CN115222829A - Mask optimization method and device based on neural network model - Google Patents
Mask optimization method and device based on neural network model Download PDFInfo
- Publication number
- CN115222829A CN115222829A CN202210902804.5A CN202210902804A CN115222829A CN 115222829 A CN115222829 A CN 115222829A CN 202210902804 A CN202210902804 A CN 202210902804A CN 115222829 A CN115222829 A CN 115222829A
- Authority
- CN
- China
- Prior art keywords
- neural network
- mask
- image
- network model
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 85
- 238000003062 neural network model Methods 0.000 title claims abstract description 84
- 238000005457 optimization Methods 0.000 title claims abstract description 53
- 238000013528 artificial neural network Methods 0.000 claims abstract description 47
- 230000008569 process Effects 0.000 claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 36
- 238000001259 photo etching Methods 0.000 claims abstract description 34
- 238000005516 engineering process Methods 0.000 claims abstract description 24
- 238000001459 lithography Methods 0.000 claims abstract description 18
- 238000007906 compression Methods 0.000 claims description 38
- 230000006837 decompression Effects 0.000 claims description 38
- 230000006835 compression Effects 0.000 claims description 26
- 230000004913 activation Effects 0.000 claims description 12
- 239000013078 crystal Substances 0.000 claims description 8
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 238000003384 imaging method Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 abstract description 72
- 238000012937 correction Methods 0.000 abstract description 11
- 238000010586 diagram Methods 0.000 description 9
- 230000003287 optical effect Effects 0.000 description 7
- 238000013461 design Methods 0.000 description 4
- 230000009467 reduction Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 229920002120 photoresistant polymer Polymers 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000009966 trimming Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/002—Image coding using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Preparing Plates And Mask In Photomechanical Process (AREA)
Abstract
The invention provides a mask optimization method and a mask optimization device based on a neural network model, wherein the method comprises the following steps: acquiring a data sample set, wherein the data sample set comprises a plurality of groups of target images and OPC mask images which are generated by utilizing an OPC method and correspond to the target images; training a neural network mask model through a data sample set; inputting the target image into a neural network mask model which is output by training to obtain a neural network model mask image; loading the neural network model mask image into a photoetching engine to obtain a wafer image; the neural network mask model is again optimized by minimizing the loss function of the wafer image and the target image. The invention provides a method for learning the OCP correction process through a neural network model and generating the neural network mask model to break through the limitation of the OCP solution space height, so that the method has wider application range, and further fine tuning of the neural network mask model is realized through learning of an Inverse Lithography Technology (ILT) on the basis.
Description
Technical Field
The invention relates to the field of chip manufacturing, in particular to a mask optimization method and device based on a neural network model.
Background
Mask optimization has been a key issue in VLSI (Very Large Scale Integration) design flows due to the mismatch between lithography systems and ever shrinking feature sizes. The mask is used for converting a target image into a mask image, the mask image is printed on a wafer through a photoetching engine to form a final wafer image, and the mask optimization aims at generating a high-quality mask image so as to improve the quality of the wafer image. Currently, the optimization of the mask is mostly realized by using an Optical Proximity Correction (OPC) technology and an Inverse Lithography Technology (ILT), and a conventional wafer image generation process is shown in fig. 1, and the main principle of the conventional wafer image generation process includes dividing a pattern edge into segments in an OPC stream and then performing shift/correction according to a mathematical model; then using a sub-resolution assist function (SRAF) to obtain a high printing adaptability mask; and then fine-tuning the wafer image by using an Inverse Lithography Technology (ILT). However, model-based OPC flows are highly limited by their solution space (the main operation of OPC is matching and finding of rule bases, accuracy is limited by the size of the database), and lack reliability for complex designs.
Therefore, how to design a mask optimization method that is not limited by the height of the solution space is highly needed in the art.
Disclosure of Invention
In order to be able to get rid of the limitation of the OPC solution space height and provide a mask image with better quality, in one aspect of the invention, a mask optimization method based on a neural network model is provided, wherein the mask is used for round wafer imaging of a lithography system, and the method comprises the following steps: acquiring a data sample set, wherein the data sample set comprises a plurality of groups of target images and OPC mask images which are generated by utilizing an OPC method and correspond to the target images; training a neural network mask model through the data sample set; inputting the target image into the neural network mask model output by training to obtain a neural network model mask image; loading the neural network model mask image into a photoetching engine to obtain a wafer image; the neural network mask model is again optimized by minimizing the loss function of the wafer image and the target image.
In one or more embodiments, the method includes performing mask optimization using an AutoEncoder neural network model, the training of the neural network mask model by the set of data samples includes: reducing the dimension of an input target image through a compression function and acquiring abstract characteristics; restoring the abstract features to a mask image by a decompression function; the compression function and the weight of the decompression function are trained by minimizing a loss function of the mask image and the corresponding OPC mask image.
In one or more embodiments, the method further comprises: determining the number of layers of hidden layers and input and output dimensions in the compression process to construct a compression function; determining the layer number and input/output dimensionality of a hidden layer in the decompression process to construct a decompression function; and respectively configuring an activation function for the compression process and the decompression process.
In one or more embodiments, the activation function includes: sigmoid function.
In one or more embodiments, said re-optimizing said neural network mask model by minimizing a loss function of said wafer image and said target image comprises: constructing a photoetching error function of the round crystal image and the target image based on a reverse photoetching technology; calculating the gradient of the photoetching error relative to the parameters of the AutoEncoder neural network model by sequentially solving partial derivatives of the mask and the parameters of the AutoEncoder neural network model; calculating the parameters of the AutoEncoder neural network model when the photoetching error is minimum by a gradient descent method; the parameters of the AutoEncoder neural network model are composed of the compression function and the weight coefficient of the decompression function.
In a second aspect of the present invention, a mask optimization apparatus based on a neural network model is provided, including: the device comprises a sample set generation module, a data acquisition module and a data acquisition module, wherein the sample set generation module is used for acquiring a plurality of groups of target images and generating OPC (optical proximity correction) mask images corresponding to the target images by using an OPC (optical proximity correction) method to obtain a data sample set; a first training module configured to train a neural network mask model through the set of data samples; the mask image generation module is used for inputting the target image into the neural network mask model output by training to obtain a neural network model mask image; the round crystal image generation module is configured for loading the neural network model mask image into a photoetching engine to obtain a round crystal image; a second training module configured to re-optimize the neural network mask model by minimizing a loss function of the wafer image and the target image.
In one or more embodiments, the first training module is further configured to: carrying out mask optimization by using an AutoEncoder neural network model; reducing the dimension of an input target image through a compression function and acquiring abstract characteristics; restoring the abstract features to a mask image by a decompression function; the weights of the compression function and the decompression function are trained by minimizing a loss function of the mask image with a corresponding OPC mask image.
In one or more embodiments, the mask optimization apparatus based on a neural network model of the present invention further includes: the model building module is configured and used for determining the number of layers and input and output dimensions of hidden layers in the compression process so as to build a compression function; determining the number of layers of hidden layers and input-output dimensions in the decompression process to construct a decompression function; and configuring activation functions for the compression process and the decompression process, respectively.
In one or more embodiments, the activation function includes: sigmoid function.
In one or more embodiments, the second training module is further configured to: constructing a photoetching error function of the wafer image and the target image based on an inverse photoetching technology; calculating the gradient of the photoetching error relative to the parameters of the AutoEncoder neural network model by sequentially solving partial derivatives of the mask and the parameters of the AutoEncoder neural network model; calculating the parameters of the AutoEncoder neural network model when the photoetching error is minimum by a gradient descent method; the parameters of the AutoEncoder neural network model consist of the compression function and the weight coefficient of the decompression function.
The beneficial effects of the invention include: based on an Optical Proximity Correction (OPC) technology, the OCP correction process is learned through a neural network model, and a neural network mask model is generated to break through the limitation of the OCP solution space height, so that the method is wider in application range; and because the neural network mask model inherits the strong learning capability of the neural network model, the invention further realizes the further optimization of the neural network mask model based on the learning of the Inverse Lithography Technology (ILT), thereby further improving the quality of the mask image generated by the neural network mask model.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other embodiments can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a schematic diagram of a conventional wafer image generation process;
FIG. 2 is a flowchart of the mask optimization method based on neural network model according to the present invention;
FIG. 3 is a schematic diagram of a first phase tuning process of the neural network model-based mask optimization method of the present invention;
FIG. 4 is a diagram illustrating a second phase tuning process of the neural network model-based mask optimization method according to the present invention;
FIG. 5 is a schematic structural diagram of a mask optimization apparatus based on a neural network model according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the following embodiments of the present invention are described in further detail with reference to the accompanying drawings.
It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are used for distinguishing two entities with the same name but different names or different parameters, and it should be noted that "first" and "second" are merely for convenience of description and should not be construed as limitations of the embodiments of the present invention, and they are not described in any way in the following embodiments.
In order to generate a high-quality mask image (namely a mask template), the invention provides a method for learning the OCP correction process of the mask image through a neural network model based on an Optical Proximity Correction (OPC) technology and generating the neural network mask model to break through the limitation of the OCP solution space height, so that the method has wider application range; and because the neural network mask model inherits the strong learning ability of the neural network model, the invention further realizes the further optimization of the neural network mask model based on the learning of the Inverse Lithography Technology (ILT), thereby further improving the quality of the mask image generated by the neural network mask model.
FIG. 2 is a flowchart of the mask optimization method based on neural network model according to the present invention. As shown in fig. 2, the work flow of the mask optimization method based on the neural network model of the present invention includes: step 100, acquiring a data sample set, wherein the data sample set comprises a plurality of groups of target images and OPC mask images which are generated by using an OPC method and correspond to the target images; step 200, training a neural network mask model through a data sample set; step 300, inputting a target image into a neural network mask model output by training to obtain a neural network model mask image; step 400, loading the neural network model mask image into a photoetching engine to obtain a wafer image; step 500 again optimizes the neural network mask model by minimizing the loss function of the wafer image and the target image. The target image is a design image, the mask image is a mask template, the wafer image is an image formed by transferring the mask image onto the wafer by using a photoetching technology, and the wafer image is not completely consistent with the mask image due to the diffraction superposition effect of light.
In a further embodiment, the neural network model adopted by the present invention is an AutoEncoder neural network model (hereinafter referred to as the AutoEncoder model for short), and the step 200 trains a neural network mask model through a data sample set, including: step 210, performing dimension reduction on the input target image through a compression function and acquiring abstract features; step 220, restoring the abstract features into a mask image through a decompression function; and a step 230 of training the compression function and the weights of the decompression function by minimizing the loss function of the mask image and the corresponding OPC mask image.
Specifically, the Autoencor model comprises two processes of encorder (compression) and decoder (decompression), wherein one or more hidden layers are established in the encorder process, dimension reduction is performed on an input current image through an encorder function, and input data are mapped to a feature space to realize extraction of abstract features; in the subsequent decoder process, mapping the abstract features back to the original space through a decoder function to obtain a reconstructed sample; and finally, training the weights of the encoder function and the decoder function by minimizing the defined loss function, thereby completing the training of the neural network mask model.
In a further embodiment, the mask optimization method based on the neural network model further includes determining the number of hidden layers and input/output dimensions in the compression process to construct a compression function; determining the number of layers of hidden layers and input-output dimensions in the decompression process to construct a decompression function; and configuring activation functions for the compression process and the decompression process, respectively. Wherein the activation function includes, but is not limited to, a sigmoid function.
In a further embodiment, step 500, again optimizing the neural network mask model by minimizing a loss function of the wafer image and the target image, comprises: step 510, constructing a photoetching error function of the wafer image and the target image based on an inverse photoetching technology; step 520, calculating the gradient of the photoetching error relative to the parameters of the AutoEncoder neural network model by sequentially solving the partial derivatives of the mask and the parameters of the AutoEncoder neural network model; step 530, calculating the parameters of the AutoEncoder neural network model when the photoetching error is minimum by a gradient descent method; the parameters of the AutoEncoder neural network model consist of compression functions and weight coefficients of decompression functions.
It can be seen from the above embodiments that the mask optimization method based on the neural network model provided by the present invention is divided into two stages, the first stage is to obtain the neural network mask model capable of outputting a mask image close to an OCP mask image after inputting a target image by learning the mask image generated by an Optical Proximity Correction (OPC) technique, and the process thereof refers to fig. 3, and fig. 3 is a schematic diagram of a first-stage tuning process of the mask optimization method based on the neural network model of the present invention; in the second stage, the learning of the optimization of the wafer image is performed through the Inverse Lithography Technology (ILT), but the difference is that the optimization of the wafer image through the Inverse Lithography Technology (ILT) is fed back to the neural network mask model, and the fine tuning of the neural network mask model is realized from another angle. Therefore, the quality of the mask image generated by using the neural network mask model of the invention is further improved, so that the circular crystal image with higher precision can be generated after the mask image is printed on a circular crystal, and the specific process is shown in fig. 4, wherein fig. 4 is a schematic diagram of a second-stage tuning process of the mask optimization method based on the neural network model. Of course, if the wafer image needs to be adjusted and optimized by combining the back lithography technology (ILT) again, the method can greatly shorten the time required by adjustment and optimization.
Application example 1-training procedure for OPC-based AutoEncoder
In the OPC-based AutoEncoder training stage, a target image Z t Inputting the model of AutoEncoder, outputting the generated mask M of AutoEncoder, and minimizing OPC-based group Truth mask M * Generated with AutoEncoderThe error between the masks M makes the mask M generated by the AutoEncoder approach the OPC-based group Truth mask M infinitely * . The process of training the AutoEncoder is shown in Table 1, and in each training iteration, a small batch of target images Z (line 2) are sampled, and the gradient Delta W of the generator is initialized to be 0 (line 3); target image Z t Is fed into the AutoEncoder model to obtain the generated mask M (line 5), the target image Z is obtained t Group Truth mask M * (line 6); by M * -M|| 2 (line 7) estimating the quality of the generated mask M; calculating the gradient of the AutoEncoder parameter W through line 8; finally, W is updated (line 10).
Application example 2-trimming of AutoEncoder model based on ILT
For the invention
And |, represent convolution and element product, respectively. To avoid confusion, all norms | · | | are calculated for the flattened vector. The hopkins partially coherent imaging system theory has been widely applied to mathematical analysis of lithographic masks. Since the Hopkins diffraction model is complex and computationally inconvenient, the original model is approximated by a weighted summation of correlation systems using Singular Value Decomposition (SVD)
Wherein h is k And w k Is the kth core and its weight. As in [7 ]]The invention selects the system
An order approximation. The equation (1) is changed to that,
the lithographic intensity corresponds to the exposure level on the photoresist, which controls the final wafer image through the photoresist model (equation (3))
Wherein Z t Is the target image, Z is the wafer image, M is the mask, | | Z t -Z|| 2 Is the target image Z t And L of Z-flattened vector of wafer image 2 And (4) error.
Although the AutoEncoder can obtain better performance by applying the technology in embodiment 1, it is still a huge challenge to train a complex AutoEncoder model with good convergence. The ILT technology is used as a guide method to finely adjust the parameters of the AutoEncoder. The main goal of ILT is to minimize lithographic errors by gradient descent.
E＝||Z t -Z|| 2 (4)
Wherein Z is the wafer image, Z, for a given mask M t Is the target image. Since the mask M and the wafer image Z are treated as a continuous value matrix in the ILT-based optimization process, the present invention applies a transformed sigmoid function to bring the pixel values close to 0 or 1.
Wherein, I th Is a binary threshold value, M b Is an incomplete binary mask. Combining the analyses of equations (1) - (3) and (4) - (6) can be concluded thatThe lower gradient indicates that the gradient is,
wherein H * Is the conjugate matrix of the original lithography kernel H. In conventional ILT flow, the mask can be optimized by iteratively decreasing the gradient until E is below a threshold.
In the fine adjustment stage of the AutoEncoder, loading the mask M generated by each AutoEncoder into a photoetching engine to obtain a wafer image Z; by minimizing the target image Z t The wafer image Z generated by the mask M generated by the AutoEncoder through the photoetching technology is infinitely close to the target image Z by the error between the wafer image Z and the wafer image Z t 。
In the process of fine-tuning the AutoEncoder, such as Algorithm 2, a small batch of target images Z (line 2) is sampled in each pre-training iteration, and the gradient Δ W of the generator is initialized to be 0 (line 3); the small batch of target images is fed into the AutoEncoder model to obtain the resulting mask (line 5). Loading each generated mask into a lithography machine to obtain a wafer image Z (line 6); estimate the quality of the wafer image by equation (4) (line 7);
by chain rule
Calculating the gradient of the photoetching error E relative to the AutoEncode parameter W (line 8); finally, W is updated (line 10).
FIG. 5 is a schematic structural diagram of a mask optimization apparatus based on a neural network model according to the present invention. In a second aspect of the present invention, a mask optimization apparatus based on a neural network model is provided, as shown in fig. 5, the mask optimization apparatus based on the neural network model of the present invention includes: a sample set generating module 10 configured to acquire a plurality of sets of target images and generate OPC mask images corresponding to the target images by using an OPC method to obtain a data sample set; a first training module 20 configured to train the neural network mask model through a set of data samples; a mask image generation module 30 configured to input the target image into a neural network mask model for training and outputting, and obtain a neural network model mask image; a wafer image generating module 40 configured to load the neural network model mask image into the lithography engine to obtain a wafer image; a second training module 50 configured to optimize the neural network mask model again by minimizing a loss function of the wafer image and the target image.
In a further embodiment, the first training module is further configured to: carrying out mask optimization by using an AutoEnco der neural network model; reducing dimensions of an input target image through a compression function and acquiring abstract features; restoring the abstract features to the mask image through a decompression function; training the weights of the compression function and the decompression function by minimizing the loss function of the mask image and the corresponding OPC mask image.
Specifically, the Autoencor model comprises two processes of encorder (compression) and decoder (decompression), wherein one or more hidden layers are established in the encorder process, dimension reduction is performed on an input current image through an encorder function, and input data are mapped to a feature space to realize extraction of abstract features; in the subsequent decoder process, mapping the abstract features back to the original space through a decoder function to obtain a reconstructed sample; and finally, training the weights of the enc oder function and the decoder function by minimizing the defined loss function, thereby completing the training of the neural network mask model.
In a further embodiment, the mask optimization apparatus based on a neural network model of the present invention further includes: a model building module 60 configured to determine the number of layers of hidden layers and input/output dimensions during a compression process to build a compression function; determining the number of layers of hidden layers and input and output dimensions in the decompression process to construct a decompression function; and configuring activation functions for the compression process and the decompression process, respectively. Wherein the activation function includes, but is not limited to, a sigmoid function.
In a further embodiment, the second training module is further configured to: constructing a photoetching error function of the wafer image and the target image based on an inverse photoetching technology; calculating the gradient of the photoetching error relative to the parameters of the AutoEncoder neural network model by sequentially solving the partial derivatives of the mask and the parameters of the AutoEnco der neural network model; calculating the parameters of the AutoEncoder neural network model when the photoetching error is minimum by a gradient descent method; the AutoEncoder neural network model parameters are composed of compression functions and weight coefficients of decompression functions.
It can be seen from the above embodiments that the mask optimization method based on the neural network model provided by the present invention is divided into two stages, the first stage is to obtain the neural network mask model capable of outputting a mask image close to an OCP mask image after inputting a target image by learning the mask image generated by an Optical Proximity Correction (OPC) technique, and the process thereof refers to fig. 3, and fig. 3 is a schematic diagram of a first-stage tuning process of the mask optimization method based on the neural network model of the present invention; in the second stage, the learning of the optimization of the wafer image is performed through the Inverse Lithography Technology (ILT), but the difference is that the optimization of the wafer image through the Inverse Lithography Technology (ILT) is fed back to the neural network mask model, and the fine tuning of the neural network mask model is realized from another angle. Therefore, the quality of the mask image generated by using the neural network mask model of the invention is further improved, so that the circular crystal image with higher precision can be generated after the mask image is printed on a circular crystal, and the specific process is shown in fig. 4, wherein fig. 4 is a schematic diagram of a second-stage tuning process of the mask optimization method based on the neural network model. Of course, if the wafer image needs to be adjusted and optimized by combining the back lithography technology (ILT) again, the time required by adjustment and optimization can be greatly shortened by using the method provided by the invention.
The foregoing is an exemplary embodiment of the present disclosure, but it should be noted that various changes and modifications could be made herein without departing from the scope of the present disclosure as defined by the appended claims. The functions, steps and/or actions of the method claims in accordance with the disclosed embodiments described herein need not be performed in any particular order. Furthermore, although elements of the disclosed embodiments of the invention may be described or claimed in the singular, the plural is contemplated unless limitation to the singular is explicitly stated.
It should be understood that, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly supports the exception. It should also be understood that "and/or" as used herein is meant to include any and all possible combinations of one or more of the associated listed items.
The numbers of the embodiments disclosed in the embodiments of the present invention are merely for description, and do not represent the advantages or disadvantages of the embodiments.
Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, of embodiments of the invention is limited to these examples; within the idea of an embodiment of the invention, also technical features in the above embodiment or in different embodiments may be combined and there are many other variations of the different aspects of the embodiments of the invention as described above, which are not provided in detail for the sake of brevity. Therefore, any omissions, modifications, substitutions, improvements, and the like that may be made without departing from the spirit and principles of the embodiments of the present invention are intended to be included within the scope of the embodiments of the present invention.
Claims (10)
1. A mask optimization method based on a neural network model, wherein the mask is used for wafer imaging of a lithography system, and the method comprises the following steps:
acquiring a data sample set, wherein the data sample set comprises a plurality of groups of target images and OPC mask images which are generated by utilizing an OPC method and correspond to the target images;
training a neural network mask model through the data sample set;
inputting the target image into the neural network mask model output by training to obtain a neural network model mask image;
loading the neural network model mask image into a photoetching engine to obtain a wafer image;
the neural network mask model is again optimized by minimizing a loss function of the wafer image and the target image.
2. The neural network model-based mask optimization method of claim 1, wherein the method comprises performing mask optimization using an AutoEncoder neural network model, and the training of the neural network mask model by the data sample set comprises:
reducing dimensions of an input target image through a compression function and acquiring abstract features;
restoring the abstract features to a mask image through a decompression function;
the weights of the compression function and the decompression function are trained by minimizing a loss function of the mask image with a corresponding OPC mask image.
3. The neural network model-based mask optimization method of claim 2, further comprising:
determining the number of layers of hidden layers and input and output dimensions in the compression process to construct a compression function;
determining the number of layers of hidden layers and input-output dimensions in the decompression process to construct a decompression function;
and respectively configuring an activation function for the compression process and the decompression process.
4. The neural network model-based mask optimization method of claim 3, wherein the activation function comprises: sigmoid function.
5. The neural network model-based mask optimization method of claim 2, wherein the re-optimizing the neural network mask model by minimizing a loss function of the wafer image and the target image comprises:
constructing a photoetching error function of the round crystal image and the target image based on an inverse photoetching technology;
calculating the gradient of the photoetching error relative to the parameters of the AutoEncoder neural network model by sequentially solving partial derivatives of the mask and the parameters of the AutoEncoder neural network model;
calculating the parameters of the AutoEncoder neural network model when the photoetching error is minimum by a gradient descent method;
the AutoEncoder neural network model parameters are composed of the compression function and the weight coefficient of the decompression function.
6. A mask optimization device based on a neural network model is characterized by comprising:
the system comprises a sample set generation module, a data acquisition module and a data acquisition module, wherein the sample set generation module is used for acquiring a plurality of groups of target images and generating OPC mask images corresponding to the target images by utilizing an OPC method to obtain a data sample set;
a first training module configured to train a neural network mask model through the set of data samples;
the mask image generation module is configured to input the target image into the neural network mask model output by training to obtain a neural network model mask image;
the wafer image generation module is configured to load the neural network model mask image into a lithography engine to obtain a wafer image;
a second training module configured to re-optimize the neural network mask model by minimizing a loss function of the wafer image and the target image.
7. The neural network model-based mask optimization device of claim 6, wherein the first training module is further configured to:
carrying out mask optimization by using an AutoEncoder neural network model;
reducing the dimension of an input target image through a compression function and acquiring abstract characteristics;
restoring the abstract features to a mask image through a decompression function;
the weights of the compression function and the decompression function are trained by minimizing a loss function of the mask image with a corresponding OPC mask image.
8. The neural network model-based mask optimization device of claim 7, further comprising:
the model building module is configured for determining the number of layers of hidden layers and input and output dimensions in a compression process so as to build a compression function;
determining the number of layers of hidden layers and input and output dimensions in the decompression process to construct a decompression function; and
and respectively configuring an activation function for the compression process and the decompression process.
9. The neural network model-based mask optimization device of claim 8, wherein the activation function comprises: sigmoid function.
10. The neural network model-based mask optimization device of claim 7, wherein the second training module is further configured to:
constructing a photoetching error function of the wafer image and the target image based on a reverse photoetching technology;
calculating the gradient of the photoetching error relative to the parameters of the AutoEncoder neural network model by sequentially solving partial derivatives of the mask and the parameters of the AutoEncoder neural network model;
calculating the parameters of the AutoEncoder neural network model when the photoetching error is minimum by a gradient descent method;
the AutoEncoder neural network model parameters are composed of the compression function and the weight coefficient of the decompression function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210902804.5A CN115222829A (en) | 2022-07-29 | 2022-07-29 | Mask optimization method and device based on neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210902804.5A CN115222829A (en) | 2022-07-29 | 2022-07-29 | Mask optimization method and device based on neural network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115222829A true CN115222829A (en) | 2022-10-21 |
Family
ID=83614472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210902804.5A Pending CN115222829A (en) | 2022-07-29 | 2022-07-29 | Mask optimization method and device based on neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115222829A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116720479A (en) * | 2023-08-10 | 2023-09-08 | 腾讯科技（深圳）有限公司 | Mask generation model training method, mask generation method and device and storage medium |
-
2022
- 2022-07-29 CN CN202210902804.5A patent/CN115222829A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116720479A (en) * | 2023-08-10 | 2023-09-08 | 腾讯科技（深圳）有限公司 | Mask generation model training method, mask generation method and device and storage medium |
| CN116720479B (en) * | 2023-08-10 | 2024-01-09 | 腾讯科技（深圳）有限公司 | Mask generation model training method, mask generation method and device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yang et al. | GAN-OPC: Mask optimization with lithography-guided generative adversarial nets | |
| CN110187609B (en) | Deep learning method for calculating photoetching | |
| JP2022518156A (en) | Layout pattern selection method and equipment | |
| US20240037803A1 (en) | Methods and systems for compressing shape data for electronic designs | |
| CN111258176A (en) | Method of fabricating a photolithographic mask for an integrated circuit | |
| Ma et al. | Lithographic source optimization based on adaptive projection compressive sensing | |
| CN108228981B (en) | OPC model generation method based on neural network and experimental pattern prediction method | |
| CN113759657A (en) | Optical proximity correction method | |
| WO2022016802A1 (en) | Physical feature map- and dcnn-based computation method for machine learning-based inverse lithography technology solution | |
| Song et al. | Inverse lithography source optimization via compressive sensing | |
| CN115222829A (en) | Mask optimization method and device based on neural network model | |
| Geng et al. | Truncated nuclear norm minimization based group sparse representation for image restoration | |
| CN110244523B (en) | Integrated photoetching method and photoetching system | |
| CN114326328A (en) | Deep learning-based simulation method for extreme ultraviolet lithography | |
| Wang et al. | Fast pixelated lithographic source and mask joint optimization based on compressive sensing | |
| Shi et al. | Fast and accurate machine learning inverse lithography using physics based feature maps and specially designed DCNN | |
| CN113255799B (en) | Atlas clustering method for mask characteristic graph | |
| CN114326329A (en) | Photoetching mask optimization method based on residual error network | |
| Shi et al. | Physics based feature vector design: a critical step towards machine learning based inverse lithography | |
| Xiao et al. | A novel optical proximity correction (OPC) system based on deep learning method for the extreme ultraviolet (EUV) lithography | |
| Wang et al. | Comparison of different lithographic source optimization methods based on compressive sensing | |
| US11899374B2 (en) | Method for determining an electromagnetic field associated with a computational lithography mask model | |
| CN113189845A (en) | Target pattern mask optimization method based on artificial expectation | |
| Cao et al. | Hybrid deep learning OPC framework with generative adversarial network | |
| Shen et al. | Enhanced DCT2-based inverse mask synthesis with initial SRAF insertion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |