Photoresist shrinkage prediction method and system, and electronic equipment

Abstract

The invention discloses a photoresist shrinkage prediction method which comprises the following steps: acquiring input data which comprises CD-SEM (Compact Disc-Scanning Electron Microscope) measurement conditions and original values of photoresist thickness and photoresist width measured by the CD-SEM; the obtained input data are input into a pre-trained prediction model, an output result of the prediction model is obtained, the prediction model is obtained through training based on a spider bee optimization algorithm and a support vector regression algorithm, and the output result is the shrinkage amount of the photoresist under CD-SEM measurement. According to the method disclosed by the invention, the photoresist shrinkage prediction model is constructed by utilizing support vector regression (SVR), and parameter optimization is carried out on the support vector regression model through a spider bee optimization algorithm, so that the regression performance of the SVR is improved, and the prediction model generated by training can achieve a more accurate prediction result in the aspect of predicting the photoresist shrinkage value; this helps to improve the metering accuracy of the CD-SEM.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to a photoresist shrinkage amount prediction method and system, an electronic device and a storage medium. BACKGROUND

[0002] With the development of semiconductor manufacturing technology, the feature size of integrated circuits is continuously reduced, and the control accuracy of the photolithography process is increasingly improved. The critical dimension scanning electron microscope (CD-SEM) is an important measurement tool in the photolithography process. However, with the reduction of the feature size, the measurement accuracy of the CD-SEM is challenged by the shrinkage of the photoresist under electron beam irradiation. Existing research shows that the photoresist shrinkage is closely related to factors such as the electron beam dose, the acceleration voltage, the chemical formula of the photoresist, and the feature size structure, as shown in FIG. 1. The photoresist with a width of CD before electron beam irradiation is changed to CD' after the irradiation of the CD-SEM electron beam, resulting in a shrinkage of ΔCD (as shown in the gray area on the right side of FIG. 1). Figure 1 Figure 1

[0003] ​​To mitigate the impact of resist shrinkage caused by electron beam irradiation on CD-SEM measurement accuracy, two mainstream solutions have been proposed in the existing technology. One is a method based on mathematical models. For example, Bunday et al. proposed 1-D and 2-D resist shrinkage models to fit actual shrinkage data. Rana et al. proposed applying artificial neural networks (ANNs) to predict resist shrinkage and predicting the unshrinkage value CD through supervised learning. Ohashi et al. proposed a resist shrinkage correction model based on electron beam energy to restore the original unshrinkage profile from the two-dimensional pattern profile extracted from CD-SEM images. Li et al. proposed an across sampling method and combined it with a shrinkage model to infer CD differences. Sugie et al. proposed improvement measures by improving the scanning strategy and frame integration algorithm, and experimentally verified their application effect on EUV photoresist. Wei et al. verified the effectiveness of the profile averaging method in alleviating pattern shrinkage during scanning electron microscopy (SEM) imaging. Another approach is to use a lower accelerating voltage or fewer scanning frames to acquire SEM images to minimize the shrinkage caused by electron beam irradiation. However, the correction efficiency and accuracy of the former method still need to be improved, while the latter method's operation inevitably introduces image noise, reducing measurement accuracy. Summary of the Invention

[0004] The embodiments of the present invention provide a new solution to solve the problem of CD-SEM metrology accuracy caused by the photoresist shrinkage effect in the prior art.

[0005] According to a first aspect of the present invention, a method for predicting photoresist shrinkage is provided, comprising: Acquiring input data, wherein the input data includes CD-SEM measurement conditions, and original values ​​of photoresist thickness and photoresist width measured by CD-SEM; The obtained input data is input into a pre-trained prediction model, and an output result of the prediction model is obtained, wherein the prediction model is trained based on an improved spider bee optimization algorithm and a support vector regression algorithm, and the output result is the shrinkage amount caused by the photoresist under CD-SEM measurement.

[0006] According to a second aspect of the present invention, a photoresist shrinkage prediction system is provided, comprising: The shrinkage prediction module is configured to input data obtained from a critical dimension scanning electron microscope into a pre-trained prediction model, and obtain an output result of the prediction model, wherein the prediction model is trained based on an improved spider optimization algorithm and a support vector regression algorithm, the output result is a shrinkage of photoresist caused by CD-SEM measurement, and the input data includes CD-SEM measurement conditions, photoresist thickness and photoresist width original values measured by CD-SEM.

[0007] According to a third aspect of the present application, an electronic device is provided, comprising at least one processor, and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method of the first aspect.

[0008] According to a fourth aspect of the present application, a storage medium is provided, which stores a computer program executable by a processor to implement the steps of the method of the first aspect.

[0009] The method of the embodiments of the present application uses support vector regression (SVR) to construct a photoresist shrinkage prediction model, and optimizes parameters of the support vector regression (SVR) model through an improved spider optimization algorithm, so as to improve the regression performance of SVR, and thus the prediction model trained can achieve more accurate prediction results in predicting photoresist shrinkage values, which helps to improve the measurement accuracy of CD-SEM. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0011] Figure 1 A schematic diagram of the shrinkage effect of photoresist caused by CD-SEM measurement in the prior art is shown; Figure 2 A flowchart of a photoresist shrinkage prediction method according to an embodiment of the present application is shown schematically; Figure 3 A schematic diagram of the principle of SVR loss calculation in the prior art is shown; Figure 4 A flowchart of a method for training a prediction model according to an embodiment of the present application is shown schematically; Figure 5A flow chart of a method for training an initial model based on SVR using an improved spider optimization algorithm is shown schematically in an embodiment of the present application; Figure 6 A comparative effect diagram of the MSE convergence curves of ISWO and other six optimization algorithms for training support vector regression models is shown schematically; Figure 7 The optimal hyperparameter combinations obtained by ISWO and other six optimization algorithms and their prediction performance indicators on the validation set are shown schematically; Figure 8 A radar comparison diagram based on four indicators of ISWO and other six optimization algorithms is shown schematically; Figure 9 A scatter plot showing the predicted values and actual values of the photoresist shrinkage amount by the ISWO-SVR model is shown schematically; Figure 10 A schematic diagram of the architecture of a photoresist shrinkage prediction system is shown schematically in an embodiment of the present application; Figure 11 A schematic diagram of the hardware structure of an electronic device is shown schematically in an embodiment of the present application; Figure 12 A schematic diagram of the hardware structure of an electronic device is shown schematically in another embodiment of the present application. DETAILED DESCRIPTION

[0012] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0013] It should be noted that the embodiments and features in the present application can be combined with each other without conflict.

[0014] In the description of the application, it needs to be understood that if the terms "center", "middle", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "axial", "radial", "circumferential" and the like indicate the orientation or positional relationship shown in the drawings, which is only for the convenience of describing the application and simplifying the description, and does not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application. The features defined as "first", "second" are used to distinguish the feature names, not to have special meanings, and in addition, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the application, unless otherwise stated, the meaning of "a plurality of" is two or more.

[0015] In the description of the application, it needs to be explained that unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the application can be understood according to the specific circumstances.

[0016] It also needs to be explained that in this paper, the terms "including", "containing" not only include those elements, but also include other elements not explicitly listed, or include elements inherent to such processes, methods, articles or devices. Without more limitation, the elements defined by the sentence "including" do not exclude the presence of other identical elements in the process, method, article or device including the elements. The terms used in this paper are generally the terms commonly used by those skilled in the art, and if they are inconsistent with commonly used terms, the terms in this paper shall prevail.

[0017] In order to make the purpose, technical scheme and advantages of the embodiments of the application more clear, the technical scheme in the embodiments of the application will be described clearly and completely in the following with reference to the drawings of the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0018] The application will be described in further detail below with reference to the drawings.

[0019] The photoresist shrinkage prediction method in the embodiment of the application can be applied in any mobile terminal device, and can also be applied on a server and a mobile electronic device, so that the device or server or terminal can use the scheme of the embodiment of the application to predict the shrinkage of the photoresist caused by CD-SEM measurement, and the prediction accuracy is high, which helps to improve the measurement accuracy of CD-SEM.

[0020] Figure 2 The flow of the photoresist shrinkage prediction method of an embodiment of the application is schematically shown, and the execution subject of the method can be a processor of a computer, a cloud server, a server in a cluster, a mobile terminal device, etc., which is not limited in the embodiment of the application. As shown in Figure 2 , it includes: Operation S1, obtaining input data, wherein the input data includes CD-SEM measurement conditions, CD-SEM measured photoresist thickness and photoresist width original values; Operation S2, inputting the obtained input data into a pre-trained prediction model, and obtaining an output result of the prediction model, wherein the prediction model is trained based on a spider optimization algorithm and a support vector regression algorithm, and the output result is the shrinkage of the photoresist caused by CD-SEM measurement.

[0021] Preferably, the CD-SEM measurement conditions can be the acceleration voltage and exposure time set by the corresponding CD-SEM device when performing measurement, and the photoresist thickness and photoresist width original values are the measurement results output by the corresponding CD-SEM device. In operation S1, the input data can be obtained from the CD-SEM device whose measurement result accuracy needs to be corrected. By inputting the measurement result and measurement condition of a CD-SEM device into the trained prediction model, the shrinkage of the photoresist caused by the measurement of the CD-SEM device can be predicted, and based on the predicted shrinkage, the measurement result output by the CD-SEM device can be corrected, thereby avoiding the measurement error caused by the shrinkage of the photoresist and improving the measurement accuracy of the CD-SEM.

[0022] In operation S2, the prediction model is trained using the spider optimization algorithm and the support vector regression algorithm. The support vector regression algorithm (SVR) is proposed by Vladimir Vapnik et al. based on the support vector machine (SVM) algorithm, which is used to solve the regression problem. Unlike traditional regression algorithms, SVR does not use the method of directly calculating the difference between the true value and the predicted value to calculate the loss, but sets a hyperparameter tolerance deviation ε to calculate the loss, as shown in Figure 3 , only the loss of the sample points outside the ε-interval is calculated. The algorithm process of SVR is as follows: Suppose the training sample set is , The purpose of SVR is to obtain a regression model as formula (1), so that the real value and the predicted value can be as close as possible.

[0023] (1) Where w is the weight vector, x is the input feature vector, and b is the bias term.

[0024] Therefore, the problem solved by the SVR algorithm can be converted into a minimum value solving problem as shown in formula (2): (2) Where C in formula (2) is a penalty factor, which can be used to control the balance between the regularization term and the error term, and is an important hyperparameter in the SVR algorithm; is an ε-insensitive loss function, the expression is as shown in formula (3): (3) That is, when the sample point is within the ε-interval band, the error between the real value and the predicted value is not calculated; when the sample point falls outside the ε-interval band, the error between the real value and the predicted value that exceeds the interval band is calculated.

[0025] In order to more conveniently solve the above minimum value solving problem, the slack variable and are introduced, and the above problem is further converted into: (4)

[0026] SVR solves the above optimization problem through the Lagrange function expressed by formula (5) (5) Where , , , are Lagrange multipliers (λi, λi*, λi**, λi***). , , ,

[0027] Solve formula (5) for , , ,​ The partial derivative of the function is obtained, and the partial derivative is set to 0 to obtain: (6) (7) (8) (9) The dual problem of SVR is obtained by substituting formula (6), (7), (8), (9) into formula (5): (10) The solution of the SVR problem is obtained by substituting formula (6) into formula (1): (11) In order to solve the regression problem of SVR for nonlinear data, in the embodiment of the present application, x is preferably mapped into a high-dimensional feature vector Then formula (6) can be written as: (12) At this time, the solution of SVR can be written as: (13) Wherein is a kernel function, preferably, the kernel function selected by the present application is a Gaussian radial basis function, which is defined as follows: (14) Wherein σ is the bandwidth parameter of the Gaussian radial basis function, and is the hyperparameter of the kernel function.

[0028] Spider Wasp Optimizer (SWO) is a meta-heuristic intelligent optimization algorithm based on the behavior of spider wasps in nature. By simulating the hunting, nesting and mating behaviors of spider wasps, the optimal solution in the optimization problem is found, and the algorithm has the characteristics of fast search speed and high solution accuracy. The algorithm process of the algorithm includes population initialization, hunting and nesting behavior, tracking and escape stage, nesting stage, mating behavior, and population reduction and memory storage. The algorithm process of the spider wasp optimization algorithm will be briefly described below.

[0029] In the population initialization stage, the spider wasp optimization algorithm uses a random initialization method to initialize the female spider wasp population, as shown in formula (15).

[0030] (15) Wherein represents the solution corresponding to the ith female spider wasp of the tth generation, and The upper limit of the search space is The lower bound of the search space, is a random vector.

[0031] The hunting and nesting behavior of the algorithm includes a search phase and a tracking and escape phase. In the search phase, the female spider bee has two ways to search for prey: the first way is to update the position by a fixed step to search for prey; the second way is to search around the position where the spider falls. After searching for prey by the two ways, the position of the female spider bee is updated using equations (16) and (18), respectively.

[0032] (16) wherein and are two random female spider bees in the t th generation population for determining the search direction, for determining the step length along the direction, and the calculation formula is: (17) wherein is a random number between 0 and 1, is a random number generated by a normal distribution.

[0033] (18) (19) (20) wherein is a random female spider bee in the t th generation population, and l is a random number between 1 and 2.

[0034] The SWO algorithm determines which search method to use to update the position of the female spider bee by equation (21).

[0035] (21) wherein and are both random numbers between 0 and 1.

[0036] When the female spider bee finds prey, it will attack the prey, but the spider will fall off the web and escape, and then there are two trends of the spider bee tracking the prey and the prey successfully escaping.

[0037] When the spider bee tracks the prey, equation (22) is used to simulate the situation of the female spider bee chasing the spider to catch them.

[0038] (22) (23) where C is a distance control factor for determining the speed of the wasp, is a random vector between [0, 1], is a random number between [0, 1], t is the current iteration number, t max is the maximum iteration number, is the tth generation of the wasp randomly selected.

[0039] When the prey successfully escapes, the trend that the distance between the wasp and the spider increases with the iteration number is simulated by equation (24).

[0040] (24) (25) where is a vector generated according to the normal distribution between -k and k, and k is calculated by equation (25).

[0041] The above two trends are balanced by equation (26).

[0042] (26) In the initial stage of the optimization process, the female wasp mainly applies the search mechanism, and with the increase of the iteration number of the optimization algorithm, the female wasp mainly applies the following and escaping mechanisms, and the transition of the two mechanisms is completed by equation (27).

[0043] (27) where p is a random number between [0, 1].

[0044] In the nest building stage, SWO selects and simulates two nest building ways of the wasp, the first way is to drag the prey into the area containing the best spider to build a nest, and the second way is to randomly select the position of a spider to build a nest. The two nest building ways can be described by equations (28) and (29).

[0045] (28) (29) (30) where is the current optimal solution, is a random number in the interval [0, 1], is a number generated according to Levy flight, is a binary vector for determining whether to apply the step size to avoid building two nests in the same position, and two vectors of random values in the interval [0, 1], , and three solutions randomly selected from the population.

[0046] The nest building method is determined by the wasp optimization algorithm through equation (31).

[0047] (31) During the optimization process of the algorithm, hunting behavior and nest building behavior are balanced through equation (32): (32) The wasp optimization algorithm generates potential solutions by simulating the mating behavior of female and male spiders, as shown in equation (33): (33) Where Crossover is the uniform crossover operator, and CR is the crossover rate. is a female wasp, is a male wasp, and the male wasp is generated by equation: (34) Where and are two numbers randomly generated according to the normal distribution, and e is the exponential constant, and are calculated according to the following formula: (35) (36) Where is the fitness function, , , is a random solution in the population.

[0048] During the iteration process, the number of female wasps is reduced through equation (37) to speed up the convergence speed of the algorithm.

[0049] (37) Where, is the minimum population size set in the algorithm.

[0050] The solution obtained by each wasp will be compared with the previous optimal solution, and if the new solution is better, the current solution will be replaced with the new solution.

[0051] In order to apply the above SVR algorithm and SWO algorithm to the embodiments of the present application to solve the problem of CD-SEM metrology accuracy caused by the photoresist shrinkage effect, the embodiments of the present application propose to apply the SVR algorithm to the photoresin shrinkage prediction task. As a possible implementation, the SVR algorithm can be used to construct a regression model for predicting the shrinkage amount of the photoresist caused by the CD-SEM measurement, and an optimization problem with the SVR model prediction accuracy as the target is constructed, and the optimization problem is solved by using the spider wasp optimization algorithm to obtain the optimal solution, so as to improve the regression performance and accuracy of the trained prediction model. As a preferred implementation, the spider wasp optimization algorithm of the embodiments of the present application is an improved spider wasp optimization algorithm, which is based on a good point set for population initialization. The good point set (GPS) is a method for optimizing the initial distribution, which can generate a point set with good coverage and high uniformity under limited resource constraints. Assuming is a unit cube in s-dimensional Euclidean space, and , then the good point set can be obtained by formula (41): (41) where the good point ; P is the smallest prime number satisfying ; and is the decimal part of .

[0052] , and the deviation of the good point set needs to satisfy formula (42) (42) where , is a constant related to r and .

[0053] Therefore, the initialization population strategy based on the good point set, i.e. formula (15) in the above SWO algorithm, can be represented by formula (43).

[0054] (43) Therefore, during the preset number of iteration solving process by using the spider wasp optimization algorithm, the population initialization can be performed based on the good point set.

[0055] More preferably, the improved spider optimization algorithm is population reduction using a cyclic population reduction technique during the iteration process. The cyclic population reduction technique is a population reduction strategy proposed in the Crown-Hoar optimization algorithm. This strategy can not only speed up the optimization process, but also ensure that the population has enough change and exploration ability during the optimization process, which can better avoid falling into local minimum and find the global optimal solution. CPR determines the number of times this process is performed during the optimization process based on the cycle variable T, as shown in equation (38). The mathematical model of the cyclic population reduction population size is as follows, which can be used to replace the population reduction algorithm of equation (37) in the above SWO algorithm process: (38) Where T is a variable for determining the number of cycles, t is the current function evaluation number, is the maximum function evaluation number, is the minimum number of individuals in the newly generated population, so that the population size cannot be less than .

[0056] More preferably, the trade-off rate is a super parameter that significantly affects the optimization ability of the spider optimization algorithm. In SWO, manually adjusting the trade-off rate limits the effective use of the algorithm. To solve this problem, the improved spider optimization algorithm of the present embodiment proposes an adaptive trade-off rate (TR) strategy for dynamically adjusting the hunting and nesting behavior and mating behavior in the spider optimization algorithm. Specifically, the adaptive trade-off rate strategy can be implemented as follows: if the best fitness of the current iteration is improved by more than a preset threshold compared to the best fitness of the last iteration , then the local development ability of the algorithm is strengthened by reducing the TR value; otherwise, if the improvement does not reach the threshold, the global search ability is enhanced by increasing the TR value.

[0057] The hunting behavior and nesting behavior of the improved spider optimization algorithm during the optimization process of the algorithm are balanced by the following formula: (39) Wherein, is the trade-off rate in the tth iteration, the adjustment range of the value is limited between 0.1 and 0.9, t is the current iteration number, is a preset threshold, is an adjustment step, is the best fitness improvement rate, and its calculation formula is as follows: (40) Where represents the best fitness in the tth iteration, represents the best fitness in the (t-1)th iteration.

[0058] wherein, Figure 4 A method for training a prediction model is shown in Fig. 1, which comprises the following steps: Figure 4 In operation S21, an initial model based on support vector regression (SVR) is generated by regression modeling the shrinkage of photoresist under CD-SEM measurement based on CD-SEM measurement conditions, photoresist thickness and photoresist width raw values measured by CD-SEM. In operation S22, the initial model is trained based on the improved wasp optimization algorithm, the training sample set and the verification set to generate the prediction model.

[0059] As a possible implementation, in operation S21, when regression modeling the shrinkage of photoresist under CD-SEM measurement, the present application is based on the nonlinear mapping relationship between CD-SEM measurement conditions, photoresist thickness and photoresist width raw values measured by CD-SEM and the shrinkage of photoresist to construct the initial model of SVR. Specifically, the initial model is constructed with the shrinkage of photoresist as the predicted value f(x), and the CD-SEM measurement conditions, photoresist thickness and photoresist width values as the input. The CD-SEM measurement conditions can be the acceleration voltage and exposure time set by CD-SEM during measurement. The photoresist thickness and photoresist width raw values are the measurement results obtained based on the corresponding acceleration voltage and exposure time. Based on this, the regression model constructed by the present application for predicting the shrinkage of photoresist under SEM measurement can be abstracted as:

[0060] wherein, f(x) is the predicted shrinkage value of photoresist, and x is a vector including the acceleration voltage set by CD-SEM, the exposure time of CD-SEM, the photoresist thickness and photoresist width raw values measured by CD-SEM.

[0061] ​According to the foregoing SVR algorithm process, in order to obtain a final model that can be used to predict the shrinkage of photoresist caused by CD-SEM measurement, the training of the foregoing regression model actually converts into a problem of solving the hyperparameters in the model, and if the optimal solution of the hyperparameters is found, the accuracy of the SVR model in solving the photoresist shrinkage prediction value problem can be effectively improved. In combination with the foregoing SVR algorithm process, the hyperparameters that need to be solved optimally and are determined by the embodiments of the present application include a penalty factor C, a hyperparameter σ of a kernel function, and a tolerance deviation width ε of an insensitive loss function. As a preferred embodiment, the embodiments of the present application use an improved spider wasp optimization algorithm to optimize and solve these hyperparameters (referred to as to-be-optimized parameters by the embodiments of the present application) of the initial model by using a training sample set and a validation set, so as to improve the accuracy of the SVR model in solving the photoresist shrinkage prediction value problem. The training sample set includes at least one group of training data, each group of training data includes an acceleration voltage of a CD-SEM, an exposure time of the CD-SEM, a measured photoresist thickness and a photoresist width original value of the CD-SEM, and a photoresist shrinkage value, wherein the photoresist shrinkage value in each group of training data is the actual shrinkage value corresponding to the measured photoresist thickness and the photoresist width original value under the corresponding CD-SEM acceleration voltage and exposure time. The validation set of the embodiments of the present application includes at least one group of validation data, each group of validation data includes an acceleration voltage of a CD-SEM, an exposure time of the CD-SEM, a measured photoresist thickness and a photoresist width original value of the CD-SEM, and a photoresist shrinkage value. Wherein, after obtaining the experimental data, the experimental data can be divided into the training sample set and the validation set by dividing the experimental data according to a ratio such as 8:2, and the experimental data can be obtained by measuring KrF photoresist with a thickness of 3300 Å, 4000 Å and 5200 Å and a critical dimension (CD) of 150 nm and 220 nm under different acceleration voltages (400 V, 500 V, 600 V, 700 V, 800 V) and different exposure times t by using a CD-SEM. When training, the embodiments of the present application set the mean square error (MSE value) of the initial model on the validation set as the fitness function value of the improved spider wasp optimization algorithm, and set the optimization target of the improved spider wasp optimization algorithm as minimizing the fitness function value, so that the problem of finding the optimal solution of the to-be-optimized parameters by the improved spider wasp optimization algorithm is converted into the problem of optimizing the following formula to continuously reduce the difference between the photoresist shrinkage value and the predicted photoresist shrinkage value:

[0062] In the formula, n represents the number of groups of validation data in the validation set, is the photoresist shrinkage value corresponding to the i th group of validation data in the validation set, Predicted shrinkage value obtained from initial model and raw values of CD-SEM accelerating voltage, CD-SEM exposure time, photoresist thickness and photoresist width measured by CD-SEM in the i-th group of verification data in the verification set.

[0063] Thus, the training of the constructed initial model based on the improved wasp optimization algorithm, the training sample set and the verification set is actually implemented as follows: the improved wasp optimization algorithm is used for an iterative solving process of a preset number of times, in each iteration process, the improved wasp optimization algorithm finds the solution of the to-be-optimized parameter corresponding to the current population by using the training data in the training sample set (for example, the solution corresponding to the t-th generation population is determined in the t-th iteration process); after the solution of the to-be-optimized parameter corresponding to the current population is determined, the initial model is further updated by using the solution of the to-be-optimized parameter, and the mean square error value of the updated initial model is calculated by using the verification set; then, the initial model corresponding to a group of to-be-optimized parameters with the minimum mean square error value is selected from the updated initial model as the optimal solution of the current population, and the optimal solution of the current population is selected as the fitness; then, the position of the solution of the to-be-optimized parameter is updated according to the calculated fitness, and after the position of the solution is updated, the next improved wasp optimization algorithm iteration process is continued on the basis of the position of the solution by using the training sample set, so as to obtain the solution of the to-be-optimized parameter corresponding to the next generation population; in this way, when the preset number of iterations is reached, the optimal solution of the to-be-optimized parameter in the entire iteration process can be found. Then, the initial model is updated by using the group of optimal solutions iterated by the improved wasp optimization algorithm, and the trained prediction model is determined by using the training sample set to train the updated initial model. It should be noted that the preset number of iterations can be set according to requirements and experience, for example, the preset number of iterations can be set to 50 times. In addition, it should be noted that the initial model of the embodiment of the present application refers to the regression model before the trained prediction model is obtained, including the model constructed at the initialization and the model in the intermediate state formed in the iteration process, and after being trained, it is referred to as a prediction model, and the initial model corresponding to each group of solutions is an SVR model obtained by updating the parameters of the initial model with each group of to-be-optimized parameters. The detailed implementation process of the improved wasp optimization algorithm for finding the solution of the to-be-optimized parameter corresponding to the current population by using the training data in the training sample set, and the detailed implementation process of the training of the regression model of the updated initial model by using the training sample set, can be implemented based on the algorithm process description of the corresponding algorithm described above and the prior art, and the process will not be described in detail.

[0064] wherein, Figure 5 The process of optimization and solving by using the improved wasp optimization algorithm is schematically shown as follows: Figure 5As shown, the algorithm process first initializes the population based on the set of good points, then determines the current population corresponding to the to-be-optimized parameter solution at the preset iteration number t max within the range, uses the improved spider wasp optimization algorithm to determine the current population corresponding to the to-be-optimized parameter solution at the current iteration number t, then updates the initial model using the current population corresponding to the to-be-optimized parameter solution, and calculates the mean square error value of the updated initial model using the verification set to find the optimal solution in the current population as the fitness of the solution, and updates the position of the to-be-optimized parameter solution according to the calculated fitness, while using the cyclic population reduction technique to reduce the population, then performing the next iteration process until reaching the preset iteration number t max , and outputting the found optimal solution. Wherein, the algorithm uses the improved scheme described in the foregoing in population initialization, hunting and nesting behavior, mating behavior weight adjustment, and population reduction.

[0065] The embodiment of the present application aims at the shrinkage error problem of photoresist caused by electron beam irradiation in CD-SEM measurement process, and proposes a prediction method of optimizing support vector regression (SVR) model based on improved spider wasp optimization algorithm (referred to as ISWO in the embodiment of the present application). In the hyperparameter optimization stage, ISWO is used to optimize the penalty factor C, the kernel function parameter sigma and the tolerance deviation width epsilon of the insensitive loss function of the SVR model. ISWO improves the global search ability and convergence efficiency of the original SWO algorithm by introducing the strategies of cyclic population reduction, dynamic weight balancing probability and good point set initialization, and improves the stability and precision of parameter optimization. Therefore, the prediction model trained by ISWO has good performance in fitting precision, error control and convergence stability. The method can not only effectively improve the prediction ability of the SVR model, but also provides a practical modeling method for the shrinkage error correction of CD-SEM.

[0066] In order to systematically evaluate the hyperparameter optimization capability of ISWO algorithm in support vector regression (SVR) model and verify its effectiveness in photoresist shrinkage prediction task, the embodiment of the present application designs comparative experiments based on other six kinds of optimization algorithms, which include SWO (i.e. the improved spider wasp optimization algorithm), RIME (a high-efficiency optimization algorithm based on the physical phenomenon of fog ice, referred to as RIME, also referred to as frost ice optimization algorithm), PSO (Particle Swarm Optimization, particle swarm optimization algorithm), DBO (Dung beetle optimizer, dung beetle optimization algorithm), GWO (Grey Wolf Optimizer, grey wolf optimization algorithm) and BKA (black-winged kite algorithm), wherein, Figure 6The ISWO and other six optimization algorithms are shown in the support vector regression (SVR) model in the process of super parameter optimization, the mean square error (MSE) convergence curve obtained in each iteration, the abscissa in the figure is the iteration number, and the ordinate is the MSE value under the corresponding iteration, from the figure, it can be seen that ISWO rapidly reduces the MSE to a lower level in the early iteration, and continues to decrease slightly in the subsequent iteration, and finally stabilizes at about 0.0034, the convergence speed and final accuracy are better than other algorithms.

[0067] To further evaluate the effect of different optimization algorithms on the optimization of SVR model super parameters, the SWO, RIME, PSO, DBO, GWO, BKA and ISWO proposed in the application are used to optimize the super parameters of the SVR model, and the performance is evaluated on the unified data set. Figure 7 The optimal super parameter combination (ε, C, σ) obtained by each algorithm and its prediction performance index on the validation set are listed, including mean square error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and determination coefficient (R²).

[0068] As shown in Figure 7 From the results in the table, it can be seen that the ISWO-SVR model performs well in all evaluation indicators. Its MSE is 0.0034, which is better than the remaining six algorithms, indicating that its prediction error is the smallest; at the same time, its MAE and MAPE are 0.0408 and 15.4993% respectively, which are at a low level, indicating that the model performs stably in the absolute value and relative proportion of prediction error; R² reaches 0.9964, close to 1, indicating that the model has a high goodness of fit and can accurately reflect the change trend of photoresist shrinkage. Compared with the PSO-SVR, although it also performs well in MSE (0.0038) and R² (0.996), its MAPE is 24.2867%, which is significantly higher than that of ISWO-SVR, indicating that it has a deficiency in relative error control. In summary, ISWO not only has stronger global search ability, but also can quickly jump out of local optimum and find better solution in a limited number of iterations. The results further verify the effectiveness and robustness of ISWO in the SVR super parameter optimization task proposed in this study, and the effectiveness of solving the photoresist shrinkage problem.

[0069] To more intuitively compare the performance of different optimization algorithms in the SVR model, Figure 8A radar chart based on four indicators (MSE, MAE, MAPE and R²) is drawn. For MSE, MAE and MAPE, the smaller the value, the better the performance, and the closer the vertex to the center of the graph. The R² value is closer to 1, so in this dimension, the farther the vertex is from the center, the stronger the fitting ability. As can be seen from the distribution in the figure, the ISWO comprehensive performance is better than other optimization algorithms.

[0070] Figure 9 Figure 1 is a scatter plot of the predicted value of the ISWO-SVR model and the actual value of the photoresist shrinkage (ΔCD). The abscissa is the actual ΔCD value, the ordinate is the predicted value, and the dashed line represents the ideal prediction result (i.e. the reference line where the predicted value is completely consistent with the actual value). As can be seen from the figure, most data points are closely distributed near the reference line, indicating that the prediction results of the model are highly consistent with the actual values. This shows that the ISWO-SVR model has good fitting ability and high prediction accuracy when modeling the photoresist shrinkage behavior. In addition, the data points are linearly aligned in the entire numerical range, without obvious systematic deviation or uneven distribution, further verifying the robustness and generalization ability of the model at different ΔCD levels.

[0071] In summary, the ISWO algorithm in the present application has excellent super parameter optimization ability and generalization performance, and has significant advantages in photoresist shrinkage modeling tasks, verifying its effectiveness and robustness in SVR model super parameter optimization.

[0072] Figure 10 Figure 1 schematically shows the photoresist shrinkage prediction system architecture of an embodiment of the present application, as shown in Figure 10 The system includes a critical dimension scanning electron microscope 100 and a computing device 200, wherein the computing device includes a shrinkage prediction module 201 for obtaining input data from the critical dimension scanning electron microscope and inputting the input data to a pre-trained prediction model, and obtaining an output result of the prediction model, wherein the prediction model is generated based on a spider optimization algorithm and support vector regression training, the output result is the shrinkage of the photoresist under SEM measurement, and the input data includes CD-SEM measurement conditions, CD-SEM measured photoresist thickness and photoresist width original values.

[0073] The specific implementation process of each module in the system and the specific training process of the prediction model can be referred to the description of the method part in the foregoing, and will not be described here. It should be noted that in other embodiments, the photoresist shrinkage prediction system can also be directly implemented as a computing device including the shrinkage prediction module 201, and the critical dimension scanning electron microscope can not be part of the system, but only as a third-party provider of input data, and the embodiments of the present application are not limited in this regard.

[0074] Figure 11 An electronic device according to an embodiment of the present application is schematically shown in the figure, which is implemented as comprising: a memory 81 for storing executable instructions; and a processor 82 communicatively connected with the memory 81, configured to execute the executable instructions stored in the memory, which when executed by the processor implement the steps of the critical dimension scanning electron microscope resist image processing method according to any one of the preceding embodiments.

[0075] In some embodiments, the present application provides a non-transitory computer readable storage medium having stored therein one or more programs comprising instructions executable by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to read and execute, for performing the critical dimension scanning electron microscope resist image processing method according to any one of the preceding embodiments.

[0076] In some embodiments, the present application also provides a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions that, when executed by a computer, cause the computer to perform the critical dimension scanning electron microscope resist image processing method according to any one of the preceding embodiments.

[0077] In some embodiments, the present application also provides an electronic device comprising at least one processor, and a memory communicatively connected with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the critical dimension scanning electron microscope resist image processing method according to any one of the preceding embodiments.

[0078] In some embodiments, the present application also provides a storage medium having stored thereon a computer program, which when executed by a processor implement the critical dimension scanning electron microscope resist image processing method according to any one of the preceding embodiments.

[0079] Figure 12 is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application, and the computing device described above can be implemented in the structure shown in the figure, as shown in the figure, the electronic device comprises: Figure 12 as shown, the electronic device comprises: one or more processors 610 and a memory 620, Figure 12 of which the processor 610 is taken as an example.

[0080] The electronic device can further include an input device 630 and an output device 640.

[0081] The processor 610, the memory 620, the input device 630, and the output device 640 can be connected by a bus or other means, Figure 12 For example, the connection by the bus is taken as an example.

[0082] The memory 620, as a non-volatile computer readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs, and modules, such as program instructions / modules corresponding to the critical dimension scanning electron microscope photoresist image processing method in the embodiment of the application. The processor 610 executes various function applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 620, that is, implements the critical dimension scanning electron microscope photoresist image processing method of the above method embodiment.

[0083] The memory 620 can include a program storage area and a data storage area, wherein the program storage area can store an operating system and application programs required by at least one function; the data storage area can store data created according to the use of the critical dimension scanning electron microscope photoresist image processing method, etc. In addition, the memory 620 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 can optionally include a memory remotely arranged with respect to the processor 610, and these remote memories can be connected to the electronic device through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0084] The input device 630 can receive input digital or character information, and generate signals related to user settings and function control of the image processing device. The output device 640 can include a display device such as a display screen.

[0085] The one or more modules are stored in the memory 620, and when executed by the one or more processors 610, perform the critical dimension scanning electron microscope photoresist image processing method in any of the above method embodiments.

[0086] The above product can perform the method provided by the embodiment of the application, has the function modules and beneficial effects corresponding to the execution method. Technical details not described in detail in the embodiment can be referred to the method provided by the embodiment of the application.

[0087] The electronic device of the embodiment of the application exists in various forms, including but not limited to: (1) Mobile communication device: The feature of this kind of device is to have mobile communication function, and to provide voice and data communication as the main target. This kind of terminal includes: smart phone (such as iPhone), multimedia phone, functional phone, and low-end phone, etc.

[0088] (2) Ultra mobile personal computer device: This kind of device belongs to the category of personal computer, has computing and processing function, and generally has mobile internet feature. This kind of terminal includes: PDA, MID and UMPC device, such as iPad.

[0089] (3) Portable entertainment device: This kind of device can display and play multimedia content. This kind of device includes: audio and video player (such as iPod), palm game machine, electronic book, and smart toy and portable car navigation device.

[0090] (4) Server: The device providing computing service, the constitution of server includes processor, hard disk, memory, system bus, etc., the server and general computer architecture are similar, but because of needing to provide high reliable service, therefore in processing capacity, stability, reliability, security, scalability, manageability, etc. require higher.

[0091] (5) Other electronic devices with data interaction function.

[0092] The device embodiments described above are only schematic, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e. may be located in one place, or may be distributed on a plurality of network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0093] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software plus a general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions essentially or say the part that contributes to the related art can be embodied in the form of software product, and the computer software product can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, server, or network device, etc.) execute the method described in each embodiment or some part of the embodiment.

[0094] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the same; although the present application has been described in detail with reference to the foregoing examples, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting photoresist shrinkage, characterized in that: include: Acquiring input data, wherein the input data includes CD-SEM measurement conditions, and original values ​​of photoresist thickness and photoresist width measured by CD-SEM; The obtained input data is input into a pre-trained prediction model, and an output result of the prediction model is obtained, wherein the prediction model is trained based on an improved spider bee optimization algorithm and a support vector regression algorithm, and the output result is the shrinkage amount caused by the photoresist under CD-SEM measurement.

2. The method according to claim 1, characterized in that The prediction model is trained by the following method: Based on the CD-SEM measurement conditions, the original values ​​of the photoresist thickness and the photoresist width measured by CD-SEM, a regression model is performed on the shrinkage of the photoresist under CD-SEM measurement to generate an initial model based on support vector regression; The constructed initial model is trained based on the improved spider bee optimization algorithm, the training sample set and the verification set to generate the prediction model.

3. The method according to claim 2, characterized in that The constructed initial model is trained based on the improved spider bee optimization algorithm, the training sample set and the validation set to generate the prediction model, including: The mean square error of the initial model on the validation set is set as the fitness function value, the optimization target is set to minimize the fitness function value, and the initial model is trained using the improved spider bee optimization algorithm based on the optimization target, the training sample set and the validation set to generate the prediction model.

4. The method according to claim 3, characterized in that The initial model is trained using the improved spider bee optimization algorithm based on the optimization objective, the training sample set, and the validation set, including: Based on the improved spider bee optimization algorithm, training sample set, validation set and optimization target, the parameters to be optimized of the initial model are optimized to determine the optimal solution of the parameters to be optimized; Based on the determined optimal solution and training sample set, the initial model is subjected to regression training to generate a prediction model.

5. The method according to claim 4, characterized in that Based on the improved spider bee optimization algorithm, training sample set, validation set and optimization target, the parameters to be optimized of the initial model are optimized, and the optimal solutions of the parameters to be optimized are determined to include: In each iteration, the following processing is performed until the preset number of iterations is reached, and the optimal solution of the parameters to be optimized is output when the preset number of iterations is reached: Using the training data in the training sample set to determine the solution of the parameters to be optimized corresponding to the current population, and using the solution of the parameters to be optimized to update the initial model; The updated initial model is used to calculate its mean square error using the validation set and a set of parameter solutions to be optimized that minimizes the mean square error is selected as the fitness; The position of the solution of the parameter to be optimized is updated according to the calculated fitness.

6. The method according to claim 5, characterized in that Before starting the iterative process, the improved spider bee optimization algorithm performs population initialization based on the good point set.

7. The method according to claim 6, characterized in that The improved spider bee optimization algorithm adopts a cyclic population reduction technique to reduce the population during the iteration process.

8. The method according to claim 6 or 7, characterized in that The hunting behavior and nesting behavior of the improved spider bee optimization algorithm are weighed by the following formula: in, is the trade-off rate in the t-th iteration, The adjustment range of the value is limited to between 0.1 and 0.9, t is the current iteration number, is the preset threshold, To adjust the step size, is the optimal fitness improvement, which is calculated as follows: in represents the best fitness in the t-th iteration, represents the best fitness in the t-1th iteration.

9. A photoresist shrinkage prediction system, characterized in that: include: The shrinkage prediction module is used to obtain input data from a critical dimension scanning electron microscope and input it into a pre-trained prediction model, and obtain the output result of the prediction model, wherein the prediction model is trained and generated based on the improved spider bee optimization algorithm and the support vector regression algorithm, and the output result is the shrinkage caused by the photoresist under SEM measurement. The input data includes CD-SEM measurement conditions, and the original values ​​of the photoresist thickness and photoresist width measured by CD-SEM.

10. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method according to any one of claims 1 to 8.

Images