Segmented frequency domain regression network for lithography alignment measurements and methods of use thereof

Abstract

The invention belongs to the technical field of semiconductor processing, and discloses a segmented frequency domain regression network for photoetching alignment measurement and a use method thereof, and the segmented frequency domain regression network comprises a preprocessing module which comprises wavelet transform and Fourier transform; the twin frequency domain regression network comprises a twin module and a data fusion module, and the twin module is used for extracting rough and fine dislocation information from different spectrograms; and the data fusion module is used for calculating offset by integrating the output of the twin module. Wavelet transformation can automatically realize segmented division of acquired signals under different measuring ranges, and the process has self-adaptability and does not need manual operation; a twin frequency domain regression network automatically fits a frequency domain filter for data in different measuring ranges, and high-precision measurement under a wide range is realized through sectional extraction of effective information and feature fusion; according to the invention, sub-nanometer photoetching alignment under millimeter magnitude can be realized.

Description

Technical Field

[0001] The invention belongs to the technical field of semiconductor processing, and in particular relates to a segmented frequency domain regression network for photolithography alignment measurement and a use method thereof. Background Art

[0002] Photolithography alignment is a key step in the photolithography process, and its purpose is to ensure that the pattern on the mask is accurately aligned with the pattern on the silicon wafer to ensure accurate overlay between the patterns after exposure. In the process of realizing the present invention, the inventors found that the existing photolithography alignment measurement algorithm has at least the following problems: it cannot solve the contradiction between large-scale measurement and high-precision measurement; it cannot automatically fit and eliminate the error model of various error sources, resulting in large measurement errors; it cannot effectively demodulate complex alignment signals, resulting in additional measurement errors brought by the algorithm; it cannot effectively deal with nonlinear quantities in the measurement system, resulting in the accuracy of the algorithm strategy being limited. Summary of the invention

[0003] The present invention aims to solve the above technical problems at least to a certain extent. To this end, the present invention aims to provide a segmented frequency domain regression network for lithography alignment measurement and a method for using the network.

[0004] The technical solution adopted by the present invention is:

[0005] A piecewise frequency domain regression network for photolithography alignment measurement, comprising:

[0006] Preprocessing module, including wavelet transform and Fourier transform;

[0007] The twin frequency domain regression network includes a twin module and a data fusion module. The twin module is used to extract coarse and fine misalignment information from different spectrum graphs; the data fusion module is used to calculate the offset by integrating the output of the twin module.

[0008] Preferably, the wavelet transform: m A, I m H, I m V, I m D]=Dwt2(I m , wname),

[0009] Among them, I m A, I m H, I m V and I m D represents the low frequency, horizontal high frequency, vertical high frequency and diagonal high frequency components of the corresponding image respectively, Dwt2 represents a single-layer two-dimensional discrete wavelet transform, and wname is the name of the wavelet kernel used.

[0010] Fourier transform is used to process the four moiré fringe images, and the amplitude and phase information in the frequency domain are obtained as follows:

[0011] Amplitude I amp =log(abs(FFTshift(FFT(I m· ))));

[0012] Phase I pha =angle(FFTshift(FFT(I m· ))),

[0013] Among them, I m For I m A, I m H, I m V or I m For any component in D, FFT is fast Fourier transform, FFTshift is used to move the zero-frequency component to the center of the array, abs is used for amplitude extraction, and angle is used for phase extraction.

[0014] Preferably, the twin module includes four branches, each branch includes a trunk branch and a mask branch, the trunk branch is used for feature extraction, and the mask branch is used to simulate the attention mechanism. The trunk branch includes a residual unit, and the mask branch includes an encoder-decoder structure. The data fusion module includes two fully connected layers, and the number of units in the two fully connected layers is 4 and 2 respectively.

[0015] A method for using the above-mentioned segmented frequency domain regression network includes:

[0016] S1, overlap the silicon wafer and the mask, and let the translation stage start to perform step scanning. At each scanning position, the camera is used to collect and image the formed moiré fringes to form a data set;

[0017] S2, taking the image as the input image and the position value of the displacement stage as the true value, inputting the frequency domain regression network for training, and completing the training by minimizing the loss through repeated training;

[0018] S3. Use the trained frequency domain regression network to perform regression prediction on the actually collected alignment signal.

[0019] Preferably, the mask surface is etched with linear or circular alignment marks. The step scan is a one-dimensional scan or a two-dimensional scan.

[0020] The present invention processes data by combining wavelet transform and a segmented frequency domain regression network composed of a twin frequency domain regression network to achieve sub-nanometer lithography alignment at the millimeter level, and the beneficial effects are:

[0021] Wavelet transform can automatically realize the segmentation of collected signals at different ranges. The process is adaptive and does not require manual operation.

[0022] The twin frequency domain regression network automatically fits the frequency domain filter to the data within different measurement ranges, extracts effective information and integrates features in a segmented manner, and achieves high-precision measurement under a wide range.

[0023] It can handle various complex alignment signals and effectively deal with nonlinear problems in strategy systems;

[0024] This method uses global signals for alignment, has good robustness, and does not have stringent requirements on beam quality and measurement environment; it has fast alignment speed and high alignment efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the network structure of the segmented frequency domain regression network of the present invention. DETAILED DESCRIPTION

[0026] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of 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.

[0027] It should be understood that it should also be noted that in the embodiment, the functions / actions that appear may be different from the order in which the drawings appear. For example, depending on the functions / actions involved, they can actually be executed substantially concurrently, or two figures shown in succession can sometimes be executed in the reverse order.

[0028] like Figure 1 As shown, a segmented frequency domain regression network for lithography alignment measurement in this embodiment includes a preprocessing module and a twin frequency domain regression network.

[0029] The preprocessing module includes wavelet transform and Fourier transform;

[0030] Wavelet transform: [I m A, I m H, I m V, I m D]=Dwt2(I m , wname),

[0031] Among them, I m A, I m H, I m V and I mD represents the low frequency, horizontal high frequency, vertical high frequency and diagonal high frequency components of the corresponding image respectively, Dwt2 represents a single-layer two-dimensional discrete wavelet transform, and wname is the name of the wavelet kernel used.

[0032] Fourier transform is used to process the four moiré fringe images, and the amplitude and phase information in the frequency domain are obtained as follows:

[0033] Amplitude I amp =log(abs(FFTshift(FFT(I m· ))));

[0034] Phase I pha =angle(FFTshift(FFT(I m· ))),

[0035] Among them, I m For I m A, I m H, I m V or I m For any component in D, FFT(.) represents fast Fourier transform, FFTshift(.) moves the zero-frequency component to the center of the array, abs(.) is used for amplitude extraction, and angle(.) is used for phase extraction. In order to ensure the integrity and consistency of frequency domain information, the amplitude I amp and Phase I pha Concatenate into two-channel magnitude-phase input images for network use.

[0036] After preprocessing with information-lossless wavelet transform and Fourier transform, the twin frequency domain regression network is used to extract the misalignment information. The twin frequency domain regression network includes a twin module and a data fusion module. The twin module is used to extract coarse and fine misalignment information from different spectrum graphs; the data fusion module is used to calculate the offset by integrating the output of the twin module.

[0037] The branch architecture of the twin module helps to isolate specific feature attributions in different spectrograms, thereby optimizing the feature extraction accuracy of the network. The twin module consists of four branches, each of which includes a trunk branch and a mask branch. The trunk branch is used for feature extraction, and the mask branch is used to simulate the attention mechanism. After performance evaluation, the residual unit was selected as the basic building block for building the trunk branch. Since the input I and output T(I) of the trunk branch have the same dimension, the encoder-decoder structure is used to simulate the feedforward and feedback attention mechanism in the mask branch to generate a mask M(x) for slightly weighted processing of the feature map T(x). Therefore, the output of the attention module (H(x)) can be expressed as:

[0038] H i，c (I) = T i，c(I)*(1+M i，c (I)),

[0039] Among them, i and c represent the pixels and channels of the feature map respectively.

[0040] After the twin module outputs four sets of results, a data fusion module is developed. The data fusion module includes two fully connected layers with 4 and 2 units respectively to generate the final perception results. In order to effectively evaluate the displacement difference in each direction, the mean square error (MSE) loss function, also known as L2 loss, is used, and the expression is as follows:

[0041]

[0042] Where X represents the regressed displacement value, F represents the Frobenius norm, the subscript gt represents the true value, n is the index number, and N is the dimension of the regressed displacement.

[0043] The method of using the above-mentioned segmented frequency domain regression network includes:

[0044] S1. Overlap the silicon wafer and the mask, etch linear or circular alignment marks on the mask surface, and let the translation stage start one-dimensional or two-dimensional step scanning. Use the camera to collect and image the formed moiré fringes at each scanning position to form a data set.

[0045] S2, taking the image as the input image and the position value of the displacement stage as the true value, inputting the above-mentioned frequency domain regression network for training, and completing the training by minimizing the loss (the standard is L1 or L2 loss) through repeated training;

[0046] S3. Use the trained frequency domain regression network to perform regression prediction on the alignment signal actually collected to realize the alignment measurement function.

[0047] The method of using this embodiment uses a segmented frequency domain regression network to improve the accuracy of the signal analysis algorithm of the lithography alignment system.

[0048] The present invention is not limited to the above-mentioned optional implementation modes. Anyone can derive other various forms of products under the inspiration of the present invention. However, no matter what changes are made in the shape or structure, all technical solutions that fall within the scope defined by the claims of the present invention fall within the protection scope of the present invention.

Claims

1. A segmented frequency domain regression network for lithography alignment measurement, characterized in that: include: Preprocessing module, including wavelet transform and Fourier transform; The twin frequency domain regression network includes a twin module and a data fusion module. The twin module is used to extract coarse and fine misalignment information from different spectrum graphs; the data fusion module is used to calculate the offset by integrating the output of the twin module.

2. The segmented frequency domain regression network according to claim 1, characterized in that: Small wave change: [I m A, I m H, I m V, I m D]=Dwt2(I m ,wnamne), Among them, I m A, I m H, I m V and I m D represents the low frequency, horizontal high frequency, vertical high frequency and diagonal high frequency components of the corresponding image respectively. Dwt2 represents a single-layer two-dimensional discrete wavelet transform, and wname is the name of the wavelet kernel used.

3. The segmented frequency domain regression network according to claim 2, characterized in that: Fourier transform is used to process the four moiré fringe images, and the amplitude and phase information in the frequency domain are obtained as follows: Amplitude I amp = log(abs(FFTshift(FFT(I m .)))) ; Phase I pha =angle(FFTshift(FFT(I m .))), Among them, I m For I m A, I m H, I m V or I m For any component in D, FFT is fast Fourier transform, FFTshift is used to move the zero-frequency component to the center of the array, abs is used for amplitude extraction, and angle is used for phase extraction.

4. The segmented frequency domain regression network according to claim 1, wherein: The twin module consists of four branches, each of which includes a trunk branch and a mask branch. The trunk branch is used for feature extraction, and the mask branch is used to simulate the attention mechanism.

5. The segmented frequency domain regression network according to claim 4, characterized in that: The trunk branch includes a residual unit, and the mask branch includes an encoder-decoder structure.

6. The segmented frequency domain regression network according to claim 4 or 5, characterized in that: The data fusion module consists of two fully connected layers, and the number of units in the two fully connected layers is 4 and 2 respectively.

7. A method for using the segmented frequency domain regression network according to any one of claims 1 to 6, characterized in that: include: S1, overlap the silicon wafer and the mask, and let the translation stage start to perform step scanning. At each scanning position, the camera is used to collect and image the formed moiré fringes to form a data set; S2, taking the image as the input image and the position value of the displacement stage as the true value, inputting the frequency domain regression network for training, and completing the training by minimizing the loss through repeated training; S3. Use the trained frequency domain regression network to perform regression prediction on the actually collected alignment signal.

8. The method of use according to claim 7, characterized in that: A linear or circular alignment mark is etched on the mask surface.

9. The method of use according to claim 7, characterized in that: The step scan is a one-dimensional scan or a two-dimensional scan.

Images