A method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction

Abstract

The present application belongs to the field of lithography and neural networks, and specifically discloses a method for extracting angularly resolved polarized scattering overlay errors based on optical diffraction, comprising: performing polarization differentiation on the frequency domain image of the objective lens back focal plane of the overlay mark to be measured to obtain a target image corresponding to the overlay mark to be measured; inputting the target image into a trained error extraction network to obtain the overlay error value of the overlay mark to be measured; wherein the error extraction network is trained based on training samples and identification labels corresponding to the training samples; the training samples are obtained by utilizing the nanostructure morphology parameters and material optical constants of the overlay mark in combination with pre-examination measurement conditions, and polarized through an established forward optical characteristic model of the overlay mark. The present application can improve the accuracy of overlay error extraction.

Description

Technical Field

[0001] The present application belongs to the field of lithography and neural networks, and more specifically, relates to an angle-resolved polarization scattering overlay error extraction method based on optical diffraction. Background Art

[0002] As semiconductor technology nodes continue to shrink, integrated circuit (IC) devices are increasingly miniaturized and three-dimensional. To ensure proper device functionality, the overlay error of patterns in different photolithography layers must be less than 1 / 5 to 1 / 3 of the feature size, ensuring proper connection between patterns in different photolithography layers. Consequently, the requirements for pattern alignment accuracy are becoming increasingly stringent at advanced technology nodes. Rapid, accurate, and stable measurement and evaluation of overlay error is crucial to ensuring semiconductor device performance.

[0003] Typical overlay error measurement methods are divided into imaging-based overlay (IBO) and diffraction-based overlay (DBO). Both methods require designing overlay marks at the same location on the front and back layers. Overlay measurement equipment locates and measures the marks, and uses the measured information to extract overlay error. With decreasing process nodes, overlay error has gradually exceeded the limits of IBO measurement, leading to its gradual decline in dominance.

[0004] The overlay mark used in the DBO technology is a specially designed periodic nano-grating structure. By detecting the diffraction light signal of the overlay mark, the overlay error to be measured is extracted from it through certain mathematical methods. The DBO method mainly includes eDBO (empirical DBO) and mDBO (model-based DBO). The mDBO method extracts parameters by calling the forward optical model multiple times and compares them with the measured diffraction signal, solving complex partial differential equations in real time, which is difficult to meet the time requirements of online in-situ measurement. The eDBO method uses the fact that the overlay error is within a certain range, and there is a linear relationship between the zero-order light reflectivity R0 of the periodic overlay mark and the overlay error OV as shown below: R0 = K·OV+b;

[0005] By utilizing this empirical linear relationship, we can get rid of the dependence on the forward optical model, thereby achieving rapid extraction of overlay errors. In subsequent studies, it was found that when the overlay error is within a certain range, there is also a linear relationship between the difference in reflectivity of the high-order diffraction light of the periodic overlay mark and the overlay error. eDBO extracts overlay errors based on the local linear relationship of the overlay optical characterization curve. This linear relationship is determined by the overlay mark structure, overlay mark material and measurement conditions. However, under the influence of complex IC manufacturing processes, it is difficult for the overlay mark to guarantee the ideal design morphology and measurement conditions, which leads to poor accuracy in overlay error extraction.

[0006] Therefore, how to improve the accuracy of overlay error extraction is a technical problem that needs to be solved urgently. Summary of the Invention

[0007] In view of the defects of the prior art, the purpose of this application is to provide an angle-resolved polarization scattering overlay error extraction method based on optical diffraction, aiming to solve the problem of inaccurate overlay error extraction.

[0008] To achieve the above objectives, the present application provides an angle-resolved polarization scattering overlay error extraction method based on optical diffraction, comprising:

[0009] Performing polarization differentiation on the frequency domain image of the objective lens back focal plane of the overlay mark to be measured to obtain a target image corresponding to the overlay mark to be measured;

[0010] Inputting the target image into the trained error extraction network to obtain the overlay error value of the overlay mark to be tested;

[0011] In which, the error extraction network is trained based on training samples and identification labels corresponding to the training samples; the training samples are obtained by using the nanostructure morphology parameters and material optical constants of the overlay mark combined with preliminary measurement conditions, and polarized through the established overlay mark forward optical property model.

[0012] The error extraction network trained in this application can automatically extract and analyze overlay errors from polarization difference images. Compared to traditional manual or semi-manual measurements, it can obtain overlay error information more quickly and accurately. Furthermore, the error extraction network is learned from a large number of training samples, so it can comprehensively consider the variations and error types of different overlay marks, improving the precision and accuracy of error assessment.

[0013] Optionally, the training method of the error extraction network includes:

[0014] Acquire a training sample and determine a theoretical value of an overlay error corresponding to the training sample, and use the theoretical value of the overlay error as the identification label;

[0015] Determining a neural network structure of an initial network, inputting a training set of the training samples into the initial network, and obtaining an overlay error extraction value corresponding to the training set;

[0016] Constructing a loss function based on the overlay error extraction value and the error theoretical value to obtain a training loss value;

[0017] The initial network is iteratively updated based on a preset number of training cycles and the number of input images per batch until the training loss value is less than a preset value, and the neural network weight file information is determined to obtain a trained error extraction network.

[0018] Optionally, obtaining a training sample and determining a theoretical value of an overlay error corresponding to the training sample includes:

[0019] Substituting the nanostructure morphology parameters, material optical constants, and preset measurement conditions of the overlay mark into the amplitude-divided analyzer angle-resolved scatterometer system model, a forward optical characteristic model of the overlay mark based on rigorous coupled wave analysis is obtained. Based on the forward optical characteristic model of the overlay mark, the incident light is analyzed to obtain a frequency domain image of the back focal plane of the cross-polarized objective lens of the overlay error.

[0020] According to the nanostructure morphology parameters and the forward optical characteristic model of the overlay mark, randomly taking values ​​of the structural parameters of the overlay mark within a preset deviation range to obtain a frequency domain image of the back focal plane of the cross-polarization objective lens and an overlay error;

[0021] The frequency domain image of the back focal plane of the cross-polarized objective lens is used as a training sample, and the overlay error is used as a theoretical extraction value.

[0022] By substituting the nanostructure topography parameters and the material optical constants into a rigorous coupled wave analysis model, the forward optical properties of the overlay mark can be accurately calculated. This approach not only considers the geometry of the overlay mark but also the optical response of the material, allowing for more accurate predictions of its optical behavior under different polarization states.

[0023] This application effectively detects and quantifies overlay errors by analyzing frequency domain images of the back focal plane of the cross-polarized objective lens. By randomly selecting values ​​for the structural parameters of the overlay mark within a preset deviation range, it can simulate the diverse error scenarios that may occur in real-world production. Using the obtained frequency domain images of the back focal plane of the cross-polarized objective lens as training samples, an error extraction network can be trained. This method can automatically extract and quantify errors from the image data of the overlay mark, improving the efficiency and accuracy of error assessment.

[0024] Optionally, the incident light is analyzed based on the forward optical characteristic model of the overlay mark to obtain a frequency domain image of the back focal plane of the cross-polarization objective lens of the overlay error, as shown in the following formula:

[0025]

[0026] in, represents the azimuth of the incident light, E input represents the electric field vector of the incident light, θ represents the incident angle, P1 represents the polarizer, P2 represents the polarization beam splitter, Indicates the rotation angle of the coordinate system relative to the incident surface The rotation matrix, J S represents the Jones matrix of the overlay mark, P1 represents the polarization beam splitter, J M Represents the reflection transformation matrix.

[0027] Specifically, each Jones matrix has the following form:

[0028]

[0029] Optionally, the nanostructure morphology parameters include: the line width of the top grating, the line width of the bottom grating, the period of the top grating, the period of the bottom grating, the wall height of the top grating, the wall height of the bottom grating, the thickness of the intermediate thin film layer, the theoretical value of the overlay error, the left wall angle and the right wall angle of the top grating and the bottom grating, and the number of slices; the material optical constant refers to the complex refractive index of the material, and the measurement conditions include the wavelength of the detection light, the polarization angle of the polarizer, the polarization beam splitter, and the numerical aperture of the objective lens.

[0030] Optionally, the process of acquiring the training samples specifically includes:

[0031] Based on the forward optical characteristic model of the overlay mark, the incident S light is passed through the polarization beam splitter to obtain the co-polarization pupil plane frequency domain image of SS and the cross-polarization pupil plane frequency domain image of SP;

[0032] The cross-polarization pupil plane frequency domain image is polarization-differentiated and used as a training sample.

[0033] Optionally, the loss function during the error extraction network training process is as shown in the following formula:

[0034]

[0035] Where MSE represents the training loss value, f(x) represents the overlay error extraction value of the nth training sample, y is the theoretical value of the overlay error, and n is the number of training samples.

[0036] Optionally, the error extraction network is a ConvNeXt network; the ConvNeXt network further includes:

[0037] Attention mechanism layer, used to enhance the network's response to features at specific locations in the image;

[0038] Global pooling layer, used to reduce the spatial dimension of features;

[0039] The fully connected layer is used to map the learned features to the final predicted overlay error value.

[0040] In a second aspect, the present application provides an electronic device comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.

[0041] In a third aspect, the present application provides a computer-readable storage medium storing a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0042] In a fourth aspect, the present application provides a computer program product. When the computer program product runs on a processor, it enables the processor to execute the method described in the first aspect or any possible implementation of the first aspect.

[0043] It can be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0044] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the existing technologies:

[0045] (1) The error extraction network trained in this application can automatically extract and analyze overlay errors from polarization differential images. Compared to traditional manual or semi-manual measurements, it can obtain overlay error information more quickly and accurately. Furthermore, the error extraction network is learned from a large number of training samples and can therefore comprehensively consider the variations and error types of different overlay marks, thereby improving the precision and accuracy of error assessment.

[0046] (2) In the present application, the polarization difference of the frequency domain image of the rear focal plane of the objective lens can provide a high-resolution representation of the overlay mark, and the subtle structural features of the overlay mark can be identified and quantified by the image difference under different polarization states. The polarization difference can enhance the detailed information of the overlay mark in the image, which is helpful for detecting and analyzing overlay errors, such as shape deviation, size error, etc.

[0047] (3) The method provided by this application does not rely on the traditional method of solving the overlay error by using the linear relationship between the positive and negative first-order light intensity difference and the overlay error, which requires the introduction of a preset deviation ±D. It can realize the overlay error measurement in a single direction based on a unit, and can achieve a smaller overlay mark area, thereby increasing the transistor density of the functional area per unit chip area. In addition, the neural network model trained by this method can resist the influence of certain changes in the overlay mark morphology parameters (process deviation) on the overlay error measurement results, and the extraction of overlay error is fast, accurate, and robust. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is one of the flow charts of the method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction provided in an embodiment of the present application;

[0049] Figure 2 This is the second flow chart of the method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction provided in an embodiment of the present application;

[0050] Figure 3 This is a schematic diagram of the morphological parameters and materials of the overlay mark nanostructure provided in the embodiments of the present application;

[0051] Figure 4 Schematic diagram of the principle of an amplitude-divided analyzer angle-resolved scatterometer provided in an embodiment of the present application;

[0052] Figure 5 The middle left image is the frequency domain image of the cross-polarization pupil plane with overlay marks, and the right image is the frequency domain image of the cross-polarization pupil plane after polarization difference;

[0053] Figure 6 This is the network structure diagram of the improved ConvNeXt;

[0054] Figure 7 The extraction result of overlay error data on the trained neural network;

[0055] Figure 8 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0057] The term "and / or" as used herein describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. The symbol " / " as used herein indicates that the related objects are in an "or" relationship, for example, A / B means either A or B.

[0058] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0059] Next, the technical solutions provided in the embodiments of this application are introduced.

[0060] Reference Figure 1 The present application provides a method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction, comprising:

[0061] Step S101. Perform polarization differentiation on the frequency domain image of the back focal plane of the objective lens of the overlay mark to be measured, to obtain a target image corresponding to the overlay mark to be measured;

[0062] Step S102: inputting the target image into the trained error extraction network to obtain the overlay error value of the overlay mark to be tested;

[0063] In which, the error extraction network is trained based on training samples and identification labels corresponding to the training samples; the training samples are obtained by using the nanostructure morphology parameters and material optical constants of the overlay mark combined with preliminary measurement conditions, and polarized through the established overlay mark forward optical property model.

[0064] In step S101, the overprinted mark typically consists of nanostructures, whose optical properties can be observed using the frequency domain image of the back focal plane of the objective lens. This image, obtained by focusing light onto the back focal plane of the objective lens, is used to demonstrate the mark's complex optical reflection and scattering properties. The overprinted mark may be polarization-dependent, meaning that its frequency domain image of the back focal plane of the objective lens will differ under different polarization states. Polarization differentiation of these images can be used to separate the optical properties under different polarization states, facilitating more accurate analysis of the mark's fine structure and surface topography.

[0065] After polarization differential processing, the target image obtained has richer information, showing the reflection and scattering of the overlay mark under different polarization states. The target image will be used as input data for subsequent error evaluation.

[0066] Step S102 is the application of the error extraction network. The general process involves loading the trained neural network weights, performing polarization differentiation on the frequency domain image of the objective lens back focal plane of the overlay mark to be measured, and then inputting the trained neural network to extract the overlay error value. The error extraction network in this embodiment is a deep learning model trained to extract overlay mark error information from the target image. By learning from training samples, the network is able to identify and quantify different types of overlay errors.

[0067] The training samples in this embodiment were obtained through simulation and experimentation, and include the nanostructure morphology parameters and material optical constants of the overlay mark, combined with preset measurement conditions. These parameters were then used to perform polarization calculations using an established forward optical property model of the overlay mark, generating a simulated frequency domain image of the objective lens' back focal plane.

[0068] During training, the error extraction network learns how to identify and extract information related to the mark overlay quality from the frequency domain image of the objective lens back focal plane, enabling the network to accurately predict the error value of the actual overlay mark, thus providing a basis for quality control and adjustments in the manufacturing process.

[0069] Specifically, refer to Figure 2 , the training process of the error extraction network is as follows:

[0070] Step S201: Acquire a training sample and determine a theoretical value of overlay error corresponding to the training sample, and use the theoretical value of overlay error as the identification label;

[0071] Step S202: Determine the neural network structure of the initial network, input the training set of the training samples into the initial network, and obtain the overlay error extraction value corresponding to the training set;

[0072] Step S203: constructing a loss function based on the overlay error extraction value and the error theoretical value to obtain a training loss value;

[0073] Step S204. The initial network is iteratively updated based on a preset number of training cycles and the number of input images per batch, until the training loss value is less than a preset value, the neural network weight file information is determined, and a trained error extraction network is obtained.

[0074] Furthermore, in the above step S201, obtaining the training sample and determining the theoretical value of the overlay error corresponding to the training sample includes:

[0075] Substituting the nanostructure morphology parameters, material optical constants, and preset measurement conditions of the overlay mark into the amplitude-divided analyzer angle-resolved scatterometer system model, a forward optical characteristic model of the overlay mark based on rigorous coupled wave analysis is obtained. Based on the forward optical characteristic model of the overlay mark, the incident light is analyzed to obtain a frequency domain image of the back focal plane of the cross-polarized objective lens of the overlay error.

[0076] According to the nanostructure morphology parameters and the forward optical characteristic model of the overlay mark, randomly taking values ​​of the structural parameters of the overlay mark within a preset deviation range to obtain a frequency domain image of the back focal plane of the cross-polarization objective lens and an overlay error;

[0077] The frequency domain image of the back focal plane of the cross-polarized objective lens is used as a training sample, and the overlay error is used as a theoretical extraction value.

[0078] Specifically, refer to Figure 3 In step S201, the nanostructure morphology parameters of the overlay mark include the line width CD1 of the top grating, the line width CD2 of the bottom grating, the period Pitch of the top grating and the bottom grating, the wall height H1 of the top grating, the wall height H3 of the bottom grating, the thickness H2 of the intermediate film layer, the theoretical value of the overlay error, the left wall angle LSWA and the right wall angle RSWA of the top grating and the bottom grating, and the number of slices slice_num (the trapezoidal grating can be differentially sliced ​​into multiple layers of gratings with different line widths); the material optical constant refers to the complex refractive index of the material, and the measurement conditions include the wavelength of the probe light, the polarization angle p1 of the polarizer, the polarization beam splitter PBS, and the numerical aperture NA of the objective lens.

[0079] Reference Figure 4 The light emitted by the light source is collimated and then passes through a high NA objective lens before converging on the sample surface. After diffraction by the sample, the diffracted light of different orders containing the sample's morphological information is collected by the objective lens and imaged onto the main detection optical path CCD through a relay lens. An auxiliary optical path is added according to the measurement requirements to determine the position of the overlay mark. The polarization beam splitter PBS will split the pupil plane frequency domain image into a co-pol co-polarization pupil plane frequency domain image and a cross-pol cross-polarization pupil plane frequency domain image. The co-pol co-polarization pupil plane frequency domain image contains a stronger signal, and its symmetric information contains the sample's morphological structure information, including height, critical dimension CD, material properties, etc., while the cross-pol cross-polarization pupil plane frequency domain image contains a weaker signal intensity, and its asymmetric information usually includes overlay information. If each device is regarded as an ideal device, the only devices that change the properties of the beam are the polarizer, the sample to be measured, and the polarization beam splitter. If the electric field intensity vector is used to describe the incident beam, the Jones matrix is ​​used to describe the system model according to the order in which the beam passes through each device, as shown in the following formula:

[0080]

[0081] in, represents the azimuth of the incident light, E input represents the electric field vector of the incident light, θ represents the incident angle, P1 represents the polarizer, P2 represents the polarization beam splitter, Indicates the rotation angle of the coordinate system relative to the incident surface The rotation matrix, J S represents the Jones matrix of the overlay mark, P1 represents the polarization beam splitter, J M Represents the reflection transformation matrix.

[0082] Specifically, each Jones matrix has the following form:

[0083]

[0084] Optionally, the process of acquiring the training samples specifically includes:

[0085] Based on the forward optical characteristic model of the overlay mark, the incident S light is passed through the polarization beam splitter to obtain the co-polarization pupil plane frequency domain image of SS and the cross-polarization pupil plane frequency domain image of SP;

[0086] The cross-polarization pupil plane frequency domain image is polarization-differentiated and used as a training sample.

[0087] Specifically, refer to Figure 5 The incident light uses S light, and the polarization beam splitter splits it into the Co-pol co-polarization pupil frequency domain image of SS and the Cross-pol cross-polarization pupil frequency domain image of SP. The Cross-pol cross-polarization pupil frequency domain image of SP is used as the training sample after polarization difference.

[0088] It should be noted that the signal strength of the cross-polarization pupil frequency domain image of the SP or PS is relatively weak, and its asymmetric information usually includes overlay information. Using a polarization beam splitter to separate the co-polarization pupil frequency domain image with a stronger signal strength reduces its ability to overwhelm the asymmetric overlay information. Secondly, simulations show that the cross-polarization pupil frequency domain image of the SP is more sensitive to overlay error OV than the cross-polarization pupil frequency domain image of the PS. Finally, after using polarization difference, the asymmetric information of the image is more obvious. Therefore, the training dataset uses the polarization-differenced cross-polarization pupil frequency domain image of the SP.

[0089] Optionally, the loss function during the error extraction network training process is as shown in the following formula:

[0090]

[0091] Where MSE represents the training loss value, f(x) represents the overlay error extraction value of the nth training sample, y is the theoretical value of the overlay error, and n is the number of training samples.

[0092] It should be noted that the mean square error loss function is usually smoother than the cross entropy loss function, which can make the training process more stable and reduce the risk of gradient explosion or disappearance. The mean square error loss function has a larger penalty for outliers and is more susceptible to outliers than the cross entropy loss function.

[0093] Optionally, the error extraction network is a ConvNeXt network; the ConvNeXt network further includes:

[0094] Attention mechanism layer, used to enhance the network's response to features at specific locations in the image;

[0095] Global pooling layer, used to reduce the spatial dimension of features;

[0096] The fully connected layer is used to map the learned features to the final predicted overlay error value.

[0097] Specifically, refer to Figure 6 The network structure in this embodiment is an improved ConvNeXt network, which specifically involves connecting a fully connected layer with an output node of 1 to the end of the original network structure to output the predicted overlay error value OV', introducing a CA coordinate attention mechanism, and adding it before the last global pooling layer. The position information is encoded into the channel attention, allowing the network to autonomously learn important regional features, improving the network's perception of effective information features, and thus allocating limited information processing resources to more important parts.

[0098] The following is a complete embodiment of the present application, and the specific implementation process is as follows:

[0099] 1. Determine the nanostructure morphology parameters and material optical constants of the overlay mark. The linewidth of the top and bottom gratings is 300 nm, the period pitch of both is 600 nm, the ideal wall height H1 of the top grating is 150 nm, the ideal wall height H3 of the bottom grating is 150 nm, the ideal thickness H2 of the intermediate film layer is 30 nm, and the ideal left wall angle LSWA and ideal right wall angle RSWA of both the top and bottom gratings are 90°. The top grating is made of photoresist, with a ridge refractive index nrd of 3 and a groove refractive index ngr of 1.0. The intermediate layer is made of silicon dioxide with a refractive index of 1.551. The bottom grating is made of silicon, with a ridge refractive index nrd of 1.48 and a groove refractive index ngr of 1.0. Load the corresponding material model file and calculate the complex refractive index of each layer.

[0100] 2. Based on the overlay mark nanostructure morphology parameters and material optical constants determined in step 1, determine the measurement conditions: a probe light wavelength of 532 nm, an objective lens numerical aperture (NA) of 0.95, a polarizer polarization angle p1 of 90° (i.e., S-light incident light), and a polarization beam splitter (PBS) setting of 0°, to obtain SP cross-pol polarized incident light. A model of the overlay mark forward optical properties based on rigorous coupled-wave analysis is established and incorporated into the system model of the amplitude-divided analyzer angle-resolved scatterometer.

[0101] The top grating wall height H1 was randomly set between 148 nm and 150 nm, the middle film layer thickness H2 was randomly set between 29 nm and 31 nm, and the bottom grating wall height H3 was randomly set between 148 nm and 150 nm. The left and right sidewall angles LSWA and RSWA were both randomly set between 85° and 90° to simulate random process errors. The overlay error (OV) varied between -10 nm and 10 nm. Fifty images were generated for each overlay error (OV) value, for a total of X0 = 1050 training samples. The simulated error range can be selected based on actual error conditions or empirical values.

[0102] 3. Determine the neural network structure. The neural network used in this embodiment to extract the overlay error is the improved ConNeXt. The end of the original network structure is connected to a fully connected layer with an output node of 1 to output the predicted overlay error value OV'. The Coordinate Attention mechanism is introduced and added before the last global pooling layer. Figure 4 The improved network structure diagram is shown in FIG. A loss function is established to characterize the deviation between the extracted value OV' of the overlay error and the theoretical value OV of the overlay error, and the mean square error loss function (MSE Loss) is used.

[0103] 4. The training dataset was split into a training set and a validation set in a ratio of 7:3. Iterative optimization was performed using the Adam optimization algorithm, with a learning rate of 0.0002, 3000 epochs, and 16 batches. The default loss value was set to 0.01. When the loss value first fell below 0.01, the training weight file was stored and recorded as the best loss value. When the subsequent loss value fell below the current best loss value, the current training weight file was stored, ultimately obtaining the neural network weights with the lowest loss value after training.

[0104] 5. Load the trained neural network weights, perform polarization differential on the frequency domain image of the objective lens back focal plane of the overlay mark to be measured, input it into the trained neural network, and extract the predicted overlay error value OV'.

[0105] Reference Figure 7 , Figure 7 The results of extracting some untrained overlay error data using a trained neural network are shown. As can be seen from the figure, the correlation coefficient and linear relationship are relatively stable, indicating that using this model for error identification can improve the accuracy of error identification results.

[0106] Reference Figure 8 Based on the methods in the above embodiments, an embodiment of the present application provides an electronic device, which may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call logic instructions in the memory 830 to execute the methods in the above embodiments.

[0107] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0108] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0109] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0110] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0111] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0112] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted via the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state drive (SSD)).

[0113] It will be understood that the various numerical numbers involved in the embodiments of the present application are merely distinctions for the convenience of description and are not intended to limit the scope of the embodiments of the present application.

[0114] It is easy for those skilled in the art to understand that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction, characterized in that: include: Performing polarization differentiation on the frequency domain image of the back focal plane of the objective lens of the overprint mark to be measured to obtain a target image corresponding to the overprint mark to be measured; Inputting the target image into the trained error extraction network to obtain the overlay error value of the overlay mark to be tested; The error extraction network is trained based on training samples and identification labels corresponding to the training samples; the training samples are polarized using the nanostructure morphology parameters and material optical constants of the overlay mark in combination with preset measurement conditions and an established forward optical property model of the overlay mark; The identification label is determined based on a theoretical value of an overlay error corresponding to a training sample, and a method for obtaining the theoretical value of the overlay error includes: Substituting the nanostructure morphology parameters, material optical constants, and preset measurement conditions of the overlay mark into the amplitude-divided analyzer angle-resolved scatterometer system model, a forward optical characteristic model of the overlay mark based on rigorous coupled wave analysis is obtained. Based on the forward optical characteristic model of the overlay mark, the incident light is analyzed to obtain a frequency domain image of the back focal plane of the cross-polarized objective lens of the overlay error. According to the nanostructure morphology parameters and the forward optical characteristic model of the overlay mark, randomly taking values ​​of the structural parameters of the overlay mark within a preset deviation range to obtain a frequency domain image of the back focal plane of the cross-polarization objective lens and an overlay error; The frequency domain image of the back focal plane of the cross-polarized objective lens is used as a training sample, and the overlay error is used as a theoretical extraction value.

2. The method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction according to claim 1, characterized in that: The training method of the error extraction network includes: Acquire a training sample and determine a theoretical value of an overlay error corresponding to the training sample, and use the theoretical value of the overlay error as the identification label; Determining a neural network structure of an initial network, inputting a training set of the training samples into the initial network, and obtaining an overlay error extraction value corresponding to the training set; Constructing a loss function based on the overlay error extraction value and the error theoretical value to obtain a training loss value; The initial network is iteratively updated based on a preset number of training cycles and the number of input images per batch until the training loss value is less than a preset value, and the neural network weight file information is determined to obtain a trained error extraction network.

3. The method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction according to claim 2, characterized in that: The incident light is analyzed based on the forward optical characteristic model of the overlay mark to obtain the frequency domain image of the back focal plane of the cross-polarization objective lens of the overlay error, as shown in the following formula: in, represents the electric field vector of the outgoing light, represents the azimuth of the incident light, represents the angle of incidence, represents the electric field vector of the incident light, represents polarizer, P2 represents polarization beam splitter, Indicates the rotation angle of the coordinate system relative to the incident surface The rotation matrix of represents the Jones matrix of the overlay mark, represents the polarization beam splitter, represents the reflection transformation matrix; Specifically, each Jones matrix has the following form: 。 4. The method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction according to claim 1, wherein: The nanostructure morphology parameters include: the line width of the top grating, the line width of the bottom grating, the period of the top grating, the period of the bottom grating, the wall height of the top grating, the wall height of the bottom grating, the thickness of the intermediate thin film layer, the theoretical value of the overlay error, the left wall angle and the right wall angle of the top grating and the bottom grating, and the number of slices; the material optical constant refers to the complex refractive index of the material, and the measurement conditions include the wavelength of the detection light, the polarization angle of the polarizer, the polarization beam splitter, and the numerical aperture of the objective lens.

5. The method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction according to claim 1, characterized in that: The process of obtaining the training samples specifically includes: Based on the forward optical characteristic model of the overlay mark, the incident S light is passed through the polarization beam splitter to obtain the co-polarization pupil plane frequency domain image of SS and the cross-polarization pupil plane frequency domain image of SP; The cross-polarization pupil plane frequency domain image is polarization-differentiated and used as a training sample.

6. The method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction according to claim 1, characterized in that: The loss function during the error extraction network training process is shown in the following formula: Among them, MSE represents the training loss value, represents the overlay error extraction value of the nth training sample, is the theoretical value of overlay error, and n is the number of training samples.

7. The method for extracting angle-resolved polarization scattering overlay errors based on optical diffraction according to claim 1, characterized in that: The error extraction network is a ConvNeXt network; The ConvNeXt network also includes: Attention mechanism layer, used to enhance the network's response to features at specific locations in the image; Global pooling layer, used to reduce the spatial dimension of features; The fully connected layer is used to map the learned features to the final predicted overlay error value.

8. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1 to 7.

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

Images