A network training method for generating an EUV optimal mask

Abstract

The application belongs to the technical field of computing lithography, and specifically discloses a network training method for generating an EUV optimal mask, which applies DUV mask technology to the generation of an EUV mask, and guides the generator to correct in the training by the mask generated by a DUV mask generation model, so that the training is advanced in an optimized direction; a compensation part of EUV mask generation is added to the generator, so as to realize maximum optimization; key pattern screening technology is simultaneously applied to target image, EUV mask and DUV mask technology, so as to ensure the speed and effect of full-chip mask optimization in the training; and the method of the application generates the mask generated by the DUV mask generation model as additional information in the training under the framework of CGAN, adds an EUV compensation part on the basis of a general generator, and then passes through a discriminator, so as to train a model capable of generating an EUV optimal mask.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of computing lithography, and particularly relates to a network training method for generating an EUV optimal mask. BACKGROUND

[0002] Optical lithography is a key step in the manufacture of semiconductor integrated circuits, which is used to accurately transfer a microcircuit pattern onto a silicon wafer, involving light source irradiation, photosensitive resist reaction and etching process. With the progress of this technology, the technology node is constantly shrinking, and the market needs to use shorter wavelength EUV lithography machines for development, that is, to use better resolution enhancement technology to meet the growing demand for chip performance.

[0003] Mask optimization technology is one of the resolution enhancement technologies. The mask plate is a carrier that carries the information of the pattern design and process technology in the mask optimization technology, and its function is similar to that of the film of a traditional camera. Through exposure, the circuit pattern is transferred to the wafer of the downstream industry to realize mass production. The precision and quality level of the mask plate directly affect the yield of the final product, and is the key to connecting industrial design and process manufacturing, that is, to explore mask optimization technology is an important means to promote the development of integrated circuit manufacturing.

[0004] With the continuous shrinking of the key size of integrated circuits, moving towards 7nm and below, researchers' attention has shifted from deep ultraviolet (DUV) lithography technology to extreme ultraviolet (EUV) lithography technology, and mask optimization technology has also changed accordingly, gradually becoming a difficult part of integrated circuit manufacturing. The difficulties mainly lie in the following aspects: different mask structures and mask materials from DUV lithography technology, more complex imaging models and optimization algorithms, and lower mask defect tolerance. In view of these difficulties, the existing EUV mask optimization methods mainly include: studying the structure and material of the EUV mask absorption layer, reducing the three-dimensional mask effect in EUV lithography, and improving the EUV lithography imaging quality; further research on the imaging model, optimization algorithm, light source mask characterization and optimization strategy of the light source mask optimization (SMO); or in view of the defect sensitivity of the EUV mask, the multi-layer film defects of the EUV mask and the defect detection and compensation technology are optimized. However, the above methods all involve the consideration of complex factors in EUV lithography technology, and each step will produce huge cost, which needs to be further improved. SUMMARY

[0005] The present invention provides a network training method for generating an EUV optimal mask, aiming to directly train the network to generate an EUV optimal mask with the goal of reducing the difference in photoresist patterns corresponding to the generated mask. Within the framework of a conditional generative adversarial network (CGAN), the mask generated by a DUV mask generation model is used as additional information input during training, and an EUV compensation part is added to the general generator. After the method is passed through a discriminator, a model capable of generating an EUV optimal mask is trained.

[0006] To solve the above problems, the present invention provides the following technical solutions:

[0007] An embodiment of the present invention provides a network training method for generating an EUV optimal mask, comprising the following steps:

[0008] Step 1: Before formal training, data preprocessing and generator pretraining are performed. The generator is the most critical component in training and consists of two parts: the general generator G' and the compensator C. Among them, data preprocessing uses the algorithm based on breadth-first search in the key graph screening technology, including frequency extraction, frequency grouping and key graph screening, which is responsible for converting the obtained target image Z t And the mask m generated by the corresponding EUV and DUV mask generation models E *、m D * As the input of the CGAN network, a minimal set of key graphics is obtained to ensure speed and effectiveness during training;

[0009] Step 2: Training starts with m E * and Z t Composed mask pair (Z t ,m E *), by Z t and the mask G(Z) generated by the guidance generator to form a mask pair (Z t ,G(Z)); where (Z t ,m E *) Mask pair represents m E *The difference between the photoresist pattern under the mask and the target image; (Z t ,G(Z)) mask pair represents the difference between the photoresist pattern under the G(Z) mask and the target image;

[0010] Step 3, (Z t ,m E *) and (Z t ,G(Z)) two sets of mask pair data are passed into the discriminator, which is a common convolutional neural network structure used to perform convergence, predict the difference between the target image and the photoresist pattern under the mask, and continuously move the generated EUV mask G(Z) to the given m E *near;

[0011] Step 4: The discriminator outputs the result to determine whether the discriminator can identify m E * and G(Z), the judgment result determines the subsequent training process of the model; the step 4 includes: step 41: the judgment result is yes, indicating that G(Z) and Z in the training result t If there is a large gap, go to step 5; Step 42: If the judgment result is no, the EUV mask generated by the generator G(Z) meets the optimization result, and the training ends;

[0012] Step 5, judging whether the photoresist pattern Z under the G(Z) mask is closer to the target image, and the judgment result determines the subsequent training process of the model; said step 5 includes: step 51: if the judgment result is yes, it means m E * and G(Z) have some differences, but the photoresist pattern Z under the G(Z) mask generated by the generator is closer to the target image. Such G(Z) also meets the optimization result, and the training ends; Step 52: If the judgment result is no, the EUV mask generated by the generator G(Z) is unqualified, and jump to step 6;

[0013] Step 6: Send the instruction to the DUV mask generation model to guide it to correct the mask G'(Z) generated in the previous round, and then further optimize it by the compensator to generate a new round of G(Z) and compare it with Z t Form a new mask pair (Z t ,G(Z)), jump to step 3.

[0014] In an optional embodiment of the present invention, the specific steps of frequency extraction, frequency grouping, and key pattern screening in step 1 are as follows:

[0015] Step 11: Perform Fourier transform on the mask pattern to obtain the diffraction spectrum, and extract the peak frequency position of each diffraction peak. and projection boundaries, and the size of the frequency contour is controlled by the growth factor γ;

[0016] Step 12: Group the main frequencies to find the inclusion relationship between the mask patterns, so that the selected key patterns can represent the imaging quality of all mask patterns as much as possible. For example, if the diffraction peak F B The peak position In F A The outline of S A (S A =P A γ), then F B Belongs to F A Grouping;

[0017] Step 13: Use the breadth-first search graph screening method to complete the traversal and obtain the screening results of the key graphs.

[0018] In an optional embodiment of the present application, the specific steps of the pre-training in step 1 are as follows:

[0019] Step 14, small batch sampling is performed on the target image, and the gradient of the general generator is initialized to zero;

[0020] Step 15, the small batch image obtained in step 1 is input into the general generator to obtain a mask G'(Z), and the compensator C is an added part in the model generator, which mainly compensates the mask G'(Z) generated by the general generator, including the compensation of mask shadow effect, stray light effect and photoresist effect;

[0021] Step 16, the mask G'(Z) is transmitted into the ILT program to generate a corresponding photoresist pattern Z by using a lithography simulator;

[0022] Step 17: setting the evaluation lithography error E as the objective function, performing gradient calculation and parameter updating.

[0023] In an optional embodiment of the present application, in the first round of formal training, the generation steps of the two pairs of masks are as follows:

[0024] Step 21, the target image Z t Guidance, the general generator G' approaches the EUV mask pattern according to the generation method of the DUV mask, denoted as G'(Z);

[0025] Step 22, G'(Z) is transmitted into the compensator C for further optimization to generate the initial mask G(Z) in the training;

[0026] Step 23, m E * and Z t , G(Z) and Z t are respectively taken as a group to form respective loss functions, that is, a mask pair, denoted as (Z t , m E * ), (Z t , G(Z)).

[0027] In an optional embodiment of the present application, step 21 adopts a joint of loss function and optimization algorithm, and the joint of MSE and DICE parameters can consider regional loss while considering the loss of a single pixel point:

[0028] MSE is responsible for measuring the square of the error between each pixel point and the true pixel point, and the average is taken according to the overall image size; i n is the number of image pixels, y E is the photoresist pattern under the mask G(Z) generated by the generator, The photoresist pattern under the mask mE* generated by the EUV mask generation model; the DICE evaluates the similarity of two samples, that is, the two masks are overlapped to calculate the ratio of the overlapped part; the DICE loss is: The epsilon in formula (4) is a very small number, which is a smoothing coefficient.

[0029] The optimization algorithm selects the gradient descent algorithm, after the first round of formal training, if the discriminator can distinguish m E * and the photoresist pattern generated under G(Z), and the photoresist pattern Z under the mask G(Z) generated by the generator is not closer to the target image Z t , the network model transmits information from the output layer to the input layer into the next round of training, and the mask m D * generated by the existing EUV mask generation model is guided, and the parameters in the generator or compensator are updated according to the gradient descent process to minimize the loss function.

[0030] An optional embodiment of the present application further comprises, before step 1:

[0031] Step 101, setting initial light source parameters: determining the light source wavelength, light source type, light source power, conversion efficiency and light source generation method of EUV lithography;

[0032] Step 102, before formal training, the target image Z t in the training library is selected by using the key pattern screening technology, and then the selected Z t is transmitted into the EUV and DUV model to generate masks m E * and m D *.

[0033] Step 103, before formal training, the generator is pre-trained in combination with the ILT program, that is, the photoresist pattern Z corresponding to G'(Z) is generated by using the lithography simulator, and the evaluation model is used to compare it with the target image Z t , until the evaluation quality of the photoresist pattern reaches the standard, and the pre-training is completed.

[0034] Step 104, taking m E * generated by the existing EUV mask generation model as a data sample, taking m D * generated by the existing DUV mask generation model as random noise, and taking the target image Z t as a guide for the generator to generate G(Z).

[0035] The embodiment of the present application also provides an electronic device, which comprises:

[0036] At least one memory for storing a computer program;

[0037] At least one processor configured to execute a program stored in the memory, when the program stored in the memory is executed, the processor is configured to execute a network training method for generating an EUV optimal mask as in the above embodiment.

[0038] The embodiment of the present application also provides a computer readable storage medium, the computer readable storage medium stores a computer program, and when the computer program runs on a processor, the processor executes a network training method for generating an EUV optimal mask as in the above embodiment.

[0039] Beneficial effects: the method for generating an EUV optimal mask provided by the present application is a simpler and more efficient method, before training, the key pattern screening technology is used to concentrate the training part on a small number of patterns with all features, the DUV mask generation model is used for guidance, and the compensator is used for further optimization, so that the complex EUV mask generation model is avoided in each round of training, and the efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0041] Figure 1 A flowchart of a network training method for generating an EUV optimal mask provided by the embodiment of the present application

[0042] Figure 2 A CGAN training flowchart provided by the embodiment of the present application.

[0043] Figure 3 A pre-training schematic diagram of a generator under the ILT program provided by the embodiment of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0045] The method of the present invention aims to directly train a network to generate optimal EUV masks by directly minimizing the discrepancies between the photoresist patterns corresponding to the generated masks. The technical solution: Within the framework of a conditional generative adversarial network (CGAN), masks generated by a DUV mask generation model are used as additional information during training. An EUV compensation component is added to the general generator, and then passed through a discriminator to train a model capable of generating optimal EUV masks.

[0046] like Figure 1 As shown, an embodiment of the present invention provides a network training method for generating an EUV optimal mask, comprising the following steps:

[0047] Step 1: Before formal training, data preprocessing and generator pretraining are performed. The generator is the most critical component in training and consists of two parts: the general generator G' and the compensator C. Among them, data preprocessing uses the algorithm based on breadth-first search in the key graph screening technology, including frequency extraction, frequency grouping and key graph screening, which is responsible for converting the obtained target image Z t And the mask m generated by the corresponding EUV and DUV mask generation models E *、m D * As the input of the CGAN network, a minimal set of key graphics is obtained to ensure speed and effectiveness during training;

[0048] Step 2: Training starts with m E * and Z t Composed mask pair (Z t ,m E *), by Z t and the mask G(Z) generated by the guidance generator to form a mask pair (Z t ,G(Z)); where (Z t ,m E *) Mask pair represents m E *The difference between the photoresist pattern under the mask and the target image; (Z t ,G(Z)) mask pair represents the difference between the photoresist pattern under the G(Z) mask and the target image; assuming there is enough model capacity and training time to converge, the discriminator can compare the two sets of data to make G(Z)≈m after training. E * ;

[0049] Step 3, (Z t ,m E *) and (Z tThe two groups of masks and data are input into the discriminator, which is a general convolutional neural network structure, to perform convergence and predict the difference between the target image and the photoresist pattern under the mask, and the generated EUV mask G(Z) is continuously given to the m E * Close to;

[0050] Step 4, the discriminator outputs the result, and judges whether the discriminator can distinguish m E *and G(Z); step 4 includes: step 41: the discrimination result is yes, indicating that G(Z) in the training result and Z t There is a big gap, then go to step 5; step 42: the discrimination result is no, the EUV mask generated by the generator G(Z) meets the optimization result, and the training is completed;

[0051] Step 5, judge whether the photoresist pattern Z under the mask G(Z) is more close to the target image; the step 5 includes: step 51: the discrimination result is yes, indicating that m E *and G(Z) exist certain differences, but the photoresist pattern Z under the mask G(Z) generated by the generator is more close to the target image, and such G(Z) also meets the optimization result, and the training is completed; step 52: the discrimination result is no, then the EUV mask generated by the generator G(Z) is unqualified, and jump to step 6;

[0052] Step 6, send instructions to the DUV mask generation model to guide it to correct the mask G'(Z) generated in the last round, and further optimize it by the compensator to generate a new round of G(Z), and compare it with Z t To form a new mask pair (Z t ,G(Z)), jump to step 3.

[0053] Specifically, step 1, before formal training, data preprocessing and generator pre-training are performed. The generator part of the embodiment is composed of a general generator G' and a compensator C. Among them, the general generator is based on Abbe lithography space imaging model as the theoretical basis, mainly composed of an encoder and a decoder, which is a stacked convolutional architecture. Before formal training, we first pre-train it to reduce the possibility of being stuck in a local minimum region, and "guide" the generator to the optimum. After completing the pre-training, a small amount of target images are sampled, and then the encoder performs hierarchical layout feature abstraction, and the decoder operates in the opposite way to complete the mask correction relative to the target. In addition, the general generator G' is responsible for correcting the mask m D *The guiding of the ILT program generates a pattern close to the EUV mask and records it as G'(Z), which is transmitted to the compensator. The compensator is composed of a compensation model, including the compensation of mask shadow effect, stray light effect and photoresist effect. Using the lithography imaging model considering the three effects, the generator can further correct the mask G'(Z) generated by the general generator, which is equivalent to further optimization of the EUV mask, recorded as G(Z). The optimized mask G(Z) will be close to the target image Z t As a set of data to the discriminator.

[0054] Data preprocessing and generator pre-training: using the key pattern screening method (breadth-first search algorithm), the key pattern set representing the least number of target images Zt can be obtained, and mE* and mD* are generated according to the least key pattern set. The mask is generated, that is, a small number of patterns are used to represent the overall pattern structure characteristics of each, so as to ensure the speed and effect of full-chip mask optimization. The generator is the most critical component in the network model, and the generator in this model is composed of two parts of general generator and compensator. In order to train a satisfactory mask generation network model, the joint training algorithm guided by the ILT program is used to initialize the general generator part, so that it can better converge in the subsequent optimization process. Before training the mask generation for a certain target image, the generator is pre-trained once.

[0055] The specific steps of frequency extraction, frequency grouping and key pattern screening in step 1 are as follows:

[0056] Step 11, Fourier transform is performed on the mask pattern to obtain the diffraction spectrum, and the peak frequency position of each diffraction peak is extracted And the projection boundary, the size of the frequency profile is controlled by the growth factor γ;

[0057] Step 12, group the main frequencies to find the inclusion relationship between the mask patterns, so that the screened key patterns can represent the imaging quality of all mask patterns as much as possible. For example, if the peak position of the diffraction peak F B is in the profile S A of F A (S A = P A γ), then F B belongs to the F A group;

[0058] Step 13, the breadth-first search pattern screening method is used to complete the traversal to obtain the screening result of the key pattern.

[0059] The generator pre-training is performed before the formal training of the model, such as Figure 3 ​The pre-training of the generator is completed by the joint training algorithm guided by the generator of the model and the ILT program, and the pre-training object is the general generator part. The specific steps of the pre-training in step 1 are as follows:

[0060] Step 14, small batch sampling is performed on the target image, and the gradient of the general generator is initialized to zero;

[0061] Step 15, the small batch image obtained in step 1 is input into the general generator to obtain a mask G'(Z); the compensator C is an added part in the model generator, and the main function is to compensate the mask G'(Z) generated by the general generator, including the compensation of mask shadow effect, stray light effect and photoresist effect;

[0062] Step 16, the mask G'(Z) is transmitted into the ILT program to generate a corresponding photoresist pattern Z by using a lithography simulator;

[0063] Step 17, setting the evaluation lithography error E as the objective function, performing gradient calculation and parameter updating.

[0064] The lithography error is:

[0065] Siomoid function:

[0066] Incomplete binary mask:

[0067] Gradient representation:

[0068] Wherein, α and β represent the steepness factor, H represents the original convolution kernel, H * represents the conjugate matrix of H.

[0069] Specifically, the compensator is an added part in the model generator, and the main function is to compensate the mask G'(Z) generated by the general generator, including the compensation of mask shadow effect, stray light effect and photoresist effect. The specific method used is:

[0070] The compensation of mask shadow effect adopts a simplified mask shadow effect model, which is equivalent to introducing an edge error with a specific width in photoresist imaging, and the shadow width B and the incident angle α s of the form:

[0071]

[0072] Wherein, B max is the maximum shadow width, α sis the azimuth angle of the mask pattern edge to the x-axis of the annular sector exposure field of the EUV lithography system, W is the width of the annular sector exposure field, F is the opening angle of the annular sector exposure field, and x is the x-axis coordinate of the exposure field position where the edge is located.

[0073] The stray light effect is characterized by a stray light point spread function PSF f and a total integral scatter factor TIS:

[0074]

[0075] wherein, is the mask surface spatial coordinate, n f is the spectral index, r min is the boundary range between the low-frequency and high-frequency phase errors.

[0076] The TIS is obtained by integrating the scattered light in all possible directions.

[0077] The compensation of the photoresist effect adopts a simplified photoresist model of a point spread function, and a photoresist point spread function PSF r is used.

[0078]

[0079] wherein, σ PSF is the blurring effect of the photoresist on the final imaging.

[0080] After the above three compensation models, the mask G'(Z) generated by the general generator is further optimized and recorded as G(Z).

[0081] Before step 1, the following steps are further included:

[0082] Step 101, setting initial light source parameters: determining the light source wavelength, light source type, light source power, conversion efficiency and light source generation method of the EUV lithography;

[0083] Step 102, before formal training, the target image Z t in the training library is selected by using a key pattern screening technology, and the selected Z t is transmitted into the EUV and DUV models to generate masks m E * and m D *;

[0084] Step 103, before formal training, the generator is pre-trained in combination with the ILT program, that is, the photoresist pattern Z corresponding to G'(Z) is generated by using the lithography simulator, and the evaluation model is used to compare it with the target image Z t , until the evaluation quality of the photoresist pattern reaches the standard, and the pre-training is completed;

[0085] Step 104, m E * As data samples, m D * As random noise, target image Z t For guiding the generator to generate G(Z).

[0086] Step 2, training starts, by m E * and Z t Form a mask pair (Z t , m E * ), Z t and the mask G(Z) generated by its guide generator form a mask pair (Z, G(Z)).

[0087] Step 2 in the first round of formal training, the generation steps of the two pairs of masks are:

[0088] Step 21: guided by the target image Z t , the general generator G' approaches the EUV mask pattern according to the generation method of the DUV mask, denoted as G'(Z);

[0089] Step 22: pass G'(Z) into the compensator C for further optimization to generate the initial mask G(Z) in training;

[0090] Step 23: m E * and Z t , G(Z) and Z t are respectively taken as a group to form their own loss function, that is, a mask pair, denoted as (Z t , m E * ), (Z t , G(Z)).

[0091] Specifically, as a bridge connecting model prediction, training target and optimization algorithm, the loss function directly affects the learning effect and final performance of the model.

[0092] The target image Z t is expressed by the joint of mean square error MSE and DICE loss function:

[0093] MSE is responsible for measuring the square of the error between each pixel point and the true pixel point, and the average is taken according to the overall image size.

[0094] Where n is the number of image pixels, y i is the photoresist pattern under the mask G(Z) generated by the generator, The mask m generated by the model generated for the EUV mask E * The photoresist pattern under the EUV mask.

[0095] The DICE loss function is responsible for focusing on the edge cases of the mask, evaluating the similarity of two samples, that is, superimposing two masks to calculate the ratio of the superimposed part.

[0096] The DICE loss is:

[0097] Where ε is a very small number, which is a smoothing coefficient.

[0098] The above two parameters are combined to consider regional loss while considering the loss of individual pixels.

[0099] The optimization algorithm selects the gradient descent algorithm. After the first round of formal training, if the discriminator can distinguish m E * And the photoresist pattern under G(Z), and the photoresist pattern Z under the mask G(Z) generated by the generator is not closer to the target image Z t , the network model will pass the information from the output layer to the input layer into the next round of training, and the mask m D * Guided, update the parameters in the general generator according to the gradient descent process to minimize the loss function.

[0100] Step 3, (Z t ,m E * ) and (Z t ,G(Z)) two sets of mask pairs are input into the discriminator. Specifically, the discriminator in this model is a general convolutional neural network structure, mainly used to perform convergence, and to predict the difference between the target image and the photoresist pattern under the mask. The generated EUV mask G(Z) is constantly close to the given m E * .

[0101] Step 4, the discriminator outputs the result, and judges whether the discriminator can distinguish m E * And G(Z), and the judgment result determines the subsequent training process of the model, which is divided into two cases: step 41: the discrimination result is yes, which means that there is a large gap between G(Z) and Z t , then go to step 5; step 42: the discrimination result is no, the EUV mask generated by the generator G(Z) meets the optimization result, and the training is completed.

[0102] Specifically, the loss function of the model is m E* and G(Z) can be seen as trying to make the mask generated by the network model closer to the mask generated by the existing EUV mask generation model. The joint of the mean square error MSE and the DICE loss function is still used.

[0103] Step 5: Determine whether the photoresist pattern Z under the G(Z) mask is closer to the target image. The determination result determines the subsequent training process of the model, which is divided into two cases: Step 51: The determination result is yes, which means that although m E * and G(Z) exist certain differences, but the photoresist pattern Z under the G(Z) mask generated by the generator is closer to the target image, and such G(Z) also satisfies the optimization result, and the training is ended; Step 52: The determination result is no, and the EUV mask generated by the generator G(Z) is unqualified, and the process jumps to step 6.

[0104] Step 6: Send instructions to the DUV mask generation model, and jump to step 3; Step 6 is not the last step of the training, but a bridge to the next round of training, which is also a key part of the present application. Mainly, the DUV mask generation model corrects the mask G'(Z) generated in the last round, and further optimizes it by the compensator to generate a new G(Z), and compare it with Z t to form a new mask pair (Z t ,G(Z)).

[0105] The embodiment of the present application also provides an electronic device, which comprises: at least one memory for storing a computer program; and at least one processor for executing the program stored in the memory, wherein the processor is configured to execute the network training method for generating an optimal EUV mask according to the above embodiment when the program stored in the memory is executed.

[0106] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, wherein the computer program is configured to make the processor execute the network training method for generating an optimal EUV mask according to the above embodiment when the computer program is executed on the processor.

[0107] Embodiment 1

[0108] The embodiment provides a network training method for generating an optimal EUV mask, which comprises the following steps:

[0109] Step 1: In the data preprocessing, the algorithm based on the breadth-first search is used to complete the frequency extraction, frequency grouping and key pattern screening, so as to obtain Z t , m E * and m D * ;

[0110] Step 2, the generator of the model is pre-trained, mainly the general generator of the model is combined with the training algorithm of ILT program, and the gradient calculation and parameter update are carried out according to the evaluated lithography error;

[0111] Step 3, the preprocessed data m E * and Z t As a set of inputs, the mask pair (Z t , m E * ) is obtained, and Z t and the G(Z) generated by the generator are inputted into the compensation model to obtain the mask pair (Z, G(Z));

[0112] Step 4, the discriminator discriminates the mask pair, and obtains the loss function determined by m E * and G(Z);

[0113] Step 5, according to the flow chart of CGAN training in Figure 2 , the parameters in the generator are updated by using gradient descent method, and the loss function of the model is minimized;

[0114] Step 6, repeat the above steps until the training is completed.

[0115] In step 1, the algorithm based on breadth-first search groups the graph by extracting the peak frequency position and the projection boundary, and sets the growth factor, so as to find out the key graph to represent all.

[0116] In step 2, the pre-training of the generator mainly pre-trains the general generator, as shown in Figure 3 , the general generator and the joint training algorithm guided by ILT program are completed together.

[0117] The ILT program contains a lithography simulator and an evaluation model, and the lithography simulator is mainly based on Abbe imaging principle.

[0118] The square of the two norm of the difference is used as the objective function:

[0119]

[0120] The Sigmoid function and the incomplete binary function are used to describe the lithography development process:

[0121]

[0122] Where, α and β are steepness factors, I thThe threshold is generally 0.5 or the position of the maximum steep point. The greater the steepness factor, the more the function approximates the ideal resist function.

[0123] Through this procedure, the mask G'(Z) generated by the general generator is converted into the corresponding resist pattern Z.

[0124] In step 3, the formal training begins:

[0125] The target image Z is generated by the existing EUV mask generation model t The mask G'(Z) is generated by the existing EUV mask generation model t together with Z t form a mask pair (Z E , m * ), which is directly fed into the discriminator;

[0126] The Z t under the reference of the DUV mask generated by the pre-trained general generator is sent to the compensator part in the generator, and the DUV mask is sequentially compensated:

[0127] Simplified mask shadow effect:

[0128]

[0129] where B max is the maximum shadow width, a s ' is the azimuth angle of the mask pattern edge to the x-axis of the annular sector exposure field of the EUV lithography system, W is the width of the annular sector exposure field, F is the opening angle of the annular sector exposure field, and x is the x-axis coordinate of the exposure field position of the edge.

[0130] Stray light effect:

[0131]

[0132] where PSF f represents the stray light point spread function, TIS represents the total volume integral scattering factor, is the mask surface spatial coordinate, n f is the spectral index, r min is the boundary range between low-frequency and high-frequency phase errors.

[0133] Resist effect:

[0134] where PSF r represents the resist point spread function, and σ PSF is the blurring effect of the resist on the final imaging.

[0135] After the above three compensation models, the mask G'(Z) generated by the general generator is further optimized and recorded as G(Z).

[0136] In step 4, the mask pair (Z t ,m E * ) and (Z, G(Z)) are passed to the discriminator, which distinguishes m E * With G(Z), its loss function is expressed as the union of mean square error MSE and DICE loss function:

[0137]

[0138] Where n is the number of image pixels, y i Generate the photoresist pattern under the mask G(Z) for the generator, Mask m generated for EUV mask generation model E * The photoresist pattern below.

[0139] The DICE loss is:

[0140] Among them, ε is a very small number, which serves as a smoothing coefficient.

[0141] The optimization algorithm selects the gradient descent algorithm.

[0142] After the first round of formal training, if the discriminator can distinguish m E * and the photoresist pattern generated under G(Z), and the photoresist pattern Z under the mask G(Z) generated by the generator is not closer to the target image Z t , the network model passes the information from the output layer to the input layer to enter the next round of training, and the mask m generated by the existing DUV mask generation model D * Guided by the gradient descent process, the parameters in the general generator are updated to minimize the loss function.

[0143] In step 5, the discriminator outputs the result according to Figure 2 The training can be divided into the following situations:

[0144] Case 1: If the discriminator cannot distinguish m E * and G(Z), indicating that the photoresist pattern under the G(Z) mask is very close to the pattern given by the model, that is, the EUV mask generated by the generator is qualified and the training is completed;

[0145] Case 2: If the discriminator can distinguish m E *and G(Z), but m E * In contrast, the photoresist pattern under the G(Z) mask is closer to the target image Z t , indicating that the EUV mask generated by the generator is more excellent, and the training is completed.

[0146] Case 3: If the discriminator can distinguish m E * and G(Z), and m E * In contrast, the photoresist pattern under the G(Z) mask is not closer to the target image Z t , indicating that the EUV mask generated by the generator is insufficiently trained, at which point the information is fed back, the parameters in the generator are updated using the gradient descent method, and the loss function of the model is minimized. Until the training is completed.

[0147] In summary, although the present application has been disclosed as above with preferred embodiments, the above preferred embodiments are not intended to limit the present application, and those of ordinary skill in the art can make various modifications and embellishments without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application is defined by the scope of the claims.

Claims

1. A network training method for generating EUV optimal masks, characterized in that: The following steps are involved: Step 1: Before formal training, data preprocessing and generator pretraining are performed. The generator is the most critical component in training and consists of two parts: the general generator G' and the compensator C. Among them, data preprocessing uses the algorithm based on breadth-first search in the key graph screening technology, including frequency extraction, frequency grouping and key graph screening, which is responsible for converting the obtained target image Z t And the mask m generated by the corresponding EUV and DUV mask generation models E *、m D * As the input of the CGAN network, a minimal set of key graphics is obtained to ensure speed and effectiveness during training; Step 2, training starts with m E * and Z t Composed mask pair (Z t ,m E *), by Z t and the mask G(Z) generated by the guidance generator to form a mask pair (Z t ,G(Z)); where (Z t ,m E *) Mask pair represents m E *The difference between the photoresist pattern under the mask and the target image; (Z t ,G(Z)) mask pair represents the difference between the photoresist pattern under the G(Z) mask and the target image; Step 3, (Z t ,m E *) and (Z t ,G(Z)) two sets of mask pair data are passed into the discriminator, which is a common convolutional neural network structure used to perform convergence, predict the difference between the target image and the photoresist pattern under the mask, and continuously move the generated EUV mask G(Z) to the given m E * near; Step 4: The discriminator outputs the result to determine whether the discriminator can identify m E * and G(Z), the judgment result determines the subsequent training process of the model; the step 4 includes: step 41: the judgment result is yes, indicating that G(Z) and Z in the training result t If there is a large gap, go to step 5; Step 42: If the judgment result is no, the EUV mask generated by the generator G(Z) meets the optimization result, and the training ends; Step 5, judging whether the photoresist pattern Z under the G(Z) mask is closer to the target image, and the judgment result determines the subsequent training process of the model; said step 5 includes: step 51: if the judgment result is yes, it means m E * and G(Z) have some differences, but the photoresist pattern Z under the G(Z) mask generated by the generator is closer to the target image. Such G(Z) also meets the optimization result, and the training ends; Step 52: If the judgment result is no, the EUV mask generated by the generator G(Z) is unqualified, and jump to step 6; Step 6: Send the instruction to the DUV mask generation model to guide it to correct the mask G'(Z) generated in the previous round, and then further optimize it by the compensator to generate a new round of G(Z) and compare it with Z t Form a new mask pair (Z t ,G(Z)), jump to step 3.

2. A network training method for generating EUV optimal mask according to claim 1, characterized in that: The specific steps for frequency extraction, frequency grouping, and key graph screening in step 1 are as follows: Step 11: Perform Fourier transform on the mask pattern to obtain the diffraction spectrum, and extract the peak frequency position of each diffraction peak. and projection boundaries, and the size of the frequency contour is controlled by the growth factor γ; Step 12: Group the main frequencies to find the inclusion relationship between the mask patterns, so that the selected key patterns can represent the imaging quality of all mask patterns as much as possible. B The peak position In F A The outline of S A (S A =P A γ), then F B Belongs to F A Grouping; Step 13: Use the breadth-first search graph screening method to complete the traversal and obtain the screening results of the key graphs.

3. The network training method for generating EUV optimal mask according to claim 1, characterized in that: The specific steps of pre-training in step 1 are: Step 14: perform mini-batch sampling on the target image and initialize the gradient of the general generator to zero; In step 15, the mini-batch images obtained in step 1 are fed into the general generator to obtain the mask G'(Z). The compensator C is an additional component in the model generator. Its main function is to compensate for the mask G'(Z) generated by the general generator, including the mask shadow effect, stray light effect, and photoresist effect. Step 16, the mask G'(Z) is transferred to the ILT program, and the corresponding photoresist pattern Z is generated using a photolithography simulator; Step 17: Set the evaluated lithography error E as the objective function, perform gradient calculation and parameter update.

4. The network training method for generating EUV optimal mask according to claim 1, characterized in that: Step 2 In the first round of formal training, the steps for generating two mask pairs are: Step 21, the target image Z t Guidance, the general generator G' follows the DUV mask generation method to approach the EUV mask pattern, denoted as G'(Z); Step 22: G'(Z) is passed to the compensator C for further optimization to generate the initial mask G(Z) in training. Step 23, m E * and Z t , G(Z) and Z t As a group, they form their own loss function, which becomes a mask pair and is recorded as (Z t ,m E * )、(Z t ,G(Z)).

5. A network training method for generating EUV optimal mask according to claim 4, characterized in that: Step 21 is represented by a combination of loss function and optimization algorithm. The combination of MSE and DICE parameters can take into account the regional loss while taking into account the loss of individual pixels: MSE is responsible for measuring the square of the error between each pixel and the true pixel, and taking the average according to the overall image size; n is the number of image pixels, y i Generate the photoresist pattern under the mask G(Z) for the generator, Mask m generated for EUV mask generation model E *The photoresist pattern under; DICE evaluates the similarity of two samples, that is, the two masks are overlapped to calculate the ratio of the overlapping parts; DICE loss is: The ε in formula (4) is a very small number and serves as a smoothing coefficient; The optimization algorithm selects the gradient descent algorithm. After the first round of formal training, if the discriminator can distinguish m E * and the photoresist pattern generated under G(Z), and the photoresist pattern Z under the mask G(Z) generated by the generator is not closer to the target image Z t , the network model passes the information from the output layer to the input layer to enter the next round of training, and the mask m generated by the existing DUV mask generation model D *Guidance, updating the parameters in the generator or compensator according to the gradient descent process to minimize the loss function.

6. The network training method for generating EUV optimal mask according to claim 1, characterized in that: Before step 1 also include: Step 101, setting initial light source parameters: determining the light source wavelength, light source type, light source power, conversion efficiency, and light source generation method for EUV lithography; Step 102: Before formal training, use key image screening technology to filter the target image Z in the training library. t Make a selection, and then use the selected Z t Generate mask m for the target input into EUV and DUV models E *、m D *; Step 103: Before formal training, the generator is pre-trained in combination with the ILT program, that is, the photolithography simulator is used to generate the photoresist pattern Z corresponding to G'(Z), and the evaluation model is used to compare it with the target image Z. t Compare until the evaluation quality of the photoresist pattern reaches the standard and the pre-training is completed; Step 104: generate the EUV mask generated by the existing EUV mask generation model. E *As a data sample, the existing DUV mask generation model generates m D * as random noise, target image Z t Used to guide the generator to generate G(Z).

7. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the network training method for generating an EUV optimal mask according to any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program runs on a processor, the processor is enabled to execute the network training method for generating an EUV optimal mask according to any one of claims 1 to 6.

Images