Apparatus and method for non-destructive measurement of semiconductor wafer surface temperature for rapid thermal processing

Abstract

This invention discloses a non-destructive measurement device and method for the surface temperature of semiconductor wafers used in rapid thermal processing. The device includes a processing cavity, a heating system, a Mueller matrix elliptic thermometry system, and a processing unit. The wafer to be tested is placed in the processing cavity, and the heating system is located at the bottom of the processing cavity for rapid thermal processing of the wafer. The Mueller matrix elliptic thermometry system is located at the top of the processing cavity and includes a light source, a polarizer, a first compensator, a beam splitter, multiple temperature-measuring light tubes, multiple angle-modulation modules, multiple first reflectors, and a multi-channel spectrometer. The processing unit performs a Fourier series expansion of the total light intensity of the multi-wavelength polarization response signals collected by the multi-channel spectrometer, solves for the Fourier coefficients of each order, obtains the Mueller matrix spectrum at different wavelengths, and inverts the wafer surface temperature based on the Mueller matrix spectrum. This invention can effectively solve the problem of inaccurate measurement in the low-temperature range of infrared pyrometers and improve the stability and accuracy of temperature measurement.

Description

Technical Field

[0001] This invention belongs to the field of semiconductor metrology and optical temperature measurement, specifically relating to a non-destructive measurement device and method for the surface temperature of semiconductor wafers used in rapid thermal processing. Background Technology

[0002] Rapid thermal processing (RTP) equipment refers to manufacturing systems specifically designed for the semiconductor industry, used to rapidly heat and cool semiconductor wafers in a controlled environment for short periods (typically seconds to minutes). Currently, there is a huge market demand for RTP equipment. Semiconductor chip nodes below 90nm and high-bandwidth memory (HBM) chips all require RTP equipment to complete high-speed, high-uniformity annealing and oxidation processes. Because the wafer rotates continuously during RTP operation, contact temperature measurement (such as thermocouples) cannot meet the requirements of online non-destructive testing applications. The commonly used method is non-contact temperature measurement (such as infrared pyrometers). However, during RTP, infrared pyrometers cannot accurately measure temperatures in the low-temperature range (80℃ to 400℃). Within this temperature range, silicon wafers are transparent in the infrared band (imaginary part k≈0), with an emissivity ε≈0 and extremely low radiation intensity, leading to unstable wafer surface temperature measurements (fluctuations of approximately 10℃). This technical bottleneck severely affects the temperature control accuracy in semiconductor manufacturing processes, potentially causing wafer deformation, stacking errors, and other problems, impacting the yield and reliability of semiconductor chips. Summary of the Invention

[0003] The purpose of this invention is to provide a non-destructive measurement device and method for the surface temperature of semiconductor wafers in rapid thermal processing, which can effectively solve the problem of inaccurate measurement in the low-temperature range of infrared pyrometers and improve the stability and accuracy of temperature measurement.

[0004] One aspect of the present invention provides a non-destructive measurement device for the surface temperature of a semiconductor wafer for rapid thermal processing, comprising a processing cavity, a radiation heating module, a Mueller matrix elliptic temperature measurement system, and a processing unit; The wafer to be tested is placed in the center of the processing cavity, and the radiation heating module is located at the bottom of the processing cavity for rapid thermal processing of the wafer in the processing cavity. The Mueller matrix elliptic temperature measurement system is located in the upper part of the processing cavity and includes a light source, a polarizer, a first compensator, a beam splitter, multiple temperature measuring light tubes, multiple angle-changing demodulation modules, multiple first reflectors, and a multi-channel spectrometer. A light source is used to provide incident light, a polarizer is used to generate linearly polarized light from the incident light, a first compensator is used to introduce a phase delay into the linearly polarized light to generate ellipsoidally polarized light with adjustable ellipsoidal state, a beam splitting module is used to split the ellipsoidally polarized light into multiple measurement channels that enter multiple temperature-sensing light tubes, the multiple temperature-sensing light tubes are optically connected to the processing cavity, multiple angle-changing demodulation modules are respectively located between the multiple temperature-sensing light tubes and the wafer, used to change the incident angle of light on the wafer surface and demodulate the polarization state of the reflected light on the wafer surface, multiple first mirrors are used to reflect the polarization state demodulated light returned by the multiple angle-changing demodulation modules and couple it to a multi-channel spectrometer, which collects it as a multi-wavelength polarization response signal; The processing unit is used to perform Fourier series expansion on the total light intensity of the multi-wavelength polarization response signals collected by the multi-channel spectrometer, solve for the Fourier coefficients of each order, obtain the Mueller matrix spectrum at different wavelengths, and invert the wafer surface temperature based on the Mueller matrix spectrum.

[0005] Another aspect of the present invention provides a non-destructive measurement method for the surface temperature of a semiconductor wafer used in rapid thermal processing. The method utilizes the aforementioned apparatus to perform non-destructive measurement of the surface temperature of the semiconductor wafer, and includes: Measure the Mueller matrix spectrum of wafers at different temperatures and establish a database of the relationship between wafer temperature and Mueller matrix spectrum; Deep learning was used to train the database to obtain a classification model of wafer temperature-Muller matrix spectra; The multi-path Mueller matrix spectrum of the wafer under test is obtained, input into a deep learning classification model, and the wafer temperature is deduced.

[0006] The non-destructive measurement device and method for semiconductor wafer surface temperature in rapid thermal processing according to the above-described aspects of the present invention can effectively solve the problem of inaccurate measurement in the low-temperature range of infrared pyrometers and improve the stability and accuracy of temperature measurement. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the present invention, the accompanying drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort: Figure 1 This is a schematic diagram of a non-destructive measurement device for semiconductor wafer surface temperature in rapid thermal processing according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a Mueller matrix elliptic thermometry system according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a beam splitter module according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a variable angle demodulation module according to an embodiment of the present invention; Figure 5 This is a Mueller matrix spectrum according to an embodiment of the present invention. Detailed Implementation

[0008] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0009] One embodiment of the present invention provides a non-destructive measurement device for the surface temperature of semiconductor wafers during rapid thermal processing, such as... Figure 1 As shown, the non-destructive temperature measurement device for semiconductor wafer surface during rapid thermal processing according to an embodiment of the present invention includes a fixture 1, a processing cavity 3, a Mueller matrix elliptic thermometry system, a quartz window 8, a radiation heating module 9, and a processing unit. The wafer 2 is fixed by the fixture 1, with its center on the central axis 4 and located at the center of the processing cavity 3. The radiation heating module 9 provides the rapid heating environment required for the RTP process and includes a heat source array 11 and a reflector tube 10. The heat source array 11 is uniformly distributed at the bottom of the processing cavity 3 for rapid thermal processing of the wafer 2. The reflector tube 10 is interspersed between the units of the heat source array 11 to assist in the thermal uniformity of the radiation surface formed by the heat source array 11. The quartz window 8 is disposed between the processing cavity 3 and the heat source array 11 to separate the processing cavity 3 from the radiation heating module 9, ensuring a high-temperature heating environment and an airtight environment. The Mueller matrix elliptic thermometry system is arranged at the top of the processing cavity 3 to acquire the optical response of the wafer 2 surface in real time. The processing unit is, for example, a... Figure 2 The computer 206 is used to receive and process the optical response signal acquired by the Mueller matrix elliptic thermometry system, and to invert the wafer surface temperature based on the Mueller matrix spectrum.

[0010] like Figure 2 As shown, the Mueller matrix elliptic thermometry system includes a light source 201, a collimating mirror 202, a first motor 203, a driver 204, a controller 205, a polarizer 207, a first compensator 208, a beam splitter 209, a first reflector 210, a thermometric light tube 6, a variable angle demodulation module 7, and a multi-channel spectrometer 213. The optical path propagation sequence is: light source 201 → collimating mirror 202 → polarizer 207 → first compensator 208 → beam splitter 209 → first reflector 210 → thermometric light tube 6 → variable angle demodulation module 7 → wafer 2 → return optical path → multi-channel spectrometer 213.

[0011] The light source 201 provides broadband stable incident light; the collimating lens 202 converts divergent light into parallel light to improve polarization stability; the polarizer 207 generates initial linearly polarized light; the first compensator 208 introduces phase delay to achieve adjustable ellipsometric polarization; the controller 205 controls the driver 204, which drives the first motor 203, which in turn drives the first compensator 208 to adjust the ellipsometric polarization state of the incident light; the beam splitter 209 divides the ellipsometric light (elliptical polarized light) into multiple (e.g., 8) measurement channels, which then enter multiple temperature-measuring light tubes 6; the first reflector 210 is used for optical path folding and spatial coupling; the temperature-measuring light tube 6 passes through the angle-changing demodulation module 7 to form an optical path with the wafer 2, ensuring stable propagation of the optical path in a high-temperature RTP environment and achieving optical path isolation and stable transmission in a high-temperature environment. The variable angle demodulation module 7 is located between the temperature measuring light tube 6 and the wafer 2. It is used for optical path modulation and signal demodulation to realize dynamic adjustment of the incident angle, polarization state demodulation and demodulation polarization state change; the multi-channel spectrometer 213 is used to collect multi-wavelength polarization response signals.

[0012] like Figure 3 As shown, the beam splitting module 209 includes multiple beam splitters 302 and a second reflecting mirror 303. The elliptically polarized light 301, after passing through the first compensator 208, is split into eight elliptically polarized beams by the beam splitters 302 and the second reflecting mirror 303, which then enter eight temperature-measuring light tubes 6. The beam splitting module 209 is located after the first compensator 208 and before the temperature-measuring light tubes 6. It receives the elliptically polarized light 301 output from the first compensator 208, achieves beam spatial separation through the multiple beam splitters 302, adjusts the propagation direction through the second reflecting mirror 303, and matches the eight optical channels before entering the eight temperature-measuring light tubes 6. The functions of the beam splitting module 209 are: to decompose a single elliptically polarized light into eight independent polarization measurement beams; each beam corresponds to a different incident sampling point; to improve the dimensionality and redundancy of the Mueller matrix measurement; and to provide a data foundation for subsequent multi-channel inversion.

[0013] like Figure 4As shown, the variable angle demodulation module 7 is located between the temperature-sensing light tube 6 and the wafer 2. It includes eight sets of components: a variable angle reflector 408, a moving mechanism, a polarizer 405, a second compensator 406, and a second motor 407, connected to the eight temperature-sensing light tubes 6. There are two variable angle reflectors 408, which are respectively positioned on the incident and reflection paths of the incident light on the wafer. They are used to synchronously change the incident and reflection angles of the light on the wafer surface, achieving coaxial or quasi-coaxial modulation of the incident and reflection paths. The moving mechanism includes an X-axis moving guide rail 402, a Z-axis moving guide rail 403, and two sliders 404. Two variable-angle reflectors 408 are respectively mounted on the two sliders 404. The two sliders 404 are symmetrically mounted on the X-axis moving guide rail 402, which is mounted on the Z-axis moving guide rail 403. The X-axis moving guide rail 402 and the Z-axis moving guide rail 403 form a two-dimensional displacement platform. The sliders 404 carry the variable-angle reflectors 408. The coordinated movement of the X-axis moving guide rail 402, the Z-axis moving guide rail 403, and the sliders 404 can achieve synchronous adjustment of the incident angle and the reflection angle. The two variable-angle reflectors 408 are driven by the two sliders 404 respectively, achieving symmetrical movement. The second compensator 406, as an adjustable polarization state analysis element, performs precise phase modulation on the polarized light returning from the wafer. The analyzer 405 is used to demodulate the polarization state of the returned light. The second motor 407 drives the second compensator 406 to adjust the elliptic polarization state of the returned light. Like the first motor 203, the second motor 407 can be driven by a separate controller, or it can share a multi-axis driver and controller with the first motor 203. All modules are rigidly connected, forming an adjustable optical angle platform. The optical path propagation sequence is: temperature-measuring light tube 6 → angle-adjustable reflector 408 (incident) → reflection / transmission from wafer 2 surface → angle-adjustable reflector 408 (recovery) → second motor 407 → second compensator 406 → analyzer 405 → light tube 401 → multi-channel spectrometer 213. The function of the angle-adjustable demodulation module 7 is as follows: the X-axis moving guide rail 402 and the Z-axis moving guide rail 403 work together to change the incident angle, achieving angle-resolved measurement; the analyzer 405 demodulates the polarization state of the returned light; the angle-adjustable reflector 408 achieves coaxial or quasi-coaxial modulation of the incident / reflection paths; and angle scanning enhances the recognizability of the Mueller matrix.

[0014] Computer 206 processes the spectrum by taking the light intensity obtained from rotating the first compensator 208 and the second compensator 406 at different rotation angles, performing Fourier analysis, and solving the overdetermined equations to obtain the Mueller matrix at different wavelengths. For example... Figure 5 The elliptic spectrum of the Mueller matrix is ​​shown, where 511: Mueller matrix M 11 Spectra of elements; 512: Mueller matrix M 12 Spectra of elements; 513: Mueller matrix M 13 Spectra of elements; 514: Mueller matrix M14 Spectra of elements; 521: Mueller matrix M 21 Spectra of elements; 522: Mueller matrix M 22 Spectra of elements; 523: Mueller matrix M 23 Spectra of elements; 524: Mueller matrix M 24 Spectra of elements; 531: Mueller matrix M 31 Spectra of elements; 532: Mueller matrix M 32 Spectra of elements; 533: Mueller matrix M 33 Spectra of elements; 534: Mueller matrix M 34 Spectra of elements; 541: Mueller matrix M 41 Spectra of elements; 542: Mueller matrix M 42 Spectra of elements; 543: Mueller matrix M 43 Spectra of elements; 544: Mueller matrix M 44 The spectrum of elements.

[0015] For isotropic structures, analyze the elements on the main diagonal: 512, 521, 522, 533, 534, 543, 544; for anisotropic structures, additionally analyze the elements on the secondary diagonal: 513, 514, 523, 524, 531, 532, 541, 542. A detailed description follows.

[0016] The incident and outgoing polarized light are modulated by time-domain polarization modulation through rotating the first compensator 208 and the second compensator 406, with a rotation angle ratio of 5:3. Based on the above optical path sequence, the formula for solving the Stokes vector of the incident light can be obtained. :

[0017] in, Let Stokes vector be the natural light. and These are the Mueller matrices for the polarizer 207 and the first compensator 208 (incident optical path phase delayer), respectively. The azimuth angle of the fast axis of the first compensator 208. It is a rotation matrix.

[0018] Therefore, the electric field component of polarized light can be observed. , and and phase difference This represents the Stokes vector of the polarized light after passing through the sample wafer. :

[0019] in, It is the Mueller matrix of the sample. This represents the incident angle. After passing through the sample, the polarized light undergoes post-processing, specifically modulation and demodulation by the second compensator 406 (outgoing light path phase delayer) and the analyzer 405, to obtain the Stokes vector of the final output polarized light information. :

[0020] in, It is the Mueller matrix of the polarizer 405. . It is the Mueller matrix of the second compensator 406. . This indicates the azimuth angle of the fast axis of the second compensator 406.

[0021] In the Mueller matrix ellipticization system, the fast axis azimuth rotation ratio of the phase retarder is 5:3 or 3:5. In this embodiment, This represents the total light intensity I received by a single channel of a multi-channel spectrometer. A Fourier series expansion is performed on the total light intensity I to improve the 4×4 matrix. For the stability of the solution to the 16 elements, 24 non-zero Fourier coefficients must be satisfied. Let the rotation angle ratio of the two phase delayers be... By using matrix transfer and the sum-difference-to-product of trigonometric functions, the angular ratio of the fast axis azimuth angles of the two phase delayers is converted into an angular velocity ratio. The highest-order nonzero Fourier coefficient can be obtained as follows: Therefore, through the non-zero Fourier coefficient ( Solve for a 4×4 matrix. 15 unknown elements (normalized to) Of these, eight Fourier coefficients were eliminated, namely... Therefore, at least one cycle is required Therefore, considering the number of sampling points, measurement efficiency, and rotational stability, the number of samples for one optical period of the theoretical Mueller matrix spectrum is 36.

[0022] Since solving the Mueller matrix spectrum analytically will yield singular solutions, the Fourier coefficient fitting method is used to solve the Mueller matrix spectrum. The optical axis azimuths P and A of polarizer 207 and analyzer 405, as well as the fast axis azimuths of the two phase retarders, are introduced. , The total light intensity can be obtained at t Expression for time:

[0023] in,

[0024] In the formula, , ( ) In the formula, and These represent the phase delay amounts of the two phase delayers, respectively. The elements of the sample Mueller matrix are shown. This is the total light intensity I. Perform a Fourier expansion on I:

[0025] in, For DC Fourier components, and Indicates a frequency of 2 Fourier coefficient at time Its phase term.

[0026] Normalize the elements of the Mueller matrix to Finally, the Mueller matrix of the sample can be solved by a linear combination of non-zero Fourier coefficients. The other 15 elements are shown in the following formula group:

[0027] in, It is the DC Fourier coefficient: (p) Since the optical axis azimuth angles P and A of polarizer 207 and analyzer 405 have different values ​​at different wavelengths, P and A are expressed as follows in the above formula set: , The total detected light intensity is expanded using a Fourier series, and the Fourier coefficients of each order are solved. and Then, linear operations are performed in conjunction with system parameters (system parameters include P, A, ...). , , and By combining the formula set (ap), the Mueller matrix of the sample can be solved. The 15 elements are used to solve for the Mueller matrix spectrum of the sample by solving for each wavelength point.

[0028] The procedure for non-destructive measurement of semiconductor wafer surface temperature for rapid thermal processing according to an embodiment of the present invention is as follows: First, the thermo-optical coefficient and thermal expansion coefficient of the wafer are obtained to ensure that the surface temperature of the wafer can be accurately inferred from the measured spectrum. Next, the Mueller matrix spectra of the wafer at different temperatures are measured, and a database of the relationship between temperature and Mueller matrix spectra is established. Then, the database is trained using deep learning to obtain a classification model of wafer temperature-Mueller matrix spectrum. Finally, the multi-channel Mueller matrix spectra obtained from actual measurements are used as a test set and input into the deep learning classification model to infer the wafer temperature.

[0029] The principle of the temperature measurement process in this invention is based on: wafer temperature change → material thermal-optical parameter change → Mueller matrix spectral response change → deep learning temperature inversion. The specific process is as follows: (1) Acquisition of thermo-optic coefficient and thermal expansion coefficient: Using optical testing instruments such as ellipsometer, the changes in optical properties such as film thickness and refractive index of wafer material are measured at multiple temperatures; the thermo-optic coefficient and thermal expansion coefficient are calculated by measuring the changes in refractive index and extinction coefficient; the multi-channel Mueller matrix spectrum of the sample wafer is measured under different temperature conditions using the measuring device of this embodiment; the actual temperature value of each temperature point is calibrated using a standard thermometer (such as a thermocouple); the Mueller matrix spectrum data at each temperature point is recorded, and a preliminary mathematical model of the temperature and Mueller matrix spectrum response of a specific material is established by using multivariate linear regression.

[0030] (2) Relational database construction: The Mueller matrix spectra measured at multiple temperature points and the corresponding known temperatures are recorded in the database; the database contains detailed information such as wafer material, thickness, and process conditions; data processing is performed, and noise and unqualified samples are removed through data preprocessing to obtain a robust database foundation; multivariate analysis: multivariate analysis (such as principal component analysis PCA) is used to extract effective spectral features and enhance the distinguishability of the Mueller matrix at different temperatures.

[0031] (3) Deep learning model training and temperature inversion: First, data preprocessing and feature extraction are performed. The acquired Mueller matrix spectrum is normalized to eliminate the influence of light intensity fluctuations under different measurement conditions. Feature selection technology is used to extract the most representative and discriminative spectral features and optimize the feature dimension of the input neural network. Second, deep learning model construction is carried out. The model architecture is selected based on deep learning model architectures such as convolutional neural network (CNN) or long short-term memory network (LSTM) because they are suitable for processing complex spectral signal sequences. Temperature-Mueller matrix pairs in relational database are used. Model training is performed, and appropriate loss functions (such as mean square error) and optimization algorithms (such as Adam) are selected. Cross-validation technology (such as K-fold cross-validation) is used to evaluate the model performance to ensure the predictive ability for both seen and unseen data. Finally, temperature inversion is performed. In actual operation, multi-channel Mueller matrix spectra of the wafer surface under test are collected in real time. The collected data is input into the trained deep learning model for temperature inversion. After temperature prediction, necessary corrections are made based on existing calibration data and physical knowledge to improve measurement accuracy.

[0032] In summary, the non-destructive measurement device and method for semiconductor wafer surface temperature in rapid thermal processing according to embodiments of the present invention replaces the infrared pyrometer in existing rapid thermal processing equipment with a Mueller matrix elliptic temperature measurement system. By utilizing a temperature measurement method combining thermal-optical coefficients and ellipticity, it overcomes the technical limitations of traditional infrared pyrometers, avoids the problem of inaccurate measurement in the low-temperature range (80℃-400℃) of infrared pyrometers, and uses deep learning to reverse characterize the temperature of the wafer surface. Combined with multi-channel Mueller matrix data, it improves the accuracy of temperature identification.

[0033] The non-destructive measurement device and method for semiconductor wafer surface temperature in rapid thermal processing embodiments of the present invention can effectively solve the problem of low-temperature infrared transparency of silicon wafers, provide accurate low-temperature temperature measurement, thereby reducing wafer deformation and stacking errors caused by differences in thermal expansion coefficients, improving the manufacturing yield and reliability of semiconductor chips, and has significant technical application value, promoting technological progress in the semiconductor industry. Furthermore, the application of the temperature measurement device and method of the present invention includes, but is not limited to, the measurement of semiconductor wafer surface temperature, and can also be used for temperature measurement of various materials whose thermo-optical coefficient is significantly greater than their thermal expansion coefficient.

[0034] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A non-destructive measurement device for the surface temperature of a semiconductor wafer used in rapid thermal processing, characterized in that, Includes a processing cavity, a radiant heating module, a Mueller matrix elliptic temperature measurement system, and a processing unit; The wafer to be tested is placed in the center of the processing cavity, and the radiation heating module is located at the bottom of the processing cavity for rapid thermal processing of the wafer in the processing cavity. The Mueller matrix elliptic temperature measurement system is located in the upper part of the processing cavity and includes a light source, a polarizer, a first compensator, a beam splitter, multiple temperature measuring light tubes, multiple angle-changing demodulation modules, multiple first reflectors, and a multi-channel spectrometer. A light source is used to provide incident light, a polarizer is used to generate linearly polarized light from the incident light, a first compensator is used to introduce a phase delay into the linearly polarized light to generate ellipsoidally polarized light with adjustable ellipsoidal state, a beam splitting module is used to split the ellipsoidally polarized light into multiple measurement channels that enter multiple temperature-sensing light tubes, the multiple temperature-sensing light tubes are optically connected to the processing cavity, multiple angle-changing demodulation modules are respectively located between the multiple temperature-sensing light tubes and the wafer, used to change the incident angle of light on the wafer surface and demodulate the polarization state of the reflected light on the wafer surface, multiple first mirrors are used to reflect the polarization state demodulated light returned by the multiple angle-changing demodulation modules and couple it to a multi-channel spectrometer, which collects it as a multi-wavelength polarization response signal; The processing unit is used to perform Fourier series expansion on the total light intensity of the multi-wavelength polarization response signals collected by the multi-channel spectrometer, solve for the Fourier coefficients of each order, obtain the Mueller matrix spectrum at different wavelengths, and invert the wafer surface temperature based on the Mueller matrix spectrum.

2. The apparatus as claimed in claim 1, characterized in that, The variable angle demodulation module includes a variable angle reflector, a moving mechanism, a second compensator, and a polarizer; There are two variable angle mirrors, which are used to synchronously change the incident angle and reflection angle of light on the wafer surface. The moving mechanism is used to move the two variable angle mirrors symmetrically. The second compensator is used to perform phase modulation on the polarized light returning from the wafer surface. The analyzer is used to demodulate the polarization state of the returning polarized light.

3. The apparatus as described in claim 2, characterized in that, The moving mechanism includes an X-axis moving guide rail, a Z-axis moving guide rail, and two sliders. Two variable-angle reflectors are respectively mounted on the two sliders. The two sliders are symmetrically mounted on the X-axis moving guide rail, and the X-axis moving guide rail is mounted on the Z-axis moving guide rail.

4. The apparatus according to any one of claims 1-3, characterized in that, The beam splitting module includes multiple beam splitters and a second reflector. The ellipsoidally polarized light after passing through the first compensator passes through the beam splitter and the second reflector, and is split into multiple ellipsoidally polarized light beams that enter the temperature measuring light tube respectively.

5. The apparatus according to any one of claims 1-3, characterized in that, The radiation heating module includes a heat source array and a reflector tube. The heat source array is evenly distributed at the bottom of the processing cavity and is used to perform rapid thermal processing on the wafer. The reflector tube is inserted between the units of the heat source array and is used to assist in the thermal uniformity of the radiation surface formed by the heat source array. The device also includes a clamp and a quartz window. The clamp is used to fix the wafer, and the quartz window is disposed between the processing cavity and the heat source array to separate the processing cavity from the radiation heating module.

6. The apparatus according to any one of claims 1-3, characterized in that, The Mueller matrix elliptic temperature measurement system also includes a collimating mirror, a first motor, a driver, and a controller; A collimating lens is positioned between the light source and the polarizer to convert the divergent light from the light source into parallel light, thereby improving polarization stability. The controller is used to control the driver to drive the first motor, causing the first compensator to rotate and change the elliptic polarization state of the incident light. The angle-modulation module also includes a second motor, which drives the second compensator to rotate and change the elliptic polarization state of the returned light.

7. A non-destructive method for measuring the surface temperature of a semiconductor wafer used in rapid thermal processing, characterized in that, The method for non-destructive measurement of the surface temperature of a semiconductor wafer using the apparatus of any one of claims 1-6 includes: Measure the Mueller matrix spectrum of wafers at different temperatures and establish a database of the relationship between wafer temperature and Mueller matrix spectrum; Deep learning was used to train the database to obtain a classification model of wafer temperature-Muller matrix spectra; The multi-path Mueller matrix spectrum of the wafer under test is obtained, input into a deep learning classification model, and the wafer temperature is deduced.

8. The method as described in claim 7, characterized in that, The Mueller matrix spectrum is obtained as follows: The total light intensity received by a single channel of the multi-channel spectrometer is expanded into a Fourier series, and the Fourier coefficients of each order are solved. The elements of the Mueller matrix of the wafer are solved by fitting the Fourier coefficients. The Mueller matrix spectrum of the wafer is obtained by solving for each wavelength point.

9. The method as described in claim 7 or 8, characterized in that, Also includes: Optical testing instruments are used to measure the changes in the optical properties of the wafer material at multiple temperatures. The thermo-optical coefficient and the coefficient of thermal expansion are calculated based on the measured changes in optical properties, ensuring that the surface temperature of the wafer can be inferred from the measured spectrum.

10. The method as described in claim 7 or 8, characterized in that, Deep learning training on a database includes: Data preprocessing and feature extraction were performed, and the obtained Mueller matrix spectrum was normalized. Feature selection techniques are used to extract representative and discriminative spectral features; Convolutional neural networks were chosen as the deep learning model. Train the model, select the loss function and optimization algorithm; Cross-validation was used to evaluate the model performance.

Images