Spectrum-based chip defect detection method and related equipment

Abstract

The invention relates to a spectrum-based chip defect detection method and related equipment. The spectrum-based chip defect detection method comprises the following steps: collecting a spectrum signal reflected from a to-be-detected chip through a preset spectrometer; performing spectral feature extraction on the spectral signal to obtain an extracted spectral feature; performing feature analysis on the extracted spectral features to obtain a feature analysis result; if the feature analysis result is that chip defect features exist in the extracted spectral features, defect area positioning is performed on the to-be-detected chip by using a preset spectral imaging technology, and a to-be-detected chip positioning area is obtained; performing defect type analysis on the positioning area of the to-be-detected chip to obtain a defect type, obtaining a defect factor based on the defect type, and making a defect solving strategy based on the defect factor; the technical problems that a traditional defect detection method often depends on physical detection and manual inspection and is low in efficiency and high in subjectivity are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of chip detection, and in particular to a chip defect detection method based on spectroscopy and related equipment. Background Art

[0002] With the rapid development of semiconductor technology, chip manufacturing processes are becoming increasingly complex, and the detection and location of chip defects are becoming increasingly important. Traditional defect detection methods often rely on physical testing and manual inspection, which are inefficient and highly subjective. This makes it impossible to detect defects in the early stages, resulting in increased production costs and reduced product quality. Therefore, it is imperative to develop an efficient and automated defect detection method to ensure high quality and low loss in chip production.

[0003] At present, although existing detection technologies such as electron microscopy and X-ray imaging have high accuracy, they have obvious limitations in detection speed and cost. In addition, with the reduction of chip size and the improvement of integration, traditional detection methods are difficult to meet the needs of real-time monitoring. As an emerging detection method, spectroscopy technology has gradually gained attention for its advantages of non-contact and rapid response. However, in practical applications, how to accurately extract spectral features and conduct effective analysis is still a major challenge for technical implementation.

[0004] In response to these problems, a spectral-based chip defect detection method has emerged. This method can efficiently identify defect features and locate the area through the collection and feature extraction of spectral signals. This method not only improves the degree of automation of detection, but also helps to achieve more efficient defect analysis and solution strategy formulation. However, current research still needs to further explore the accuracy of spectral feature extraction and the intelligence of defect analysis algorithms to ensure its reliability and application value in complex production environments. Summary of the invention

[0005] The main purpose of the present invention is to provide a spectral-based chip defect detection method and related equipment, which solves the technical problems that traditional defect detection methods often rely on physical detection and manual inspection, and have low efficiency and strong subjectivity.

[0006] To achieve the above object, the present invention provides a chip defect detection method based on spectrum, comprising the following steps: The spectral signal reflected from the chip to be detected is collected by a preset spectrometer; Extracting spectral features from the spectral signal to obtain extracted spectral features; Performing feature analysis on the extracted spectral features to obtain feature analysis results; If the characteristic analysis result shows that there is a chip defect characteristic in the extracted spectral characteristic, the defect area of ​​the chip to be detected is located by using a preset spectral imaging technology to obtain a location area of ​​the chip to be detected; The defect type analysis is performed on the positioning area of ​​the chip to be inspected to obtain the defect type, and the defect factor is obtained based on the defect type, and a defect resolution strategy is formulated based on the defect factor.

[0007] Furthermore, the collecting of the spectral signal reflected from the chip to be detected by a preset spectrometer includes: Irradiating the surface of the chip to be detected with polychromatic light emitted by a preset broadband light source; Using a preset spectrometer to collect the polychromatic light reflected from the surface of the chip to be detected, and obtain a scattered spectrum signal corresponding to the polychromatic light; Calibrate the scattered spectrum signal to obtain a calibration spectrum signal; Performing time-frequency analysis on the calibration spectrum signal to obtain an instantaneous frequency signal; The instantaneous frequency signal is subjected to multi-scale decomposition to obtain a spectrum signal.

[0008] Further, extracting spectral features from the spectral signal to obtain extracted spectral features includes: De-noising and smoothing the spectral signal to obtain a de-noised and smoothed spectral signal; Performing multi-level decomposition on the denoised and smoothed spectral signal to obtain a multi-level decomposed spectral signal; wherein the multi-level decomposed spectral signal includes a frequency domain component, a time domain component and a scale component; Performing feature enhancement on the multi-level decomposition spectral signal to obtain a feature enhanced spectral signal; Decoupling the feature-enhanced spectral signal by using a variational autoencoder to obtain a decoupled independent spectral feature; Performing feature fusion on the decoupled independent spectral features to obtain a feature fusion spectral signal; Spectral feature recognition and extraction are performed on the characteristic fusion spectral signal to obtain extracted spectral features.

[0009] Further, feature analysis is performed on the extracted spectral features to obtain feature analysis results, including: Performing multi-dimensional analysis on the extracted spectral features by using a preset tensor decomposition technology to obtain a chip feature sub-tensor; Performing nonlinear geometric analysis on the chip characteristic subtensor to obtain intrinsic curvature characteristics; Performing fractional Stockwell transformation on the intrinsic curvature feature to obtain a time-frequency-scale joint distribution feature; Performing anisotropic diffusion filtering on the time-frequency-scale joint distribution features to obtain an enhanced multi-scale feature map; Perform feature extraction on the enhanced multi-scale feature map to obtain a high-order statistical feature tensor; Performing tensor sparse coding on the high-order statistical feature tensor to obtain sparse representation coefficients; Performing online dictionary learning on the sparse representation coefficients to obtain an adaptive feature dictionary; Perform feature analysis on the adaptive feature dictionary to obtain a feature analysis result.

[0010] Furthermore, the defect area of ​​the chip to be inspected having chip defect characteristics is located by using a preset spectral imaging technology to obtain the location area of ​​the chip to be inspected, including: Acquire a spectrum signal of the chip to be detected corresponding to the chip defect feature, and use the spectrum signal of the chip to be detected corresponding to the chip defect feature as a reconstructed spectrum signal; The reconstructed spectral signal is spatially and spectrally jointly reconstructed by spectral imaging technology to obtain a spectral image; Performing defect area detection on the spectral image to obtain a preliminary defect area image; Performing morphological processing on the preliminary defect region to obtain a morphological defect region image; Binarizing the morphological defect region image to obtain a candidate defect region mask; Extracting the skeleton structure of the candidate defect region mask to obtain a topological skeleton of the defect region; Based on a preset region growing algorithm, the topological skeleton of the defect region is located to obtain each defect boundary and the coordinates corresponding to the defect boundary; The chip positioning area to be inspected is obtained based on the defect boundaries and the coordinates corresponding to the defect boundaries.

[0011] Furthermore, the defect type analysis is performed on the positioning area of ​​the chip to be inspected to obtain the defect type; a defect factor is obtained based on the defect type, and a defect solution strategy is formulated based on the defect factor, including: Perform multi-scale time-frequency analysis on the positioning area of ​​the chip to be detected by continuous wavelet transform technology to obtain a time-frequency characteristic map of the defect area; Extracting texture features from the time-frequency feature map of the defect area to obtain a texture descriptor of the defect area; Performing feature unmixing and dimensionality reduction on the texture descriptor of the defect area to obtain independent feature components; Using a preset density peak clustering algorithm to perform non-parametric clustering on the independent characteristic components to obtain defect types; Performing factor analysis on the defect type using a preset partial least squares regression technique to find a linear relationship between the defect type and a characteristic variable affecting the defect type, thereby obtaining a factor variable affecting the defect; Using Lasso regularization technology to perform sparsity processing on the factor variables affecting the defect to obtain a defect factor; Constructing a fuzzy cognitive map based on the defect factors, and obtaining a causal relationship network corresponding to the defect factors based on the fuzzy cognitive map; A defect resolution strategy is developed based on the causal network.

[0012] Furthermore, a preset density peak clustering algorithm is used to perform non-parametric clustering on the independent characteristic components to obtain defect types, including: Using a preset density peak clustering algorithm to calculate the probability density distribution of the independent feature components, and obtain a density distribution map of the multi-dimensional defect feature space; Performing fast convolution on the density distribution map of the multi-dimensional defect feature space to obtain a smoothed defect density field; Performing multi-scale analysis on the smoothed defect density field to obtain a scale space representation of the defect density field; Extracting topological features from the scale-space representation of the defect density field to obtain a persistence map of the defect density field; Analyze the persistence diagram of the defect density field to obtain a candidate defect type center and a spatial neighborhood corresponding to the candidate defect type center; wherein the spatial neighborhood is a specific spatial range affected or associated with the candidate defect type center; Performing global consistency optimization on the candidate defect type center and the spatial neighborhood corresponding to the candidate defect type center to obtain a preliminary defect type; Performing defect type verification on the preliminary defect type to obtain a defect type verification result; If the defect type verification result is within a preset defect type range, the preliminary defect type is used as the defect type; wherein the defect type includes surface defects, structural defects, material defects and process defects.

[0013] The present invention also provides a chip defect detection device based on spectrum, comprising: A reflection module, used to collect the spectral signal reflected from the chip to be detected through a preset spectrometer; An extraction module, used for extracting spectral features from the spectral signal to obtain extracted spectral features; An analysis module, used for performing feature analysis on the extracted spectral features to obtain feature analysis results; A positioning module, used for locating the defective area of ​​the chip to be detected by using a preset spectral imaging technology to obtain a positioning area of ​​the chip to be detected if the characteristic analysis result shows that there is a chip defect characteristic in the extracted spectral characteristic; The formulation module is used to analyze the defect type of the positioning area of ​​the chip to be inspected, obtain the defect type, obtain the defect factor based on the defect type, and formulate a defect resolution strategy based on the defect factor.

[0014] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0016] The spectral-based chip defect detection method provided by the present invention comprises the following steps: collecting spectral signals reflected from a chip to be detected by a preset spectrometer; extracting spectral features of the spectral signals to obtain extracted spectral features; performing feature analysis on the extracted spectral features to obtain feature analysis results; if the feature analysis results show that there are chip defect features in the extracted spectral features, locating the defect area of ​​the chip to be detected by using a preset spectral imaging technology to obtain a chip positioning area to be detected; performing defect type analysis on the chip positioning area to obtain the defect type, and obtaining a defect factor based on the defect type, and formulating a defect solution strategy based on the defect factor; the method solves the technical problems that traditional defect detection methods often rely on physical detection and manual inspection, and have low efficiency and strong subjectivity, realizes intelligent defect processing, improves the efficiency of defect management, and can effectively identify chip defect features through precise spectral feature extraction and analysis, improves the accuracy of detection results, and reduces the probability of false alarms and missed alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the steps of a chip defect detection method based on spectroscopy in one embodiment of the present invention; Figure 2 is a structural block diagram of a chip defect detection device based on spectroscopy in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0018] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] like Figure 1 As shown, Figure 1 A chip defect detection method based on spectroscopy in one embodiment of the present invention comprises the following steps: Step S1, collecting the spectral signal reflected from the chip to be detected by a preset spectrometer.

[0021] Specifically, when implementing the step of "collecting the spectral signal reflected from the chip to be tested through a preset spectrometer", it is necessary to first select a suitable spectrometer with high sensitivity and a wide wavelength range to ensure that all spectral signals reflected from the chip surface can be captured. Next, place the chip to be tested in the detection area of ​​the spectrometer to ensure that its surface is clean to reduce the impact of interference signals. The spectrometer then emits a light source of a specific wavelength to illuminate the chip, and the reflected light is collected by the detector of the spectrometer to form complete spectral data. These spectral signals are then converted into digital form and prepared for subsequent feature extraction and analysis. For example, in a high-tech semiconductor production environment, manufacturers can use this method to monitor the quality of chips in real time so that defects can be discovered and located in a timely manner during the production process, thereby reducing cost waste caused by defects and improving product quality.

[0022] Step S2, extracting spectral features from the spectral signal to obtain extracted spectral features.

[0023] Specifically, when implementing the step of "extracting spectral features from the spectral signal to obtain extracted spectral features", it is first necessary to preprocess the spectral signal collected from the chip to be detected to remove noise and background interference, so as to ensure that the extracted spectral features are more accurate and reliable. The preprocessing process may include technical means such as signal smoothing, denoising and normalization. Next, use an appropriate algorithm to extract features from the preprocessed spectral signal. Common methods include principal component analysis (PCA), waveform analysis and other machine learning techniques, which can effectively extract representative spectral features. When extracting features, it is necessary to pay attention to spectral changes related to chip defects, such as changes in reflectivity, absorption features at specific wavelengths, etc. These changes can reflect possible defects on the chip surface. For example, if there are tiny scratches or cracks on the chip surface, the spectral signal will show abnormal reflection or absorption features at specific wavelengths. By identifying these features, the system can quickly determine the presence of defects. By building a feature database, the system can compare the extracted spectral features with known defect features to improve the accuracy of recognition. In high-tech semiconductor manufacturing environments, this spectral feature extraction technology can be applied to online testing of newly produced chips to ensure that potential defects can be identified and located before the chips leave the factory, reducing subsequent problems caused by defects. In addition, by extracting and analyzing features on a large number of chips, manufacturers can continuously optimize their production processes, thereby improving overall product quality, reducing production costs, and achieving more efficient production management.

[0024] Step S3, performing feature analysis on the extracted spectral features to obtain feature analysis results.

[0025] Specifically, when implementing the step of "performing feature analysis on the extracted spectral features to obtain feature analysis results", the extracted spectral features need to be first input into the feature analysis system. The system can use a variety of analysis algorithms, such as support vector machines (SVM), decision trees or neural networks, which can effectively process complex data sets and extract meaningful information from them. By building a training model, the system can compare the spectral features with the existing defect database, analyze the correlation between them and different defect types, and thus identify potential defect features. During the feature analysis process, the analysis system will focus on the key wavelengths and characteristic patterns in the spectral features, which may represent specific types of defects. For example, a significant reduction in reflectivity at certain wavelengths may mean that there are cracks or contamination on the chip surface. By conducting in-depth analysis of these features, the system can generate feature analysis results, indicating whether the chip has defects and their possible types. This analysis can not only identify the existence of defects, but also provide a basis for subsequent defect type analysis. In high-tech semiconductor manufacturing environments, the application of this feature analysis is particularly critical. For example, when a manufacturer monitors a batch of chips on the production line and finds that the spectral characteristics deviate greatly from normal values, the feature analysis system can quickly identify the defect mode and issue an alarm to remind operators to take measures. In addition, by accumulating a large number of spectral features and analysis results, manufacturers can continuously optimize and adjust production processes to achieve higher production efficiency and lower defect rates, ultimately ensuring the stability and reliability of product quality.

[0026] Step S4: if the characteristic analysis result shows that there is a chip defect characteristic in the extracted spectral characteristic, the defect area of ​​the chip to be detected is located by using a preset spectral imaging technology to obtain a location area of ​​the chip to be detected.

[0027] Specifically, when implementing the step of "if the characteristic analysis result shows that there are chip defect characteristics in the extracted spectral characteristics, then using the preset spectral imaging technology to locate the defect area of ​​the chip to be detected to obtain the location area of ​​the chip to be detected", it is first necessary to determine whether the characteristic analysis result indicates the existence of defect characteristics. If the analysis result shows a signal consistent with the known defect pattern, the system will start the preset spectral imaging technology to locate the defect area of ​​the chip to be detected in detail. This process involves scanning the chip to be detected through a specific spectral imaging device to obtain higher resolution spectral data in order to accurately locate the defect area. Spectral imaging technology combines multiple wavelengths of light into an image, which can generate a rich spectral information map on the chip surface, thereby helping the system to identify specific defect areas. By analyzing the reflection characteristics of different positions in the spectral map, the system can accurately calibrate the location of the defect. For example, if the reflection characteristics of a certain area are significantly different from those of a normal area, the system will highlight the area through color or marking to form a defect location map. In high-tech semiconductor manufacturing environments, the application of this process is crucial. For example, when the characteristic analysis results of a batch of chips confirm that there are defects, manufacturers can quickly use spectral imaging technology to scan the chips and accurately identify the defect location, thereby reducing rework or scrap rates. In addition, by conducting a detailed defect type analysis of the located area, manufacturers can develop targeted solutions to further improve production efficiency and product quality. This not only enables timely response to production problems, but also provides data support for future production processes to continuously optimize manufacturing processes and improve product reliability.

[0028] Step S5, performing defect type analysis on the positioning area of ​​the chip to be inspected to obtain the defect type, and obtaining a defect factor based on the defect type, and formulating a defect resolution strategy based on the defect factor.

[0029] Specifically, when implementing the step of "analyzing the defect type of the detected chip location area to obtain the defect type, obtaining the defect factor based on the defect type, and formulating a defect solution strategy based on the defect factor", it is first necessary to analyze the defect area that has been located to determine the correspondence between the spectral characteristics of the area and the specific defect type. The system will use the established defect database to compare the features extracted from the spectral imaging with the known defect patterns to identify the specific defect type, such as scratches, cracks, contamination or other physical defects. This process may involve the application of machine learning or pattern recognition algorithms, which automatically classify and analyze the features of different areas on the chip surface. When the defect type is determined, the system will further analyze the root cause of the defect and extract the relevant defect factors. For example, if the defect type is a scratch, the defect factor may include wear of the production equipment or errors in the operation process. By analyzing these factors, the system can determine the source of the defect, thereby providing a basis for subsequent production optimization. Based on the analysis results of the defect factor, the system will formulate a targeted defect solution strategy. For example, if the analysis shows that the defect factor is related to the wear of the production equipment, the solution strategy may include replacing or repairing the relevant equipment, or optimizing the production process to reduce the recurrence of similar defects. In the semiconductor manufacturing environment, this analysis process can effectively help manufacturers improve product quality, reduce rework and scrap rates, and also provide an important reference for subsequent production improvements. For example, after a batch of chips has scratches, the analysis results show that the defect factor is related to equipment friction, then the solution may be to replace parts or optimize the process to prevent the problem from happening again.

[0030] In a specific embodiment, the collecting of the spectral signal reflected from the chip to be detected by a preset spectrometer includes: Irradiating the surface of the chip to be detected with polychromatic light emitted by a preset broadband light source; Using a preset spectrometer to collect the polychromatic light reflected from the surface of the chip to be detected, and obtain a scattered spectrum signal corresponding to the polychromatic light; Calibrate the scattered spectrum signal to obtain a calibration spectrum signal; Performing time-frequency analysis on the calibration spectrum signal to obtain an instantaneous frequency signal; The instantaneous frequency signal is subjected to multi-scale decomposition to obtain a spectrum signal.

[0031] Specifically, when implementing the step of "collecting the spectral signal reflected from the chip to be detected by a preset spectrometer", it is first necessary to use a preset broadband light source to emit polychromatic light to uniformly illuminate the surface of the chip to be detected. This process is very critical because the polychromatic light source can cover a wide range of wavelengths, so that subsequent detection can capture a variety of spectral features and ensure the comprehensiveness and accuracy of the spectral data. Subsequently, the polychromatic light reflected from the surface of the chip to be detected is collected by the preset spectrometer, and the spectrometer obtains the light signal reflected from the chip through a highly sensitive detector to form a corresponding scattered spectral signal. These scattered spectral signals contain information about tiny defects on the surface of the chip to be detected, and can provide important basic data for subsequent analysis.

[0032] Next, the collected scattered spectrum signal needs to be calibrated, which is a key step to ensure the accuracy of the measurement results. By comparing the known spectrum data of the standard sample, the system can adjust the systematic error and environmental noise in the spectrum signal to obtain a calibrated spectrum signal. This calibration not only improves the reliability of the data, but also provides a more accurate signal basis for subsequent analysis.

[0033] After obtaining the calibrated spectral signal, the system will perform time-frequency analysis on it. This analysis process can extract the changing characteristics of the signal in time and frequency, thereby obtaining the instantaneous frequency signal. The analysis of the instantaneous frequency signal can reveal the dynamic changes of the chip surface characteristics and provide important information for further defect analysis. Finally, the system will perform multi-scale decomposition on the instantaneous frequency signal. This process involves decomposing the signal into components at different scales in order to extract more detailed spectral signals. In the multi-scale decomposition, mathematical tools such as wavelet transform can be used to separate the details and main features of the signal, thereby effectively reducing the impact of noise and highlighting useful information. These spectral signals will provide an important basis for subsequent defect detection and analysis. The application of this process is particularly significant in high-tech semiconductor manufacturing environments. For example, during the production process, manufacturers can monitor the spectral signals of chips in real time to facilitate early detection of potential defects. If a signal abnormality is found in a certain band during the collection of spectral signals, after calibration and analysis, the manufacturer can quickly locate the specific defect type and take necessary corrective measures. This technology not only improves the efficiency of detection, but also provides strong support for the optimization of the production process, thereby ensuring product quality and reducing production costs. In this way, it can be ensured that each batch of chips meets strict quality standards and reduce economic losses caused by defects.

[0034] In a specific embodiment, extracting spectral features from the spectral signal to obtain extracted spectral features includes: De-noising and smoothing the spectral signal to obtain a de-noised and smoothed spectral signal; Performing multi-level decomposition on the denoised and smoothed spectral signal to obtain a multi-level decomposed spectral signal; wherein the multi-level decomposed spectral signal includes a frequency domain component, a time domain component and a scale component; Performing feature enhancement on the multi-level decomposition spectral signal to obtain a feature enhanced spectral signal; Decoupling the feature-enhanced spectral signal by using a variational autoencoder to obtain a decoupled independent spectral feature; Performing feature fusion on the decoupled independent spectral features to obtain a feature fusion spectral signal; Spectral feature recognition and extraction are performed on the characteristic fusion spectral signal to obtain extracted spectral features.

[0035] Specifically, when implementing the step of "extracting spectral features from the spectral signal to obtain extracted spectral features", the collected spectral signals need to be denoised and smoothed first. Since the spectral signals may be interfered by various noises, such as environmental noise and equipment noise, it is crucial to denoise and smooth these signals. This process usually uses filtering techniques, such as median filtering or wavelet transform, which can effectively eliminate noise and obtain clearer denoised and smoothed spectral signals. In this way, the effective information in the signal is retained, laying the foundation for subsequent feature extraction. After denoising and smoothing, the system will perform multi-level decomposition on the denoised and smoothed spectral signal. This process decomposes the signal into different components to facilitate better analysis and feature extraction. The results of the multi-level decomposition will include frequency domain components, time domain components and scale components, which represent the characteristics of the signal at different time and frequency scales, respectively, and help to reveal the intrinsic structure and pattern of the spectral signal. This decomposition can not only help understand the components of the spectral signal, but also provide a basis for feature enhancement. After obtaining the multi-level decomposed spectral signal, the next step is to enhance its features, with the goal of improving the identifiability of effective features in the signal. By using feature enhancement techniques, such as contrast enhancement or weighted processing, the contrast of features can be improved, making the subsequent decoupling process more efficient. In this step, the system will identify those key features and make them more prominent through enhancement techniques. Next, the system decouples the feature-enhanced spectral signal through a variational autoencoder, which uses the ability of the autoencoder to decompose the spectral signal into multiple independent features. This decoupling process can simplify the components of complex signals into independent features that are unrelated to each other, which helps to better understand the inherent relationships and feature structures in the signal. Through this process, the system can extract more representative decoupled independent spectral features, laying a good foundation for feature fusion. After obtaining the decoupled independent spectral features, the next step is feature fusion, which integrates different independent features to produce a comprehensive feature fusion spectral signal. Feature fusion usually uses methods such as weighted averaging or principal component analysis, which can effectively improve the expressiveness and stability of features. This fusion helps to more comprehensively capture the characteristics of the chip to be tested, especially in a complex spectral environment, and the fused signal can better reflect the true state of the chip. Finally, the system performs spectral feature recognition and extraction on the characteristic fusion spectral signal. This step aims to identify and extract key spectral features from the fusion signal to provide support for subsequent defect analysis and identification. In a high-tech semiconductor manufacturing environment, for example, when extracting spectral features from a batch of chips, this series of processing can effectively identify potential defect features, thereby providing an important basis for quality control in the production process.Through this consistent feature extraction process, manufacturers can quickly and accurately assess the quality of the chips being tested, ensuring that the final product meets strict quality standards.

[0036] In a specific embodiment, the extracted spectral features are subjected to feature analysis to obtain feature analysis results, including: Performing multi-dimensional analysis on the extracted spectral features by using a preset tensor decomposition technology to obtain a chip feature sub-tensor; Performing nonlinear geometric analysis on the chip characteristic subtensor to obtain intrinsic curvature characteristics; Performing fractional Stockwell transformation on the intrinsic curvature feature to obtain a time-frequency-scale joint distribution feature; Performing anisotropic diffusion filtering on the time-frequency-scale joint distribution features to obtain an enhanced multi-scale feature map; Perform feature extraction on the enhanced multi-scale feature map to obtain a high-order statistical feature tensor; Performing tensor sparse coding on the high-order statistical feature tensor to obtain sparse representation coefficients; Performing online dictionary learning on the sparse representation coefficients to obtain an adaptive feature dictionary; Perform feature analysis on the adaptive feature dictionary to obtain a feature analysis result.

[0037] Specifically, when implementing the step of "performing feature analysis on the extracted spectral features to obtain feature analysis results", it is first necessary to perform multi-dimensional analysis on the extracted spectral features through a preset tensor decomposition technology. This process can convert complex spectral features into chip feature sub-tensors that are easier to handle. In this process, the system will use the tensor decomposition method to extract the potential associations and structural information between the features, thereby providing high-dimensional data support for subsequent analysis. After obtaining the chip feature sub-tensor, the system will perform nonlinear geometric analysis on it to extract the intrinsic curvature feature, which describes the geometric form of the data in its original space and helps us understand the form of tiny defects on the chip surface and their mutual relationship. Next, the intrinsic curvature feature is subjected to a fractional-order Stockwell transform to obtain the time-frequency-scale joint distribution feature. This transform not only combines time and frequency information, but also provides a richer feature description, making the dynamic changes of the spectral signal more obvious. On this basis, the process of anisotropic diffusion filtering is used to further remove noise and enhance the main features of the signal, thereby obtaining an enhanced multi-scale feature map. This step ensures the clarity and recognizability of the feature map, making the subsequent analysis more effective. After obtaining the enhanced multi-scale feature map, the system will continue to extract features to obtain a high-order statistical feature tensor. This tensor contains deeper statistical characteristics of the signal and can reflect the distribution and change trend of the features at different scales. Subsequently, the system will perform tensor sparse coding on the high-order statistical feature tensor, which aims to obtain sparse representation coefficients by retaining the main components and ignoring redundant information. This process effectively reduces the data dimension and facilitates subsequent calculations and processing. Finally, online dictionary learning of the sparse representation coefficients can generate an adaptive feature dictionary, which can effectively capture the correlation and pattern between features and provide a more flexible tool for subsequent feature analysis. After feature analysis, the system finally obtains feature analysis results, which can not only be used to judge the quality of the chip, but also provide support for defect detection in actual production. For example, in the production process of semiconductor chips, after this series of analysis steps, manufacturers can accurately identify tiny defects that affect product performance, thereby optimizing the production process and improving overall product quality and yield. By combining the above methods and technologies, manufacturers can ensure that they stay ahead in the fierce market competition.

[0038] In a specific embodiment, a predetermined spectral imaging technology is used to locate a defective area of ​​a chip to be inspected having chip defect characteristics, and a location area of ​​the chip to be inspected is obtained, including: Acquire a spectrum signal of the chip to be detected corresponding to the chip defect feature, and use the spectrum signal of the chip to be detected corresponding to the chip defect feature as a reconstructed spectrum signal; The reconstructed spectral signal is spatially and spectrally jointly reconstructed by spectral imaging technology to obtain a spectral image; Performing defect area detection on the spectral image to obtain a preliminary defect area image; Performing morphological processing on the preliminary defect region to obtain a morphological defect region image; Binarizing the morphological defect region image to obtain a candidate defect region mask; Extracting the skeleton structure of the candidate defect region mask to obtain a topological skeleton of the defect region; Based on a preset region growing algorithm, the topological skeleton of the defect region is located to obtain each defect boundary and the coordinates corresponding to the defect boundary; The chip positioning area to be inspected is obtained based on the defect boundaries and the coordinates corresponding to the defect boundaries.

[0039] Specifically, when implementing the step of "using the preset spectral imaging technology to locate the defect area of ​​the chip to be detected with chip defect features and obtain the location area of ​​the chip to be detected", it is first necessary to obtain the spectral signal of the chip to be detected corresponding to the chip defect features. This process is crucial because the spectral signal not only contains the basic information of the chip, but also reveals its potential defect features. Through high-precision spectral imaging technology, we can capture the spectral response of the chip to be detected at different wavelengths, and then use these spectral signals as reconstructed spectral signals. Subsequently, the reconstructed spectral signal is reconstructed spatially and spectrally using spectral imaging technology. This step aims to combine the spectral signal with its spatial information to generate a high-quality spectral image. In this process, the system will apply a variety of algorithms, such as principal component analysis (PCA) and independent component analysis (ICA), to extract features related to chip defects. This reconstruction not only improves the resolution of the image, but also enhances the visibility of the defect features, laying the foundation for subsequent defect area detection. Once the spectral image is obtained, the system will perform defect area detection on it and generate a preliminary defect area image. At this stage, machine learning algorithms, such as support vector machines (SVM) or convolutional neural networks (CNN), are usually applied to identify and locate defects in spectral images. By analyzing the relationship between spectral features and defect features, the system can effectively identify possible defect areas. Next, the system will perform morphological processing on the preliminary defect area image to further improve the clarity of the defect area. This process includes dilation, corrosion, and opening and closing operations, the purpose of which is to remove noise and fill small holes to obtain a cleaner morphological defect area image. After morphological processing, the system will binarize the image to distinguish the defect area from the background to obtain a candidate defect area mask. The key to this step is to select a suitable threshold to ensure maximum detection of the defect area. Next, the system will perform skeleton structure extraction on the candidate defect area mask, a process aimed at simplifying the defect area into its topological skeleton. By applying a refinement algorithm, such as the Zhang-Suen algorithm, the system can extract the main structural features from the complex defect area, which will help the subsequent defect location and analysis. Based on the obtained topological skeleton of the defect area, the system will then use a preset region growing algorithm to locate the defect area. The region growing algorithm expands from the seed point and gradually includes adjacent points with similar characteristics to the seed point, thereby obtaining the boundaries of each defect. In this way, the system can accurately identify the boundaries of each defect and its corresponding coordinate information. Finally, based on the coordinates corresponding to the boundaries of each defect and the defect boundaries, the system will obtain the positioning area of ​​the chip to be tested. The determination of this positioning area can not only be used for subsequent defect analysis and processing, but also as a basis for further optimizing the chip manufacturing process.For example, in the manufacturing process of semiconductor chips, this positioning method can timely identify defects that may occur in the production process, so that corresponding improvement measures can be taken to ensure the quality and reliability of the chips. In addition, if this technology is applied to the quality control of microelectronic products, it will greatly improve production efficiency and product yield. In short, through this series of precise steps and algorithms, the use of spectral imaging technology to effectively detect and locate chip defects not only improves the accuracy of defect identification, but also provides an important basis for subsequent defect repair and quality control. The successful implementation of this method demonstrates the application potential of spectral imaging technology in modern manufacturing, especially its importance in improving product quality and production efficiency.

[0040] In a specific embodiment, the defect type analysis is performed on the positioning area of ​​the chip to be inspected to obtain the defect type; a defect factor is obtained based on the defect type, and a defect solution strategy is formulated based on the defect factor, including: Perform multi-scale time-frequency analysis on the positioning area of ​​the chip to be detected by continuous wavelet transform technology to obtain a time-frequency characteristic map of the defect area; Extracting texture features from the time-frequency feature map of the defect area to obtain a texture descriptor of the defect area; Performing feature unmixing and dimensionality reduction on the texture descriptor of the defect area to obtain independent feature components; Using a preset density peak clustering algorithm to perform non-parametric clustering on the independent characteristic components to obtain defect types; Performing factor analysis on the defect type using a preset partial least squares regression technique to find a linear relationship between the defect type and a characteristic variable affecting the defect type, thereby obtaining a factor variable affecting the defect; Using Lasso regularization technology to perform sparsity processing on the factor variables affecting the defect to obtain a defect factor; Constructing a fuzzy cognitive map based on the defect factors, and obtaining a causal relationship network corresponding to the defect factors based on the fuzzy cognitive map; A defect resolution strategy is developed based on the causal network.

[0041] Specifically, when implementing the step of "analyzing the defect type of the chip to be detected and obtaining the defect type", we first need to perform a detailed defect analysis on the chip to be detected, which will help us identify different defect types. For example, on a chip used for high-performance computing, there may be multiple types of defects such as short circuits, open circuits or crystal defects. Through this analysis, we can obtain defect factors, that is, various factors that affect the occurrence of these defects, which may be related to material properties, environmental factors or equipment parameters in the chip manufacturing process. Next, based on the defect type, we formulate a defect solution strategy. To this end, we first need to perform multi-scale time-frequency analysis on the chip to be detected through continuous wavelet transform technology. This process will help us extract the time-frequency feature map of the defect area and better analyze the changing characteristics of the defect signal in time and frequency. For example, in the time-frequency feature map, the correlation between certain specific frequency components and the defect type may be observed, which provides a basis for in-depth understanding of the defect. After obtaining the time-frequency feature map of the defect area, the system will extract texture features to obtain the texture descriptor of the defect area. These descriptors can quantify the shape, distribution and texture pattern of the defect, providing key feature information for subsequent analysis. Subsequently, the texture descriptors of the defective area are subjected to feature unmixing and dimensionality reduction to obtain independent feature components. This step helps to remove redundant information and extract more effective features for subsequent clustering analysis. The preset density peak clustering algorithm is used to perform non-parametric clustering on the independent feature components, which can effectively classify similar defect types into one category, thereby obtaining more accurate defect types. This process is crucial to improving the accuracy of defect classification because it can automatically determine the number and form of clusters based on the distance relationship between features. In addition, by performing factor analysis on the defect type using the preset partial least squares regression technique, we can find the linear relationship between the defect type and the characteristic variables that affect the defect type. This analysis helps to reveal which characteristic variables affect the occurrence of defects to what extent. After identifying the factor variables that affect the defects, the system will use the Lasso regularization technology to perform sparse processing on these factor variables to obtain defect factors. The purpose of this step is to select the most influential factors from many features, thereby simplifying the model and improving the interpretability of the analysis. Next, based on these defect factors, we construct a fuzzy cognitive map to more intuitively display the relationship between different factors. Finally, based on the fuzzy cognitive map, we obtain the causal relationship network corresponding to the defect factors, analyze the interaction between the factors and their impact on the defect type, and then develop a targeted defect solution strategy. For example, in a certain type of chip, if we find that temperature and humidity are important factors affecting the defect type, we may recommend controlling the temperature and humidity in the production environment to reduce the defect rate.This coherent process not only provides a scientific basis for defect analysis, but also provides a clear direction for subsequent improvement measures.

[0042] In a specific embodiment, a preset density peak clustering algorithm is used to perform non-parametric clustering on the independent characteristic components to obtain defect types, including: Using a preset density peak clustering algorithm to calculate the probability density distribution of the independent feature components, and obtain a density distribution map of the multi-dimensional defect feature space; Performing fast convolution on the density distribution map of the multi-dimensional defect feature space to obtain a smoothed defect density field; Performing multi-scale analysis on the smoothed defect density field to obtain a scale space representation of the defect density field; Extracting topological features from the scale-space representation of the defect density field to obtain a persistence map of the defect density field; Analyze the persistence diagram of the defect density field to obtain a candidate defect type center and a spatial neighborhood corresponding to the candidate defect type center; wherein the spatial neighborhood is a specific spatial range affected or associated with the candidate defect type center; Performing global consistency optimization on the candidate defect type center and the spatial neighborhood corresponding to the candidate defect type center to obtain a preliminary defect type; Performing defect type verification on the preliminary defect type to obtain a defect type verification result; If the defect type verification result is within a preset defect type range, the preliminary defect type is used as the defect type; wherein the defect type includes surface defects, structural defects, material defects and process defects.

[0043] Specifically, when implementing the step of "using a preset density peak clustering algorithm to perform non-parametric clustering on the independent feature components to obtain the defect type", it is first necessary to calculate the probability density distribution of the independent feature components to obtain a density distribution map of the multidimensional defect feature space. The purpose of this step is to identify the areas with higher density in the data set by calculating the density of each feature point in the multidimensional space, which usually indicate the presence of a certain type of defect. Suppose we are testing a high-performance computing chip. By calculating the density distribution of its surface features, we can identify the distribution of these features on the chip surface. Next, the density distribution map of the multidimensional defect feature space is fast convolved to obtain a smoothed defect density field. The role of fast convolution is to eliminate noise in the density distribution, making the density field smoother, so that it is easier to identify the real defect area. For example, by applying Gaussian convolution, we can smooth the sharp changes in defect features, making the changes in the density field more continuous and smooth. Afterwards, the smoothed defect density field is multi-scale analyzed to obtain a scale space representation of the defect density field. The purpose of this step is to capture defects of different sizes and shapes by observing the changes in the density field at different scales. For example, at a smaller scale, we may find subtle surface defects, while at a larger scale, we may identify a larger range of structural defects. After obtaining the scale space representation, the system will extract topological features to generate a persistence map of the defect density field. The purpose of topological feature extraction is to identify important structures in the density field, such as high-density peaks and low-density valleys. The persistence map helps us understand the changes and persistence of defect features at multiple scales. By analyzing the persistence map, we can identify those defect features that remain stable at different scales. Next, the persistence map of the defect density field is analyzed to obtain candidate defect type centers and the spatial neighborhoods corresponding to the candidate defect type centers. Candidate defect type centers refer to those points with higher density in the density field, which represent potential defect centers to a certain extent. The spatial neighborhood refers to the specific spatial range affected or associated by these candidate centers. In this way, we can identify each defect type center and its corresponding spatial range. In order to improve the accuracy of recognition, we perform global consistency optimization on the candidate defect type centers and their corresponding spatial neighborhoods to obtain preliminary defect types. The purpose of global consistency optimization is to ensure that the identification of each defect type is globally consistent by considering the distribution of the entire feature space, thereby avoiding local errors. For example, by applying the minimum cut algorithm in graph theory, we can ensure that the division of each defect type is globally consistent. After obtaining the preliminary defect type, the system will verify the defect type to obtain the defect type verification result. The verification process usually includes comparison with the preset defect model to ensure that the identified defect type actually exists.For example, by comparing with the already marked defect data, we can verify the accuracy of the identification results. If the defect type verification result is within the preset defect type range, the preliminary defect type is used as the final defect type. The preset defect types usually include surface defects, structural defects, material defects and process defects. In this way, we can ensure that the identified defect types are of practical significance and can guide subsequent repairs and improvements. For example, if a material defect is identified on a high-performance computing chip, we can further analyze its cause and take improvement measures in the production process to avoid the recurrence of similar defects. In short, by using the preset density peak clustering algorithm to perform non-parametric clustering of independent feature components, we can accurately identify different types of defects and ensure the accuracy and reliability of the identification results through a series of verification and optimization steps. This method not only improves the efficiency of defect detection, but also provides an important basis for subsequent defect repair and quality control.

[0044] The above describes the chip defect detection method based on spectroscopy in the embodiment of the present invention. The following describes the chip defect detection device based on spectroscopy in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a chip defect detection device based on spectroscopy includes: A reflection module 21 is used to collect the spectral signal reflected from the chip to be detected through a preset spectrometer; An extraction module 22 is used to extract spectral features from the spectral signal to obtain extracted spectral features; An analysis module 23 is used to perform feature analysis on the extracted spectral features to obtain feature analysis results; A positioning module 24 is used to locate the defective area of ​​the chip to be detected by using a preset spectral imaging technology to obtain a positioning area of ​​the chip to be detected if the characteristic analysis result shows that there is a chip defect characteristic in the extracted spectral characteristic; The formulation module 25 is used to analyze the defect type of the positioning area of ​​the chip to be inspected, obtain the defect type, obtain the defect factor based on the defect type, and formulate a defect solution strategy based on the defect factor.

[0045] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the description in the above method embodiment, which will not be repeated here.

[0046] Reference Figure 3 The present invention also provides a computer device in an embodiment, wherein the internal structure of the computer device can be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0047] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0048] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0049] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0050] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0051] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A chip defect detection method based on spectroscopy, characterized in that: The following steps are involved: The spectral signal reflected from the chip to be detected is collected by a preset spectrometer; Extracting spectral features from the spectral signal to obtain extracted spectral features; Performing feature analysis on the extracted spectral features to obtain feature analysis results; If the characteristic analysis result shows that there is a chip defect characteristic in the extracted spectral characteristic, the defect area of ​​the chip to be detected is located by using a preset spectral imaging technology to obtain a location area of ​​the chip to be detected; The defect type analysis is performed on the positioning area of ​​the chip to be inspected to obtain the defect type, and the defect factor is obtained based on the defect type, and a defect resolution strategy is formulated based on the defect factor.

2. The spectral-based chip defect detection method according to claim 1, characterized in that: The collecting of the spectral signal reflected from the chip to be detected by a preset spectrometer includes: Irradiating the surface of the chip to be detected with polychromatic light emitted by a preset broadband light source; Using a preset spectrometer to collect the polychromatic light reflected from the surface of the chip to be detected, and obtain a scattered spectrum signal corresponding to the polychromatic light; Calibrate the scattered spectrum signal to obtain a calibration spectrum signal; Performing time-frequency analysis on the calibration spectrum signal to obtain an instantaneous frequency signal; The instantaneous frequency signal is subjected to multi-scale decomposition to obtain a spectrum signal.

3. The spectral-based chip defect detection method according to claim 1, characterized in that: Extracting spectral features from the spectral signal to obtain extracted spectral features includes: De-noising and smoothing the spectral signal to obtain a de-noised and smoothed spectral signal; Performing multi-level decomposition on the denoised and smoothed spectral signal to obtain a multi-level decomposed spectral signal; wherein the multi-level decomposed spectral signal includes a frequency domain component, a time domain component and a scale component; Performing feature enhancement on the multi-level decomposition spectral signal to obtain a feature enhanced spectral signal; Decoupling the feature-enhanced spectral signal by using a variational autoencoder to obtain a decoupled independent spectral feature; Performing feature fusion on the decoupled independent spectral features to obtain a feature fusion spectral signal; Spectral feature recognition and extraction are performed on the characteristic fusion spectral signal to obtain extracted spectral features.

4. The spectral-based chip defect detection method according to claim 1, characterized in that: Performing feature analysis on the extracted spectral features to obtain feature analysis results includes: Performing multi-dimensional analysis on the extracted spectral features by using a preset tensor decomposition technology to obtain a chip feature sub-tensor; Performing nonlinear geometric analysis on the chip characteristic subtensor to obtain intrinsic curvature characteristics; Performing fractional Stockwell transformation on the intrinsic curvature feature to obtain a time-frequency-scale joint distribution feature; Performing anisotropic diffusion filtering on the time-frequency-scale joint distribution features to obtain an enhanced multi-scale feature map; Perform feature extraction on the enhanced multi-scale feature map to obtain a high-order statistical feature tensor; Performing tensor sparse coding on the high-order statistical feature tensor to obtain sparse representation coefficients; Performing online dictionary learning on the sparse representation coefficients to obtain an adaptive feature dictionary; Perform feature analysis on the adaptive feature dictionary to obtain a feature analysis result.

5. The spectral-based chip defect detection method according to claim 1, characterized in that: The defect area of ​​the chip to be inspected with chip defect characteristics is located by using the preset spectral imaging technology to obtain the location area of ​​the chip to be inspected, including: Acquire a spectrum signal of the chip to be detected corresponding to the chip defect feature, and use the spectrum signal of the chip to be detected corresponding to the chip defect feature as a reconstructed spectrum signal; The reconstructed spectral signal is spatially and spectrally jointly reconstructed by spectral imaging technology to obtain a spectral image; Performing defect area detection on the spectral image to obtain a preliminary defect area image; Performing morphological processing on the preliminary defect region to obtain a morphological defect region image; Binarizing the morphological defect region image to obtain a candidate defect region mask; Extracting the skeleton structure of the candidate defect region mask to obtain a topological skeleton of the defect region; Based on a preset region growing algorithm, the topological skeleton of the defect region is located to obtain each defect boundary and the coordinates corresponding to the defect boundary; The chip positioning area to be inspected is obtained based on the defect boundaries and the coordinates corresponding to the defect boundaries.

6. The spectral-based chip defect detection method according to claim 1, characterized in that: The defect type analysis is performed on the positioning area of ​​the chip to be inspected to obtain the defect type; A defect factor is obtained based on the defect type, and a defect solution strategy is formulated based on the defect factor, including: Perform multi-scale time-frequency analysis on the positioning area of ​​the chip to be detected by continuous wavelet transform technology to obtain a time-frequency characteristic map of the defect area; Extracting texture features from the time-frequency feature map of the defect area to obtain a texture descriptor of the defect area; Performing feature unmixing and dimensionality reduction on the texture descriptor of the defect area to obtain independent feature components; Using a preset density peak clustering algorithm to perform non-parametric clustering on the independent characteristic components to obtain defect types; Performing factor analysis on the defect type using a preset partial least squares regression technique to find a linear relationship between the defect type and a characteristic variable affecting the defect type, and obtaining a factor variable affecting the defect; Using Lasso regularization technology to perform sparsity processing on the factor variables affecting the defect to obtain a defect factor; Constructing a fuzzy cognitive map based on the defect factors, and obtaining a causal relationship network corresponding to the defect factors based on the fuzzy cognitive map; A defect resolution strategy is developed based on the causal network.

7. The spectral-based chip defect detection method according to claim 6, characterized in that: The independent characteristic components are non-parametrically clustered using a preset density peak clustering algorithm to obtain defect types, including: Using a preset density peak clustering algorithm to calculate the probability density distribution of the independent feature components, and obtain a density distribution map of the multi-dimensional defect feature space; Performing fast convolution on the density distribution map of the multi-dimensional defect feature space to obtain a smoothed defect density field; Performing multi-scale analysis on the smoothed defect density field to obtain a scale space representation of the defect density field; Extracting topological features from the scale-space representation of the defect density field to obtain a persistence map of the defect density field; Analyze the persistence diagram of the defect density field to obtain a candidate defect type center and a spatial neighborhood corresponding to the candidate defect type center; wherein the spatial neighborhood is a specific spatial range affected or associated with the candidate defect type center; Performing global consistency optimization on the candidate defect type center and the spatial neighborhood corresponding to the candidate defect type center to obtain a preliminary defect type; Performing defect type verification on the preliminary defect type to obtain a defect type verification result; If the defect type verification result is within a preset defect type range, the preliminary defect type is used as the defect type; wherein the defect type includes surface defects, structural defects, material defects and process defects.

8. A chip defect detection device based on spectroscopy, characterized in that: include: A reflection module, used to collect the spectral signal reflected from the chip to be detected through a preset spectrometer; An extraction module, used for extracting spectral features from the spectral signal to obtain extracted spectral features; An analysis module, used for performing feature analysis on the extracted spectral features to obtain feature analysis results; A positioning module, for locating a defective area of ​​the chip to be detected by using a preset spectral imaging technology to obtain a positioning area of ​​the chip to be detected if the characteristic analysis result shows that there is a chip defect characteristic in the extracted spectral characteristic; The formulation module is used to analyze the defect type of the positioning area of ​​the chip to be inspected, obtain the defect type, obtain the defect factor based on the defect type, and formulate a defect resolution strategy based on the defect factor.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images