features of the chip layout is an urgent problem to be solved. Summary of the Invention [Problem to be solved by the invention]
[0006] The present application provides a training for a chip layout encoder, a method and apparatus for screening a chip layout, and is a technical means for extracting layout features of a chip layout and screening the chip layout, the technical means including the following contents: [Means for solving the problem]
[0007] In one aspect, a training method for use in a chip layout encoder is provided, the method being performed by electronic equipment, the method comprising: obtaining a sample chip layout and an initial encoder; performing a geometric transformation on the sample chip layout to obtain at least one reference chip layout; extracting sample layout features of the sample chip layout and reference layout features of each reference chip layout using the initial encoder; training the initial encoder based on layout features of the sample chip layout and layout features of each of the reference chip layouts to obtain a chip layout encoder, the chip layout encoder being used to extract layout features of a target chip layout; or training the initial encoder based on the sample layout features and the reference layout features to obtain a trained encoder.
[0008] In another aspect, a method for screening a chip layout is provided, the method being performed by an electronic device, the method comprising: obtaining a plurality of target chip layouts and a chip layout encoder (i.e., a trained encoder), the chip layout encoder (i.e., the trained encoder) being obtained by training with a training method used for the chip layout encoder; extracting target layout features for each target chip layout using the chip layout encoder (i.e., the trained encoder); performing a clustering process on the plurality of target chip layouts based on each target layout feature to obtain a plurality of target clustering groups, in which at least one target chip layout is included in any one of the target clustering groups; For any one of the target clustering groups, screening a key chip layout from among the target chip layouts included in the any one of the target clustering groups.
[0009] In another aspect, there is provided a training apparatus for use in a chip layout encoder, the apparatus comprising: an acquisition module used to acquire a sample chip layout and an initial encoder; a transformation module used for performing a geometric transformation on the sample chip layout to obtain at least one reference chip layout; an extraction module adapted to extract sample layout features of the sample chip layout and reference layout features of each reference chip layout using the initial encoder; a training module used for training the initial encoder based on layout features of the sample chip layout and layout features of each of the reference chip layouts to obtain a chip layout encoder, the chip layout encoder being used for extracting layout features of a target chip layout, or the training module is used for training the initial encoder based on the sample layout features and the reference layout features to obtain a trained encoder.
[0010] In another aspect, there is provided an apparatus for screening chip layouts, the apparatus comprising: An acquisition module used for acquiring a plurality of target chip layouts and a chip layout encoder (i.e., a trained encoder), the chip layout encoder (i.e., the trained encoder) being obtained by training according to a training method used for the chip layout encoder; an extraction module adapted to extract target layout features for each target chip layout using the chip layout encoder (i.e., the trained encoder); A clustering module is used for performing a clustering process on the plurality of target chip layouts based on each target layout feature to obtain a plurality of target clustering groups, in which at least one target chip layout is included in any one of the target clustering groups; a screening module, which is used for, for any one of the target clustering groups, screening a key chip layout from among the target chip layouts included in the any one of the target clustering groups.
[0011] In another aspect, an electronic device is provided, the electronic device including a processor and a memory, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor, such that the electronic device implements a training method for use in the chip layout encoder or implements a screening method for the chip layout.
[0012] In another aspect, a non-volatile computer-readable storage medium is further provided, in which at least one computer program is stored, and the at least one computer program is loaded and executed by a processor to cause an electronic device to realize a training method for use in the chip layout encoder or a screening method for the chip layout.
[0013] In another aspect, a computer program is further provided, the computer program being at least one, and when the at least one computer program is loaded and executed by a processor, the electronic device realizes a training method for use in the chip layout encoder or a screening method for the chip layout.
[0014] In yet another aspect, a computer program product is further provided, in which at least one computer program is stored, and when the at least one computer program is loaded and executed by a processor, an electronic device realizes a training method for use in the chip layout encoder or a screening method for the chip layout. Effect of the Invention
[0015] The technical solution provided by the present application is to perform a geometric transformation on a sample chip layout to obtain a reference chip layout, and use the layout features of the sample chip layout and the layout features of the reference chip layout to train an initial encoder to obtain a chip layout encoder, so that the chip layout encoder can output similar layout features for the chip layouts before and after the geometric transformation. Because the chip layouts before and after the geometric transformation belong to the same chip layout type, the chip layout encoder can focus on extracting layout features that can distinguish the chip layout type, and then, when performing clustering and screening processes on the chip layout based on the layout features, it is advantageous in improving the accuracy of the clustering result and reducing the redundancy of the screening result. [Brief description of the drawings]
[0016] [Figure 1] FIG. 1 is a schematic diagram of an implementation environment of a training method or a chip layout screening method used in a chip layout encoder provided by an embodiment of the present application. [Diagram 2] 1 is a flowchart of a training method used in a chip layout encoder provided by an embodiment of the present application. [Diagram 3] FIG. 2 is a schematic diagram showing a geometric transformation performed on a sample chip layout provided by an embodiment of the present application. [Figure 4] 1 is a schematic diagram of the Unet framework provided by an embodiment of the present application. [Diagram 5] 1 is a flowchart of a chip layout screening method provided by an embodiment of the present application; [Figure 6] FIG. 2 is a schematic diagram of a chip layout encoder provided by an embodiment of the present application, and clustering and screening based on the chip layout encoder; [Figure 7] FIG. 2 is a structural schematic diagram of a training device used in a chip layout encoder provided by an embodiment of the present application; [Figure 8]FIG. 2 is a structural schematic diagram of a chip layout screening device provided by an embodiment of the present application. [Figure 9] FIG. 2 is a structural schematic diagram of a terminal device provided by an embodiment of the present application; [Figure 10] FIG. 2 is a structural schematic diagram of a server provided by an embodiment of the present application; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0017] In order to make the objectives, technical means and advantages of the present application clearer, the embodiments of the present application will be described in more detail below in conjunction with the drawings.
[0018] Fig. 1 is a schematic diagram of an implementation environment of a training method for a chip layout encoder or a screening method for a chip layout provided in an embodiment of the present application. As shown in Fig. 1, the implementation environment includes a terminal device 101 and a server 102. Here, the training method for a chip layout encoder or a screening method for a chip layout provided in an embodiment of the present application may be executed by the terminal device 101, may be executed by the server 102, or may be executed by the terminal device 101 and the server 102 jointly.
[0019] The terminal device 101 may be a smartphone, a game host, a desktop computer, a tablet computer, a portable laptop computer, a smart TV, a smart in-vehicle device, a smart voice interaction device, a smart home appliance, etc. The server 102 may be any one of a single server, a server cluster consisting of multiple servers, a cloud computing platform, and a virtualization center, and the embodiment of the present application is not limited thereto. The server 102 may be communicatively connected to the terminal device 101 via a wired network or a wireless network. The server 102 may have functions such as data processing, data storage, and data transmission and reception, and the embodiment of the present application is not limited thereto. The number of the terminal device 101 and the server 102 is not limited, and may be one or more.
[0020] The training method or screening method for chip layout encoder provided in the embodiments of the present application can be realized based on artificial intelligence technology. Artificial intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, sense the environment, obtain knowledge, and use the knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology of computer science, which aims to grasp the essence of intelligence and create a kind of new smart machine that can respond in a manner similar to human intelligence. Artificial intelligence studies the design principles and realization methods of various smart machines, enabling machines to have the functions of sensing, reasoning, and decision-making.
[0021] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields and includes both hardware and software technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technology mainly includes several major directions such as computer vision technology, voice processing technology, natural language processing technology, machine learning / deep learning, autonomous driving, and smart transportation.
[0022] In the technical field of integrated circuits, an integrated circuit layout is a kind of ordinary layout, and an integrated circuit layout is also called a chip layout. In general, a light source mask needs to be optimized for the chip layout to obtain an optimized mask layout, so that the error between the imaged layout obtained by exposing the optimized mask layout on a wafer by operating a photolithography device and the chip layout is reduced, thereby improving the chip performance and increasing the chip yield.
[0023] When optimizing the source mask for the chip layout, it is necessary to extract the layout features of the chip layout. Therefore, how to train a chip layout encoder and use the chip layout encoder to extract the layout features of the chip layout is an urgent problem to be solved.
[0024] The embodiment of the present application provides a training method for a chip layout encoder, which can be applied in the above implementation environment and can solve the problem of obtaining a chip layout encoder by training. In the example of the flowchart of the training method for a chip layout encoder provided by the embodiment of the present application shown in Figure 2, for convenience of description, the terminal device 101 or the server 102 that executes the training method for a chip layout encoder in the embodiment of the present application is called an electronic device, and the method may be executed by an electronic device. As shown in Figure 2, the method includes the following steps:
[0025] Step 201: Obtain a sample chip layout and an initial encoder.
[0026] A sample chip layout is an integrated circuit (IC) layout or a sub-region in an IC layout. Usually, the number of sample chip layouts is multiple, and the processing manner of each sample chip layout is the same.
[0027] In the embodiment of the present application, an arbitrary IC layout can be obtained, and the IC layout can be taken as a sample chip layout. Or, the IC layout can be uniformly divided into a number of sub-regions, and each sub-region can be obtained, and any sub-region can be taken as a sample chip layout. Optionally, some sub-regions can be manually screened from each sub-region (e.g., screened out sub-regions that have no defect points, a defect point being a part of an IC layout that does not satisfy a process window requirement) and these sub-regions can be taken as a sample chip layout, or each sub-region can be classified, and a chip layout type corresponding to each sub-region can be uniformly sampled and screened from the sub-regions corresponding to each chip layout type to obtain a sample chip layout. Here, the chip layout type can include, but is not limited to, a contact hole type, a logic layout type, and a dense line type, etc.
[0028] As can be seen, when dividing an IC layout uniformly, there is a possibility that a defective point is exactly on the dividing line. In this case, when dividing an IC layout, one defective point will be divided into at least two parts, and the different parts are in different sub-regions. Since a sub-region contains a part of the defective point, which is an incomplete defective point, it is difficult to identify the defective point from the sub-region, so that the sub-region containing the defective point is easily taken as the sample chip layout, which causes the quality of the sample chip layout to be reduced.
[0029] In one possible implementation, each defect point in the IC layout can be identified first. The embodiment of the present application does not limit the identification method of the defect points, and for example, each defect point in the IC layout can be identified by a manual identification method, or each defect point in the IC layout can be identified by a photolithography simulation method. Then, an image division model is used to perform smart division on the IC layout to obtain each sub-region. By distributing the defect points in the center of the sub-region as much as possible, rather than distributing them on the edge of the sub-region through the smart division, the defect points can be quickly and accurately identified from the sub-region when screening the sample chip layout from each sub-region, and the screened sample chip layout can be avoided as a sub-region having a defect point, and the quality of the sample chip layout can be improved. Thus, when the chip layout encoder is subsequently trained by the sample chip layout, the training effect of the chip layout encoder can be improved.
[0030] In the embodiment of the present application, the sample chip layout is image data and belongs to a pixelated representation mode. By dividing the IC layout and screening the sample chip layout, various chip layout types such as contact hole type, logic layout type, and dense line type can be represented by pixels. Here, the pixel size of the sample chip layout needs to be equal to or larger than the wavelength of the light source / (4×numerical aperture), that is, the pixel size of the sample chip layout needs to meet the minimum resolution of the photolithography device.
[0031] The sample chip layout is used to train the initial encoder to obtain a chip layout encoder. The embodiment of the present application does not limit the structure, size, etc. of the initial encoder. For example, the initial encoder is an encoder in a deep learning framework based on a U network, and is abbreviated as an encoder in the Unet framework. The encoder in the Unet framework can effectively perform a dimensionality reduction process on high-dimensional data (e.g., image data), thereby extracting data features. The Unet framework is a kind of convolutional autoencoder, which is mainly configured based on two parts. The first part is an encoder, which is configured of at least two convolution blocks, and the encoder can be the initial encoder in the embodiment of the present application. The second part is a decoder, which is configured of at least two deconvolution blocks, and the number of convolution blocks is the same as the number of deconvolution blocks.
[0032] Step 202: Perform a geometric transformation on the sample chip layout to obtain at least one reference chip layout.
[0033] The geometric transformation includes at least one type of transformation, such as an inversion transformation and a rotation transformation. At least one type of transformation, such as an inversion transformation and a rotation transformation, can be performed on the sample chip layout to obtain a reference chip layout, where the inversion transformation is a mirror inversion process mentioned below, and the rotation transformation is a rotation process mentioned below.
[0034] Illustratively, step 202 includes performing a mirror flip process on the sample chip layout to obtain a symmetric chip layout, and taking the symmetric chip layout as at least one reference chip layout. As can be understood, the symmetric chip layout here may also be called the sample chip layout after the mirror flip process, and is abbreviated as the flipped chip layout. That is, the sample chip layout is subjected to a mirror flip process to obtain the flipped chip layout. The at least one reference chip layout includes the flipped chip layout. The implementation of the mirror flip process will be described below, but will not be described in detail here.
[0035] Illustratively, step 202 includes performing a rotation process on the sample chip layout to obtain a rotated chip layout, and taking the rotated chip layout as at least one reference chip layout. As can be understood, the rotated chip layout here may also be called the sample chip layout after the rotation process, and is abbreviated as the rotated chip layout. That is, the sample chip layout is rotated to obtain the rotated chip layout. The at least one reference chip layout includes the rotated chip layout. The implementation of the rotation process will be described below, but will not be described in detail here.
[0036] In one possible implementation, step 202 includes performing a mirror flip process on the sample chip layout to obtain a symmetric chip layout, performing a rotation process on the sample chip layout and the symmetric chip layout to obtain a rotated chip layout, and taking the symmetric chip layout and the rotated chip layout as at least one reference chip layout. As can be understood, the symmetric chip layout here may also be referred to as the sample chip layout after the mirror flip process, and is abbreviated as the flipped chip layout. The rotated chip layout here may also be referred to as the sample chip layout after the rotation process and the symmetric chip layout after the rotation process, and is abbreviated as the rotated chip layout. That is, performing a mirror flip on the sample chip layout to obtain a flipped chip layout, and performing a rotation on the sample chip layout and the flipped chip layout to obtain a rotated chip layout. The at least one reference chip layout includes the flipped chip layout and the rotated chip layout.
[0037] In the embodiment of the present application, a mirror inversion process is performed on the sample chip layout according to the mirror information, thereby realizing an inversion transformation on the sample chip layout, obtaining a symmetrical chip layout, and using the symmetrical chip layout as a reference chip layout. The mirror information can represent the position and orientation of the mirror, where the orientation of the mirror is the same as the normal direction of the mirror, and the position of the mirror is represented by the position of a reference point (e.g., a center point) on the mirror. When the mirror is oriented in the horizontal direction, it indicates that the normal direction of the mirror is horizontal, and in this case, a left or right mirror inversion process can be realized on the sample chip layout. When the mirror is oriented in the vertical direction, it indicates that the normal direction of the mirror is vertical, and in this case, a top or bottom mirror inversion process can be realized on the sample chip layout.
[0038] It can be understood that there is at least one piece of mirror surface information. For any one piece of mirror surface information, a mirror inversion process is carried out on the sample chip layout according to the mirror surface information to obtain a symmetric chip layout, and the symmetric chip layout is used as the reference chip layout. In this manner, different reference chip layouts can be determined according to different mirror surface information, and the number of reference chip layouts is at least one.
[0039] By performing a rotation process on the sample chip layout at least once according to the rotation parameters, a rotation transformation on the sample chip layout can be realized, and a rotated chip layout obtained after each rotation process can be obtained. That is, first, a first rotation process is performed on the sample chip layout according to the rotation parameters, and the sample chip layout obtained after the first rotation process is the rotated chip layout obtained after the first rotation process. Next, a second rotation process is performed on the rotated chip layout obtained after the first rotation process according to the rotation parameters, and a rotated chip layout obtained after the second rotation process is obtained. Next, a third rotation process is performed on the rotated chip layout obtained after the second rotation process according to the rotation parameters, and a rotated chip layout obtained after the third rotation process is obtained. In this manner, at least one rotation process is realized on the sample chip layout.
[0040] Similarly, the rotation transformation for the symmetric chip layout can be realized by performing a rotation process on the symmetric chip layout at least once according to the rotation parameter, and a rotated chip layout obtained after each rotation process can be obtained. The rotated chip layout obtained after each rotation process can be each reference chip layout. It is necessary to explain that the rotation parameter is used to indicate the angle that needs to be rotated for each rotation process. For example, the rotation parameter is a target angle that needs to be rotated for each rotation process, in this case, the sample chip layout can be rotated once for each target angle to obtain one rotated chip layout, and the symmetric chip layout can also be rotated once for each target angle to obtain one rotated chip layout.
[0041] Referring to Fig. 3, Fig. 3 is a schematic diagram of a geometric transformation performed on a sample chip layout provided by an embodiment of the present application. In the embodiment of the present application, on the one hand, the sample chip layout can be rotated once every 45 degrees to obtain two rotated chip layouts, and on the other hand, the sample chip layout is first subjected to a right mirror flip process to obtain a symmetric chip layout, and then the symmetric chip layout can be rotated once every 45 degrees to obtain two rotated chip layouts.
[0042] In the embodiment of the present application, the symmetric chip layout and the rotated chip layout can be used as the reference chip layout. As can be understood, first, a rotation process is performed on the sample chip layout to obtain a rotated chip layout, and then a mirror flip process is performed on the sample chip layout and the rotated chip layout to obtain a symmetric chip layout. Then, the symmetric chip layout and the rotated chip layout can be used as the reference chip layout, and the quantity of the reference chip layout is at least one. As can be understood, the rotated chip layout here can also be called the sample chip layout after the rotation process, and is abbreviated as the rotated chip layout. The symmetric chip layout here can also be called the sample chip layout after the mirror flip process, and the rotated chip layout after the mirror flip process, and is abbreviated as the rotated chip layout. That is, the sample chip layout is rotated to obtain the rotated chip layout, and the sample chip layout and the rotated chip layout are mirror flipped to obtain the flipped chip layout. At least one reference chip layout includes the rotated chip layout and the flipped chip layout. The implementation of the rotation process and the mirror flip process has been described above, and will not be described in detail here.
[0043] In the embodiments of the present application, performing mirror reversal processing on a chip layout (such as the sample chip layout, symmetric chip layout, and rotated chip layout mentioned above) refers to adjusting the position information of each pixel point in the chip layout according to the mirror information, so that the adjusted position information of any one pixel point and the position information of the pixel point satisfy mirror symmetry. The symmetric chip layout obtained after the mirror reversal processing includes the adjusted position information of each pixel point.
[0044] Based on the same principle, performing a rotation process on a chip layout refers to adjusting the position information of each pixel point in the chip layout according to a rotation parameter so that the adjusted position information of any one pixel point and the position information of the pixel point satisfy the rotation transformation. The rotated chip layout obtained after the rotation process includes the adjusted position information of each pixel point.
[0045] It can be seen that performing geometric transformation on the sample chip layout is to adjust the position information of each pixel point in the sample chip layout, and the reference chip layout includes the adjusted position information of each pixel point. It can be understood that the adjacent relationship between pixel points in the sample chip layout is the same as the adjacent relationship between pixel points in the reference chip layout, and the adjacent relationship between pixel points refers to the relationship between every two pixel points whether they are adjacent or not. That is, if two pixel points in the sample chip layout are adjacent, these two pixel points in the reference chip layout are also adjacent.
[0046] In the embodiment of the present application, the sample chip layout can be subjected to mirror flipping and rotation to obtain a reference chip layout, the symmetric chip layout can be subjected to rotation to obtain a reference chip layout, and the rotated chip layout can be subjected to mirror flipping to obtain a reference chip layout. The reference chip layout is determined by a plurality of methods, which can improve the diversity of the reference chip layout, and the chip layout encoder obtained by training using the reference chip layout can be endowed with a relatively strong generalization ability, and the expression ability of the layout features can be improved. Thus, the accuracy of the clustering result can be improved when the clustering process is performed on the chip layout based on the layout features.
[0047] Step 203: Extract sample layout features of the sample chip layout and reference layout features of each reference chip layout using the initial encoder.
[0048] Optionally, the sample layout features may be referred to as layout features of the sample chip layout, and the reference layout features may be referred to as layout features of the reference chip layout. Extracting layout features using an initial encoder corresponds to extracting layout features by an initial encoder. In this case, step 203 may be said to extract layout features of the sample chip layout and layout features of each reference chip layout by an initial encoder.
[0049] A sample chip layout is input to an initial encoder, and the initial encoder performs feature extraction on the sample chip layout to obtain layout features of the sample chip layout. Similarly, for any one reference chip layout, the reference chip layout is input to an initial encoder, and the initial encoder performs feature extraction on the reference chip layout to obtain layout features of the reference chip layout. The following focuses on the method in which the initial encoder performs feature extraction on the sample chip layout, but the method in which the initial encoder performs feature extraction on the reference chip layout is similar to the method in which the initial encoder performs feature extraction on the sample chip layout, and therefore will not be described in detail again below.
[0050] In one possible implementation, the step 203 of "extracting layout features of the sample chip layout using the initial encoder" includes performing multiple downsampling operations on the sample chip layout by (i.e., using) the initial encoder to obtain respective downsampling features, where each downsampling feature is a downsampling feature obtained in each downsampling operation, and the layout features of the sample chip layout are the sample layout features and the downsampling features obtained in the final downsampling operation.
[0051] In the embodiment of the present application, the initial encoder includes at least two convolution blocks, and the input of the next convolution block is the output of the previous convolution block. Based on this, the sample chip layout is input to the initial encoder, and downsampling is performed multiple times by at least two convolution blocks, and the output of each convolution block is one downsampled feature. Optionally, the output of each convolution block has a half reduction in characteristic dimension and a double increase in the number of channels compared to the input of the convolution block. The final output of the initial encoder may be a one-dimensional feature vector, for example, the dimension of the one-dimensional feature vector is (1,1,1024). The one-dimensional feature vector is a layout feature of the sample chip layout.
[0052] In an exemplary embodiment, the step of "performing a plurality of downsampling processes on the sample chip layout using the initial encoder to obtain each downsampling feature" includes the steps of performing a first downsampling process on the sample chip layout using the initial encoder to obtain a first downsampling feature, and performing a next downsampling process on any one of the downsampling features obtained in any one of the downsampling processes using the initial encoder to obtain a downsampling feature obtained in the next downsampling process.
[0053] That is, after inputting the sample chip layout into the initial encoder, the initial encoder maps the sample chip layout as an initial feature, and a first convolution block of the at least two convolution blocks performs a downsampling process on the initial feature of the sample chip layout to obtain a first downsampling feature. Then, a second convolution block of the at least two convolution blocks performs a downsampling process on the first downsampling feature to obtain a second downsampling feature, and so on, until a final downsampling feature is obtained, which is the layout feature of the sample chip layout.
[0054] That is, after inputting the sample chip layout into the initial encoder, the first convolution block of the at least two convolution blocks performs a downsampling process on the initial feature mapped from the sample chip layout to obtain the first downsampling feature. For the i-th convolution block (i is a positive integer greater than 1), the i-th convolution block performs a downsampling process on the i-1-th downsampling feature to obtain the i-th downsampling feature. Assuming there are M convolution blocks in total, the M-th downsampling feature obtained by the M-th convolution block is the layout feature of the sample chip layout.
[0055] By performing multiple downsampling processes on the sample chip layout, the dimension of the downsampling features is continuously reduced, and the expressive ability of the downsampling features is improved, that is, the expressive ability of the layout features of the sample chip layout is improved.
[0056] Optionally, one convolution block includes one convolution layer, the convolution layer is composed of at least one 3×3 filter, and the number of filters included in the previous convolution block is less than the number of filters included in the next convolution block, where the convolution layer is used to perform a convolution process on the initial feature or the downsampled feature of the sample chip layout to obtain a convolution feature. If the convolution block only includes a convolution layer, the convolution process corresponds to a downsampled process, and the convolution feature corresponds to the downsampled feature.
[0057] Optionally, one convolution block includes a convolution layer and a batch normalization layer connected in series. The batch normalization layer is used to perform a batch normalization process on the convolution feature to obtain a batch normalized feature. If the convolution block includes only a convolution layer and a batch normalization layer, the downsampling process includes a convolution process and a batch normalization process, and the batch normalized feature corresponds to the downsampling feature.
[0058] Optionally, one convolution block includes a convolution layer, a batch normalization layer, and an activation layer connected in series. The activation layer is used to perform an activation process on the batch normalization feature to obtain an activation feature. If the convolution block only includes a convolution layer, a batch normalization layer, and an activation layer, the downsampling process includes a convolution process, a batch normalization process, and an activation process, and the activation feature corresponds to the downsampling feature. Here, the activation layer can adopt a rectified linear unit (ReLU), and the linear rectification function is also called a rectified linear unit.
[0059] Referring to Figure 4, Figure 4 is a schematic diagram of the Unet framework provided in the embodiment of the present application. The Unet framework includes an encoder, which is the initial encoder in the embodiment of the present application. The encoder includes eight convolution blocks, each of which includes a convolution layer, a batch normalization layer, and an activation layer that are connected in series, and the activation layer adopts a rectified linear unit. These eight convolution blocks include 8, 16, 32, 64, 128, 256, 512, and 1024 3x3 filters in order.
[0060] In FIG. 4, the sample chip layout shown in FIG. 3, a symmetric chip layout, and a rotated chip layout are shown, and all of these layouts are input into the Unet framework, and the layout features of these layouts can be extracted by the encoder included in the Unet framework.
[0061] Taking the sample chip layout as an example, the sample chip layout is input into the Unet framework, and the encoder included in the Unet framework performs the downsampling process on the sample chip layout eight times to obtain eight downsampling features. Assuming that the dimensions of the sample chip layout are (256,256,1), the dimensions of these eight downsampling features are (128,128,8), (64,64,16), (32,32,32), (16,16,64), (8,8,128), (4,4,256), and (1,1,1024), respectively. Here, the last downsampling feature is the layout feature of the sample chip layout, so the dimensions of the layout feature of the sample chip layout are (1,1,1024), which is one one-dimensional feature vector.
[0062] Step 204: Train an initial encoder based on the sample layout features and the reference layout features to obtain a trained encoder.
[0063] As can be understood, the trained encoder may also be called a chip layout encoder. On this basis, step 204 may be said to train an initial encoder based on the layout features of the sample chip layout and the layout features of each reference chip layout to obtain a chip layout encoder.
[0064] Optionally, the trained encoder is used to generate a chip layout and / or extract features of the chip layout.
[0065] In the embodiment of the present application, the first loss can be determined based on the layout features of the sample chip layout and the layout features of each reference chip layout. Optionally, for any one of the reference chip layouts, the distance between the layout features of the reference chip layout and the layout features of the sample chip layout can be calculated, the distance is called the distance corresponding to the reference chip layout, and a weighted calculation (e.g., weighted addition, weighted averaging, etc.) can be performed on the distance corresponding to each reference chip layout, and the weighted calculation result can be the first loss. Or, an average calculation can be performed on the layout features of each reference chip layout to obtain an average layout feature, and the distance between the average layout feature and the layout feature of the sample chip layout can be calculated, and the distance can be called the first loss.
[0066] For example, the following formula (1) can be employed to perform an averaging calculation on the layout features of each reference chip layout to obtain an average layout feature.
[0067]
number
number
[0068] In addition, Encoder in Equation (1) represents a feature extraction operation, and Σ represents a symbol of an addition function. Therefore, [Equation 3] represents the average layout feature, and [Equation 4] represents the layout feature of the reference chip layout obtained after performing the n-th geometric transformation on the i-th sample chip layout.
[0069]
number
number
[0070] Based on formula (1), the following formula (2) can be adopted to calculate the distance between the average layout feature and the layout feature of the sample chip layout to obtain the first loss.
[0071]
number
[0072] Here, Loss rotation represents the first loss. Encoder(x i ) represents the layout feature of the i-th sample chip layout. The total number of sample chip layouts is N+1. Σ represents the symbol of the addition function.
[0073] After calculating the first loss, the first loss is taken as the loss of the initial encoder, and the loss of the initial encoder is used to train the initial encoder through a gradient descent method, so as to obtain a trained initial encoder. Training the initial encoder is to reduce the loss of the initial encoder by adjusting the weights and biases of the neurons in the initial encoder. If the trained initial encoder meets the training end condition, the trained initial encoder is taken as the chip layout encoder (i.e., the trained encoder). If the trained initial encoder does not meet the training end condition, the trained initial encoder is taken as the initial encoder for the next training round, and the initial encoder is trained for the next training round according to the training method used for the chip layout encoder provided in the embodiment of the present application, until a chip layout encoder is obtained.
[0074] The embodiment of the present application is not limited to satisfying the training end condition. For example, the training end condition is satisfied when the number of training times reaches a target number (for example, 500 times). Or, when training the initial encoder using the loss of the initial encoder by the gradient descent method, it is necessary to determine the gradient of the loss of the initial encoder, and the parameters of the initial encoder are adjusted based on the gradient, so that the gradient of the loss of the initial encoder determined in the next training is smaller than the gradient of the loss of the initial encoder determined in the current training. When the gradient difference between the gradient of the loss of the initial encoder determined in the next training and the gradient of the loss of the initial encoder determined in the current training is smaller than the difference threshold, it is determined that the training end condition is satisfied. That is, the gradient difference is obtained by subtracting the gradient of the loss of the initial encoder determined in the next training from the gradient of the loss of the initial encoder determined in the current training, and if the gradient difference is smaller than the difference threshold, the training end condition is satisfied. At this time, the training end condition is satisfied corresponding to the descent range of the gradient of the loss of the initial encoder being within the range of the difference threshold. The embodiments of the present application do not limit the method of determining the difference threshold, and illustratively, the difference threshold is determined according to human experience, or a data range is artificially set, and the difference threshold is randomly generated data within the data range.
[0075] It is necessary to explain that since the first loss is determined based on the layout features of the sample chip layout and the layout features of each reference chip layout, the layout features of the sample chip layout extracted by the initial encoder and the layout features of the reference chip layout can be made closer to each other by training the initial encoder using the first loss. Since the reference chip layout is obtained by performing a geometric transformation on the sample chip layout, the reference chip layout and the sample chip layout belong to the same type of chip layout. In the process of training the initial encoder, the layout features extracted by the initial encoder can continue to have a tendency to express the chip layout category, so that the chip layout encoder finally obtained by training mainly extracts layout features that can express the chip layout category. The layout features extracted by the chip layout encoder can express the chip layout before the geometric transformation and the chip layout after the geometric transformation, so that the layout features extracted by the chip layout encoder have a characteristic that does not change before and after the geometric transformation. Therefore, the first loss is a loss that can express that does not change before and after the geometric transformation.
[0076] In one possible implementation, the training method for the chip layout encoder further includes step 205, where step 205 is performed after step 203.
[0077] Step 205: Determine a reconstructed chip layout based on the layout characteristics of the sample chip layout, i.e., determine a reconstructed chip layout based on the sample layout characteristics, where the reconstructed chip layout is a chip layout reconstructed based on the layout characteristics of the sample chip layout.
[0078] In the embodiment of the present application, the layout features of the sample chip layout are input to the initial decoder, and the initial decoder reconstructs a new chip layout according to the layout features of the sample chip layout to obtain a reconstructed chip layout. The embodiment of the present application does not limit the structure, size, etc. of the initial decoder, and for example, the initial decoder is a decoder in the Unet framework.
[0079] In one possible implementation, step 205 includes performing multiple upsampling processes on the layout features of the sample chip layout to obtain each upsampling feature, where each upsampling feature is an upsampling feature obtained in each upsampling process. Alternatively, performing multiple upsampling processes on the sample layout features to obtain each upsampling feature obtained in each upsampling process. The reconstructed chip layout is a chip layout obtained based on the upsampling feature obtained in the last upsampling process.
[0080] In an embodiment of the present application, the initial decoder includes at least two deconvolution blocks, where the output of the previous deconvolution block is the input of the next deconvolution block. Optionally, the output of each deconvolution block has a double increase in characteristic size and half the number of channels compared to the input of the deconvolution block. The characteristic size ultimately output by the initial decoder matches the size of the sample chip layout input to the initial encoder.
[0081] That is, after the layout feature of the sample chip layout is input to the initial decoder, a first deconvolution block of the at least two deconvolution blocks performs an upsampling process on the layout feature of the sample chip layout to obtain a first upsampling feature. Then, a second deconvolution block of the at least two deconvolution blocks performs an upsampling process on the first upsampling feature to obtain a second upsampling feature, and so on, until a final upsampling feature is obtained, which can be mapped as a chip layout, and the chip layout is a reconstructed chip layout.
[0082] That is, after the layout features of the sample chip layout are input to the initial decoder, the first deconvolution block among the at least two deconvolution blocks performs an upsampling process on the layout features of the sample chip layout to obtain a first upsampling feature. For the i-th deconvolution block (i is a positive integer greater than 1), the i-1-th upsampling feature is performed by the i-th deconvolution block to obtain the i-th upsampling feature. Assuming that there are M deconvolution blocks in total, the M-th upsampling feature obtained by the M-th deconvolution block can be mapped as a reconstructed chip layout.
[0083] Optionally, one deconvolution block includes one deconvolution layer, the deconvolution layer is composed of at least one 3×3 filter, and the number of filters included in the previous deconvolution block is greater than the number of filters included in the next deconvolution block, where the deconvolution layer is used to perform deconvolution processing on the layout features of the sample chip layout or the upsampling features to obtain the deconvolution features. If the deconvolution block only includes a deconvolution layer, the deconvolution processing corresponds to an upsampling processing, and the deconvolution features correspond to the upsampling features.
[0084] Optionally, one deconvolution block includes a deconvolution layer and a batch normalization layer connected in series. The batch normalization layer is used to perform a batch normalization process on the deconvolution features to obtain batch normalized features. If the deconvolution block includes only a deconvolution layer and a batch normalization layer, the upsampling process includes a deconvolution process and a batch normalization process, and the batch normalized features correspond to the upsampling features.
[0085] Optionally, one deconvolution block includes a deconvolution layer, a batch normalization layer, and an activation layer connected in series. The activation layer is used to perform an activation process on the batch normalization feature to obtain an activation feature. If the deconvolution block only includes a deconvolution layer, a batch normalization layer, and an activation layer, the upsampling process includes a deconvolution process, a batch normalization process, and an activation process, and the activation feature corresponds to the upsampling feature. Here, the activation layer can adopt a Leaky-ReLU activation function or a sigmoid activation function.
[0086] The step of selectively performing multiple upsampling processes on the layout features of the sample chip layout to obtain each upsampling feature includes steps A1 to A4 shown below.
[0087] Step A1: A first upsampling process is performed on the layout feature of the sample chip layout (i.e., the sample layout feature) to obtain a first upsampling feature. The layout feature of the sample chip layout (i.e., the sample layout feature) is obtained by performing multiple downsampling processes on the sample chip layout.
[0088] In step 203, it is mentioned that the layout feature of the sample chip layout is obtained by performing downsampling processing on the sample chip layout multiple times, so it will not be described in detail again here. In the embodiment of the present application, after the layout feature of the sample chip layout is input to the initial decoder, the first deconvolution block of the at least two deconvolution blocks performs upsampling processing on the layout feature of the sample chip layout to obtain a first upsampling feature.
[0089] Step A2: For any one of upsampling features obtained in any one of the upsampling processes, a downsampling feature corresponding to any one of the upsampling features is obtained, and when the downsampling process is performed on the sample chip layout to obtain a downsampling feature corresponding to any one of the upsampling features, the number of downsampling processes is a first number, and when the upsampling process is performed on the layout features of the sample chip layout (i.e., the sample layout features) to obtain any one of the upsampling features, the number of upsampling processes is a second number, and the sum of the first number and the second number is a target number.
[0090] In the embodiment of the present application, the number of convolution blocks in the initial encoder is the same as the number of deconvolution blocks in the initial decoder, so the total number of downsampling operations performed by the initial encoder is the same as the total number of upsampling operations performed by the initial decoder. The target number is the total number of downsampling operations performed by the initial encoder. For example, if the initial encoder performs a total of 8 downsampling operations, the target number is 8.
[0091] For the i-th upsampling feature obtained in the i-th upsampling process (i is a positive integer equal to or greater than 1), the second number of times is i. Since the sum of the first number of times and the second number of times is the target number of times, if the target number of times is M, the first number of times is Mi. It can be determined that the downsampling feature corresponding to the i-th upsampling feature is the Mi-th downsampling feature obtained by performing the Mi-th downsampling process on the sample chip layout.
[0092] For example, the target number is 8, and the downsampling feature corresponding to the first upsampling feature is the seventh downsampling feature, the downsampling feature corresponding to the second upsampling feature is the sixth downsampling feature, and so on.
[0093] Step A3: splicing any one of the upsampling features and a downsampling feature corresponding to any one of the upsampling features to obtain a spliced feature corresponding to any one of the upsampling features.
[0094] The i-th upsampling feature and the Mi-th downsampling feature can be spliced together to obtain a spliced feature corresponding to the i-th upsampling feature.
[0095] Step A4: Perform upsampling processing on the spliced feature corresponding to any one of the upsampling features to obtain a next upsampling feature of any one of the upsampling features.
[0096] An upsampling process is performed on the spliced feature corresponding to the i-th upsampled feature to obtain the i+1-th upsampled feature.
[0097] It is necessary to explain that when the last upsampling feature is obtained, the last upsampling feature can be mapped as a reconstructed chip layout; the last upsampling feature and the first downsampling feature can be spliced together to obtain a spliced feature corresponding to the last upsampling feature, and the spliced feature corresponding to the last upsampling feature can be mapped as a reconstructed chip layout.
[0098] Referring to Fig. 4, the Unet framework shown in Fig. 4 includes a decoder, which is the initial decoder in the embodiment of the present application. The decoder includes eight deconvolution blocks, each of which includes a deconvolution layer, a batch normalization layer, and an activation layer connected in series. These eight deconvolution blocks include 1024, 512, 256, 128, 64, 32, 16, and 1 3x3 filter in order, where the activation layers in the first seven deconvolution blocks are all Leaky-ReLU activation functions, and the activation layer in the eighth deconvolution block is a sigmoid activation function. The encoder in the Unet framework has been described above, and will not be described in detail again here.
[0099] In the embodiment of the present application, the encoder in the Unet framework can perform the downsampling process on the sample chip layout eight times to obtain eight downsampling features; perform the upsampling process on the eighth downsampling feature to obtain the first upsampling feature; obtain the second upsampling feature based on the first upsampling feature and the seventh downsampling feature; obtain the third upsampling feature based on the second upsampling feature and the sixth downsampling feature, and so on until the eighth upsampling feature is obtained based on the seventh upsampling feature and the first downsampling feature.
[0100] Here, when obtaining the second upsampling feature based on the first upsampling feature and the seventh downsampling feature, the first upsampling feature and the seventh downsampling feature are spliced together to obtain a spliced feature corresponding to the first upsampling feature, and an upsampling process is performed on the spliced feature corresponding to the first upsampling feature to obtain the second upsampling feature. The method of determining the third upsampling feature to the eighth upsampling feature is similar to the method of determining the second upsampling feature, and will not be described in detail again here.
[0101] Assuming that the layout features of the sample chip layout have dimensions (1,1,1024) and are one-dimensional feature vectors, the dimensions of the eight upsampling features are (2,2,1024), (4,4,512), (8,8,256), (16,16,128), (32,32,64), (64,64,32), (128,128,16), and (256,256,1), respectively. The dimensions of the reconstructed chip layout obtained based on the eighth upsampling feature are (256,256,1), which is consistent with the dimensions of the sample chip layout.
[0102] In the embodiment of the present application, determining the next upsampling feature of any one upsampling feature based on the downsampling feature corresponding to the upsampling feature essentially means splicing together the output of the convolution block in the initial encoder and the output of the deconvolution block in the initial decoder to jointly serve as the input of the next deconvolution block, which can improve the expression ability of the upsampling feature obtained after the deconvolution block performs the upsampling process, thereby improving the accuracy of the reconstructed chip layout and making the reconstructed chip layout closer to the sample chip layout.
[0103] Optionally, the reconstructed chip layout output by the initial decoder is a binary mask layout. The binary mask layout includes a plurality of pixel points, and the value of each pixel point represents that the pixel point is transparent or opaque. Optionally, if the value of a pixel point is a first value, the pixel point represents that the pixel point is transparent, and if the value of a pixel point is a second value, the pixel point represents that the pixel point is opaque. The first value is any numerical value, for example, the first value is 0. The second value is any numerical value different from the first value, for example, the second value is 1. That is, if the first value is 0 and the second value is 1, when the value of a pixel point on the binary mask layout is 1, the pixel point represents that the pixel point is opaque, and when the value of a pixel point on the binary mask layout is 0, the pixel point represents that the pixel point is transparent.
[0104] If a reconstructed chip layout can be obtained, step 204 includes training an initial encoder based on the sample chip layout, the reconstructed chip layout, layout features of the sample chip layout (i.e., sample layout features), and layout features of each reference chip layout (i.e., reference layout features) to obtain a chip layout encoder (i.e., a trained encoder).
[0105] The loss of the initial encoder can be determined based on the sample chip layout, the reconstructed chip layout, the layout features of the sample chip layout, and the layout features of each reference chip layout. The loss of the initial encoder is used to train an initial encoder to obtain a chip layout encoder.
[0106] Optionally, training an initial encoder based on the sample chip layout, the reconstructed chip layout, the layout features of the sample chip layout, and the layout features of each reference chip layout to obtain a chip layout encoder includes steps B1 to B3 as follows.
[0107] Step B1: Determine a first loss based on the layout features of the sample chip layout (i.e., the sample layout features) and the layout features of each reference chip layout (i.e., the reference layout features). Here, the process of determining the first loss has already been described in step 204, so it will not be described in detail again here.
[0108] Step B2: Determine a second loss based on the sample chip layout and the reconfigured chip layout.
[0109] In the embodiment of the present application, at the pixel level, the sample chip layout and the reconstructed chip layout are compared, and the distance between the sample chip layout and the reconstructed chip layout is obtained, and the distance is taken as the second loss.
[0110] Optionally, the second loss is determined by employing equation (3) below.
[0111]
number
[0112] Here, Loss rebuild represents the second loss. x i represents the i-th sample chip layout. xi(pred) represents the reconstructed chip layout corresponding to the i-th sample chip layout. The total number of sample chip layouts is N+1. Σ represents the symbol of the addition function.
[0113] Step B3: Train an initial encoder based on the first loss and the second loss to obtain a chip layout encoder (i.e., a trained encoder).
[0114] In the embodiment of the present application, the first loss and the second loss can be subjected to arithmetic processing such as addition, weighted addition, averaging, and weighted averaging, and the arithmetic processing result can be determined as the initial encoder loss. For example, the initial encoder loss is expressed as Loss=Loss rebuild +Loss rotation We use the loss of the initial encoder to train the initial encoder and obtain a chip layout encoder.
[0115] Since the second loss is determined based on the sample chip layout and the reconstructed chip layout, the second loss is used to train the initial encoder, so that the initial encoder extracts the layout features of the sample chip layout, and then the initial decoder reconstructs a reconstructed chip layout that is closer to the sample chip layout based on the layout features of the sample chip layout, thereby ensuring that the layout features of the sample chip layout can accurately represent the sample chip layout, and improving the accuracy of the layout features of the sample chip layout.
[0116] It is necessary to mention that in the process of using the loss of the initial encoder to train the initial encoder, the loss of the initial encoder can be further used to synchronously train the initial decoder, thereby improving the training effect. In application, the reconstructed chip layout can be further determined according to the layout characteristics of the reference chip layout. The reconstructed chip layout is a chip layout reconstructed according to the layout characteristics of the reference chip layout. The reference chip layout and the reconstructed chip layout are used to determine the third loss, and the loss of the initial encoder is determined using at least one of the first loss, the second loss, and the third loss, so that the loss of the initial encoder is used to train the initial encoder and obtain a chip layout encoder. Here, the method of determining the reconstructed chip layout is similar to the method of determining the reconstructed chip layout, and the method of determining the third loss is similar to the method of determining the second loss, so that they will not be described in detail again here.
[0117] It is necessary to explain that the information (including but not limited to user device information and user personal information, etc.), data (including but not limited to data used for analysis, stored data, and displayed data, etc.), and signals related to this application are all authorized by the user or fully authorized by each party, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the sample chip layouts, etc. related to this application are all obtained with sufficient authorization.
[0118] The above method performs a geometric transformation on a sample chip layout to obtain a reference chip layout, and uses layout features of the sample chip layout and the reference chip layout to train an initial encoder to obtain a chip layout encoder, so that the chip layout encoder can output layout features that are close to the chip layouts before and after the geometric transformation. Because the chip layouts before and after the geometric transformation belong to the same chip layout type, the chip layout encoder can focus on extracting layout features that can distinguish the chip layout type, and then, when performing clustering and screening processes on the chip layout based on the layout features, it is advantageous in improving the accuracy of the clustering result and reducing the redundancy of the screening result.
[0119] The embodiment of the present application further provides a chip layout screening method, which can be applied in the above implementation environment, and can accurately screen a key chip layout from a plurality of target chip layouts. In the example of the flowchart of the chip layout screening method provided by the embodiment of the present application shown in Figure 5, for convenience of description, the terminal device 101 or the server 102 that executes the chip layout screening method in the embodiment of the present application is called an electronic device, and the method may be executed by an electronic device. As shown in Figure 5, the method includes the following steps:
[0120] Step 501: Obtain multiple target chip layouts and a trained encoder.
[0121] The trained encoder may be referred to as a chip layout encoder. Step 501 may be said to obtain a number of target chip layouts and a chip layout encoder.
[0122] In the embodiment of the present application, any one of the target chip layouts is an IC layout or a sub-region obtained by dividing an IC layout, and the method of determining the target chip layout is the same as the method of determining the sample chip layout, and the description of step 201 can be referred to, so that it will not be described in detail again here. The chip layout encoder is obtained by training according to the training method used for the chip layout encoder related to FIG. 2, and the description of steps 201 to 205 can be referred to, so that it will not be described in detail again here.
[0123] Step 502: Extract target layout features for each target chip layout using the trained encoder.
[0124] The target layout features may also be referred to as layout features of the target chip layout. Step 502 may also refer to extracting layout features of each target chip layout using a chip layout encoder.
[0125] Input any one of the target chip layouts into the chip layout encoder, and perform feature extraction on the target chip layout through the chip layout encoder to obtain the layout features of the target chip layout. Here, the method of determining the layout features of the target chip layout is similar to the method of determining the layout features of the sample chip layout, and the relevant description of step 203 can be referred to, so that the detailed description will not be repeated here.
[0126] Step 503: Perform a clustering process on the multiple target chip layouts according to each target layout feature to obtain multiple target clustering groups, where at least one target chip layout is included in any one of the target clustering groups.
[0127] The layout features of the target chip layouts can represent the target chip layouts, and concentrating the multiple target chip layouts into multiple target clustering groups can be realized by calculating the layout features of each target chip layout.
[0128] Exemplarily, for any two target chip layouts, the distance between the layout features of the two target chip layouts can be calculated according to a distance formula. If the distance between the layout features of the two target chip layouts is smaller than the distance threshold, it means that the two target chip layouts are similar, and the two target chip layouts are concentrated into the same initial clustering group. If the distance between the layout features of the two target chip layouts is equal to or greater than the distance threshold, it means that the two target chip layouts are not similar, and the two target chip layouts are concentrated into different initial clustering groups. In this manner, multiple target chip layouts can be concentrated into multiple initial clustering groups, and each initial clustering group includes at least one target chip layout. Optionally, each initial clustering group is a target clustering group, thereby realizing the concentration of multiple target chip layouts into multiple target clustering groups. The embodiment of the present application does not limit the distance formula, and exemplarily, the distance formula can be a Euclidean space distance formula or a cosine distance formula, etc. The value of the distance threshold can be set according to experience.
[0129] In one possible implementation form, step 503 includes steps 5031 to 5034.
[0130] Step 5031: Obtain a plurality of first layout features, and one first layout feature is used to represent a clustering center of one first clustering group.
[0131] The embodiment of the present application does not limit the method by which the electronic device acquires the first layout feature. For example, a plurality of first layout features may be arranged on the electronic device. Or, a user may input a plurality of first layout features to the electronic device. Or, the electronic device may concentrate a plurality of target chip layouts into a plurality of initial clustering groups, and for any one of the initial clustering groups, may average the layout features of each target chip layout included in the initial clustering group, and the obtained result may be one first layout feature. In this case, the number of first layout features is the same as the number of the initial clustering groups. Or, according to the method of steps 5031 to 5038, a plurality of target chip layouts are clustered a plurality of times, and a plurality of clustering groups are obtained after the last clustering of the plurality of clusterings, and the layout features of each target chip layout included in each clustering group are averaged, and the obtained result is each first layout feature. One first layout feature is a desired clustering center of one first clustering group, and can represent the clustering center of the first clustering group.
[0132] Step 5032: Calculate the distance between the layout feature of each target chip layout (i.e., the target layout feature) and each first layout feature. Illustratively, the distance between the layout feature of any one of the target chip layouts and any one of the first layout features can be calculated according to a Euclidean space distance formula, a cosine distance formula, or the like.
[0133] Step 5033: For any one of the target chip layouts, select a minimum first distance from the distances between the layout features (i.e., the target layout features) of any one of the target chip layouts and each first layout feature, and concentrate the any one of the target chip layouts into a first clustering group corresponding to the first layout feature corresponding to the first distance.
[0134] In the embodiment of the present application, the smaller the distance between the layout feature of the target chip layout and the first layout feature, the more similar the layout feature of the target chip layout is to the first layout feature, and the two can represent the same type of chip layout. Based on this principle, a minimum first distance can be selected from the distances between the layout feature of any one of the target chip layouts and each of the first layout features, and the first layout feature corresponding to the first distance is the first layout feature that is most similar to the layout feature of the target chip layout among the first layout features. The target chip layout can be concentrated into a first clustering group corresponding to the first layout feature corresponding to the first distance.
[0135] In this manner, multiple target chip layouts can be concentrated into multiple first clustering groups, and the target chip layouts in the same first clustering group correspond to the same type of chip layouts, so that the target chip layouts can be clustered according to their chip layout types in this manner.
[0136] Step 5034: If each of the first clustering groups satisfies the clustering termination condition, each of the first clustering groups is set as a target clustering group.
[0137] The embodiment of the present application does not limit that any one of the first clustering groups satisfies the clustering end condition. Illustratively, any one of the first clustering groups satisfies the clustering end condition when the number of clustering times corresponding to the first clustering group reaches a set number of times. For example, if the number of clustering times corresponding to one first clustering group is 50 times, the first clustering group satisfies the clustering end condition when the set number of times is reached. It is necessary to explain, other implementation forms in which any one of the first clustering groups satisfies the clustering end condition are further described below, and will not be described in detail here.
[0138] Optionally, for any one of the first clustering groups, an actual clustering center of the first clustering group can be determined based on the layout features of each target chip layout in the first clustering group. Illustratively, an averaging operation is performed on the layout features of each target chip layout in the i-th first clustering group according to the following formula (4) to obtain the actual clustering center of the first clustering group.
[0139]
number
[0140] Here, μ i represents the actual clustering center of the i-th clustering group. C i represents the number of target chip layouts included in the i-th first clustering group, x represents the layout features of the target chip layout, and Σ is the function symbol of the summation function.
[0141] In one possible implementation, a first clustering number of clustering processes are performed on a plurality of target chip layouts, and each first clustering group is obtained after the last clustering process of the first clustering number of clustering processes. At this time, the first clustering number-1 clustering process has already been performed on the plurality of target chip layouts, and each clustering group can be obtained after the last clustering process of the first clustering number-1 clustering process. That is, if M (M is a positive integer) clustering processes are performed on the plurality of target chip layouts, and the last clustering process of the M clustering processes is called the Mth clustering process, each first clustering group is obtained after the Mth clustering process, and each clustering group is obtained after the M-1th clustering process. In an embodiment of the present application, an actual clustering center of each clustering group can be obtained. If an error between the actual clustering center of each clustering group and the actual clustering center of each first clustering group is within a set range, it is determined that each first clustering group satisfies a clustering end condition.
[0142] In another possible implementation form, the error of the first clustering group is determined according to the layout features of each target chip layout in any one of the first clustering groups and the actual clustering center of the first clustering group, and the sum of the errors of each first clustering group is determined to obtain the clustering total error. Optionally, the sum of squared differences between the layout features of each target chip layout in any one of the first clustering groups and the actual clustering center of the first clustering group is calculated according to the following formula (5), to obtain the error of the first clustering group, and the sum of the errors of each first clustering group is the clustering total error.
[0143]
number
[0144] Here, Error represents the total clustering error. N represents the number of the first clustering group. μ i represents the actual clustering center of the i-th clustering group. C i represents the number of target chip layouts included in the i-th first clustering group, x represents the layout features of the target chip layout, and Σ is the function symbol of the summation function.
[0145] In one possible implementation form, when the clustering total error of each of the first clustering groups is smaller than a set total error, it is determined that each of the first clustering groups satisfies the clustering end condition. Alternatively, a clustering process is performed a first clustering number of times on a plurality of target chip layouts to obtain each of the first clustering groups. At this time, it is possible to obtain a clustering total error of each of the clustering groups obtained after performing a clustering process (first clustering number - 1) on the target chip layout. If a difference between the clustering total error of each of the clustering groups and the clustering total error of each of the first clustering groups is within a set range, it is determined that each of the first clustering groups satisfies the clustering end condition.
[0146] When each of the first clustering groups satisfies the clustering termination condition, the first clustering groups are set as the target clustering groups.
[0147] Optionally, after step 5033, steps 5035 to 5038 are further included.
[0148] Step 5035: if each of the first clustering groups does not satisfy the clustering termination condition, determine a second layout feature for each of the first clustering groups based on the layout features of each target chip layout in each of the first clustering groups, and the second layout feature is used to represent a clustering center of the second clustering group.
[0149] Each of the first clustering groups not satisfying the clustering end condition corresponds to the existence of a first clustering group that does not satisfy the clustering end condition among the multiple first clustering groups. At this time, based on the layout feature (i.e., the target layout feature) of each target chip layout in any one of the first clustering groups, the actual clustering center of the first clustering group can be determined, and the actual clustering center of the first clustering group can be taken as one second layout feature. This second layout feature is the desired clustering center of one second clustering group, and can represent the clustering center of the second clustering group.
[0150] Step 5036: Calculate the distance between the layout feature of each target chip layout (i.e., the target layout feature) and each second layout feature. Illustratively, the distance between the layout feature of any one of the target chip layouts and any one of the second layout features can be calculated according to a Euclidean space distance formula, a cosine distance formula, or the like.
[0151] Step 5037: For any one of the target chip layouts, select a minimum second distance from the distances between the layout features (i.e., the target layout features) of any one of the target chip layouts and each second layout feature, and concentrate the any one of the target chip layouts into a second clustering group corresponding to the second layout feature corresponding to the second distance.
[0152] The smaller the distance between the layout feature of the target chip layout and the second layout feature, the more similar the layout feature of the target chip layout is to the second layout feature, and the two can represent the same type of chip layout. Based on this principle, a minimum second distance can be selected from the distances between the layout feature of any one of the target chip layouts and each second layout feature. The second layout feature corresponding to the second distance is the second layout feature that is most similar to the layout feature of the target chip layout among the second layout features. The target chip layout can be concentrated into a second clustering group corresponding to the second layout feature corresponding to the second distance.
[0153] In this manner, multiple target chip layouts can be concentrated into multiple second clustering groups, and each target chip layout in the same second clustering group corresponds to a chip layout of the same type, so that each target chip layout can be clustered according to the type of chip layout in this manner.
[0154] Step 5038: If each of the second clustering groups satisfies the clustering termination condition, each of the second clustering groups is set as a target clustering group.
[0155] When each second clustering group satisfies the clustering termination condition, each second clustering group is taken as a target clustering group. If each second clustering group does not satisfy the clustering termination condition, the multiple target chip layouts can be clustered again according to the method of steps 5035 to 5038, and the process continues until each target clustering group is obtained.
[0156] By clustering the multiple target chip layouts multiple times using the method of steps 5031 to 5038, the clustering center of the clustering group can be continuously changed, and the target chip layouts in the same clustering group become more and more similar, while the target chip layouts in different clustering groups become more and more dissimilar, until the clustering end condition is met. Satisfying the clustering end condition corresponds to the clustering group converging, and the clustering center of the clustering group no longer fluctuates. Therefore, the error of the target clustering group obtained after clustering the multiple target chip layouts multiple times using the method of steps 5031 to 5038 is relatively small, and the accuracy is relatively high.
[0157] Step 504: For any one of the target clustering groups, screen key chip layouts from among the target chip layouts included in the any one of the target clustering groups.
[0158] In the embodiment of the present application, each target chip layout included in the target clustering group belongs to the same chip layout type, therefore, each target chip layout included in any one target clustering group can be randomly sampled to obtain a key chip layout. The number of key chip layouts is at least one. By clustering each target chip layout, obtaining each target clustering group, and screening the key chip layout from the target chip layout included in each target clustering group, it is possible to avoid manually selecting the key chip layout from all target chip layouts, and prevent the interference of human experience, random selection, etc. on the final result.
[0159] In one possible implementation form, step 504 includes a step of obtaining a distance between layout features of each target chip layout in any one of the target clustering groups and a clustering center of any one of the target clustering groups, and a step of uniformly sampling each target chip layout included in any one of the target clustering groups based on the distance between the layout features of each target chip layout in any one of the target clustering groups and the clustering center of any one of the target clustering groups to obtain multiple key chip layouts.
[0160] In the embodiment of the present application, the clustering center of any one of the target clustering groups may be the desired clustering center of the target clustering group, or may be the actual clustering center of the target clustering group. Here, the calculation method of the desired clustering center of the target clustering group can refer to the relevant content of the determination of the second layout feature, and the realization principles of both are similar. The calculation method of the actual clustering center of the target clustering group can refer to the relevant content of the determination of the actual clustering center of the first clustering group, and the realization principles of both are similar.
[0161] For any one of the target clustering groups, the distance between the layout feature of any one of the target chip layouts included in the target clustering group and the clustering center of the target clustering group can be calculated according to the Euclidean space distance formula, the cosine distance formula, etc. In this manner, the distance between the layout feature of each target chip layout in the target clustering group and the clustering center of the target clustering group can be obtained.
[0162] Optionally, determine a maximum distance from the distance between the layout feature of each target chip layout in the target clustering group and the clustering center of the target clustering group, and obtain a sampling interval by dividing the maximum distance by the sampling number. Screen a sampling distance whose distance is an integer multiple of the sampling interval from the distance between the layout feature of each target chip layout in the target clustering group and the clustering center of the target clustering group. Sample one key chip layout from the target chip layout corresponding to the sampling distance.
[0163] For example, assume that there are M target clustering groups and the sampling number is L+1. For any one target clustering group, determine the maximum distance from the distance between the layout feature of each target chip layout in the target clustering group and the clustering center of the target clustering group, and divide the maximum distance by L+1 to obtain the sampling interval. From the distance between the layout feature of each target chip layout in the target clustering group and the clustering center of the target clustering group, screen the sampling distances whose distances are 0, 1*sampling interval, 2*sampling convolution, ..., and (L+1)*sampling interval, and sample a total of L+2 target chip layouts from the target chip layouts corresponding to these sampling distances. Here, the layout feature of the target chip layout corresponding to the sampling distance 0 is the clustering center of the target clustering group, and the distance between the layout feature of the target chip layout corresponding to the sampling distance "(L+1)*sampling interval" and the clustering center of the target clustering group is the maximum distance. Since one target clustering group can sample L+2 target chip layouts, M target clustering groups can sample a total of M×(L+2) target chip layouts.
[0164] In one possible implementation form, after step 504, the method further includes a step of performing light source mask optimization on the key chip layout to obtain a target light source and a mask layout corresponding to the key chip layout, and a step of performing mask optimization on other chip layouts based on the target light source to obtain a mask layout corresponding to the other chip layouts, where the other chip layouts are target chip layouts other than the key chip layout among each target chip layout.
[0165] In the embodiment of the present application, performing source mask optimization on the key chip layout corresponds to performing source and mask combined optimization (SMO) on the key chip layout, and a target source and a mask layout corresponding to the key chip layout can be obtained. Here, SMO is an important resolution enhancement technology for realizing nanometer (e.g., 28 nanometers, even smaller nanometers) integrated circuits. By screening the key chip layout from the target chip layout and performing SMO on the key chip layout, the number of key chip layouts can be made smaller than the number of target chip layouts, thereby improving the speed of SMO. In addition, by clustering the target chip layout and screening the key chip layout from the target chip layouts included in each clustering group, it can be ensured that the key chip layout covers various chip layout types. In addition, since the redundancy phenomenon of the key chip layout is relatively small, the accuracy of the target source obtained by performing SMO on the key chip layout is relatively high, thereby improving the efficiency and effect of SMO.
[0166] SMO is a type of source optimization (SO) technology. SMO technology can adjust the intensity and direction of incident light by changing the intensity distribution of the light source, so that the quality of the imaged layout obtained by exposing the mask layout on the wafer using the light source is relatively high, which contributes to improving the chip yield. When SMO is performed on a key chip layout according to the method of the embodiment of the present application, the speed and effect of SMO can be improved, so that the embodiment of the present application can improve the photolithography resolution and increase the photolithography process window. In addition, the target chip layout in the embodiment of the present application is a layout expressed by pixels, so it is relatively versatile and can be applied to chip layout types such as contact hole type, logic layout type, and dense line type.
[0167] When performing any one SMO, an arbitrary light source is used as an initial light source, a photolithography device is operated, and a key chip layout is exposed on a photoresist based on the initial light source to obtain an intermediate mask layout. The photolithography device is further operated, and the intermediate mask layout is exposed on a wafer using the initial light source to obtain an imaged layout. The imaged layout is compared with the key chip layout to obtain an error between the imaged layout and the key chip layout.
[0168] Optionally, if the error between the imaging layout and the key chip layout satisfies the optimization condition, the initial light source is the target light source, and the intermediate mask layout is the mask layout corresponding to the key chip layout. The mask layout corresponding to the key chip layout is used to expose the mask layout corresponding to the key chip layout on the wafer by operating the photolithography device using the target light source, and obtain the imaging layout corresponding to the key chip layout. If the error between the imaging layout and the key chip layout does not satisfy the optimization condition, the initial light source is adjusted according to the error between the imaging layout and the key chip layout, and an adjusted initial light source is obtained. The adjusted initial light source is the initial light source of the next SMO, and at least one SMO as described above is performed according to the initial light source, and continues until the target light source and the mask layout corresponding to the key chip layout are obtained.
[0169] Optionally, satisfying the optimization condition is that the error between the imaging layout and the key chip layout is within a set error range, or satisfying the optimization condition is that the gradient of the error between the imaging layout and the key chip layout is smaller than a set gradient threshold, in which case, when adjusting the initial light source based on the error between the imaging layout and the key chip layout, the initial light source can be adjusted based on the gradient of the error between the imaging layout and the key chip layout.
[0170] Then, a photolithography device is operated to expose another chip layout onto a photoresist based on the target light source, and a mask layout corresponding to the other chip layout is obtained. The mask layout corresponding to the other chip layout is used for exposing the mask layout corresponding to the other chip layout onto a wafer by using a target light source through the photolithography device to obtain an imaging layout corresponding to the other chip layout.
[0171] Optionally, if the key chip layout and the other chip layout are obtained by dividing one chip layout, in this case, the light source mask optimization is performed on the key chip layout to obtain a target light source, and then a photolithography device is operated to expose the chip layout onto a photoresist according to the target light source, and a mask layout corresponding to the chip layout is obtained. The mask layout corresponding to the chip layout is used for exposing the mask layout corresponding to the chip layout onto a wafer by using a target light source through the photolithography device, and obtaining an imaging layout corresponding to the chip layout.
[0172] It is necessary to explain that the information (including but not limited to user device information and user personal information, etc.), data (including but not limited to data used for analysis, stored data, and displayed data, etc.), and signals related to this application are all authorized by the user or fully authorized by each party, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the target chip layout, etc. related to this application are all obtained with sufficient authorization.
[0173] The chip layout encoder in the above method focuses on extracting layout features that can distinguish chip layout types. Therefore, when the layout features of each target chip layout are extracted using the chip layout encoder and clustering processing is performed on multiple target chip layouts based on the layout features of each target chip layout, target chip layouts of the same chip layout type can be accurately concentrated in the same target clustering group, and target chip layouts of different chip layout types can be concentrated in different target clustering groups, thereby improving the accuracy of the clustering result. When screening key chip layouts from each target chip layout included in any one target clustering group, the redundancy of the key chip layout can be reduced and the screening quality can be improved.
[0174] The above describes the training method and the screening method of chip layout used in the chip layout encoder according to the embodiment of the present application from the viewpoint of method steps, which will be further described below in conjunction with Fig. 6. Fig. 6 is a schematic diagram of the chip layout encoder according to the embodiment of the present application and clustering and screening based on the chip layout encoder.
[0175] In the embodiment of the present application, a sample data set is obtained, and the sample data set is used to train an initial encoder according to the training method used in the chip layout encoder related to FIG. 2 to obtain a chip layout encoder. The sample data set in the embodiment of the present application includes two open source data sets. One data set includes 4877 chip layouts, and the 4877 chip layouts or some chip layouts can be extracted from the 4877 chip layouts to be sample chip layouts. The other data set includes 5394 chip layouts, and the 5394 chip layouts or some chip layouts can be extracted from the 5394 chip layouts to be sample chip layouts. The sample chip layouts are used to train the initial encoder. In addition, the sample data set in the embodiment of the present application further includes some contact hole type chip layouts, logic layout type chip layouts, and dense line type chip layouts, etc.
[0176] After training and obtaining the chip layout encoder, the chip layout encoder can be used to screen key chip layouts from multiple target chip layouts. The structure of the chip layout encoder can refer to the relevant description in FIG. 4, and will not be described in detail again here.
[0177] In an embodiment of the present application, a chip layout encoder can be used to extract layout features of each target chip layout. Then, each target chip layout is clustered based on the layout features of each target chip layout to obtain a plurality of target clustering groups. At least one target chip layout is included in any one target clustering group. Then, a key chip layout is screened from each target clustering group, for example, for any one target clustering group, a key chip layout can be screened from each target chip layout included in the target clustering group.
[0178] The chip layout encoder focuses on extracting layout features that can distinguish chip layout types, and therefore, when the layout features of each target chip layout are extracted using the chip layout encoder, and clustering processing is performed on a plurality of target chip layouts based on the layout features of each target chip layout, the accuracy of the clustering result can be improved, and when screening key chip layouts from each target clustering group, the redundancy of the key chip layouts can be reduced, and screening quality can be improved.
[0179] FIG. 7 shows a structural schematic diagram of a training device used in the chip layout encoder provided by the embodiment of the present application. As shown in FIG. 7, the device includes: an acquisition module 701 used to acquire a sample chip layout and an initial encoder; A transformation module 702 is used to perform a geometric transformation on the sample chip layout to obtain at least one reference chip layout; An extraction module 703 used for extracting layout features of the sample chip layout and layout features of each reference chip layout by the initial encoder, or an extraction module 703 used for extracting sample layout features of the sample chip layout and reference layout features of each reference chip layout using the initial encoder; and a training module 704 that is used to train an initial encoder based on layout features of a sample chip layout (i.e., sample layout features) and layout features of each reference chip layout (i.e., reference layout features) to obtain a chip layout encoder (i.e., a trained encoder), where the chip layout encoder is used to extract layout features of a target chip layout.
[0180] In one possible implementation, the conversion module 702 is used to perform a mirror flip process on the sample chip layout to obtain a symmetric chip layout, perform a rotation process on the sample chip layout and the symmetric chip layout to obtain a rotated chip layout, and use the symmetric chip layout and the rotated chip layout as at least one reference chip layout. Alternatively, the conversion module 702 is used to perform a mirror flip process on the sample chip layout to obtain a flipped chip layout, and perform a rotation process on the sample chip layout and the flipped chip layout to obtain a rotated chip layout, where the at least one reference chip layout includes the flipped chip layout and the rotated chip layout.
[0181] In one possible implementation, the conversion module 702 is used to perform a mirror flip process on the sample chip layout to obtain a symmetric chip layout, and the symmetric chip layout is used as at least one reference chip layout, or to perform a rotation process on the sample chip layout to obtain a rotated chip layout, and the rotated chip layout is used as at least one reference chip layout. Alternatively, the conversion module 702 is used to perform a mirror flip process on the sample chip layout to obtain a flipped chip layout, and the at least one reference chip layout includes the flipped chip layout, or to perform a rotation process on the sample chip layout to obtain a rotated chip layout, and the at least one reference chip layout includes the rotated chip layout.
[0182] In one possible implementation, the conversion module 702 is used to perform a rotation process on the sample chip layout to obtain a rotated chip layout, perform a mirror flip process on the sample chip layout and the rotated chip layout to obtain a symmetric chip layout, and take the symmetric chip layout and the rotated chip layout as at least one reference chip layout. Alternatively, the conversion module 702 is used to perform a rotation on the sample chip layout to obtain a rotated chip layout, and perform a mirror flip process on the sample chip layout and the rotated chip layout to obtain an inverted chip layout, where the at least one reference chip layout includes the rotated chip layout and the inverted chip layout.
[0183] In one possible implementation, the extraction module 703 is used to perform multiple downsampling operations on the sample chip layout by the initial encoder (i.e., using ), and obtain downsampling features obtained in each downsampling operation. The layout features of the sample chip layout are the downsampling features obtained in the last downsampling operation.
[0184] In one possible implementation, the apparatus includes: A reconfiguration module used for determining a reconfigured chip layout based on layout characteristics of the sample chip layout, the reconfigured chip layout being a chip layout reconfigured based on the layout characteristics of the sample chip layout; and The method further includes a training module 704 that is used to train an initial encoder based on the sample chip layout, the reconstructed chip layout, layout features of the sample chip layout, and layout features of each reference chip layout to obtain a chip layout encoder.
[0185] In one possible implementation, the reconstruction module is used to perform multiple upsampling processes on layout features of the sample chip layout to obtain each upsampling feature, and the reconstructed chip layout is a chip layout obtained based on the upsampling features obtained in the last upsampling process.
[0186] In one possible implementation form, the reconstruction module performs a first upsampling process on layout features of the sample chip layout to obtain a first upsampling feature, where the layout features of the sample chip layout are obtained by performing multiple downsampling processes on the sample chip layout; and for any one of the upsampling features obtained in any one of the upsampling processes, obtain a downsampling feature corresponding to any one of the upsampling features, and when performing the downsampling process on the sample chip layout to obtain a downsampling feature corresponding to any one of the upsampling features, the downsampling feature is obtained. The number of times of ring processing is a first number, and when performing an upsampling process on a layout feature of a sample chip layout to obtain any one of the upsampling features, the number of times of upsampling processing is a second number, and the sum of the first number and the second number is a target number; the method is used for splicing any one of the upsampling features and a downsampling feature corresponding to any one of the upsampling features to obtain a spliced feature corresponding to any one of the upsampling features; and performing an upsampling process on the spliced feature corresponding to any one of the upsampling features to obtain a next upsampling feature of any one of the upsampling features.
[0187] In one possible implementation, the training module 704 is used for determining a first loss based on layout features of the sample chip layout and layout features of each reference chip layout, determining a second loss based on the sample chip layout and the reconstructed chip layout, and training an initial encoder based on the first loss and the second loss to obtain a chip layout encoder.
[0188] The above-mentioned device performs a geometric transformation on a sample chip layout to obtain a reference chip layout, and uses layout features of the sample chip layout and the reference chip layout to train an initial encoder to obtain a chip layout encoder, so that the chip layout encoder can output similar layout features for the chip layouts before and after the geometric transformation. Because the chip layouts before and after the geometric transformation belong to the same chip layout type, the chip layout encoder is advantageous in that it focuses on extracting layout features that can distinguish the chip layout type, and then performs clustering and screening on the chip layouts based on the layout features, thereby improving the clustering accuracy and reducing the redundancy of the screening result.
[0189] It should be understood that the device provided in the above Fig. 7 only illustrates the division of each of the above functional modules when implementing its functions, but in actual applications, the above functions can be assigned to different functional modules to be completed as necessary, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the above-described functions. In addition, the device provided in the above embodiment belongs to the same idea as the method embodiment, and the specific implementation process of the device can be referred to the method embodiment, so it will not be described in detail again here.
[0190] FIG. 8 shows a schematic diagram of the structure of a chip layout screening device provided in an embodiment of the present application. As shown in FIG. 8, the device includes: An acquisition module 801 for acquiring a plurality of target chip layouts and a chip layout encoder (i.e., a trained encoder), the chip layout encoder being obtained by training according to a training method used for the chip layout encoder; an extraction module 802, which is used to extract layout features (i.e., target layout features) of each target chip layout using a chip layout encoder; A clustering module 803 is used for performing clustering processing on a plurality of target chip layouts based on layout features (i.e., each target layout feature) of each target chip layout to obtain a plurality of target clustering groups, in which at least one target chip layout is included in any one of the target clustering groups; a screening module 804, which is used for, for any one of the target clustering groups, screening key chip layouts from among the target chip layouts included in any one of the target clustering groups.
[0191] In one possible implementation form, the clustering module 803 is used to obtain a plurality of first layout features, where one first layout feature is used to represent a clustering center of one first clustering group; calculate a distance between the layout features of each target chip layout and each first layout feature; for any one target chip layout, select a minimum first distance from the distances between the layout features of any one target chip layout and each first layout feature, and concentrate any one target chip layout into a first clustering group represented by the first layout feature corresponding to the first distance; and if each first clustering group satisfies a clustering termination condition, each first clustering group is used as a target clustering group.
[0192] In one possible implementation form, the clustering module 803 is further used for: determining, for any one of the first clustering groups, a second layout feature based on the layout features of each target chip layout in any one of the first clustering groups if each of the first clustering groups does not satisfy the clustering termination condition, wherein the second layout feature is used to represent a clustering center of one of the second clustering groups; calculating a distance between the layout features of each of the target chip layouts and each of the second layout features; selecting, for any one of the target chip layouts, a minimum second distance from the distances between the layout features of any one of the target chip layouts and each of the second layout features, and concentrating the any one of the target chip layouts into a second clustering group represented by the second layout feature corresponding to the second distance; and if each of the second clustering groups satisfies the clustering termination condition, selecting each of the second clustering groups as a respective target clustering group.
[0193] In one possible implementation form, the screening module 804 is used for obtaining a distance between layout features of each target chip layout in any one of the target clustering groups and a clustering center of any one of the target clustering groups, and uniformly sampling each target chip layout included in any one of the target clustering groups based on the distance between the layout features of each target chip layout in any one of the target clustering groups and the clustering center of any one of the target clustering groups to obtain multiple key chip layouts.
[0194] In one possible implementation, the apparatus includes: The system further includes an optimization module for performing light source mask optimization on the key chip layout to obtain a target light source and a mask layout corresponding to the key chip layout, and performing mask optimization on other chip layouts based on the target light source to obtain a mask layout corresponding to the other chip layout, where the other chip layouts are target chip layouts other than the key chip layout among each target chip layout.
[0195] The chip layout encoder in the above device focuses on extracting layout features that can distinguish chip layout types. Therefore, when the layout features of each target chip layout are extracted using the chip layout encoder and clustering processing is performed on multiple target chip layouts based on the layout features of each target chip layout, target chip layouts of the same chip layout type can be accurately concentrated in the same target clustering group, and target chip layouts of different chip layout types can be concentrated in different target clustering groups, thereby improving the accuracy of the clustering result. When screening key chip layouts from each target chip layout included in any one target clustering group, the redundancy of the key chip layout can be reduced, and screening quality can be improved.
[0196] It should be understood that the device provided in the above Fig. 8 only illustrates the division of each of the above functional modules when implementing its functions, but in actual applications, the above functions can be assigned to different functional modules as necessary to complete them, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the above-described functions. In addition, the device provided in the above embodiment belongs to the same idea as the method embodiment, and the specific implementation process of the device can be referred to the method embodiment, and will not be described in detail again here.
[0197] 9 shows a structural block diagram of a terminal device 900 provided by an exemplary embodiment of the present application. The terminal device 900 includes: a processor 901 and a memory 902.
[0198] The processor 901 may include one or more processing cores, such as a 4-core processor and an 8-core processor. The processor 901 may be realized by adopting at least one hardware form among DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 901 may include a main processor and a coprocessor, the main processor is a processor used for processing data under a wake-up state and is also called a CPU (Central Processing Unit), and the coprocessor is a low-power processor used for processing data under a standby state. In some embodiments, the processor 901 may be integrated with a GPU (Graphics Processing Unit), which is used for rendering and drawing contents that need to be displayed on a display screen. In some embodiments, the processor 901 may further include an AI (Artificial Intelligence) processor, which is used for processing computation operations related to machine learning.
[0199] The memory 902 may include one or more computer-readable storage media, which may be non-transitory. The memory 902 may further include high-speed random access memory and non-volatile memory, such as one or more magnetic disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 902 is used to store at least one computer program, which is executed by the processor 901 to cause the terminal device 900 to realize the training method for the chip layout encoder or the screening method for the chip layout provided by the method embodiments of the present application.
[0200] In some embodiments, the terminal device 900 further includes a selectable display screen 905 .
[0201] The display screen 905 is used to display a UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 905 is a touch display screen, the display screen 905 further has the ability to collect touch signals on or above the surface of the display screen 905. The touch signals can be input as control signals to the processor 901 for processing. At this time, the display screen 905 can further be used to provide soft buttons and / or virtual buttons, also called soft keyboards, and / or virtual keyboards. In some embodiments, the display screen 905 may be one and is disposed on the front panel of the terminal device 900, and in some other embodiments, the display screen 905 may be at least two and is disposed on different surfaces of the terminal device 900, or is designed to be folded. In some other embodiments, the display screen 905 may be a flexible display screen and is disposed on a curved or folded surface of the terminal device 900. Furthermore, the display screen 905 may also be configured as a non-rectangular irregular shape, i.e., a irregular screen. The display screen 905 may be manufactured by adopting materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode). Exemplarily, chip layouts such as a sample chip layout, a rotated chip layout, and a symmetric chip layout may be displayed by the display screen 905.
[0202] As will be appreciated by those skilled in the art, the structure shown in FIG. 9 does not constitute a limitation on the terminal device 900, which may include more or fewer components than shown, combine certain components, or employ a different component arrangement.
[0203] 10 is a structural diagram of a server provided by an embodiment of the present application, the server 1000 may have a relatively large difference due to different configurations or performances, and may include one or more processors 1001 and one or more memories 1002. Here, at least one computer program is stored in the one or more memories 1002, and the at least one computer program is loaded and executed by the one or more processors 1001, so that the server 1000 realizes the training method for the chip layout encoder or the screening method for the chip layout provided by the embodiment of the above method, and for example, the processor 1001 is a CPU. Of course, the server 1000 may further include components such as a wired or wireless network interface, a keyboard, and an input / output interface for input and output, and the server 1000 may further include other components used to realize the functions of the device, which will not be described in detail here.
[0204] In an exemplary embodiment, the present invention further provides a non-volatile computer readable storage medium, in which at least one computer program is stored, which is loaded and executed by a processor, so that the electronic device realizes the training method for use in any one of the chip layout encoders or the screening method for chip layouts.
[0205] Optionally, the computer-readable storage medium may be a Read-Only Memory (ROM), a Random Access Memory (RAM), a Compact Disc Read-Only Memory (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, and the like.
[0206] In an exemplary embodiment, a computer program is further provided, the computer program being at least one, and when the at least one computer program is loaded and executed by a processor, the electronic device realizes a training method or a chip layout screening method used in any one of the above chip layout encoders.
[0207] In an exemplary embodiment, a computer program product is further provided, the computer program product having stored therein at least one computer program, the at least one computer program being loaded and executed by a processor in an electronic device to implement a training method for use in any one of the chip layout encoders described above or a chip layout screening method.
[0208] As should be understood, "plurality" as referred to herein refers to two or more than two. "And / or" describes a relationship between related objects and indicates that three types of relationships may exist, for example, A and / or B may indicate three cases: A exists alone, A and B exist together, and B exists alone. The character " / " generally indicates that the related objects before and after are in a type of "or" relationship.
[0209] The numbers of the above embodiments of the present application are used only for the purpose of description and do not represent the superiority or inferiority of the embodiments.
[0210] The above are only illustrative examples of the present application, and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present application should be included within the protection scope of the present application.
Claims
1. 1. A training method for a chip layout encoder, performed by electronic equipment, said method comprising: obtaining a sample chip layout and an initial encoder; performing a geometric transformation on the sample chip layout to obtain at least one reference chip layout; extracting sample layout features of the sample chip layout and reference layout features of each reference chip layout using the initial encoder; training the initial encoder based on the sample layout features and the reference layout features to obtain a trained encoder.
2. The step of performing a geometric transformation on the sample chip layout to obtain at least one reference chip layout comprises: performing mirror inversion on the sample chip layout to obtain an inverted chip layout; 2. The method of claim 1, further comprising: performing a rotation on the sample chip layout and the flipped chip layout to obtain a rotated chip layout, wherein the at least one reference chip layout includes the flipped chip layout and the rotated chip layout.
3. The step of performing a geometric transformation on the sample chip layout to obtain at least one reference chip layout comprises: performing a mirror reflection on the sample chip layout to obtain a flipped chip layout, wherein the at least one reference chip layout includes the flipped chip layout; Alternatively, the method of claim 1 further comprises a step of performing a rotation on the sample chip layout to obtain a rotated chip layout, wherein the at least one reference chip layout includes the rotated chip layout.
4. The step of performing a geometric transformation on the sample chip layout to obtain at least one reference chip layout comprises: performing a rotation on the sample chip layout to obtain a rotated chip layout; 2. The method of claim 1, further comprising: performing a mirror flip on the sample chip layout and the rotated chip layout to obtain an inverted chip layout, wherein the at least one reference chip layout includes the rotated chip layout and the inverted chip layout.
5. extracting sample layout features of the sample chip layout using the initial encoder, 5. The method of claim 1, further comprising: performing a plurality of downsampling operations on the sample chip layout using the initial encoder to obtain downsampling features obtained in each downsampling operation, wherein the sample layout features are downsampling features obtained in a final downsampling operation.
6. The step of performing a plurality of downsampling processes on the sample chip layout using the initial encoder and obtaining downsampling features obtained in each downsampling process includes: performing a first downsampling process on the sample chip layout using the initial encoder to obtain a first downsampling feature; and performing a next downsampling process on any one downsampling feature obtained in any one downsampling process using the initial encoder to obtain a downsampling feature obtained in the next downsampling process.
7. The method comprises: determining a reconstructed chip layout based on the sample layout characteristics; training the initial encoder based on the sample layout features and the reference layout features to obtain a trained encoder; 7. The method of claim 1, further comprising: training the initial encoder based on the sample chip layout, the reconstructed chip layout, the sample layout features, and the reference layout features to obtain a trained encoder.
8. determining a reconstructed chip layout based on the sample layout characteristics; 8. The method of claim 7, further comprising: performing a plurality of upsampling processes on the sample layout features to obtain upsampling features obtained in each upsampling process, wherein the reconstructed chip layout is a chip layout obtained based on the upsampling features obtained in a final upsampling process.
9. The step of performing a plurality of upsampling processes on the sample layout features and obtaining upsampling features obtained by each upsampling process includes: performing a first upsampling process on the sample layout features to obtain first upsampling features, the sample layout features being obtained by performing multiple downsampling processes on the sample chip layout; a step of acquiring a downsampling feature corresponding to any one of the upsampling features obtained by any one of the upsampling processes, wherein the number of downsampling processes is a first number when the downsampling process is performed on the sample chip layout to acquire the downsampling feature corresponding to any one of the upsampling features, the number of upsampling processes is a second number when the upsampling process is performed on the sample layout feature to acquire the any one of the upsampling features, and the sum of the first number and the second number is a target number; splicing the one of the upsampling features and the downsampling feature corresponding to the one of the upsampling features to obtain a spliced feature corresponding to the one of the upsampling features; and performing an upsampling process on a spliced feature corresponding to the one upsampling feature to obtain a next upsampling feature of the one upsampling feature.
10. training the initial encoder based on the sample chip layout, the reconstructed chip layout, the sample layout features, and the reference layout features to obtain a trained encoder; determining a first loss based on the sample layout feature and the reference layout feature; determining a second loss based on the sample chip layout and the reconstructed chip layout; training the initial encoder based on the first loss and the second loss to obtain a trained encoder.
11. 1. A method for screening chip layouts, performed by electronic equipment, the method comprising: obtaining a plurality of target chip layouts and a trained encoder, the trained encoder being obtained by training according to a training method for a chip layout encoder according to any one of claims 1 to 10; extracting target layout features for each target chip layout using the trained encoder; A step of performing a clustering process on the plurality of target chip layouts based on each target layout feature to obtain a plurality of target clustering groups, in which at least one target chip layout is included in any one of the target clustering groups; For any one target clustering group, screening a key chip layout from among each target chip layout included in the one target clustering group.
12. The step of performing a clustering process on the plurality of target chip layouts based on each target layout feature to obtain a plurality of target clustering groups includes: obtaining a plurality of first layout features, each of which is used to represent a clustering center of a first clustering group; calculating a distance between each of the target layout features and each of the first layout features; For any one of the target chip layouts, selecting a smallest first distance from distances between the target layout features of the one of the target chip layouts and each of the first layout features, and concentrating the one of the target chip layouts into a first clustering group corresponding to the first layout feature corresponding to the first distance; If each first clustering group satisfies a clustering termination condition, the first clustering group is set as a target clustering group.
13. The method comprises: If each of the first clustering groups does not satisfy the clustering termination condition, for any one of the first clustering groups, determining one second layout feature based on each target layout feature in the one of the first clustering groups, wherein the one second layout feature is used to represent a clustering center of one of the second clustering groups; calculating a distance between each of the target layout features and each of the second layout features; For any one of the target chip layouts, selecting a minimum second distance from the distances between the target layout features of the one of the target chip layouts and each of the second layout features, and concentrating the one of the target chip layouts into a second clustering group corresponding to the second layout feature corresponding to the second distance; The method of claim 12 , further comprising the step of: if each second clustering group satisfies the clustering termination condition, selecting each second clustering group as the respective target clustering group.
14. The step of screening a key chip layout from each target chip layout included in any one of the target clustering groups includes: obtaining a distance between a layout feature of each target chip layout in the one target clustering group and a clustering center of the one target clustering group; The method according to any one of claims 11 to 13, further comprising: a step of uniformly sampling each target chip layout included in any one of the target clustering groups based on the distance between the layout features of each target chip layout in any one of the target clustering groups and the clustering center of the any one of the target clustering groups, to obtain a plurality of key chip layouts.
15. After the step of screening a key chip layout from each target chip layout included in any one of the target clustering groups, performing a light source mask optimization on the key chip layout to obtain a target light source and a mask layout corresponding to the key chip layout; The method according to any one of claims 11 to 14, further comprising: performing mask optimization on other chip layouts based on the target light source to obtain mask layouts corresponding to the other chip layouts, wherein the other chip layouts are target chip layouts other than the key chip layout among the target chip layouts.
16. 1. A training device for use in a chip layout encoder, the device comprising: an acquisition module used to acquire a sample chip layout and an initial encoder; a transformation module used for performing a geometric transformation on the sample chip layout to obtain at least one reference chip layout; an extraction module used to extract sample layout features of the sample chip layout and reference layout features of each reference chip layout using the initial encoder; a training module used for training the initial encoder based on the sample layout features and the reference layout features to obtain a trained encoder.
17. 1. A chip layout screening apparatus, comprising: an acquisition module used to acquire a plurality of target chip layouts and a trained encoder, the trained encoder being obtained by training according to the training method for chip layout encoders according to any one of claims 1 to 10; an extraction module adapted to extract target layout features of each target chip layout using the trained encoder; A clustering module is used to perform a clustering process on the plurality of target chip layouts based on each target layout feature to obtain a plurality of target clustering groups, wherein at least one target chip layout is included in any one of the target clustering groups; and a screening module for screening a key chip layout from among each target chip layout included in any one target clustering group, for each target clustering group.
18. An electronic device comprising a processor and a memory, wherein at least one computer program is stored in the memory, and wherein the at least one computer program is loaded and executed by the processor, thereby causing the electronic device to realize the training method used in a chip layout encoder according to any one of claims 1 to 10, or the chip layout screening method according to any one of claims 11 to 15.
19. A non-volatile computer-readable storage medium having at least one computer program stored therein, the at least one computer program being loaded and executed by a processor to cause an electronic device to implement the training method used in a chip layout encoder according to any one of claims 1 to 10, or the chip layout screening method according to any one of claims 11 to 15.
20. A computer program product, wherein at least one computer program is stored in the computer program product, and when the at least one computer program is loaded and executed by a processor, an electronic device realizes the training method used in a chip layout encoder according to any one of claims 1 to 10, or the chip layout screening method according to any one of claims 11 to 15.
21. A computer program, the computer program being at least one, wherein the at least one computer program is loaded and executed by a processor, causing an electronic device to realize the training method used in a chip layout encoder according to any one of claims 1 to 10, or the chip layout screening method according to any one of claims 11 to 15.
Citation Information
Patent Citations
Information processing device, information processing method, and information processing program
JP2021137127A
Grain identification method, device, and computer storage medium
JP2021507389A
Bacterial species identification support method, multi-colony learning model generation method, bacterial species identification support device and computer program
JP2022046265A
Classifying patterns in electronic circuit layouts using machine learning-based coding
JP2022533704A
Predicting local layout effects in circuit design patterns
US20200380088A1
