Euv lithography multilayer film defect detection method based on dark field microscopy images

Abstract

This invention relates to the field of extreme ultraviolet (EUV) lithography mask technology, and more particularly to a method for detecting defects in multilayer films using EUV lithography masks. The method includes: S1: simulating dark-field microscopic images of multilayer film defects using EUV lithography masks; S2: training a neural network using the simulated dark-field microscopic images to obtain a mapping model between the dark-field microscopic images and defect morphological parameters; S3: acquiring dark-field microscopic images of multilayer film defects using a dark-field microscope, and inputting the images into the mapping model to obtain the defect morphological parameters. This invention achieves high-throughput detection of various morphological parameters on the surface and inside of multilayer film defects without damaging the mask sample. Furthermore, it has low requirements for the illumination system and possesses promising prospects for industrial applications.

Description

Technical Field

[0001] This invention relates to the field of extreme ultraviolet (EUV) lithography mask technology, and more particularly to a method for detecting defects in multilayer films of EUV lithography masks. Background Technology

[0002] Photolithography is a mass production technology used in semiconductor manufacturing to transfer integrated circuit patterns. Extreme ultraviolet (EUV) lithography masks are key components in EUV lithography equipment used to define integrated circuit patterns, achieving light field modulation through reflection. The multilayer films in EUV lithography masks are the core structure for achieving reflection, typically composed of two or more different materials stacked alternately. Multilayer film defects are unique to EUV lithography masks, mainly caused by protrusions and depressions on the substrate, as well as particles that fall during deposition, leading to localized deformation of the multilayer film structure. These defects interfere with the amplitude and phase of the reflected light from the EUV lithography mask, thus adversely affecting photolithographic imaging.

[0003] In industry, defect compensation can effectively improve the quality of photolithography imaging. However, defect detection is crucial for compensation and traceability in this process. Therefore, to achieve the manufacture of "zero-defect" masks, a method with high throughput and accurate defect detection is urgently needed. Prior technology 1 (Kwon HJ, Harris-Jones J, Teki R, et al. Printability of native blank defects and programmed defects and their stack structures[C]. Photomask Technology 2011. SPIE, 2011, 8166: 60-69.) uses transmission electron microscopy to detect mask defects. Although it can obtain complete information about the defects, the sample preparation process is destructive, and the tested mask cannot be used for production. Prior technology 2 (Son D, He L, Satake M, et al. Wafer printability simulation of EUV mask defects using mask SEM and AFM[C]. Metrology, Inspection, and Process Control XXXVIII. SPIE, 2024, 12955: 669-674.) and prior technology 3 (Cho W, Price D, Morgan PA, et al. Classification and printability of EUV mask defects from SEM images[C]. International Conference on Extreme Ultraviolet Lithography 2017. SPIE, 2017, 10450: 28-35.) respectively used atomic force microscopy and scanning electron microscopy to detect surface deformation of multilayer films, but it was difficult to obtain information on defects located inside the multilayer film. Prior technology 4 (Chen Y, Lin Y, Chen R, et al. EUV multilayer defect characterization via cycle-consistent learning[J]. Optics Express, 2020, 28(12): 18493-18506.) is based on microcoherent extreme ultraviolet scattering microscopy and combined with neural networks and cycle-consistent learning to achieve comprehensive detection of defect information.However, this technology employs a scanning imaging method, which has limited throughput, and its dependence on synchrotron radiation sources restricts its application in industrial environments. Prior technology 5 (Zheng H, Li S, Cheng W, et al. Phase defect characterization using generative adversarial networks for extreme ultraviolet lithography[J]. Applied Optics, 2023, 62(5): 1243-1252.) is based on a high numerical aperture extreme ultraviolet band mask detection system and combines it with neural networks to achieve defect detection. However, due to the use of bright-field imaging, the signal-to-noise ratio is low, and it also relies on synchrotron radiation sources, which limits its practicality. Prior technology 6 (Yamane T, Tanaka T, Terasawa T, et al. Phase defect analysis with actinic full-field EUVL mask blank inspection[C]. Photomask Technology 2011. SPIE, 2011, 8166:52-59.) is based on dark-field microscopy and combines it with defect signal intensity analysis to achieve defect localization and qualitative estimation of the equivalent volume diameter of the defect sphere. However, this technology has not fully utilized the image information acquired by the two-dimensional detector, thus making it difficult to achieve comprehensive detection of the defect geometry and internal structure. Summary of the Invention

[0004] This invention addresses the shortcomings of existing extreme ultraviolet (EUV) lithography mask multilayer film defect detection technologies in terms of detection comprehensiveness, throughput, and hardware industrial applicability. It provides an EUV lithography mask multilayer film defect detection method based on dark-field microscopy images, achieving comprehensive defect detection with high throughput and industrial applicability.

[0005] The extreme ultraviolet lithography mask defect detection method provided by this invention specifically includes the following steps: S1: Dark-field microscopic images of defects in multilayer films using simulated extreme ultraviolet lithography masks, prepared for neural network training; preferably, step S1 specifically includes the following steps: S11: Constructing a dark-field microscope model: The illumination method adopted is Circle illumination, with a partial coherence factor of 0.26, a wavelength of 13.5 nm, and an incident angle of 0°. The numerical aperture of the projection objective is set to 0.27, and the numerical aperture of the shading lens is set to 0.1. The magnification of the optical system is set to 1200×. These parameter settings ensure that the optical response of the simulated dark-field microscopy image is close to that of an actual dark-field microscope.

[0006] S12: Constructing a mask model incorporating defects in multilayer films: The extreme ultraviolet (EUV) lithography mask consists of a molybdenum / silicon multilayer film and a quartz substrate, with defects positioned at the mask's center. Defects are represented by Gaussian-shaped local deformations of the multilayer film, determined by both surface and bottom Gaussian deformations. The deformation of the intermediate layers is calculated using linear interpolation. Y(Shape, Hbot, Wbot, Htop, Wtop, Blayer) represents the defect's morphological parameters. Specifically, the defect's unevenness type (Shape) is determined by random binary sampling; the bottom height (Hbot) is randomly selected from 6-20 nm; the bottom half-width (Wbot) is randomly selected from 15-50 nm; the top height (Htop) is randomly selected from 0.5-5 nm; the top half-width (Wtop) is randomly selected from 35-70 nm; and the number of buried layers (Blayer) is a random integer value from 0-77 layers. A total of m defect models are generated. Defects with Gaussian deformations and randomly generated within reasonable parameter ranges can essentially cover the diversity of actual multilayer film defects.

[0007] S13: Simulate and acquire dark-field microscopic images of multilayer film defects with different degrees of defocus. This is because defocus aberration modulates the phase field of the scattered signal and, to some extent, amplifies or cancels the phase field distortion caused by defects, thereby improving the resolvability of defect phase features in dark-field microscopic images. This includes the following sub-steps: S131: Calculate the near-field electric field distribution corresponding to the defect using the waveguide method or the finite-difference time-domain method.

[0008] S132: Based on Abbe imaging theory and combined with dark-field microscopy imaging model, the near-field electric field is imaged and calculated; during the imaging calculation process, defocus aberration is loaded into the pupil to simulate the axial position offset of the projection system. The axial position offset is set with a total of 5 defocus positions, namely APO1=−400nm, APO2=−200nm, APO3=0nm, APO4=200nm, and APO5=400nm, so as to obtain dark-field microscopic images under different defocus conditions.

[0009] S133: Set the signal-to-noise ratio of the dark-field microscopy image to be randomly sampled within the range of 30–110 dB; based on the signal-to-noise ratio sampling results, inject Gaussian noise into the calculated ideal dark-field microscopy image to simulate the noise effect in actual dark-field microscopy imaging.

[0010] S134: The calculated dark-field microscopic image is downsampled by average pooling according to the pixel size of the EUV imaging sensor to simulate the receiving process of scattered light intensity in the photosensitive area of ​​a single pixel in the imaging sensor, wherein the pixel size is 12μm.

[0011] S2: Train a neural network using simulated dark-field microscopic images to obtain a mapping model between dark-field microscopic images and defect morphology parameters; preferably, step S2 specifically includes the following steps: S21: Construct a neural network model, which includes an upsampling module, a feature extraction module, and a multilayer perceptron module. The upsampling module uses a linear interpolation algorithm to interpolate the input image to 224×224 pixels. The feature extraction module is a residual convolutional neural network. The multilayer perceptron module includes an input layer, a hidden layer, and an output layer connected in sequence. The input layer contains 150 neurons, the hidden layer contains 78 neurons, and the output layer contains 6 neurons. Activation layers are set between each layer.

[0012] S22: The mean squared error is used as the loss function and the Adam optimization algorithm is used to optimize it; the 5×m dark field microscopic images obtained from step S1 simulation and the corresponding m sets of morphological parameters are used as the training set to train the convolutional neural network and obtain the trained neural network model.

[0013] S23: Construct a post-processing model to round the floating-point values ​​representing the defect convexity / concave type and the number of buried layers in the neural network output to obtain the final prediction result; and constrain the top height based on the predicted value of the number of buried layers. When Blayer > 0, the top height is set to zero. This is because the convexity / concave type and the number of buried layers are discrete values, requiring integer outputs to directly correspond to the actual defect type. Furthermore, when Blayer > 0, the top deformation is meaningless; the constraint ensures that the output parameters are physically reasonable and avoids prediction errors.

[0014] S3: Acquire dark-field microscopic images of multilayer film defects using a dark-field microscope, and input them into a mapping model to obtain the morphological parameters of the defects. Using a dark-field microscope can shield bright-field signals and reduce low-spatial-frequency noise, thereby improving the signal-to-noise ratio of multilayer film defects and reducing the need for exposure dose, thus increasing detection throughput. Furthermore, dark-field microscopy has lower requirements for the illumination system and does not require unique illumination angles; therefore, commercially available laser plasma sources or discharge plasma sources can be used. Preferably, step S3 specifically includes the following steps: S31: The reference point position was read using a 27× low-magnification dark-field microscope; the entire mask area was rapidly scanned to mark areas where multilayer film defects might exist; subsequently, these areas were individually imaged at high magnification, obtaining high-magnification defocused dark-field microscopic images with a magnification of 1200× and axial position offsets of the projection system, namely APO1=−400nm, APO2=−200nm, APO3=0nm, APO4=200nm, and APO5=400nm. Because the low-magnification mode has a large field of view, scanning a large area of ​​the mask can be completed with fewer mechanical movements, thereby reducing the time overhead of mechanical movement and further improving the throughput of defect detection.

[0015] S32: Input the high-magnification dark field image acquired in S31 into the mapping model to obtain the corresponding defect morphology parameters, including the defect concavity / convexity type, bottom height, bottom half-width, top height, top half-width, and number of buried layers.

[0016] Compared with the prior art, the present invention can achieve the following beneficial effects: This invention, based on the penetrating characteristics of dark-field microscopy, enables the detection of surface and internal morphological parameters of defects in multilayer films without damaging the mask sample. These parameters include defect type, bottom height, bottom half-width at half-maximum (HWHM), top height, top HWHM, and the number of buried layers. Furthermore, the high throughput and rapid magnification switching capabilities of dark-field microscopy, coupled with the fact that it does not require fully coherent illumination, make the defect detection method feasible for implementation in industrial production environments. In addition, the introduction of a multi-channel defocus detection mechanism effectively acquires phase change information caused by defects, thereby improving the accuracy of defect morphological parameter detection. Moreover, this invention employs a single neural network model to simultaneously perform defect parameter classification and regression tasks, reducing system complexity and improving the ease of implementation and stability of the detection system. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for detecting defects in multilayer films using extreme ultraviolet lithography masks based on dark-field microscopic images, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of acquiring dark-field microscopic images with different degrees of defocus according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a neural network model provided according to an embodiment of the present invention. Detailed Implementation

[0018] In the following description, embodiments of the invention will be described with reference to the accompanying drawings. In the description below, the same modules are denoted by the same reference numerals. Where the same reference numerals are used, their names and functions are also the same. Therefore, their detailed description will not be repeated.

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0020] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0021] Figure 1 The flowchart of the method for detecting defects in multilayer films using extreme ultraviolet lithography masks based on dark-field microscopic images provided in an embodiment of the present invention is shown. S1: Dark-field micrograph of defects in multilayer films using simulated extreme ultraviolet lithography masks, specifically including the following steps: S11: Constructing a dark-field microscope model: The illumination method is Circle illumination, where the partial coherence factor of Circle illumination is 0.26, the wavelength is set to 13.5nm, and the incident angle is 0°; the numerical aperture of the projection objective is set to 0.27, and the numerical aperture of the shading lens is set to 0.1; the magnification of the optical system is set to 1200×. S12: Construct a mask model containing multilayer film defects: The extreme ultraviolet lithography mask consists of a molybdenum / silicon multilayer film and a quartz substrate, with defects located at the center of the mask; the defects are represented by Gaussian-shaped local deformations of the multilayer film, determined by the surface Gaussian deformation and the bottom Gaussian deformation, while the deformation of the intermediate layers is calculated through linear interpolation; Y(Shape, Hbot, Wbot, Htop, Wtop, Blayer) represents the morphological parameters of the defects, where the defect convexity type is Shape, which is randomly sampled from binary samples; the bottom height of the defect is Hbot, randomly selected from 6-20nm; the bottom half-width is Wbot, randomly selected from 15-50nm; the top height is Htop, randomly selected from 0.5-5nm; the top half-width is Wtop, randomly selected from 35-70nm; and the number of buried layers is Blayer, randomly selected from 0-77 layers; the total number of defect models is m. S13: Simulate and obtain dark-field microscopic images of multilayer film defects with different degrees of defocus, including the following sub-steps: S131: Calculate the near-field electric field distribution corresponding to the defect using the waveguide method or the finite-difference time-domain method; S132: Based on Abbe imaging theory and combined with a dark-field microscopy imaging model, the near-field electric field is calculated for imaging. During the imaging calculation, defocus aberration is applied to the pupil to simulate the axial position shift of the projection system. The axial position shift is set at five defocus positions: APO1 = −400nm, APO2 = −200nm, APO3 = 0nm, APO4 = 200nm, and APO5 = 400nm, thereby obtaining dark-field microscopic images (e.g., under different defocus conditions) for different conditions. Figure 2 (as shown) S133: Set the signal-to-noise ratio of the dark-field microscopy image to be randomly sampled within the range of 30–110 dB; based on the signal-to-noise ratio sampling results, inject Gaussian noise into the calculated ideal dark-field microscopy image to simulate the noise effect in actual dark-field microscopy imaging; S134: The calculated dark-field microscopic image is downsampled by average pooling according to the pixel size of the EUV imaging sensor to simulate the receiving process of scattered light intensity in the photosensitive area of ​​a single pixel in the imaging sensor, wherein the pixel size is 12μm.

[0022] S2: Train a neural network using simulated dark-field microscopic images to obtain a mapping model between dark-field microscopic images and defect morphology parameters. This includes the following steps: S21: Construct a neural network model, which includes an upsampling module, a feature extraction module, and a multilayer perceptron module. The upsampling module uses a linear interpolation algorithm to interpolate the input image to 224×224 pixels. The feature extraction module is a residual convolutional neural network. The multilayer perceptron module includes a sequentially connected input layer, hidden layer, and output layer. The input layer contains 150 neurons, the hidden layer contains 78 neurons, and the output layer contains 6 neurons. Activation layers (such as...) are set between each layer. Figure 3 (as shown) S22: The mean squared error is used as the loss function and the Adam optimization algorithm is used to optimize it; the 5×m dark field microscopic images obtained from step S1 simulation and the corresponding m sets of morphological parameters are used as the training set to train the convolutional neural network and obtain the trained neural network model.

[0023] S3: Acquire dark-field microscopic images of defects in multilayer films using a dark-field microscope, and input them into a mapping model to obtain the morphological parameters of the defects. This includes the following steps: S31: Use a dark-field microscope in 27× low magnification mode to read the reference point position; perform a rapid scan of the entire mask area to mark areas where multilayer film defects may exist; then, perform high-magnification imaging on these areas one by one to obtain high-magnification defocused dark-field micrographs with a magnification of 1200× and axial position offsets of the projection system, namely APO1=−400nm, APO2=−200nm, APO3=0nm, APO4=200nm, and APO5=400nm; S32: Input the high-magnification dark field image acquired in S31 into the mapping model to obtain the corresponding defect morphology parameters, including the defect concavity / convexity type, bottom height, bottom half-width, top height, top half-width, and number of buried layers.

[0024] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0025] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for detecting defects in multilayer films of extreme ultraviolet lithography masks based on dark-field microscopy images, characterized in that, Specifically, the steps include the following: S1: Dark-field micrograph of defects in multilayer films using simulated extreme ultraviolet lithography masks; S2: Train a neural network using simulated dark-field microscopy images to obtain a mapping model between dark-field microscopy images and defect morphology parameters; S3: Use a dark-field microscope to acquire dark-field microscopic images of defects in multilayer films, and input them into the mapping model to obtain the morphological parameters of the defects.

2. The method for detecting defects in multilayer films using extreme ultraviolet lithography masks based on dark-field microscopy images according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11: Constructing a dark-field microscope model: The illumination method is Circle illumination, where the partial coherence factor of Circle illumination is 0.26, the wavelength is set to 13.5nm, and the incident angle is 0°; the numerical aperture of the projection objective is set to 0.27, and the numerical aperture of the shading lens is set to 0.1; the magnification of the optical system is set to 1200×. S12: Construct a mask model containing multilayer film defects: The extreme ultraviolet lithography mask consists of a molybdenum / silicon multilayer film and a quartz substrate, with defects located at the center of the mask; the defects are represented by Gaussian-shaped local deformations of the multilayer film, determined by the surface Gaussian deformation and the bottom Gaussian deformation, while the deformation of the intermediate layers is calculated through linear interpolation; Y(Shape, Hbot, Wbot, Htop, Wtop, Blayer) represents the morphological parameters of the defects, where the defect convexity type is Shape, which is randomly sampled from binary samples; the bottom height of the defect is Hbot, randomly selected from 6-20nm; the bottom half-width is Wbot, randomly selected from 15-50nm; the top height is Htop, randomly selected from 0.5-5nm; the top half-width is Wtop, randomly selected from 35-70nm; and the number of buried layers is Blayer, randomly selected from 0-77 layers; the total number of defect models is m. S13: Simulate and obtain dark-field microscopic images of multilayer film defects with different degrees of defocus, including the following sub-steps: S131: Calculate the near-field electric field distribution corresponding to the defect using the waveguide method or the finite-difference time-domain method; S132: Based on Abbe imaging theory and combined with dark-field microscopy imaging model, the near-field electric field is imaged and calculated; during the imaging calculation process, defocus aberration is loaded into the pupil to simulate the axial position shift of the projection system. The axial position shift is set with a total of 5 defocus positions, namely APO1=−400nm, APO2=−200nm, APO3=0nm, APO4=200nm, and APO5=400nm, so as to obtain dark-field microscopic images under different defocus conditions; S133: Set the signal-to-noise ratio of the dark-field microscopy image to be randomly sampled within the range of 30–110 dB; based on the signal-to-noise ratio sampling results, inject Gaussian noise into the calculated ideal dark-field microscopy image to simulate the noise effect in actual dark-field microscopy imaging; S134: The calculated dark-field microscopic image is downsampled by average pooling according to the pixel size of the imaging sensor to simulate the receiving process of scattered light intensity in the photosensitive area of ​​a single pixel in the imaging sensor, wherein the pixel size is 12μm.

3. The method for detecting defects in multilayer films using extreme ultraviolet lithography masks based on dark-field microscopy images according to claim 1, characterized in that, Step S2 specifically includes the following steps: S21: Construct a neural network model, which includes an upsampling module, a feature extraction module, and a multilayer perceptron module. The upsampling module uses a linear interpolation algorithm to interpolate the input image to 224×224 pixels. The feature extraction module is a residual convolutional neural network. The multilayer perceptron module includes an input layer, a hidden layer, and an output layer connected in sequence. The input layer contains 150 neurons, the hidden layer contains 78 neurons, and the output layer contains 6 neurons. Activation layers are set between each layer. S22: The mean squared error is used as the loss function and the Adam optimization algorithm is used to optimize it; the 5×m dark field microscopic images and the corresponding m sets of morphological parameters obtained from the simulation in step S1 are used as the training set to train the convolutional neural network and obtain the trained neural network model. S23: Construct a post-processing model, round the floating-point values ​​representing the defect type and the number of buried layers in the neural network output to obtain the final prediction result; and constrain the top height based on the predicted value of the number of buried layers. When Blayer>0, the top height is set to zero.

4. The method for detecting defects in multilayer films using extreme ultraviolet lithography masks based on dark-field microscopy images according to claim 1, characterized in that, Step S3 specifically includes the following steps: S31: Use a dark-field microscope in 27× low magnification mode to read the reference point position; perform a rapid scan of the entire mask area to mark areas where multilayer film defects may exist; then, perform high-magnification imaging on these areas one by one to obtain high-magnification defocused dark-field micrographs with a magnification of 1200× and axial position offsets of the projection system, namely APO1=−400nm, APO2=−200nm, APO3=0nm, APO4=200nm, and APO5=400nm; S32: Input the high-magnification dark field image acquired in S31 into the mapping model to obtain the corresponding defect morphology parameters, including the defect concavity / convexity type, bottom height, bottom half-width, top height, top half-width, and number of buried layers.

Images