XPS spectrogram analysis method and system, computer equipment and medium

Abstract

The invention provides an XPS spectrogram analysis method and system, computer equipment and a medium, and belongs to the field of surface analysis, and the method comprises the steps: obtaining to-be-recognized XPS spectrogram data; feature extraction of different scales is carried out on the XPS spectrogram data to be recognized according to the spatial dimension from high to low, and local region features corresponding to the physical attributes of all spectral peaks in the XPS spectrogram data to be recognized are obtained; performing up-sampling on the local region features to obtain a reconstructed spectrogram; fusing the local region features and the reconstructed spectrogram to obtain a fused feature map; performing physical attribute fitting on the fused feature map, matching physical attributes with a chemical shift reference range, and determining a chemical state; and constructing a fitting peak according to the physical attribute of the spectrum peak, and generating a background curve according to the chemical state of the XPS spectrogram data to be identified. According to the method, reliable chemical state information can be extracted without manual operation and fitting of detection personnel, and the reliability of an energy spectrum analysis result is ensured while the energy spectrum analysis efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of surface analysis, and particularly relates to an XPS spectrum analysis method and system, a computer device and a medium. BACKGROUND

[0002] X-ray photoelectron spectroscopy (XPS) is a powerful tool for studying the chemical composition and chemical state of the material surface. By analyzing the kinetic energy of photoelectrons, the type, content and chemical environment information of elements can be obtained. In high-resolution XPS spectra, the same element in different chemical states will produce spectral peaks with specific binding energy shifts. Spectral peak fitting is a key step for analyzing complex spectra and quantitatively obtaining information of each chemical state. However, XPS spectral peak fitting is strongly influenced by manual operation, is highly subjective, and depends on the experience of the detection personnel, and the detection quality is poor.

[0003] In view of the above defects, the prior art often uses XPS Peak, CasaXPS and other spectral peak fitting software to assist in energy spectrum analysis, but these fitting software also rely on manual operation and experience judgment of the detection personnel to determine the analysis results, and the reliability is low. SUMMARY

[0004] In order to solve the problem that the existing spectral peak fitting software relies on manual operation and experience judgment of the detection personnel to determine the analysis results, and the reliability is low, the present application provides an XPS spectrum analysis method, system, computer device and medium.

[0005] In order to achieve the above purpose, the present application provides the following technical solutions: An XPS spectrum analysis method, comprising: obtaining XPS spectrum data to be identified; performing feature extraction of different scales on the XPS spectrum data to be identified according to spatial dimensions from high to low, to obtain local region features corresponding to physical properties of each spectral peak in the XPS spectrum data to be identified; performing up-sampling on the local region features to obtain a reconstructed spectrum; fusing the local region features and the reconstructed spectrum to obtain a fusion feature map; performing physical property fitting of the spectral peak on the fusion feature map, matching the physical properties of the spectral peak with a preset chemical shift reference range, and determining the chemical state of the XPS spectrum data to be identified; constructing a fitting peak according to the physical properties of the spectral peak, and generating a background curve according to the chemical state of the XPS spectrum data to be identified.

[0006] Optionally, the application provides an XPS spectrum analysis method, wherein the chemical state is obtained by processing the to-be-identified XPS spectrum data through a deep learning network model; the deep learning network model comprises an encoder, a decoder, a skip connection layer, and an output layer; and the skip connection layer is connected to the encoder and the decoder.

[0007] Optionally, the XPS spectrum analysis method provided by the application further comprises: obtaining an XPS spectrum data training set; inputting the XPS spectrum data training set into an untrained deep learning network model to determine the chemical state, combining a label corresponding to the XPS spectrum data training set, and a regularization penalty term to calculate a loss function, and adjusting the deep learning network model according to the loss function calculation result, wherein the regularization penalty term comprises a regularization penalty term corresponding to a physical and chemical rule of a peak half-peak width range, a regularization penalty term corresponding to a physical and chemical rule of a theoretical energy difference of a specific element core level spin-orbit splitting, and a regularization penalty term corresponding to a physical and chemical rule of a theoretical area ratio of a specific element core level spin-orbit splitting.

[0008] Optionally, the XPS spectrum analysis method provided by the application further comprises: adding Gaussian noise and Poisson noise to the XPS spectrum data training set to obtain the XPS spectrum data training set after noise addition; performing random energy axis translation processing on the XPS spectrum data training set to obtain the XPS spectrum data training set after energy axis translation; performing baseline tilt processing on the XPS spectrum data training set to obtain the XPS spectrum data training set after baseline tilt processing.

[0009] Optionally, the XPS spectrum analysis method provided by the application further comprises: performing energy axis calibration on the energy points in the to-be-identified XPS spectrum data, and performing intensity normalization processing on the intensity values in the to-be-identified XPS spectrum data to obtain the preprocessed to-be-identified XPS spectrum data.

[0010] Optionally, the XPS spectrum analysis method provided by the application further comprises:

[0011] Optionally, the XPS spectrum analysis method provided by the application further comprises: based on the pre-set core level spin-orbit splitting theoretical rule corresponding to the specific element, the peak pairs of the fitting peak are associated with the same chemical component.

[0012] The application further provides an XPS spectrum analysis system, comprising: a data acquisition module configured to acquire XPS spectrum data to be identified; a chemical state prediction module configured to perform feature extraction of different scales on the XPS spectrum data to be identified according to spatial dimensions from high to low, to obtain local region features corresponding to physical properties of each spectrum peak in the XPS spectrum data to be identified; to perform up-sampling on the local region features to obtain a reconstructed spectrum; to fuse the local region features and the reconstructed spectrum to obtain a fusion feature map; and to perform physical property fitting of spectrum peaks on the fusion feature map, to match the physical properties of the spectrum peaks with a preset chemical shift reference range, and to determine a chemical state of the XPS spectrum data to be identified. an analysis result generation module configured to construct a fitting peak according to the physical properties of the spectrum peaks, and to generate a background curve according to the chemical state of the XPS spectrum data to be identified.

[0013] The application further provides a computer device comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any one of the XPS spectrum analysis methods.

[0014] The application further provides a computer readable storage medium, wherein the storage medium stores a computer program, and the computer program can implement the steps of any one of the XPS spectrum analysis methods when loaded by a processor.

[0015] The XPS spectrum analysis method provided by the application has the following beneficial effects: The XPS spectrum analysis method provided by the application can extract local features and up-sampling features from XPS spectrum data to be identified, fuse the spectrum peak physical property local features and the reconstructed spectrum, fit the peak parameters from the fusion results, determine the chemical state in combination with a chemical shift rule library, and determine reliable chemical state information without manual operation of a detection personnel, thereby improving the efficiency of energy spectrum analysis and ensuring the reliability of the energy spectrum analysis results. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the application and the design scheme thereof, the following will briefly introduce the drawings required by the embodiments. The drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0017] Figure 1 One of the XPS spectrum analysis method schematic diagrams provided by the embodiments of the application; Figure 2 One of the XPS spectrum analysis method schematic diagrams provided by the embodiments of the application; Figure 3 Figure 3 is a schematic diagram of an XPS spectrum analysis method provided by an embodiment of the present application; Figure 4 Figure 4 is a schematic diagram of an XPS spectrum analysis method provided by an embodiment of the present application; Figure 5 Figure 5 is a schematic diagram of an XPS spectrum analysis method provided by an embodiment of the present application; Figure 6 Figure 6 is an XPS spectrum analysis flowchart provided by an embodiment of the present application; Figure 7 Figure 7 is a schematic diagram of an XPS spectrum analysis system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to enable a person skilled in the art to better understand the technical solutions of the present application and to implement the same, the present application will be described in detail below with reference to the drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and cannot be used to limit the protection scope of the present application.

[0019] Existing XPS spectrum peak fitting is mainly realized manually by detection personnel, but the spectrum peak shape is affected by many factors such as signal-to-noise ratio, instrument broadening and intrinsic line shape, for example, Doniach-Šunjić line shape of metal or sp2 hybrid carbon which may contain asymmetry, etc. It is difficult to select a suitable and uniform peak shape function from Gaussian, Lorentz, Voigt, Doniach-Šunjić, etc. and spectrum analysis errors are easily caused by improper selection.

[0020] In addition, XPS spectrum peak fitting is highly subjective and depends on the experience of detection personnel. The selection of Linear, Shirley, Tougaard background subtraction methods, the initial setting and constraint of fitting parameters based on the FWHM range, peak area ratio, spin-orbit splitting parameters are largely dependent on the professional knowledge and experience of the operator, resulting in a lack of standardization and repeatability of the results. Inexperienced detection personnel may misjudge noise as a peak, improperly subtract the background, set FWHM too wide or with a sharp change without chemical basis, incorrectly handle or ignore the spin-orbit splitting rules corresponding to the fixed rules of energy difference and area ratio of specific core level peak pairs, add too many synthetic peaks without physical meaning in pursuit of "perfect" mathematical fitting, do not display residual plots to evaluate the quality of fitting, etc. Different analysts may draw significantly different conclusions about the same spectrum.

[0021] In addition, the manual fitting method also has the problems of time-consuming, laborious and low efficiency for complex multi-peak overlapping spectrum, confusion between the mathematical function for describing the observed spectral peak and the deconvolution for removing the instrument broadening effect to obtain a spectrum closer to the intrinsic line shape, and more difficult to accurately identify and fit each component due to the influence of low signal-to-noise ratio and low energy resolution of the spectrum, and misjudgment of the chemical state.

[0022] In view of the defects of manual fitting by the detection personnel, the spectrum fitting software such as XPS Peak, CasaXPS can be used for assistance, but this method still depends on the operation and judgment of the detection personnel, and has not realized the automatic, intelligent and standardized analysis in the true sense, which limits the reliability and comparability of the XPS technical analysis results.

[0023] The XPS spectrum analysis method provided by the application can learn the mapping relationship between the XPS spectrum characteristics and the chemical state information by the deep neural network, and can fuse the XPS physical and chemical rule constraints in the training process, so as to automatically, standardize and intelligently analyze the input XPS spectrum with high precision, extract reliable chemical state information, overcome the subjectivity and limitations in the traditional manual fitting process, and avoid the problems of error-prone, low efficiency, lack of standardization and difficulty in processing complex peak shape and low quality data in the XPS spectrum analysis.

[0024] Embodiment 1 The application provides an XPS spectrum analysis method, specifically as Figure 1 shown, comprising the following steps: Step 11, obtaining XPS spectrum data to be identified.

[0025] Step 12, performing feature extraction of different scales on the XPS spectrum data to be identified according to the spatial dimension from high to low, to obtain local region features corresponding to the physical properties of each spectral peak in the XPS spectrum data to be identified; performing up-sampling on the local region features to obtain a reconstructed spectrum; fusing the local region features and the reconstructed spectrum to obtain a fusion feature map; performing physical property fitting of the spectral peak on the fusion feature map, matching the physical properties of the spectral peak with a preset chemical shift reference range, and determining the chemical state of the XPS spectrum data to be identified.

[0026] Step 13, constructing a fitting peak according to the physical properties of the spectral peak, and generating a background curve according to the chemical state of the XPS spectrum data to be identified.

[0027] In addition, in the XPS spectrum analysis method provided by the application, the chemical state is obtained by processing the XPS spectrum data to be identified by a deep learning network model; the deep learning network model comprises an encoder, a decoder, a skip connection layer and an output layer.

[0028] The skip connection layer connects the encoder and decoder, specifically including the following steps: By using the encoder path, features of the XPS spectral data to be identified are extracted from high to low spatial dimensions to obtain local region features.

[0029] The reconstructed spectrum is obtained by upsampling local region features using a decoder.

[0030] By using skip connection layers, local region features and reconstructed spectral maps are fused to obtain a fused feature map.

[0031] The physical properties of the spectral peaks are fitted to the fused feature map by the output layer, and the physical properties of the spectral peaks are matched with the preset chemical shift reference range to determine the chemical state of the XPS spectral data to be identified.

[0032] Furthermore, based on the above-described embodiments, such as Figure 2 As shown, in the XPS spectrum analysis method provided by the present invention, before step 12, the method further includes: Step 19: Perform energy axis calibration on the energy points in the XPS spectrum data to be identified, and normalize the intensity values ​​in the XPS spectrum data to be identified to obtain preprocessed XPS spectrum data to be identified.

[0033] Specifically, the XPS spectral analysis method provided by this invention first receives XPS high-resolution spectral data containing energy coordinates and corresponding intensity values. Where I is intensity and E is binding energy or kinetic energy, such as discrete energy points. and their corresponding strength values Subsequently, the spectral data was analyzed. Standardization is performed to obtain preprocessed spectral data. The standardization process can include intensity normalization and energy axis calibration, respectively, for energy points in the spectral data. and strength value Normalization is performed. Intensity normalization can be achieved through min-max normalization, as shown in formula (1), which scales the intensity values ​​to a specific range, such as the interval [range]. middle:

[0034] (1) in, These are the intensity values ​​from the original spectral data. This represents the minimum intensity value in the original spectral data. This represents the maximum intensity value in the original spectral data. This is the intensity value after normalization. Furthermore, the energy value can be calibrated according to the energy axis, as shown in formula (2):

[0035] (2) wherein E is an energy value of the original spectrum data, is an energy value after normalization processing, is a reference energy value based on actual needs of a person skilled in the art, for example, a 1s contaminated carbon peak position of C. After the spectrum data is preprocessed, the spectrum data is input into a pre-trained deep learning neural network model M for analysis and prediction of sequence data in the XPS spectrum. The deep learning neural network model can adopt a one-dimensional convolutional neural network (1D CNN) architecture, which realizes feature extraction in the spectrum data by sequentially connecting one-dimensional convolutional layers, batch normalization layers, and nonlinear activation layers. Among them, the operation of the convolutional layer is shown in formula (3):

[0036] (3) wherein is an input feature map, for example, a spectral segment in the spectrum data, is a convolution kernel, for example, a trained filter, represents a convolution operation, which outputs a new feature map, extracts peak shape features and other local features of the spectrum data, and processes sequence dependencies through a recurrent neural network (RNN) or its variants, long short-term memory (LSTM) and gated recurrent unit (GRU).

[0037] For example, the deep learning neural network model can adopt a one-dimensional U-Net architecture, which realizes feature extraction and chemical state prediction by an encoder path, a decoder path, and a skip connection layer. The encoder path is provided with at least one down-sampling module, which extracts local features of the spectrum data and reduces the energy dimension through convolution and pooling operations, while the decoder path is provided with at least one up-sampling module, which is reconstructed through up-sampling or transposed convolution operations, thereby recovering the energy dimension, and realizing multi-scale feature fusion through the skip connection layer connecting the corresponding layers of the encoder path and the decoder path. Finally, the decoder path performs chemical state prediction based on the multi-scale feature fusion result.

[0038] In addition, the deep learning neural network model also includes an activation function layer, for example, a ReLU (Rectified Linear Unit) activation function is introduced to introduce nonlinearity, as shown in formula (4): (4) When the to-be-identified XPS spectrum data is predicted by the deep learning neural network model, the prediction result is further analyzed to form the final analysis result.

[0039] The XPS spectrum analysis method provided by the application can extract and fuse local features and up-sampling features of XPS spectrum data to be identified, and predict the chemical state, so that reliable chemical state information can be extracted without manual fitting operation of the detector, thereby improving the spectrum analysis efficiency and ensuring the reliability of the spectrum analysis result.

[0040] Based on the above-mentioned embodiments, as shown in Figure 3 The XPS spectrum analysis method provided by the application further comprises the following steps before step 12: Step 14: obtaining an XPS spectrum data training set.

[0041] Step 15: inputting the XPS spectrum data training set into an untrained deep learning network model to determine the chemical state, combining the corresponding label of the XPS spectrum data training set and a regularization penalty term to calculate a loss function, and adjusting the deep learning network model according to the loss function calculation result, wherein the regularization penalty term comprises a regularization penalty term corresponding to a physical and chemical rule of a peak half-peak width range, a regularization penalty term corresponding to a physical and chemical rule of a theoretical energy difference of a specific element core level spin-orbit splitting, and a regularization penalty term corresponding to a physical and chemical rule of a theoretical area ratio of a specific element core level spin-orbit splitting.

[0042] Specifically, in the XPS spectrum analysis method provided by the application, the regularization penalty term based on the X-ray photoelectron spectroscopy physical and chemical rule constraint is involved in the loss function calculation, so as to ensure that the trained model not only learns the mapping relationship between the XPS spectrum features and the chemical state information, but also realizes the chemical state analysis under the constraint of the XPS physical and chemical rule.

[0043] For example, in the model training stage, a data set containing a large number of input spectrum and corresponding target output is input into the deep learning model for model training. The corresponding target output includes not only the chemical state label corresponding to the input spectrum, but also the spectrum peak physical attribute value or peak parameter fitted by the XPS spectrum analysis expert in advance, and the component spectrum. Subsequently, the training prediction result predicted by the model is combined with the chemical state label and other information to optimize the calculation of the loss function, and the parameters of the deep learning model are adjusted based on the loss function calculation result, such as the weight w and the bias b in the model, so as to finally obtain a deep learning neural network capable of predicting XPS spectrum data.

[0044] Since the loss function in the model training stage of the XPS spectrum analysis method provided by the application introduces the regularization penalty term based on the XPS spectrum physical and chemical rule constraint, the trained model can meet the requirements of the XPS spectrum physical and chemical constraint, thereby ensuring the accuracy of the chemical state prediction and improving the XPS spectrum analysis effect.

[0045] The regularization penalty term determined based on the XPS energy spectrum physical and chemical constraints can include a regularization penalty term of a preset reasonable range of a full width at half maximum (FWHM) of a spectral peak, a regularization penalty term of a theoretical energy difference of a spin-orbit splitting of a core energy level of a specific element, and a regularization penalty term of a theoretical energy difference of a spin-orbit splitting of a core energy level of a specific element. The specific element is determined by a person skilled in the art based on actual XPS spectrum analysis requirements, for example, a C element, and the present application is not limited thereto.

[0046] For example, the loss function in the model training process can be designed in a composite form as shown in formula (5): (5) wherein, is a goodness-of-fit loss, used to measure the difference between the predicted reconstructed total spectrum and the input spectrum , or in a supervised case, the difference between the spectrum and the real component superimposed spectrum, is a parameter loss determined based on an accurate parameter label, used to measure the difference between the predicted peak parameter and the real parameter , is a physical and chemical constraint loss, used to punish the predicted results that do not conform to the physical law, and is a weight factor of different losses participating in the loss function calculation.

[0047] Specifically, the goodness-of-fit loss can be calculated by a mean-square error (MSE), as shown in formula (6): (6) wherein, is the goodness-of-fit loss, is the i-th reconstructed total spectrum, is the i-th input spectrum, and N is the total number of energy points in the spectrum. The physical and chemical constraint loss can be determined based on the following constraint conditions:

[0048] For the half width of the predicted peak in the training prediction result , when it exceeds the pre-set reasonable range, it is punished; for the predicted spin-orbit splitting pair in the training prediction result, for example , , the predicted energy difference and the theoretical value energy difference a deviation of the predicted area ratio from the theoretical area ratio , for example, a theoretical area ratio of 2:1 or 3:2 A deviation of the predicted area ratio (7) wherein, and are the predicted energy difference and the theoretical energy difference, respectively, and are the predicted area ratio and the theoretical area ratio, respectively, and are weight factors for different physical and chemical constraint conditions participating in loss calculation. is the spin-orbit coupling loss, which is part of the physical constraint loss, and by punishing the deviations of the energy difference ΔE and the area ratio R from the theoretical values, the model is forced to learn and comply with the physical rules of spin-orbit splitting.

[0049] Due to the XPS spectrum analysis method provided by the present application, the regularization penalty term is designed based on the peak half-width, the energy difference of the core level spin-orbit splitting, and the area ratio of the core level spin-orbit splitting. When the loss function is calculated, the peak half-width, the energy difference, and the area ratio can be considered at the same time. The trained model can simultaneously satisfy the physical and chemical constraints of the peak half-width, the energy difference, and the area ratio, further ensuring the accuracy of the chemical state prediction.

[0050] On the basis of the above-mentioned embodiments, as shown in Figure 4 , the XPS spectrum analysis method provided by the present application further comprises, before step 15: Step 16, adding Gaussian noise and Poisson noise to the XPS spectrum data training set to obtain the XPS spectrum data training set after noise addition.

[0051] Step 17, performing random energy axis translation processing on the XPS spectrum data training set to obtain the XPS spectrum data training set after energy axis translation.

[0052] Step 18, performing baseline tilt processing on the XPS spectrum data training set to obtain the XPS spectrum data training set after baseline tilt.

[0053] Specifically, the XPS spectrum analysis method provided by the present application can further perform data enhancement on the training spectrum in the model training stage through random transformation and the like, for example, adding Gaussian noise or Poisson noise, performing random energy axis translation, performing baseline tilt or drift, inputting the spectrum after data enhancement into the model for training. The level of Gaussian noise or Poisson noise, the random energy axis translation distance, and the baseline tilt or drift amplitude are set by a person skilled in the art based on actual needs, and the present application does not make any limitation. In addition, a deconvolution enhancement module can also be set in the model. After the model is trained in a specific manner, the instrument response function or the original-broadened spectrum pair can be determined, as shown in formula (8), and the intrinsic spectrum can also be solved or estimated, thereby realizing the prediction of the intrinsic spectrum :

[0054] (8) wherein, is the intrinsic spectrum, is the added noise.

[0055] Since the XPS spectrum analysis method provided by the present application can further perform data enhancement on the training set in the model training stage, simulate the Gaussian noise or Poisson noise, energy axis translation, baseline tilt and the like that may exist in actual spectrum analysis, and ensure that the model after training can also obtain accurate results for spectrum data with related interference, the generalization ability of the model is improved.

[0056] In addition, in the XPS spectrum analysis method provided by the present application, the physical properties of the spectrum peak include the center binding energy position of the chemical component peak, the half-peak width of the chemical component fitting peak, the peak area of the chemical component fitting peak, and the relative intensity of the chemical component fitting peak.

[0057] Based on the above embodiments, as shown in Figure 5 , the XPS spectrum analysis method provided by the present application further comprises the following steps after step 131: Step 132: based on the pre-set theoretical rules of core level spin-orbit splitting corresponding to specific elements, the peak pairs of the fitting peak are associated with the same chemical component.

[0058] Specifically, in the XPS spectrum analysis method provided by the present application, after the deep neural network model determines the chemical state prediction result based on the to-be-recognized XPS spectrum data, the chemical state prediction result can be further analyzed to generate background curve estimation, fitting peak parameters of each chemical component, and chemical state identification analysis information.

[0059] For example, the deep neural network model can further analyze the input spectrum after prediction, perform the actions of calculating the background estimation result and identifying the chemical state identification, and output the analysis result For the background estimation requirement, the background curve can be estimated implicitly or explicitly by a deep neural network model, for example, for Shirley background, its value at energy E is related to the signal integration of the region with higher kinetic energy or lower binding energy, the conceptual representation is shown in equation (9):

[0060] (9) At this time, the model outputs the background curve estimation result And the model can also output the fitting peak of each chemical component k

[0061] (10) At this time, the peak parameters of the corresponding peak peak position , FWHM, area , peak shape descriptor} are extracted from each fitting peak These peak parameters can be directly output by the model or obtained after post-processing based on the chemical state prediction result, which is not limited by the present application. For example, the fitting peak parameters can include the center binding energy position, half-peak width, peak area and relative intensity of each chemical component peak.

[0062] And based on the peak position and peak shape characteristics, the model can also specify the corresponding specific chemical state for each component k, and calculate the confidence In addition, spin-orbit splitting verification can also be performed, for example, automatically identifying peak pairs that meet the energy difference and area ratio rules and attributing them to spin-orbit splitting components of the same element. For example, by analyzing the spin-orbit splitting verification unit in the analysis module, automatically identifying peak pairs in the prediction result that meet the specific element core energy level spin-orbit splitting theoretical rules, such as energy difference rules and area ratio rules, and associating them as the same chemical component.

[0063] When the model completes the analysis of the spectrum data, the above analysis information can also be output through the output interface connected to the analysis module, for example, through a graphical display interface to simultaneously or optionally display the original spectrum, the total fitting curve, the fitting peak curve of each chemical component, the background curve and the residual curve, or output the parameter report in the form of text or table, listing the peak parameters, chemical state identification and optional confidence evaluation of each chemical component. For example, the original data, fitting line, component peak, background, residual and other results are displayed graphically, and the calculation of the residual is shown in equation (11):

[0064] (11) Where, is the residual, is the total fitted spectrum, is the original spectrum. In addition, a quality evaluation index such as the determination coefficient (R-squared) can also be fitted, and the determination coefficient calculation is shown in formula (12):

[0065] (12) wherein, is the average value of the input spectrum intensity, is the determination coefficient.

[0066] The XPS spectrum analysis method provided by the present application can further analyze the spectrum after completing the chemical state prediction, further reducing the workload of the detection personnel, avoiding the interference of manual operation on the analysis result, and improving the accuracy of XPS spectrum analysis.

[0067] Example 2 Based on example 1, taking the spectrum identification and analysis of the 2p oxidation state of the Si element as an example, the present application further provides a specific spectrum analysis example: First, the XPS data acquisition and labeling process in the model training stage, the data can include high-resolution Si 2p spectrum of the surface of silicon wafer with different oxidation degrees, such as naturally oxidized silicon wafer, thermally oxidized silicon wafer, and chemically treated silicon wafer. And using XPS instruments with known energy resolution and characteristics, such as XPS instruments equipped with monochromatic Al Kα source for spectrum data acquisition and synthesis, using Voigt function simulation of symmetric peak model, combined with Doniach-Šunjić line model to simulate Si 2p asymmetry physical model to generate synthetic Si 2p spectrum containing different proportions of elemental silicon , , , , components, wherein each oxidation state component contains spin-orbit split doublet in accordance with physical rules and , such as energy difference of about 0.6 eV and intensity ratio of about 2:1. The synthesized spectrum data needs to cover different signal-to-noise ratios, linear or Shirley background types and peak overlap degrees.

[0068] When the spectrum data acquisition is completed, at least two experienced XPS analysis experts independently perform spectrum peak fitting, such as using traditional fitting software such as CasaXPS, strictly performing background subtraction such as Shirley background, and for each chemical state, such as , , , , , and The peak pairs are parameterized for peak position, FWHM, area, etc., to ensure that the spin-orbit splitting parameters meet the physical rules. Subsequently, two XPS analysis experts cross-verify the fitting results, and reach a consensus as the "ground truth" of training. At this time, the simulation data comes with accurate labels corresponding to the "ground truth" determined by cross-verification of two XPS analysis experts, and the labeling information constitutes the target output Y in the training set.

[0069] Secondly, the data preprocessing process is performed on the collected or generated Si 2p spectrum data to obtain . Specifically, the energy axis is calibrated with the measured contaminant carbon C 1s peak, for example, set to 284.8 eV as a reference for calibration, or for conductive samples, calibrated with the Fermi edge reference. Moreover, the intensity value of each spectrum is scaled to the [0, 1] interval by the minimum-maximum normalization processing shown in formula (1). In addition, data augmentation can also be performed, which is not limited by the present application.

[0070] Then, a one-dimensional U-Net architecture is selected to construct a neural network model M for spectrum analysis, and the spectrum analysis process is as shown in Figure 6 . First, the input XPS spectrum data is standardized by the data preprocessing module, then the deep learning model for XPS spectrum features is constructed by the model construction module, and the model training is performed by the model training module using the labeled XPS data set, and the intelligent spectrum analysis module automatically performs background subtraction, peak fitting and other spectrum processing. Finally, the fitting curve and parameter table are displayed by the result output module. When the model training is completed, the XPS spectrum data to be recognized is input, standardized by the data preprocessing module, predicted by the trained model, and the background subtraction, peak fitting and other spectrum processing are performed by the intelligent spectrum analysis module according to the prediction result, and the result output module is output and displayed.

[0071] Specifically, the neural network model M includes an encoder (Encoder) path, a bottleneck layer (Bottleneck), a decoder (Decoder) path and an output layer (Output Layer).

[0072] wherein the encoder path is composed of multiple down-sampling modules, each of which usually contains two or more repeating units, each unit containing a one-dimensional convolution layer, a batch normalization layer, and a nonlinear activation layer based on a ReLU activation function or a Rectified Linear Unit activation function. The encoder path extracts multi-scale features of the spectrum and reduces the energy dimension or the spatial dimension step by step. The bottleneck layer, which connects the intermediate layers of the encoder and the decoder, contains convolution units. The decoder path is composed of multiple up-sampling modules, each of which contains an up-sampling operation such as transposed convolution or linear interpolation, and realizes the splicing of the feature maps of the corresponding layers of the decoder and the encoder through a skip connection, thereby preserving high-resolution details, and also includes convolution units composed of convolution layers, batch normalization layers, and ReLU activation layers. The decoder path gradually recovers the energy dimension and generates fine predictions by combining low-level and high-level features. The output layer is the last layer of the model, which can output a predicted spectrum with the same number of energy points as the input spectrum, and the predicted spectrum is the sum of the fitting peaks of each chemical state component. The spectrum of each component can also be directly output and background . and background .

[0073] In the model training process, the preprocessed Si 2p spectrum sequence is input into the 1D U-Net model for training. The composite loss function of the model is determined by formula (13) on the basis of formula (5):

[0074] (13) wherein is a hyperparameter weight, the goodness-of-fit loss is used to measure the similarity between the predicted total spectrum and the real input spectrum or the input spectrum fitted by an expert , and the similarity can be determined by calculating the loss through the mean square error as shown in formula (6). The physical and chemical constraint loss is a physical and chemical constraint regularization term. For example, the component spectrum output by the model is post-processed to extract the Si 2p doublet parameters, and a penalty term is calculated on the basis of formula (7) as shown in formula (14):

[0075] (14) wherein k traverses all identified to , or all components containing doublets are determined, and the energy difference The voltage is controlled to fluctuate around 0.6 eV, and the area ratio R is controlled to fluctuate around 2:1. and Specifically, these are the hyperparameter weights involved in the loss function calculation. Specifically, in the 2p level of Si, due to spin-orbit coupling, it splits into two sub-levels with very similar energies. "and" ", for these two sub-levels" "and" The energy difference between the two sub-level peaks is a basically fixed physical constant, hardly changing with changes in the chemical environment such as oxidation state. R represents the energy difference between the two sub-level peaks. "and" The ratio of the peak areas of the two peaks is 2:1 for the 2p energy level of Si.

[0076] In addition, soft constraints on the FWHM range can be added. Model training can be performed using gradient descent with the Adam optimizer. Specific training strategies include mini-batch gradient descent, setting an appropriate learning rate, training for several epochs, monitoring model performance using a validation set, and applying early stopping to prevent overfitting. The training platform can be TensorFlow or PyTorch.

[0077] Once the model training is complete, the Si 2p spectrum to be analyzed is input into the trained model M. The model directly outputs the predicted total fitted spectrum. and corresponding to Spectra of each chemical state component obtained from decomposition and background Then, for each output component spectrum... Find the peak point and determine the peak position. And determine the peak area by fitting the peak shape or calculating numerical integrals. Estimate the half-peak width And for Other components, identification and Peak pairs are used to verify whether their energy difference and area ratio are within the allowable error range, for example, energy difference. Whether it is within the range of [0.55, 0.65] eV, and the area ratio. Whether it is within the range [1.8, 2.2], and according to peak Based on the preset chemical shift reference range, the chemical state is identified. And the confidence can be given based on the probability output by the model or the fitting quality of post-processing .

[0078] Finally, the result output and visualization module, the output information can include the original spectrum in the graphical interface, the total fitting curve, the fitting peak of each chemical state, the background curve, and the residual curve determined based on formula (11); in the parameter table interface, for example, the peak position, FWHM, relative area percentage of each chemical state identified and the compliance check result and confidence of the spin-orbit splitting parameter; the determination coefficient fitting goodness index.

[0079] Among them, the calculation of the relative area percentage is shown in formula (15): (15) The results of the deep learning model fitting in this embodiment are compared with the results of manual fitting by experts, as shown in Table 1: Table 1 Comparison of fitting results As can be seen from the above table, by constructing and training a deep learning model based on 1D U-Net, this embodiment can quickly and automatically perform high-precision fitting and chemical state identification on Si 2p spectrum. Compared with the traditional manual method, the analysis time is significantly shortened, the consistency and standardization of the results are improved, and by incorporating physical constraints into the model, the physical and chemical rationality of the analysis results is ensured, especially in correctly handling spin-orbit splitting. The method has a significant advantage in processing a large amount of XPS data and achieving high-throughput analysis.

[0080] Embodiment 3 The application also provides an XPS spectrum analysis system, as shown in Figure 7 , comprising: A data acquisition module 31 for acquiring XPS spectrum data to be identified.

[0081] A chemical state prediction module 32 for performing feature extraction of different scales on the XPS spectrum data to be identified according to the spatial dimension from high to low, obtaining local region features corresponding to the physical properties of each spectral peak in the XPS spectrum data to be identified; performing up-sampling on the local region features to obtain a reconstructed spectrum; fusing the local region features and the reconstructed spectrum to obtain a fusion feature map; performing physical property fitting of the spectral peak on the fusion feature map, matching the physical properties of the spectral peak with a pre-set chemical shift reference range, and determining the chemical state of the XPS spectrum data to be identified.

[0082] ​The analysis result generation module 33 is configured to construct a fitting peak according to the physical attribute of the spectrum peak, and generate a background curve according to the chemical state of the XPS spectrum data to be identified.

[0083] The present application further provides a computer device, including a memory, a processor and a computer program stored in the memory, and the processor executes the computer program to implement the steps in the XPS spectrum analysis method embodiment. The specific implementation method can be referred to the method embodiment, which will not be repeated here.

[0084] Further, the present application further provides a non-transitory computer readable storage medium containing instructions, and the storage medium stores a computer program. For example, the storage medium containing instructions can be executed by the processor of the computer device to complete the above method. For example, the non-transitory computer readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk and an optical data storage device, etc. The computer program is executed by the processor, and the steps in the XPS spectrum analysis method embodiment can be implemented. The specific implementation method can be referred to the method embodiment, which will not be repeated here.

[0085] Those skilled in the art should understand that the embodiments of the present application can provide a method, a system or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a computer program product in the form of being implemented on one or more computer usable storage media containing computer usable program code (including but not limited to disk storage, CD-ROM, optical storage, etc.).

[0086] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system) and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The device for implementing the functions specified in one flow or multiple flows and / or blocks. Figure 1 The device for implementing the functions specified in one flow or multiple flows and / or blocks.

[0087] These computer program instructions can also be stored in a computer readable storage medium which can guide the computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction devices, which implement the functions specified in the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 the function specified in the one or more blocks.

[0088] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable data processing devices to generate a computer implemented process, so that the instructions executed on the computer or other programmable data processing devices provide a process for implementing the flow Figure 1 one or more processes and / or blocks Figure 1 the function specified in the one or more blocks.

[0089] It should be noted that the above detailed description of the present application can enable those skilled in the art to more fully understand the present application, but is not in any way limiting the present application. Therefore, although the present application has been described in detail in the present specification and examples, it should be understood by those skilled in the art that modifications or equivalent replacements can still be made to the present application; and all technical solutions and improvements that do not deviate from the spirit and scope of the present application are encompassed within the protection scope of the present application patent. Any reference signs in the claims should not be considered as limiting the claims involved. Simple changes or equivalent replacements of technical solutions that can be obviously obtained by those skilled in the art within the technical scope disclosed by the present application all belong to the protection scope of the present application.

Claims

1. An XPS spectral analysis method, characterized in that, include: Obtain the XPS spectrum data to be identified; The XPS spectrum data to be identified is subjected to feature extraction at different scales according to spatial dimensions from high to low, to obtain local region features corresponding to the physical properties of each spectral peak in the XPS spectrum data to be identified; the local region features are upsampled to obtain a reconstructed spectrum; the local region features and the reconstructed spectrum are fused to obtain a fused feature map; the physical properties of the spectral peaks are fitted to the fused feature map, and the physical properties of the spectral peaks are matched with a preset chemical shift reference range to determine the chemical state of the XPS spectrum data to be identified; Fitting peaks are constructed based on the physical properties of the spectral peaks, and background curves are generated based on the chemical state of the XPS spectrum data to be identified.

2. The XPS spectrum analysis method according to claim 1, characterized in that, The chemical state is obtained by processing the XPS spectrum data to be identified using a deep learning network model; the deep learning network model includes an encoder, a decoder, a skip connection layer, and an output layer; the skip connection layer connects the encoder and the decoder.

3. The XPS spectral analysis method according to claim 2, characterized in that, Before performing multi-scale feature extraction on the XPS spectral data to be identified, the process also includes: Obtain the XPS spectral data training set; The XPS spectrum training set is input into an untrained deep learning network model to determine the chemical state. The loss function is calculated by combining the labels corresponding to the XPS spectrum training set with regularization penalty terms. The deep learning network model is then adjusted based on the loss function calculation results. The regularization penalty terms include regularization penalty terms corresponding to the physicochemical rules of the half-peak width range of the spectral peak, regularization penalty terms corresponding to the physicochemical rules of the theoretical energy difference of the spin-orbit splitting of the core energy level of a specific element, and regularization penalty terms corresponding to the physicochemical rules of the theoretical area ratio of the spin-orbit splitting of the core energy level of a specific element.

4. The XPS spectrum analysis method according to claim 3, characterized in that, Before inputting the XPS spectral data training set into an untrained deep learning network model for chemical state determination, the process also includes: Gaussian noise and Poisson noise are added to the XPS spectrum training set to obtain a noise-added XPS spectrum training set. The XPS spectrum training set is subjected to random energy axis shifting to obtain the energy axis shifted XPS spectrum training set; The XPS spectral data training set is subjected to baseline tilting processing to obtain a baseline-tilted XPS spectral data training set.

5. The XPS spectrum analysis method according to claim 1, characterized in that, Before performing feature extraction at different scales according to spatial dimensions from high to low on the XPS spectral data to be identified, the process also includes: Energy axis calibration is performed on the energy points in the XPS spectrum data to be identified, and intensity normalization is performed on the intensity values ​​in the XPS spectrum data to be identified to obtain preprocessed XPS spectrum data to be identified.

6. The XPS spectrum analysis method according to claim 1, characterized in that, The physical properties of the spectral peaks include the central binding energy position of the chemical component peak, the half-width at half maximum (WHM) of the fitted peak, the peak area of ​​the fitted peak, and the relative intensity of the fitted peak.

7. The XPS spectrum analysis method according to claim 1, characterized in that, After constructing fitted peaks based on the physical properties of the spectral peaks and generating a background curve based on the chemical state of the XPS spectrum data to be identified, the method further includes: Based on the pre-set theoretical rules of core energy level spin-orbit splitting corresponding to specific elements, the peak pairs of the fitted peaks are associated with the same chemical composition.

8. An XPS spectral analysis system, characterized in that, include: The data acquisition module is used to acquire the XPS spectrum data to be identified; The chemical state prediction module is used to extract features from the XPS spectrum data to be identified at different scales according to the spatial dimension from high to low, to obtain local region features corresponding to the physical properties of each spectral peak in the XPS spectrum data to be identified; upsample the local region features to obtain a reconstructed spectrum; fuse the local region features and the reconstructed spectrum to obtain a fused feature map; fit the physical properties of the spectral peaks to the fused feature map, and match the physical properties of the spectral peaks with a preset chemical shift reference range to determine the chemical state of the XPS spectrum data to be identified; The analysis result generation module is used to construct fitting peaks based on the physical properties of the spectral peaks and generate background curves based on the chemical state of the XPS spectrum data to be identified.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the XPS spectral analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to perform the steps of the XPS spectral analysis method according to any one of claims 1 to 7.

Images