Semiconductor critical dimension measuring method
By acquiring through-focus scanning images under multimodal optical conditions and combining them with a three-dimensional vector optical imaging model and a dual prediction model, the problems of expensive equipment and easy sample damage in traditional methods are solved, realizing low-cost, high-precision measurement of semiconductor critical dimensions, which is suitable for online non-destructive testing.
Patent Information
- Application Number
- CN202511642969.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-11
- Publication Date
- 2026-02-10
AI Technical Summary
Existing technologies cannot efficiently, cost-effectively, and non-destructively measure critical semiconductor dimensions at the submicron level and below. Furthermore, traditional methods suffer from problems such as expensive equipment, complex operation, and easy sample damage.
Through-focus scanning images were acquired under multimodal optical illumination conditions. A three-dimensional vector optical imaging model and a dual prediction model (CNN regression and KNN matching) were combined to construct a measured and simulated dataset. Line width was measured through dual-model collaborative verification, and the model was dynamically optimized to improve accuracy and robustness.
It achieves low-cost, high-precision measurement of critical semiconductor dimensions with an error of less than 15nm, suitable for online non-destructive testing, improving the reliability of measurement results and the space utilization of equipment.
Smart Images
Figure FT_1 
Figure FT_2
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor critical dimension measurement, and more particularly to a method for measuring semiconductor critical dimensions. Background Technology
[0002] In the manufacturing of photomasks, photolithography processes, and subsequent packaging and bonding of integrated circuits, real-time monitoring of critical dimensions (i.e., linewidth) is necessary. Linewidth is a core parameter reflecting the width of feature lines in integrated circuits and directly affects chip performance. As manufacturing processes shrink to the sub-micron level, the development of traditional metrology technologies faces the following challenges: Optical microscopes: limited by the optical diffraction limit (resolution ≥ λ / 2), they cannot resolve features < 200 nm.
[0003] Scanning electron microscopy (CD-SEM): It has nanometer-level resolution, but it is expensive, requires an ultra-high vacuum environment, and the electron beam can easily damage the sample, so it cannot be used for online non-destructive testing. Optical scattering measurement (OCD) relies on complex optical models and parameter fitting, and is only applicable to periodic structures (such as gratings), and cannot measure isolated lines or irregular patterns.
[0004] Atomic force microscopy (AFM): provides three-dimensional topographic information, but has extremely low scanning efficiency (minutes for a single area) and the probe is prone to wear.
[0005] X-ray technology: It achieves atomic-level precision, but relies on synchrotron radiation sources, making it slow and costly.
[0006] CD-SEM (Critical Dimension Scanning Electron Microscope) is expensive, requires an ultra-high vacuum environment and specialized operation for maintenance, and electron beam scanning can easily damage samples, making it unsuitable for online non-destructive testing. OCD (Optical Critical Dimension Measurement) has a complex algorithm modeling, requiring the pre-establishment of an accurate optical model and fitting with a large number of parameters. It is only applicable to periodic structures (such as gratings) and cannot measure isolated lines or irregular shapes.
[0007] AFM (Atomic Force Microscopy) suffers from extremely low measurement efficiency (minutes per area) due to its single-point scanning mode, which cannot meet the high-throughput requirements of production lines. Probe wear introduces additional errors, requiring frequent calibration.
[0008] Traditional optical microscopes, limited by the optical diffraction limit (resolution ≥ λ / 2), cannot resolve features < 200 nm and rely on threshold algorithms based on the "optimal focal plane" to achieve edge positioning errors > 50 nm at the submicron scale.
[0009] In traditional optical microscopy, it is generally considered necessary to acquire images at the "optimal focus" position for meaningful analysis, based on the premise that the target is most realistically represented only at this position. A common thresholding algorithm, based on the image at the optimal focal plane position, defines the location of the largest gray-level gradient change in the image as an edge. However, this method is unreliable when measuring lines smaller than 1µm. This is because the edge features of these lines are often smaller than 200nm, and optical microscopy is generally considered unsuitable for measuring features smaller than half the illumination wavelength. Summary of the Invention
[0010] To address the aforementioned technical problems, the purpose of this invention is to provide a method for measuring critical dimensions of semiconductors.
[0011] To achieve the above objectives, the present invention adopts the following technical solution: A method for measuring critical dimensions of semiconductors includes the following steps: Step 1: Acquire through-focus scanning images of the target linewidth under multimodal optical illumination conditions and construct a measured image database; Step 2: Simulate the through-focus scanning image under the same illumination conditions based on the three-dimensional vector optical imaging model. After calibrating the model parameters with actual data, generate targeted process deviation data to supplement the database. Step 3: After integrating the measured and simulated data, construct a structured dataset, and input it as a group of through-focus scan images with the same linewidth under various optical conditions, and output it as physical size parameters. Step 4: Use the dataset to train the dual prediction model in parallel: 1) CNN regression prediction model, based on a convolutional neural network trained on real-world and simulated databases, directly regresses and predicts the linewidth value Lcnn after inputting a real-time focus image; 2) KNN data matching model defines the through-focus images with different line widths in the dataset as independent categories. By calculating the Euclidean distance or cosine similarity between the real-time acquired images and the database images in the feature space, it determines the structural similarity and outputs the nearest neighbor category calibration line width value Lknn. Step 5: Based on the results of the two models, proceed to dynamic arbitration fusion, compare the difference between |Lknn−Lcnn| and the preset threshold δ, where δ=2×target measurement accuracy; If the difference is less than or equal to δ, a weighted fusion output Lfinal=w is used. Lknn+(1−w) Lcnn, with weights w∈[0,1]; If the difference is greater than δ, it is judged as an abnormal divergence, and the manual review process is automatically triggered to prevent erroneous output.
[0012] Preferably, in the semiconductor critical dimension measurement method, the multimodal optical illumination conditions in step 1 include ultraviolet, visible, infrared spectra, TE / TM / circular polarization, and variable aperture / NA.
[0013] Preferably, the semiconductor critical dimension measurement method uses the RCWA algorithm in its three-dimensional vector optical imaging model.
[0014] Preferably, in the semiconductor critical dimension measurement method, the actual data in step 2 includes material optical constants and aberration coefficients.
[0015] Preferably, in the semiconductor critical dimension measurement method, the semiconductor critical dimension measurement is less than 200nm.
[0016] Preferably, in the semiconductor critical dimension measurement method, the total error of the semiconductor critical dimension is <15nm.
[0017] Preferably, the semiconductor critical dimension measurement method acquires a measured image database by adjusting the wide / narrow band spectral light source (ultraviolet-visible-infrared), polarization direction (TE / TM / circular polarization), aperture shape / position, and objective lens numerical aperture (NA), and simultaneously using a monochrome camera with light source switching to acquire high signal-to-noise ratio single-band pass-focus scanning sequences or a color camera with wide spectral light source to acquire RGB three-band pass-focus scanning signals in a single acquisition, thereby constructing a measured image database.
[0018] Preferably, in the semiconductor critical dimension measurement method, after step 5 is completed, the system continuously collects subsequent test data, dynamically optimizes the fusion weights of the CNN regression prediction model and the KNN data matching model results, and iteratively updates the model and database to gradually improve measurement accuracy and robustness.
[0019] By means of the above-described solution, the present invention has at least the following advantages: 1. The equipment used in the invention has low hardware cost and high upper limit of measurement accuracy.
[0020] 2. The invention can be coupled with the current overlay measurement equipment to realize the measurement of CD and overlay on a single device, thereby improving the space utilization of the Fab yellow light area.
[0021] 3. The invention adopts a dual-model collaborative verification mechanism (KNN + CNN) to improve the credibility of measurement results.
[0022] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description
[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0024] Figure 1 This is a schematic diagram of the through-focus scanning image process of the present invention; Figure 2 This is a line width diagram of the through-focus scanning image of the present invention. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0026] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0027] Example 1 This invention constructs linewidth feature signals based on full-focus scanning, establishes the correlation between the target full-focus scan image and the target size features, and ensures measurement reliability through a dual-model collaborative verification mechanism.
[0028] A method for measuring critical dimensions of semiconductors includes the following steps: Step 1: Acquire through-focus scanning images of the target linewidth under multimodal optical illumination conditions and construct a measured image database; Step 2: Simulate the through-focus scanning image under the same illumination conditions based on the three-dimensional vector optical imaging model. After calibrating the model parameters with actual data, generate targeted process deviation data to supplement the database. Step 3: After integrating the measured and simulated data, construct a structured dataset. The input consists of a group of through-focus scan images with the same linewidth under various optical conditions. The output labels are physical dimension parameters (left wall width d1, top width d2, right wall width d3, line height d4, or a subset thereof, such as...). Figure 2 (as shown) Step 4: Use the dataset to train the dual prediction model in parallel: 1) CNN regression prediction model, based on a convolutional neural network trained on real-world and simulated databases, directly regresses and predicts the linewidth value Lcnn after inputting a real-time focus image; 2) KNN data matching model defines the through-focus images with different line widths in the dataset as independent categories. By calculating the Euclidean distance or cosine similarity between the real-time acquired images and the database images in the feature space, it determines the structural similarity and outputs the nearest neighbor category calibration line width value Lknn. Step 5: Based on the results of the two models, proceed to dynamic arbitration fusion, compare the difference between |Lknn−Lcnn| and the preset threshold δ, where δ=2×target measurement accuracy; If the difference is less than or equal to δ, a weighted fusion output Lfinal=w is used. Lknn+(1−w) Lcnn, with weights w∈[0,1]; If the difference is greater than δ, it is judged as an abnormal divergence, and the manual review process is automatically triggered to prevent erroneous output; Step 6: The system continuously collects subsequent test data, dynamically optimizes the fusion weights of the CNN regression prediction model and the KNN data matching model, and iteratively updates the model and database to gradually improve measurement accuracy and robustness.
[0029] The initial weight of this invention is w=0.5.
[0030] Construction of a through-focus scan image, such as Figure 1 The image is formed by the coherent superposition of diffracted light of different orders on the image plane. The optimal focal plane position is where all diffraction orders have the minimum optical path difference (or near-zero phase difference), maximizing the intensity and minimizing the size of the central diffraction spot (Airy disk), thus achieving the highest resolution and clearest image plane. However, out-of-focus images also contain feature information about the target. To further improve the signal-to-noise ratio of the image signal, a full-focus scan image is constructed. This full-focus scan image incorporates information from all focal planes of the optical discretization (CD), containing more information about the actual physical morphology of the measured target.
[0031] By adjusting the lens's z-axis position, all images near the sharp focal plane are acquired. Along a direction parallel to the linewidth, the target linewidth image is projected as a one-dimensional curve. Each one-dimensional curve can be viewed as a graph with a size of 1. Images of N are superimposed in order of z position to form the final full-focus scan image.
[0032] Establish a model database There are two ways to establish a linewidth model database: one is to build a basic database by directly acquiring data from the equipment; the other is to build a simulation model database based on rigorous vector imaging theory. Method one offers higher data accuracy, but acquiring a complete dataset requires process coordination, resulting in a long modeling cycle and high cost. Method two is low-cost, but the modeling process is complex, especially for aberration simulation modeling. This means that the linewidth simulation model often cannot accurately account for the influence of equipment-specific aberrations, making it less accurate than Method one.
[0033] Combining the characteristics of the two modeling methods described above, a comprehensive database is established. For some routine data collection, direct acquisition can be used, acquiring through-focus scanning images under multimodal optical conditions (UV-Vis-IR spectroscopy, TE / TM / circular polarization, variable aperture / NA) to construct a high-precision measured database. However, for some uncommon and difficult-to-acquire anomalous data, simulation of process anomalies (such as edge asymmetry and dimensional deviations) is performed based on a 3D vector optical imaging model. After calibrating the model parameters, the database is supplemented to ensure that the final database covers both routine and anomalous operating conditions.
[0034] The multimodal optical illumination conditions in step 1 of this invention include ultraviolet, visible, infrared spectra, TE / TM / circular polarization, and variable aperture / NA.
[0035] The measured image database is constructed by adjusting the wide / narrow band spectral light source (ultraviolet-visible-infrared), polarization direction (TE / TM / circular polarization), aperture shape / position and objective lens numerical aperture (NA), and simultaneously using a monochrome camera with light source switching to acquire high signal-to-noise ratio single-band pass-focus scanning sequences or a color camera with wide spectral light source to acquire RGB three-band pass-focus scanning signals in a single acquisition.
[0036] The RCWA algorithm is used in the three-dimensional vector optical imaging model of this invention.
[0037] The actual data in step 2 of this invention includes material optical constants and aberration coefficients.
[0038] In this invention, the critical dimension measurement of semiconductors is less than 200nm, and the total error of the critical dimension of semiconductors is less than 15nm.
[0039] The dual-channel architecture leverages the interpretability of KNN to compensate for the black-box limitations of CNN, while threshold arbitration enables quantifiable evaluation of the results' reliability, ensuring industrial-grade reliability for submicron scale measurements.
[0040] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0041] In the description of this application, it should be noted that the terms "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use. They are used only for the convenience of describing this application and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0042] Furthermore, terms such as "horizontal" and "vertical" do not imply that components must be absolutely horizontal or vertical, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.
[0043] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set up," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.
[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for measuring critical dimensions of semiconductors, Its features are, Includes the following steps: Step 1: Acquire through-focus scanning images of the target linewidth under multimodal optical illumination conditions and construct a measured image database; Step 2: Simulate the through-focus scanning image under the same illumination conditions based on the three-dimensional vector optical imaging model. After calibrating the model parameters with actual data, generate targeted process deviation data to supplement the database. Step 3: After integrating the measured and simulated data, construct a structured dataset, and input it as a group of through-focus scan images with the same linewidth under various optical conditions, and output it as physical size parameters. Step 4: Use the dataset to train the dual prediction model in parallel: 1) CNN regression prediction model, based on a convolutional neural network trained on real-world and simulated databases, directly regresses and predicts the linewidth value Lcnn after inputting a real-time focus image; 2) KNN data matching model defines the through-focus images with different line widths in the dataset as independent categories. By calculating the Euclidean distance or cosine similarity between the real-time acquired images and the database images in the feature space, it determines the structural similarity and outputs the nearest neighbor category calibration line width value Lknn. Step 5: Based on the results of the two models, proceed to dynamic arbitration fusion, compare the difference between |Lknn−Lcnn| and the preset threshold δ, where δ=2×target measurement accuracy; If the difference is less than or equal to δ, a weighted fusion output Lfinal=w is used. Lknn+(1−w) Lcnn, with weights w∈[0,1]; If the difference is greater than δ, it is judged as an abnormal divergence, and the manual review process is automatically triggered to prevent erroneous output.
2. The semiconductor critical dimension measurement method according to claim 1, characterized in that: The multimodal optical illumination conditions in step 1 include ultraviolet, visible, infrared spectra, TE / TM / circular polarization, and variable aperture / NA.
3. The semiconductor critical dimension measurement method according to claim 1, characterized in that: The 3D vector optical imaging model uses the RCWA algorithm.
4. The semiconductor critical dimension measurement method according to claim 1, characterized in that: The actual data in step 2 includes material optical constants and aberration coefficients.
5. The semiconductor critical dimension measurement method according to claim 1, characterized in that: Semiconductor critical dimension measurement is less than 200nm.
6. A method for measuring critical dimensions of a semiconductor according to claim 1 or 5, characterized in that: The total error of the critical dimension of the semiconductor is <15nm.
7. The semiconductor critical dimension measurement method according to claim 1, characterized in that: The measured image database was constructed by adjusting the wide / narrow band spectral light source, polarization direction, aperture shape / position, and objective lens numerical aperture, and simultaneously using a monochrome camera with light source switching to acquire high signal-to-noise ratio single-band pass-focus scanning sequences or a color camera with wide spectral light source to acquire RGB three-band pass-focus scanning signals in a single acquisition.
8. The semiconductor critical dimension measurement method according to claim 1, characterized in that: After step 5 is completed, the system continues to collect subsequent test data, dynamically optimizes the fusion weights of the CNN regression prediction model and the KNN data matching model, and iteratively updates the model and database to gradually improve measurement accuracy and robustness.