Film thickness measurement system and method based on differential confocal sensor
By introducing inclination angle measurement technology into the laser differential confocal sensor system, the inclination angle of the surface to be measured in real time, and combining the Gaussian hybrid model and the expectation maximization algorithm, the measurement error caused by inclination in laser differential confocal lens technology is solved. and the problem of limited range, improving the accuracy and range of ultra-thin material thickness measurement.
Patent Information
- Application Number
- CN202411139837.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-08-20
AI Technical Summary
When measuring the thickness of ultra-thin materials, the existing laser differential confocal lens technology increases measurement error and limits the measurement range due to the inclination of the measured part.
Using a film thickness measurement system based on differential confocal sensor, the inclination measurement technology is introduced, and laser, polarization beam splitter, 1/4 wave plate, objective positioner and microscopic objective lens assembly are used to combine the surface array camera and beam splitter lens , measure the inclination angle of the surface to be measured in real time, and separate the upper and lower surface differential confocal signals of the material to be measured through the Gaussian hybrid model and the expected maximization algorithm to calculate the film thickness.
The accuracy and range of ultra-thin material thickness measurement is improved, the time required for measurement is reduced, and the measurement error problem when the optical axis of the object to be measured cannot be completely perpendicular to the surface to be measured.
Smart Images

Figure CN118654585B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of optical measurement, and in particular relates to a film thickness measurement system and method based on a differential confocal sensor. Background Art
[0002] Ultra-thin materials are widely used in optoelectronics, architecture, biology, medicine, aerospace and other fields. The thickness measurement is of great significance for optimizing performance, meeting diverse application requirements and ensuring product quality.
[0003] At present, the thickness measurement of ultra-thin materials is mainly divided into two types of methods: contact and non-contact. Among them, contact measurement is mainly contact probe measurement, such as magnetic thickness measurement, eddy current thickness measurement and resistance thickness measurement. Although these measurement methods have high accuracy, they measure the thickness of materials by directly contacting the surface of the material, which is easy to scratch the material and destroy the surface finish, especially optical components and biological materials. At the same time, the wear of the probe will also affect the measurement accuracy and repeatability.
[0004] The development of measuring instruments has been moving towards precision, integration and intelligence. The process of high-end equipment manufacturing also puts forward higher requirements on the processing accuracy, measurement speed and accuracy of the workpiece thickness. Therefore, non-contact measurement is very necessary. Among them, optical measurement is used in the thickness measurement of many ultra-thin materials due to its advantages such as high precision, high resolution and visualization. For example, the ellipsometry method determines the thickness of the measured piece by measuring the change in the polarization state of light after reflection on the sample surface, but this method has high requirements on the sample surface, requires the optical constants of the material to be known, and the algorithm is complex when measuring complex multi-layer structures. The laser triangulation method uses a laser beam to illuminate the sample surface and calculates the thickness of the sample by detecting the position change of the reflected light, but its accuracy is affected by the surface reflection characteristics and its scope of application is limited. The fluorescence microscopy measurement method uses the fluorescent dye in the sample to emit light under excitation of a specific wavelength, and measures the thickness by detecting the intensity and distribution of the fluorescence signal, but this method requires pretreatment of the measured material (such as dyeing). Spectral interferometry is based on the interference phenomenon. It divides the light source into two beams: one beam is irradiated to the sample surface, and the other beam is used as a reference light. The two beams are recombined in the interferometer to produce interference fringes. The phase and fringe spacing of the interference fringes are related to the thickness of the sample surface. By analyzing the changes in the interference pattern, the thickness of the sample material can be accurately measured. Common spectral interferometry methods include Michelson interferometer and white light interferometer measurement. However, spectral interferometry relies on complex signal processing and data analysis, and requires a high-precision spectrometer. It is also sensitive to environmental conditions (such as temperature, vibration, air flow, etc.). Spectral confocal method is based on confocal microscopy technology. In the confocal optical system, the light source is irradiated onto the sample through a pinhole, and the reflected light is detected through another pinhole. Only the light at the focal position can pass through the second pinhole, so high longitudinal resolution can be achieved. The reflected light is decomposed into spectra of different wavelengths by a spectroscopic device (such as a grating or prism), and the spectrometer analyzes the spectral information of these wavelengths. Light of different wavelengths has different focal positions in the sample, and high-precision thickness measurement is achieved by analyzing the changes in the spectrum. Compared with confocal microscopy, laser differential confocal lens technology has higher high-axial tomography characteristics and calculates the center thickness of the material through ray tracing algorithms. However, in the fields of precision machining, optical component manufacturing and micro-nano technology, even a slight deviation in the tilt angle may affect the performance and reliability of the final product. When the measured object is tilted, it will inevitably lead to an increase in measurement errors, compressing the measurement range of the measured object's thickness. Summary of the invention
[0005] The present invention aims to solve the technical problems in the prior art that when using laser differential confocal lens technology to measure the thickness of ultra-thin materials, the measurement error increases and the measurement range is limited due to the tilt of the measured object, and provides a film thickness measurement system and method based on a differential confocal sensor. The present invention is based on differential confocal microscopy technology and introduces inclination measurement technology, which can measure the inclination of the surface to be measured in real time, solves the problem of large measurement errors caused when the optical axis of the measured object cannot be completely perpendicular to the measured surface, and further expands the measurement range of ultra-thin material thickness using differential confocal microscopy technology.
[0006] In order to solve the above technical problems, the technical solutions of the present invention are as follows:
[0007] A film thickness measurement system based on a differential confocal sensor includes, in the direction of the optical path, a laser, a polarization beam splitter, a quarter wave plate, an objective lens positioner, and a microscope objective lens assembly;
[0008] The parallel light beam emitted by the laser changes the polarization state of the light after passing through the polarization beam splitter and the 1 / 4 wave plate, and the polarized light is focused on the surface of the material to be tested through the objective lens positioner and the microscope objective lens assembly. The objective lens positioner and the microscope objective lens in the microscope objective lens assembly move along the axial direction of the microscope objective lens under the support of the objective lens positioner to realize axial scanning. After the light beam is reflected by the material to be tested, it is emitted from the pupil plane of the microscope objective lens. The emitted light beam is divided into two light beams by the first beam splitter, one light beam enters the area array camera for analyzing the tilt angle of the material to be tested; the other light beam is divided into two light beams by the second beam splitter, one light beam enters the post-focus pinhole detector after being focused by the first focusing lens, and the other light beam enters the pre-focus pinhole detector after being focused by the second focusing lens, and the differential confocal signal curves of the upper and lower surfaces of the material to be tested extracted by the pre-focus pinhole detector and the post-focus pinhole detector are separated to obtain the difference in the coordinates of the peak value of the light spot, and the film thickness of the material to be tested is calculated in combination with the tilt angle and the refractive index of the material to be tested.
[0009] In the above technical solution, the laser is a single-mode fiber laser, and its output light is monochromatic light with a wavelength of 642nm.
[0010] In the above technical solution, the light beam emitted from the pupil plane of the microscope objective is split into two light beams in a ratio of 1:1 by the first beam splitter.
[0011] In the above technical solution, the material to be tested is placed on a laboratory table.
[0012] In the above technical solution, the area array camera is a CCD camera.
[0013] A film thickness measurement method based on a differential confocal sensor, the applicable system is the film thickness measurement system based on a differential confocal sensor of the present invention, and the measurement method comprises the following steps:
[0014] Step 1: Process the spot image received by the array camera, calculate the spot centroid, perform angle recognition through the mapping relationship table of spot distribution and tilt angle calibrated in advance, and obtain the real-time tilt angle;
[0015] Step 2: Use Gaussian mixture model (GMM) and expectation maximization (EM) algorithm to separate the overlapping differential confocal signal curves of the upper and lower surfaces of the material to be tested extracted by the pre-focus pinhole detector and the post-focus pinhole detector, obtain the difference in the peak coordinates of the spot, and calculate the film thickness of the material to be tested by combining the tilt angle obtained in step 1 and the refractive index of the material to be tested.
[0016] In the above technical solution, in step 1, the density-based clustering (DBSCAN) algorithm is used to calculate the centroid of the light spot.
[0017] In the above technical solution, in step 1, the steps of obtaining the tilt angle are as follows:
[0018] According to the calculated spot centroid, the tilt angle is obtained by analyzing the peak position of multiple spot images. , Spot peak coordinates , the peak coordinate of the spot With tilt angle Use polynomial fitting between and define the function , so that for any input spot peak coordinate , predict the corresponding tilt angle represents the predicted tilt angle;
[0019] The expression of the defined polynomial regression model is:
[0020] ;
[0021] The polynomial regression model is composed of multiple spot peak coordinates composed of, is the peak coordinate of the light spot, for The order of for The order of is the highest order polynomial, are the parameters of the model;
[0022] Convert the original data into polynomial features, is the input feature matrix, Every behavior of ], through polynomial feature transformation, a new polynomial feature matrix is obtained , each of its rows is [1, , , ,......];
[0023] Then, a polynomial regression model is used to fit the polynomial feature matrix Angle of inclination to target value , the loss function of the polynomial regression model is the least squares error, and the loss function expression is as follows:
[0024] ;
[0025] In the formula, represents the loss function, For the fitting sample samples, is the number of samples, For the The true tilt angle of the samples is For the The predicted tilt angle of samples;
[0026] When the loss function is minimized, there is a unique solution. The actual tilt angle of the samples With predicted tilt angle The difference between the two is the smallest. The parameters of each item in It is determined;
[0027] Finally, according to the expression of the polynomial regression model, the parameters Value and spot peak coordinates Substitute it and get the corresponding predicted tilt angle , the predicted tilt angle The actual tilt angle .
[0028] In the above technical solution, step 2 specifically includes the following steps:
[0029] The formula of the Gaussian mixture model is as follows:
[0030] ;
[0031] in, The Gaussian distribution is represented by Gaussian components, Representative The weights of the Gaussian components, Representative The mean of the Gaussian components, It is The covariance matrix of the Gaussian components, For the The multivariate normal density function has Gaussian components, is the vector of input data points;
[0032] Define a model for a given input data and find Gaussian mixture model parameters that explain the observed data, wherein the Gaussian mixture model parameters include mean, covariance, and mixture weights;
[0033] For a Gaussian mixture model with K components, the dataset , n represents the number of data points, and the log-likelihood function L is given by the product of the probability density of each data point, which is defined by the Gaussian mixture model as follows:
[0034] ;
[0035] in, Represents the set of parameters of the Gaussian mixture model, including the mean, covariance and mixture weights of all Gaussian components, Represents the number of data sets, Representative The weights of the Gaussian components, For the The first The multivariate normal density function has Gaussian components, Representative The mean of the Gaussian components, It is The covariance matrix of the Gaussian components, Represents a data set, which is a set of n data points A collection of input data, each data point Represents a vector, representing a point in the data space; represents the number of Gaussian components in the Gaussian mixture model, represents the log-likelihood function, represents the likelihood function;
[0036] Using the expectation-maximization algorithm Maximize the log-likelihood function to estimate, first randomly initialize the model parameters, and then iterate between the following two steps;
[0037] Step 1: Expected Step:
[0038] Compute the expected log-likelihood of the model with respect to the distribution of the latent variables given the observed data and the current estimates of the model parameters;
[0039] Step 2: Maximization step:
[0040] Update the parameters of the model to maximize the log-likelihood of the observed data, given the latent variables estimated in step 1;
[0041] Update the model parameters as follows:
[0042] a. Update each component:
[0043] ;
[0044] in, is the posterior probability of each Gaussian component, Representative data sets, n represents the number of data sets, Represents the updated The mean of the Gaussian components;
[0045] b. Update the covariance of each component:
[0046] ;
[0047] in, Represents the updated The covariance of the Gaussian components;
[0048] c. Update the mixing weights:
[0049] ;
[0050] in, Represents the updated The weights of the Gaussian components;
[0051] Repeating steps 1 and 2 to ensure convergence to the local maximum of the log-likelihood function, the superposition distribution of the differential confocal signal curves on the upper and lower surfaces of the material to be tested can be separated into and Two spot peaks, Represents the peak value of the light spot on the surface of the material to be tested. Represents the peak value of the spot on the lower surface of the material to be tested. The pre-focus optical path difference is obtained according to the peak coordinates of the spot. , the optical path difference after focusing is , the optical path difference is obtained by averaging ;
[0052] Finally, the film thickness of the material to be tested is calculated according to the following formula: ;
[0053] ;
[0054] in, represents the optical path difference, is the refractive index of the material to be tested, Represents the object's tilt angle relative to the principal optical axis.
[0055] The beneficial effects of the present invention are:
[0056] The thin film thickness measurement system based on the differential confocal sensor of the present invention uses a spectroscope to divide the light beam of the test piece into two parts. One part is received by an array camera such as a CCD camera, and the image processing is performed on the spot image received by the CCD camera to calculate the centroid of the spot. The angle recognition is performed through the mapping relationship between the spot distribution and the tilt angle that has been calibrated in advance to obtain the real-time tilt angle; another part of the light beam is received by a photosensitive element, and the algorithm is used to separate the overlapping differential confocal signal curves of the upper and lower surfaces of the test piece to obtain the difference in the coordinates of the spot peak, and the thickness of the ultra-thin material is accurately calculated by combining the tilt angle and the refractive index of the test piece. Therefore, the present invention is based on differential confocal microscopy technology and introduces the tilt angle measurement technology, which can calibrate the tilt angle in real time and reduce the time required for measurement. Introducing the tilt angle into the thickness calculation of ultra-thin materials further improves the accuracy of ultra-thin material thickness measurement and expands the measurement range of ultra-thin material thickness.
[0057] The thin film thickness measurement method based on the differential confocal sensor of the present invention is based on the differential confocal microscopy technology and introduces the inclination measurement technology. It can measure the inclination of the surface to be measured in real time, solves the problem of large measurement errors caused when the optical axis of the object to be measured cannot be completely perpendicular to the surface to be measured, and further expands the measurement range of the differential confocal microscopy technology for the thickness of ultra-thin materials. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0059] Figure 1 Schematic diagram of the structure of the film thickness measurement system based on differential confocal sensor.
[0060] Figure 2 This is the internal optical path diagram when the device under test is tilted.
[0061] Figure 3 Schematic diagram of the deflection of the reflected light beam when the surface to be measured is tilted.
[0062] Figure 4 This is a diagram of overlapping peaks and the effect after separation.
[0063] The reference numerals in the figures indicate:
[0064] 1-laser, 2-polarization beam splitter, 3-1 / 4 wave plate, 4-objective positioner and microscope objective assembly, 5-material to be tested, 6-laboratory table, 7-area array camera, 8-first beam splitter, 9-second beam splitter, 10-first focusing lens, 11-second focusing lens, 12-post-focus pinhole detector, 13-pre-focus pinhole detector. DETAILED DESCRIPTION
[0065] The inventive concept of the present invention is: the optical measurement methods and improved confocal microscope technology in the prior art have solved the measurement needs of ultra-thin material thickness in many fields, but they still have some problems that need to be solved, such as pre-treatment of the measured object, the need for a better measurement environment, and the need for the measured object to maintain a certain angle with the optical axis. To this end, in view of the above-mentioned problems in measuring the thickness of ultra-thin materials, the present invention introduces inclination measurement technology based on differential confocal microscopy technology, which can measure the inclination of the surface to be measured in real time, solves the problem of large measurement errors caused when the optical axis of the object to be measured is not completely perpendicular to the surface to be measured, and further expands the measurement range of ultra-thin material thickness by differential confocal microscopy technology.
[0066] The present invention adopts improved differential confocal microscope technology to solve the problem of increased measurement error caused by tilt of the test piece and the problem of expanding the thickness measurement range of ultra-thin materials. A beam splitter is used to divide the light beam of the test piece into two parts, one of which is received by a planar array camera such as a CCD camera, and the image of the spot image received by the CCD camera is processed to calculate the centroid of the spot. The angle is identified by the mapping relationship between the pre-calibrated spot distribution and the tilt angle to obtain the real-time tilt angle; another part of the light beam is received by a photosensitive element, and the algorithm is used to separate the differential confocal signal curves of the upper and lower surfaces of the overlapped test piece to obtain the difference in the peak coordinates of the spot, and the thickness of the ultra-thin material is accurately calculated by combining the tilt angle and the refractive index of the test piece.
[0067] The thin film thickness measurement system based on the differential confocal sensor of the present invention has a main structure that after the parallel light beam emitted from the laser passes through a polarization beam splitter and a 1 / 4 wave plate, the emitted parallel light passes through a microscope objective probe and is focused on the surface of the object to be measured. The microscope objective can be moved along the axis of the microscope objective under the support of an objective positioner to achieve axial scanning. Since the object to be measured is an ultra-thin material, the light beam will be transmitted inside the material. Both the upper surface at the focus and the lower surface at the defocus point will be reflected. After being reflected by the material to be measured, the light beam is emitted from the pupil plane of the microscope objective and is divided into two light beams at a ratio of 1:1 by a non-polarization beam splitter. One light beam enters a planar array camera for analyzing the tilt angle of the object to be measured. After passing through the non-polarization beam splitter, one light beam enters the pre-focus and post-focus pinhole detectors through a focusing lens respectively for analyzing the thickness of the material to be measured.
[0068] The present invention is described in detail below with reference to the accompanying drawings.
[0069] Combination Figure 1 Specifically, the film thickness measurement system based on the differential confocal sensor of the present invention includes, in the direction of the optical path, a laser 1, a polarization beam splitter 2, a quarter wave plate 3, and an objective lens positioner and a microscope objective lens assembly 4; the laser 1 is a single-mode fiber laser, and its output light is monochromatic light with a wavelength of 642 nm.
[0070] The output light of the single-mode fiber laser 1 is monochromatic light with a wavelength of 642 nm. After the parallel light is emitted, the parallel light beam passes through the polarization beam splitter 2 and the 1 / 4 wave plate 3 to change the polarization state of the light. The polarized light is focused on the surface of the material to be tested 5 through the objective lens positioner and the microscope objective lens assembly 4. The material to be tested 5 is placed on the experimental table 6. The objective lens positioner and the microscope objective lens in the microscope objective lens assembly 4 move along the axial direction of the microscope objective lens under the support of the objective lens positioner to achieve axial scanning. After the light beam is reflected by the material to be tested 5, it is emitted from the pupil plane of the microscope objective lens. The emitted light beam is divided into two light beams by the first beam splitter 8 at a ratio of 1:1. One light beam enters the area array camera 7. The area array The camera 7 is a CCD camera, which is used to analyze the tilt angle of the material to be tested 5; the other beam is divided into two beams by the second beam splitter 9, one of which is focused by the first focusing lens 10 and enters the post-focus pinhole detector 12, and the other beam is focused by the second focusing lens 11 and enters the pre-focus pinhole detector 13. The overlapping differential confocal signal curves of the upper and lower surfaces of the material to be tested 5 extracted by the pre-focus pinhole detector 13 and the post-focus pinhole detector 12 are separated to obtain the difference in the coordinates of the light spot peaks, and the film thickness of the material to be tested 5 is calculated in combination with the tilt angle and the refractive index of the material to be tested 5. To calculate the film thickness, it is necessary to separate the overlapping differential confocal signal curves of the upper and lower surfaces of the material to be tested 5 extracted before and after focusing, respectively. The overlapping signal belongs to the Gaussian mixture model (GMM). The two Gaussian distributions of the upper and lower surfaces of the material to be tested 5 are superimposed together. The Gaussian mixture model (GMM) and the expectation maximization (EM) algorithm are used for separation, and the film thickness of the material to be tested 5 is calculated using the formula.
[0071] The film thickness measurement method based on the differential confocal sensor of the present invention is applicable to the film thickness measurement system based on the differential confocal sensor of the present invention, and comprises the following steps:
[0072] Step 1: Process the spot image received by the array camera 7, calculate the spot centroid, perform angle recognition through the mapping relationship between the spot distribution and the tilt angle calibrated in advance, and obtain the real-time tilt angle;
[0073] Step 2: Use Gaussian mixture model (GMM) and expectation maximization (EM) algorithm to separate the overlapping differential confocal signal curves of the upper and lower surfaces of the material to be tested 5 extracted by the pre-focus pinhole detector 13 and the post-focus pinhole detector 12, respectively, to obtain the difference in the peak coordinates of the spot, and combine the tilt angle obtained in step 1 and the refractive index of the material to be tested 5 to calculate the film thickness of the material to be tested 5.
[0074] In the above measurement method, it is necessary to separate the overlapping differential confocal signal curves of the upper and lower surfaces of the material to be tested 5 extracted before and after focusing, respectively. The overlapping signal belongs to the Gaussian mixture model (GMM). The two Gaussian distributions of the upper and lower surfaces of the material to be tested 5 are superimposed together. The Gaussian mixture model (GMM) and the expectation maximization (EM) algorithm are used for separation to obtain the difference in the coordinates of the light spot peaks, and the film thickness of the material to be tested 5 is calculated using a formula in combination with the obtained tilt angle and the refractive index of the material to be tested 5. The following specifically describes the tilt angle calculation process of step 1 and the film thickness calculation process of the material to be tested 5 in step 2.
[0075] (1) Establish a fitting table for the mapping relationship between the main light spot coordinates and the tilt angle;
[0076] When the measuring beam is focused on the tilted surface to be measured, the deflection of the reflected beam is as follows: Figure 3 As shown in the figure, when the surface to be measured has no tilt angle and is placed horizontally at the focal position of the sensor, the focused light beam of the sensor is reflected at the focal position, and the light beam returning to the microscope objective is re-collimated and emitted from the exit pupil surface of the microscope objective. Since the incident light path and the exit light path are completely symmetrical, the light beam incident at an angle of α is still reflected at an angle of α after being reflected by the surface to be measured, and re-enters the microscope objective. Therefore, the distribution of the collimated light beam emitted from the exit pupil surface of the microscope objective is also consistent with the incident light beam, and still maintains a centrally symmetrical Gaussian distribution.
[0077] When the surface to be measured is tilted, there is a certain angle between the normal line of the surface to be measured and the central axis of the incident light beam. ,by The beam incident at an angle of The reflected light is reflected at an angle and re-enters the microscope objective. When the inclination angle of the surface to be measured is too large, the reflected light beam will be blocked by the aperture diaphragm of the microscope objective, and the spot distribution on the exit pupil surface of the microscope objective will no longer be a Gaussian distribution symmetrical about the central axis. Taking advantage of the fact that the light spot on the exit pupil surface of the microscope objective will present different distribution characteristics when the surface to be measured is at different inclination angles, an area array camera is introduced into the system structure for inclination measurement, and the angle is identified through the mapping relationship between the spot distribution and the inclination angle. As the surface to be measured tilts, the reflected light enters the microscope objective from different positions, causing the shape of the spot image captured by the area array camera to change. By analyzing the correspondence between the spot shape and the inclination angle, accurate measurement of the inclination angle of the surface to be measured can be achieved.
[0078] In order to analyze the spot distribution on the exit pupil of the microscope objective when the surface to be measured is tilted, and to effectively simulate it, a model is built based on the Huygens-Fresnel diffraction integral formula. The collimated Gaussian beam is focused on the surface to be measured, reflected from the tilted surface to be measured, and re-collimated by the microscope objective. The model can simulate the reflection effect of the beam after the surface to be measured is tilted by adjusting the light field phase of the beam focused on the surface to be measured and changing the wavefront shape.
[0079] Light field distribution on the pupil surface of the microscope objective in the emission direction :
[0080]
[0081]
[0082]
[0083] in, is a constant, The tilt angle of the surface to be measured, is the inclination direction of the surface to be measured, is the pupil function of the microscope objective on the detection path, is the focal length of the objective lens, is the defocus amount, It is the light field distribution before reaching the material under test for reflection; is plural, is the wave number, expressed as 2π / λ, where λ is the wavelength of light;
[0084] The light spot images at different tilt angles are generated by the above formula, which shows that as the tilt angle of the surface to be measured changes, the shape of the light spot image also changes.
[0085] The spot image is composed of multiple spots superimposed after reflection from the upper and lower surfaces of the material to be tested. It is necessary to process the collected spot image, separate the spots, and obtain the important spot coordinate information close to the center;
[0086] Density-based clustering (DBSCAN) algorithm, which uses the density connectivity of classes to quickly discover classes of arbitrary shapes. It can discover clusters of arbitrary shapes, identify noise points, and exclude them from clusters. Instead of assigning all data points to a cluster like the hierarchical clustering algorithm, it does not need to specify the number of clusters in advance, but automatically determines the number of clusters based on the density of the data.
[0087] The steps of the DBSCAN algorithm are as follows:
[0088] a. From the sample set Randomly select an unlabeled sample object that is not included in any set ,calculate of Neighborhood, the formula is as follows:
[0089] ;
[0090] in, represents a dataset, Yes and The distance between points.
[0091] If the following formula is satisfied, then the point is a core point; otherwise, it is marked as a non-core point and other points are selected for discrimination. Repeat the above steps until a core point is found. If not found, add the unclassified samples to the noise set.
[0092] ;
[0093] in, Represents the minimum number of points, indicating the minimum number of neighbors required for a point to become a core point, including the point itself. Specifically, if the point of Number of points in the neighborhood At least , then Considered a core point.
[0094] b. At the core point Cluster ,Will All points in the neighborhood are added to the cluster ;
[0095] c. Iteratively examine all unlabeled points in the neighborhood and expand the cluster. If a neighborhood point As a core point, the points in its area that are not included in the set are added to the cluster middle;
[0096] d. Repeat the above steps until all samples are classified into the specified set;
[0097] e. Output clusters C1, C2, ..., Cn and noise set O; calculate the centroid of the light spot :
[0098] ;
[0099] According to the calculated spot centroid, a large number of tilt angles and spot peak coordinates are obtained by analyzing the peak positions of these spot images. By establishing a good fitting relationship, a mapping relationship between the tilt angle and the spot peak position can be established. With tilt angle Use polynomial fitting to find a function , so that for any input , the corresponding tilt angle can be predicted represents the predicted tilt angle;
[0100] The expression of the defined polynomial regression model is:
[0101] ;
[0102] The polynomial regression model is composed of multiple spot peak coordinates composed of, is the peak coordinate of the light spot, for The order of for The order of is the highest order polynomial, are the parameters of the model; in order to perform polynomial regression, we first need to convert the original data into polynomial features. is the input feature matrix, Every behavior of ]. Through polynomial feature transformation, we get a new polynomial feature matrix , each of its rows is [1, , , ,......].
[0103] Use a polynomial regression model to fit a polynomial feature matrix Angle of inclination to target value The loss function of the polynomial regression model is the least squares error, and the loss function expression is as follows:
[0104] ;
[0105] In the formula, represents the loss function, For the fitting sample samples, is the number of samples, For the The true tilt angle of the samples is For the The predicted tilt angle of samples, the purpose of the loss function is to make the actual tilt angle With predicted tilt angle The difference between them is the smallest;
[0106] When the loss function is minimized, there is a unique solution. The actual tilt angle of the samples With predicted tilt angle The difference between the two is the smallest. The parameters of each item in The value is determined;
[0107] Finally, according to the expression of the polynomial regression model, the parameters and the spot peak coordinates Substitute it and get the corresponding predicted tilt angle , then the predicted tilt angle at this time It can be considered as the actual tilt angle .
[0108] (2) Separate the overlapping signals before and after focusing;
[0109] The light intensity signals received by the pre-focus pinhole detector 13 and the post-focus pinhole detector 12 will cause overlapping peaks due to the reflection from the upper and lower surfaces of the material 5 to be tested. This phenomenon can be attributed to the separation of the Gaussian mixture model (GMM) problem.
[0110] The Gaussian mixture model (GMM) formula is as follows:
[0111] ;
[0112] in, The Gaussian distribution is represented by Gaussian components, Representative The weights of the Gaussian components; Representative The mean of the Gaussian components, It is The covariance matrix of the Gaussian components, For the The multivariate normal density function has Gaussian components, is the vector of input data points;
[0113] We want to find the GMM parameters, including the mean, covariance, and mixing weights, that explain the observed data; to do this, we first need to define the likelihood of the model given the input data.
[0114] For a GMM with K components, the dataset , n represents the number of data points, and the log-likelihood function L is given by the product of the probability density of each data point, which is defined by GMM as follows:
[0115] ;
[0116] in, Represents the set of parameters of the Gaussian mixture model, including the mean, covariance and mixture weights of all Gaussian components, Represents the number of data sets, Representative The weights of the Gaussian components, For the The first The multivariate normal density function has Gaussian components, Representative The mean of the Gaussian components, It is The covariance matrix of the Gaussian components, Represents a data set, which is a set of n data points A collection of input data, each data point Represents a vector, representing a point in the data space; represents the number of Gaussian components in the Gaussian mixture model. The GMM parameter is is a linear combination of Gaussian components, represents the log-likelihood function, represents the likelihood function.
[0117] The parameters of GMM can be obtained by Maximize the log-likelihood function to estimate. The log-likelihood function is highly nonlinear and difficult to maximize analytically. Use the expectation-maximization (EM) algorithm to solve this problem.
[0118] The EM algorithm is a powerful method for finding maximum likelihood estimates of parameters in statistical models that depend on unobserved latent variables.
[0119] The EM algorithm first randomly initializes the model parameters; then it iterates between the following two steps:
[0120] 1. Expected step (e step):
[0121] Compute the expected log-likelihood of the model with respect to the distribution of the latent variable given the observed data and the current estimates of the model parameters; this step includes estimating the probability of the latent variable.
[0122] 2. Maximization step (m-step): Update the parameters of the model to maximize the log-likelihood of the observed data, given the latent variables estimated in the e-step.
[0123] Update the model parameters as follows:
[0124] a. Update each component:
[0125] ;
[0126] in, is the posterior probability of each Gaussian component, Representative data sets, n represents the number of data sets, Represents the updated The mean of the Gaussian components;
[0127] b. Update the covariance of each component:
[0128] ;
[0129] in, Represents the updated The covariance of the Gaussian components, the other parameters have the same meanings as above;
[0130] c. Update the mixing weights:
[0131] ;
[0132] in, Represents the updated The weight of the Gaussian component, the other parameters have the same meaning as above;
[0133] Repeating these two steps ensures convergence to the local maximum of the likelihood function.
[0134] After the above two steps, Figure 4 The superposition distribution in is separated into and Two spot peaks, represents the peak value of the light spot on the surface of the material to be tested 5, Represents the peak value of the spot on the lower surface of the material to be tested 5. According to the coordinates of the peak value of the spot, the pre-focus difference can be obtained. , the post-focus difference is , the average processing results in the optical path difference D = .
[0135] (3) Calculate the film thickness of the material to be tested 5 :
[0136] See also Figure 2 When the material to be tested 5 is tilted, part of the light emitted by the optical axis is reflected by the upper surface of the material to be tested 5 through reflected light 1, and part of it is transmitted, reflected by the lower surface of the material to be tested 5, and then emitted through reflected light 2.
[0137] The incident light angle is , the object's inclination angle relative to the principal optical axis is , the refractive index is , the optical path difference calculated in the previous step is , the film thickness of the material 5 to be tested can be calculated according to the following formula :
[0138] ;
[0139] In summary, the thin film thickness measurement system and method based on the differential confocal sensor of the present invention is based on differential confocal microscopy technology and introduces inclination measurement technology, which can measure the inclination of the surface to be measured in real time, solves the problem of large measurement errors caused when the optical axis of the object to be measured cannot be completely perpendicular to the surface to be measured, and further expands the measurement range of differential confocal microscopy technology for the thickness of ultra-thin materials.
[0140] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.
Claims
1. A film thickness measurement method based on a differential confocal sensor, characterized in that: The applicable system comprises, in the direction of the optical path, a laser (1), a polarization beam splitter (2), a quarter wave plate (3), and an objective lens positioner and a microscope objective lens assembly (4); The parallel light beam emitted by the laser (1) changes the polarization state of the light after passing through the polarization beam splitter (2) and the quarter wave plate (3). The polarized light is focused on the surface of the material to be tested (5) through the objective lens positioner and the microscope objective lens assembly (4). The objective lens positioner and the microscope objective lens in the microscope objective lens assembly (4) move along the axial direction of the microscope objective lens under the support of the objective lens positioner to achieve axial scanning. After the light beam is reflected by the material to be tested (5), it is emitted from the pupil plane of the microscope objective lens. The emitted light beam is split into two light beams by the first beam splitter (8). One light beam enters the area array camera (7) and is used to analyze the material to be tested ( 5); the other beam is divided into two beams by a second beam splitter (9), one beam is focused by a first focusing lens (10) and then enters a post-focus pinhole detector (12), and the other beam is focused by a second focusing lens (11) and then enters a pre-focus pinhole detector (13), the overlapped differential confocal signal curves of the upper and lower surfaces of the material to be tested (5) extracted by the pre-focus pinhole detector (13) and the post-focus pinhole detector (12) are separated, the difference in the coordinates of the light spot peak is obtained, and the film thickness of the material to be tested (5) is calculated in combination with the tilt angle and the refractive index of the material to be tested (5); The measuring method comprises the following steps: Step 1: Process the spot image received by the array camera (7), calculate the spot centroid, perform angle recognition based on the mapping relationship between the spot distribution and the tilt angle calibrated in advance, and obtain the real-time tilt angle; The density-based clustering algorithm is used to calculate the centroid of the light spot. According to the calculated centroid of the light spot, the peak position of multiple light spot images is analyzed to obtain the tilt angle θ and the light spot peak coordinates (x, y). The light spot peak coordinates (x, y) and the tilt angle θ are fitted using a polynomial, and the function F(x, y) is defined so that for any input light spot peak coordinates (x, y), the corresponding tilt angle is predicted. represents the predicted tilt angle; The expression of the defined polynomial regression model is: The polynomial regression model is composed of multiple spot peak coordinates (x, y), where x, y are the spot peak coordinates, j is the order of x, k is the order of y, d is the highest order of the polynomial, and a jk are the parameters of the model; Convert the original data into polynomial features. Let X be the input feature matrix, and each line of X is [x_i, y_i]. Through polynomial feature conversion, a new polynomial feature matrix X_poly is obtained, and each line of it is [1, x_i, y_i, x_i^2, ...]; Then, a polynomial regression model is used to fit the polynomial feature matrix X_poly and the target value inclination angle θ. The loss function of the polynomial regression model is the least squares error, and the loss function expression is as follows: In the formula, L represents the loss function, i is the i-th sample in the fitting sample, n is the number of samples, and θ i is the true tilt angle of the i-th sample, is the predicted tilt angle of the i-th sample; When the loss function is minimized, there is a unique solution, where the true tilt angle θ of the i-th sample is i With predicted tilt angle The difference between the two is the smallest. The parameter a of each item in jk The value is determined; Finally, according to the expression of the polynomial regression model, the parameter a jk Substitute the value and the spot peak coordinates (x, y) to get the corresponding predicted tilt angle The predicted tilt angle That is the actual tilt angle θ; Step 2: using a Gaussian mixture model and an expectation maximization algorithm to separate the overlapping differential confocal signal curves of the upper and lower surfaces of the material to be tested (5) extracted by the pre-focus pinhole detector (13) and the post-focus pinhole detector (12), respectively, to obtain the difference in the coordinates of the light spot peaks, and combining the tilt angle obtained in step 1 and the refractive index of the material to be tested (5), to calculate the film thickness of the material to be tested (5); The formula of the Gaussian mixture model is as follows: Among them, k represents the Gaussian distribution composed of k Gaussian components, ω k represents the weight of the kth Gaussian component, μ k represents the mean of the kth Gaussian component, ∑ k is the covariance matrix of the kth Gaussian component, N(x; μ k ,∑ k ) is the multivariate normal density function of the kth Gaussian component, and x is the vector of input data points; Define a model for a given input data and find Gaussian mixture model parameters that explain the observed data, wherein the Gaussian mixture model parameters include mean, covariance, and mixture weights; For a Gaussian mixture model with K components, the dataset X = {x1, .., x n }, n represents the number of data points, and the log-likelihood function is given by the product of the probability density of each data point, which is defined by the Gaussian mixture model as follows: Among them, β represents the parameter set of the Gaussian mixture model, including the mean, covariance and mixing weight of all Gaussian components, n represents the number of data sets, ω k represents the weight of the kth Gaussian component, N(x i ;μ k ,∑ k ) is the multivariate normal density function of the kth Gaussian component of the ith data set, μ k represents the mean of the kth Gaussian component, Σ k is the covariance matrix of the kth Gaussian component, X represents the data set, which is a matrix consisting of n data points x i A collection of input data, each data point x i represents a vector, indicating a point in the data space; K represents the number of Gaussian components in the Gaussian mixture model, l(β|X) represents the log-likelihood function, and L(β|X) represents the likelihood function; The expectation maximization algorithm is used to estimate the log-likelihood function of β by first randomly initializing the model parameters and then iterating between the following two steps; Step 1: Expected Step: Compute the expected log-likelihood of the model with respect to the distribution of the latent variables given the observed data and the current estimates of the model parameters; Step 2: Maximization step: Update the parameters of the model to maximize the log-likelihood of the observed data, given the latent variables estimated in step 1; Update the model parameters as follows: a. Update each component: in, is the posterior probability of each Gaussian component, x i represents the i-th data set, n represents the number of data sets, represents the mean of the updated k-th Gaussian component; b. Update the covariance of each component: in, represents the covariance of the updated k-th Gaussian component; c. Update the mixing weights: in, Represents the weight of the updated k-th Gaussian component; Repeating steps 1 and 2 to ensure convergence to the local maximum of the log-likelihood function, the superimposed distribution of the differential confocal signal curves of the upper and lower surfaces of the material to be tested (5) can be separated into two spot peaks x1 and x2, wherein x1 represents the spot peak of the upper surface of the material to be tested (5), and x2 represents the spot peak of the lower surface of the material to be tested (5). According to the spot peak coordinates, the optical path difference before focus is D1, and the optical path difference after focus is D2. The optical path difference D=(D1+D2) / 2 is obtained by averaging. Finally, the film thickness d of the material to be tested (5) is calculated according to the following formula; Wherein, D represents the optical path difference, n is the refractive index of the material to be measured (5), and θ represents the inclination angle of the object relative to the main optical axis.
2. The film thickness measurement method based on differential confocal sensor according to claim 1, characterized in that: The laser (1) is a single-mode fiber laser, and its output light is monochromatic light with a wavelength of 642 nm.
3. The film thickness measurement method based on differential confocal sensor according to claim 1, characterized in that: The light beam emitted from the pupil plane of the microscope objective is split into two light beams at a ratio of 1:1 by a first beam splitter (8).
4. The film thickness measurement method based on differential confocal sensor according to claim 1, characterized in that: The material to be tested (5) is placed on a test bench (6).
5. The film thickness measurement method based on differential confocal sensor according to claim 1, characterized in that: The area array camera (7) is a CCD camera.
