Epitaxial wafer photoluminescence test data processing method and system

Abstract

This application proposes a method and system for processing photoluminescence test data of epitaxial wafers, belonging to the field of new material testing. The method includes: performing photoluminescence testing on a target epitaxial wafer to obtain a spectral atlas and several wavelength intensity sets; processing the obtained discrete parameters of several wavelength intensities to predict the epitaxial wafer roughness and obtain a first epitaxial wafer roughness; acquiring microscopic images to identify and obtain a second epitaxial wafer roughness and obtain a test confidence level; performing epitaxial wafer quality analysis to obtain initial epitaxial wafer quality parameters; and performing confidence compensation based on the test confidence level to obtain the final epitaxial wafer quality parameters. This application solves the technical problem in existing technologies where photoluminescence test data processing of epitaxial wafers only analyzes spectral intensity and peak position, failing to accurately assess epitaxial wafer quality. It achieves the technical effect of improving the accuracy of epitaxial wafer quality assessment by predicting surface roughness through multi-wavelength intensity discreteness analysis and combining it with microscopic image verification.

Description

Technical Field

[0001] This invention relates to the field of new material testing, and in particular to a method and system for processing photoluminescence test data of epitaxial wafers. Background Technology

[0002] Epitaxial wafers are a crucial basic material in semiconductor device manufacturing, and their quality directly affects the performance and reliability of the final devices. Photoluminescence testing, as an important method for epitaxial wafer quality inspection, can analyze the optical properties and material quality of epitaxial wafers.

[0003] Currently, existing epitaxial wafer photoluminescence testing data processing focuses on analyzing spectral intensity and peak position. Specifically, the luminous efficiency of the epitaxial wafer is determined by measuring the peak intensity of the photoluminescence spectrum, and the band structure and compositional uniformity of the material are evaluated by analyzing the peak wavelength. However, this single-dimensional analytical method has significant limitations. On the one hand, the surface roughness of the epitaxial wafer significantly affects the photoluminescence test results. When a rough structure exists on the epitaxial wafer surface, both the incident laser and the emitted fluorescence are scattered, resulting in a decrease in the collected signal intensity and random variations at different test locations. On the other hand, existing technologies lack effective verification mechanisms to ensure the reliability of the test results. Relying solely on spectral data for quality assessment is prone to misjudgment and cannot comprehensively reflect the true quality status of the epitaxial wafer. Summary of the Invention

[0004] This invention addresses the technical problem in existing technologies where epitaxial wafer photoluminescence test data processing only analyzes spectral intensity and peak position, failing to accurately assess epitaxial wafer quality. It provides a method and system for processing epitaxial wafer photoluminescence test data to solve this problem.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a method for processing photoluminescence test data of an epitaxial wafer, comprising: performing photoluminescence testing on a target epitaxial wafer to obtain a spectral atlas; randomly selecting several wavelengths; extracting the intensity of several wavelengths at different positions on the epitaxial wafer within the spectral atlas to obtain several wavelength intensity sets, wherein the several wavelengths include the main peak wavelength; calculating the rate of change of the several wavelength intensity sets; processing to obtain several wavelength intensity discrete parameters; performing epitaxial wafer roughness prediction to obtain a first epitaxial wafer roughness; acquiring a microscopic image of the target epitaxial wafer; identifying and obtaining a second epitaxial wafer roughness; verifying it with the first epitaxial wafer roughness to obtain a test confidence level; performing epitaxial wafer quality analysis based on the spectral atlas, the first epitaxial wafer roughness, and the second epitaxial wafer roughness to obtain initial epitaxial wafer quality parameters; and performing confidence compensation based on the test confidence level to obtain the epitaxial wafer quality parameters.

[0006] Secondly, the present invention provides an epitaxial wafer photoluminescence test data processing system, comprising: a spectral data acquisition module, used to perform photoluminescence testing on a target epitaxial wafer, obtain a spectral atlas, randomly select several wavelengths, extract the intensity of several wavelengths at different positions on the epitaxial wafer within the spectral atlas, and obtain several wavelength intensity sets, wherein the several wavelengths include the main peak wavelength; a roughness prediction module, used to calculate the rate of change of the several wavelength intensity sets, process to obtain several wavelength intensity discrete parameters, perform epitaxial wafer roughness prediction, and obtain a first epitaxial wafer roughness; a confidence verification module, used to acquire a microscopic image of the target epitaxial wafer, identify and obtain a second epitaxial wafer roughness, verify it with the first epitaxial wafer roughness, and obtain a test confidence level; and a quality parameter output module, used to perform epitaxial wafer quality analysis based on the spectral atlas, the first epitaxial wafer roughness, and the second epitaxial wafer roughness, obtain initial epitaxial wafer quality parameters, perform confidence compensation based on the test confidence level, and obtain epitaxial wafer quality parameters.

[0007] The beneficial effects of this invention are: Photoluminescence testing is performed on the target epitaxial wafer to obtain a spectral atlas. Several wavelengths are randomly selected, and the intensities of these wavelengths at different locations on the epitaxial wafer are extracted from the spectral atlas to obtain several wavelength intensity sets. Among these, several wavelengths include the main peak wavelength, thereby obtaining photoluminescence data of the epitaxial wafer at multiple locations, providing basic data for subsequent analysis of the surface uniformity of the epitaxial wafer. The rate of change of several wavelength intensity sets is calculated, and several wavelength intensity discrete parameters are obtained. Epitaxial wafer roughness is predicted to obtain the first epitaxial wafer roughness. By analyzing the dispersion of light intensity between different locations, the influence of surface roughness on the photoluminescence signal can be quantitatively evaluated, realizing roughness prediction based on optical test data. Microscopic images of the target epitaxial wafer are acquired, and the roughness of the second epitaxial wafer is identified and verified against the roughness of the first epitaxial wafer to obtain a test confidence level. By introducing an independent microscopic image analysis method, the optical prediction results can be cross-validated, improving the reliability and accuracy of roughness assessment. Epitaxial wafer quality analysis is performed based on spectral atlases, the roughness of the first epitaxial wafer, and the roughness of the second epitaxial wafer to obtain initial epitaxial wafer quality parameters. Confidence compensation is performed based on the test confidence level to obtain the epitaxial wafer quality parameters. Thus, by integrating multi-dimensional information and adjusting compensation according to the confidence level, accurate and reliable epitaxial wafer quality assessment results are output.

[0008] The above technical solution solves the technical problem in the existing technology that the photoluminescence test data processing of epitaxial wafers only analyzes spectral intensity and peak position and cannot accurately evaluate the quality of epitaxial wafers. It achieves the technical effect of predicting surface roughness through multi-wavelength intensity dispersion analysis and verifying it with microscopic images, thereby improving the accuracy of epitaxial wafer quality evaluation. Attached Figure Description

[0009] Figure 1 A flowchart illustrating a method for processing photoluminescence test data of an epitaxial wafer provided by the present invention; Figure 2 This is a schematic diagram of the structure of an epitaxial wafer photoluminescence test data processing system provided by the present invention.

[0010] In the attached diagram, the components represented by each number are as follows: Spectral data acquisition module 11, roughness prediction module 12, confidence verification module 13, and quality parameter output module 14. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0013] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0014] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for processing epitaxial wafer photoluminescence test data, including: S1. Perform photoluminescence testing on the target epitaxial wafer to obtain a spectral set. Randomly select several wavelengths and extract the intensity of several wavelengths at different positions on the epitaxial wafer from the spectral set to obtain several wavelength intensity sets, including the main peak wavelength.

[0015] Specifically, the target epitaxial wafer refers to the semiconductor epitaxial wafer to be tested, such as GaN-based LED epitaxial wafers, AlGaN-based ultraviolet LED epitaxial wafers, InGaN-based visible light LED epitaxial wafers, and other compound semiconductor epitaxial wafers. First, photoluminescence testing is performed on the target epitaxial wafer to obtain basic spectral data. Photoluminescence testing is a non-destructive optical characterization technique. By irradiating the surface of the target epitaxial wafer with an excitation light source, the charge carriers in the target epitaxial wafer are excited. When the excited-state charge carriers recombine, they emit fluorescence of characteristic wavelengths. By detecting and analyzing these fluorescence signals with a spectrometer, spectral information reflecting the optical properties and structural characteristics of the target epitaxial wafer can be obtained, resulting in a spectral atlas. In the specific testing process, multiple different locations on the target epitaxial wafer are scanned point-by-point, with the excitation beam sequentially irradiating each preset location on the surface of the target epitaxial wafer. A complete photoluminescence spectrum is obtained at each location. These spectra from different spatial locations together constitute a spectral atlas. This atlas not only contains spectral intensity information for each location but also retains the corresponding spatial coordinate information, thereby establishing the spatial distribution characteristics of the optical properties of the target epitaxial wafer.

[0016] Next, several wavelengths are selected from the spectral atlas for focused analysis. The wavelength of the main peak is essential, as it corresponds to the primary emission peak of the epitaxial wafer material and directly reflects its bandgap characteristics and fundamental optical properties. For example, for GaN-based blue LED epitaxial wafers, the main peak wavelength is typically around 450 nm. In addition to the main peak wavelength, several wavelengths near the main peak, such as 445 nm and 455 nm, should also be selected. This selection strategy provides more comprehensive spectral characteristic information, and multi-wavelength analysis improves the reliability of the data.

[0017] Subsequently, for several selected wavelengths, the intensity values ​​of these wavelengths at all test locations on the epitaxial wafer are extracted from the spectral atlas, thus obtaining several wavelength intensity sets. Each wavelength intensity set contains the intensity data sequence of that specific wavelength at different spatial locations, reflecting the uniformity distribution characteristics of the emission of the target epitaxial wafer at that wavelength. Theoretically, if the surface of the target epitaxial wafer is smooth and the material properties are uniform, the emission intensity of the same wavelength at different locations should be relatively stable; however, if there is a roughness problem on the surface of the target epitaxial wafer, it will cause a light scattering effect, resulting in random variations in the intensity of the same wavelength at different locations, providing a data basis for subsequent prediction of surface roughness through intensity variation characteristics.

[0018] S2. Calculate the rate of change of the plurality of wavelength intensity sets, process to obtain a plurality of wavelength intensity discrete parameters, perform epitaxial wafer roughness prediction, and obtain the first epitaxial wafer roughness.

[0019] Specifically, based on several wavelength intensity sets obtained, the spatial variation characteristics of each wavelength intensity are calculated through statistical analysis methods, thereby achieving quantitative prediction of the surface roughness of the target epitaxial wafer.

[0020] First, the rate of change of several wavelength intensity sets is calculated. Specifically, according to a preset spatial order on the target epitaxial wafer (e.g., scanning order from left to right, top to bottom), the rate of change of intensity data between adjacent positions within each wavelength intensity set is calculated. For any wavelength λ, the intensity change amplitude between adjacent test positions is calculated, i.e., the rate of change R = |I(i+1) - I(i)| / I(i), where I(i) represents the intensity value at the i-th position, and I(i+1) represents the intensity value at the next adjacent position. By performing this calculation on all adjacent positions, the set of wavelength intensity rate of change for that wavelength can be obtained. This process is repeated for several selected wavelengths to ultimately obtain several sets of wavelength intensity rate of change.

[0021] Next, statistical analysis is performed on the obtained sets of wavelength intensity change rates to calculate the corresponding rate of change dispersion parameters. Rate of change dispersion parameters are statistical indicators reflecting the degree of data dispersion. For example, the standard deviation of multiple wavelength intensity change rates within each wavelength intensity change rate set is calculated as the rate of change dispersion parameter for that set. Assuming the wavelength intensity change rate set for a certain wavelength of 450nm is {0.05, 0.12, 0.08, 0.15, 0.06, 0.11, 0.09, 0.13, 0.07, 0.10}, the calculated standard deviation σ = 0.033 is the rate of change dispersion parameter for that wavelength. By calculating the rate of change dispersion parameters for each selected wavelength, several wavelength intensity dispersion parameters are finally obtained.

[0022] Subsequently, based on the obtained discrete wavelength intensity parameters, a pre-trained epitaxial wafer roughness prediction network is used to predict the roughness of the epitaxial wafer, obtaining the first epitaxial wafer roughness. This pre-trained epitaxial wafer roughness prediction network establishes a mapping relationship between several discrete wavelength intensity parameters and surface roughness based on a large amount of historical sample data. The discrete wavelength intensity parameters of the current target epitaxial wafer are used as input feature vectors and fed into the pre-trained epitaxial wafer roughness prediction network. Through multi-layer nonlinear transformation processing, the corresponding first epitaxial wafer roughness is output.

[0023] By using a roughness prediction method based on the statistical characteristics of spectral data, it is possible to evaluate the surface morphology, providing support for the quality analysis of epitaxial wafers.

[0024] S3. Acquire a microscopic image of the target epitaxial wafer, identify and obtain the roughness of the second epitaxial wafer, verify it with the roughness of the first epitaxial wafer, and obtain the test confidence level.

[0025] Specifically, in order to verify the accuracy of the predicted roughness of the first epitaxial wafer, the roughness of the second epitaxial wafer was obtained by independent microscopic imaging, and the test confidence level was established through comparative analysis.

[0026] First, microscopic images of the target epitaxial wafer are acquired. High-resolution imaging of the target epitaxial wafer surface is performed using microscopy systems such as atomic force microscopy (AFM), scanning electron microscopy (SEM), or optical microscopy, obtaining microscopic images that clearly reflect the surface morphology. These microscopic images directly reveal the three-dimensional morphology information of the target epitaxial wafer surface, including microstructural features such as surface undulations, particle distribution, and defect morphology. This information provides a reliable data source for the direct measurement of surface roughness.

[0027] Next, a pre-trained epitaxial wafer roughness recognition network is used to analyze and process the acquired microscopic images. This network, built on a convolutional neural network, can automatically identify surface morphology features in the microscopic images and output the corresponding epitaxial wafer roughness. The network is trained through supervised learning using a large number of sample microscopic images and corresponding epitaxial wafer roughness values, establishing a mapping relationship from image features to roughness parameters. By inputting the microscopic image of the current target epitaxial wafer into this network, the epitaxial wafer roughness determined based on the microscopic image can be obtained, serving as the second epitaxial wafer roughness.

[0028] Then, the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer are compared and verified to obtain the test confidence level. The reliability of the roughness of the first epitaxial wafer is evaluated by calculating the similarity between the two roughnesses. The closer the roughness of the first and second epitaxial wafers are, the higher the accuracy of the roughness predicted by the photoluminescence test data, and the greater the corresponding test confidence level. Conversely, when the roughness of the first and second epitaxial wafers differs significantly, it indicates a certain deviation in the prediction, and the test confidence level is relatively low. This test confidence level will serve as an important parameter in subsequent epitaxial wafer quality analysis for quality assessment.

[0029] S4. Perform epitaxial wafer quality analysis based on the spectral atlas, the roughness of the first epitaxial wafer, and the roughness of the second epitaxial wafer to obtain initial epitaxial wafer quality parameters. Perform confidence compensation based on the test confidence level to obtain the epitaxial wafer quality parameters.

[0030] Specifically, firstly, the roughness of the first and second epitaxial wafers is weighted and fused based on the test confidence level. Specifically, the test confidence level is used as the first weight, and 1 is subtracted from the first weight to obtain the second weight. Then, a weighted average is calculated using the first and second weights to obtain the final epitaxial wafer roughness. When the test confidence level is high, it indicates that the predicted roughness of the first epitaxial wafer has high reliability, and therefore it is given a larger weight; when the test confidence level is low, more reliance is placed on the roughness of the second epitaxial wafer obtained from the microscopic image. This adaptive weighting strategy fully leverages the advantages of both epitaxial wafer roughness values, resulting in a more accurate and reliable roughness assessment.

[0031] Next, quality parameter analysis is performed based on the spectral atlas and the fused epitaxial wafer roughness. According to the spectral characteristic information in the spectral atlas, spectral analysis is used to obtain the spectral epitaxial wafer quality parameters. Specifically, by analyzing the spectral data at each test location in the spectral atlas, spectral characteristic indicators are extracted, including luminescence intensity uniformity, peak position stability, and full width at half maximum (FWHM) variation. Luminescence intensity uniformity is quantified by calculating the ratio of the standard deviation to the average value of the main peak intensity at different locations; a smaller ratio indicates more uniform luminescence of the epitaxial wafer. Peak position stability is assessed by statistically analyzing the deviation of the main peak wavelength at each test location; a smaller deviation indicates more uniform material composition. FWHM variation reflects the material crystallinity quality by analyzing the consistency of the spectral peak shape; a more stable FWHM indicates better crystal quality. These spectral characteristic indicators are then comprehensively processed using a pre-defined scoring model to obtain the spectral epitaxial wafer quality parameters, represented as a standardized quality score.

[0032] Simultaneously, based on the fused epitaxial wafer roughness, a pre-constructed roughness quality assessment model is used to convert the epitaxial wafer roughness into a corresponding quality score, obtaining the roughness spectrum epitaxial wafer quality parameters. This roughness quality assessment model establishes a mapping relationship between the surface roughness of the epitaxial wafer and the quality score; a higher score is output when the roughness is low, and a lower score is output when the roughness is high. This roughness quality assessment model is obtained through supervised learning training by collecting roughness data and corresponding quality score labels from a large number of historical epitaxial wafer samples. During training, epitaxial wafer samples with known roughness values ​​are used as input, and quality scores obtained from expert evaluation or standard testing are used as labels. A nonlinear mapping function from roughness to quality score is established using regression algorithms or neural networks to obtain the roughness quality assessment model. After the roughness quality assessment model is trained, it can automatically predict the corresponding quality score based on new epitaxial wafer roughness input.

[0033] Subsequently, the mean values ​​of the quality parameters of the spectral epitaxial wafer and the rough spectral epitaxial wafer were calculated to obtain the average quality score of the epitaxial wafer quality parameters obtained through the two paradigms, which served as the initial epitaxial wafer quality parameters. These initial epitaxial wafer quality parameters comprehensively considered information from both spectral characteristics and surface morphology, providing a more comprehensive reflection of the overall quality of the epitaxial wafer. Then, a confidence compensation coefficient was configured based on the test confidence level to compensate for the initial epitaxial wafer quality parameters. When the test confidence level was high, indicating strong reliability of the test process, the confidence compensation coefficient was close to 1, allowing for minor adjustments to the initial epitaxial wafer quality parameters. When the test confidence level was low, indicating some uncertainty in the test, the confidence compensation coefficient was correspondingly reduced, correcting the initial epitaxial wafer quality parameters and lowering the quality score of the target epitaxial wafer. The confidence-corrected epitaxial wafer quality parameters were obtained by multiplying the initial epitaxial wafer quality parameters by the confidence compensation coefficient. The epitaxial wafer quality parameters not only integrate multi-dimensional test information but also take into account the reliability of the test process, providing more accurate, reliable, and credible quantitative results for epitaxial wafer quality assessment and effectively improving the accuracy of quality inspection.

[0034] By combining photoluminescence spectral analysis with microscopic image recognition, the surface roughness of epitaxial wafers can be predicted by analyzing the variation characteristics of photoluminescence intensity at different spatial locations. Furthermore, test confidence levels and confidence compensation are established, which effectively improves the accuracy and reliability of quality assessment and provides strong technical support for the quality control and process optimization of semiconductor epitaxial wafers.

[0035] Furthermore, photoluminescence testing is performed on the target epitaxial wafer to obtain a spectral atlas. Several wavelengths are randomly selected, and the intensities of several wavelengths at different locations on the epitaxial wafer are extracted from the spectral atlas to obtain several wavelength intensity sets, including: S11. Perform photoluminescence testing on the target epitaxial wafer to obtain a spectral atlas, wherein the spectral atlas includes spectral maps of multiple locations on the target epitaxial wafer; S12. Randomly select several wavelengths, including the main peak wavelength and several wavelengths adjacent to the main peak wavelength. S13. Based on several wavelengths, extract the intensity of several wavelengths at different positions on the epitaxial wafer from the spectral atlas to obtain several wavelength intensity sets.

[0036] In a preferred embodiment, firstly, photoluminescence testing is performed on the target epitaxial wafer to obtain a spectral atlas. Specifically, multiple locations on the target epitaxial wafer are scanned point-by-point, with an excitation beam sequentially illuminating each location on the surface of the target epitaxial wafer. A complete photoluminescence spectrum is obtained at each location. The spectral atlases from multiple locations on the target epitaxial wafer form a spectral atlas, containing spectral intensity information for each location and retaining the corresponding spatial coordinate information, thereby establishing the spatial distribution characteristics of the optical properties of the target epitaxial wafer.

[0037] Then, several wavelengths were randomly selected from the spectral range corresponding to the spectral atlas for focused analysis. These wavelengths included the main peak wavelength and several wavelengths adjacent to it. The main peak wavelength corresponds to the primary emission peak of the epitaxial wafer material, directly reflecting the material's bandgap characteristics and basic optical properties. The wavelengths adjacent to the main peak wavelength are usually located on either side of the main peak, providing supplementary information about the main peak's characteristics. Combining multiple wavelengths for analysis can improve the reliability and stability of the data.

[0038] Next, based on several determined wavelengths, the intensities of several wavelengths at different locations on the epitaxial wafer are extracted from the spectral atlas to obtain several wavelength intensity sets. Specifically, for each selected wavelength, the spectral data of all locations in the spectral atlas are traversed, and the intensity value at that wavelength is accurately obtained through interpolation or fitting methods, ultimately forming a wavelength intensity set for that wavelength. By performing this process on all selected wavelengths, several wavelength intensity sets are obtained. Each wavelength intensity set contains the intensity data sequence of that specific wavelength at different spatial locations, reflecting the luminescence uniformity distribution characteristics of the target epitaxial wafer at that wavelength.

[0039] Furthermore, the rate of change of the plurality of wavelength intensity sets is calculated, and several wavelength intensity discrete parameters are obtained through processing. Epitaxial wafer roughness prediction is then performed to obtain the first epitaxial wafer roughness, including: S21. According to the preset spatial order on the target epitaxial wafer, calculate the change amplitude of each wavelength intensity and the neighboring wavelength intensity of the next position point in several wavelength intensity sets to obtain several wavelength intensity change rate sets. S22. Calculate the rate of change discrete parameters of several wavelength intensity change rate sets to obtain several wavelength intensity discrete parameters. S23. Based on several wavelength intensity discrete parameters, predict the roughness of the epitaxial wafer to obtain the roughness of the first epitaxial wafer.

[0040] In a preferred embodiment, according to a preset spatial order on the target epitaxial wafer, the variation amplitude of the intensity of each wavelength in a plurality of wavelength intensity sets and the intensity of the neighboring wavelength at the next position point is calculated to obtain a plurality of wavelength intensity variation rate sets. For example, according to a preset scanning order (such as a spatial order from left to right or from top to bottom) during photoluminescence testing, the variation rate between adjacent position points is calculated for the intensity data in each wavelength intensity set. For any wavelength λ, the intensity variation amplitude between adjacent position points is calculated, i.e., the variation rate R = |I(i+1) - I(i)| / I(i), where I(i) represents the intensity value of the i-th position point under the wavelength, and I(i+1) represents the intensity value of the next adjacent position point under the wavelength. By performing such calculations on all adjacent position points for the wavelength, a wavelength intensity variation rate set for the wavelength is obtained. The above processing is performed on a plurality of selected wavelengths respectively to finally obtain a plurality of wavelength intensity variation rate sets.

[0041] Subsequently, statistical analysis is performed on the obtained sets of wavelength intensity change rates to calculate the rate of change dispersion parameters, thus obtaining several wavelength intensity dispersion parameters. Specifically, a statistical index of the dispersion degree of the rate of change data in each wavelength intensity change rate set is calculated. For example, the standard deviation of the data in each wavelength intensity change rate set is calculated as the dispersion parameter for that wavelength. In addition to the standard deviation, other statistical indices can also be used as dispersion parameters, such as the coefficient of variation. By calculating the ratio of the standard deviation to the mean, the influence of differences in intensity magnitudes across different wavelengths is eliminated, and the coefficient of variation is obtained as the wavelength intensity change rate dispersion parameter; alternatively, the range, i.e., the difference between the maximum and minimum values ​​in the wavelength intensity change rate set, can be used to reflect the dispersion range of the data as the wavelength intensity change rate dispersion parameter. By calculating the rate of change dispersion parameters for each selected wavelength, several wavelength intensity dispersion parameters are finally obtained. These wavelength intensity dispersion parameters reflect the randomness and non-uniformity of intensity changes at each wavelength. The larger the dispersion parameter, the more random the intensity change, which usually corresponds to higher surface roughness. It should be noted that although there are multiple statistical indicators available, one of them should be selected as the unified calculation standard during the specific testing process to ensure the comparability of the discrete parameters of the rate of change of each wavelength. At the same time, the same statistical indicator should also be used to calculate the discrete parameters during the training process of the epitaxial wafer roughness prediction network to ensure the consistency between the training data and the actual application.

[0042] Subsequently, based on several discrete wavelength intensity parameters, a pre-trained epitaxial wafer roughness prediction network is used to predict the roughness of the epitaxial wafer, obtaining the first epitaxial wafer roughness. Specifically, several discrete wavelength intensity parameters of the current target epitaxial wafer are used as input feature vectors and input into the pre-trained epitaxial wafer roughness prediction network. This epitaxial wafer roughness prediction network is built based on machine learning, and is trained through supervised learning using a large number of historical samples of discrete wavelength intensity parameters and corresponding epitaxial wafer roughness data, establishing a mapping relationship between discrete parameters and surface roughness.

[0043] By obtaining the roughness of the first epitaxial wafer, indirect prediction of surface roughness based on photoluminescence spectral data was achieved, providing important parameters for epitaxial wafer quality analysis. This method avoids the potential damage to the epitaxial wafer surface caused by direct contact measurement, while fully utilizing the intensity spatial distribution information during photoluminescence testing, laying the foundation for subsequent multi-dimensional quality assessment.

[0044] Furthermore, based on several wavelength intensity discrete parameters, the roughness of the epitaxial wafer is predicted to obtain the first epitaxial wafer roughness, including: S231. Obtain a pre-trained epitaxial wafer roughness prediction network, wherein the epitaxial wafer roughness prediction network is constructed based on machine learning, and is obtained by supervising training using several sample wavelength intensity discrete parameter sets and sample epitaxial wafer roughness sets as training data, and the training data is collected from epitaxial wafer test data over a historical period. S232. Input the plurality of wavelength intensity discrete parameters into the epitaxial wafer roughness prediction network and predict and output the first epitaxial wafer roughness.

[0045] In a preferred embodiment, firstly, epitaxial wafer test data from a historical period is collected, and a large number of epitaxial wafer samples of different types and qualities are selected as training samples. For each epitaxial wafer sample, photoluminescence testing and surface roughness standard measurement are performed to obtain complete training data. For photoluminescence testing, each epitaxial wafer sample is processed using the same method as the current test, including selecting several wavelengths, extracting the intensity values ​​of each wavelength at different locations, calculating the wavelength intensity change rate set, and using a unified statistical index to calculate the wavelength intensity discrete parameters for each wavelength. By performing this processing on all training samples, several sets of sample wavelength intensity discrete parameters are formed, with each set of several sample wavelength intensity discrete parameters corresponding to a multi-wavelength discrete parameter feature vector for one epitaxial wafer sample. For surface roughness standard measurement, a high-precision surface roughness measuring device is used to accurately measure the same epitaxial wafer samples to obtain the true roughness value of each sample. These true roughness values ​​are then uniformly standardized to form a sample epitaxial wafer roughness set, which serves as label data for supervised learning.

[0046] Next, an epitaxial wafer roughness prediction network was constructed based on the prepared training data. This network was built using a machine learning algorithm and a deep neural network architecture. It used a set of discrete parameters representing several sample wavelength intensities as input features and a set of sample epitaxial wafer roughness as supervised labels for training. During supervised training, the internal weights and bias parameters were continuously adjusted using the backpropagation algorithm to gradually learn and optimize the nonlinear mapping relationship between the input features and the output roughness, ultimately obtaining an epitaxial wafer roughness prediction network capable of accurately predicting the surface roughness of epitaxial wafers.

[0047] Subsequently, several discrete wavelength intensity parameters of the current target epitaxial wafer are input into the epitaxial wafer roughness prediction network to predict and output the first epitaxial wafer roughness. Specifically, the obtained discrete wavelength intensity parameters are used to form a feature vector, which serves as the input to the epitaxial wafer roughness prediction network. After receiving the input, the epitaxial wafer roughness prediction network performs forward propagation calculations through its internal multi-layer neuron structure and activation functions. After undergoing multi-layer nonlinear transformation processing, the corresponding roughness prediction value, i.e., the first epitaxial wafer roughness, is finally generated at the output layer.

[0048] Furthermore, a microscopic image of the target epitaxial wafer is acquired, the roughness of the second epitaxial wafer is identified and verified against the roughness of the first epitaxial wafer, and the test confidence level is obtained, including: S31. Acquire a microscopic image of the target epitaxial wafer; S32. Obtain an epitaxial wafer roughness recognition network, wherein the epitaxial wafer roughness recognition network is constructed using a convolutional neural network and obtained by supervised training using a set of sample microscopic images and a set of sample epitaxial wafer roughness as training data. S33. Input the microscopic image into the epitaxial wafer roughness recognition network, and recognize and output the second epitaxial wafer roughness; S34. Verify the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer to obtain the test confidence level.

[0049] In a preferred embodiment, firstly, a high-resolution microscopic imaging device is used to image and acquire a microscopic image of the target epitaxial wafer surface. Suitable high-resolution microscopic imaging devices include atomic force microscopes, scanning electron microscopes, or optical microscopes. During imaging, it is ensured that the microscopic image clearly reflects the morphological features of the target epitaxial wafer surface, including surface undulations, particle distribution, defect morphology, and other microstructural information. The acquired microscopic image should have sufficient resolution and contrast to provide a reliable data foundation for subsequent image recognition and analysis.

[0050] Subsequently, an epitaxial wafer roughness recognition network was developed. Specifically, firstly, a large number of epitaxial wafer samples were acquired along with their corresponding roughness measurement data. For each epitaxial wafer sample, its microscopic image was acquired under the same imaging conditions, and the true roughness value of the sample was obtained using a roughness measurement device. By performing this process on a large number of samples, a set of sample microscopic images and a set of sample epitaxial wafer roughness were formed. A convolutional neural network was constructed, which has a multi-layer structure of convolutional layers, pooling layers, and fully connected layers. During network training, the set of sample microscopic images was used as input, and the set of sample epitaxial wafer roughness was used as labels for supervised learning. By continuously optimizing the parameters of the convolutional neural network, a mapping relationship from image features to roughness values ​​was established, resulting in the epitaxial wafer roughness recognition network.

[0051] Next, the microscopic image of the target epitaxial wafer is input into the epitaxial wafer roughness recognition network, which identifies and outputs the second epitaxial wafer roughness. Specifically, the acquired microscopic image is input into the pre-trained epitaxial wafer roughness recognition network. The epitaxial wafer roughness recognition network automatically extracts local features and global topography information of the image through its convolutional layer structure. After multi-layer feature transformation and nonlinear processing, it finally generates the corresponding roughness prediction value, i.e., the second epitaxial wafer roughness, in the output layer.

[0052] Next, the roughness of the first and second epitaxial wafers was verified to obtain the test confidence level. Specifically, the reliability of the spectroscopic prediction results was evaluated by calculating the similarity between the roughness values ​​of the first and second epitaxial wafers. The closer the two roughness values ​​are, the higher the accuracy of the first epitaxial wafer roughness prediction based on photoluminescence spectroscopy data, and the greater the corresponding test confidence level. Conversely, a large difference between the two roughness values ​​indicates a certain deviation in the spectroscopic prediction, and the test confidence level is relatively low. This test confidence level will be used as a weighting parameter in subsequent epitaxial wafer quality analysis.

[0053] Furthermore, the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer are verified to obtain the test confidence level, including: S341. Calculate the similarity between the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer, and use it as the test confidence level.

[0054] In a preferred embodiment, the test confidence level is obtained by calculating the reliability of the spectral prediction results based on the similarity measure of the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer. For example, the relative error between the two roughness values ​​is calculated as the similarity between the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer, i.e., similarity = 1 - |roughness of the first epitaxial wafer - roughness of the second epitaxial wafer| / max(roughness of the first epitaxial wafer, roughness of the second epitaxial wafer).

[0055] Assuming the roughness of the first epitaxial wafer is 0.8 nm and the roughness of the second epitaxial wafer is 0.85 nm, then the similarity = 1 - |0.8 - 0.85| / max(0.8, 0.85) = 1 - 0.05 / 0.85 ≈ 0.94, corresponding to a test confidence level of 0.94. If the roughness of the first epitaxial wafer is 0.6 nm and the roughness of the second epitaxial wafer is 1.2 nm, then the similarity = 1 - |0.6 - 1.2| / max(0.6, 1.2) = 1 - 0.6 / 1.2 = 0.5, corresponding to a test confidence level of 0.5.

[0056] By calculating the similarity between the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer, the test confidence level is obtained, providing a reference for subsequent quality analysis. The closer the two roughness values ​​are, the higher the similarity between the constructed vectors, indicating better accuracy of the spectral method prediction and a higher corresponding test confidence level; when the two roughness values ​​differ significantly, the vector similarity is low, and the test confidence level decreases accordingly.

[0057] Furthermore, epitaxial wafer quality analysis is performed based on the spectral atlas, the roughness of the first epitaxial wafer, and the roughness of the second epitaxial wafer to obtain initial epitaxial wafer quality parameters. Confidence compensation is then performed based on the test confidence level to obtain further epitaxial wafer quality parameters, including: S41. Based on the test confidence level, configure the first weight and the second weight, and perform a weighted calculation on the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer to obtain the roughness of the epitaxial wafer. S42. Based on the spectral atlas, identify and obtain the quality parameters of the spectral epitaxial wafer; based on the roughness of the epitaxial wafer, classify and obtain the quality parameters of the rough spectral epitaxial wafer. S43. Based on the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters, process to obtain initial epitaxial wafer quality parameters, combine with the test confidence level for confidence compensation, and process to obtain epitaxial wafer quality parameters.

[0058] In a preferred embodiment, a first weight and a second weight are configured based on the test confidence level, and a weighted calculation is performed on the roughness of the first and second epitaxial wafers to obtain the epitaxial wafer roughness. Specifically, the test confidence level is directly used as the first weight, and the second weight is obtained by subtracting the test confidence level from 1, i.e., first weight = test confidence level, second weight = 1 - test confidence level. The final epitaxial wafer roughness is calculated by a weighted average method, i.e., epitaxial wafer roughness = first weight × first epitaxial wafer roughness + second weight × second epitaxial wafer roughness. When the test confidence level is high, it indicates that the first epitaxial wafer roughness predicted by the spectroscopic method has high reliability, so it is given a larger weight in the weighted calculation; when the test confidence level is low, more reliance is placed on the second epitaxial wafer roughness obtained by the microscopic imaging method. Through this adaptive weighting strategy, the advantages of the two testing methods can be fully utilized to obtain more accurate and reliable roughness evaluation results.

[0059] Then, quality parameter analysis is performed based on the spectral atlas and the weighted fused epitaxial wafer roughness. First, the spectral epitaxial wafer quality parameters are identified and obtained from the spectral atlas. By analyzing the spectral data at various points in the spectral atlas, spectral feature indicators are extracted, including key parameters such as luminescence intensity uniformity, peak stability, and full width at half maximum (FWHM) variation. These spectral feature indicators are then comprehensively processed using a pre-defined scoring model to obtain standardized spectral epitaxial wafer quality parameters. The pre-defined scoring model can be a machine learning-based scoring model or a data mapping table based on historical scoring data. For the machine learning-based scoring model, by collecting a large number of historical epitaxial wafer samples' spectral feature indicators and corresponding expert-assessed quality scores, supervised learning methods are used to train and construct multi-layer neural networks or support vector machines, establishing a non-linear mapping relationship from multi-dimensional spectral feature vectors to quality scores, thus obtaining the pre-defined scoring model. For the data mapping table based on historical scoring data, based on accumulated experience, a correspondence table between the numerical range of spectral feature indicators and quality scores is established. The spectral features of the current epitaxial wafer are converted into standardized quality scores through table lookup or linear interpolation, serving as the pre-defined scoring model. Simultaneously, based on the obtained epitaxial wafer roughness, a pre-constructed roughness quality assessment model is used for classification processing to obtain roughness spectrum epitaxial wafer quality parameters. This roughness quality assessment model establishes a mapping relationship between epitaxial wafer roughness and quality score, and can convert roughness values ​​into corresponding quality scores.

[0060] Subsequently, based on the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters, initial epitaxial wafer quality parameters are obtained. These are then combined with the test confidence level for confidence compensation to obtain the final epitaxial wafer quality parameters. Specifically, firstly, the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters are fused and calculated using a weighted average or linear combination method to obtain the initial epitaxial wafer quality parameters. These initial quality parameters comprehensively consider information from both spectral characteristics and surface morphology. Next, a confidence compensation coefficient is configured based on the test confidence level to compensate for the initial epitaxial wafer quality parameters. When the test confidence level is high, the confidence compensation coefficient is close to 1, resulting in a small adjustment to the initial epitaxial wafer quality parameters; when the test confidence level is low, the confidence compensation coefficient is correspondingly reduced, resulting in a conservative correction to the initial epitaxial wafer quality parameters. By multiplying the initial epitaxial wafer quality parameters by the confidence compensation coefficient, the confidence-corrected epitaxial wafer quality parameters are obtained.

[0061] Furthermore, based on the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters, initial epitaxial wafer quality parameters are obtained through processing. Confidence compensation is then performed using the test confidence level to obtain further epitaxial wafer quality parameters, including: S431. Calculate the initial epitaxial wafer quality parameters based on the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters; S432. Configure the confidence compensation coefficient based on the test confidence level; S433. Using the confidence compensation coefficient, the initial epitaxial wafer quality parameters are compensated and calculated to obtain the epitaxial wafer quality parameters.

[0062] In a preferred embodiment, an initial epitaxial wafer quality parameter is calculated based on the spectral epitaxial wafer quality parameter and the rough spectral epitaxial wafer quality parameter. Since both the spectral epitaxial wafer quality parameter and the rough spectral epitaxial wafer quality parameter are expressed in a standardized quality score format, the two quality scores are fused using an arithmetic mean: Initial epitaxial wafer quality parameter = (spectral epitaxial wafer quality parameter + rough spectral epitaxial wafer quality parameter) / 2. Through this fusion method, the initial epitaxial wafer quality parameter can comprehensively reflect quality information from both spectral characteristics and surface morphology dimensions, providing basic data for subsequent confidence compensation.

[0063] Then, the confidence compensation coefficient is configured based on the test confidence level. Specifically, a mapping relationship is established between the test confidence level and the confidence compensation coefficient. When the test confidence level is high, it indicates that the test process is highly reliable and the data quality is good. In this case, the configured confidence compensation coefficient is close to 1, allowing for minor adjustments to the initial epitaxial wafer quality parameters. When the test confidence level is low, it indicates that there is some uncertainty in the test. In this case, a smaller confidence compensation coefficient is configured to achieve a conservative correction to the initial epitaxial wafer quality parameters. The configuration of the confidence compensation coefficient can use linear mapping, piecewise functions, or other mathematical functions to establish the correspondence between the test confidence level and the compensation coefficient.

[0064] Subsequently, a confidence compensation coefficient is used to compensate for the initial epitaxial wafer quality parameters, thus obtaining the epitaxial wafer quality parameters. Specifically, by multiplying the initial epitaxial wafer quality parameters by the confidence compensation coefficient, quality parameter correction based on test reliability is achieved, i.e., epitaxial wafer quality parameter = initial epitaxial wafer quality parameter × confidence compensation coefficient. The epitaxial wafer quality parameters after compensation calculation not only integrate multi-dimensional test information but also fully consider the reliability of the testing process, providing more accurate, reliable, and credible quantitative results for epitaxial wafer quality evaluation.

[0065] Example 2, as Figure 2 As shown, based on the same inventive concept as the epitaxial wafer photoluminescence test data processing method provided in Embodiment 1, this embodiment of the invention also provides an epitaxial wafer photoluminescence test data processing system, including: The spectral data acquisition module 11 is used to perform photoluminescence testing on the target epitaxial wafer, obtain a spectral atlas, randomly select several wavelengths, extract the intensity of several wavelengths at different positions on the epitaxial wafer within the spectral atlas, and obtain several wavelength intensity sets, wherein several wavelengths include the main peak wavelength. Roughness prediction module 12 is used to calculate the rate of change of the plurality of wavelength intensity sets, process to obtain a plurality of wavelength intensity discrete parameters, perform epitaxial wafer roughness prediction, and obtain the first epitaxial wafer roughness. The confidence verification module 13 is used to acquire a microscopic image of the target epitaxial wafer, identify and obtain the roughness of the second epitaxial wafer, verify it with the roughness of the first epitaxial wafer, and obtain the test confidence. The quality parameter output module 14 is used to perform epitaxial wafer quality analysis based on the spectral atlas, the roughness of the first epitaxial wafer, and the roughness of the second epitaxial wafer to obtain initial epitaxial wafer quality parameters, and to perform confidence compensation based on the test confidence level to obtain the epitaxial wafer quality parameters.

[0066] Furthermore, the execution steps of the spectral data acquisition module 11 include: Photoluminescence testing is performed on the target epitaxial wafer to obtain a spectral atlas, wherein the spectral atlas includes spectral maps of multiple locations on the target epitaxial wafer; Several wavelengths are randomly selected, including the main peak wavelength and several wavelengths adjacent to the main peak wavelength. Based on several wavelengths, the intensities of several wavelengths at different locations on the epitaxial wafer are extracted from the spectral atlas to obtain several wavelength intensity sets.

[0067] Furthermore, the execution steps of the roughness prediction module 12 include: According to the preset spatial order on the target epitaxial wafer, calculate the variation amplitude of each wavelength intensity and the neighboring wavelength intensity of the next position point in several wavelength intensity sets to obtain several wavelength intensity variation rate sets; Calculate the rate of change discrete parameters of several wavelength intensity change rate sets to obtain several wavelength intensity discrete parameters; Based on several discrete parameters of wavelength intensity, the roughness of the epitaxial wafer is predicted to obtain the roughness of the first epitaxial wafer.

[0068] Furthermore, the execution steps of the roughness prediction module 12 also include: A pre-trained epitaxial wafer roughness prediction network is obtained, wherein the epitaxial wafer roughness prediction network is constructed based on machine learning, and is obtained by supervising training using several sample wavelength intensity discrete parameter sets and sample epitaxial wafer roughness sets as training data. The training data is collected from epitaxial wafer test data over a historical period. The discrete parameters of several wavelength intensities are input into the epitaxial wafer roughness prediction network to predict and output the first epitaxial wafer roughness.

[0069] Furthermore, the execution steps of the confidence verification module 13 include: Acquire microscopic images of the target epitaxial wafer; An epitaxial wafer roughness recognition network is obtained, wherein the epitaxial wafer roughness recognition network is constructed using a convolutional neural network and obtained by supervised training using a set of sample microscopic images and a set of sample epitaxial wafer roughness as training data; The microscopic image is input into the epitaxial wafer roughness recognition network, and the second epitaxial wafer roughness is recognized and output. The roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer were verified to obtain the test confidence level.

[0070] Furthermore, the execution steps of the confidence verification module 13 also include: The similarity between the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer is calculated as the test confidence level.

[0071] Furthermore, the execution steps of the quality parameter output module 14 include: Based on the test confidence level, a first weight and a second weight are configured, and the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer are weighted and calculated to obtain the roughness of the epitaxial wafer. Based on the spectral atlas, the quality parameters of the spectral epitaxial wafer are identified and obtained; based on the roughness of the epitaxial wafer, the quality parameters of the rough spectral epitaxial wafer are classified and obtained. Based on the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters, initial epitaxial wafer quality parameters are obtained through processing. Confidence compensation is then performed using the test confidence level to obtain the final epitaxial wafer quality parameters.

[0072] Furthermore, the execution steps of the quality parameter output module 14 also include: The initial epitaxial wafer quality parameters are calculated based on the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters. Configure the confidence compensation coefficient based on the test confidence level; The confidence compensation coefficient is used to compensate for the initial epitaxial wafer quality parameters to obtain the epitaxial wafer quality parameters.

[0073] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0074] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0076] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0077] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0078] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0079] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for processing photoluminescence test data of epitaxial wafers, characterized in that, The method includes: Photoluminescence testing is performed on the target epitaxial wafer to obtain a spectral atlas. Several wavelengths are randomly selected, and the intensities of several wavelengths at different positions on the epitaxial wafer are extracted from the spectral atlas to obtain several wavelength intensity sets, including the main peak wavelength. Calculate the rate of change of the plurality of wavelength intensity sets, process to obtain a plurality of wavelength intensity discrete parameters, perform epitaxial wafer roughness prediction, and obtain the first epitaxial wafer roughness; Microscopic images of the target epitaxial wafer are acquired, the roughness of the second epitaxial wafer is identified and verified with the roughness of the first epitaxial wafer, and the test confidence level is obtained. Based on the spectral atlas, the roughness of the first epitaxial wafer, and the roughness of the second epitaxial wafer, the quality of the epitaxial wafer is analyzed to obtain initial epitaxial wafer quality parameters. Confidence compensation is then performed based on the test confidence level to obtain the final epitaxial wafer quality parameters.

2. The method for processing epitaxial wafer photoluminescence test data according to claim 1, characterized in that, Photoluminescence testing was performed on the target epitaxial wafer to obtain a spectral atlas. Several wavelengths were randomly selected, and the intensities of these wavelengths at different locations on the epitaxial wafer were extracted from the spectral atlas to obtain several wavelength intensity sets, including: Photoluminescence testing is performed on the target epitaxial wafer to obtain a spectral atlas, wherein the spectral atlas includes spectral maps of multiple locations on the target epitaxial wafer; Several wavelengths are randomly selected, including the main peak wavelength and several wavelengths adjacent to the main peak wavelength. Based on several wavelengths, the intensities of several wavelengths at different locations on the epitaxial wafer are extracted from the spectral atlas to obtain several wavelength intensity sets.

3. The method for processing epitaxial wafer photoluminescence test data according to claim 1, characterized in that, Calculate the rate of change of the plurality of wavelength intensity sets, process to obtain a plurality of wavelength intensity discrete parameters, perform epitaxial wafer roughness prediction, and obtain the first epitaxial wafer roughness, including: According to the preset spatial order on the target epitaxial wafer, calculate the variation amplitude of each wavelength intensity and the neighboring wavelength intensity of the next position point in several wavelength intensity sets to obtain several wavelength intensity variation rate sets; Calculate the rate of change discrete parameters of several wavelength intensity change rate sets to obtain several wavelength intensity discrete parameters; Based on several discrete parameters of wavelength intensity, the roughness of the epitaxial wafer is predicted to obtain the roughness of the first epitaxial wafer.

4. The epitaxial wafer photoluminescence test data processing method according to claim 3, characterized in that, Based on several discrete wavelength intensity parameters, the roughness of the epitaxial wafer is predicted to obtain the first epitaxial wafer roughness, including: A pre-trained epitaxial wafer roughness prediction network is obtained, wherein the epitaxial wafer roughness prediction network is constructed based on machine learning, and is obtained by supervising training using several sample wavelength intensity discrete parameter sets and sample epitaxial wafer roughness sets as training data. The training data is collected from epitaxial wafer test data over a historical period. The discrete parameters of several wavelength intensities are input into the epitaxial wafer roughness prediction network to predict and output the first epitaxial wafer roughness.

5. The method for processing epitaxial wafer photoluminescence test data according to claim 1, characterized in that, Acquire microscopic images of the target epitaxial wafer, identify and obtain the roughness of the second epitaxial wafer, verify it with the roughness of the first epitaxial wafer, and obtain test confidence levels, including: Acquire microscopic images of the target epitaxial wafer; An epitaxial wafer roughness recognition network is obtained, wherein the epitaxial wafer roughness recognition network is constructed using a convolutional neural network and obtained by supervised training using a set of sample microscopic images and a set of sample epitaxial wafer roughness as training data; The microscopic image is input into the epitaxial wafer roughness recognition network, and the second epitaxial wafer roughness is recognized and output. The roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer were verified to obtain the test confidence level.

6. The method for processing epitaxial wafer photoluminescence test data according to claim 5, characterized in that, The roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer are verified to obtain the test confidence level, including: The similarity between the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer is calculated as the test confidence level.

7. The method for processing epitaxial wafer photoluminescence test data according to claim 1, characterized in that, Based on the aforementioned spectral atlas, the roughness of the first epitaxial wafer, and the roughness of the second epitaxial wafer, epitaxial wafer quality analysis is performed to obtain initial epitaxial wafer quality parameters. Confidence compensation is then performed based on the aforementioned test confidence level to obtain further epitaxial wafer quality parameters, including: Based on the test confidence level, a first weight and a second weight are configured, and the roughness of the first epitaxial wafer and the roughness of the second epitaxial wafer are weighted and calculated to obtain the roughness of the epitaxial wafer. Based on the spectral atlas, the quality parameters of the spectral epitaxial wafer are identified and obtained; based on the roughness of the epitaxial wafer, the quality parameters of the rough spectral epitaxial wafer are classified and obtained. Based on the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters, initial epitaxial wafer quality parameters are obtained through processing. Confidence compensation is then performed using the test confidence level to obtain the final epitaxial wafer quality parameters.

8. The method for processing epitaxial wafer photoluminescence test data according to claim 7, characterized in that, Based on the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters, initial epitaxial wafer quality parameters are obtained through processing. Confidence compensation is then performed using the test confidence level to obtain further epitaxial wafer quality parameters, including: The initial epitaxial wafer quality parameters are calculated based on the spectral epitaxial wafer quality parameters and the rough spectral epitaxial wafer quality parameters. Configure the confidence compensation coefficient based on the test confidence level; The confidence compensation coefficient is used to compensate for the initial epitaxial wafer quality parameters to obtain the epitaxial wafer quality parameters.

9. A data processing system for epitaxial wafer photoluminescence testing, characterized in that, The system is used to implement the epitaxial wafer photoluminescence test data processing method as described in any one of claims 1 to 8, the system comprising: The spectral data acquisition module is used to perform photoluminescence testing on the target epitaxial wafer, obtain a spectral atlas, randomly select several wavelengths, extract the intensity of several wavelengths at different positions on the epitaxial wafer within the spectral atlas, and obtain several wavelength intensity sets, wherein several wavelengths include the main peak wavelength; The roughness prediction module is used to calculate the rate of change of the plurality of wavelength intensity sets, process to obtain a plurality of wavelength intensity discrete parameters, perform epitaxial wafer roughness prediction, and obtain the first epitaxial wafer roughness. The confidence verification module is used to acquire microscopic images of the target epitaxial wafer, identify and obtain the roughness of the second epitaxial wafer, verify it with the roughness of the first epitaxial wafer, and obtain the test confidence level. The quality parameter output module is used to perform epitaxial wafer quality analysis based on the spectral atlas, the roughness of the first epitaxial wafer, and the roughness of the second epitaxial wafer to obtain initial epitaxial wafer quality parameters, and to perform confidence compensation based on the test confidence level to obtain the epitaxial wafer quality parameters.

Images