Image processing method and device, computer equipment and computer readable storage medium

Abstract

The invention provides an image processing method and device, computer equipment and a computer readable storage medium, and belongs to the technical field of semiconductors. The method comprises the following steps: obtaining a developed pattern and an etched pattern; processing the developed pattern and the etched pattern to respectively obtain a developed pattern contour and an etched pattern contour; obtaining a target etching deviation between each candidate data point on the developed pattern contour and a corresponding data point on the etched pattern contour; obtaining M candidate features of each candidate data point on the developed graph contour, wherein M is an integer greater than 1; obtaining an etching deviation correlation index of each candidate feature according to the candidate feature of each candidate data point on the developed pattern contour and the target etching deviation thereof; according to the etching deviation relevancy index of each candidate feature, M1 target features are determined from the M candidate features, and M1 is an integer larger than or equal to 1 and smaller than or equal to M.

Description

Technical Field

[0001] The present disclosure relates to the field of semiconductor technology, and in particular, to an image processing method, apparatus, computer device, and computer-readable storage medium. Background Art

[0002] In the integrated circuit manufacturing process, etching is the process of selectively removing the portion of the thin film layer that is not masked by the patterned mask layer, so as to obtain a pattern on the thin film that is completely consistent with the patterned mask layer. Physical simulation of the etching process is difficult. In the process of etching the material, both the dry etching process and the wet etching process are coupled by multiple factors, such as the shape and density of the mask pattern, the diffusion of the etching material, the flow of the etchant, the interface stratification, the chemical reaction, etc. Whether the deviation caused by etching can be accurately predicted in order to transfer the mask pattern to be transferred is one of the key steps in forming a semiconductor device with performance that meets the requirements. However, the amount of data used to train the etching deviation model in the related art is large, and there may be features in the data that have a low correlation with the etching deviation, resulting in poor accuracy of the etching deviation model obtained by training with the data. Summary of the invention

[0003] The disclosed embodiment provides an image processing method, which includes: obtaining a post-development graphic and a post-etching graphic; processing the post-development graphic and the post-etching graphic to obtain a post-development graphic contour and a post-etching graphic contour, respectively; obtaining a target etching deviation between each candidate data point on the post-development graphic contour and a corresponding data point on the post-etching graphic contour; obtaining M candidate features of each candidate data point on the post-development graphic contour, where M is an integer greater than 1; obtaining an etching deviation correlation index of each candidate feature based on the candidate feature of each candidate data point on the post-development graphic contour and its target etching deviation; and determining M1 target features from the M candidate features based on the etching deviation correlation index of each candidate feature, where M1 is an integer greater than or equal to 1 and less than or equal to M.

[0004] According to another aspect of the present disclosure, there is provided an image processing device, the device comprising: an obtaining unit for obtaining a post-development graphic and a post-etching graphic; a processing unit for processing the post-development graphic and the post-etching graphic to obtain a post-development graphic contour and a post-etching graphic contour, respectively; the obtaining unit is further used to obtain a target etching deviation between each candidate data point on the post-development graphic contour and a corresponding data point on the post-etching graphic contour; the obtaining unit is further used to obtain M candidate features of each candidate data point on the post-development graphic contour, M being an integer greater than 1; the obtaining unit is further used to obtain an etching deviation correlation index of each candidate feature based on the candidate feature of each candidate data point on the post-development graphic contour and its target etching deviation; a determining unit is used to determine M1 target features from the M candidate features based on the etching deviation correlation index of each candidate feature, M1 being an integer greater than or equal to 1 and less than or equal to M.

[0005] According to another aspect of the present disclosure, there is provided a computer device, comprising one or more processors; and a memory configured to store one or more programs, so that when the one or more programs are executed by the one or more processors, the computer device implements the image processing method in any embodiment of the present disclosure.

[0006] According to another aspect of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, wherein the computer program is suitable for being loaded and executed by a processor so that a computer device having the processor executes the image processing method in any embodiment of the present disclosure.

[0007] According to yet another aspect of the present disclosure, a computer program product is provided. When the computer program is executed by a processor, the image processing method in any embodiment of the present disclosure is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flowchart of an image processing method provided by an embodiment of the present disclosure is shown.

[0009] Figure 2 A schematic diagram showing a pattern after development in an embodiment of the present disclosure.

[0010] Figure 3 A schematic diagram showing a pattern after etching in an embodiment of the present disclosure.

[0011] Figure 4 A schematic diagram showing a graphic profile after development and a graphic profile after etching in an embodiment of the present disclosure.

[0012] Figure 5 A schematic diagram showing etching deviation in an embodiment of the present disclosure.

[0013] Figure 6 A schematic diagram showing etching deviation in another embodiment of the present disclosure.

[0014] Figure 7 A schematic diagram showing a target etching deviation between each candidate data point on a post-development graphic profile and a corresponding data point on a post-etched graphic profile in an embodiment of the present disclosure.

[0015] Figure 8 A schematic diagram showing the aperture effect in an embodiment of the present disclosure.

[0016] Fig. 9 Schematic diagram showing the micro-loading effect in an embodiment of the present disclosure.

[0017] Fig.10 A schematic diagram showing a density kernel function in an embodiment of the present disclosure.

[0018] Fig.11 A schematic diagram showing a visible kernel function in an embodiment of the present disclosure.

[0019] Fig.12 A schematic diagram showing an occlusion kernel function in an embodiment of the present disclosure is shown.

[0020] Fig.13 A schematic diagram showing a graphic profile after development and a graphic profile after etching in another embodiment of the present disclosure.

[0021] Fig.14 A schematic diagram showing a graphic profile after development and a graphic profile after etching in yet another embodiment of the present disclosure.

[0022] Fig.15 A schematic diagram showing the distribution of etching deviation in an embodiment of the present disclosure is shown.

[0023] Fig.16 A flowchart of another image processing method provided by an embodiment of the present disclosure is shown.

[0024] Fig.17 A schematic diagram of an image processing device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0025] like Figure 1 As shown, the image processing method provided by the embodiment of the present disclosure may include the following steps.

[0026] In S110 , a post-development pattern (After Developing Inspection, ADI) and a post-etching pattern (After Etching Inspection, AEI) are obtained.

[0027] For example, Figure 2 As shown in Figure 1, the ADI graph is obtained by SEM (scanning electron microscope). Figure 3 Shown is the AEI pattern obtained by SEM.

[0028] Figure 3 A schematic diagram of a pattern after etching in an embodiment of the present disclosure is shown. This figure is a SEM image of AEI, which needs to be processed to obtain the outline of AEI.

[0029] In S120, the developed pattern and the etched pattern are processed to obtain a developed pattern outline and an etched pattern outline, respectively.

[0030] For example, for the above Figure 2 The ADI graph and Figure 3 The AEI graphs shown in the figure are respectively subjected to contour detection processing to obtain ADI contour and AEI contour. Figure 4 As shown, 41 represents the ADI profile and 42 represents the AEI profile.

[0031] In S130, a target etching deviation between each candidate data point on the post-development pattern profile and a corresponding data point on the post-etched pattern profile is obtained.

[0032] In the disclosed embodiment, the ADI profile and the AEI profile are formed by connecting a plurality of data points in series. The candidate data points on the ADI profile may include all the data points on the ADI profile, or may be some of the data points extracted from all the data points on the ADI profile, such as some of the data points extracted at equal intervals or at non-equal intervals. For example, for the straight line portion on the ADI profile, a longer interval can be used to extract data points as candidate data points, and for the non-straight line portion on the ADI profile (such as line end, corner, arc, etc.), a shorter interval can be used to extract data points as candidate data points, thereby reducing the amount of data calculation and ensuring that the etching deviation model trained by the training data set obtained by the extracted candidate data points can have a higher etching deviation prediction accuracy.

[0033] In an exemplary embodiment, obtaining a target etching deviation between each candidate data point on the post-development graphic contour and a corresponding data point on the post-etched graphic contour includes: obtaining a curvature of the candidate data point; drawing a straight line intersecting the post-etched graphic contour along the direction of the curvature of the candidate data point; determining a data point corresponding to the candidate data point from the intersection points on the post-etched graphic contour; and using the distance between the candidate data point and its corresponding data point as the target etching deviation of the candidate data point.

[0034] from Figure 4 It can be seen that the ADI profile and the AEI profile are different. Therefore, in order to accurately locate the candidate data points on the ADI profile and the corresponding data points on the AEI profile for calculating the target etching deviation, the embodiment of the present disclosure calculates the curvature of each candidate data point, and then draws a straight line along the curvature direction of the candidate data point. The straight line may have one or two intersection points with the AEI profile. When there are two intersection points, any one of the intersection points or the intersection point closer to the candidate data point can be selected as the data point corresponding to the candidate data point. Then calculate the distance between the candidate data point on the ADI profile and the corresponding data point on the AEI profile as the target etching deviation of the candidate data point, that is, the real etching deviation. By calculating the curvature of the candidate data point, the embodiment of the present disclosure can accurately determine the candidate data points and corresponding data points on the ADI profile and AEI profile with irregular shapes (for example, corners, arcs, etc.).

[0035] In the embodiment of the present disclosure, the curvature kernel function may be set as follows:

[0036]

[0037] In the above formula, k represents the curvature of a candidate data point on the ADI contour, (x, y) represents the abscissa and ordinate of the candidate data point, respectively, and y' and y" represent the first-order derivative and second-order derivative of y with respect to x, respectively.

[0038] The first-order derivative of y with respect to x can be calculated using the following formula:

[0039]

[0040] In the above formula, (x1, y1) represents the horizontal coordinate and vertical coordinate of a candidate data point located before the candidate data point (x, y) on the ADI profile, and (x2, y2) represents the horizontal coordinate and vertical coordinate of a candidate data point located after the candidate data point (x, y) on the ADI profile, that is, the first-order derivative of the candidate data point is approximately calculated by the slope. It can be understood that the calculation method of the curvature of each candidate data point in the present disclosure is not limited to the above example.

[0041] like Figure 7 As shown, the dashed line 71 represents the ADI profile, the solid line 72 represents the AEI profile, and 73 represents the target etching deviation between each candidate data point on the ADI profile and the corresponding data point on the ADI profile.

[0042] In S140, M candidate features of each candidate data point on the developed pattern contour are obtained, where M is an integer greater than 1. The method for obtaining the M candidate features of each candidate data point on the ADI contour can be, for example, referred to the following Figures 10 to 12 , but the present disclosure does not limit the type and method of obtaining the M candidate features.

[0043] It should be noted that there is no restriction on the execution order of the above S130 and S140 , and they can be executed in parallel, or S130 can be executed first and then S140 , or S140 can be executed first and then S130 .

[0044] In S150, an etching deviation correlation index of each candidate feature is obtained according to the candidate feature of each candidate data point on the post-development pattern contour and its target etching deviation.

[0045] In an exemplary embodiment, an etching deviation correlation index of each candidate feature is obtained based on the candidate feature of each candidate data point on the post-development graphic contour and its target etching deviation, including: processing the M candidate features of each candidate data point through an etching deviation model to obtain an absolute etching deviation prediction value of each candidate data point; adding noise to each candidate feature of each candidate data point respectively, processing the M candidate features after adding noise to each candidate data point through the etching deviation model to obtain a noisy etching deviation prediction value of each candidate feature of each candidate data point; obtaining an etching deviation correlation index of each candidate feature based on the absolute etching deviation prediction value of each candidate data point, the noisy etching deviation prediction value of each candidate feature of each candidate data point, and the target etching deviation of each candidate data point.

[0046] In an exemplary embodiment, the post-development graphic contour includes N candidate data points, where N is an integer greater than 1. Among them, according to the absolute etching deviation prediction value of each candidate data point, the noisy etching deviation prediction value of each candidate feature of each candidate data point and the target etching deviation of each candidate data point, the etching deviation correlation index of each candidate feature is obtained, including: obtaining the absolute etching deviation value of the ith candidate data point according to the absolute etching deviation prediction value of the ith candidate data point and the target etching deviation of the ith candidate data point; i is an integer greater than or equal to 1 and less than or equal to N, and j is an integer greater than or equal to 1 and less than or equal to M; obtaining the noisy etching deviation value of the jth candidate feature of the ith candidate data point according to the noisy etching deviation prediction value of the jth candidate feature of the ith candidate data point and the target etching deviation of the ith candidate data point; obtaining the difference value of the jth candidate feature of the ith candidate data point according to the noisy etching deviation value of the jth candidate feature of the ith candidate data point and the absolute etching deviation value of the ith candidate data point; obtaining the etching deviation correlation index of the jth candidate feature according to the difference value of the jth candidate feature of the N candidate data points.

[0047] The etching deviation model in the embodiments of the present disclosure refers to a model that can be used to predict etching deviation. The etching deviation model can be obtained by training a machine learning model with a training data set. The trained machine learning model can be, for example, an ANN (Artificial Neural Network) model, a GBDT (Gradient Boosting Decision Tree) model, or any other suitable model. The present disclosure does not limit this. In the following embodiments, the GBDT model is used as an example. The effect of integrated tree models such as GBDT is better.

[0048] The M candidate features of each candidate data point are designed based on experience. Some candidate features may have a strong correlation with the etching deviation, while some candidate features may have no correlation or a weak correlation with the etching deviation. The method provided in the embodiment of the present disclosure screens and compresses the M candidate features, and selects M1 target features that have a strong correlation with the etching deviation.

[0049] For example, the M candidate features of each candidate data point on the ADI profile can be input into the GBDT model. The GBDT model processes the M candidate features of each candidate data point, outputs the absolute etching deviation prediction value of each candidate data point, and subtracts the absolute etching deviation prediction value of the i-th candidate data point from the target etching deviation of the i-th candidate data point to obtain the absolute etching deviation value of the i-th candidate data point. The absolute etching deviation values ​​of N candidate data points can be expressed as [errOOB11, errOOB12, …, errOOB1 N ] Among them, errOOB1 i represents the absolute etching deviation prediction value of the i-th candidate data point. The ADI profile here can be all ADI profiles of all ADI graphics in the data set, and N is the number of all candidate data points on all ADI profiles of all ADI graphics in the data set.

[0050] Then randomly select the jth candidate feature F for all samples (one sample is a candidate data point on the ADI contour) j Add noise interference. For example, when no noise is added, the i-th candidate data point is represented as [F1, F2, …F j , …F M ], where F j represents the jth candidate feature in the i-th candidate data point. After adding noise interference to the j-th candidate feature of the i-th candidate data point, it is expressed as [F1, F2, …F j +n j , F j+1 ,…F M ], nj represents the noise added to the jth candidate feature of the i-th candidate data point. j +n j , …F M ] is input into the GBDT model, and the GBDT model calculates [F1, F2, …F j +n j , …F M ] is processed and the noise-added etching deviation value of the jth candidate feature of the i-th candidate data point is output, which is expressed as errOOB2 ij , the noise etching deviation value of the M candidate features of the i-th candidate data point can be expressed as [errOOB2 i1 ,errOOB1 i2 ,…,errOOB1 iM ]. In a similar manner, the noisy etching deviation values ​​of the M candidate features of the N candidate data points can be obtained.

[0051] For example, the etching deviation correlation index FI of the jth candidate feature can be calculated by the following formula:j :

[0052]

[0053] It should be noted that the method for obtaining the etching deviation correlation index of each candidate feature is not limited to the above example, for example, it can also be obtained by the following method:

[0054]

[0055] Alternatively, you can obtain it as follows:

[0056]

[0057] In S160 , M1 target features are determined from the M candidate features according to the etching deviation correlation index of each candidate feature, where M1 is an integer greater than or equal to 1 and less than or equal to M.

[0058] The etching deviation correlation index [FI1, FI2, ..., FI j ,…,FI M ], the etching deviation correlation index can be used to indicate the correlation between the candidate feature and the etching deviation. The larger the etching deviation correlation index is, the greater the correlation between the corresponding candidate feature and the etching deviation is, that is, the more important this candidate feature is for the prediction of the etching deviation. The disclosed embodiment can arrange the M candidate features in ascending or descending order according to the M etching deviation correlation indicators. For example, taking descending order as an example, a predetermined number of candidate features are selected as target features, and the predetermined number can be equal to M1. Alternatively, a predetermined percentage of candidate features can be selected as target features to obtain M1 target features. The predetermined percentage can be set to balance accuracy and computational complexity, and the disclosed embodiment does not limit this.

[0059] The image processing method provided by the embodiment of the present disclosure obtains the target etching deviation between each candidate data point on the ADI profile and the corresponding data point on the AEI profile, and obtains the etching deviation correlation index of each candidate feature according to the M candidate features and their target etching deviations of each candidate data point on the ADI profile, so that M1 target features with strong correlation with the etching deviation can be screened out from the M candidate features according to the etching deviation correlation index of the M candidate features, and the candidate features with weak correlation with the etching deviation can be screened out, thereby reducing the amount of data calculation. When the etching deviation model is subsequently trained and predicted according to the M1 target features of each candidate data point on the ADI profile, on the one hand, the training speed and prediction speed of the etching deviation model can be improved, and the occupied computing resources and network transmission resources can be reduced; on the other hand, the accuracy of etching deviation prediction can be improved, so that the simulation accuracy can be improved during semiconductor process modeling and simulation, and more accurate process simulation results can be obtained.

[0060] The predicted etch bias in the embodiment of the present disclosure may include two types: negative etch bias and positive etch bias. Figure 5 As shown in (a), a photoresist layer is formed on the surface of a substrate 51, and the photoresist layer is patterned according to a photolithography pattern (not shown in the figure), and a mask layer 52 is formed on the surface of the substrate 51 to expose a portion of the surface of the substrate 51. The mask layer 52 or the boundary line of the photolithography pattern is used as the ADI profile. Figure 5 As shown in (b), ions and radicals are injected, and the substrate 51 is etched with the mask layer 52 as a mask, and a corresponding etched groove is formed in the substrate 51, and the boundary line of the etched groove on the surface of the substrate 51 is used as the AEI profile. After the etched groove is formed, the mask layer 52 can be removed to facilitate the subsequent detection of the etched groove and obtain the corresponding AEI profile. The presence of a preset etching area and a non-etching area in the substrate 51 can be determined by the patterned mask layer. The preset etching area is the area in the substrate that is desired to be etched during the actual etching process, and the preset non-etching area is the area in the substrate that is not desired to be etched during the actual etching process. Figure 5 A negative etch bias 53 is formed in the etched groove, that is, the etched groove includes an area that is not desired to be etched. Figure 6 As shown in (a), a mask layer 62 is formed on the surface of the substrate 61 to expose a portion of the surface of the substrate 61. Figure 6 As shown in (b), ions and free radicals are injected, and the substrate 61 is etched using the mask layer 62 as a mask to form corresponding etching grooves in the substrate 61. Figure 6A positive etch bias 63 is formed in the etched groove, that is, the etched groove does not completely include the area desired to be etched.

[0061] There is a loading effect in the etching process, which refers to the effect of decreased etching rate or uneven distribution caused by the consumption of local etching gas being greater than the supply. Loading effects can be divided into three types: macro loading, micro loading, and loading effects related to the aspect ratio of etching (ARDE, also known as aperture effect). When designing each candidate feature of each candidate data point on the ADI profile, the disclosed embodiment can comprehensively consider the loading effect of etching, such as considering the micro loading effect and the aperture effect.

[0062] Figure 8 A schematic diagram showing the aperture effect in the embodiment of the present disclosure. The aperture effect refers to the fact that the width of the spacing between the photolithography pattern or the mask layer will affect the etching depth. Figure 8 As shown in (a), it is assumed that different photolithography patterns 81 have the same line width W, but different spacings S. Figure 8 (b) It can be seen that the etching depth corresponding to the photolithography pattern with a wide pitch is deeper, and the etching depth corresponding to the photolithography pattern with a narrow pitch is shallower.

[0063] Fig. 9 Schematic diagram showing the micro-loading effect in the embodiment of the present disclosure. The micro-loading effect refers to the density of the photolithography pattern affecting the etching depth. Fig. 9 As shown in (a), it is assumed that the lithographic patterns 91 on the left have the same pitch and line width W2, and the lithographic patterns on the right have the same pitch and line width, but the density of the lithographic patterns on the left and right sides is different, the lithographic patterns on the left are dense, and the lithographic patterns on the right are sparse. Fig. 9 (b) It can be seen that the etching depth in the pattern-dense area is smaller than that in the pattern-sparse area.

[0064] Fig.10 A schematic diagram of the density kernel function in the embodiment of the present disclosure is shown. Fig.10 As shown in the figure, we can select any point of interest from the ADI profile as a candidate data point, and then draw a circle with the point of interest as the center and r1 as the radius (the area within the circle is the interaction range of the point of interest), and calculate the litho pattern (e.g. Fig.10 The area ratio of the black rectangle in the figure to the circle is the local density of the point of interest. The radius r1 can be set based on experience, and the present disclosure does not limit this. The local density of the candidate data point on the ADI profile can be obtained through the density kernel function designed in the embodiment of the present disclosure, and the local density can be used as a candidate feature of the candidate data point.

[0065] Fig.11 Schematic diagram showing the visible kernel function in the embodiment of the present disclosure. Fig.11 As shown in Figure 1, we can select any point of interest from the ADI contour as a candidate data point, and then draw a circle with the point of interest as the center and r2 as the radius. During the circle drawing process, if it is blocked by the photolithography pattern, the drawing is stopped, and the following can be obtained: Fig.11 The interaction range of the point of interest shown is calculated, and the area ratio of the interaction range to the circle is calculated to be the visibility of the point of interest. The radius r2 can be set based on experience, and the present disclosure does not limit this. The visibility of the candidate data point on the ADI contour can be obtained through the visible kernel function designed in the embodiment of the present disclosure, and the visibility can be used as a candidate feature of the candidate data point.

[0066] Fig.12 FIG. 1 is a schematic diagram showing an occlusion kernel function in an embodiment of the present disclosure. Fig.12 As shown, you can arbitrarily select an interest point from the ADI contour as a candidate data point, and then draw a circle with the interest point as the center and r3 as the radius. The overlapping area of ​​the circle and the lithography pattern corresponding to the ADI contour where it is located is the interaction range of the interest point. The area ratio of the interaction range to the circle is calculated, which is the occlusion degree of the interest point. The radius r3 can be set based on experience, and the present disclosure does not limit this. The occlusion kernel function designed by the embodiment of the present disclosure can obtain the occlusion degree of the candidate data point on the ADI contour, and the occlusion degree can be used as a candidate feature of the candidate data point. It should be noted that Figures 10 to 12 The three kernel functions shown are only used to illustrate how to design candidate features of candidate data points in the present disclosure, but the present disclosure is not limited thereto.

[0067] Fig.13 and Fig.14 Schematic diagrams of the ADI profile and the AEI profile are shown. The dotted line represents the ADI profile, and the solid line represents the AEI profile. Fig.13 and Fig.14 And combined with the above Figure 7It can be seen that the shapes of ADI profiles and AEI profiles are various, depending on the actual design requirements of semiconductor devices. At the same time, most of the ADI profiles and AEI profiles are straight lines or approximate straight lines, and a small number are non-straight lines such as line ends, corners, and arcs. At the same time, the line ends and corners have large etch bias during etching. This will result in many more candidate data points for the straight line part than for the non-straight line part. This data imbalance will cause the etching bias model to be sensitive to training data during the training of the etching bias model, and the effects will vary greatly for different training data, resulting in model overfitting. The trained etching bias model has a better prediction effect on the etching bias of the straight line part of the ADI profile, but a poor effect on the etching bias of the non-straight line part of the ADI profile.

[0068] In an exemplary embodiment, the method provided by the embodiment of the present disclosure also includes: determining a target data point whose target etching deviation is greater than an etching deviation threshold from the candidate data points; performing interpolation processing on the target data point to obtain an interpolated data point; obtaining an interpolated etching deviation between the interpolated data point and a corresponding data point on the post-etching graphic contour; placing the target data point and its target etching deviation and the interpolated data point and its interpolated etching deviation into a training data set; and using the training data set to train to obtain an etching deviation model.

[0069] Assuming that the etching bias threshold is 6nm, interpolation is performed near candidate data points with a target etch bias greater than 6nm to ensure that there are enough samples for candidate data points with few samples. For example, there are 20 candidate data points at the line end. In order to make more small sample data, interpolation is performed among the 20 points to obtain 30 points, and the interpolated etching bias of the corresponding interpolated data points is calculated and added to the training data set.

[0070] In the embodiments of the present disclosure, the etching deviation threshold can be set according to actual needs. For example, the mean of the actual etching deviations of the candidate data points of all straight line parts on the ADI profile can be statistically obtained and the mean is used as the etching deviation threshold. However, the present disclosure does not limit the setting method of the etching deviation threshold. For example, the etching deviation threshold can also be set based on experience. When the target etching deviation of a candidate data point on the ADI profile is greater than the etching deviation threshold, it can be determined as a data point of the non-linear part of the ADI profile, and the candidate data points of these non-linear parts are extracted as target data points, and the target data points are interpolated, that is, the candidate data points of the non-linear part such as the line end and the corner part are interpolated to obtain interpolated data points, and the target data points of the non-linear part and the interpolated data points are used as a training data set, thereby expanding the sample data of the non-linear parts such as the line end and the corner, balancing the sample data of the non-linear part and the sample data of the linear part, so that the etching deviation model obtained by training based on the training data set not only has a good prediction effect on the etching deviation of the linear part of the ADI profile, but also has a good prediction effect on the etching deviation of the non-linear part of the ADI profile, solving the problem that the etching deviation model is sensitive to data and preventing the etching deviation model from having a long tail effect during training. In addition, through interpolation processing, the sample data is increased, more sample data is obtained, and using more sample data to train the etching deviation model can also solve the overfitting problem, because more samples can allow the etching deviation model to learn more and more effective features.

[0071] In the embodiment of the present disclosure, placing the target data point and the interpolation data point into the training data set may be placing M candidate features of the target data point and the interpolation data point or M1 target features into the training data set. When M1 target features are placed into the training data set, compared with placing M candidate features, the complexity of the etching deviation model and the amount of computation during model training can be reduced, the training efficiency and speed can be improved, and the prediction accuracy of the etching deviation can also be guaranteed.

[0072] In an exemplary embodiment, obtaining an interpolated etching deviation between the interpolated data point and a corresponding data point on the post-etching graphic contour includes: obtaining the curvature of the interpolated data point; drawing a straight line intersecting the post-etching graphic contour along the direction of the curvature of the interpolated data point; determining a data point corresponding to the interpolated data point from the intersection points on the post-etching graphic contour; and using the distance between the interpolated data point and its corresponding data point as the interpolated etching deviation of the interpolated data point.

[0073] In the disclosed embodiment, for the interpolation data points of the non-straight portion on the ADI profile, it is necessary to determine the data points on the AEI profile corresponding to the interpolation data points. The curvature of each interpolation data point can be calculated, and a straight line intersecting the AEI profile can be drawn along the curvature direction. The intersection point may be one or two. When there are two, any one or the intersection point closest to the interpolation data point can be selected as the corresponding data point of the interpolation data point, and the distance between the interpolation data point and the corresponding data point on the AEI profile is calculated as the interpolation etching deviation of the interpolation data point, thereby eliminating the need for interpolation processing of the non-straight portion on the AEI profile.

[0074] Fig.15 A schematic diagram showing the distribution of etching deviation in an embodiment of the present disclosure is shown. Fig.15 The horizontal axis in is the value of the etching deviation, in nm (nanometer), and the vertical axis is the number of candidate data points on the ADI profile whose etching deviation falls within the corresponding interval. Fig.15 It can be seen that the distribution of candidate data points within the value range of each etching deviation is not balanced and has large differences.

[0075] In an exemplary embodiment, the method provided by the embodiment of the present disclosure may also include: clustering the candidate data points according to the feature vectors of the candidate data points, the feature vectors of the candidate data points including M candidate features or M1 target features; performing a first uniform sampling on the candidate data points in each clustering category, and placing the first uniformly sampled candidate data points and their target etching deviations into a training data set; performing a second uniform sampling on the remaining candidate data points after the first uniform sampling according to the target etching deviation partitions of the remaining candidate data points, and placing the second uniformly sampled candidate data points and their target etching deviations into the training data set; and using the training data set to train to obtain an etching deviation model.

[0076] In the embodiment of the present disclosure, the k-means algorithm can be used to cluster the candidate data points on the ADI profile, and k (k is a positive integer greater than 1, k can be set according to actual needs, and the present disclosure does not limit it) initial centroids are selected as initial clusters (cluster, i.e., cluster categories); for each candidate data point's feature vector, the nearest centroid is calculated, and its cluster category is marked as the cluster corresponding to the centroid; the centroids corresponding to the k clusters are recalculated; and the execution is repeated until the centroids no longer change to obtain k cluster categories. It can be understood that the present disclosure does not limit the algorithm used for clustering processing. In the following embodiments, k-means is used as an example, but it is not limited to this.

[0077] In the disclosed embodiment, all candidate data points on the ADI profile are k-means clustered with their respective feature vectors, and then the candidate data points (i.e., samples) in each cluster category are uniformly sampled. For example, assuming that there are a total of 1,000,000 candidate data points and there are 10 cluster categories after k-means clustering, 500 candidate data points are sampled from each of the 10 cluster categories, and the feature vectors and target etching deviations of the 5,000 candidate data points obtained by the first sampling are put into the training data set as training samples. Through the first uniform sampling, on the one hand, the amount of data can be reduced to speed up the training speed of the etching deviation model; on the other hand, it can ensure that the proportion of samples for each category is relatively uniform, preventing the etching deviation model from having a good effect on a certain type of graphics in the ADI profile and a poor effect on other graphics.

[0078] Then the remaining candidate data points that have not been sampled for the first time are distributed according to the target etch bias histogram (e.g. Fig.15 The histogram distribution shown in the figure is uniformly sampled, for example, 500 samples are randomly taken from the remaining candidate data points with target etch bias values ​​between 0.1-0.3 nanometers, 500 samples are sampled from the remaining candidate data points with target etch bias values ​​between 0.3-0.6nm, and so on. For samples, some samples have large target etch bias and some have small target etch bias. In the training data, the data amounts of samples with different target etch bias are different, which will cause the etching bias model obtained through training to be inclined towards samples with larger data amounts. For example, 80% of the target etch bias in the training data are between 0.1-1.1nm, so it may be difficult for the model to learn the values ​​of the target etch bias in other intervals. Therefore, the embodiment of the present disclosure divides the target etch bias value range, for example, taking an interval of 0.4nm as an interval, and sampling each interval, such as 0.0-0.4nm, 0.4-0.8nm, and 0.8-1.2nm intervals, so that the number of samples in each target etching bias interval can be more balanced, thereby improving the prediction accuracy of the etching bias model.

[0079] In the embodiment of the present disclosure, for the verification data set, in order to keep it consistent with the actual application scenario, no data interpolation and sampling processing may be performed to ensure that the test results of the verification data set can truly reflect the application effect of the etching deviation model.

[0080] The method provided by the embodiment of the present disclosure analyzes ADI and AEI, considers the micro-load effect and aperture effect for each candidate data point of the ADI contour, uses empirical knowledge to generate a series of candidate features, and selects target features that are strongly related to etching bias through feature compression; then, the generated input domain and value range space are sampled and balanced, and finally the target features of the processed samples and the target etching bias are combined into a training data set and a verification data set, and a regression model is trained by GBDT as an etching bias model. The etching bias model obtained by training in the above manner has a high accuracy, and the predicted AEI contour pattern is highly similar to the real AEI contour pattern. The input domain here refers to ADI graphics with different characteristics, and the value range refers to data pairs (ADI & -AEI) with different etch biases, that is, the training data is both diverse in type and the distribution of various etch biases is relatively uniform. In order to obtain an etching bias model that is easier to train and more robust.

[0081] like Fig.16 As shown, the image processing method provided by the embodiment of the present disclosure may include the following steps.

[0082] In S201, a post-development pattern profile and a post-etching pattern profile are obtained.

[0083] In the disclosed embodiment, in the data preparation stage, the ADI profile and the AEI profile are obtained. The ADI profile can be obtained by directly taking a photo and extracting the profile in the wafer, or it can be generated by simulation using a photolithography model. The AEI profile can be obtained by taking a photo on the wafer and extracting the contour.

[0084] In S202, a target etching deviation between each candidate data point on the post-development pattern profile and a corresponding data point on the post-etched pattern profile is obtained.

[0085] For each candidate data point on the ADI profile, the target etching deviation with respect to the corresponding data point on the AEI profile is calculated and obtained as the actual etching deviation, which is used as a label when the etching deviation model is trained.

[0086] In S203, M candidate features are obtained for each candidate data point on the contour of the developed graphic.

[0087] In the disclosed embodiment, multiple kernel functions such as density kernel function, visibility kernel function, occlusion kernel function, curvature kernel function, etc. can be designed through artificial experience combined with micro-load effect and aperture effect, and M candidate features are calculated for each candidate data point on the ADI contour as the feature vector of the candidate data point.

[0088] In S204, feature compression is performed on the M candidate features of each candidate data point on the contour of the developed pattern, and M1 target features that are strongly correlated with the etching deviation are screened.

[0089] In S205, sample balancing is performed on the candidate data points having M1 target features on the contour of the developed image to obtain a training data set.

[0090] In S206 , an etching deviation model is obtained by training using the training data set.

[0091] In S207 , M1 target features of each data point on the contour of the pattern to be predicted after development are input into the etching deviation model to obtain the etching deviation prediction value of each data point.

[0092] In S208, each data point on the contour of the pattern to be predicted after etching is determined according to each data point on the contour of the pattern to be predicted after development and its predicted time deviation value.

[0093] In S209, the data points on the contour of the pattern to be predicted after etching are connected to obtain the contour of the pattern to be predicted after etching.

[0094] In the disclosed embodiment, the prepared training data set is trained using the GBDT model, and the error between the target etch bias and the predicted bias (etching deviation prediction value) of the samples in the training data set is used as the loss. The GBDT model parameters are updated by back propagation until the model converges and the error reaches the minimum value and no longer decreases. After maintaining the model parameters, when it is necessary to predict the AEI contour, the feature vector corresponding to each data point on the ADI contour to be predicted is input into the trained etching deviation model to obtain the etching deviation prediction value, and the data points on the ADI contour after being offset based on the etching deviation prediction value are connected in series as a contour to obtain the AEI contour to be predicted.

[0095] The method provided by the embodiment of the present disclosure selects target features through feature compression. The target features are strongly correlated with the etching deviation model, thereby reducing the computational complexity and amount of computation, making the training of the etching deviation model easier and shortening the training time, while improving the prediction accuracy of the etching deviation; then sampling and interpolating the samples solves the problem of unbalanced sample data, prevents overfitting of the model, and improves the prediction accuracy of the etching deviation.

[0096] The embodiment of the present disclosure adopts a feature screening method to prevent the unreasonable factors existing in the feature kernel function designed by artificial experience from affecting the model effect. At the same time, the data is uniformly sampled considering the graphic features to prevent the model from overfitting due to the long-tail effect. The trained etching deviation model has better robustness and is closer to the real graphics in special graphics prediction. Based on the same inventive concept, an image processing device is also provided in the embodiment of the present disclosure, as described in the following embodiments. Since the principle of solving the problem in the embodiment of the device is similar to that in the above-mentioned method embodiment, the implementation of the embodiment of the device can refer to the implementation of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0097] Fig.17 A schematic diagram of an image processing device in an embodiment of the present disclosure is shown. Fig.17 As shown, the image processing device 170 may include an acquisition unit 171, a processing unit 172, and a determination unit 173. The acquisition unit 171 is used to obtain a post-development graphic and a post-etching graphic. The processing unit 172 is used to process the post-development graphic and the post-etching graphic to obtain a post-development graphic contour and a post-etching graphic contour, respectively. The acquisition unit 171 is also used to obtain a target etching deviation between each candidate data point on the post-development graphic contour and a corresponding data point on the post-etching graphic contour. The acquisition unit 171 is also used to obtain M candidate features for each candidate data point on the post-development graphic contour, where M is an integer greater than 1. The acquisition unit 171 is also used to obtain an etching deviation correlation index of each candidate feature based on the candidate feature of each candidate data point on the post-development graphic contour and its target etching deviation. The determination unit 173 is used to determine M1 target features from the M candidate features based on the etching deviation correlation index of each candidate feature, where M1 is an integer greater than or equal to 1 and less than or equal to M.

[0098] The computer device in the embodiment of the present disclosure may include: one or more processors, memory and input / output interface. The processor, memory and input / output interface are connected via a bus. The memory is used to store a computer program, which includes program instructions; the input / output interface is used to receive data and output data, such as for data exchange between a host machine and a computer device, or for data exchange between various virtual machines in a host machine; the processor is used to execute the program instructions stored in the memory.

[0099] Among them, the processor can perform the following operations: obtain a post-development graphic and a post-etching graphic; process the post-development graphic and the post-etching graphic to obtain a post-development graphic contour and a post-etching graphic contour respectively; obtain a target etching deviation between each candidate data point on the post-development graphic contour and a corresponding data point on the post-etching graphic contour; obtain M candidate features of each candidate data point on the post-development graphic contour, where M is an integer greater than 1; obtain an etching deviation correlation index of each candidate feature based on the candidate feature of each candidate data point on the post-development graphic contour and its target etching deviation; determine M1 target features from the M candidate features based on the etching deviation correlation index of each candidate feature, where M1 is an integer greater than or equal to 1 and less than or equal to M.

[0100] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor and the input and output interface. A portion of the memory may also include a non-volatile random access memory. In a specific implementation, the computer device may execute the implementation methods provided by the various steps in any of the above method embodiments through its built-in functional modules. For details, please refer to the implementation methods provided by the various steps in the figure shown in the above method embodiments, which will not be repeated here.

[0101] The embodiments of the present disclosure provide a computer device, including a processor, an input / output interface, and a memory. The processor obtains a computer program in the memory to execute each step of the method shown in any of the above embodiments.

[0102] The embodiment of the present disclosure also provides a computer-readable storage medium, which stores a computer program, and the computer-readable storage medium stores a program product capable of implementing the above method of the present disclosure. The computer program is suitable for being loaded by the processor and executing the image processing method provided in each step of any of the above embodiments.

[0103] The present disclosure also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the method provided in any of the above-mentioned optional modes.

Claims

1. An image processing method, characterized in that: include: Obtaining a developed pattern and an etched pattern; Processing the developed pattern and the etched pattern to obtain a developed pattern outline and an etched pattern outline respectively; Obtaining a target etching deviation between each candidate data point on the post-development graphic contour and a corresponding data point on the post-etched graphic contour; Obtaining M candidate features for each candidate data point on the contour of the developed graphic, where M is an integer greater than 1; Obtaining an etching deviation correlation index for each candidate feature according to the candidate feature and the target etching deviation of each candidate data point on the post-development graphic contour; According to the etching deviation correlation index of each candidate feature, M1 target features are determined from the M candidate features, where M1 is an integer greater than or equal to 1 and less than or equal to M.

2. The method according to claim 1, characterized in that According to the candidate feature of each candidate data point on the post-development graphic contour and its target etching deviation, an etching deviation correlation index of each candidate feature is obtained, including: Processing the M candidate features of each candidate data point through the etching deviation model to obtain an absolute etching deviation prediction value of each candidate data point; Adding noise to each candidate feature of each candidate data point respectively, processing the M candidate features after adding noise to each candidate data point through the etching deviation model, and obtaining the noisy etching deviation prediction value of each candidate feature of each candidate data point; According to the absolute etching deviation prediction value of each candidate data point, the noisy etching deviation prediction value of each candidate feature of each candidate data point and the target etching deviation of each candidate data point, the etching deviation correlation index of each candidate feature is obtained.

3. The method according to claim 2, characterized in that The developed graphic contour includes N candidate data points, where N is an integer greater than 1; According to the absolute etching deviation prediction value of each candidate data point, the noisy etching deviation prediction value of each candidate feature of each candidate data point, and the target etching deviation of each candidate data point, the etching deviation correlation index of each candidate feature is obtained, including: Obtaining an absolute etching deviation value of the i-th candidate data point according to the absolute etching deviation prediction value of the i-th candidate data point and the target etching deviation of the i-th candidate data point; i is an integer greater than or equal to 1 and less than or equal to N, and j is an integer greater than or equal to 1 and less than or equal to M; Obtaining a noisy etching deviation value of the jth candidate feature of the ith candidate data point according to the noisy etching deviation prediction value of the jth candidate feature of the ith candidate data point and the target etching deviation of the ith candidate data point; Obtaining a difference value of the jth candidate feature of the ith candidate data point according to the noised etching deviation value of the jth candidate feature of the ith candidate data point and the absolute etching deviation value of the ith candidate data point; According to the difference value of the j-th candidate feature of the N candidate data points, the etching deviation correlation index of the j-th candidate feature is obtained.

4. The method according to claim 1, characterized in that Obtaining a target etching deviation between each candidate data point on the post-development graphic contour and a corresponding data point on the post-etched graphic contour, comprising: Obtaining the curvature of the candidate data point; Draw a straight line intersecting the outline of the etched pattern along the direction of the curvature of the candidate data point; Determine a data point corresponding to the candidate data point from the intersection points on the etched graphic contour; The distance between the candidate data point and its corresponding data point is used as the target etching deviation of the candidate data point.

5. The method according to claim 1, characterized in that Also includes: Determining a target data point whose target etching deviation is greater than an etching deviation threshold value from the candidate data points; Performing interpolation processing on the target data points to obtain interpolation data points; Obtaining an interpolated etching deviation between the interpolated data point and a corresponding data point on the post-etching graphic contour; Putting the target data point and its target etching deviation and the interpolated data point and its interpolated etching deviation into a training data set; The training data set is used to train and obtain an etching deviation model.

6. The method according to claim 5, characterized in that Obtaining an interpolated etching deviation between the interpolated data point and a corresponding data point on the post-etching graphic contour, comprising: Obtaining the curvature of the interpolation data point; Draw a straight line intersecting the outline of the etched pattern along the direction of the curvature of the interpolation data point; Determining data points corresponding to the interpolation data points from the intersection points on the etched graphic contour; The distance between the interpolation data point and its corresponding data point is used as the interpolation etching deviation of the interpolation data point.

7. The method according to any one of claims 1 to 6, characterized in that: Also includes: Clustering the candidate data points according to the feature vectors of the candidate data points, where the feature vectors of the candidate data points include M candidate features or M1 target features; Performing a first uniform sampling of candidate data points in each cluster category, and putting the first uniformly sampled candidate data points and their target etching deviations into a training data set; For the remaining candidate data points after the first uniform sampling, performing a second uniform sampling according to the target etching deviation partitions of the remaining candidate data points, and placing the candidate data points and their target etching deviations obtained by the second uniform sampling into the training data set; The training data set is used to train and obtain an etching deviation model.

8. An image processing device, characterized in that: include: An obtaining unit, used for obtaining a developed pattern and an etched pattern; A processing unit, used for processing the developed pattern and the etched pattern to obtain a developed pattern outline and an etched pattern outline respectively; The obtaining unit is further used to obtain a target etching deviation between each candidate data point on the post-development graphic contour and a corresponding data point on the post-etched graphic contour; The obtaining unit is further used to obtain M candidate features for each candidate data point on the contour of the developed graphic, where M is an integer greater than 1; The obtaining unit is further used to obtain an etching deviation correlation index of each candidate feature according to the candidate feature and the target etching deviation of each candidate data point on the post-development graphic contour; The determination unit is used to determine M1 target features from M candidate features according to the etching deviation correlation index of each candidate feature, where M1 is an integer greater than or equal to 1 and less than or equal to M.

9. A computer device, characterized in that: include: one or more processors; The memory is configured to store one or more programs, and when the one or more programs are executed by the one or more processors, the computer device implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed on a computer, the computer is caused to execute the method according to any one of claims 1 to 7.

Images