CN114488719B - OPC method based on three-dimensional feature reinforcement - Google Patents

Abstract

The invention belongs to the technical field of optical proximity correction, and discloses an OPC method based on three-dimensional feature reinforcement. The invention solves the problem of large single calculation amount caused by the integral optimization of the existing optimization method through local optimization; the exposure dose distribution is optimized by a local optimization method aiming at three-dimensional height characteristics, so that the problem of missing of the height optimization in the existing proper matrix optimization method is solved; meanwhile, the optimal threshold value can be automatically found, and the problem that the gray gradient threshold value cannot be set manually is solved.

Description

An OPC method based on three-dimensional feature enhancement

Technical field

The invention belongs to the technical field of optical proximity correction, and more specifically, relates to an OPC method based on three-dimensional feature enhancement.

Background technique

Photolithography is an important process in many fields such as semiconductor integrated circuits, MEMS, micro-nano optics, etc. It is the etching or deposition of masks (such as silicon dioxide) on the surface of semiconductor wafers to facilitate localized diffusion of impurities. a processing technology. Its expensive equipment manufacturing and maintenance costs account for about half of the total integrated circuit production costs. Among them, mask manufacturing costs are huge. In order to reduce costs, maskless lithography technologies such as electron beam lithography, ion beam lithography and scanning laser lithography are gradually emerging.

The scanning laser lithography system mainly includes: laser light source, focus modulation system, stepping system, scanning system and silicon wafer coated with photoresist. Among them, the laser beam is focused to the spot size of the target size, and the beam power is modulated while scanning the photoresist surface (the exposure depth of the grayscale photoresist increases nonlinearly with the exposure energy, commonly used as an s-shaped curve approximation), and then in Scan with the set step size in the scanning direction. After scanning for one period, step once in the step direction to continue scanning in the reverse direction.

In the process of scanning lithography imaging, the imaging needs to be optimized due to the superposition of energy from other scanning areas. In addition, as the critical dimension (CD) continues to decrease, the feature size of scanning laser lithography is restricted. In a scanning laser lithography system, the exposure dose at each position can be precisely controlled, so the superimposed energy can be planned and compensated by simulating and calculating the exposure dose distribution, so that imaging resolution enhancement can be achieved.

Optical proximity correction (OPC) technology for scanning laser lithography usually uses photolithography machine and photoresist parameters to build a nonlinear numerical model, and uses nonlinear programming methods or gradient-based optimization algorithms to find the optimal exposure dose distribution. Thereby minimizing the difference between the desired pattern and the exposed pattern (mainly including pattern error (PE) and edge placement error (EPE)), the output pattern is optimized. In 3D thick film lithography, optimization of height is of great significance. 3D structures not only include dimensional accuracy on the horizontal plane, but also include dimensional accuracy on the height. Changes in height are more susceptible to the influence of the diffraction limit, resulting in unclear edges and excessive edge fillet sizes, which ultimately affect product quality. Therefore, the height parameters of 3D structures often play a key role in the function of the structure.

Although the existing vector matrix-based 3D lithography optimization method reduces the overall calculation cost, it still has the following defects: (1) The optimization object is the overall optimization of the entire exposure area, and the optimization of the pattern edges is also processed. , without focusing on optimizing the pattern edge (that is, the interface between the target pattern and the etched part during photolithography), it is impossible to further improve the optimization speed and accuracy. (2) There is a lack of local optimization of the z-axis (i.e., height direction) of 3D lithography, and the structural height characteristics are not fully optimized.

Contents of the invention

By providing an OPC method based on three-dimensional feature enhancement, the present invention solves the problems in the prior art that the high-level feature optimization is lacking and the optimization speed and accuracy need to be improved.

The present invention provides an OPC method based on three-dimensional feature enhancement, which mainly includes the following steps:

Step S1: Obtain an optimized photoresist function model based on the photoresist exposure data and the chemical reaction function of the photoresist; convert the target pattern into a target pixelated pattern according to the photolithography machine parameters;

Step S2: Use the Sobel operator to perform edge extraction on the target pixelated pattern under two different gradient thresholds to obtain the corresponding edge feature matrix, and use the value of the edge feature matrix as the value of the target pattern matrix;

Step S3: Obtain the initial imaging pattern matrix according to the photolithography machine parameters and the optimized photoresist function model;

Step S4: Construct the first cost function and constraint conditions, and update the values of the exposure dose distribution and imaging pattern matrix according to the constraint conditions;

Step S5: Automatically update the gradient threshold according to the update judgment conditions;

Step S6: Determine whether the loop end condition is satisfied; if not, return to step S2; if satisfied, end the loop.

Preferably, step S1 includes the following sub-steps:

Step S11: Establish a second cost function based on the actual measured photoresist exposure data and the chemical reaction function of the photoresist;

The second cost function is expressed as:

Among them, H represents the second cost function, PR represents the actual measured photoresist exposure data, and Sig(·) represents the chemical reaction function of the photoresist;

Step S12: According to the second cost function, use an optimization algorithm to optimize the photoresist parameters until the second cost function is minimum and obtain optimal photoresist parameters. The photoresist parameters include etching Speed and etch threshold;

Step S13: Obtain an optimized photoresist function model based on the optimal etching speed and the optimal etching threshold;

The optimized photoresist function model is expressed as:

Among them, a represents the optimal etching speed, t _r represents the optimal etching threshold;

Step S14: Input photolithography machine parameters to convert the target pattern into a target pixelated pattern.

Preferably, before using the Sobel operator for edge extraction, the method further includes: performing Gaussian smoothing on the target pixelated pattern.

Preferably, the Gaussian smooth convolution kernel is as follows:

0.075 0.124 0.075 0.124 0.204 0.124 0.075 0.124 0.075

By using the Gaussian smoothing convolution kernel and the target pixelated pattern to perform a convolution operation, Gaussian smoothing processing is performed on the target pixelated pattern.

Preferably, the step S2 includes the following sub-steps:

Step S21: Use the Sobel operator to perform a convolution operation with the target pixelated pattern to obtain a pattern gradient change map; the Sobel operator convolution kernel is as follows:

x direction:

y direction:

G＝|Gx|+|Gy|

Among them, Gx represents the gradient size of each pixel in the x-axis direction obtained by convolving the target pixelated pattern with the Sobel operator convolution kernel in the x direction, and Gy represents the convolution of the target pixelated pattern with the Sobel operator in the y direction. The gradient size of each pixel in the y-axis direction obtained after kernel convolution; G represents the height gradient size of each pixel, and the value of G is the sum of the absolute values of the gradient values of each pixel in the x- and y-axis directions;

Step S22: Based on the pattern gradient change map, according to the two set gradient thresholds β ₁ and β ₂ , extract two gradient distribution matrices greater than the gradient threshold as edge feature matrices S _z1 and S _z2 respectively;

For each gradient threshold, if S _ij ≥ β, then S _Zij =S _ij , otherwise, S _Zij =0;

Among them, S _ij represents the coordinate value of the pixel point in the i-th row and j-th column in the pattern gradient change map, and S _zij represents the coordinate value of the pixel point in the i-th row and jth column in the edge feature matrix;

The value of the edge feature matrix is used as the value of the target pattern matrix Z(x, y).

Preferably, step S3 includes the following sub-steps:

Step S31: Perform pixelation processing on the target pattern according to the lithography machine parameters to obtain an exposure dose distribution matrix:

Among them, E(x, y) represents the exposure dose distribution matrix. The initial value of the exposure dose distribution matrix comes from the value of the corresponding pixel position of the target pattern matrix Z(x, y). (x, y) represents the position of an exposure point. Coordinates, the position coordinate of a single exposure point is equal to the position coordinate of the corresponding pixel point, θ represents the unconstrained optimization variable;

Step S32: Obtain the Gaussian beam matrix:

Among them, B(x, y) represents the Gaussian beam matrix, P represents the overall exposure power, and ω ₀ is the radius of the laser spot at the focal plane;

Step S33: Obtain the exposure energy distribution matrix according to the exposure dose distribution matrix and the Gaussian beam matrix:

Among them, D(x, y) represents the exposure energy distribution matrix, Represents the convolution symbol;

Step S34: Obtain the initial imaging pattern matrix according to the exposure energy distribution matrix and the optimized photoresist function model:

in, Represents the initial imaging pattern matrix.

Preferably, in step S4, the first cost function and constraints are constructed in combination with an optimization algorithm, and the exposure dose distribution matrix and imaging pattern matrix are iteratively optimized:

Among them, F _Z represents the pattern error after photoresist imaging, F _E represents the total output dose of the system, F represents the first cost function, Z (x, y) represents the target pattern matrix, represents the imaging pattern matrix, and E(x, y) represents the exposure dose distribution matrix;

Iteratively optimize the exposure dose distribution matrix:

Among them, s represents the updated step size in the optimization algorithm.

Preferably, in step S5, the update judgment condition includes a first condition and a second condition; the first condition is F ₂ -F ₁ >0, and the second condition is F ₂ -F ₁ ≤0; F ₁ represents the first cost function constructed with the edge feature matrix under the gradient threshold β ₁ , and F ₂ represents the first cost function constructed with the edge feature matrix under the gradient threshold β ₂ ;

If the first condition is met, the gradient threshold is automatically updated according to the following principles:

β ₂ =β ₁ , β ₁ =β ₁ -α(F ₂ -F ₁ )

If the second condition is met, the gradient threshold is automatically updated according to the following principles:

β ₁ =β ₂ , β ₂ =β ₂ +α(F ₂ -F ₁ )

Among them, α represents the set step size.

Preferably, in step S6, if the loop ends, the smaller value of β ₁ and β ₂ is taken as the optimal gradient threshold, the edge feature matrix corresponding to the optimal gradient threshold is taken as the optimal edge feature matrix, and the The exposure dose distribution matrix and exposure energy distribution matrix corresponding to the optimal edge feature matrix are regarded as the global optimal distribution.

Preferably, in step S6, the loop end condition is that the number of optimization times is reached or the value of the first cost function is less than the optimization threshold.

One or more technical solutions provided in the present invention have at least the following technical effects or advantages:

In the present invention, an optimized photoresist function model is obtained based on the photoresist exposure data and the chemical reaction function of the photoresist; the target pattern is converted into a target pixelated pattern according to the photolithography machine parameters; the Sobel operator is used to calculate the target The edges of the pixelated pattern are extracted under two different gradient thresholds to obtain the corresponding edge feature matrix. The value of the edge feature matrix is used as the value of the target pattern matrix. According to the photolithography machine parameters and the optimized photoresist function model, we obtain Initial imaging pattern matrix; construct the first cost function and constraint conditions, update the exposure dose distribution and imaging pattern matrix values according to the constraint conditions; automatically update the gradient threshold according to the update judgment conditions; determine whether the loop end condition is met after the update ; If not satisfied, return to the edge extraction step; if satisfied, end the loop. The traditional exposure dose distribution solution is based on the entire pattern, that is, the optimization object is the overall optimization of the entire exposure area, resulting in a large amount of single calculation and slow optimization speed. The present invention solves the edges of features, uses the Sobel operator to extract the edge matrix, and performs targeted optimization, which can improve the optimization effect and speed. The present invention performs three-dimensional structural optimization of exposure dose distribution based on a highly optimized method of Sobel operator edge extraction, and solves the problem of missing height optimization in existing appropriate matrix optimization methods. The present invention performs edge extraction under two different gradient thresholds, automatically updates the gradient threshold according to the update judgment conditions, and then obtains the optimal gradient threshold. That is, the present invention proposes an automatic search for the optimal threshold method, which solves the problem of grayscale The gradient threshold is artificially unable to set the optimal value.

Description of the drawings

Figure 1 is a flow chart of an OPC method based on three-dimensional feature enhancement provided by an embodiment of the present invention.

Detailed ways

In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation modes.

This embodiment provides an OPC method based on three-dimensional feature enhancement, see Figure 1, which includes the following steps:

Step S1: Obtain an optimized photoresist function model based on the photoresist exposure data and the chemical reaction function of the photoresist; convert the target pattern into a target pixelated pattern according to the photolithography machine parameters.

Among them, the step S1 includes the following sub-steps:

Step S11: Establish a second cost function based on the actual measured photoresist exposure data and the chemical reaction function of the photoresist;

The second cost function is expressed as:

Among them, H represents the second cost function, PR represents the actual measured photoresist exposure data, including exposure energy and film retention rate; Sig(·) represents the chemical reaction function of the photoresist.

Step S12: According to the second cost function, use an optimization algorithm to optimize the photoresist parameters until the second cost function is minimum and obtain optimal photoresist parameters. The photoresist parameters include etching Speed and etch threshold.

Step S13: Obtain an optimized photoresist function model based on the optimal etching speed and the optimal etching threshold;

The optimized photoresist function model is expressed as:

Among them, a represents the optimal etching speed, and _tr represents the optimal etching threshold.

Step S14: Input photolithography machine parameters to convert the target pattern into a target pixelated pattern.

Among them, the lithography machine parameters include resolution, scanning speed, stepping speed, light source size, etc.

Step S2: Use the Sobel operator to perform edge extraction on the target pixelated pattern under two different gradient thresholds to obtain the corresponding edge feature matrix, and use the value of the edge feature matrix as the value of the target pattern matrix.

Among them, the step S2 includes the following sub-steps:

Step S21: Use the Sobel operator to perform a convolution operation with the target pixelated pattern to obtain a pattern gradient change map; the Sobel operator convolution kernel is as follows:

x direction:

y direction:

G＝|Gx|+|Gy|

Among them, Gx represents the gradient size of each pixel in the x-axis direction obtained by convolving the target pixelated pattern with the Sobel operator convolution kernel in the x direction, and Gy represents the convolution of the target pixelated pattern with the Sobel operator in the y direction. The gradient size of each pixel in the y-axis direction obtained after kernel convolution; G represents the height gradient size of each pixel, and the value of G is the sum of the absolute values of the gradient values of each pixel in the x- and y-axis directions.

Step S22: Based on the pattern gradient change map, according to the two set gradient thresholds β ₁ and β ₂ , extract two gradient distribution matrices greater than the gradient threshold as edge feature matrices S _z1 and S _z2 respectively;

For each gradient threshold, if S _ij ≥ β, then S _Zij =S _ij , otherwise, S _Zij =0;

Among them, S _ij represents the coordinate value of the pixel point in the i-th row and j-th column in the pattern gradient change map, and S _zij represents the coordinate value of the pixel point in the i-th row and jth column in the edge feature matrix;

The value of the edge feature matrix is used as the value of the target pattern matrix Z(x, y).

Step S3: Obtain an initial imaging pattern matrix based on the photolithography machine parameters and the optimized photoresist function model.

Among them, the step S3 includes the following sub-steps:

Step S31: Perform pixelation processing on the target pattern according to the lithography machine parameters to obtain an exposure dose distribution matrix:

Among them, E(x, y) represents the exposure dose distribution matrix. The initial value of the exposure dose distribution matrix comes from the value of the corresponding pixel position of the target pattern matrix Z(x, y). (x, y) represents the position of an exposure point. Coordinates, the position coordinates of a single exposure point are equal to the position coordinates of the corresponding pixel point, and θ represents the unconstrained optimization variable.

Step S32: Obtain the Gaussian beam matrix:

Among them, B (x, y) represents the Gaussian beam matrix, P represents the overall exposure power, and ω ₀ is the radius of the laser spot at the focal plane; the mathematical form of the Gaussian beam matrix is related to the parameters of the lithography machine, because the parameters of the lithography machine are fixed. , so the value of the Gaussian beam matrix is also fixed in the next steps.

Step S33: Obtain the exposure energy distribution matrix according to the exposure dose distribution matrix and the Gaussian beam matrix:

Among them, D(x, y) represents the exposure energy distribution matrix, Represents the convolution symbol;

Step S34: Obtain the initial imaging pattern matrix according to the exposure energy distribution matrix and the optimized photoresist function model:

in, Represents the initial imaging pattern matrix.

Step S4: Construct a first cost function and constraint conditions, and update the exposure dose distribution and imaging pattern matrix values according to the constraint conditions.

Specifically, combined with the optimization algorithm, the first cost function and constraints are constructed, and the exposure dose distribution matrix and imaging pattern matrix are iteratively optimized:

Among them, F _Z represents the pattern error after photoresist imaging, F _E represents the total output dose of the system, F represents the first cost function, Z (x, y) represents the target pattern matrix, represents the imaging pattern matrix, and E(x, y) represents the exposure dose distribution matrix.

Iteratively optimize the exposure dose distribution matrix:

Among them, s represents the updated step size in the optimization algorithm.

Step S5: Automatically update the gradient threshold according to the update judgment condition.

Wherein, the update judgment condition includes a first condition and a second condition; the first condition is F ₂ -F ₁ > 0, and the second condition is F ₂ -F ₁ ≤ 0; F ₁ represents the gradient threshold The first cost function constructed with the edge feature matrix under β ₁ , F ₂ represents the first cost function constructed with the edge feature matrix under the gradient threshold β ₂ .

If the first condition is met, the gradient threshold is automatically updated according to the following principles:

β ₂ =β ₁ , β ₁ =β ₁ -α(F ₂ -F ₁ )

If the second condition is met, the gradient threshold is automatically updated according to the following principles:

β ₁ =β ₂ , β ₂ =β ₂ +α(F ₂ -F ₁ )

Among them, α represents the set step size.

Step S6: Determine whether the loop end condition is satisfied; if not, return to step S2; if satisfied, end the loop.

Among them, if the loop is ended, the smaller value of β ₁ and β ₂ is taken as the optimal gradient threshold, the edge feature matrix corresponding to the optimal gradient threshold is taken as the optimal edge feature matrix, and the optimal edge feature matrix is The corresponding exposure dose distribution matrix and exposure energy distribution matrix are used as the global optimal distribution.

The loop end condition is that the number of optimization times is reached or the value of the first cost function is less than the optimization threshold.

In addition, in a preferred solution, before using the Sobel operator for edge extraction, the method further includes: performing Gaussian smoothing on the target pixelated pattern.

Specifically, the Gaussian smooth convolution kernel is as follows:

0.075 0.124 0.075 0.124 0.204 0.124 0.075 0.124 0.075

By using the Gaussian smoothing convolution kernel and the target pixelated pattern to perform a convolution operation, Gaussian smoothing processing is performed on the target pixelated pattern.

The present invention uses the Sobel operator to process the grayscale image. After extracting the image edge, the exposure dose distribution of the image edge is obtained by performing reverse photolithography calculation on the image edge, because the computational complexity of the solution varies with the individual pixels. The number increases rapidly. In this process, the method provided by the present invention reduces the number of pixel points that need to be solved, thereby reducing the computational complexity. In addition, the size of the gradient threshold affects the size of the cost function, and the appropriate gradient thresholds for different patterns are not necessarily the same. The appropriate gradient threshold cannot be determined in advance. Faced with the difficulty of artificially setting the threshold, the present invention adopts It adopts an automatic optimization algorithm. When two different initial gradient thresholds are given, the gradient threshold can be automatically adjusted to the optimal value through the optimization step, further reducing the difficulty in solving the problem. Therefore, the technical solution provided by the present invention has the advantages of speed, efficiency, and simple operation, and can greatly increase the speed of reverse lithography.

An OPC method based on three-dimensional feature enhancement provided by the embodiment of the present invention at least includes the following technical effects:

The present invention solves the problem of large single calculation amount caused by existing optimization methods for overall optimization through local optimization; provides a local optimization method for three-dimensional height characteristics to optimize exposure dose distribution, and solves the problem of existing appropriate matrix optimization methods. It highly optimizes the missing problem and can automatically find the optimal threshold, solving the problem that the gray gradient threshold cannot be manually set to the optimal value.

Finally, it should be noted that the above specific embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to examples, those of ordinary skill in the art will understand that the technical solutions of the present invention can be carried out. Modifications or equivalent substitutions without departing from the spirit and scope of the technical solution of the present invention shall be included in the scope of the claims of the present invention.

Claims (6)

1. An OPC method based on three-dimensional feature enhancement, which is characterized by including the following steps: Step S1: Obtain an optimized photoresist function model based on the photoresist exposure data and the chemical reaction function of the photoresist; convert the target pattern into a target pixelated pattern according to the photolithography machine parameters; The step S1 includes the following sub-steps: Step S11: Establish a second cost function based on the actual measured photoresist exposure data and the chemical reaction function of the photoresist; The second cost function is expressed as: Among them, H represents the second cost function, PR represents the actual measured photoresist exposure data, and Sig(·) represents the chemical reaction function of the photoresist; Step S12: According to the second cost function, use an optimization algorithm to optimize the photoresist parameters until the second cost function is minimum and obtain optimal photoresist parameters. The photoresist parameters include etching Speed and etch threshold; Step S13: Obtain an optimized photoresist function model based on the optimal etching speed and the optimal etching threshold; The optimized photoresist function model is expressed as: Among them, a represents the optimal etching speed, t _r represents the optimal etching threshold; Step S14: Enter the photolithography machine parameters to convert the target pattern into the target pixelated pattern; Step S2: Use the Sobel operator to perform edge extraction on the target pixelated pattern under two different gradient thresholds to obtain the corresponding edge feature matrix, and use the value of the edge feature matrix as the value of the target pattern matrix; The step S2 includes the following sub-steps: Step S21: Use the Sobel operator to perform a convolution operation with the target pixelated pattern to obtain a pattern gradient change map; the Sobel operator convolution kernel is as follows: x direction: y direction: G＝|Gx|+|Gy| Among them, Gx represents the gradient size of each pixel in the x-axis direction obtained by convolving the target pixelated pattern with the Sobel operator convolution kernel in the x direction, and Gy represents the convolution of the target pixelated pattern with the Sobel operator in the y direction. The gradient size of each pixel in the y-axis direction obtained after kernel convolution; G represents the height gradient size of each pixel, and the value of G is the sum of the absolute values of the gradient values of each pixel in the x- and y-axis directions; Step S22: Based on the pattern gradient change map, according to the two set gradient thresholds β ₁ and β ₂ , extract two gradient distribution matrices greater than the gradient threshold as edge feature matrices S _z1 and S _z2 respectively; For each gradient threshold, if S _ij ≥ β, then S _Zij =S _ij , otherwise, S _Zij =0; Among them, S _ij represents the coordinate value of the pixel point in the i-th row and j-th column in the pattern gradient change map, and S _zij represents the coordinate value of the pixel point in the i-th row and jth column in the edge feature matrix; Use the value of the edge feature matrix as the value of the target pattern matrix Z(x,y); Step S3: Obtain the initial imaging pattern matrix according to the photolithography machine parameters and the optimized photoresist function model; The step S3 includes the following sub-steps: Step S31: Perform pixelation processing on the target pattern according to the lithography machine parameters to obtain an exposure dose distribution matrix: Among them, E(x, y) represents the exposure dose distribution matrix. The initial value of the exposure dose distribution matrix comes from the value of the corresponding pixel position of the target pattern matrix Z(x, y). (x, y) represents the position of an exposure point. Coordinates, the position coordinate of a single exposure point is equal to the position coordinate of the corresponding pixel point, θ represents the unconstrained optimization variable; Step S32: Obtain the Gaussian beam matrix: Among them, B(x,y) represents the Gaussian beam matrix, P represents the overall exposure power, and ω ₀ is the radius of the laser spot at the focal plane; Step S33: Obtain the exposure energy distribution matrix according to the exposure dose distribution matrix and the Gaussian beam matrix: Among them, D(x, y) represents the exposure energy distribution matrix, Represents the convolution symbol; Step S34: Obtain the initial imaging pattern matrix according to the exposure energy distribution matrix and the optimized photoresist function model: in, Represents the initial imaging pattern matrix; Step S4: Construct the first cost function and constraint conditions, and update the values of the exposure dose distribution and imaging pattern matrix according to the constraint conditions; In the step S4, the optimization algorithm is combined to construct the first cost function and constraint conditions, and the exposure dose distribution matrix and imaging pattern matrix are iteratively optimized: Among them, F _Z represents the pattern error after photoresist imaging, F _E represents the total output dose of the system, F represents the first cost function, Z (x, y) represents the target pattern matrix, represents the imaging pattern matrix, and E(x, y) represents the exposure dose distribution matrix; Iteratively optimize the exposure dose distribution matrix: Among them, s represents the updated step size in the optimization algorithm; Step S5: Automatically update the gradient threshold according to the update judgment conditions; Step S6: Determine whether the loop end condition is satisfied; if not, return to step S2; if satisfied, end the loop. 2. The OPC method based on three-dimensional feature enhancement according to claim 1, characterized in that, before using the Sobel operator for edge extraction, it further includes: performing Gaussian smoothing on the target pixelated pattern. 3. The OPC method based on three-dimensional feature enhancement according to claim 2, characterized in that the Gaussian smooth convolution kernel is as follows: 0.075 0.124 0.075 0.124 0.204 0.124 0.075 0.124 0.075 By using the Gaussian smoothing convolution kernel and the target pixelated pattern to perform a convolution operation, Gaussian smoothing processing is performed on the target pixelated pattern. 4. The OPC method based on three-dimensional feature enhancement according to claim 1, characterized in that, in step S5, the update judgment condition includes a first condition and a second condition; the first condition is F ₂ -F ₁ >0, the second condition is F ₂ -F ₁ ≤ 0; F ₁ represents the first cost function constructed using the edge feature matrix under the gradient threshold β ₁ , and F ₂ represents the edge feature matrix under the gradient threshold β ₂ The first cost function constructed; If the first condition is met, the gradient threshold is automatically updated according to the following principles: β ₂ =β ₁ , β ₁ =β ₁ -α(F ₂ -F ₁ ) If the second condition is met, the gradient threshold is automatically updated according to the following principles: β ₁ =β ₂ , β ₂ =β ₂ +α(F ₂ -F ₁ ) Among them, α represents the set step size. 5. The OPC method based on three-dimensional feature enhancement according to claim 4, characterized in that, in step S6, if the loop is ended, the smaller value in _β1 and _β2 is taken as the optimal gradient threshold, and the optimal gradient threshold is taken. The edge feature matrix corresponding to the optimal gradient threshold is used as the optimal edge feature matrix, and the exposure dose distribution matrix and exposure energy distribution matrix corresponding to the optimal edge feature matrix are used as the global optimal distribution. 6. The OPC method based on three-dimensional feature enhancement according to claim 1, characterized in that in step S6, the loop end condition is that the number of optimization times is reached or the value of the first cost function is less than the optimization threshold.