Holographic spectrum-based defect detection method, electronic equipment and product

Abstract

The embodiment of the invention relates to the field of semiconductors, and discloses a defect detection method based on holographic spectrum.The method comprises the steps that an optical image of the surface of a substrate and spectral signals under multiple wavebands corresponding to each pixel in the optical image are obtained, constructing a holographic spectral data cube based on the optical image and the spectral signal; acquiring a spectral curve of each pixel according to the holographic spectral data cube; extracting a spectral feature vector through the spectral curve, and generating a material distribution diagram based on the spectral feature vector; and comparing the material distribution diagram with a preset standard sample material distribution diagram, and identifying an abnormal region. Therefore, accurate positioning and material judgment of the defect area are achieved, high-sensitivity defect recognition can be achieved under transparent, high-light-transmittance and low-contrast materials, and the accuracy and the automation degree of defect detection are remarkably improved.

Description

Technical Field

[0001] This application relates to the semiconductor field, and more particularly to a defect detection method, electronic device, and product based on holographic spectroscopy. Background Technology

[0002] In high-precision industrial fields such as microelectronics manufacturing, semiconductor testing, and precision materials analysis, defect detection is a crucial step in ensuring device reliability and process stability. Traditional defect finding and detection methods mainly rely on grayscale images or bright-field imaging technology, identifying surface anomalies through two-dimensional visual comparison. However, with the increasing complexity of material systems and device structures, these methods have gradually revealed significant limitations.

[0003] First, traditional visual inspection is insensitive to differences in material composition. This is especially true for transparent or highly transparent materials (such as silicon nitride (SiN), silicon dioxide). In the case of (etc.), because their refractive indices are similar and their image contrast is low, defects and background areas are difficult to distinguish in grayscale space, which can easily lead to missed detections or misjudgments.

[0004] Secondly, two-dimensional images are susceptible to noise and pattern interference. For samples with complex lithographic patterns or periodic textures, the background signal of the image is strong and changes drastically, often causing the detection algorithm to fail to accurately separate the real defect area, resulting in a high false detection rate.

[0005] Furthermore, existing visual methods lack the ability to identify the physical or chemical properties of defects. Traditional images only reflect changes in spatial brightness and cannot distinguish the causes of defects, such as structural collapse, contamination particles, oxide layer formation, or process residues, making it difficult to support process optimization and root cause analysis.

[0006] Furthermore, the testing process relies heavily on manual labor. In the testing of multiple batches of wafers or microstructure samples, manual comparison and judgment lead to low search efficiency and highly subjective results, which is not conducive to achieving large-scale automated testing and consistent quality control. Summary of the Invention

[0007] One objective of this application is to provide a defect detection method, electronic device, and product based on holographic spectroscopy, at least to address the problems of low accuracy, susceptibility to noise and background interference, and inability to perform material property analysis in existing detection methods based on two-dimensional grayscale or bright-field images for transparent, highly transparent, and low-contrast materials. To achieve the above objective, some embodiments of this application provide the following aspects:

[0008] In a first aspect, this application provides a defect detection method based on holographic spectroscopy, the method comprising:

[0009] An optical image of the substrate surface and spectral signals in multiple bands corresponding to each pixel in the optical image are acquired, and a holographic spectral data cube is constructed based on the optical image and spectral signals.

[0010] The spectral curve of each pixel is obtained based on the holographic spectral data cube;

[0011] Spectral feature vectors are extracted from the spectral curves, and a material distribution map is generated based on the spectral feature vectors.

[0012] The material distribution map is compared with the material distribution map of a preset standard sample to identify abnormal areas.

[0013] Secondly, some embodiments of this application also provide an electronic device, the electronic device comprising: one or more processors; and a memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method described above.

[0014] Thirdly, some embodiments of this application also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the method described above.

[0015] Compared with related technologies, the solution provided in this application, based on holographic spectroscopy, achieves joint representation of the sample in both spatial and spectral dimensions during the acquisition process. Furthermore, based on spectral clustering, material classification, and anomaly detection algorithms, spectral features of different regions are decomposed and compared, thereby achieving precise localization and material determination of defect areas. This solution enables highly sensitive defect identification even in transparent, high-transmittance, and low-contrast materials, while also distinguishing between physical structural anomalies and differences in chemical composition. This not only significantly improves the accuracy and automation of defect detection but also expands the interpretability and traceability of detection results, providing new tools for process monitoring and quality management. Attached Figure Description

[0016] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings represent similar elements. Unless otherwise stated, the drawings are not to be scaled.

[0017] Figure 1 A flowchart illustrating a defect detection method based on holographic spectroscopy, provided as an exemplary embodiment of this disclosure;

[0018] Figure 2 A schematic diagram of the structure of a defect spectral database provided as an exemplary embodiment of this disclosure;

[0019] Figure 3 This is an exemplary structural diagram of an electronic device provided for some embodiments of this application. Detailed Implementation

[0020] 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, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Figure 1 A flowchart of a defect detection method based on holographic spectroscopy provided for an exemplary embodiment of this disclosure, the method comprising:

[0022] S101. Obtain an optical image of the substrate surface and spectral signals in multiple bands corresponding to each pixel in the optical image, and construct a holographic spectral data cube based on the optical image and spectral signals.

[0023] Specifically, a holographic spectral detection system integrating visible to near-infrared wavelengths (400nm to 1100nm) can be used to image and acquire data on the surface of the substrate under test. This detection system employs a pushbroom hyperspectral camera, coupled with a uniform and stable broadband light source and a reflective optical lens, to achieve line-by-line scanning of the substrate surface. During imaging, the system simultaneously records the two-dimensional planar coordinates (X, Y) of each pixel and its spectral reflectance data in multiple wavelength bands. Based on this, the acquired spatial image data and spectral channel data are integrated to form a three-dimensional hyperspectral data structure containing both spatial and spectral dimensions, namely a holographic data cube (HDC).

[0024] Furthermore, in this process, a standard white board and dark field reference signal can be acquired first for subsequent system calibration. After acquisition, through data stacking and spectral calibration, spectral reflectance information of each spatial location within a continuous wavelength range is obtained, thereby achieving a three-dimensional data representation of "image and spectrum fusion". This cubic data structure can simultaneously reflect the spatial distribution characteristics and spectral response properties of the sample, providing a foundation for subsequent material identification and defect analysis.

[0025] S102. Obtain the spectral curve of each pixel based on the holographic spectral data cube.

[0026] Specifically, in the constructed HDC, each pixel corresponds to a spectral reflectance curve. By reading the spectral dimension data at a specified coordinate position, the reflectance variation sequence of that pixel across all wavelength channels can be obtained, thus forming a complete spectral curve to reflect the optical response characteristics of the substrate surface material at different wavelengths. For example, the reflectance sequence at coordinates (x,y) for 300 wavelengths can be extracted.

[0027] S103. Extract spectral feature vectors from the spectral curves and generate a material distribution map based on the spectral feature vectors.

[0028] Specifically, after obtaining the spectral curves of each pixel, all curves are standardized and features are extracted. Principal component analysis (PCA) or other spectral dimensionality reduction algorithms can be used to transform high-dimensional spectral data into several principal feature vectors, thereby preserving the key trends in spectral variation and reducing data dimensionality. Subsequently, unsupervised clustering analysis is performed based on the extracted spectral feature vectors. For example, K-Means or Gaussian mixture model algorithms are used to automatically classify pixels with similar spectral characteristics into the same category, thereby achieving the distinction between different material regions.

[0029] Furthermore, the clustering results described above can be mapped back to the original spatial coordinate system to generate a corresponding material distribution map. In this map, regions with similar spectral characteristics are assigned the same identifier color, visually reflecting the spatial distribution of different materials or film layers on the substrate surface. This material distribution map can reveal material differences that are difficult to identify using traditional visual inspection, providing a basis for subsequent anomaly identification.

[0030] S104. Compare the material distribution map with the preset standard sample material distribution map to identify abnormal areas.

[0031] Specifically, after obtaining the material distribution map of the sample to be tested, it is compared with the material distribution map of a pre-established standard sample. The standard sample can be a qualified wafer or substrate that has undergone process verification, and its material distribution is typical and stable. During the comparison process, the system detects areas where the material distribution deviates, i.e., abnormal areas, based on pixel-level or region-level classification results.

[0032] If a region is absent from the standard sample or has a significantly different material type, it is identified as an anomaly. Furthermore, by comparing differences in spectral characteristics, anomaly regions can be further subdivided into possible types such as contamination, scratches, and film anomalies. For example, if the spectral curve of an anomaly region shows an abnormal shift in absorption or reflection peaks in a specific wavelength band, it may correspond to changes in film thickness or composition. Finally, anomaly regions can be spatially located and labeled on a material distribution map to generate a defect distribution map.

[0033] Through this comparison and recognition process, even if some anomalies do not show obvious morphological differences in visible light images, they can be accurately detected through differences in spectral features, thereby significantly improving the sensitivity and reliability of detection.

[0034] In the above embodiments, holographic spectroscopy technology enables the joint representation of the sample in both spatial and spectral dimensions during the acquisition process. Based on spectral clustering, material classification, and anomaly detection algorithms, spectral features of different regions are decomposed and compared, thereby achieving precise localization and material determination of defect areas. This approach enables highly sensitive defect identification even in transparent, high-transmittance, and low-contrast materials, while also distinguishing between physical structural anomalies and differences in chemical composition. It significantly improves the accuracy and automation of defect detection, expands the interpretability and traceability of detection results, and provides new tools for process monitoring and quality management.

[0035] In one embodiment, the step of acquiring an optical image of the substrate surface and spectral signals in multiple bands corresponding to each pixel in the optical image, and constructing a holographic spectral data cube based on the image and spectral signals specifically includes:

[0036] S201. Scan the surface of the substrate line by line to obtain two-dimensional plane pixel coordinates and spectral channel data under multiple corresponding wavebands;

[0037] S202. Based on the pixel coordinates and the corresponding spectral channel data, form three-dimensional hyperspectral data containing spatial dimensions (X, Y) and spectral dimensions (λ);

[0038] Wherein, the XY plane represents a two-dimensional image of the substrate under a single wavelength, and the Z axis is the spectral dimension, used to characterize the reflectivity or light intensity distribution under different wavelengths;

[0039] S203. Construct a holographic spectral data cube using the three-dimensional hyperspectral data.

[0040] Specifically, in step S201, the surface of the substrate to be detected is scanned line by line, and an optical image of the substrate surface and the spectral signals of each pixel in the optical image under multiple bands are obtained using a hyperspectral imaging device.

[0041] In its implementation, the hyperspectral camera operates using a pushbroom scanning method, acquiring reflected light signals in the 400nm–1100nm wavelength range through a spectral dispersive element. Each scan line corresponds to a set of spectral channel data, including spatial coordinates (X, Y) and spectral dimension (λ).

[0042] For example, during a single scan, the system can obtain two-dimensional spatial information with a spatial resolution of 2560×2160 pixels, as well as reflectance or light intensity data for approximately 300 spectral channels, thereby forming a three-dimensional data matrix I(x,y,λ); where x,y: represent the spatial coordinates of pixels on the substrate surface; λ: represent the wavelength range of the spectral channels (400nm~1100nm).

[0043] The two-dimensional spatial data obtained by line-by-line scanning is stacked with the corresponding spectral channel data to construct a three-dimensional hyperspectral data structure containing spatial dimensions (X, Y) and spectral dimensions (λ). In this data structure, the XY plane represents the two-dimensional image at a single wavelength, and the Z-axis (λ) represents the spectral dimension, used to characterize the reflectance or light intensity distribution at different wavelengths. Through the above data stacking, a complete holographic spectral data cube is formed, and its storage format can adopt the ENVI standard HDR+IMG binary file structure, which facilitates subsequent algorithm calls and spectral analysis.

[0044] The method of this embodiment can simultaneously obtain the spatial morphology and spectral properties of the substrate surface in the same detection process, enabling high-sensitivity detection of minute defects in low-contrast, transparent, or composite material samples.

[0045] In one embodiment, the step of obtaining the spectral curve of each pixel based on the holographic spectral data cube specifically includes:

[0046] The holographic spectral data cube is standardized, including dark field signal correction, white field normalization, and noise filtering.

[0047] Specifically, the step of obtaining the spectral curves of each pixel based on the holographic spectral data cube includes standardizing the raw spectral data and extracting spectra to ensure data accuracy and comparability. To eliminate the effects of system noise, ambient light interference, and light source inhomogeneity, the raw hyperspectral data is preprocessed, including dark current correction, whiteboard calibration, and smoothing filtering.

[0048] 1) Dark current correction

[0049] A dark-field image (I_{dark}) under no-light conditions is acquired to characterize the sensor's noise floor. Dark current correction is performed on the raw spectral image (I_{raw}), calculated using the following formula:

[0050] I_{corrected}=I_{raw}-I_{dark}

[0051] I_{corrected} represents the corrected spectral signal. The above formula can effectively eliminate the influence of detector background noise and dark current drift on the spectral signal.

[0052] 2) Whiteboard calibration

[0053] The reflection image (I_{white}) acquired using a standard white sample is normalized to obtain the relative reflectivity (R(λ)) at each wavelength. I_{sample} is the measured reflected light signal of the substrate under test at wavelength λ, calculated using the following formula:

[0054] R(λ)=(I_{sample}-I_{dark}) / (I_{white}-I_{dark})

[0055] The above formula can correct the non-uniformity of light source intensity distribution and system response, ensuring that the spectral reflectance of different samples is comparable.

[0056] 3) Smoothing Filter

[0057] The calibrated spectral data is smoothed to reduce random noise and preserve characteristic peak shapes. The Savitzky-Golay filtering algorithm, with a window width of 9 nm, can be used to suppress noise fluctuations while maintaining spectral characteristics and peak position information.

[0058] For the holographic spectral data cube after the above standardization process, the corresponding spectral curves are extracted according to the pixel coordinates (x, y). The spectral curve of each pixel consists of the reflectance (R(λ)) under multiple wavelength channels, which is used to characterize the optical response characteristics of the pixel in different bands. The extracted spectral curves can be further used for spectral clustering, feature matching, or anomaly detection to achieve intelligent identification of regions with different materials or defect regions.

[0059] The spectral standardization and extraction method in this embodiment can significantly improve the signal-to-noise ratio and stability of holographic spectral data, ensuring the accuracy and repeatability of spectral features in subsequent defect analysis.

[0060] In one embodiment, the step of extracting spectral feature vectors from the spectral curve and generating a material distribution map based on the spectral feature vectors specifically includes:

[0061] S301. For the spectral curve of each pixel, extract the spectral feature vector, wherein the spectral feature vector includes the peak value, slope and band intensity of the spectral curve;

[0062] S302. Based on the spectral feature vector, the K-means clustering algorithm is used to cluster the spectral feature vector to generate a material distribution map;

[0063] S303. Based on the material distribution map, determine the classification areas of different materials on the substrate surface, wherein the classification areas correspond to different material types.

[0064] Specifically, the spectral curve reflects the reflectance or light intensity distribution of a sample at different wavelengths, and different materials exhibit significant differences in their spectral reflectance characteristics. Therefore, several typical parameters can be extracted from each spectral curve, such as: the position of the spectral peak (representing the center wavelength of the main absorption or reflection band), the slope of the spectral curve (reflecting the trend of the spectrum changing with wavelength), and the spectral intensity of each key band (used to characterize the reflectance at specific wavelengths). By combining these parameters, a multi-dimensional spectral feature vector can be formed to quantitatively characterize the optical response properties of each pixel.

[0065] Subsequently, based on the extracted spectral feature vectors, an unsupervised clustering algorithm (such as K-Means) is used to perform cluster analysis on the global pixel data. This algorithm divides pixels into several regions with adjacent cluster centers based on the similarity of their spectral features. Each cluster corresponds to a material type with similar spectral characteristics, such as metal contamination, silicon-based regions, oxide films, polymer residues, etc. After clustering, the clustering results can be mapped back to the spatial coordinate system of the samples, thereby generating a two-dimensional material distribution map.

[0066] Furthermore, the material distribution map illustrates the spatial distribution of different material types on the substrate surface. Each region represents a cluster of a specific material type. By comparing the clustering results with the spectral distribution of standard samples, the composition, uniformity, and potential abnormal areas of the sample surface can be visually identified. Latent defects that are not easily observed directly in optical images (such as slight contamination or abnormal material composition) can also be distinguished in the material distribution map through differences in their spectral characteristics.

[0067] In the above embodiments, features can be automatically extracted based on spectral information, classification and clustering can be completed, and material distribution maps can be generated, thereby achieving high-precision identification of material differences in micro-regions on the substrate surface. This not only improves detection sensitivity but also provides a data foundation for subsequent defect identification and film thickness inversion.

[0068] In one embodiment, the step of determining abnormal areas by comparing the material distribution map with a preset standard sample material distribution map specifically includes:

[0069] S401. Obtain the material distribution map of the preset standard sample;

[0070] S402. Calculate the cluster center of each classification region in the material distribution map;

[0071] S403. Compare the cluster centers with the cluster centers of the preset standard sample material distribution map to obtain the cluster center deviation;

[0072] S404. If the cluster center deviation exceeds a preset threshold, the corresponding region is marked as an abnormal region, and an abnormal region set is obtained.

[0073] Specifically, the standard sample material distribution map is a reference spectrum obtained using the same holographic spectral detection method and data processing flow for substrates under qualified or normal process conditions. This spectrum records the typical distribution characteristics of each material category in the standard sample and its spectral cluster centers, which can be used as a comparison benchmark for subsequent defect detection.

[0074] Subsequently, for the material distribution map of the sample under test, the cluster center of each classification region is calculated. The mean vector of the spectral feature vectors of the pixels in each classification region can be obtained as the cluster center point of that region. This cluster center comprehensively reflects the representative spectral response characteristics of the region, and is used to describe the average state of its material properties.

[0075] Next, the cluster centers of each classification region of the sample to be tested are compared with the cluster centers of the corresponding material categories in the material distribution map of the standard sample. By calculating the difference in spectral feature distance or similarity between the two, such as by using Euclidean distance, cosine similarity, or spectral angle mapping (SAM), the cluster center deviation value of each material category is obtained.

[0076] Furthermore, when the deviation of a cluster center exceeds a preset threshold, it indicates that the spectral characteristics of that region differ significantly from those of the standard sample. In this case, the corresponding region is marked as an anomalous region, and all marked regions are grouped into an anomalous region set. Anomalous regions may correspond to different types of defects, including material contamination, process residues, abnormal film composition, or non-uniform material distribution.

[0077] Through the steps described above, this embodiment enables automated anomaly detection by comparing standard samples. This method can not only identify macroscopic defects but also detect microscopic material differences that are difficult to perceive with the naked eye through spectral feature deviations, exhibiting high sensitivity and high stability. Compared to traditional image detection methods based on grayscale or color differences, this method has significant advantages in identifying latent defects such as changes in optical properties and compositional shifts.

[0078] In one embodiment, the method further includes:

[0079] The spectral information of the abnormal region is compared with the defect template information in the defect spectral database to identify the relevant defects;

[0080] Specifically, in the detection process, once an abnormal region is located, the average spectral curve or feature vector of all pixels within that region is extracted. Next, this spectral information is compared with all templates in the defect spectral database to calculate the matching degree. The matching algorithm can employ spectral angle matching (SAM), correlation coefficient calculation, or utilize a pre-trained classification model using machine learning algorithms (such as Support Vector Machines (SVM), neural networks, etc.).

[0081] This classification model uses the spectral features of standard defect samples as training data and can output the probability or confidence level that an unknown spectrum belongs to a specific defect type in the database. Finally, the system assigns the defect type with the highest matching degree or the highest classification confidence (e.g., "copper residue" or "photoresist residue") to the abnormal region, thereby completing the automatic identification and classification of defects.

[0082] In this embodiment, the defect type is determined by identifying abnormal areas, providing process engineers with direct and actionable clues for tracing the source, greatly shortening the problem diagnosis cycle. Standardized database comparison and algorithmic classification replace manual microscopic inspection and experience-based judgment, ensuring the efficiency, consistency, and repeatability of defect identification.

[0083] Furthermore, in one embodiment, the method for constructing the defect spectral database may include the following steps:

[0084] First, various typical process defects are artificially created on a standard silicon substrate to serve as standard samples for subsequent spectral acquisition. Specific defect types include: forming gold (Au) or platinum (Pt) particles via sputtering to simulate metal residue defects; performing incomplete development or cleaning after spin-coating photoresist to simulate organic contamination defects; and precisely controlling the oxide layer thickness using dry etching to simulate abnormal oxide layer defects.

[0085] Subsequently, a hyperspectral imaging (HSI) system consistent with the online detection system was used to scan all the prepared defect standards. To ensure the statistical validity and robustness of the subsequent training model, spectral data from no fewer than 100 sample points were collected for each defect type.

[0086] The acquired raw spectral data is then processed to extract key features that can effectively distinguish different defect types. For example, the reflectance ratio at specific wavelengths can be extracted. For instance, for metallic residues, the ratio of reflectance at 600 nm to reflectance at 550 nm (R(600nm) / R(550nm)) can be calculated as a feature value; for organic matter, the slope of its spectral curve near 950 nm can be calculated as a feature value. These calculated feature values ​​are combined into a feature vector to characterize the sample point, such as... Figure 2 .

[0087] Finally, the extracted feature vectors are used to train a classification model. The feature vectors are used as input to the model, and mature machine learning algorithms such as Support Vector Machine (SVM) or Random Forest (RF) are employed for training. After training, the model can analyze the feature vectors of an unknown sample and output the probability that it belongs to a certain defect type defined in the database.

[0088] In the above embodiments, by constructing a standardized defect database and classifying defects based on comparison and algorithms, the efficiency, consistency and repeatability of defect determination are ensured.

[0089] Furthermore, when new defect types appear on the production line that are not included in the database, operators or process engineers can manually annotate them and add their spectral characteristics to the database. This online update mechanism allows the database to be continuously enriched and improved, enabling it to adapt to process changes and maintain high identification accuracy.

[0090] In one embodiment, after the step of obtaining the spectral curve of each pixel based on the holographic spectral data cube, the method further includes:

[0091] S501. The spectral curve is fitted and analyzed using an optical thin film model to retrieve the thickness parameters of the corresponding film layer.

[0092] Specifically, firstly, for the expected film structure on the substrate under test (e.g., on a Si substrate) Single-layer film, or Si / (SiN multilayer film stack), pre-built one or more optical thin film model libraries. This model library is based on physical optics principles, such as the transfer matrix method (TMM), and can accurately calculate the theoretical reflection spectrum of the structure across the entire spectral range based on the input optical constants (refractive index n, extinction coefficient k) and thickness (d) of the film material.

[0093] After obtaining the actual measured spectral curve for each pixel, the corresponding optical model is invoked, and an iterative optimization algorithm (such as least squares method or machine learning-based regression model) is used to fit the measured spectral curve with the theoretical spectral curve generated by the model. The goal of this algorithm is to find a set of optimal film thickness and optical constant parameters that minimizes the sum of squared residuals between the theoretical and measured spectra.

[0094] Once the fitting converges, the optimal parameter solution obtained is the physical parameter of the film corresponding to that pixel, with the key parameter being the film thickness at that point. Through this process, a precise inversion from spectral data to nanometer-level thickness information is achieved.

[0095] S502. Generate a film thickness distribution map based on the inverted film thickness data, and evaluate the film thickness and uniformity through the film thickness distribution map.

[0096] Specifically, after performing the inversion process S501 on all pixels or regions of interest (ROIs) on the substrate surface, a two-dimensional thickness data matrix with the same size as the optical image is obtained, where the value of each matrix element represents the film thickness at the corresponding spatial location. This thickness data matrix can be visualized, for example using pseudo-color mapping, to generate an intuitive "film thickness distribution map." In this map, different colors directly correspond to different film thicknesses, making the spatial variation of thickness immediately apparent.

[0097] This film thickness distribution map allows for qualitative observation, identifying macroscopic uniformity of the film across the entire substrate, the presence of edge effects, and process issues such as center overthrow or underthrow. It also enables quantitative statistical analysis. For example, it can calculate key process parameters such as the average thickness, standard deviation (as a quantitative indicator of uniformity), and maximum / minimum values ​​for the entire map, and compare them with preset process control specifications. Any area with a thickness value exceeding the specifications can be automatically identified and marked as a "thickness anomaly defect," thus enabling the detection of "functional defects" that are completely invisible to traditional visual methods.

[0098] In this embodiment, not only can the material type be identified, but the thickness of the film layer on the substrate can also be detected, greatly enriching the dimensions and depth of the detection information. This embodiment can accurately identify and quantify functional defects caused by uneven coating, etching residue, abnormal growth, etc., which have extremely low contrast or are even invisible under conventional optical images, significantly improving the comprehensiveness of the detection. Furthermore, by generating a global film thickness distribution map, direct and visualized data feedback can be provided for the process stability and uniformity of the film deposition equipment, providing strong support for advanced process control and yield improvement.

[0099] In one embodiment, the method further includes:

[0100] The film thickness distribution map is compared with the film thickness distribution map of a preset standard sample to identify film-related defects.

[0101] Specifically, the preset standard sample film thickness distribution map can be an actual measurement map of a product manufactured using the same process and confirmed to be qualified, or it can be an idealized digital template generated according to process design specifications (such as target thickness ± tolerance range). The comparison process can employ an image subtraction algorithm, that is, spatially aligning the film thickness map of the sample to be tested with the standard film thickness map, and then calculating the thickness difference between the two pixel by pixel to generate a "thickness difference map". In this difference map, the value of each pixel represents the deviation of the thickness of the sample to be tested at that point from the standard value.

[0102] Subsequently, the system can binarize the "thickness difference map" according to a preset judgment threshold (e.g., the thickness fluctuation range allowed by the process specifications, such as ±3nm or ±5%). Any pixel whose absolute deviation exceeds the threshold is identified and marked as a film-related defect. These defects can be further classified into specific process problems such as "film too thick," "film too thin," "poor uniformity," and "sharp edge drop." Finally, the system can output a defect report that marks the location, area, and degree of deviation of all film-related defects.

[0103] In this embodiment, film thickness anomalies that are invisible under traditional optical imaging but directly affect the electrical or optical performance of devices can be reliably detected. By quantitatively comparing the measurement results with standards, the stability and uniformity of key process steps such as coating and etching can be objectively and accurately evaluated, providing a data foundation for achieving more stringent process control. Compared to spot measurement with a film thickness gauge, this method enables comprehensive, rapid, and non-contact inspection of the entire substrate and automatically marks all non-compliant areas, greatly improving inspection efficiency and coverage.

[0104] In one embodiment, the optical thin film model is established based on the Fresnel equation or the transfer matrix method to describe the reflection and transmission relationship of incident light at multiple interfaces between air / film / substrate. The reflectivity as a function of wavelength λ is determined by the film thickness d, refractive index n(λ), and incident angle θ. The model is fitted by calculating the deviation between the theoretical reflectivity curve and the measured spectral curve.

[0105] Specifically, to achieve quantitative inversion from spectral data to physical film thickness, an optical thin film reflection model is first established. This model can be constructed based on the Fresnel equation or the Transfer Matrix Method (TMM) to describe the reflection and transmission behavior of incident light at multiple interfaces between air / film / substrate.

[0106] Specifically, when broadband light illuminates the surface of a sample containing a thin film structure, the incident light undergoes partial reflection and transmission between the interfaces of each layer, resulting in multiple interferences within the film. This interference effect leads to periodic variations in reflectivity at different wavelengths, and this spectral fluctuation is closely related to parameters such as film thickness, refractive index, and optical loss. Therefore, the actual thickness of the film can be deduced by analyzing the reflectivity spectrum curve. In theoretical modeling, reflectivity R(λ) can be expressed as a function of wavelength λ, primarily depending on parameters such as film thickness d, refractive index n(λ), extinction coefficient k(λ), and incident angle θ. For single-layer film structures, the model can describe the complex amplitude coefficients of reflection and transmission using Fresnel equations; for multilayer film structures, the transfer matrix method can be used, representing the optical properties of each layer as a matrix, and obtaining the overall reflectivity of the entire structure through matrix multiplication. This method can accommodate optical behavior under different numbers of film layers, material types, and angles, resulting in high modeling accuracy.

[0107] The refractive index and extinction coefficient of the film can be determined based on known materials (e.g., ...). The optical constants (such as SiN, ITO, etc.) are obtained from optical constant databases in the visible and near-infrared regions, or obtained through experimental calibration. The incident angle θ is usually set under normal or small-angle incident conditions to simplify calculations. The theoretical reflectivity curves for different film thicknesses can be calculated using the above model, providing a basis for subsequent parameter inversion.

[0108] In the above embodiments, the establishment of the model realizes the mapping from spectral reflectance data to the physical structure parameters of the film, enabling the holographic spectral detection system not only to determine the surface reflectance intensity, but also to deduce the true thickness distribution inside the film, significantly enhancing the analytical capabilities of the detection system.

[0109] In one embodiment, the inversion of the film thickness parameters is achieved using a nonlinear least squares optimization algorithm, and a film thickness confidence map is generated based on the residuals to evaluate the reliability of the film thickness inversion results.

[0110] Specifically, after establishing the optical model, the measured pixel spectral curves are fitted and analyzed with the theoretically calculated reflectance curves to obtain the film thickness parameters that best match the measured data. A nonlinear least squares optimization algorithm can be used as the core algorithm for the fitting solution. During the fitting process, the film thickness d is used as the main parameter to be determined, while the refractive index n(λ) and extinction coefficient k(λ) are known constants. An error function is constructed by calculating the difference between the theoretical curve R_th(λ,d) and the measured curve R_exp(λ):

[0111]

[0112] The algorithm iteratively optimizes to find the film thickness value d_opt that minimizes the error E(d), which is the film thickness parameter of the pixel obtained through inversion. In each iteration, the algorithm updates the estimated film thickness and recalculates the theoretical reflectivity curve until the difference between the two is less than a preset convergence threshold. This approach achieves high inversion accuracy while maintaining computational efficiency, making it suitable for batch fitting of large-scale pixels. Furthermore, to improve inversion stability and convergence speed, the initial film thickness can be limited to a range (e.g., 10nm to 1000nm), and the step size can be adaptively adjusted using local gradient information. For multilayer film structures, constraints can be introduced between the thicknesses of each layer to prevent inversion errors caused by parameter coupling. In this step, the corresponding film thickness value can be obtained at each pixel, thus forming a pixel-level film thickness data matrix.

[0113] After obtaining the inversion film thickness value for each pixel, the fitting residual and its statistical characteristics are further calculated to evaluate the reliability of the inversion results.

[0114] Specifically, a "film thickness confidence map" can be generated based on the sum of squared fitting errors or the standard deviation of residuals. The smaller the residual and the higher the curve fit, the higher the reliability of the film thickness result for that pixel; conversely, if the residual is large or the curve fit is low, the pixel may be affected by noise, material anomalies, or multilayer interference uncertainties, and its film thickness value has low reliability.

[0115] Film thickness confidence maps are typically displayed in grayscale or pseudocolor, providing a direct visual representation of the reliability of film thickness inversion results for each region. In subsequent film thickness distribution analysis, the system can assign weights based on the confidence level, filtering out or interpolating low-confidence regions to improve the overall accuracy of the analysis. Simultaneously, reconstructing the film thickness results of each pixel using spatial coordinates generates a "film thickness distribution map." This map uses different colors or grayscale levels to represent film thickness variations, clearly displaying film uniformity, thickness gradient direction, and local anomalies. When the film thickness in certain regions significantly deviates from the average or exhibits abrupt changes within a short distance, it can be determined that the region contains film formation anomalies, thickness inhomogeneities, or optical interference defects. Through joint analysis of the film thickness distribution map and confidence map, the system can not only achieve quantitative measurement of film thickness but also perform self-checking and quality assessment of the measurement results, thereby ensuring the reliability of the analysis.

[0116] This embodiment introduces an optical thin film model and, through precise fitting of the theoretical model with measured spectra, can accurately invert the film thickness at each pixel, achieving non-contact, non-destructive full-field film thickness measurement. Furthermore, residual analysis generates a film thickness confidence map, which not only identifies areas of abnormal film thickness but also automatically assesses the reliability of the inversion results. Moreover, it can simultaneously obtain film thickness and uniformity information for the entire sample surface in a single scan, providing an efficient and reliable technical means for film control and defect early warning during the production process.

[0117] In one embodiment, the method further includes generating a test report based on the analysis results.

[0118] Specifically, based on the holographic spectral data cube and spectral analysis results obtained in the aforementioned embodiments, a test report for defect assessment and process diagnosis is further generated. This test report integrates defect spatial distribution, material classification, film thickness consistency, and statistical analysis results, used for visual and quantitative evaluation of the tested object. Based on defect identification results and holographic spectral data analysis, a test report containing multidimensional information is generated.

[0119] In one embodiment, the report includes a defect location distribution map to show the spatial distribution of abnormal areas.

[0120] Specifically, based on the defect classification results, a spatial distribution map of the abnormal area is drawn in the two-dimensional coordinate system of the substrate to intuitively display the distribution density and location relationship of the defects, which is used to assist in localization and source tracing analysis.

[0121] In one embodiment, the report further includes a film thickness distribution map to assess the uniformity of film thickness.

[0122] Specifically, by performing spectral fitting on the interference fringes or reflectance curves in the full-spectrum data, the film thickness corresponding to each pixel is calculated, and a film thickness distribution map is generated to evaluate the uniformity of the film thickness and the stability of the process.

[0123] In one embodiment, the report may also include the defect type at the defect location.

[0124] Specifically, the distribution map is used to label the defect areas with types such as "copper residue", "photoresist residue", and "medium inhomogeneity" to quickly identify abnormal areas with different causes.

[0125] In one embodiment, the report also includes global statistical results, including the number of defects, area percentage, average film thickness, and thickness deviation.

[0126] Specifically, the system automatically calculates indicators such as the total number of defects, the percentage of defect area, the average film thickness, and the thickness deviation, generating a global statistical table to provide quantitative basis for process optimization and quality tracking.

[0127] The test report generation method described in the above embodiments can realize defect visualization and location, material qualitative analysis and thickness consistency assessment in a single test, greatly improving the interpretability and traceability of test results, and providing reliable data support for subsequent process adjustments and yield improvement.

[0128] Furthermore, some embodiments of this application also provide an electronic device. The electronic device can be various forms of digital computer, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, etc. The electronic device can also be various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices.

[0129] The electronic device includes: one or more processors; and a memory storing computer program instructions that, when executed, cause the processor to perform the steps of the methods provided in any one or more of the above embodiments. Figure 3 An exemplary structural diagram of the electronic device is disclosed. The electronic device includes one or more processors 1101, a memory 1102, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some other embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple electronic devices can be connected, each providing some of the necessary operations. The components, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0130] The electronic device may further include an input device 1103 and an output device 1104. The processor 1101, memory 1102, input device 1103 and output device 1104 may be connected by a bus or other means, as shown in the figure, which is connected by a bus.

[0131] Input device 1103 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the electronic device, such as a touch screen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 1104 may include a display device, auxiliary lighting device (e.g., LED), and haptic feedback device (e.g., vibration motor). The display device may include, but is not limited to, a liquid crystal display, a light-emitting diode display, and a plasma display. In some embodiments, the display device may be a touch screen.

[0132] To provide interaction with the user, the electronic device can be a computer. The computer has: a display device (e.g., a cathode ray tube or LCD monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback); and input from the user can be received in any form (e.g., voice input or tactile input).

[0133] In this embodiment, a computer-readable medium stores a computer program / instructions that, when executed by a processor, implement the steps of the methods provided in any one or more of the above embodiments. This computer-readable medium may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into that device. The aforementioned computer-readable medium carries one or more computer-readable instructions.

[0134] The memory 1102 can serve as a non-transitory computer-readable storage medium, used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The processor 1101 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions, and modules stored in the memory 1102, thereby implementing the program instructions / modules corresponding to the methods provided in any one or more of the embodiments described above in this application.

[0135] The memory 1102 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 1102 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 1102 may optionally include memory remotely located relative to the processor 1101, and these remote memories can be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0136] It should be noted that the computer-readable medium described in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. Computer-readable media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections having one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0137] Computer-readable media include permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory, static random access memory, dynamic random access memory, other types of random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other memory technologies, read-only optical discs, digital versatile optical discs or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0138] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including local area networks (LANs) or wide area networks (WANs), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0139] In the above embodiments, all or part of the implementation can be achieved through software, hardware, firmware, or any combination thereof. For example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In some embodiments, the software program of this application can be executed by a processor to implement the above steps or functions. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, magnetic or optical drives, floppy disks, and similar devices. In addition, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.

[0140] The computer program product provided in this application includes one or more computer programs / instructions. When executed by a processor, these computer programs / instructions generate, in whole or in part, the processes or functions described in this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0141] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0142] The scope of this application is defined by the appended claims rather than the foregoing description, and is therefore intended to encompass all variations falling within the meaning and scope of equivalents of the claims. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device in software or hardware. Terms such as "first," "second," etc., are used only for distinguishing descriptions and do not indicate any particular order, nor should they be construed as indicating or implying relative importance.

[0143] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily made by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims, and the above embodiments should be regarded as exemplary and non-limiting.

Claims

1. A defect detection method based on holographic spectroscopy, characterized in that, The method includes: An optical image of the substrate surface and spectral signals in multiple bands corresponding to each pixel in the optical image are acquired, and a holographic spectral data cube is constructed based on the optical image and spectral signals. The spectral curve of each pixel is obtained based on the holographic spectral data cube; Spectral feature vectors are extracted from the spectral curves, and a material distribution map is generated based on the spectral feature vectors. The material distribution map is compared with the material distribution map of a preset standard sample to identify abnormal areas.

2. The method according to claim 1, characterized in that, The steps of acquiring an optical image of the substrate surface and spectral signals in multiple bands corresponding to each pixel in the optical image, and constructing a holographic spectral data cube based on the image and spectral signals specifically include: The substrate surface is scanned line by line to obtain two-dimensional planar pixel coordinates and spectral channel data for multiple corresponding wavelength bands; Based on the pixel coordinates and the corresponding spectral channel data, three-dimensional hyperspectral data containing spatial dimensions (X, Y) and spectral dimensions (λ) is formed; Wherein, the XY plane represents a two-dimensional image of the substrate at a single wavelength, and the Z-axis is the spectral dimension, used to characterize the reflectivity at different wavelengths; A holographic spectral data cube is constructed using the three-dimensional hyperspectral data.

3. The method according to claim 1, characterized in that, The step of obtaining the spectral curve of each pixel based on the holographic spectral data cube specifically includes: The holographic spectral data cube is standardized, including dark field signal correction, white field normalization, and noise filtering. From the standardized holographic spectral data cube, the corresponding spectral curves are extracted according to the pixel spatial coordinates. Each spectral curve consists of reflectance data from multiple wavelength channels, which is used to characterize the optical response characteristics of the pixel in different wavelength bands.

4. The method according to claim 1, characterized in that, The step of extracting spectral feature vectors from the spectral curves and generating material distribution maps based on the spectral feature vectors specifically includes: For the spectral curve of each pixel, a spectral feature vector is extracted, wherein the spectral feature vector includes the peak value, slope, and band intensity of the spectral curve; Based on the spectral feature vector, the K-means clustering algorithm is used to cluster the spectral feature vector to generate a material distribution map; Based on the material distribution map, the classification areas of different materials on the substrate surface are determined, wherein the classification areas correspond to different material types.

5. The method according to claim 1, characterized in that, The step of comparing the material distribution map with the material distribution map of a preset standard sample to determine abnormal areas specifically includes: Obtain the material distribution map of the preset standard sample; Calculate the cluster center of each category region in the material distribution map; The cluster center is compared with the cluster center of the preset standard sample material distribution map to obtain the cluster center deviation; If the cluster center deviation exceeds a preset threshold, the corresponding region is marked as an abnormal region, and an abnormal region set is obtained.

6. The method according to claim 1, characterized in that, After the step of obtaining the spectral curve of each pixel based on the holographic spectral data cube, the method further includes: The spectral curves are fitted and analyzed using an optical thin film model to retrieve the thickness parameters of the corresponding film layer. A film thickness distribution map is generated based on the inverted film thickness data, and the film thickness and uniformity are evaluated using the film thickness distribution map.

7. The method according to claim 6, characterized in that, The method further includes: The spectral information of the abnormal region is compared with the defect template information in the defect spectral database to identify the relevant defects; And / or, The film thickness distribution map is compared with the film thickness distribution map of a preset standard sample to identify film-related defects.

8. The method according to claim 7, characterized in that, The method further includes generating a test report based on the analysis results, the report including: Defect location distribution map, used to show the spatial distribution of abnormal areas; And / or film thickness distribution map, used to assess the uniformity of film thickness; and / or the type of defect at the location of the defect; And / or global statistical results, including the number of defects, area ratio, average film thickness, and thickness deviation.

9. An electronic device, characterized in that, The electronic device includes: One or more processors; and A memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method as described in any one of claims 1 to 8.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 8.

Images