Training method for multiple solution problems in machine learning, X-ray measurement method and device

Abstract

The training method for multiple solutions in machine learning, the X-ray measurement method and device are applied in the field of nano-measurement technology. X-rays are used to obtain the scattering spectrum of the target part to be detected; the scattering spectrum is analyzed by the measurement model to obtain the key size of the device to be detected; the measurement model is obtained by inputting the data set of the sample device into the measurement model to be trained based on the joint loss function; the loss factor of the joint loss function includes the loss factor of the predicted value based on the key size of the sample device, the Babinnet complementary loss factor of the predicted value, the centrosymmetric loss factor of the predicted value, and the Babinnet complementary centrosymmetric loss factor of the predicted value. By combining physical prior information to construct a joint loss function in the model training process, the loss of all possible structures corresponding to the same scattering spectrum is calculated, and the multiple solution problem is solved. The trained measurement model can perform real-time and high-precision X-ray scattering measurement calculations.

Description

Technical Field

[0001] The present invention relates to the field of nano-measurement technology, and in particular to a training method for multi-solution problems in machine learning, and an X-ray measurement method and device. Background Art

[0002] X-ray scattering measurement technology is a powerful tool for measuring the critical dimension (CD) of nanodevices.

[0003] However, due to the Babinet theorem and the lack of phase information in the scattering process, the same scattering pattern will correspond to multiple possible target structures. The prediction result in the conventional deep learning training process is only one of the multiple possible target structures, and it is impossible to accurately obtain the same configuration as the target structure. Therefore, when using conventional deep learning training methods to train models, when dealing with complex structures containing a large number of parameters, problems such as long iteration time, high computational cost, and poor measurement accuracy will occur.

[0004] Therefore, how to solve the multiple solution problem in the model training process and realize the high-precision, real-time X-ray scattering measurement and calculation of the key dimensions of nanodevices using deep learning methods is an urgent problem to be solved. Summary of the invention

[0005] The embodiment of the present application provides an X-ray scattering measurement method for realizing high-precision and real-time measurement and calculation of key dimensions of a target part to be inspected.

[0006] In a first aspect, an embodiment of the present application provides an X-ray scattering measurement method, comprising:

[0007] Use X-rays to obtain the scattering spectrum of the target part to be inspected;

[0008] Analyzing the scattering pattern using a measurement model to obtain a critical dimension of the device to be detected;

[0009] The measurement model is obtained by inputting a data set of sample devices into the measurement model to be trained, and training the measurement model to be trained based on a joint loss function; the loss factors in the joint loss function include a loss factor based on a predicted value of a critical dimension of the sample device, a Babinet complementary loss factor of the predicted value, a centrally symmetric loss factor of the predicted value, and a Babinet complementary centrally symmetric loss factor of the predicted value.

[0010] By combining physical prior information with the training process of the measurement model to construct a joint loss function and calculate the losses of all possible structures corresponding to the same scattering spectrum, the problem of premature model convergence caused by multiple solutions is solved, and the training of deep learning models with low loss values ​​is achieved. The trained measurement model can be used to perform real-time and high-precision X-ray scattering measurements of nanostructured samples.

[0011] In some embodiments, the calculation formula of the joint loss function is as follows:

[0012]

[0013] Among them, y p The critical dimensions of the sample device output by the model prediction are: for y p The corresponding original structural loss factor, y pb The sample device has y p The critical dimension when the value of the Babinet complementary structure is ; for y pb The corresponding Babinet complementary loss factor, y pc The sample device has y p The critical dimension when the value is a centrosymmetric structure; for y pc The corresponding centrosymmetric loss factor; y pbc The sample device has y p The critical dimensions of the Babinet complementary and center-rotation structures are: for y pbc The corresponding Babinet complementary centrosymmetric loss factor; y t is the label value of the sample device.

[0014] In some embodiments, the joint loss function further includes a total variation regularization term, which is used to enhance the continuity between key size data points of the continuous distribution.

[0015] In some embodiments, the calculation formula of the total variation regularization term is as follows:

[0016]

[0017] in, is the total variation regularization term, N is the number of consecutive adjacent key size data points, y pi is the value corresponding to the i-th key size data point.

[0018] In some embodiments, the data set of the sample device is constructed by:

[0019] Generate a sample device through computer simulation and determine the critical dimensions of the sample device;

[0020] Performing data processing on the sample device to obtain a scattering spectrum of the sample device;

[0021] The sample device is processed by zero mean value to update the critical dimensions of the sample device related to the relative position coordinates, thereby forming a scattering spectrum of the sample device and a sample corresponding to the critical dimensions of the sample device.

[0022] In some embodiments, the performing data processing on the sample device to obtain a scattering spectrum of the sample device includes:

[0023] Based on the scattering spectrum intensity distribution, obtaining scattering data of the sample device at multiple angles;

[0024] The multi-angle scattering data are clipped to at least partially remove repeated data and zero-value point data in the scattering data, and the clipped scattering data are spliced ​​to obtain a scattering spectrum of the sample device.

[0025] In some embodiments, the scattering spectrum intensity distribution is calculated by first-order Born approximation and Fourier transform, and the calculation formula is as follows:

[0026]

[0027] Where q represents the scattering vector, r represents the position vector, ρ(r) is the grating shape function, P is the grating constant, and β is the scale factor. bkg (q) is the background noise, I noise (q) is other noise.

[0028] In a second aspect, an embodiment of the present application provides an X-ray scattering measurement device, comprising:

[0029] An acquisition module, used to acquire a scattering spectrum of a target object to be detected by using X-rays;

[0030] A processing module, used for analyzing the scattering spectrum using a measurement model to obtain a critical dimension of the device to be detected;

[0031] The processing module is also used for the measurement model, which is obtained by inputting a data set of sample devices into the measurement model to be trained, and training the measurement model to be trained based on a joint loss function; the joint loss function is used to select a loss factor of the predicted value of the key dimension of the device to be tested, the Babinet complementary loss factor of the predicted value, the central symmetric loss factor of the predicted value, and the Babinet complementary central symmetric loss factor of the predicted value.

[0032] By combining physical prior information with the training process of the measurement model to construct a joint loss function and calculate the losses of all possible structures corresponding to the same scattering spectrum, the problem of premature model convergence caused by multiple solutions is solved, and the training of deep learning models with low loss values ​​is achieved. The trained measurement model can be used to perform real-time and high-precision X-ray scattering measurements of nanostructured samples.

[0033] In some embodiments, the processing module is specifically used to:

[0034] The calculation formula of the joint loss function is as follows:

[0035]

[0036] Among them, y p The critical dimensions of the sample device output by the model prediction are the original structure. for y p The corresponding original structural loss factor, y pb The sample device has y p The critical dimension when the value of the Babinet complementary structure is ; for y pb The corresponding Babinet complementary loss factor, y pc The sample device has y p The critical dimension when the value is a centrosymmetric structure; for y pc The corresponding centrosymmetric loss factor; y pbc The sample device has y p The critical dimensions of the Babinet complementary and center-rotation structures are: for y pbc The corresponding Babinet complementary centrosymmetric loss factor; y t is the label value of the sample device.

[0037] In some embodiments, the processing module is specifically used to:

[0038] The joint loss function also includes a total variation regularization term, which is used to strengthen the continuity between key size data points of the continuous distribution.

[0039] In some embodiments, the processing module is specifically used to:

[0040] The calculation formula of the total variation regularization term is as follows:

[0041]

[0042] in, is the total variation regularization term, N is the number of consecutive adjacent key size data points, ypi is the value corresponding to the i-th key size data point.

[0043] In some embodiments, the processing module is specifically used to:

[0044] The data set of the sample device is constructed in the following manner, including:

[0045] Generate a sample device by simulation and determine the position coordinates and key dimensions of the sample device;

[0046] Performing data processing on the sample device to obtain a scattering spectrum of the sample device;

[0047] The sample device is processed by zero mean value to update the critical dimension of the sample device related to the relative position coordinates, thereby forming a scattering spectrum of the sample device and a sample corresponding to the position coordinates and the critical dimension of the sample device.

[0048] In some embodiments, the processing module is specifically used to: perform data processing on the sample device to obtain a scattering spectrum of the sample device, including:

[0049] Based on the scattering spectrum intensity distribution, obtaining scattering data of the sample device at multiple angles;

[0050] The repeated data and zero-value point data in the multi-angle scattering data are clipped, and the clipped multi-angle scattering data are spliced ​​to obtain the scattering spectrum of the sample device, and the scattering spectrum is normalized.

[0051] In some embodiments, the processing module is specifically used to calculate the intensity distribution of the scattering spectrum by first-order Born approximation and Fourier transform, and the calculation formula is as follows:

[0052]

[0053] Where q represents the scattering vector, r represents the position vector, ρ(r) is the grating shape function, P is the grating constant, β is the scale factor, and I bkg (q) is the background noise, I noise (q) is other noise.

[0054] In a third aspect, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes any of the X-ray scattering measurement methods described in the first aspect.

[0055] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program executable by a computer device. When the program is run on the computer device, the computer device executes any X-ray scattering measurement method described in the first aspect above. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0057] Figure 1 A schematic diagram of a flow chart of an X-ray scattering measurement method provided by an embodiment of the present invention;

[0058] Figure 2 A schematic diagram of the structure of a transmission small-angle X-ray scattering measurement and detection system provided by an embodiment of the present invention;

[0059] Figure 3 A schematic diagram of a model architecture of a measurement model provided by an embodiment of the present invention;

[0060] Figure 4 A schematic diagram of error distribution of a key dimension prediction result provided by an embodiment of the present invention;

[0061] Figure 5 A schematic cross-sectional view of a nano-grating sample structure and three structures having the same scattering spectrum provided by an embodiment of the present invention;

[0062] Figure 6 A schematic diagram of curve comparison of an unimproved MAE loss function and an improved joint loss function trained on the same data set and model provided in an embodiment of the present invention;

[0063] Figure 7 A schematic diagram of X-ray scattering data of a sample at a single angle provided by an embodiment of the present invention;

[0064] Figure 8 A schematic diagram of a final scattering spectrum provided by an embodiment of the present invention;

[0065] Fig. 9 A schematic diagram of a flow chart of an X-ray scattering measurement method provided by an embodiment of the present invention;

[0066] Fig.10 A schematic diagram of the structure of an X-ray scattering measurement device provided in an embodiment of the present invention;

[0067] Fig.11 A schematic diagram of the structure of a computing device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] First, the nouns mentioned in this application are further explained:

[0070] Critical Dimension (CD) is usually used to describe the most critical dimensional parameters on semiconductor chips or other microelectronic components. These dimensional parameters are crucial to the performance and function of semiconductor devices and may include line width, spacing, pattern position and alignment on photomasks or wafers.

[0071] Babinet principle: In Fraunhofer diffraction, the diffraction patterns produced by two diffraction screens with complementary amplitudes on the receiving screen far away from the diffraction center are the same.

[0072] Phase information missing: Information missing refers to the lack of phase information when converting measurement data into a spectrum during X-ray scattering measurements of nanodevices, resulting in the spectrum not containing position information but only structural information.

[0073] Zero mean standardization: It is a commonly used preprocessing method that can convert data into a set of data with a mean of zero, eliminating the error caused by the mean and making it easier to compare and analyze.

[0074] Small Angle X-ray Scattering (SAXS) refers to the scattering phenomenon that occurs when X-rays are irradiated onto a sample within a small angle range of 2° to 5° around the incident X-ray beam. Since the scattering of electromagnetic waves follows the inverse law - the larger the effective size of the irradiated object relative to the incident wave, the smaller the scattering angle of the electromagnetic wave, the main research subject of SAXS is submicroscopic structures of hundreds to thousands of angstroms (polymers, proteins, sol-gels, ultrafine powders, porous carbon, etc.). In short, SAXS technology has become one of the main tools for submicroscopic structure research with its unique advantages.

[0075] Critical dimensions are usually used to describe the most critical dimensional parameters on semiconductor chips or other microelectronic components. These dimensional parameters are crucial to the performance and function of semiconductor devices. They can include line width, spacing, pattern position and alignment on photomasks or wafers. X-ray scattering measurement technology is a powerful tool for measuring critical dimensions (CD) of nano devices. The focus of scattering measurement is to solve the inverse mapping problem from scattering patterns to critical dimensions of the sample to be measured.

[0076] In the prior art, iterative optimization algorithms, such as genetic algorithms and evolutionary algorithms, are generally used to solve the inverse mapping problem of X-ray scattering measurements. When dealing with complex structures containing a large number of parameters, due to the inherent limitations of such algorithms, problems such as long iteration time, high computational cost, and falling into local optimality will occur. With the development of deep learning technology, it has also been widely used in the field of X-ray scattering measurement, such as scattering spectrum classification and structure classification. However, due to problems such as Babinet's theorem and missing phase information in the scattering process, the same scattering spectrum will correspond to multiple possible target structures. The prediction result in the conventional deep learning training process is only one of the multiple possible target structures, and the same configuration as the target structure cannot be accurately obtained, resulting in a slower decrease in the loss function, causing the deep learning model to converge prematurely, and it is impossible to use conventional deep learning training methods to obtain a low-error model.

[0077] The present application embodiment provides a process of an X-ray scattering measurement method, such as Figure 1 As shown, the following steps are included:

[0078] Step S101, using X-rays to obtain a scattering spectrum of a target object to be inspected;

[0079] Among them, the X-ray scattering measurement and detection system can adopt a transmission small-angle X-ray scattering measurement and detection system, such as Figure 2 As shown, it includes a light source 201 , a nano-grating sample 202 and a detector 203 .

[0080] Step S102, analyzing the scattering pattern using the measurement model to obtain the critical dimensions of the device to be tested;

[0081] Specifically, the scattering pattern is input into the measurement model to obtain the predicted value of the critical dimension of the device to be tested. The model architecture diagram of the measurement model is shown in Figure 3As shown, it includes an input layer, a feature extraction layer, a connection layer and an output layer. Among them, the input layer can be at least one of the adaptive pooling methods such as adaptive average pooling or adaptive maximum pooling, which performs pooling processing on the scattering spectrum, reduces the parameter amount of the input data, and changes the size of the input data to adapt to the input size of the feature extraction layer; the feature extraction layer can be at least one of the network models such as ResNet, CNN, DenseNet, and fully connected networks, or use at least one of the machine learning models such as support vector regression model and Gaussian regression model, which is used to extract features from the scattering spectrum and obtain feature vectors. The connection layer can be a fully connected network, which is used to connect the feature extraction layer and the output layer, and is used to change the size of the feature vector to adapt to the size of the output vector. The output layer can be a fully connected layer that meets the size of the output vector combined with a sigmoid activation function. The output of the fully connected layer is a key size vector, and the sigmoid activation function normalizes the output so that the output size is normalized to the range of 0 to 1, eliminating the prediction error caused by the order of magnitude difference of the key size of different data points of the sample.

[0082] Step S103, the measurement model is obtained by inputting the data set of sample devices into the measurement model to be trained, and training the measurement model to be trained based on the joint loss function; the joint loss function is used to select the loss factor of the predicted value of the key size of the device to be tested, the Babinet complementary loss factor of the predicted value, the central symmetric loss factor of the predicted value and the Babinet complementary central symmetric loss factor of the predicted value.

[0083] Specifically, the Babinet theorem and the invariance of the intensity distribution of the centrosymmetric structure are combined as physical prior conditions to construct a joint loss function, which contains the loss values ​​of all possible target structures corresponding to the same scattering spectrum. The correct structure prediction is obtained by calculating the joint loss function, such as Figure 4 As shown, this is the error distribution diagram of the critical dimension (CD) prediction results in the present application. It can be seen that through the joint loss function constructed in the present application, the critical dimension error of 99.04% of the data is less than 0.7nm, the average absolute error of the critical dimension on the test set is 0.1277nm, and the goodness of fit of the critical dimension on the test set is 0.99356, which achieves a strong correlation between the predicted value and the target value, and can achieve high-precision measurement of the critical dimension.

[0084] By combining physical prior information with the training process of the measurement model to construct a joint loss function and calculate the losses of all possible structures corresponding to the same scattering spectrum, the problem of premature model convergence caused by multiple solutions is solved, and the training of deep learning models with low loss values ​​is achieved. The trained measurement model can be used to perform real-time and high-precision X-ray scattering measurements of nanostructured samples.

[0085] The calculation formula of the joint loss function is shown in Formula 1 below:

[0086]

[0087] Among them, y p The critical dimensions of the sample device output by the model prediction are: for y p The corresponding original structural loss factor, y pb The sample device has y p The critical dimension when the value of the Babinet complementary structure is ; for y pb The corresponding Babinet complementary loss factor, y pc The sample device has y p The critical dimension when the value is a centrosymmetric structure; for y pc The corresponding centrosymmetric loss factor; y pbc The sample device has y p The critical dimensions of the Babinet complementary and center-rotation structures are: for y pbc The corresponding Babinet complementary centrosymmetric loss factor; y t is the label value of the sample device. The loss function can be one of the loss functions such as MAE, MSE, Smooth L1, etc. pb ,y pc ,y pbc Can be calculated by the following formula:

[0088] y pb =Py p ...Formula 2

[0089] y pc = flip(y p )...Formula 3

[0090] y pbc =P-flip(y p )...Formula 4

[0091] In the above formulas 3, 4, and 5, P is the period of the nano-grating sample of this embodiment, that is, the grating constant, and flip means that the key dimension vectors are arranged in reverse order, that is, the center rotates 180°. Figure 5As shown, the cross-sectional diagram of the nano-grating sample structure and three structures with the same scattering spectrum, in which a is the cross-sectional diagram of the original structure of the sample device, b is the cross-sectional diagram of the original structure of the sample device after the center is rotated 180°, c is the cross-sectional diagram of the original structure a of the sample device after the Babinet complementation operation, d is the cross-sectional diagram of the cross-sectional diagram of the sample device c after the center is rotated 180°, and is also the cross-sectional diagram of the cross-sectional diagram b after the Babinet complementation operation. a, b, c, d have the same scattering spectrum.

[0092] like Figure 6 As shown, this is a curve comparison diagram of the unimproved MAE loss function on the verification set of this embodiment and the improved joint loss function trained on the same data set and model, where the solid line is the unimproved MAE loss function curve on the verification set of this embodiment, with a higher final convergence position, early convergence, and a larger error; the dotted line is the improved joint loss function curve, with a lower loss convergence position, more sufficient model training, and a smaller error.

[0093] In some embodiments, since the data points in the nanodevice sample are continuous, in order to make the prediction results of the measurement model smoother and reduce the overall error of the sample, the joint loss function may also include a total variation regularization term, which is used to strengthen the continuity between the key size data points of the continuous distribution, thereby reducing the overall error of the sample. The calculation formula of the total variation regularization term is as follows:

[0094]

[0095] in, is the total variation regularization term, N is the number of consecutive adjacent key size data points, y pi is the value corresponding to the i-th key size data point. The formula of the joint loss function after adding the regularization term is as follows:

[0096]

[0097] In some implementations, a data set of a sample device is constructed in the following manner, including: generating a sample device through computer simulation and determining a critical dimension of the sample device; performing data processing on the sample device to obtain a scattering map of the sample device; performing zero-mean processing on the sample device to update the critical dimensions of the sample device related to the relative position coordinates, thereby forming a scattering map of the sample device and a sample corresponding to the critical dimensions of the sample device.

[0098] Specifically, for example, a computer simulation method is used to generate one-dimensional periodic nano-grating sample device data, and the sample device number includes the key dimensions and scattering patterns of each sample device. For example, 100,000 data are generated by computer simulation, which are divided into training set, test set, and verification set in a ratio of 8:1:1. The deep learning model is trained on the training set and the test set, and the performance of the deep learning model is evaluated on the test set. Among them, the samples are arranged periodically in one direction and are stacked by multiple layers of rectangles in another direction. Each layer of rectangles is a data point, and the width of each rectangle is the key dimension of this embodiment. Zero mean processing is performed on the key dimensions related to the relative position coordinates of the nanostructure sample device. In this embodiment, the mean of the coordinates of the center position of each layer of rectangles of the grating sample is calculated, and the offset between the center position of the grating sample and the origin of the coordinates, that is, the center position of the grating sample is subtracted from this mean, eliminating the additional error caused by the translation coordinate transformation, that is, eliminating the problem that the same scattering pattern caused by the translation coordinate transformation will correspond to multiple possible target structures.

[0099] Data processing is performed on the sample device to obtain a scattering spectrum of the sample device, including: obtaining multi-angle scattering data of the sample device based on the intensity distribution of the scattering spectrum; performing cropping processing on the computer-simulated multi-angle scattering data to at least partially remove repeated data and zero-value point data in the computer-simulated scattering data, and splicing the cropped scattering data to obtain a computer-simulated scattering spectrum of the sample device.

[0100] Specifically, the scattering data of the sample device at various angles are preprocessed, such as Figure 7 The figure shows the X-ray scattering data of the sample at a single angle. Cut out the symmetrically repeated data in the scattering data at each angle, take the right half of the data, take the middle 4 rows of pixel data of each angle scattering data, and merge and splice them into the following figure: Figure 8 The final scattering spectrum shown is obtained by normalizing the intensity distribution of the processed scattering spectrum so that the intensity distribution is within the range of 0 to 1.

[0101] Specifically, the scattering spectrum intensity distribution can be calculated by first-order Born approximation and Fourier transform, and the calculation formula is as follows:

[0102]

[0103] Where q represents the scattering vector, r represents the position vector, ρ(r) is the grating shape function, and P is the grating constant, such as Figure 2 P, β is the scaling factor. bkg (q) is the background noise, I noise (q) is other noise.

[0104] like Fig. 9 FIG. 1 is a flow chart of an X-ray scattering measurement method provided in an embodiment of the present application, which specifically includes the following steps:

[0105] Step S901, constructing an X-ray scattering data set;

[0106] Step S902, scattering spectrum preprocessing;

[0107] Step S903, sample position zero mean processing;

[0108] Step S904, measurement model training;

[0109] Step S905, constructing a joint loss function;

[0110] Specifically, the joint loss function consists of four parts, namely the original loss function, the Babinet complementary loss function, the central symmetric loss function, and the Babinet complementary central symmetric loss function.

[0111] Step S906, determining whether the measurement model is converged;

[0112] Step S907, obtaining the critical dimensions of the device to be inspected.

[0113] By combining physical prior information with the training process of the measurement model to construct a joint loss function and calculate the losses of all possible structures corresponding to the same scattering spectrum, the problem of premature model convergence caused by multiple solutions is solved, and the training of deep learning models with low loss values ​​is achieved. The trained measurement model can be used to perform real-time and high-precision X-ray scattering measurements of nanostructured samples.

[0114] Based on the same technical concept, the present application embodiment provides a structural schematic diagram of an X-ray scattering measurement device, such as Fig.10 As shown, the device 1000 includes:

[0115] Acquisition module 1001, used to acquire the scattering spectrum of the target object to be detected using X-rays

[0116] A processing module 1002 is used to analyze the scattering spectrum using a measurement model to obtain a critical dimension of the device to be detected;

[0117] The processing module 1002 is also used for the measurement model, which is obtained by inputting a data set of sample devices into the measurement model to be trained, and training the measurement model to be trained based on a joint loss function; the joint loss function is used to select a loss factor of a predicted value of a key dimension of the device to be tested, a Babinet complementary loss factor of the predicted value, a centrally symmetric loss factor of the predicted value, and a Babinet complementary centrally symmetric loss factor of the predicted value.

[0118] By combining physical prior information with the training process of the measurement model to construct a joint loss function and calculate the losses of all possible structures corresponding to the same scattering spectrum, the problem of premature model convergence caused by multiple solutions is solved, and the training of deep learning models with low loss values ​​is achieved. The trained measurement model can be used to perform real-time and high-precision X-ray scattering measurements of nanostructured samples.

[0119] In some embodiments, the processing module 1002 is specifically used to:

[0120] The calculation formula of the joint loss function is as follows:

[0121]

[0122] Among them, y p The critical dimensions of the sample device output by the model prediction are: for y p The corresponding original structural loss factor, y pb The sample device has y p The critical dimension when the value of the Babinet complementary structure is ; for y pb The corresponding Babinet complementary loss factor, y pc The sample device has y p The critical dimension when the value is a centrosymmetric structure; for y pc The corresponding centrosymmetric loss factor; y pbc The sample device has y p The critical dimensions of the Babinet complementary and center-rotation structures are: for y pbc The corresponding Babinet complementary centrosymmetric loss factor; y t is the label value of the sample device.

[0123] In some embodiments, the processing module 1002 is specifically used to:

[0124] The joint loss function also includes a total variation regularization term, which is used to strengthen the continuity between key size data points of the continuous distribution.

[0125] In some embodiments, the processing module 1002 is specifically used to:

[0126] The calculation formula of the total variation regularization term is as follows:

[0127]

[0128] in, is the total variation regularization term, N is the number of consecutive adjacent key size data points, ypi is the value corresponding to the i-th key size data point.

[0129] In some embodiments, the processing module 1002 is specifically used to:

[0130] The data set of the sample device is constructed in the following manner, including:

[0131] Generate a sample device through computer simulation and determine the critical dimensions of the sample device;

[0132] Performing data processing on the sample device to obtain a scattering spectrum of the sample device;

[0133] The critical dimensions of the sample device related to the relative position coordinates are updated by zero-mean processing on the sample device, thereby forming a scattering spectrum of the sample device and a sample corresponding to the critical dimensions of the sample device. In some embodiments, the processing module 1002 is specifically used to: perform data processing on the sample device to obtain the scattering spectrum of the sample device, including:

[0134] Based on the scattering spectrum intensity distribution, obtaining scattering data of the sample device at multiple angles;

[0135] The multi-angle scattering data are clipped to at least partially remove repeated data and zero-value point data in the scattering data, and the clipped scattering data are spliced ​​to obtain a scattering spectrum of the sample device.

[0136] In some embodiments, the processing module 1002 is specifically used to calculate the scattering spectrum intensity distribution through first-order Born approximation and Fourier transform, and the calculation formula is as follows:

[0137]

[0138] Where q represents the scattering vector, r represents the position vector, ρ(r) is the grating shape function, P is the grating constant, β is the scale factor, and I bkg (q) is the background noise, I noise (q) is other noise.

[0139] By combining physical prior information with the training process of the measurement model to construct a joint loss function and calculate the losses of all possible structures corresponding to the same scattering spectrum, the problem of premature model convergence caused by multiple solutions is solved, and the training of deep learning models with low loss values ​​is achieved. The trained measurement model can be used to perform real-time and high-precision X-ray scattering measurements of nanostructured samples.

[0140] Based on the same technical concept, the embodiment of the present application provides a computer device, such as Fig.11As shown, it includes at least one processor 1101 and a memory 1102 connected to the at least one processor. The specific connection medium between the processor 1101 and the memory 1102 is not limited in the embodiment of the present application. Fig.11 For example, the processor 1101 and the memory 1102 are connected via a bus. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0141] In the embodiment of the present application, the memory 1102 stores instructions that can be executed by at least one processor 1101 , and the at least one processor 1101 can perform the steps of the above-mentioned X-ray scattering measurement method by executing the instructions stored in the memory 1102 .

[0142] Among them, the processor 1101 is the control center of the computer device, and various interfaces and lines can be used to connect various parts of the computer device, and by running or executing instructions stored in the memory 1102 and calling data stored in the memory 1102, the critical dimensions of the target part to be inspected can be measured. Optionally, the processor 1101 may include one or more processing units, and the processor 1101 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 1101. In some embodiments, the processor 1101 and the memory 1102 may be implemented on the same chip, and in some embodiments, they may also be implemented separately on independent chips.

[0143] Processor 1101 can be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware processor for execution, or can be executed by a combination of hardware and software modules in the processor.

[0144] Memory 1102, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules. Memory 1102 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), programmable read-only memory (Programmable Read Only Memory, PROM), read-only memory (Read Only Memory, ROM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), magnetic memory, disk, optical disk, etc. Memory 1102 is any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 1102 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0145] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium storing a computer program executable by a computer device. When the program runs on the computer device, the computer device executes the steps of the above-mentioned X-ray scattering measurement method.

[0146] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0147] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0148] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0150] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. An X-ray scattering measurement method, characterized in that: include: Use X-rays to obtain the scattering spectrum of the target part to be inspected; Analyzing the scattering pattern using a measurement model to obtain a critical dimension of the device to be detected; The measurement model is obtained by inputting a data set of a sample device into the measurement model to be trained, and training the measurement model to be trained based on a joint loss function; the loss factors in the joint loss function include a loss factor based on a predicted value of a critical dimension of the sample device, a Babinet complementary loss factor of the predicted value, a central symmetric loss factor of the predicted value, and a Babinet complementary central symmetric loss factor of the predicted value. The calculation formula of the joint loss function is as follows: Among them, y p The critical dimensions of the sample device output by the model prediction are: for y p The corresponding loss factor, y pb The sample device has y p The critical dimension when the value of the Babinet complementary structure is ; for y pb The corresponding Babinet complementary loss factor, y pc The sample device has y p The critical dimension when the value is a centrosymmetric structure; for y pc The corresponding centrosymmetric loss factor; y pbc The sample device has y p The critical dimensions of the Babinet complementary and center-rotation structures are: for y pbc The corresponding Babinet complementary centrosymmetric loss factor; y t is the label value of the sample device.

2. The method according to claim 1, characterized in that The joint loss function also includes a total variation regularization term, which is used to strengthen the continuity between key size data points of the continuous distribution.

3. The method according to claim 2, characterized in that The calculation formula of the total variation regularization term is as follows: in, is the total variation regularization term, N is the number of consecutive adjacent key size data points, is the value corresponding to the i-th key size data point.

4. The method according to claim 1, characterized in that The data set of the sample device is constructed in the following manner, including: Generate a sample device through computer simulation and determine the critical dimensions of the sample device; Performing data processing on the sample device to obtain a scattering spectrum of the sample device; The sample device is processed by zero mean value to update the critical dimensions of the sample device related to the relative position coordinates, thereby forming a scattering spectrum of the sample device and a sample corresponding to the critical dimensions of the sample device.

5. The method according to claim 4, characterized in that The step of performing data processing on the sample device to obtain a scattering spectrum of the sample device includes: Based on the scattering spectrum intensity distribution, obtaining scattering data of the sample device at multiple angles; The multi-angle scattering data are clipped to at least partially remove repeated data and zero-value point data in the scattering data, and the clipped scattering data are spliced ​​to obtain a scattering spectrum of the sample device.

6. The method according to claim 5, characterized in that The intensity distribution of the scattering spectrum is calculated by first-order Born approximation and Fourier transform, and the calculation formula is as follows: Where q represents the scattering vector, r represents the position vector, ρ(r) is the grating shape function, P is the grating constant, β is the scale factor, and I bkg (q) is the background noise, I noise (q) is other noise, x represents the position coordinate in space, δ represents the impulse function, n is the number of δ functions, and A is the range of integration.

7. An X-ray scattering measurement device, characterized in that: include: An acquisition module, used to acquire a scattering spectrum of a target object to be detected by using X-rays; A processing module, used for analyzing the scattering spectrum using a measurement model to obtain a critical dimension of the device to be detected; The processing module is also used for the measurement model to be obtained by inputting a data set of sample devices into the measurement model to be trained, and training the measurement model to be trained based on a joint loss function; the joint loss function is used to select a loss factor of a predicted value of a critical dimension of the device to be detected, a Babinet complementary loss factor of the predicted value, a central symmetric loss factor of the predicted value, and a Babinet complementary central symmetric loss factor of the predicted value, and the calculation formula of the joint loss function is as follows: Among them, y p The critical dimensions of the sample device output by the model prediction are: for y p The corresponding loss factor, y pb The sample device has y p The critical dimension when the value of the Babinet complementary structure is ; for y pb The corresponding Babinet complementary loss factor, y pc The sample device has y p The critical dimension when the value is a centrosymmetric structure; for y pc The corresponding centrosymmetric loss factor; y pbc The sample device has y p The critical dimensions of the Babinet complementary and center-rotation structures are: for y pbc The corresponding Babinet complementary centrosymmetric loss factor; y t is the label value of the sample device.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of any one of the methods of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that: It stores a computer program executable by a computer device. When the program is run on the computer device, the computer device executes the steps of any method described in claims 1 to 6.

Images