Full-automatic wafer film thickness detection method and system based on spectral reflection

Abstract

The present application relates to the technical field of semiconductor wafer, specifically including full-automatic wafer film thickness detection method and system based on spectral reflection, including: obtaining wafer surface reflection spectrum signal, extracting spectral feature information and wavelength parameter; identifying polarization characteristic parameter, interface interference characteristic parameter, introducing the wavelength parameter as adjustment variable, determining wavelength dependent term, generating global film thickness distribution atlas, solving the technical problems that in the detection scene of multi-layer thin film or interface roughness, the influence of incident light polarization state on reflection signal is not fully considered, reflection signal feature extraction distortion, the film thickness obtained by reflection characteristic backstepping deviates from the actual film thickness, identifying polarization characteristic parameter and interface interference characteristic parameter, correcting thin film refractive index to determine the thickness of each detection point, effectively improving the precision of thickness detection, reducing the influence of spectral signal noise and interface interference, realizing the technical effect of accurate presentation of wafer thin film global film thickness distribution.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor wafers, and particularly relates to a full-automatic wafer film thickness detection method and system based on spectral reflection. BACKGROUND

[0002] In the field of semiconductor manufacturing, the thickness uniformity of wafer films directly affects the electrical performance and yield of chips. With the chip manufacturing process breaking through to the micro-nanometer level, the requirements for detection precision, efficiency and automation degree continue to increase. Spectral reflection refers to the analysis of the reflection characteristics of thin films on different wavelengths of light to deduce the thickness.

[0003] Current wafer film thickness detection based on spectral reflection does not fully consider the influence of incident light polarization state on the reflection signal, ignores the correlation between thin film layer interface interference characteristics and wavelength, and leads to detection results being easily disturbed in the case of multi-layer thin films or high roughness interfaces. In addition, factors such as spectral signal noise, device power supply ripple in the detection process will introduce system errors, and it is difficult to achieve accurate characterization of the global film thickness through a single spectral feature. The generated film thickness distribution map often has local deviations, which increases the risk of yield loss caused by inaccurate film thickness detection in chip production.

[0004] In summary, in the detection scene of multi-layer thin films or rough interfaces in the prior art, the influence of incident light polarization state on the reflection signal is not fully considered, the reflection signal feature extraction is distorted, and the film thickness obtained by reflecting the characteristics deviates from the actual film thickness. SUMMARY

[0005] The present application provides a full-automatic wafer film thickness detection method and system based on spectral reflection, which aims to solve the technical problem in the prior art that in the detection scene of multi-layer thin films or rough interfaces, the influence of incident light polarization state on the reflection signal is not fully considered, the reflection signal feature extraction is distorted, and the film thickness obtained by reflecting the characteristics deviates from the actual film thickness.

[0006] In view of the above problems, the technical scheme of the present application is:

[0007] In a first aspect, the application provides a full-automatic wafer film thickness detection method based on spectral reflection, wherein the method comprises: using a laser light source to emit incident light to a wafer film, synchronously acquiring wafer surface reflection spectrum signals, extracting spectral feature information and wavelength parameters; based on the polarization state of the incident light, combining the wavelength parameters, identifying polarization feature parameters in the wafer surface reflection spectrum signals; based on the interlayer interface of the wafer film, combining the spectral feature information, identifying interface interference feature parameters in the wafer surface reflection spectrum signals; based on the polarization feature parameters in the wafer surface reflection spectrum signals, the interface interference feature parameters in the wafer surface reflection spectrum signals, introducing the wavelength parameters as adjustment variables to determine wavelength-dependent terms; determining the thickness values of each detection point of the wafer film according to the corrected film refractive index of the wavelength-dependent terms, and generating a full-domain film thickness distribution map.

[0008] Preferably, the spectral feature information includes reflection peak intensity, reflection valley depth, characteristic peak position and spectral half-width, and the wavelength parameter covers the tunable wavelength range of the laser light source.

[0009] Preferably, the polarization feature parameters include s-polarized light reflectivity, p-polarized light reflectivity and polarization state extinction ratio, the s-polarized light reflectivity and the p-polarized light reflectivity are determined by respectively collecting the reflection light intensity of the corresponding polarization direction through a polarization beam splitter, and the polarization state extinction ratio is the ratio of the s-polarized light reflectivity to the p-polarized light reflectivity.

[0010] Preferably, the interface interference feature parameters include interlayer reflection peak position offset, interface scattering intensity factor and interference fringe contrast, the interlayer reflection peak position offset is determined by the spectral peak position difference of the reflection light formed by different film layer interfaces, the interface scattering intensity factor is positively correlated with the interface roughness and is determined by conversion through the baseline drift of the reflection spectrum signal, and the interference fringe contrast is the difference value ratio of the reflection peak intensity to the reflection valley depth.

[0011] Preferably, the film refractive index corrected according to the wavelength-dependent terms is used for joint control analysis of a filter array corresponding to a light splitting member under grating+prism combination, a full-waveband spectral feature matrix is set, the matching degree of the full-waveband spectral feature matrix is introduced, and only when the matching degree meets the convergence termination condition of the mean square error loss function, the multi-modal denoising network converges to the global optimal solution.

[0012] Preferably, the noise suppression of the wafer surface reflection spectrum signal adopts a multi-modal denoising network based on Transformer: the original spectrum signal and the wafer surface topography image are taken as dual-modal input, time series modeling is performed through a Transformer encoder, edge features and texture features of the wafer surface topography image are extracted through a convolutional neural network, and then the dual-modal features are spliced and input into a full connection layer, a voltage ripple coefficient is determined using voltage closed-loop feedback data of a Cuk converter, and the voltage ripple coefficient is analyzed in correlation with a mean square error loss function, so as to set a spectrum peak shift error and a reflectivity measurement error as convergence termination conditions of the mean square error loss function.

[0013] Preferably, the voltage ripple coefficient is determined using voltage closed-loop feedback data of a Cuk converter, and a dynamic adjustment threshold of the Cuk converter is set: when it is detected that the spectrum peak shift error exceeds a preset threshold, duty cycle adjustment of the Cuk converter is automatically triggered, the suppression of high-frequency ripple is enhanced by reducing the cutoff frequency of an LC filter network, the voltage ripple coefficient is controlled within a ripple tolerance threshold, and the reflectance peak position measurement error is corrected to a peak position accuracy reference.

[0014] Preferably, for system error of multi-batch wafer detection, a long-term voltage stability parameter of the Cuk converter is included in an error tracing system, when detection error of any batch presents systematic deviation, by comparing detection data in different voltage drift intervals, the influence weight of power supply stability on thin film refractive index correction term is located, and the calibration coefficient of the wavelength-dependent term is optimized; a double error compensation mechanism is set by the power supply stability control of the Cuk converter and the noise suppression of the multi-modal denoising network.

[0015] Preferably, based on local features of the global film thickness distribution map, in combination with abnormal points of the wafer surface reflection spectrum signal, a defect candidate area is identified, a defect classifier is trained, the input is film thickness gradient and spectrum absorption peak shift of the defect candidate area, and the output is defect type; the average thickness, thickness standard deviation and maximum thickness deviation of the wafer thin film are determined through the global film thickness distribution map, the defect classification positioning annotation is performed in combination with the defect candidate area and the corresponding defect type, and visual prompting is performed through the defect classification positioning annotation.

[0016] In a second aspect, the application provides a full-automatic wafer film thickness detection system based on spectral reflection, wherein the system comprises: a feature extraction module that uses a laser light source to emit incident light to a wafer film, synchronously acquires wafer surface reflection spectrum signals, extracts spectral feature information and wavelength parameters; a feature parameter identification module that identifies polarization feature parameters in the wafer surface reflection spectrum signals based on the polarization state of the incident light in combination with the wavelength parameters; identifies interface interference feature parameters in the wafer surface reflection spectrum signals based on the film layer interface of the wafer film in combination with the spectral feature information; a wavelength dependence item determination module that determines a wavelength dependence item based on the polarization feature parameters in the wafer surface reflection spectrum signals, the interface interference feature parameters in the wafer surface reflection spectrum signals, and introduces the wavelength parameters as adjusting variables; and a thickness value determination module that determines thickness values of each detection point of the wafer film according to the film refractive index corrected by the wavelength dependence item, and generates a full-domain film thickness distribution map.

[0017] To sum up, one or more technical solutions provided in the application extract spectral features and wavelength parameters, identify polarization feature parameters and interface interference feature parameters in combination with the polarization state of incident light and the characteristics of the film layer interface, determine a wavelength dependence item by introducing the wavelength parameters as adjusting variables, determine the thickness of each detection point after correcting the film refractive index, effectively improve the accuracy of thickness detection, reduce the influence of spectral signal noise and interface interference, and achieve the technical effect of accurately presenting the full-domain film thickness distribution of the wafer film. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of a full-automatic wafer film thickness detection method based on spectral reflection is provided for the application.

[0019] Figure 2 A structural diagram of a full-automatic wafer film thickness detection system based on spectral reflection is provided for the application.

[0020] Legend of reference signs: feature extraction module M100, feature parameter identification module M200, wavelength dependence item determination module M300, and thickness value determination module M400. DETAILED DESCRIPTION

[0021] In Example 1, the application is specifically described below in combination with the drawings, as shown in the drawings, the application provides a full-automatic wafer film thickness detection method based on spectral reflection, wherein the method comprises: Figure 1

[0022] S1: using a laser light source to emit incident light to a wafer film, synchronously acquiring wafer surface reflection spectrum signals, extracting spectral feature information and wavelength parameters.

[0023] ​Specifically, the laser light source has the characteristics of good monochromaticity and strong directivity, is suitable as an incident light source for spectral reflectance detection, can ensure stable wavelength and concentrated energy of incident light, and reduce interference of unstable light source characteristics on the reflected signal; synchronous acquisition of wafer surface reflectance signal refers to real-time acquisition of light signals reflected from the wafer film surface by a spectral detector and the like at the same time of laser incidence, which contains spectral information related to the film thickness; spectral characteristic information is a key index reflecting the characteristics of the film extracted from the reflected spectrum signal, such as reflection peak intensity, reflection valley depth, etc., and wavelength parameter refers to the wavelength range and specific wavelength value covered by the incident light and reflected light.

[0024] The execution steps are as follows: a laser light source is used to emit incident light to the wafer film, for example, a tunable laser light source with a wavelength range of 300nm-1000nm is used to ensure that the key wavelength interval required for wafer film detection can be covered; when the incident light irradiates the wafer film surface, a spectrometer and other detection devices are started synchronously to acquire the reflected spectrum signal; spectral characteristic information such as reflection peak intensity, reflection valley depth, characteristic peak position and spectral half-width is extracted from the reflected spectrum signal, and the tunable wavelength range of the laser light source is recorded as the wavelength parameter, commonly, the tunable wavelength range of the laser light source is, for example, 300nm-1000nm; by acquiring the original reflected spectrum signal and key parameters, data support is provided for subsequent feature parameter identification and thickness calculation.

[0025] S2: based on the polarization state of the incident light, in combination with the wavelength parameter, identify the polarization characteristic parameter in the wafer surface reflectance signal; based on the film layer interface of the wafer film, in combination with the spectral characteristic information, identify the interface interference characteristic parameter in the wafer surface reflectance signal.

[0026] Specifically, the polarization state of the incident light refers to the vibration direction characteristics of the incident light, such as linear polarization, circular polarization, etc.; the polarization characteristic parameter is a parameter that can reflect the polarization characteristics of the reflected spectrum signal, including s-polarized light reflectivity, p-polarized light reflectivity and polarization state extinction ratio, etc.; the film layer interface refers to the interface between different layers of the wafer film; the interface interference characteristic parameter is a parameter reflecting the interference characteristics of the reflected light at the film layer interface, such as layer reflection peak position shift, interface scattering intensity factor and interference fringe contrast, etc.

[0027] Execution step: in identifying the polarization characteristic parameter, based on the known polarization state of the incident light, such as linearly polarized light, the vibration direction of which is 90° to the incident plane for s polarization, parallel to the incident plane for p polarization, combined with the extracted wavelength parameter, the reflection intensity of s polarization direction and p polarization direction is collected through the polarization beam splitter, for example, at a wavelength of 500 nm, the s polarization light reflection intensity is 0.6, and the p polarization light reflection intensity is 0.3, and then the s polarization light reflectivity is 0.6, the p polarization light reflectivity is 0.3, and the polarization state extinction ratio is 0.6 / 0.3=2.

[0028] In identifying the interface interference characteristic parameter, according to the interlayer interface situation of the wafer film, such as the interface of two layers of film, combined with the spectral characteristic information, the interlayer reflection peak position offset is determined, and further, such as the first layer interface reflection peak position at 520 nm, the second layer interface reflection peak position at 530 nm, the offset is 10 nm, the interface scattering intensity factor is converted through the baseline drift of the reflection spectrum signal, the interference fringe contrast is determined, and further, such as the reflection peak intensity is 0.7, the reflection valley depth is 0.3, and the contrast is (0.7-0.3) / 0.7; By digging the polarization characteristics and interface interference characteristics in the reflection spectrum signal, the limitation of relying on single spectral characteristics in the conventional method is broken through, which provides more comprehensive and accurate characteristics for subsequent refractive index correction and thickness calculation, effectively reduces the detection deviation caused by not considering the polarization and interface interference factors, and improves the adaptability to complex film structure.

[0029] S3: based on the polarization characteristic parameter in the wafer surface reflection spectrum signal and the interface interference characteristic parameter in the wafer surface reflection spectrum signal, introducing the wavelength parameter as an adjusting variable to determine the wavelength dependence; S4: according to the corrected thin film refractive index of the wavelength dependence, the thickness value of each detection point of the wafer film is determined, and the global film thickness distribution map is generated.

[0030] Specifically, the wavelength dependence refers to the regular relationship between the physical properties (such as refractive index) of the film and the wavelength, by integrating the polarization characteristic parameters (such as s / p polarization light reflectivity, polarization state extinction ratio) and the interface interference characteristic parameters (such as interlayer reflection peak position offset, interference fringe contrast), and taking the wavelength parameter as a dynamic adjusting variable, the correlation between these characteristic parameters and the wavelength is established, and the wavelength dependence is determined; The thin film refractive index is the key parameter for calculating the thickness of the thin film, which will change with the wavelength, and its correction through the wavelength dependence can improve the accuracy; The global film thickness distribution map is to present the thickness value of all detection points on the wafer in a visual way, and intuitively reflect the overall distribution of the film thickness.

[0031] Execution step: when determining the wavelength dependence, integrate the identified polarization characteristic parameters and interface interference characteristic parameters as adjustment variables, for the polarization characteristic parameters, such as in the wavelength range of 400-800 nm, the s-polarized light reflectivity increases linearly from 0.5 to 0.7, the p-polarized light reflectivity increases from 0.3 to 0.5, and the polarization state extinction ratio is stable at 1.67; for the interface interference characteristic parameters, such as the interlayer reflection peak position offset increases from 5 nm to 15 nm with the increase of wavelength, and the interference fringe contrast increases from 0.4 to 0.6; taking the wavelength parameter (400-800 nm) as the adjustment variable, a mathematical model is established by multivariate regression analysis algorithm, which can complete parameter fitting in a short time and ensure efficiency.

[0032] In the determination of the thickness value and the generation of the map, the wavelength dependence is used to correct the refractive index of the thin film, such as at a wavelength of 500 nm, the uncorrected refractive index is 1.45, and the corrected refractive index is 1.47. Combined with the thin film thickness calculation formula in the spectral reflection principle, further, based on the interference equation: thickness = (λ x phase difference) / (4π x corrected refractive index), the thickness value of each 10μm x 10μm detection point on the wafer is obtained, and the thickness data of multiple detection points are plotted into a global film thickness distribution map through image processing technology; by establishing the wavelength dependence to correct the refractive index, the calculation deviation caused by the fixed value of the refractive index in the conventional method is solved, and the global film thickness distribution map effectively improves the detection precision and global characterization ability under the complex thin film structure.

[0033] Further, the spectral feature information and the wavelength parameter are extracted, and the method of the present application comprises:

[0034] The spectral feature information includes reflection peak intensity, reflection valley depth, characteristic peak position, and spectral half-width, and the wavelength parameter covers the tunable wavelength range of the laser light source.

[0035] Specifically, the spectral feature information is a key feature index extracted from the wafer surface reflection spectrum signal, wherein the reflection peak intensity refers to the maximum value of the reflection intensity in the spectrum, reflecting the reflection ability of the light at a specific wavelength; the reflection valley depth is the minimum value of the reflection intensity in the spectrum, corresponding to the reflection peak intensity, and together reflecting the fluctuation characteristics of the spectrum; the characteristic peak position is the wavelength position corresponding to the reflection peak, which is an important reference point for analyzing the optical characteristics of the thin film; the spectral half-width is the wavelength width of the reflection peak at half the peak intensity, which is used to describe the width of the peak and is related to the uniformity and other characteristics of the thin film; the wavelength parameter refers to the wavelength range and specific wavelength values within the range that can be adjusted by the laser light source, and covering the tunable wavelength range of the laser light source means that the extracted wavelength parameter completely contains all the wavelength intervals that the light source can output, ensuring that the subsequent analysis can be based on all available wavelength information of the light source.

[0036] The execution step is: when the incident light emitted by the laser light source irradiates the wafer film, the spectral detection device synchronously collects the reflected spectrum signal, analyzes the signal through a signal processing algorithm, extracts the above-mentioned spectral characteristic information, and records the adjustable wavelength range of the laser light source, such as taking the adjustable wavelength range of 300nm-1000nm as the wavelength parameter, to provide basic data support for subsequent characteristic parameter identification; the spectral characteristic information such as the reflection peak intensity and the reflection valley depth is directly related to the optical properties of the film, the characteristic peak position is related to the film thickness, and the spectral half-width can reflect the quality state of the film; the wavelength parameter covers the adjustable range of the light source, which ensures that the wavelength information is complete when the characteristic parameters are identified in combination with the polarization state and the film interlayer interface in the subsequent analysis, avoids incomplete characteristic extraction due to missing wavelength information, and further affects the detection precision, and is the initial data acquisition and extraction link of the multi-feature fusion detection system.

[0037] Further, the method of the application further comprises:

[0038] The polarization characteristic parameters include s-polarized light reflectivity, p-polarized light reflectivity, and polarization state extinction ratio. The s-polarized light reflectivity and the p-polarized light reflectivity are determined by collecting the reflection intensity of the corresponding polarization direction through a polarization beam splitter, and the polarization state extinction ratio is the ratio of the s-polarized light reflectivity to the p-polarized light reflectivity.

[0039] Specifically, the polarization characteristic parameter is a key index for describing the polarization characteristics of the wafer surface reflection spectrum signal. The s-polarized light reflectivity is the ratio of the reflection intensity of the s-polarized light with the vibration direction perpendicular to the incident plane to the incident light intensity, and the p-polarized light reflectivity is the ratio of the reflection intensity of the p-polarized light with the vibration direction parallel to the incident plane to the incident light intensity. The polarization state extinction ratio is the ratio of the s-polarized light reflectivity to the p-polarized light reflectivity, which is used to quantify the difference between the reflection characteristics of the two polarization states. The polarization beam splitter is an optical element that can separate light with different polarization directions, and through it, the reflection intensities of s-polarized light and p-polarized light are collected, and then the corresponding reflectivity is obtained.

[0040] The execution step is: when the incident light (including s and p polarizations) irradiates the wafer film surface, the reflected light carrying the polarization information propagates to the polarization beam splitter, the polarization beam splitter separates the s polarized light and the p polarized light in the reflected light, so that the two enter the corresponding detector, for example, the incident light intensity is I0, after the polarization beam splitter, the s polarized light reflection intensity is detected as Is=0.6I0, then the s polarized light reflectivity is 0.6I0 / I0=0.6; the p polarized light reflection intensity is detected as Ip=0.3I0, then the p polarized light reflectivity is 0.3I0 / I0=0.3, at this time the polarization extinction ratio is 0.6 / 0.3=2; in the above step, the polarization characteristics in the reflection spectrum are accurately extracted, the polarization beam splitter is used to realize the separation and collection of different polarization states of the reflected light, which provides data for the subsequent analysis of the correlation between the polarization characteristics and the film thickness combined with the wavelength parameters, and further, in the multi-layer film or high roughness interface scene, the reflection characteristics of s polarized light and p polarized light are significantly different, these parameters can effectively reflect the optical anisotropy and interface characteristics of the film, and make up for the feature extraction distortion problem caused by polarization, and lay a key polarization feature foundation for improving the film thickness detection accuracy.

[0041] Further, the method of the application further comprises:

[0042] The interface interference characteristic parameters include interlayer reflection peak position offset, interface scattering intensity factor and interference fringe contrast, the interlayer reflection peak position offset is determined by the peak position difference of the reflection light formed by different film layer interfaces, the interface scattering intensity factor is positively correlated with the interface roughness and is determined by conversion of the baseline drift of the reflection spectrum signal, and the interference fringe contrast is the difference value ratio of the reflection peak intensity and the reflection valley depth.

[0043] Specifically, the interface interference characteristic parameters are key parameters reflecting the interference characteristics of the reflection light of different interlayer interfaces of the wafer film, wherein the interlayer reflection peak position offset refers to the wavelength position difference between the characteristic peaks formed by the reflection light of different film layer interfaces on the spectrum, which directly reflects the interference difference of the reflection light of each layer interface; the interface scattering intensity factor is an index for quantifying the scattering degree of the film interlayer interface, and the numerical value is positively correlated with the interface roughness, that is, the rougher the interface is, the larger the factor value is, which can be obtained by conversion of the baseline drift in the reflection spectrum signal; the interference fringe contrast is the ratio of the difference value of the reflection peak intensity and the reflection valley depth to the reflection peak intensity (or to the reflection valley depth), which is used to describe the clarity of the interference fringe, and the higher the contrast is, the more obvious the interference characteristics are.

[0044] Execution step: when laser is incident to the multilayer wafer film, the interface between different layers reflects light, and the reflected light of each interface forms a corresponding reflection peak in the spectrum. By analyzing the spectrum signal, the wavelength position of these peaks can be determined. For example, the first layer interface reflection peak is located at 450 nm, and the second layer interface reflection peak is located at 460 nm. The layer reflection peak position offset is 10 nm. This parameter can reflect the optical path difference of different layer interfaces and provide a basis for analyzing the relationship between layer thickness.

[0045] For the interface scattering intensity factor, if the baseline of the reflection spectrum signal drifts by 0.1 due to interface scattering, the factor can be obtained through a preset conversion relationship, such as scattering intensity factor = 0.5 x baseline drift. When the interface roughness increases, the baseline drift increases, and the factor value also rises, thereby realizing indirect characterization of the interface roughness. The calculation of the interference fringe contrast is based on the extracted reflection peak intensity and reflection valley depth. Assuming that the reflection peak intensity is 0.8 and the reflection valley depth is 0.2, the difference between them is 0.6. Taking the reflection peak intensity as the reference, the proportion is 0.6 / 0.8 = 0.75. The higher this value, the more significant the interference phenomenon, and the more conducive to the reverse calculation of the film thickness through the interference characteristics. Preferably, in the case of a multilayer film or an interface roughness, the optical properties of the film interlayer interface are fully utilized to determine the interface interference details, which provides key interface characteristics for subsequent refractive index correction combined with spectral feature information and wavelength parameters, reduces the detection deviation caused by interface interference, and significantly improves the detection adaptability of complex film structures.

[0046] Further, according to the film refractive index corrected based on the wavelength dependence, the thickness value of each detection point of the wafer film is determined. The method of the present application further comprises:

[0047] According to the film refractive index corrected based on the wavelength dependence, the filter array corresponding to the light splitting member under the grating + prism combination is analyzed, and a full-band spectral feature matrix is set. The matching degree of the full-band spectral feature matrix is introduced, and only when the matching degree meets the convergence termination condition of the mean square error loss function, the multi-modal denoising network converges to the global optimal solution.

[0048] Specifically, the thin film refractive index after the wavelength-dependent item correction refers to the refractive index value obtained by correcting the original thin film refractive index by the wavelength-dependent item with the wavelength parameter as the adjusting variable, which dynamically adjusts with the change of wavelength and is more in line with the actual optical characteristics; the light splitting component under the combination of grating + prism is an optical assembly that splits the composite light into different wavelength monochromatic light, wherein the grating splits light through diffraction, and the prism splits light through refraction, and the combination of the two can improve the light splitting precision and wavelength coverage range; the filter array is composed of multiple filters of different wavelengths, which is used for selectively transmitting light of a specific wavelength, and cooperates with the light splitting component to realize fine screening of the spectrum.

[0049] The full-band spectral feature matrix is a data structure formed by integrating the spectral feature information (such as the reflection peak intensity and reflection valley depth at each wavelength) in the full-band range in the form of a matrix, which is used to comprehensively characterize the spectral characteristics; the matching degree of the full-band spectral feature matrix refers to the similarity between the actually collected spectral feature matrix and the pre-set standard spectral feature matrix, and the higher the matching degree, the more reliable the spectral data; the convergence termination condition of the mean square error loss function refers to the condition that the network stops training when the mean square error loss value of the multi-modal denoising network is reduced to below the pre-set threshold and no longer changes significantly, which is used to ensure that the network denoising effect reaches the expectation; the multi-modal denoising network is a neural network that can simultaneously process multiple types of input data (such as spectral signals and image signals), which is used to suppress noise and converge to a global optimal solution to represent the state of the network model reaching the best denoising effect.

[0050] Execution steps: link the thin film refractive index after the wavelength-dependent item correction with the light splitting component of the grating + prism combination, such as the corrected refractive index value at every 10 nm interval within the 400 nm-1000 nm wavelength range; the light splitting component performs light splitting processing on the reflection spectrum, and cooperates with the filter array to screen light of different wavelengths, such as filters containing wavelengths of 400 nm, 450 nm, …, 1000 nm; through the joint control analysis of the two, the spectral feature information corresponding to each wavelength within the full-band is integrated into a full-band spectral feature matrix, such as a 100x5 matrix, 100 representing 100 wavelength points, and 5 representing 5 corresponding spectral features; determine the matching degree of the full-band spectral feature matrix and the standard matrix, if the matching degree reaches the pre-set threshold (such as 90%), and the mean square error loss function value of the multi-modal denoising network is reduced to below 0.001 (satisfying the convergence termination condition), then the network converges to a global optimal solution.

[0051] Through the joint control analysis of the light splitting member and the filter array, the accuracy and comprehensiveness of the spectral feature matrix are improved, and the full-band spectral feature matrix provides more abundant spectral basis for subsequent thickness calculation; and the convergence condition of the matching degree and the mean square error loss function can ensure that the multi-modal denoising network achieves the best denoising effect, reduces the interference of spectral signal noise on the detection result, improves the signal-to-noise ratio of the spectral signal, and thus improves the matching degree of the full-band spectral feature matrix to meet the standard, and lays a high-quality data foundation for subsequent accurate determination of wafer film thickness.

[0052] Further, the method of the application further comprises:

[0053] The noise suppression of the wafer surface reflection spectrum signal adopts a multi-modal denoising network based on Transformer: the original spectrum signal and the wafer surface topography image are taken as dual-modal inputs, time series modeling is performed through a Transformer encoder, and then the edge features and texture features of the wafer surface topography image are extracted through a convolutional neural network for dual-modal feature splicing, and then input into a full connection layer, the voltage ripple coefficient is determined using the voltage closed-loop feedback data of the Cuk converter, and the voltage ripple coefficient is analyzed for correlation with the mean square error loss function, and the spectral peak shift error and reflectivity measurement error are used to set the convergence termination condition of the mean square error loss function.

[0054] Specifically, the multi-modal denoising network based on Transformer is a noise suppression model that integrates the Transformer architecture and multi-modal data processing capabilities, where multi-modal refers to processing two different types of inputs, namely the original spectrum signal and the wafer surface topography image; the Transformer encoder models the time series data (such as the sequence of spectrum signals changing with wavelength) input through a self-attention mechanism, capturing long-distance dependencies in the data; the convolutional neural network is good at extracting local features of images, and is used to extract edge features and texture features including film layer boundary profiles from the wafer surface topography image; dual-modal feature splicing is to combine the spectral features processed by the Transformer encoder and the image features extracted by the convolutional neural network into a unified feature vector to realize the fusion of multi-source information; the full connection layer is a level in which each neuron in the neural network is connected to all neurons in the previous layer, and is used for nonlinear transformation and final denoising processing output of the spliced features.

[0055] The Cuk converter is a DC-DC power converter with step-up and step-down functions, and the voltage closed-loop feedback data refers to the process data of real-time monitoring and adjusting the output voltage through the feedback mechanism; the voltage ripple coefficient is an index for measuring the size of the alternating current component in the output voltage, and the smaller the ripple coefficient, the better the voltage stability; the correlation analysis is used to determine the correlation between the voltage ripple coefficient and the mean square error loss function to evaluate the influence of the power supply ripple on the network error; the spectral peak position shift error refers to the deviation of the detected reflection peak wavelength position from the true position, and the reflectivity measurement error refers to the deviation of the detected reflectivity value from the true value, both of which are used as the basis for setting the convergence termination condition of the mean square error loss function, that is, when the two errors output by the network are within the preset range, the loss function stops converging.

[0056] The execution steps are as follows: collect the original spectrum signal and wafer surface topography image, and input them into the multimodal denoising network as dual-mode input, wherein the original spectrum signal is input into the Transformer encoder to learn the time correlation of different wavelength reflection intensities through self-attention mechanism, such as the dependence relationship between the reflection peak intensity change of any wave band and the adjacent wave band; at the same time, the wafer surface topography image is input into the convolutional neural network, and the edge features and texture features are extracted through convolutional layers and pooling layers in turn; for example, the edge features are clear contour lines of the interlayer interface of the thin film layer, and the texture features are light and dark texture distribution on the surface due to roughness; the spectral feature vector output by the Transformer is spliced with the image feature vector output by the convolutional neural network to form a fusion feature vector, which is input into the full connection layer for processing, and the denoised spectrum signal is output.

[0057] It is known that the voltage ripple of the laser light source affects the light source intensity and spectral characteristics, and further causes the spectral peak position to drift, affecting the film thickness measurement accuracy. In the process of using the laser light source to emit incident light, the voltage ripple coefficient is obtained by real-time acquisition of the voltage closed-loop feedback data of the Cuk converter, such as the fluctuation data of the output voltage within the range of 12V±0.1V. Specifically, for a sinusoidal ripple, the effective value is about 1 / Therefore, the effective value of the alternating current ripple is (0.1V÷ ) V, and thus the voltage ripple coefficient is determined as: alternating current ripple effective value ÷ direct current component = (0.1V÷ ) ÷ 12V × 100%, and accordingly, through correlation analysis, it is found that when the ripple coefficient increases, the mean square error loss function value rises, indicating that the power supply ripple has a significant impact on the detection error. Based on this, preferably, the convergence termination condition of the mean square error loss function is set as: spectral peak position shift error ≤0.5nm and reflectivity measurement error ≤1%; when the network training meets this condition, the iteration is stopped, and at this time, the noise interference such as reflection peak jitter caused by the power supply ripple in the denoised spectrum signal output is effectively suppressed.

[0058] Further, the voltage closed-loop feedback data of the Cuk converter is used to determine the voltage ripple coefficient, and the method comprises:

[0059] The voltage closed-loop feedback data of the Cuk converter is used to determine the voltage ripple coefficient, and the dynamic adjustment threshold of the Cuk converter is set: when the spectral peak position offset error is detected to exceed the preset threshold, the duty cycle adjustment of the Cuk converter is automatically triggered, the cutoff frequency of the LC filter network is reduced to enhance the suppression of high-frequency ripples, and the voltage ripple coefficient is controlled within the ripple tolerance threshold, and the corresponding reflection peak position measurement error is corrected to the peak position accuracy reference.

[0060] Specifically, the voltage closed-loop feedback data of the Cuk converter refers to the data that the Cuk converter monitors the output voltage in real time through the feedback loop during the working process, and returns the deviation information of the actual voltage from the target voltage to the control unit, which is used for dynamically adjusting the output voltage to maintain stability; the voltage ripple coefficient is the ratio of the AC ripple component to the DC component in the output voltage, which is a key indicator for measuring voltage stability; the dynamic adjustment threshold is a critical value of the spectral peak position offset error preset for triggering the adjustment mechanism of the Cuk converter, and when the actual error exceeds the value, the adjustment process is started; the duty cycle adjustment refers to adjusting the size and stability of the output voltage by changing the ratio of the conduction time to the period of the switch tube in the Cuk converter; the LC filter network is a filter circuit composed of inductors and capacitors, which is used to filter out the AC ripples in the output voltage, and reducing the cutoff frequency of the LC filter network means improving the low-pass performance of the filter, thereby suppressing high-frequency ripples; the ripple tolerance threshold is the maximum value of the allowed voltage ripple coefficient, and within this range, the detection system can work normally; the peak position accuracy reference is the target control value of the reflection peak position measurement error, that is, the accuracy standard that needs to be reached after correction.

[0061] The execution steps are as follows: the Cuk converter continuously collects voltage closed-loop feedback data to obtain the current voltage ripple coefficient, at the same time, if the dynamic adjustment threshold of the preset spectral peak position offset error is 1 nm, the ripple tolerance threshold is 0.005, and the peak position accuracy reference is ±0.3 nm; when the spectral peak position offset error at any wavelength is detected to reach 1.2 nm (exceeding 1 nm), the duty cycle adjustment of the Cuk converter is triggered, the switch tube duty cycle is adjusted, the cutoff frequency of the LC filter network is reduced, and the filtering effect on high-frequency ripples is enhanced. Specifically, the lower the cutoff frequency, the stronger the attenuation ability of high-frequency ripples above the cutoff frequency, and although the duty cycle of the Cuk converter does not directly determine the inductance parameters and capacitance parameters, it can indirectly affect the frequency distribution of ripples by changing the switching frequency or energy transmission efficiency. After the adjustment is completed, the voltage ripple coefficient is reduced until it is lower than the ripple tolerance threshold 0.005, and the corresponding reflection peak position measurement error is corrected to be within the peak position accuracy reference limit of ±0.3 nm.

[0062] Preferably, in the environment with serious high-frequency ripple interference, the influence of high-frequency ripple cannot be fully reflected by the ripple coefficient alone. The frequency distribution of the ripple is detected by a spectrum analyzer, and the spectrum analysis of the ripple is combined to ensure that the high-frequency component is effectively suppressed, effectively guaranteeing the consistency of the detection accuracy. By linking the spectral peak position deviation error with the voltage regulation of the Cuk converter, a real-time voltage ripple suppression closed loop is formed, solving the problem of measurement deviation of the reflection peak position caused by the ripple of the equipment power supply, and significantly improving the stability of the spectral feature extraction, providing a more reliable basis for the film thickness conversion based on the peak position parameter.

[0063] Further, the method of the present application comprises:

[0064] Systematic errors in multi-batch wafer detection, long-term voltage stability parameters of the Cuk converter are included in the error traceability system. When the detection error of any batch shows systematic deviation, the influence weight of power supply stability on the thin film refractive index correction term is located by comparing the detection data in different voltage drift intervals, and the calibration coefficient of the wavelength-dependent term is optimized; the power supply stability control of the Cuk converter and the noise suppression of the multi-modal denoising network are set as a double error compensation mechanism.

[0065] Specifically, the systematic error in multi-batch wafer detection refers to the regularity and repeatability error caused by the detection system itself, such as performance drift after long-term operation of the equipment, rather than random error; the long-term voltage stability parameter of the Cuk converter is an index for measuring the stability of the output voltage of the Cuk converter during long-term operation, usually represented by the voltage drift within a certain time; the error traceability system is a systematic framework for tracking error sources and analyzing error propagation paths, and the long-term voltage stability parameter is included in it, which can clearly show the influence of power supply stability on the detection error; systematic deviation refers to the overall deviation of the detection error of a batch of wafers in a certain direction, such as generally larger or smaller, rather than random deviation of individual detection points; the voltage drift interval is the division of voltage drift over time into different ranges for analyzing the differences in detection data under different drift levels; the influence weight is a parameter that quantifies the influence of power supply stability on the thin film refractive index correction term, and the greater the weight, the more significant the influence of power supply stability on the correction term; the calibration coefficient of the wavelength-dependent term is a parameter for adjusting the calculation accuracy of the wavelength-dependent term, and optimizing this coefficient can improve the accuracy of the refractive index correction; the double error compensation mechanism refers to compensating for the detection error by simultaneously controlling the power supply stability of the Cuk converter (hardware level) and suppressing the noise of the multi-modal denoising network (algorithm level), which works together to reduce the overall error.

[0066] Execution step: continuously record the long-term voltage stability parameters of the Cuk converter, such as the voltage drift within 100 hours of continuous operation of ±0.2V, and incorporate it into the error traceability system; when detecting a batch of wafers, it is found that the film thickness values of all detection points in this batch are on average 2nm higher than the standard value, i.e. there is a systematic deviation, at this time the voltage drift data during the detection of this batch is retrieved, which is divided into multiple intervals, such as 0-0.05V, 0.05-0.1V, 0.1-0.2V, and the detection data corresponding to different intervals are compared: for example, when the voltage drift is 0V-0.05V, the film thickness detection error is 0.5nm on average; when the drift is 0.1V-0.2V, the error is 2.5nm on average, thereby locating the influence weight of power supply stability on the correction term of thin film refractive index, based on this, the calibration coefficient of wavelength dependence term is optimized, so that the corrected refractive index is more close to the actual value.

[0067] At the same time, the dynamic voltage regulation of the Cuk converter reduces the influence of power supply ripple on detection; combined with the noise suppression of the multi-modal denoising network on the spectral signal, a double error compensation mechanism is formed, preferably, the thickness detection system error between the same batch is consistent in terms of standard deviation, combined with the double error compensation mechanism of different batch wafer films, the system error of multiple batches of wafer detection can be controlled, effectively solving the error accumulation problem caused by equipment performance drift and environmental interference in long-term detection, ensuring the reliability and comparability of the thickness detection results of different batches of wafer films, and meeting the high requirements of large-scale production on detection stability.

[0068] Further, the method of the present application further comprises:

[0069] Based on the local features of the global film thickness distribution map, combined with the abnormal points of the wafer surface reflectance spectrum signal, the defect candidate area is identified, and the defect classifier is trained, the input is the film thickness gradient of the defect candidate area and the spectral absorption peak shift, and the output is the defect type; through the global film thickness distribution map, the average thickness, thickness standard deviation and maximum thickness deviation of the wafer thin film are determined, combined with the defect candidate area and the corresponding defect type for defect classification positioning labeling, and the defect classification positioning labeling is visualized to remind.

[0070] In particular, the local feature of the global film thickness distribution map refers to the film thickness distribution characteristics that the local region presents obvious differences with the surrounding region in the global film thickness distribution map, such as sudden increase, sudden decrease or abnormal fluctuation of local thickness, etc.; the abnormal point of the wafer surface reflection spectrum signal refers to the point deviating from the normal law in the reflection spectrum signal, for example, the reflectivity at a certain wavelength suddenly abnormally increases or decreases, the reflection peak position appears unexpected large deviation, etc.; the defect candidate region refers to the wafer thin film region where the defect may exist which is preliminarily determined by analyzing the local feature and the abnormal point; the defect classifier is a model based on machine learning, which is used to identify and classify the type of the defect candidate region; the film thickness gradient refers to the change rate of the thin film thickness in the defect candidate region, reflecting the speed of thickness change; the spectral absorption peak shift refers to the wavelength position difference between the spectral absorption peak of the defect candidate region and the absorption peak of the normal region; the defect type includes film thickness unevenness, local over-thickness, local over-thin, abnormal absorption caused by interface particle contamination, etc.

[0071] The average thickness refers to the arithmetic mean of the thickness of all detection points of the wafer thin film, reflecting the overall thickness level of the thin film; the thickness standard deviation is used to measure the dispersion degree of the thickness of each detection point from the average thickness, reflecting the uniformity of the thickness; the maximum thickness deviation refers to the maximum difference between the thickness and the average thickness of all detection points, reflecting the extreme deviation of the thickness; the defect classification and positioning marking refers to marking the position, type and related parameters of the defect candidate region; the visual reminder refers to presenting the marked defect information in the form of intuitive graphics, colors, etc. on the display device.

[0072] The execution steps are as follows: based on the global film thickness distribution map, the local feature is analyzed, for example, it is found that the film thickness of a certain region deviates from the surrounding film thickness, for example, the film thickness suddenly decreases from 200 nm to 150 nm, forming an obvious local feature; at the same time, combined with the wafer surface reflection spectrum signal, if the reflection peak at 600 nm in the spectrum corresponding to the region suddenly shifts to 620 nm, it is determined as an abnormal point, and the region is identified as a defect candidate region by comprehensively considering the two; the film thickness gradient and the spectral absorption peak shift of the defect candidate region are extracted, and these data are taken as input to train the defect classifier; the defect classifier learns a large number of samples, and when the above parameters are input, it can output the defect type as local thin film thickness abnormal thinning.

[0073] The average thickness, thickness standard deviation and maximum thickness deviation of the wafer thin film are obtained through the global film thickness distribution map; combined with the identified defect candidate region and the corresponding local thin film thickness abnormal thinning type, the region is classified and positioned marked on the map, for example, the region position is marked with a yellow square, and the local thinning is marked, such as film thickness gradient 10 nm / μm, absorption peak shift 20 nm; the marked map is displayed on the screen in a visual way, and the color contrast and text description are used for reminding.

[0074] Through the above steps, a closed loop is achieved from film thickness detection to defect identification and presentation. By combining the global film thickness distribution map and spectral signal anomalies, defect candidate regions are accurately identified. The trained defect classifier can efficiently determine the defect type, while parameters such as average thickness and thickness standard deviation provide an overall reference for defect assessment. Defect classification, location labeling, and visualization alerts improve the efficiency of defect handling, effectively assisting in quality control during chip production and reducing yield losses caused by untimely defect handling.

[0075] In summary, the beneficial effects of the embodiments of this application are:

[0076] This application utilizes a laser source to emit incident light onto a wafer thin film, simultaneously acquiring the wafer surface reflection spectrum signal and extracting spectral feature information and wavelength parameters. Based on the polarization state of the incident light and combined with the wavelength parameters, polarization feature parameters in the wafer surface reflection spectrum signal are identified. Based on the interlayer interfaces of the wafer thin film and combined with spectral feature information, interface interference feature parameters in the wafer surface reflection spectrum signal are identified. Based on the polarization feature parameters and interface interference feature parameters in the wafer surface reflection spectrum signal, wavelength parameters are introduced as adjustment variables to determine wavelength dependence terms. The thickness value of each detection point of the wafer thin film is determined using the thin film refractive index corrected for wavelength dependence terms, generating a global film thickness distribution map. This application provides a fully automated wafer film thickness detection method and system based on spectral reflection. By extracting spectral features and wavelength parameters, and combining the polarization state of the incident light and the interlayer characteristics of the thin film, polarization feature parameters and interfacial interference feature parameters are identified. Wavelength parameters are introduced as adjustment variables to determine wavelength dependence. After correcting the refractive index of the thin film, the thickness of each detection point is determined. This effectively improves the accuracy of thickness detection, reduces the influence of spectral signal noise and interfacial interference, and achieves the technical effect of accurately presenting the global thickness distribution of wafer thin films.

[0077] Example 2, based on the same inventive concept as the fully automated wafer film thickness detection method based on spectral reflectance in the foregoing examples, such as... Figure 2 As shown in the embodiment of this application, a fully automated wafer film thickness detection system based on spectral reflectance is provided, wherein the system includes:

[0078] Feature extraction module M100: Uses a laser light source to emit incident light onto the wafer thin film, simultaneously acquires the reflection spectrum signal of the wafer surface, and extracts spectral feature information and wavelength parameters.

[0079] The characteristic parameter identification module M200: based on the polarization state of the incident light, in combination with the wavelength parameter, identify the polarization characteristic parameter in the wafer surface reflection spectrum signal; based on the interface between the thin film layers of the wafer thin film, in combination with the spectral characteristic information, identify the interface interference characteristic parameter in the wafer surface reflection spectrum signal.

[0080] The wavelength dependence determination module M300: based on the polarization characteristic parameter in the wafer surface reflection spectrum signal, the interface interference characteristic parameter in the wafer surface reflection spectrum signal, introduce the wavelength parameter as an adjusting variable to determine the wavelength dependence.

[0081] The thickness value determination module M400: according to the thin film refractive index corrected by the wavelength dependence, determine the thickness value of each detection point of the wafer thin film, and generate a global film thickness distribution map.

[0082] Further, the feature extraction module M100 is used to execute the following method:

[0083] The spectral characteristic information includes reflection peak intensity, reflection valley depth, characteristic peak position and spectral half-width, and the wavelength parameter covers the tunable wavelength range of the laser light source.

[0084] Further, the characteristic parameter identification module M200 is used to execute the following method:

[0085] The polarization characteristic parameter includes s-polarized light reflectivity, p-polarized light reflectivity and polarization state extinction ratio, the s-polarized light reflectivity and p-polarized light reflectivity are determined by collecting the reflection light intensity of the corresponding polarization direction through the polarization beam splitter, and the polarization state extinction ratio is the ratio of the s-polarized light reflectivity and the p-polarized light reflectivity.

[0086] Further, the characteristic parameter identification module M200 is used to execute the following method:

[0087] The interface interference characteristic parameter includes interlayer reflection peak position offset, interface scattering intensity factor and interference fringe contrast, the interlayer reflection peak position offset is determined by the spectral peak position difference formed by the reflection light of different thin film layer interfaces, the interface scattering intensity factor is positively correlated with the interface roughness, and is determined by conversion through the baseline drift of the reflection spectrum signal, and the interference fringe contrast is the difference value of the reflection peak intensity and the reflection valley depth.

[0088] Further, the thickness value determination module M400 is further used to execute the following method:

[0089] According to the wavelength-dependent item corrected thin film refractive index, the filter array corresponding to the light splitting member under the combination of the grating + prism is analyzed, and a full-band spectral feature matrix is set; the matching degree of the full-band spectral feature matrix is introduced, and only when the matching degree meets the convergence termination condition of the mean square error loss function, the multi-modal denoising network converges to the global optimal solution.

[0090] Further, the thickness value determination module M400 is also used to execute the following method:

[0091] The noise suppression of the wafer surface reflection spectrum signal adopts a multi-modal denoising network based on Transformer: the original spectrum signal and the wafer surface topography image are taken as dual-modal input, time series modeling is performed through a Transformer encoder, at the same time, edge features and texture features of the wafer surface topography image are extracted through a convolutional neural network, and then the dual-modal features are spliced and input into a full connection layer, a voltage ripple coefficient is determined using voltage closed-loop feedback data of a Cuk converter, and the voltage ripple coefficient is analyzed in correlation with a mean square error loss function, so as to set a convergence termination condition of the mean square error loss function by using spectrum peak shift error and reflectance measurement error.

[0092] Further, the thickness value determination module M400 is also used to execute the following method:

[0093] The voltage ripple coefficient is determined using voltage closed-loop feedback data of a Cuk converter, and a dynamic adjustment threshold of the Cuk converter is set: when it is detected that the spectrum peak shift error exceeds a preset threshold, the duty cycle adjustment of the Cuk converter is automatically triggered, the suppression of high-frequency ripple is enhanced by reducing the cutoff frequency of the LC filter network, so that the voltage ripple coefficient is controlled within a ripple tolerance threshold, and the reflection peak position measurement error is corrected to a peak position accuracy reference.

[0094] Further, the thickness value determination module M400 is also used to execute the following method:

[0095] For the system error of multi-batch wafer detection, the long-term voltage stability parameter of the Cuk converter is included in the error tracing system, when the detection error of any batch presents a systematic deviation, the influence weight of power supply stability on the thin film refractive index correction term is located by comparing the detection data in different voltage drift intervals, the calibration coefficient of the wavelength-dependent item is optimized; a double error compensation mechanism is set by the power supply stability control of the Cuk converter and the noise suppression of the multi-modal denoising network.

[0096] Further, the thickness value determination module M400 is also used to execute the following method:

[0097] Based on the local features of the global film thickness distribution map, in combination with the abnormal points of the wafer surface reflection spectrum signal, a defect candidate area is identified, a defect classifier is trained, the input is the film thickness gradient and the spectrum absorption peak offset of the defect candidate area, and the output is the defect type; the average thickness, thickness standard deviation and maximum thickness deviation of the wafer film are determined through the global film thickness distribution map, the defect classification positioning annotation is performed in combination with the defect candidate area and the corresponding defect type, and the defect classification positioning annotation is visualized for reminding.

[0098] In summary, any step can be stored as computer instructions or programs in an unrestricted computer memory and can be called and recognized by an unrestricted computer processor, and no redundant limitation is made here.

[0099] Further, the above technical solutions only represent the preferred technical solutions of the technical solutions of the embodiments of the present application, and some changes made by the person skilled in the art to some parts thereof all embody the principles of the novel embodiments of the present application. Obviously, the person skilled in the art can make various modifications and changes to the present application without departing from the scope of the present application.

Claims

1. A fully automatic wafer film thickness detection method based on spectral reflection, characterized in that, The method comprises: Using a laser light source to emit incident light to a wafer film, synchronously acquiring wafer surface reflection spectrum signals, extracting spectral feature information and wavelength parameters; Based on the polarization state of the incident light, combined with the wavelength parameters, the polarization characteristic parameters in the wafer surface reflection spectrum signals are identified; based on the interlayer interface of the wafer film, combined with the spectral feature information, the interface interference characteristic parameters in the wafer surface reflection spectrum signals are identified; Based on the polarization characteristic parameters in the wafer surface reflection spectrum signals, the interface interference characteristic parameters in the wafer surface reflection spectrum signals, and the wavelength parameters as an adjusting variable, a wavelength-dependent term is introduced to determine the thickness value of each detection point of the wafer film, and a global film thickness distribution map is generated; According to the film refractive index corrected by the wavelength-dependent term, the thickness value of each detection point of the wafer film is determined, and the method further comprises: According to the film refractive index corrected by the wavelength-dependent term, the filter array corresponding to the light splitting member under the grating+prism combination is analyzed, and a full-band spectral feature matrix is set; The matching degree of the full-band spectral feature matrix is introduced, and only when the matching degree meets the convergence termination condition of the mean square error loss function, the multi-modal denoising network converges to the global optimal solution. Extracting spectral feature information and wavelength parameters, the method comprises:

2. The full automatic wafer film thickness detection method based on spectral reflection according to claim 1, wherein, The spectral feature information includes reflection peak intensity, reflection valley depth, characteristic peak position and spectral half-width, and the wavelength parameter covers the tunable wavelength range of the laser light source. The polarization characteristic parameters include s-polarized light reflectivity, p-polarized light reflectivity and polarization state extinction ratio, the s-polarized light reflectivity and p-polarized light reflectivity are determined by collecting the reflection intensity of the corresponding polarization direction through a polarization beam splitter, and the polarization state extinction ratio is the ratio of the s-polarized light reflectivity to the p-polarized light reflectivity.

3. The full automatic wafer film thickness detection method based on spectral reflection according to claim 1, wherein, The interface interference characteristic parameters include interlayer reflection peak position offset, interface scattering intensity factor and interference fringe contrast, the interlayer reflection peak position offset is determined by the spectral peak position difference formed by the reflection light of different film layer interfaces, the interface scattering intensity factor is positively correlated with the interface roughness, and is determined by conversion through the baseline drift of the reflection spectrum signal, and the interference fringe contrast is the difference value ratio of the reflection peak intensity to the reflection valley depth.

4. The fully automatic wafer film thickness detection method based on spectral reflection according to claim 1, wherein, The method further comprises:

5. The fully automatic wafer film thickness detection method based on spectral reflection according to claim 1, wherein, The noise suppression of the wafer surface reflection spectrum signal adopts a multi-modal denoising network based on Transformer: the original spectrum signal and the wafer surface topography image are taken as dual-modal input, time series modeling is performed through a Transformer encoder, at the same time, edge features and texture features of the wafer surface topography image are extracted through a convolutional neural network, then the dual-modal feature splicing is performed, and then the full connection layer is input, the voltage ripple coefficient is determined by using the voltage closed-loop feedback data of the Cuk converter, the correlation analysis of the voltage ripple coefficient and the mean square error loss function is performed, and the convergence termination condition of the mean square error loss function is set by the spectral peak position offset error and the reflectivity measurement error. ​ 6. The full automatic wafer film thickness detection method based on spectral reflection according to claim 5, wherein, A method for determining a voltage ripple coefficient using voltage closed-loop feedback data of a Cuk converter, the method comprising: A method for determining a voltage ripple coefficient using voltage closed-loop feedback data of a Cuk converter, setting a dynamic adjustment threshold of the Cuk converter: when a spectral peak position offset error is detected to exceed a preset threshold, automatically triggering a duty cycle adjustment of the Cuk converter, enhancing the suppression of high-frequency ripples by reducing the cutoff frequency of the LC filter network, and controlling the voltage ripple coefficient within a ripple tolerance threshold, corresponding to a peak position measurement error correction to a peak position accuracy reference.

7. The fully automatic wafer film thickness detection method based on spectral reflection according to claim 6, wherein, The method comprises: Systematic errors in wafer detection of multiple batches, incorporating long-term voltage stability parameters of the Cuk converter into the error traceability system, when the detection error of any batch shows systematic deviation, by comparing the detection data of different voltage drift intervals, positioning the influence weight of power supply stability on the film refractive index correction term, and optimizing the calibration coefficient of the wavelength-dependent term; The power supply stability control of the Cuk converter and the noise suppression of the multi-modal denoising network set a double error compensation mechanism.

8. The fully automatic wafer film thickness detection method based on spectral reflection according to claim 1, wherein, The method further comprises: Based on the local features of the global film thickness distribution map, combined with the abnormal points of the wafer surface reflectance spectrum signal, identifying defect candidate regions, and training a defect classifier, the input is the film thickness gradient and the spectral absorption peak offset of the defect candidate region, and the output is the defect type; Through the global film thickness distribution map, the average thickness, thickness standard deviation and maximum thickness deviation of the wafer film are determined, combined with the defect candidate region and the corresponding defect type for defect classification and positioning labeling, and the defect classification and positioning labeling is visualized to remind.

9. A fully automatic wafer film thickness detection system based on spectral reflection, characterized in that, A system for implementing the steps of the fully automatic wafer film thickness detection method based on spectral reflectance according to any one of claims 1-8, the system comprising: A feature extraction module: using a laser light source to emit incident light to the wafer film, synchronously acquiring the wafer surface reflectance spectrum signal, and extracting spectral feature information and wavelength parameters; A feature parameter identification module: based on the polarization state of the incident light, combined with the wavelength parameters, identifying the polarization feature parameters in the wafer surface reflectance spectrum signal; based on the film layer interface of the wafer film, combined with the spectral feature information, identifying the interface interference feature parameters in the wafer surface reflectance spectrum signal; A wavelength-dependent term determination module: based on the polarization feature parameters in the wafer surface reflectance spectrum signal, the interface interference feature parameters in the wafer surface reflectance spectrum signal, and introducing the wavelength parameters as adjustment variables, the wavelength-dependent term is determined; A thickness value determination module: according to the film refractive index corrected by the wavelength-dependent term, the thickness values of each detection point of the wafer film are determined, and a global film thickness distribution map is generated.

Images