Compensation model creation method, photolithography exposure simulation method, device and equipment

Abstract

The present invention discloses a compensation model creation method, a lithography exposure simulation method, a device and an apparatus, wherein the compensation model creation method comprises: determining the mask pattern transmission parameters corresponding to the test mask pattern according to the chip pattern sample; and collecting the exposure image data set corresponding to the test mask pattern; inputting the mask pattern transmission parameters into the constructed neural network model to obtain the predicted compensation parameters; inputting the predicted compensation parameters and the mask pattern transmission parameters into the computational lithography model to obtain the simulated exposure image; performing error comparison between the simulated exposure image and the real exposure image in the exposure image data set to obtain the comparison error; optimizing the neural network model according to the comparison error until the comparison error is not greater than the preset error, thereby obtaining the compensation neural network model. The training of the compensation neural network model in the present application is simple, easy to implement and has high accuracy, which is conducive to improving the accuracy of the simulation of the computational lithography model.

Description

Technical Field

[0001] The present invention relates to the field of photolithography exposure technology, and in particular to a compensation model creation method and device for photolithography exposure simulation, a photolithography exposure simulation method and device, and an electronic device. Background Art

[0002] Computational lithography uses computers to simulate the photochemical reactions and physical processes of lithography to simulate lithography exposure images, and uses simulated exposure images to assist chip design, analysis, prediction and correction to improve chip development cycle and processing yield. In the actual lithography exposure process, the node / size of high-end chip patterns is much smaller than the wavelength of the exposure light, which will produce diffraction effects at the edges of the photomask patterns; therefore, in computational lithography, it is necessary to compensate for the edge diffraction effects during the exposure process accordingly, so as to ensure the accuracy of the simulation of the exposure process and avoid distortion of the simulated exposure images.

[0003] One of the effective technical means to achieve computational lithography compensation is to add fitting functions or lookup tables to the edges of mask patterns for spectrum compensation. However, with the reduction in key dimensions and the increase in complexity of high-end chips, the coupling information between different patterns on the mask plate has a greater impact on the exposure image, and the higher-order effects of the coupling effect cannot be ignored. Therefore, when determining the compensation signal in the computational lithography process, the numerical simulation process based on physical partial differential equations has the problem of difficulty in convergence, the fitting function has the problem of difficulty in construction, and the lookup table has the problem of difficulty in creation. Therefore, how to calculate the compensation signal in the lithography process is one of the important issues that the industry urgently needs to solve. Summary of the invention

[0004] The purpose of the present invention is to provide a compensation model creation method and device for lithography exposure simulation, a lithography exposure simulation method and device, and an electronic device. Through neural network training, a compensation neural network model that can accurately predict the compensation signal in the computational lithography process can be constructed. The training process is simple and easy to implement, which ensures that the computational lithography model can simulate the lithography exposure process more accurately, and provides a reliable theoretical basis for the photomask pattern design, test verification and correction of the chip design pattern.

[0005] In order to solve the above technical problems, the present invention provides a compensation model creation method for lithography exposure simulation, comprising:

[0006] According to the chip pattern sample, determining the corresponding test mask pattern and the mask pattern transmission parameter corresponding to the test mask pattern; and collecting the exposure image data set corresponding to the test mask pattern;

[0007] Constructing a neural network model, inputting the mask pattern transmission parameters into the neural network model, and obtaining the predicted compensation parameters output by the neural network model;

[0008] The predicted compensation parameters and the mask pattern transmission parameters are superimposed and input into a computational lithography model for exposure simulation to obtain a simulated exposure image;

[0009] Performing an error comparison between the simulated exposure image and a real exposure image in the exposure image data set to obtain a comparison error;

[0010] The neural network model is optimized according to the contrast error, and the step of inputting the mask pattern transmission parameters into the neural network model to obtain the predicted compensation parameters output by the neural network model is repeatedly executed according to the optimized neural network model, until the contrast error is no greater than the preset error, thereby obtaining a trained compensation neural network model.

[0011] In an optional embodiment of the present application, the process of determining the corresponding test mask pattern according to the chip pattern sample includes:

[0012] Extracting multiple groups of pattern features according to the chip pattern samples;

[0013] Performing similarity calculation between each group of pattern features, and classifying multiple groups of pattern features that meet the set similarity into the same category;

[0014] Calculate the average value of the pattern features of each group in each category to obtain feature measurement values ​​that respectively characterize the pattern features of each group in each category;

[0015] The test mask pattern is determined according to each group of the feature metric values.

[0016] In an optional embodiment of the present application, determining the test mask pattern according to each group of feature metric values ​​includes:

[0017] Gaussian white noise is added to each group of the characteristic metric values, and the test mask pattern is determined according to the characteristic metric values ​​after the Gaussian white noise is added.

[0018] In an optional embodiment of the present application, constructing a neural network model includes:

[0019] Construct a 5-layer CNN neural network model.

[0020] In an optional embodiment of the present application, collecting an exposure image data set corresponding to the test mask pattern includes:

[0021] Acquiring the exposure image data set including real exposure images at different focal depths under different illumination angles corresponding to the test mask pattern;

[0022] Accordingly, the predicted compensation parameters and the mask pattern transmission parameters are superimposed and input into a computational lithography model for exposure simulation to obtain a simulated exposure image, including:

[0023] The predicted compensation parameters and the mask pattern transmission parameters are superimposed and then input into a computational lithography model for exposure simulation to obtain simulated exposure images corresponding to different illumination angles and different focal depths.

[0024] In an optional embodiment of the present application, the predicted compensation parameter and the mask pattern transmission parameter are superimposed and then input into a computational lithography model for exposure simulation to obtain a simulated exposure image, including:

[0025] Superimposing the predicted compensation parameter and the mask pattern transmission parameter to obtain a mask transmission parameter;

[0026] The mask transmission parameters are input into the computational lithography model , obtaining a simulated photoresist photosensitivity representing the simulated exposure image; wherein, Pixel point on the exposure surface The simulated exposure intensity; Pixel point on the exposure surface The simulated photoresist sensitivity is is the optical kernel function, represents the effective truncated optical kernel function; represents the convolution operator; is the mask transmission parameter, ; is the position point on the test mask pattern The mask pattern transmission parameters, , is the position point on the test mask pattern The prediction compensation parameters of represents the photoresist response function.

[0027] A photolithography exposure simulation method, comprising:

[0028] Determining a corresponding mask pattern and a mask transmission parameter corresponding to the mask pattern according to a chip design pattern of the chip to be processed;

[0029] Inputting the mask transmission parameters into a pre-created compensation neural network model to obtain predicted compensation parameters output by the compensation neural network model; wherein the compensation neural network model is obtained by the compensation model creation method for lithography exposure simulation as described in any one of the above items;

[0030] The mask transmission parameters and the predicted compensation parameters are superimposed and input into a computational lithography model to obtain a simulated exposure image corresponding to the chip design pattern.

[0031] A compensation model creation device for photolithography exposure simulation, comprising:

[0032] A data acquisition module, used to determine the corresponding test mask pattern and the mask pattern transmission parameter corresponding to the test mask pattern according to the chip pattern sample; and to acquire an exposure image data set corresponding to the test mask pattern;

[0033] A model building module, used to build a neural network model, input the mask pattern transmission parameters into the neural network model, and obtain the predicted compensation parameters output by the neural network model;

[0034] An image simulation module, used for superimposing the predicted compensation parameters and the mask pattern transmission parameters and inputting them into a computational lithography model for exposure simulation to obtain a simulated exposure image;

[0035] An error comparison module, used for performing an error comparison between the simulated exposure image and a real exposure image in the exposure image data set to obtain a comparison error;

[0036] A model optimization module is used to optimize the neural network model according to the contrast error, and repeatedly execute the step of inputting the mask pattern transmission parameters into the neural network model according to the optimized neural network model to obtain the predicted compensation parameters output by the neural network model, until the contrast error is no greater than a preset error, thereby obtaining a trained compensation neural network model.

[0037] A photolithography exposure simulation device, comprising:

[0038] A parameter module, used to determine a corresponding mask pattern and a mask transmission parameter corresponding to the mask pattern according to a chip design pattern of a chip to be processed;

[0039] A compensation module, used for inputting the mask transmission parameters into a pre-created compensation neural network model to obtain predicted compensation parameters output by the compensation neural network model; wherein the compensation neural network model is obtained by the compensation model creation method for lithography exposure simulation as described in any one of the above items;

[0040] The simulation module is used to superimpose the mask transmission parameters and the predicted compensation parameters and input them into the computational lithography model to obtain a simulated exposure image corresponding to the chip design pattern.

[0041] An electronic device, comprising:

[0042] Memory for storing computer programs;

[0043] A processor is used to execute the computer program to implement the steps of the compensation model creation method for lithography exposure simulation as described in any one of the above items and / or the steps of the lithography exposure simulation method as described above.

[0044] The present invention provides a compensation model creation method and device for lithography exposure simulation, a lithography exposure simulation method and device, and an electronic device. The compensation model creation method includes: determining a corresponding test mask pattern and a mask pattern transmission parameter corresponding to the test mask pattern according to a chip pattern sample; and collecting an exposure image data set corresponding to the test mask pattern; constructing a neural network model, inputting the mask pattern transmission parameter into the neural network model, and obtaining predicted compensation parameters output by the neural network model; superimposing the predicted compensation parameters and the mask pattern transmission parameter and inputting them into a computational lithography model for exposure simulation to obtain a simulated exposure image; performing an error comparison between the simulated exposure image and a real exposure image in the exposure image data set to obtain a comparison error; optimizing the neural network model according to the comparison error, and repeatedly executing the step of inputting the mask pattern transmission parameter into the neural network model according to the optimized neural network model to obtain the predicted compensation parameters output by the neural network model, until the comparison error is no greater than a preset error, thereby obtaining a trained compensation neural network model.

[0045] In the present application, a constructed neural network model is used to predict the compensation signal to obtain predicted compensation parameters, and then a computational lithography model is used to simulate the lithography exposure process based on the predicted compensation parameters in combination with the mask pattern transmission parameters corresponding to the test mask pattern. The accuracy of the simulated exposure images obtained by the simulation is compared with the actual exposure images in the exposure image data set, which is equivalent to verifying the accuracy of the predicted compensation parameters predicted by the compensation neural network model. Based on this comparison error, the compensation neural network model is optimized, and finally a compensation neural network model that can predict accurate and reliable compensation parameters is obtained.

[0046] The training process of the compensation neural network model in the present application does not need to collect output sample data, which reduces the difficulty of data collection for model training, and then reduces the difficulty of model training, which also reduces the difficulty of determining the compensation signal in computational lithography; and in the model creation process, it is similar to using the computational lithography model as a discrimination model and the compensation neural network model as a generation model, so that an adversarial network is formed between the two, and the computational lithography model is used to implement the training of the compensation neural network model. The trained compensation neural network model can accurately predict the compensation parameters in the computational lithography model in practical applications, so that a more accurate compensation neural network model can be obtained and a more accurate computational lithography model can be obtained based on the compensation neural network model, ensuring that the computational lithography model simulates the lithography exposure process more accurately, and provides a reliable theoretical basis for the photomask pattern design, test verification and correction of the chip design pattern. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A schematic flow chart of a method for creating a compensation model for lithography exposure simulation provided in an embodiment of the present application;

[0049] Figure 2 A schematic diagram of transmission parameter coordinates corresponding to local position points on a test mask provided in an embodiment of the present application;

[0050] Figure 3 A schematic diagram of a process flow of a photolithography exposure simulation method provided in an embodiment of the present application;

[0051] Figure 4 A schematic diagram of the framework structure of a compensation model creation device for lithography exposure simulation provided in an embodiment of the present application;

[0052] Figure 5 A schematic diagram of the framework structure of a lithography exposure simulation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The core of the present invention is to provide a compensation model creation method and device for lithography exposure simulation, as well as a lithography exposure simulation method and device. Through training, a neural network model is created that can accurately predict and calculate the compensation parameters in the lithography model. The training process is simple and fast, and the accuracy and reliability of the compensation neural network model are guaranteed, which is conducive to improving the accuracy of exposure image simulation.

[0054] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0055] like Figure 1 As shown, Figure 1 A schematic flow chart of a method for creating a compensation model for lithography exposure simulation provided in an embodiment of the present application.

[0056] In a specific embodiment of the present application, the compensation model creation method for lithography exposure simulation may include:

[0057] S11: Determine a corresponding test mask pattern and a mask pattern transmission parameter corresponding to the test mask pattern according to the chip pattern sample.

[0058] It is understandable that in photolithography exposure technology, the chip pattern is the pattern that needs to be finally formed after photolithography exposure of the wafer or chip, etc.; and after the chip pattern has been determined, it is obvious that the mask pattern corresponding to the chip pattern can be determined accordingly.

[0059] In addition, the mask pattern transmission parameters in the present embodiment are equivalent to parameter data characterizing the test mask pattern on the test mask plate; specifically, the test mask plate obviously includes a light-blocking area and a hollow light-transmitting area (that is, a non-light-blocking area); therefore, if each position point on the test mask plate belongs to the light-blocking area, the mask pattern transmission parameter corresponding to the position point is 1, and when the position point belongs to the light-transmitting area, the mask pattern transmission parameter of the position point is 0.

[0060] On this basis, in an optional implementation of this embodiment, the process of determining the test mask pattern may include:

[0061] S111: extracting multiple groups of pattern features according to chip pattern samples;

[0062] S112: performing similarity calculation between each group of pattern features, and classifying multiple groups of pattern features that meet the set similarity into the same category;

[0063] S113: calculating the average value of each group of pattern features of each category to obtain each group of feature measurement values ​​that respectively represent each group of pattern features of each category;

[0064] S114: Determine a test mask pattern according to each set of feature metric values.

[0065] In the process of extracting multiple groups of pattern features based on the chip pattern samples in this embodiment, the pattern features may include features of multiple dimensions such as pattern shape, pattern size, pattern spacing, etc.; and the pattern features refer to the shape, size and spacing of the area that is not etched and removed on the chip pattern sample; wherein, the pattern shape feature may be characterized by the perimeter size and / or area size of the figure; and the pattern size may be characterized by the length dimension, width dimension, etc. of the area that is not etched and removed on the chip; and the pattern spacing is the spacing between two adjacent areas that are not etched and removed on the chip.

[0066] It should be noted that in order to improve the generalization ability of the model finally trained in this application, when selecting chip pattern samples, multiple chip patterns with different characteristics can be selected as samples to jointly extract pattern features, and then the test mask pattern determined by multiple chip pattern samples can be jointly determined based on the pattern features extracted from each different chip pattern. Therefore, on the one hand, the subsequent model training process based on the test mask pattern is not a single training for a single chip pattern sample, which can effectively avoid the problem of overfitting of the trained model; on the other hand, multiple chip pattern samples can be trained with only one test mask pattern, which simplifies the model training process to a certain extent.

[0067] On this basis, in order to further simplify the multiple groups of pattern features extracted according to the chip pattern samples, in this embodiment, each group of pattern features is first classified.

[0068] In practical applications, the process of classifying each group of pattern features according to the similarity between each group of pattern features may specifically include:

[0069] 1) Randomly select a set of pattern features as the initial center features.

[0070] 2) Substitute each group of pattern features into the classification formula In , each group of pattern features and initial center features that satisfy the above classification formula are classified into the same category, where is the initial center of the current classification, is the feature serial number of each different feature in each group of pattern features, is the number of features of each group of pattern features, is the nth eigenvalue of the initial central feature, is the nth feature value of the unclassified pattern features except the initial center feature; is a given threshold.

[0071] 3) After completing the classification of pattern features of one category, a new initial central feature can be randomly selected from each group of pattern features of the remaining unclassified categories, and the above step 2) can be repeated to classify each group of pattern features of the remaining unclassified categories, until all groups of pattern features have completed the classification.

[0072] It should be noted that the size of the given threshold can be set based on human experience; and if the number of categories divided into each group of pattern features is relatively small, the given threshold can be appropriately reduced, and then the categories of each group of pattern features can be reclassified, and when the number of categories divided into each group of pattern features is relatively large, the given threshold can be appropriately increased. Of course, in practical applications, the classification of each group of pattern features is not limited to the above implementation method. For example, clustering algorithms can also be used to automatically classify, and this application does not make specific restrictions on this.

[0073] In the above steps 1) to 3), the square of the Euclidean distance is used as the data to measure the similarity between two pattern features; in practical applications, it is not excluded to use other methods to calculate the similarity between two groups of pattern features; for example, Manhattan distance or distress similarity is used as a measure of the similarity between two groups of pattern features, and the classification between each group of pattern features can also be achieved, which is not specifically limited in this embodiment. In addition, in practical applications, this application does not exclude the direct use of clustering algorithms to achieve the classification between each group of pattern features, and this is not specifically limited in this application.

[0074] On the basis of realizing the division of each group of pattern features, the average value of each group of pattern features of the same class is further calculated, which is equivalent to re-determining the central feature of each group of pattern features of the same class by averaging, and using the central feature as the feature metric value characterizing the multiple groups of pattern features of this class, and finally determining the test mask pattern for subsequent model training with the feature metric value. That is to say, in this embodiment, one central feature represents multiple groups of pattern features in the same class, and subsequent model training is realized in this way, thereby greatly reducing the number of pattern features, that is, simplifying the pattern features, thereby reducing the difficulty of subsequent model training.

[0075] Furthermore, in the process of determining the test mask pattern based on the feature measurement values ​​of each group of pattern features, the relative distance between two adjacent light-blocking areas in the test mask pattern should be no less than the minimum visible distance to isolate the coupling effects of different features, thereby ensuring that the measurement data can directly respond to the features.

[0076] In addition, after determining the test mask pattern, the mask pattern transmission parameters corresponding to each test mask pattern can also be determined. However, it is further considered that the characteristic measurement value of the test mask pattern is determined based on multiple groups of pattern features and is incorporated into the real pattern features. Therefore, in order to further improve the authenticity of the mask pattern transmission parameters corresponding to the test mask pattern in the present application, in the process of determining the test mask pattern according to each group of characteristic measurement values, Gaussian white noise can also be added to each group of characteristic measurement values, that is, each characteristic value in each group of characteristic measurement values ​​is increased or decreased, so that each group of characteristic measurement values ​​after adding Gaussian white noise is closer to the real pattern features, and thus the test mask pattern determined based on each group of characteristic measurement values ​​after adding Gaussian white noise is more authentic and reliable.

[0077] S12: Acquire an exposure image data set corresponding to the test mask pattern.

[0078] It is understandable that when the test mask pattern is determined, the mask plate corresponding to the test mask pattern can be obtained; the light beam output by the light source is irradiated on the test mask plate, and the data set composed of the real exposure image formed after passing through the test mask plate.

[0079] In the process of actually acquiring the exposure image data set, the exposure image data set including real exposure images at different focal depths under different illumination angles corresponding to the test mask pattern may be fully acquired.

[0080] It should be noted that the real exposure image in the present application refers to the photosensitivity of the photoresist on each pixel on the chip surface when the light source is irradiated onto the chip surface through the test mask.

[0081] S13: Build a neural network model.

[0082] In practical applications, the neural network model in this application can be a 5-layer CNN network model.

[0083] S14: Inputting the mask pattern transmission parameters into the neural network model to obtain the predicted compensation parameters output by the neural network model.

[0084] As described above, the test mask pattern in this embodiment can be determined by the groups of feature measurement values ​​after adding Gaussian white noise as mentioned above. Therefore, the mask pattern transmission parameters of the test mask pattern are obviously corresponding to the transmission parameters of the test mask pattern determined by the pattern features after adding Gaussian white noise adjustment. Using this as input to the neural network model and training the neural network model is beneficial to improving the accuracy of the neural network model.

[0085] S15: superimposing the predicted compensation parameters and the mask pattern transmission parameters and inputting them into a computational lithography model for exposure simulation to obtain a simulated exposure image.

[0086] In practical applications, the computational lithography model used in this embodiment can be determined based on the simulation technology actually used.

[0087] In an optional embodiment of the present application, the computational lithography model can be expressed as: ;in, Pixel point on the exposure surface The simulated exposure intensity; Pixel point on the exposure surface The simulated photoresist sensitivity is is the optical kernel function, represents the effective truncated optical kernel function; represents the convolution operator; is the mask transmission parameter, represents the photoresist response function.

[0088] It can be understood that in practical applications, the plane where the test mask is located and the exposure surface of the chip can be regarded as two parallel planes, thereby establishing a plane rectangular coordinate system XY that is parallel to the two planes. The coordinate values ​​of each position point on the test mask in the rectangular coordinate system are expressed as The coordinates of each pixel point on the exposure plane in the rectangular coordinate system are represented by express.

[0089] also, That is, the mask transmission parameter obtained by superimposing the above-mentioned prediction compensation parameter and the mask pattern transmission parameter, ; For the test mask position point The mask pattern transmission parameters, and , For the test mask position point The prediction compensation parameters.

[0090] like Figure 2 As shown, Figure 2 The figure shows the corresponding positions of the local area points on the test mask plate. and The size of Figure 2 It can be seen that at each location There are only two values ​​0 and 1, and It varies between [0, 1) and is based on Figure 2It can be seen that the prediction compensation parameters in the present application are mainly used to compensate for the transmission response of the boundary area between the light-blocking area and the light-transmitting area in the test mask pattern.

[0091] After the mask transmission parameters obtained by superimposing the predicted compensation parameters and the mask image transmission parameters are substituted into the computational lithography model, the corresponding photosensitivity of each pixel on the chip can be simulated and calculated. However, in the actual photolithography exposure process, the photoresist needs to be exposed to light, so it is necessary to further analyze the photoresist response function. , converting the simulated photosensitivity of each pixel on the chip into simulated photoresist photosensitivity The simulated photoresist sensitivity intensity of each pixel on the chip can also represent the simulated exposure pattern.

[0092] S16: Perform an error comparison between the simulated exposure image and the real exposure image in the exposure image dataset to obtain a comparison error.

[0093] As described above, in the present application, after the predicted compensation parameters are obtained by using the neural network model for prediction, if the predicted compensation parameters can accurately characterize the diffraction effect during the mask exposure process, then after the predicted compensation parameters are superimposed on the mask pattern transmission parameters, the obtained mask transmission parameters can also accurately describe the actual transmission effect of the test mask pattern during the exposure process. The simulated exposure image thus obtained should be consistent with the actual exposure image substrate collected in S2 above. On the contrary, when the simulated exposure image and the actual exposure image are inconsistent, it means that the predicted compensation parameters predicted by the neural network model are inaccurate. Therefore, the audit network model can be further optimized and adjusted based on the difference between the simulated exposure image and the actual exposure image.

[0094] When comparing the errors between the simulated exposure image and the real exposure image, the cost function can be used to compare the errors between the simulated exposure image and the real exposure image. Specifically, the illumination intensity of each pixel point corresponding to the simulated exposure image and the real exposure image is introduced into the cost function. In , when the calculation result of the cost function is relatively large, it means that the difference between the simulated exposure image and the real exposure image is relatively large, and the predicted compensation parameters are inaccurate; is the illumination intensity of the pixel of the real exposure image.

[0095] S17: Determine whether the comparison error is greater than the preset error. If so, proceed to S18. If not, obtain a trained compensation neural network model.

[0096] S18: Optimize the neural network model according to the comparison error, obtain the optimized neural network model, and enter S14.

[0097] When it is determined based on the comparison error between the simulated exposure image and the real exposure image that the difference between the two is relatively large, the neural network model can be directionally optimized and improved based on the difference between the two, and after each iterative optimization of the neural network model, the predicted compensation parameters can be re-predicted, and the accuracy of the predicted compensation parameters can be re-judged by calculating the lithography model and the cost function until the real exposure image and the simulated exposure image are basically completely consistent. In this case, the predicted compensation parameters can be considered to be accurate parameters, which means that the compensation neural model used to predict the predicted compensation parameters is accurate and reliable. In subsequent practical applications, after accurate and reliable predicted compensation parameters are obtained based on the prediction of the compensation neural network model, the computational lithography model using the predicted compensation parameters is also more accurate and reliable, providing a reliable data basis for testing and correcting chip design patterns.

[0098] On this basis, the compensation neural network model in the present application is a model for predicting the compensation parameters in the computational lithography model. In order to achieve the optimization of the compensation neural network model in the present application, the exposure simulation function of the computational lithography model is used to judge the accuracy of the predicted compensation parameters predicted by the compensation neural network model, which is similar to using the compensation neural network model as a generation model and the computational lithography model as a discrimination model. An adversarial neural network is formed between the two. The computational lithography model is used to repeatedly judge the accuracy of the output value of the compensation neural network model to gradually achieve the optimization iteration of the compensation neural network model. The optimized and iterated compensation neural network model can in turn improve the accuracy of the computational lithography model, and ultimately improve the accuracy of the lithography exposure process simulation.

[0099] Furthermore, the method of training the compensatory neural network model in the present application does not require the pre-collection of output data samples, which greatly reduces the difficulty of collecting data samples in model training, that is, reduces the difficulty of model training.

[0100] To summarize, in the process of training the compensation neural network model in the present application, the computational lithography model is used to simulate the predicted compensation parameters obtained by the compensation neural network model in combination with the mask pattern transmission parameters corresponding to the test mask pattern, and the exposure image data set is used to perform an error comparison on the accuracy of the simulated exposure image obtained by the simulation, which is equivalent to verifying the accuracy of the predicted compensation parameters predicted by the compensation neural network model, and using this comparison error as a basis to optimize the compensation neural network model, and finally obtain a compensation neural network model that can predict accurate and reliable compensation parameters; in the present application, there is no need to collect output sample data, which reduces data collection for model training. difficulty; and in the model creation process, it is similar to using the computational lithography model as the discrimination model and the compensation neural network model as the generation model, so that an adversarial network is formed between the two, and the computational lithography model is used to realize the training of the compensation neural network model. The trained compensation neural network model can accurately predict the compensation parameters in the computational lithography model in practical applications, so as to obtain a more accurate compensation neural network model and a more accurate computational lithography model based on the compensation neural network model, thereby ensuring that the computational lithography model can more accurately simulate the lithography exposure process, and provide a reliable theoretical basis for the photomask pattern design, test verification and correction of the chip design pattern.

[0101] Based on any of the above embodiments, Figure 3 As shown, Figure 3 A schematic diagram of a process flow of a lithography exposure simulation method provided in an embodiment of the present application. The present application also provides an embodiment of a lithography exposure simulation method, the lithography exposure simulation method comprising:

[0102] S21: Determine a corresponding mask pattern and a mask transmission parameter corresponding to the mask pattern according to a chip design pattern of the chip to be processed.

[0103] S22: Inputting the mask transmission parameters into a pre-created compensation neural network model to obtain predicted compensation parameters output by the compensation neural network model; wherein the compensation neural network model is obtained by the compensation model creation method of lithography exposure simulation as described in any one of the above items;

[0104] S23: The mask transmission parameters and the predicted compensation parameters are superimposed and input into the computational lithography model to obtain a simulated exposure image corresponding to the chip design pattern.

[0105] The compensation neural network model obtained by the compensation model creation method based on the above-mentioned lithography exposure simulation in the present application can accurately predict the compensation parameters. When the calculation lithography model is used to simulate the lithography exposure process of the processed chip, the mask transmission parameters and compensation parameters corresponding to the processed chip are combined to achieve a more accurate simulation of the lithography exposure process, thereby obtaining a simulated exposure image that can truly represent the design pattern of the chip.

[0106] It can be understood that the compensation neural network model obtained by the above-mentioned embodiments in the present application is not necessarily able to realize the prediction of compensation parameters in the lithography exposure simulation of all chips to be processed; in practical applications, the chip design pattern can be feature extracted to obtain the pattern features of the chip design pattern. If the similarity between the pattern features and the pattern features of the chip pattern samples used in advance to train the compensation neural network model is not less than the similarity threshold, the compensation neural network model can be used to predict the compensation parameters. Conversely, if the similarity between the pattern features of the chip design pattern and the pattern features of the chip pattern samples used in advance to train the compensation neural network model is relatively small, the compensation neural network model needs to be recreated.

[0107] The compensation model creation device for lithography exposure simulation provided by an embodiment of the present invention is introduced below. The compensation model creation device for lithography exposure simulation described below and the compensation model creation method for lithography exposure simulation described above can refer to each other.

[0108] Figure 4 A structural block diagram of a compensation model creation device for lithography exposure simulation provided by an embodiment of the present invention, referring to Figure 4 The compensation model creation device for lithography exposure simulation may include:

[0109] The data acquisition module 110 is used to determine the corresponding test mask pattern and the mask pattern transmission parameter corresponding to the test mask pattern according to the chip pattern sample; and to acquire the exposure image data set corresponding to the test mask pattern;

[0110] A model building module 120, for building a neural network model, inputting the mask pattern transmission parameters into the neural network model, and obtaining the predicted compensation parameters output by the neural network model;

[0111] An image simulation module 130 is used to superimpose the predicted compensation parameters and the mask pattern transmission parameters and input them into a computational lithography model to perform exposure simulation to obtain a simulated exposure image;

[0112] An error comparison module 140 is used to perform an error comparison between the simulated exposure image and the real exposure image in the exposure image data set to obtain a comparison error;

[0113] The model optimization module 150 is used to optimize the neural network model according to the contrast error, and repeatedly execute the step of inputting the mask pattern transmission parameters into the neural network model according to the optimized neural network model to obtain the predicted compensation parameters output by the neural network model, until the contrast error is no greater than the preset error, thereby obtaining a trained compensation neural network model.

[0114] In an optional embodiment of the present application, the data acquisition module 100 is specifically used to extract multiple groups of pattern features based on the chip pattern samples; perform similarity calculations between the pattern features in each group, and classify the multiple groups of pattern features that meet the set similarity into the same category; calculate the average value between the pattern features in each group in each category to obtain feature measurement values ​​that respectively characterize the pattern features in each group in each category; and determine the test mask pattern based on the feature measurement values ​​of each group.

[0115] In an optional embodiment of the present application, the data acquisition module 110 is specifically configured to add Gaussian white noise to each group of the feature metric values, and determine the test mask pattern according to the feature metric values ​​after adding the Gaussian white noise.

[0116] In an optional embodiment of the present application, the model construction module 120 is specifically used to construct a 5-layer CNN neural network model.

[0117] In an optional embodiment of the present application, the data acquisition module 110 is specifically used to acquire the exposure image data set including real exposure images at different focal depths under different illumination angles corresponding to the test mask pattern;

[0118] Accordingly, the image simulation module 130 is specifically used to superimpose the predicted compensation parameters and the mask pattern transmission parameters and input them into the computational lithography model to perform exposure simulation, so as to obtain simulated exposure images corresponding to different illumination angles and different focal depths.

[0119] In an optional embodiment of the present application, the image simulation module 130 is specifically used to superimpose the predicted compensation parameters and the mask pattern transmission parameters to obtain the mask transmission parameters; input the mask transmission parameters into the computational lithography model , obtaining a simulated photoresist photosensitivity representing the simulated exposure image; wherein, Pixel point on the exposure surface The simulated exposure intensity; Pixel point on the exposure surface The simulated photoresist sensitivity is is the optical kernel function, represents the effective truncated optical kernel function; represents the convolution operator; is the mask transmission parameter, ; is the position point on the test mask pattern The mask pattern transmission parameters, , is the position point on the test mask pattern The prediction compensation parameters of represents the photoresist response function.

[0120] The present application also discloses an embodiment of a photolithography exposure simulation device, Figure 5 The structural block diagram of the lithography exposure simulation device provided by the embodiment of the present invention is shown in FIG. Figure 5 The photolithography exposure simulation device may include:

[0121] The parameter acquisition module 210 is used to determine the corresponding mask pattern and the mask transmission parameters corresponding to the mask pattern according to the chip design pattern of the chip to be processed;

[0122] A compensation operation module 220, used for inputting the mask transmission parameters into a pre-created compensation neural network model to obtain the predicted compensation parameters output by the compensation neural network model; wherein the compensation neural network model is obtained by the compensation model creation method of the lithography exposure simulation as described in any one of the above items;

[0123] The simulation operation module 230 is used to superimpose the mask transmission parameters and the predicted compensation parameters and input them into the computational lithography model to obtain a simulated exposure image corresponding to the chip design pattern.

[0124] The present application also discloses an embodiment of an electronic device, which may include:

[0125] Memory for storing computer programs;

[0126] A processor is used to execute a computer program to implement the steps of the compensation model creation method for lithography exposure simulation as described in any one of the above items and / or the steps of the lithography exposure simulation method as described above.

[0127] The memory may include random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0128] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "includes", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment comprising a series of elements are inherent to the elements. In the absence of more restrictions, the elements limited by the sentence "comprising one..." do not exclude the presence of other identical elements in the process, method, article or equipment comprising the elements. In addition, the above-mentioned technical solution provided in the embodiment of the present application is consistent with the corresponding technical solution in the prior art in principle, and the part is not described in detail, so as to avoid too much redundancy.

[0129] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core ideas of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the scope of protection of the present invention.

Claims

1. A method for creating a compensation model for lithography exposure simulation, characterized in that: include: Determining, according to the chip pattern sample, a corresponding test mask pattern and a mask pattern transmission parameter corresponding to the test mask pattern; and collecting an exposure image data set corresponding to the test mask pattern; Constructing a neural network model, inputting the mask pattern transmission parameters into the neural network model, and obtaining the predicted compensation parameters output by the neural network model; The predicted compensation parameters and the mask pattern transmission parameters are superimposed and then input into a computational lithography model for exposure simulation to obtain a simulated exposure image; Performing an error comparison between the simulated exposure image and a real exposure image in the exposure image data set to obtain a comparison error; The neural network model is optimized according to the contrast error, and the step of inputting the mask pattern transmission parameters into the neural network model to obtain the predicted compensation parameters output by the neural network model is repeatedly executed according to the optimized neural network model, until the contrast error is no greater than the preset error, thereby obtaining a trained compensation neural network model.

2. The method for creating a compensation model for lithography exposure simulation according to claim 1, wherein: The process of determining the corresponding test mask pattern according to the chip pattern sample includes: Extracting multiple groups of pattern features according to the chip pattern samples; Performing similarity calculation between each group of pattern features, and classifying multiple groups of pattern features that meet the set similarity into the same category; Calculate the average value of the pattern features of each group in each category to obtain feature measurement values ​​that respectively characterize the pattern features of each group in each category; The test mask pattern is determined according to each group of the feature metric values.

3. The method for creating a compensation model for lithography exposure simulation according to claim 2, wherein: Determining the test mask pattern according to each group of feature metric values ​​includes: Gaussian white noise is added to each group of the characteristic metric values, and the test mask pattern is determined according to the characteristic metric values ​​after the Gaussian white noise is added.

4. The method for creating a compensation model for lithography exposure simulation according to claim 1, wherein: Build a neural network model, including: Construct a 5-layer CNN neural network model.

5. The method for creating a compensation model for lithography exposure simulation according to claim 1, wherein: Acquiring an exposure image data set corresponding to the test mask pattern, including: Acquiring the exposure image data set including real exposure images at different focal depths under different illumination angles corresponding to the test mask pattern; Accordingly, the predicted compensation parameters and the mask pattern transmission parameters are superimposed and input into a computational lithography model for exposure simulation to obtain a simulated exposure image, including: The predicted compensation parameters and the mask pattern transmission parameters are superimposed and then input into a computational lithography model for exposure simulation to obtain simulated exposure images corresponding to different illumination angles and different focal depths.

6. The method for creating a compensation model for lithography exposure simulation according to any one of claims 1 to 5, characterized in that: The predicted compensation parameter and the mask pattern transmission parameter are superimposed and input into a computational lithography model for exposure simulation to obtain a simulated exposure image, including: Superimposing the predicted compensation parameter and the mask pattern transmission parameter to obtain a mask transmission parameter; The mask transmission parameters are input into the computational lithography model , obtaining a simulated photoresist photosensitivity representing the simulated exposure image; wherein, Pixel point on the exposure surface The simulated exposure intensity; Pixel point on the exposure surface The simulated photoresist sensitivity is is the optical kernel function, represents the effective truncated optical kernel function; represents the convolution operator; is the mask transmission parameter, ; is the position point on the test mask pattern The mask pattern transmission parameters, , is the position point on the test mask pattern The prediction compensation parameters of represents the photoresist response function.

7. A photolithography exposure simulation method, characterized in that: include: Determining a corresponding mask pattern and a mask transmission parameter corresponding to the mask pattern according to a chip design pattern of the chip to be processed; Inputting the mask transmission parameters into a pre-created compensation neural network model to obtain predicted compensation parameters output by the compensation neural network model; wherein the compensation neural network model is obtained by the compensation model creation method for lithography exposure simulation according to any one of claims 1 to 6; The mask transmission parameters and the predicted compensation parameters are superimposed and input into a computational lithography model to obtain a simulated exposure image corresponding to the chip design pattern.

8. A compensation model creation device for lithography exposure simulation, characterized in that: include: A data acquisition module, used to determine a corresponding test mask pattern and a mask pattern transmission parameter corresponding to the test mask pattern according to the chip pattern sample; and collecting an exposure image data set corresponding to the test mask pattern; A model building module, used to build a neural network model, input the mask pattern transmission parameters into the neural network model, and obtain the predicted compensation parameters output by the neural network model; An image simulation module, used for superimposing the predicted compensation parameters and the mask pattern transmission parameters and inputting them into a computational lithography model for exposure simulation to obtain a simulated exposure image; An error comparison module, used for performing an error comparison between the simulated exposure image and a real exposure image in the exposure image data set to obtain a comparison error; A model optimization module is used to optimize the neural network model according to the contrast error, and repeatedly execute the step of inputting the mask pattern transmission parameters into the neural network model according to the optimized neural network model to obtain the predicted compensation parameters output by the neural network model, until the contrast error is no greater than a preset error, thereby obtaining a trained compensation neural network model.

9. A photolithography exposure simulation device, characterized in that: include: A parameter module, used to determine a corresponding mask pattern and a mask transmission parameter corresponding to the mask pattern according to a chip design pattern of a chip to be processed; A compensation module, used for inputting the mask transmission parameter into a pre-created compensation neural network model to obtain a predicted compensation parameter output by the compensation neural network model; wherein the compensation neural network model is obtained by the compensation model creation method for lithography exposure simulation according to any one of claims 1 to 6; The simulation module is used to superimpose the mask transmission parameters and the predicted compensation parameters and input them into the computational lithography model to obtain a simulated exposure image corresponding to the chip design pattern.

10. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to execute the computer program to implement the steps of the compensation model creation method for lithography exposure simulation according to any one of claims 1 to 6 and / or the steps of the lithography exposure simulation method according to claim 7.

Images