Method and apparatus for optical proximity correction of ion implantation layers
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAXINCHENG (HANGZHOU) TECH CO LTD
- Filing Date
- 2026-06-23
- Publication Date
- 2026-07-21
Smart Images
Figure CN122431055A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor integrated circuit manufacturing, and in particular to an optical proximity correction method and apparatus for ion implantation layers. Background Technology
[0002] At advanced nodes in large-scale integrated circuit manufacturing (such as 7nm and below), the resolution of photolithography processes is approaching the optical limit. Optical proximity effect can cause significant deviations between the lithographic wafer profile and the target pattern. To solve this problem, model-based optical proximity correction (OPC) must be performed on the mask data to ensure that the lithographic wafer profile matches the design target.
[0003] However, the photolithography process for ion implantation layers has unique characteristics. Due to process integration limitations and cost control requirements, ion implantation layers cannot introduce a bottom anti-reflective coating (BARC) to eliminate the influence of the underlying pattern, as is done with critical layers. Instead, a top anti-reflective coating (TARC) can be introduced above the photoresist, which can only mitigate, not completely eliminate, the interference of the underlying pattern on the light intensity distribution within the photoresist. In actual production, wafer surfaces often have complex morphologies containing silicon and silicon oxide. The size, spacing, and relative position of the underlying pattern to the ion implantation layer can significantly alter the light intensity distribution within the photoresist, leading to critical dimension deviations and three-dimensional morphology anomalies after photolithography, and even device failure.
[0004] In existing technologies, two main approaches are used to address the aforementioned problems. One approach is compensation based on geometric rules, which involves pre-correcting the target pattern according to empirical rules before running traditional OPC. However, this method can only handle simple one-dimensional graphics. For complex two-dimensional graphics or fine structures in advanced processes, the correction accuracy cannot meet the process requirements. The other approach is rigorous optical simulation, which uses the finite-difference time-domain (FDTD) method or rigorous coupled-wave analysis (RCWA) to calculate the three-dimensional light intensity distribution and accurately simulate the influence of the underlying pattern. Although this method has higher accuracy, the computational load is extremely large, and it can only handle patterns in the micrometer range, making it unsuitable for full-chip OPC with a size exceeding 10 mm.
[0005] Therefore, how to improve the optical proximity correction accuracy of the ion implantation layer while ensuring operating speed is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0006] The purpose of this invention is to provide an optical proximity correction method and apparatus for ion implantation layers, so as to solve the problems of low accuracy, slow speed and inability to be applied to large-size full-chip applications in the prior art.
[0007] To address the aforementioned technical problems, this invention provides an optical proximity correction method for ion-implanted layers, comprising:
[0008] Step 1: Receive the full chip layout to be corrected; the full chip layout to be corrected includes the ion implantation layer pattern and the underlying pattern;
[0009] Step 2: Divide the ion implantation layer pattern into edge segments and place control points at the center of each edge segment;
[0010] Step 3: Based on the parameter requirements of the pre-trained optical proximity correction convolutional neural network model, the full chip layout to be corrected is divided into several correction units.
[0011] Step 4: Based on the current ion implantation layer mask pattern, the ion implantation optical image corresponding to each correction unit is obtained using a pre-trained two-dimensional optical proximity correction simulation model; wherein, the initial current ion implantation layer mask pattern is obtained according to the ion implantation layer pattern.
[0012] Step 5: Rasterize the lower-layer pattern of each correction unit to obtain the corresponding lower-layer rasterized image;
[0013] Step 6: Input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image;
[0014] Step 7: Use a preset threshold to extract the simulated contour of the corresponding correction unit from the output optical image;
[0015] Step 8: Calculate the edge placement error corresponding to each edge segment based on the simulated contour;
[0016] Step 9: Calculate the distance that the corresponding edge segment needs to be moved based on the edge placement error, and move the edge segment to obtain the updated current ion implantation layer mask pattern;
[0017] Step 10: Repeat steps 4 to 9 until the edge placement error of all edge segments is less than the preset threshold, and obtain the final optical proximity correction mask pattern of each correction unit.
[0018] Step 11: Stitch together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern.
[0019] Optionally, in the optical proximity correction method for the ion implantation layer, edge segmentation of the ion implantation layer pattern includes:
[0020] On the edge of the ion implantation layer pattern, a first dividing point is placed based on the projection position of the endpoint of the corresponding adjacent lower layer pattern;
[0021] A second cutting point is placed on the edge of the ion implantation layer pattern based on the first cutting point and the preset critical distance.
[0022] Optionally, in the optical proximity correction method for the ion implantation layer, the underlying pattern includes at least one of an active layer pattern and a polysilicon layer pattern.
[0023] Optionally, in the optical proximity correction method for the ion implantation layer, the initial current ion implantation layer mask pattern is the ion implantation layer pattern.
[0024] Optionally, in the optical proximity correction method for the ion implanted layer, calculating the distance that the corresponding edge segment needs to be moved based on the edge placement error includes:
[0025] The distance that the corresponding edge segment needs to be moved is calculated based on the edge placement error using the following formula:
[0026] D = -f × (1 / MEF) × EPE;
[0027] Where f is the feedback coefficient, MEF is the mask error amplification coefficient of the corresponding edge segment, and EPE is the edge placement error of the corresponding edge segment.
[0028] Optionally, in the optical proximity correction method for the ion implantation layer, the optical proximity correction convolutional neural network model is established and trained through the following steps:
[0029] A multi-input single-output convolutional neural network model is established, comprising a first input path, a second input path, and an output path. The first input path is used to receive an ion-implanted optical image obtained based on an ion-implanted layer mask pattern and using a pre-trained two-dimensional optical proximity correction simulation model. The second input path is used to receive a lower-layer rasterized image corresponding to the lower-layer pattern. The output path is used to output a corrected optical image that takes into account the influence of the lower-layer pattern.
[0030] Multiple sets of training samples are acquired, each set of training samples corresponds to a modeling pattern, including a set of input image pairs and a target image; the input image pairs include an ion implantation optical image obtained based on the ion implantation layer pattern in the modeling pattern and a pre-trained two-dimensional optical proximity correction simulation model, and a lower-layer rasterized image of the corresponding lower-layer pattern in the modeling pattern; the target image is the actual photolithographic optical image of the corresponding modeling pattern.
[0031] Using the target image as a reference, the parameters of the convolutional neural network model are trained by minimizing the deviation between the output image of the convolutional neural network model and the target image.
[0032] Optionally, in the optical proximity correction method for the ion implantation layer, the method for obtaining the training samples includes:
[0033] Design a modeling pattern and manufacture a test mask containing the modeling pattern; the modeling pattern includes an ion implantation layer pattern and a corresponding lower layer pattern, including the line width and spacing of different ion implantation layer patterns, the line width and spacing of different lower layer patterns, and a geometric combination of the spacing between different ion implantation layer patterns and the lower layer patterns.
[0034] The test mask is used to perform photolithography to obtain a photolithographic wafer, and the key dimensions of each modeling pattern on the photolithographic wafer are measured.
[0035] A rigorous optical model is established, and a three-dimensional rigorous simulation optical image of each modeled pattern is calculated using the rigorous optical model, and the optical image after vertical diffusion is also calculated.
[0036] For the vertically diffused optical image, the target plane position, target vertical diffusion length, and target contour threshold are determined based on the measurement key dimensions of the corresponding modeling pattern.
[0037] The target image is obtained by intensity scaling of the two-dimensional rigorous simulation optical image of each modeling pattern at the corresponding target plane position; the simulation key dimension of the target image under the target contour threshold is consistent with the corresponding measurement key dimension value.
[0038] Optionally, in the optical proximity correction method for the ion implantation layer, calculating the optical image after vertical diffusion includes:
[0039] The optical image after vertical diffusion is calculated using the following formula:
[0040] I d (x,y,z)=I(x,y,z) exp(-z² / L²);
[0041] Where Id(x,y,z) is the optical image after vertical diffusion, I(x,y,z) is the three-dimensional rigorously simulated optical image, L is the vertical diffusion length, and z is the spatial coordinate in the thickness direction. This is a convolution operation.
[0042] Optionally, in the optical proximity correction method for the ion implantation layer, intensity scaling is performed on the two-dimensional rigorous simulation optical image of each modeled pattern at the corresponding target plane position to obtain the target image, including:
[0043] For each of the modeled patterns, the two-dimensional rigorous simulation optical image at the corresponding target plane position is intensity scaled according to the following formula to obtain the target image:
[0044] ;
[0045] Where I0(x,y) is the target image, and s is the scaling factor. This refers to the two-dimensional rigorously simulated optical image.
[0046] An optical proximity correction device for an ion-implanted layer includes:
[0047] The receiving module is used in step one to receive the full chip layout to be corrected; the full chip layout to be corrected includes an ion implantation layer pattern and a lower layer pattern;
[0048] The segmentation module is used in step two to segment the ion implantation layer pattern into edge segments and place control points at the center of each edge segment.
[0049] The partitioning module is used in step three to divide the full chip layout to be corrected into several correction units according to the parameter requirements of the pre-trained optical proximity correction convolutional neural network model.
[0050] The two-dimensional OPC module is used in step four to obtain the ion implantation optical image corresponding to each correction unit based on the current ion implantation layer mask pattern and using a pre-trained two-dimensional optical proximity correction simulation model; wherein, the initial current ion implantation layer mask pattern is obtained according to the ion implantation layer pattern.
[0051] The rasterization module is used in step five to rasterize the lower-layer pattern of each of the correction units to obtain the corresponding lower-layer rasterized image.
[0052] The convolutional neural module is used in step six to input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image.
[0053] The contour cropping module is used in step seven to crop the simulated contour of the corresponding correction unit in the output optical image using a preset threshold.
[0054] The EPE module is used in step eight to calculate the edge placement error corresponding to each edge segment based on the simulated contour.
[0055] The edge segment adjustment module is used in step nine to calculate the distance that the corresponding edge segment needs to be moved based on the edge placement error, and to move the edge segment to obtain the updated current ion implantation layer mask pattern.
[0056] The loop module is used in step ten to loop steps four to nine until the edge placement error of all edge segments is less than a preset threshold, so as to obtain the final optical proximity correction mask pattern of each correction unit.
[0057] The stitching module is used in step eleven to stitch together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern.
[0058] The present invention provides an optical proximity correction method for an ion implantation layer, comprising the following steps: Step 1, receiving a full-chip layout to be corrected, the full-chip layout including an ion implantation layer pattern and a lower layer pattern; Step 2, segmenting the ion implantation layer pattern into edge segments and placing control points at the center of each edge segment; Step 3, dividing the full-chip layout to be corrected into several correction units according to the parameter requirements of a pre-trained optical proximity correction convolutional neural network model; Step 4, obtaining the ion implantation optical image corresponding to each correction unit based on the current ion implantation layer mask pattern using a pre-trained two-dimensional optical proximity correction simulation model; wherein the initial current ion implantation layer mask pattern is obtained based on the ion implantation layer pattern; Step 5, rasterizing the lower layer pattern of each correction unit to obtain the corresponding lower layer rasterized image. Step 6: Input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image; Step 7: Use a preset threshold to extract the simulated contour of the corresponding correction unit from the output optical image; Step 8: Calculate the edge placement error corresponding to each edge segment based on the simulated contour; Step 9: Calculate the distance the corresponding edge segment needs to move according to the edge placement error, and move the edge segment to obtain the updated current ion implantation layer mask pattern; Step 10: Repeat steps 4 to 9 until the edge placement error of all edge segments is less than the preset threshold to obtain the final optical proximity correction mask pattern of each correction unit; Step 11: Stitch together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern.
[0059] This invention utilizes a dual-input (ion-implanted optical image and lower-layer rasterized image) convolutional neural network model to fully learn the influence of the lower-layer pattern on the lithographic intensity of the ion-implanted layer. This significantly improves the inference speed of optical proximity correction, enhances simulation accuracy, optimizes the lithography process window, and reduces potential defects. Furthermore, by combining separate correction units with subsequent stitching techniques, OPC for the entire chip-scale ion-implanted layer can be completed in a short time, broadening the application scenarios of this invention, meeting mass production requirements, and allowing direct integration into existing semiconductor manufacturing mask data generation pipelines. This results in good compatibility and low production line modification costs. This invention also provides an optical proximity correction device for ion-implanted layers with the aforementioned beneficial effects. Attached Figure Description
[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0061] Figure 1 A flowchart illustrating a specific embodiment of the optical proximity correction method for ion implantation layers provided by the present invention;
[0062] Figure 2 A partial schematic diagram of the full-chip layout to be corrected, representing a specific embodiment of the optical proximity correction method for ion implantation layers provided by the present invention.
[0063] Figure 3 The flowchart of the establishment and training of the optical proximity correction convolutional neural network model of a specific embodiment of the optical proximity correction method for ion implantation layers provided by the present invention is shown below.
[0064] Figure 4 A flowchart illustrating a method for obtaining training samples in a specific embodiment of the optical proximity correction method for ion implantation layers provided by the present invention.
[0065] Figure 5 A one-dimensional intensity image of a cross section at a certain location in a two-dimensional rigorously simulated optical image, which is a specific embodiment of the optical proximity correction method for ion implantation layers provided by the present invention;
[0066] Figure 6 for Figure 5 A comparison diagram of scaled one-dimensional intensity images;
[0067] Figure 7This is a schematic diagram of a specific embodiment of the optical proximity correction device for ion implantation layers provided by the present invention.
[0068] Figure label:
[0069] 10-Receiving module; 20-Segmentation module; 30-Division module; 40-2D OPC module; 50-Rasterization module; 60-Convolutional neural module; 70-Contour extraction module; 80-EPE module; 90-Edge segment adjustment module; 100-Loop module; 110-Synthesis module. Detailed Implementation
[0070] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0071] The core of this invention is to provide an optical proximity correction method for ion-implanted layers, and a flowchart of one specific implementation is shown below. Figure 1 As shown, this is referred to as Specific Implementation Method One, which includes:
[0072] S101: Receive the full chip layout to be corrected; the full chip layout to be corrected includes the ion implantation layer pattern and the underlying pattern.
[0073] Specifically, the underlying pattern includes at least one of an active layer pattern and a polysilicon layer pattern.
[0074] In this specific embodiment, the specific types of the underlying pattern are given, covering the most important underlying pattern types of ion implantation layers in advanced processes (well implantation corresponds to the active layer, and source / drain implantation corresponds to the polysilicon layer), so that the technical solution can be directly adapted to the actual production scenario and the industrial applicability of the present invention is improved.
[0075] S102: Divide the ion implantation layer pattern into edge segments and place control points at the center of each edge segment.
[0076] As one specific implementation method, this step includes:
[0077] A1: Place a first dividing point on the edge of the ion implantation layer pattern based on the projection position of the endpoint of the corresponding adjacent lower layer pattern.
[0078] You can refer to this. Figure 2The adjacent lower layer pattern refers to the lower layer pattern within a certain range around the ion implantation layer pattern. Of course, the lower layer pattern may overlap with the ion implantation layer pattern, that is, it may be inside or outside the projection of the ion implantation layer pattern. Figure 2 The corresponding situations are drawn in the figure. The first dividing point in the figure is filled with pure black, and the second dividing point is filled with diagonal lines.
[0079] A2: Based on the first cutting point and the preset critical distance, place a second cutting point on the edge of the ion implantation layer pattern.
[0080] The critical distance refers to the minimum distance that the optical proximity correction process can calculate and adjust; therefore, the distance between adjacent split points should not be less than the critical distance.
[0081] In this specific embodiment, the placement of the cutting point is divided into two steps. First, the first cutting point is placed to ensure that the edge area most significantly affected by the lower layer pattern is accurately cut, avoiding insufficient local correction due to overly coarse cutting. Second, after ensuring that the endpoint edges of the lower layer pattern are accurately cut, the second cutting point is placed to avoid generating excessively short edge segments. This ensures the accuracy of correction while controlling the amount of computation, balancing accuracy and efficiency.
[0082] S103: Based on the parameter requirements of the pre-trained optical proximity correction convolutional neural network model, the full chip layout to be corrected is divided into several correction units.
[0083] In other words, in this step, the full chip layout to be corrected is divided into several correction units. The resolution within a single correction unit should be consistent with the pixel resolution of the input and output images of the optical proximity correction convolutional neural network model, and the physical size of a single pixel should also be consistent with the physical size of the pixels in the input and output images of the optical proximity correction convolutional neural network model.
[0084] S104: Based on the current ion implantation layer mask pattern, the ion implantation optical image corresponding to each of the correction units is obtained using a pre-trained two-dimensional optical proximity correction simulation model; wherein, the initial current ion implantation layer mask pattern is obtained according to the ion implantation layer pattern.
[0085] The two-dimensional optical proximity correction simulation model is a traditional OPC model. In other words, it is an optical image obtained by only considering the ion implantation optical image and not considering other underlying structures.
[0086] In one specific implementation, the initial current ion implantation layer mask pattern is the ion implantation layer pattern.
[0087] In this specific embodiment, the ion implantation layer pattern is directly used as the initial current ion implantation layer mask pattern, which can unify the initial conditions for iterative correction, avoid iterative divergence caused by initial value deviation, ensure that the correction process converges stably to the target contour, and at the same time reduce the amount of computation and improve computational efficiency.
[0088] S105: Rasterize the lower-layer pattern of each of the correction units to obtain the corresponding lower-layer rasterized image.
[0089] There is no strict order between steps S104 and S105. The order can be adjusted or they can be performed simultaneously according to the actual situation. This invention does not limit this.
[0090] S106: Input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image.
[0091] In this step, the optical proximity correction convolutional neural network model is a model that outputs a single-channel output image based on a two-channel input image. The specific training method is described below.
[0092] S107: Use a preset threshold to extract the simulated contour of the corresponding correction unit from the output optical image.
[0093] S108: Calculate the edge placement error corresponding to each edge segment based on the simulated contour.
[0094] The edge placement error is the deviation between the simulated profile and the target pattern (ion implantation layer pattern) along the normal direction of the edge segment.
[0095] S109: Calculate the distance that the corresponding edge segment needs to be moved based on the edge placement error, and move the edge segment to obtain the updated current ion implantation layer mask pattern.
[0096] In a preferred embodiment, this step of calculating the distance the corresponding edge segment needs to move based on the edge placement error includes:
[0097] The distance that the corresponding edge segment needs to move is calculated based on the edge placement error using the following formula (1):
[0098] D = -f × (1 / MEF) × EPE; (1)
[0099] Where f is the feedback coefficient, MEF is the mask error amplification coefficient of the corresponding edge segment, and EPE is the edge placement error of the corresponding edge segment.
[0100] In this preferred embodiment, a mask error amplification factor (MEF) is introduced to quantify the correspondence between mask edge movement and wafer contour changes, avoiding overcorrection or undercorrection. On the other hand, this preferred embodiment can adjust the iteration convergence speed at any time through the feedback coefficient f, balancing high correction accuracy and fewer iterations.
[0101] S110: Repeat steps four to nine until the edge placement error of all edge segments is less than the preset threshold, and obtain the final optical proximity correction mask pattern of each correction unit.
[0102] S111: The final optical proximity correction mask patterns of all correction units are stitched together to obtain the full-chip optical proximity correction mask pattern.
[0103] Of course, the stitching of the final optical proximity correction mask pattern in this step is based on the relative positions of the corresponding correction units.
[0104] The present invention provides an optical proximity correction method for an ion implantation layer, comprising the following steps: Step 1, receiving a full-chip layout to be corrected, the full-chip layout including an ion implantation layer pattern and a lower layer pattern; Step 2, segmenting the ion implantation layer pattern into edge segments and placing control points at the center of each edge segment; Step 3, dividing the full-chip layout to be corrected into several correction units according to the parameter requirements of a pre-trained optical proximity correction convolutional neural network model; Step 4, obtaining the ion implantation optical image corresponding to each correction unit based on the current ion implantation layer mask pattern using a pre-trained two-dimensional optical proximity correction simulation model; wherein the initial current ion implantation layer mask pattern is obtained based on the ion implantation layer pattern; Step 5, rasterizing the lower layer pattern of each correction unit to obtain the corresponding lower layer rasterized image. Step 6: Input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image; Step 7: Use a preset threshold to extract the simulated contour of the corresponding correction unit from the output optical image; Step 8: Calculate the edge placement error corresponding to each edge segment based on the simulated contour; Step 9: Calculate the distance the corresponding edge segment needs to move according to the edge placement error, and move the edge segment to obtain the updated current ion implantation layer mask pattern; Step 10: Repeat steps 4 to 9 until the edge placement error of all edge segments is less than the preset threshold to obtain the final optical proximity correction mask pattern of each correction unit; Step 11: Stitch together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern. This invention utilizes a dual-input (ion-implanted optical image and lower-layer rasterized image) convolutional neural network model to fully learn the influence of the lower-layer pattern on the lithography intensity of the ion-implanted layer. This not only significantly improves the inference speed of optical proximity correction but also substantially enhances simulation accuracy, improves the lithography process window, and reduces potential defects. Furthermore, by combining the techniques of separate correction by segmentation correction units and subsequent splicing, OPC of the ion-implanted layer on a full-chip scale can be completed in a short time. This broadens the application scenarios of the invention, meets mass production requirements, and can be directly integrated into existing semiconductor manufacturing mask data generation pipelines, offering good compatibility and low production line modification costs.
[0105] As one specific implementation method, the optical proximity correction convolutional neural network model is established and trained through the following steps, referred to as Specific Implementation Method Two, and the corresponding flowchart is shown below. Figure 3 As shown, it includes:
[0106] S201: Establish a multi-input single-output convolutional neural network model, wherein the convolutional neural network model includes a first input path, a second input path, and an output path; wherein, the first input path is used to receive an ion-implanted optical image obtained based on the ion-implanted layer mask pattern and using a pre-trained two-dimensional optical proximity correction simulation model; the second input path is used to receive the lower-layer rasterized image corresponding to the lower-layer pattern, and the output path is used to output a corrected optical image considering the influence of the lower-layer pattern.
[0107] Specifically, in this step, a convolutional neural network model is established. This model includes multiple input images M1, M2, M3, ... and an output image M0, as shown in formula (2). For each modeling pattern, the input image M1 is the optical image considering only the ion implantation layer, that is, the ion implantation optical image, which can be obtained by simulation using the traditional optical proximity correction optical model (that is, the two-dimensional optical proximity correction simulation model), which has a relatively fast calculation speed; the input images M2, M3, ... are the lower-layer rasterized images of the corresponding active layer, polysilicon layer, and other lower-layer patterns.
[0108] M0=G(M1,M2,M3,…); (2)
[0109] Wherein, G represents the convolutional neural network model parameters of the convolutional neural network model.
[0110] S202: Obtain multiple sets of training samples, each set of training samples corresponding to a modeling pattern, including a set of input image pairs and a target image; the input image pairs include an ion implantation optical image obtained based on the ion implantation layer pattern in the modeling pattern and a pre-trained two-dimensional optical proximity correction simulation model, and a lower-layer rasterized image of the corresponding lower-layer pattern in the modeling pattern; the target image is the actual photolithographic optical image of the corresponding modeling pattern.
[0111] Of course, there is no strict order between steps S201 and S202, and they can be adjusted according to the actual situation.
[0112] S203: Using the target image as a reference, train the parameters of the convolutional neural network model by minimizing the deviation between the output image of the convolutional neural network model and the target image.
[0113] The deviation between the output image and the target image can be pixels and deviation. The target image I0(x,y) of the modeling pattern is trained with the parameters of the convolutional neural network model according to the objective function shown in formula (3), that is, the parameters of G in formula (2).
[0114] Obj=∑ j,k w jk· |M0(j,k)-I0(j,k)|2 ; (3)
[0115] Where M0(j,k) and I0(j,k) represent the intensities of the j-th row and k-th column pixels in the output and target images of the convolutional neural network, respectively, and w jk This represents the weight of the corresponding pixel.
[0116] During training, the ion implantation layer pattern in a single training sample is input from the first input path of the convolutional neural network model, the lower rasterized image is input from the second input path of the convolutional neural network model, and the output image of the output path of the convolutional neural network model is compared with the target image.
[0117] The convolutional neural network model in this preferred embodiment adopts a dual-input architecture, which separates the optical effects of the ion implantation layer itself from the influence of the underlying pattern. This allows the model to specifically learn the modulation rules of the underlying pattern and has extremely strong generalization ability. On the other hand, the pixel-level training target has higher accuracy than the traditional CD-level training and can accurately reproduce the lithographic contours of complex two-dimensional graphics.
[0118] As one specific implementation method, a flowchart illustrating one particular implementation of the training sample acquisition method is shown below. Figure 4 As shown, this is referred to as Specific Implementation Method Three, which includes:
[0119] S301: Design a modeling pattern and manufacture a test mask containing the modeling pattern; the modeling pattern includes an ion implantation layer pattern and a corresponding lower layer pattern, including the line width and spacing of different ion implantation layer patterns, the line width and spacing of different lower layer patterns, and a geometric combination of the spacing between different ion implantation layer patterns and the lower layer patterns.
[0120] In this step, a modeling pattern is designed and a test mask containing the modeling pattern is fabricated. The modeling pattern includes ion implantation layer patterns and corresponding underlying patterns, such as active layer and polysilicon layer patterns. The modeling pattern contains a combination of geometric features such as the linewidth and spacing of different ion implantation layer patterns, as well as the linewidth of different underlying patterns and the spacing between different ion implantation layer patterns and their underlying patterns.
[0121] S302: Perform photolithography using the test mask to obtain a photolithographic wafer, and obtain the measurement key dimensions of each of the modeling patterns on the photolithographic wafer.
[0122] This step involves running a photolithography process to obtain a photolithographically lithographically patterned wafer. A scanning electron microscope can then be used on the photolithographically lithographically patterned wafer to obtain the critical dimension (CD) of the modeled pattern, which is the key dimension W to be measured. CD.
[0123] S303: Establish a rigorous optical model, and calculate the three-dimensional rigorous simulation optical image of each modeled pattern using the rigorous optical model, and calculate the optical image after vertical diffusion.
[0124] Specifically, in this step, a rigorous optical model based on algorithms such as Finite-Difference Time-Domain (FDTD) or Rigorous Coupled-Wave Analysis (RCWA) is established, and a three-dimensional rigorous simulation optical image I(x,y,z) for each modeled pattern is calculated. The rigorous simulation optical image fully considers the influence of the underlying pattern on the light intensity distribution. Then, the optical image I after vertical diffusion is obtained according to formula (4). d (x,y,z).
[0125] I d (x,y,z)=I(x,y,z) exp(-z² / L²); (4)
[0126] Among them, I d (x,y,z) represents the optical image after vertical diffusion, I(x,y,z) represents the three-dimensional rigorously simulated optical image, L represents the vertical diffusion length, and z represents the spatial coordinates in the thickness direction. This is a convolution operation.
[0127] In this specific embodiment, the non-uniform scattering problem of photoresist, which was not considered in the rigorous optical simulation, is corrected, which greatly improves the consistency between the simulated light intensity distribution and the actual distribution. Furthermore, the complex beam propagation equation is replaced with a simple convolution operation, which improves the computational efficiency by more than an order of magnitude.
[0128] S304: For the optical image after vertical diffusion, determine the target plane position, target vertical diffusion length, and target contour threshold according to the measurement key dimensions of the corresponding modeling pattern.
[0129] Specifically, for the optical image I after vertical diffusion d (x,y,z), the optical image of a given z-plane. A two-dimensional image contains several rows and columns of pixels, such as 512*512 pixels. As shown in formula (5), for a two-dimensional optical image... By using the contour threshold T to truncate, the two-dimensional simulation contour of each modeling pattern can be obtained, and thus the critical dimension SCD of the simulation of the modeling pattern can be obtained. Then, the vertical diffusion length L, the Z-plane position and the threshold T are optimized using the cost function shown in formula (6).
[0130] ; (5)
[0131] in, The image is a two-dimensional optical image, and T is the contour threshold.
[0132] ; (6)
[0133] Among them, SCD j and WCD j Let w represent the critical dimensions for simulation and measurement of the j-th modeling pattern, respectively. j This indicates the corresponding weight.
[0134] In the following text, L0 can be used to represent the target vertical diffusion length (i.e., the optimized vertical diffusion length), Z0 can be used to represent the target planar position (i.e., the optimized planar position), and T0 can be used to represent the target contour threshold (i.e., the optimized contour threshold).
[0135] S305: Intensity scaling is performed on the two-dimensional rigorous simulation optical image of each modeling pattern at the corresponding target plane position to obtain the target image; the simulation key dimension of the target image under the target contour threshold is consistent with the corresponding measurement key dimension value.
[0136] You can refer to this. Figure 5 and Figure 6 ,in Figure 5 The figure shows a one-dimensional intensity image of a cross-section at a certain location in a two-dimensional rigorously simulated optical image, along with examples of the corresponding measured and simulated key dimensions. It can be seen that the simulated key dimension is slightly smaller than the measured key dimension, and T0 in the figure is the corresponding target contour threshold. Figure 6 Showing Figure 5 The image obtained by scaling the one-dimensional intensity image (dashed line) is the same as the simulated key dimension obtained under the threshold T0.
[0137] Furthermore, this step includes:
[0138] The two-dimensional rigorous simulation optical image of each modeled pattern at the corresponding target plane position is intensity scaled according to the following formula (7) to obtain the target image:
[0139] ; (7)
[0140] Where I0(x,y) is the target image, and s is the scaling factor. This refers to the two-dimensional rigorously simulated optical image.
[0141] Scaling using Equation (7) can quickly eliminate the systematic deviation between simulation and actual process without modifying the complex and rigorous optical model parameters. On the other hand, it also ensures that the key dimensions of the target image of each training sample are completely consistent with the wafer measurement value under the target threshold, providing an accurate benchmark for training the neural convolution model.
[0142] Of course, if the training samples need to be mass-produced using the rigorous optical model in this preferred embodiment, then the number of pixels in each input image and output image must be consistent with the target image in this specific embodiment.
[0143] This preferred embodiment solves the problem of insufficient high-precision, large-batch samples required for training convolutional neural network models. It does not rely on massive and expensive wafer measurement data. On the other hand, by combining measurement data with rigorous simulation results, the generated samples are highly consistent with the actual process, ensuring that the trained model can be directly used in production, thus improving the practicality of the invention.
[0144] The optical proximity correction device for ion implantation layers provided in the embodiments of the present invention will be described below. The optical proximity correction device for ion implantation layers described below can be referred to in correspondence with the optical proximity correction method for ion implantation layers described above.
[0145] Figure 7 This is a structural block diagram of an optical proximity correction device for an ion implantation layer provided in an embodiment of the present invention, with reference to... Figure 7 The optical proximity correction device for the ion implantation layer may include:
[0146] The receiving module 10 is used in step one to receive the full chip layout to be corrected; the full chip layout to be corrected includes an ion implantation layer pattern and a lower layer pattern;
[0147] The segmentation module 20 is used in step two to segment the ion implantation layer pattern into edge segments and place control points at the center point of each edge segment.
[0148] The partitioning module 30 is used in step three to divide the full chip layout to be corrected into several correction units according to the parameter requirements of the pre-trained optical proximity correction convolutional neural network model.
[0149] The two-dimensional OPC module 40 is used in step four to obtain the ion implantation optical image corresponding to each correction unit based on the current ion implantation layer mask pattern and using a pre-trained two-dimensional optical proximity correction simulation model; wherein, the initial current ion implantation layer mask pattern is obtained according to the ion implantation layer pattern.
[0150] The rasterization module 50 is used in step five to rasterize the lower-layer pattern of each of the correction units to obtain the corresponding lower-layer rasterized image.
[0151] Convolutional neural module 60 is used in step six to input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image.
[0152] The contour cropping module 70 is used in step seven to crop the simulated contour of the corresponding correction unit in the output optical image using a preset threshold.
[0153] EPE module 80 is used in step eight to calculate the edge placement error corresponding to each edge segment based on the simulation contour.
[0154] The edge segment adjustment module 90 is used in step nine to calculate the distance that the corresponding edge segment needs to be moved according to the edge placement error, and move the edge segment to obtain the updated current ion implantation layer mask pattern.
[0155] The loop module 100 is used to loop steps four to nine in step ten until the edge placement error of all edge segments is less than a preset threshold, so as to obtain the final optical proximity correction mask pattern of each correction unit.
[0156] The stitching module 110 is used in step eleven to stitch together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern.
[0157] In one specific implementation, the segmentation module 20 includes:
[0158] The first slicing unit is used to place a first slicing point on the edge of the ion implantation layer pattern based on the projection position of the endpoint of the corresponding adjacent lower layer pattern;
[0159] The second segmentation unit is used to place a second segmentation point on the edge of the ion implantation layer pattern based on the first segmentation point and a preset critical distance.
[0160] In one specific implementation, the edge segment adjustment module 90 includes:
[0161] The calculation and movement unit is used to calculate the distance that the corresponding edge segment needs to be moved based on the edge placement error using the following formula:
[0162] D = -f × (1 / MEF) × EPE;
[0163] Where f is the feedback coefficient, MEF is the mask error amplification coefficient of the corresponding edge segment, and EPE is the edge placement error of the corresponding edge segment.
[0164] In one specific implementation, the optical proximity correction convolutional neural network model is built and trained by a device that performs the following steps:
[0165] A convolutional neural network (CNN) model is established to build a multi-input single-output CNN model. The CNN model includes a first input path, a second input path, and an output path. The first input path is used to receive an ion-implanted optical image obtained based on the ion-implanted layer mask pattern and using a pre-trained two-dimensional optical proximity correction simulation model. The second input path is used to receive the lower-layer rasterized image corresponding to the lower-layer pattern. The output path is used to output a corrected optical image that takes into account the influence of the lower-layer pattern.
[0166] The sample acquisition module is used to acquire multiple sets of training samples, each set of training samples corresponding to a modeling pattern, including a set of input image pairs and a target image; the input image pairs include an ion implantation optical image obtained based on the ion implantation layer pattern in the modeling pattern using a pre-trained two-dimensional optical proximity correction simulation model, and a lower-layer rasterized image of the corresponding lower-layer pattern in the modeling pattern; the target image is the actual photolithographic optical image of the corresponding modeling pattern.
[0167] The training module is used to train the parameters of the convolutional neural network model by minimizing the deviation between the output image of the convolutional neural network model and the target image, using the target image as a reference.
[0168] As one specific implementation, the training sample acquisition device includes:
[0169] The design and manufacturing module is used to design modeling patterns and manufacture test masks containing the modeling patterns; the modeling patterns include ion implantation layer patterns and corresponding lower layer patterns, including geometric feature combinations of line widths and spacings of different ion implantation layer patterns, line widths and spacings of different lower layer patterns, and the distances from different ion implantation layer patterns to the lower layer patterns.
[0170] The photolithography module is used to perform photolithography using the test mask to obtain a photolithographic wafer and to acquire the measurement key dimensions of each of the modeled patterns on the photolithographic wafer.
[0171] The optical model module is used to establish a rigorous optical model, calculate a three-dimensional rigorous simulation optical image of each modeled pattern using the rigorous optical model, and calculate the optical image after vertical diffusion.
[0172] The positioning module is used to determine the target plane position, target vertical diffusion length, and target contour threshold of the vertically diffused optical image based on the key measurement dimensions of the corresponding modeling pattern.
[0173] The scaling module is used to perform intensity scaling on the two-dimensional rigorous simulation optical image of each modeling pattern at the corresponding target plane position to obtain the target image; the simulation key dimension of the target image under the target contour threshold is consistent with the corresponding measurement key dimension value.
[0174] As one specific implementation, the optical model module includes:
[0175] The vertical diffusion unit is used to calculate the optical image after vertical diffusion using the following formula:
[0176] Id(x,y,z)=I(x,y,z) exp(-z² / L²);
[0177] Where Id(x,y,z) is the optical image after vertical diffusion, I(x,y,z) is the three-dimensional rigorously simulated optical image, L is the vertical diffusion length, and z is the spatial coordinate in the thickness direction. This is a convolution operation.
[0178] In one specific implementation, the scaling module includes:
[0179] The scaling calculation unit is used to scale the intensity of the two-dimensional rigorous simulation optical image of each modeled pattern at the corresponding target plane position according to the following formula to obtain the target image:
[0180] IO(x,y)=s×Idz0(x,y);
[0181] Wherein, IO(x,y) is the target image, s is the scaling factor, and Idz0(x,y) is the two-dimensional strictly simulated optical image.
[0182] The present invention provides an optical proximity correction device for an ion implantation layer, comprising: a receiving module 10, used in step one to receive a full-chip layout to be corrected; the full-chip layout to be corrected includes an ion implantation layer pattern and a lower layer pattern; a segmentation module 20, used in step two to segment the ion implantation layer pattern into edge segments and place control points at the center of each edge segment; a division module 30, used in step three to divide the full-chip layout to be corrected into several correction units according to the parameter requirements of a pre-trained optical proximity correction convolutional neural network model; a two-dimensional OPC module 40, used in step four to obtain the ion implantation optical image corresponding to each correction unit based on the current ion implantation layer mask pattern and using a pre-trained two-dimensional optical proximity correction simulation model; wherein the initial current ion implantation layer mask pattern is obtained according to the ion implantation layer pattern; a rasterization module 50, used in step five to rasterize the lower layer pattern of each correction unit to obtain the corresponding lower layer rasterized image; and a convolutional neural network model. Block 60, used in step six, inputs the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image; Contour cropping module 70, used in step seven, uses a preset threshold to crop the simulated contour of the corresponding correction unit in the output optical image; EPE module 80, used in step eight, calculates the edge placement error corresponding to each edge segment based on the simulated contour; Edge segment adjustment module 90, used in step nine, calculates the distance that the corresponding edge segment needs to move according to the edge placement error, and moves the edge segment to obtain the updated current ion implantation layer mask pattern; Looping module 100, used in step ten, loops steps four to nine until the edge placement error of all edge segments is less than the preset threshold to obtain the final optical proximity correction mask pattern of each correction unit; Stitching module 110, used in step eleven, stitches together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern. This invention utilizes a dual-input (ion-implanted optical image and lower-layer rasterized image) convolutional neural network model to fully learn the influence of the lower-layer pattern on the lithography intensity of the ion-implanted layer. This not only significantly improves the inference speed of optical proximity correction but also substantially enhances simulation accuracy, improves the lithography process window, and reduces potential defects. Furthermore, by combining the techniques of separate correction by segmentation correction units and subsequent splicing, OPC of the ion-implanted layer on a full-chip scale can be completed in a short time. This broadens the application scenarios of the invention, meets mass production requirements, and can be directly integrated into existing semiconductor manufacturing mask data generation pipelines, offering good compatibility and low production line modification costs.
[0183] The optical proximity correction device for the ion implantation layer in this embodiment is used to implement the aforementioned optical proximity correction method for the ion implantation layer. Therefore, the specific implementation of the optical proximity correction device for the ion implantation layer can be found in the embodiment section of the optical proximity correction method for the ion implantation layer above. For example, the receiving module 10, the segmentation module 20, the division module 30, the two-dimensional OPC module 40, the rasterization module 50, the convolutional neural module 60, the contour extraction module 70, the EPE module 80, the edge segment adjustment module 90, the loop module 100, and the merging module 110 are respectively used to implement steps S101, S102, S103, S104, S105, S106, S107, S108, S109, S110, and S111 in the aforementioned optical proximity correction method for the ion implantation layer. Therefore, its specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.
[0184] The present invention also provides an optical proximity correction device for an ion-implanted layer, comprising:
[0185] Memory, used to store computer programs;
[0186] A processor is configured to execute the computer program to implement the steps of any of the above-described optical proximity correction methods for ion implantation layers. The optical proximity correction method for ion implantation layers provided by this invention includes the following steps: Step 1: Receiving a full-chip layout to be corrected; the full-chip layout to be corrected includes an ion implantation layer pattern and a lower layer pattern; Step 2: Dividing the ion implantation layer pattern into edge segments and placing control points at the center points of each edge segment; Step 3: Dividing the full-chip layout to be corrected into several correction units according to the parameter requirements of a pre-trained optical proximity correction convolutional neural network model; Step 4: Obtaining the ion implantation optical image corresponding to each correction unit based on the current ion implantation layer mask pattern using a pre-trained two-dimensional optical proximity correction simulation model; wherein the initial current ion implantation layer mask pattern is obtained based on the ion implantation layer pattern; Step 5: Rasterizing the lower layer pattern of each correction unit to obtain the corresponding lower layer rasterized image. Step 6: Input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image; Step 7: Use a preset threshold to extract the simulated contour of the corresponding correction unit from the output optical image; Step 8: Calculate the edge placement error corresponding to each edge segment based on the simulated contour; Step 9: Calculate the distance the corresponding edge segment needs to move according to the edge placement error, and move the edge segment to obtain the updated current ion implantation layer mask pattern; Step 10: Repeat steps 4 to 9 until the edge placement error of all edge segments is less than the preset threshold to obtain the final optical proximity correction mask pattern of each correction unit; Step 11: Stitch together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern. This invention utilizes a dual-input (ion-implanted optical image and lower-layer rasterized image) convolutional neural network model to fully learn the influence of the lower-layer pattern on the lithography intensity of the ion-implanted layer. This not only significantly improves the inference speed of optical proximity correction but also substantially enhances simulation accuracy, improves the lithography process window, and reduces potential defects. Furthermore, by combining the techniques of separate correction by segmentation correction units and subsequent splicing, OPC of the ion-implanted layer on a full-chip scale can be completed in a short time. This broadens the application scenarios of the invention, meets mass production requirements, and can be directly integrated into existing semiconductor manufacturing mask data generation pipelines, offering good compatibility and low production line modification costs.
[0187] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described optical proximity correction methods for ion implantation layers. The optical proximity correction method for ion implantation layers provided by this invention includes the following steps: Step 1: Receiving a full-chip layout to be corrected; the full-chip layout to be corrected includes an ion implantation layer pattern and a lower layer pattern; Step 2: Dividing the ion implantation layer pattern into edge segments and placing control points at the center points of each edge segment; Step 3: Dividing the full-chip layout to be corrected into several correction units according to the parameter requirements of a pre-trained optical proximity correction convolutional neural network model; Step 4: Obtaining the ion implantation optical image corresponding to each correction unit based on the current ion implantation layer mask pattern using a pre-trained two-dimensional optical proximity correction simulation model; wherein the initial current ion implantation layer mask pattern is obtained based on the ion implantation layer pattern; Step 5: Rasterizing the lower layer pattern of each correction unit to obtain the corresponding lower layer rasterized image. Step 6: Input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image; Step 7: Use a preset threshold to extract the simulated contour of the corresponding correction unit from the output optical image; Step 8: Calculate the edge placement error corresponding to each edge segment based on the simulated contour; Step 9: Calculate the distance the corresponding edge segment needs to move according to the edge placement error, and move the edge segment to obtain the updated current ion implantation layer mask pattern; Step 10: Repeat steps 4 to 9 until the edge placement error of all edge segments is less than the preset threshold to obtain the final optical proximity correction mask pattern of each correction unit; Step 11: Stitch together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern. This invention utilizes a dual-input (ion-implanted optical image and lower-layer rasterized image) convolutional neural network model to fully learn the influence of the lower-layer pattern on the lithography intensity of the ion-implanted layer. This not only significantly improves the inference speed of optical proximity correction but also substantially enhances simulation accuracy, improves the lithography process window, and reduces potential defects. Furthermore, by combining the techniques of separate correction by segmentation correction units and subsequent splicing, OPC of the ion-implanted layer on a full-chip scale can be completed in a short time. This broadens the application scenarios of the invention, meets mass production requirements, and can be directly integrated into existing semiconductor manufacturing mask data generation pipelines, offering good compatibility and low production line modification costs.
[0188] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0189] It should be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0190] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0191] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0192] The optical proximity correction method, apparatus, device, and storage medium for ion implantation layers provided by this invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of this invention.
Claims
1. An optical proximity correction method for an ion-implanted layer, characterized in that, include: Step 1: Receive the full chip layout to be corrected; the full chip layout to be corrected includes the ion implantation layer pattern and the underlying pattern; Step 2: Divide the ion implantation layer pattern into edge segments and place control points at the center of each edge segment; Step 3: Based on the parameter requirements of the pre-trained optical proximity correction convolutional neural network model, the full chip layout to be corrected is divided into several correction units. Step 4: Based on the current ion implantation layer mask pattern, the ion implantation optical image corresponding to each correction unit is obtained using a pre-trained two-dimensional optical proximity correction simulation model; wherein, the initial current ion implantation layer mask pattern is obtained according to the ion implantation layer pattern. Step 5: Rasterize the lower-layer pattern of each correction unit to obtain the corresponding lower-layer rasterized image; Step 6: Input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image; Step 7: Use a preset threshold to extract the simulated contour of the corresponding correction unit from the output optical image; Step 8: Calculate the edge placement error corresponding to each edge segment based on the simulated contour; Step 9: Calculate the distance that the corresponding edge segment needs to be moved based on the edge placement error, and move the edge segment to obtain the updated current ion implantation layer mask pattern; Step 10: Repeat steps 4 to 9 until the edge placement error of all edge segments is less than the preset threshold, and obtain the final optical proximity correction mask pattern of each correction unit. Step 11: Stitch together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern.
2. The optical proximity correction method for ion-implanted layers as described in claim 1, characterized in that, Edge segmentation of the ion implantation layer pattern includes: On the edge of the ion implantation layer pattern, a first dividing point is placed based on the projection position of the endpoint of the corresponding adjacent lower layer pattern; A second cutting point is placed on the edge of the ion implantation layer pattern based on the first cutting point and the preset critical distance.
3. The optical proximity correction method for ion-implanted layers as described in claim 1, characterized in that, The underlying pattern includes at least one of an active layer pattern and a polysilicon layer pattern.
4. The optical proximity correction method for ion-implanted layers as described in claim 1, characterized in that, The initial current ion implantation layer mask pattern is the ion implantation layer pattern.
5. The optical proximity correction method for ion-implanted layers as described in claim 1, characterized in that, The distance that the corresponding edge segment needs to be moved is calculated based on the edge placement error, including: The distance that the corresponding edge segment needs to be moved is calculated based on the edge placement error using the following formula: D = -f × (1 / MEF) × EPE; Where f is the feedback coefficient, MEF is the mask error amplification coefficient of the corresponding edge segment, and EPE is the edge placement error of the corresponding edge segment.
6. The optical proximity correction method for ion-implanted layers as described in claim 1, characterized in that, The optical proximity correction convolutional neural network model is established and trained through the following steps: A multi-input single-output convolutional neural network model is established, comprising a first input path, a second input path, and an output path. The first input path is used to receive an ion-implanted optical image obtained based on an ion-implanted layer mask pattern and using a pre-trained two-dimensional optical proximity correction simulation model. The second input path is used to receive a lower-layer rasterized image corresponding to the lower-layer pattern. The output path is used to output a corrected optical image that takes into account the influence of the lower-layer pattern. Multiple sets of training samples are acquired, each set of training samples corresponds to a modeling pattern, including a set of input image pairs and a target image; the input image pairs include an ion implantation optical image obtained based on the ion implantation layer pattern in the modeling pattern and a pre-trained two-dimensional optical proximity correction simulation model, and a lower-layer rasterized image of the corresponding lower-layer pattern in the modeling pattern; the target image is the actual photolithographic optical image of the corresponding modeling pattern. Using the target image as a reference, the parameters of the convolutional neural network model are trained by minimizing the deviation between the output image of the convolutional neural network model and the target image.
7. The optical proximity correction method for ion-implanted layers as described in claim 6, characterized in that, The method for obtaining the training samples includes: Design a modeling pattern and manufacture a test mask containing the modeling pattern; the modeling pattern includes an ion implantation layer pattern and a corresponding lower layer pattern, including the line width and spacing of different ion implantation layer patterns, the line width and spacing of different lower layer patterns, and a geometric combination of the spacing between different ion implantation layer patterns and the lower layer patterns. The test mask is used to perform photolithography to obtain a photolithographic wafer, and the key dimensions of each modeling pattern on the photolithographic wafer are measured. A rigorous optical model is established, and a three-dimensional rigorous simulation optical image of each modeled pattern is calculated using the rigorous optical model, as well as the optical image after vertical diffusion; For the vertically diffused optical image, the target plane position, target vertical diffusion length, and target contour threshold are determined based on the measurement key dimensions of the corresponding modeling pattern. The target image is obtained by intensity scaling of the two-dimensional rigorous simulation optical image of each modeling pattern at the corresponding target plane position; the simulation key dimension of the target image under the target contour threshold is consistent with the corresponding measurement key dimension value.
8. The optical proximity correction method for ion-implanted layers as described in claim 7, characterized in that, Calculate the optical image after vertical diffusion, including: The optical image after vertical diffusion is calculated using the following formula: I d (x,y,z)=I(x,y,z) exp(-z² / L²); Where Id(x,y,z) is the optical image after vertical diffusion, I(x,y,z) is the three-dimensional rigorously simulated optical image, L is the vertical diffusion length, and z is the spatial coordinate in the thickness direction. This is a convolution operation.
9. The optical proximity correction method for ion-implanted layers as described in claim 7, characterized in that, The target image is obtained by intensity scaling of the two-dimensional rigorous simulation optical image of each modeled pattern at the corresponding target plane position, including: For each of the modeled patterns, the two-dimensional rigorous simulation optical image at the corresponding target plane position is intensity scaled according to the following formula to obtain the target image: ； Where I0(x,y) is the target image, and s is the scaling factor. This refers to the two-dimensional rigorously simulated optical image.
10. An optical proximity correction device for an ion-implanted layer, characterized in that, include: The receiving module is used in step one to receive the full chip layout to be corrected; the full chip layout to be corrected includes an ion implantation layer pattern and a lower layer pattern; The segmentation module is used in step two to segment the ion implantation layer pattern into edge segments and place control points at the center of each edge segment. The partitioning module is used in step three to divide the full chip layout to be corrected into several correction units according to the parameter requirements of the pre-trained optical proximity correction convolutional neural network model. The two-dimensional OPC module is used in step four to obtain the ion implantation optical image corresponding to each correction unit based on the current ion implantation layer mask pattern and using a pre-trained two-dimensional optical proximity correction simulation model; wherein, the initial current ion implantation layer mask pattern is obtained according to the ion implantation layer pattern. The rasterization module is used in step five to rasterize the lower-layer pattern of each of the correction units to obtain the corresponding lower-layer rasterized image. The convolutional neural module is used in step six to input the corresponding ion implantation optical image and the lower-layer rasterized image into the optical proximity correction convolutional neural network model to obtain the output optical image. The contour cropping module is used in step seven to crop the simulated contour of the corresponding correction unit in the output optical image using a preset threshold. The EPE module is used in step eight to calculate the edge placement error corresponding to each edge segment based on the simulation contour. The edge segment adjustment module is used in step nine to calculate the distance that the corresponding edge segment needs to be moved based on the edge placement error, and to move the edge segment to obtain the updated current ion implantation layer mask pattern. The loop module is used in step ten to loop steps four to nine until the edge placement error of all edge segments is less than a preset threshold, so as to obtain the final optical proximity correction mask pattern of each correction unit. The stitching module is used in step eleven to stitch together the final optical proximity correction mask patterns of all correction units to obtain the full-chip optical proximity correction mask pattern.