Data detection method and system, detection device, computer equipment and storage medium

Abstract

The invention provides a data detection method and system, a detection device, computer equipment and a storage medium, and the method comprises the steps: obtaining a detection signal of each wafer surface in a planetary reaction cavity, and cutting the detection signal based on a trigger signal to obtain a first signal sequence and a second signal sequence; processing the first signal sequence by adopting a classification model to obtain a material type on each wafer; querying a corresponding optical model based on the material type; and performing iterative optimization on the optical model, the first signal sequence and the second signal sequence by adopting a decoupling algorithm to obtain a temperature detection value and a reflectivity detection value. According to the method, the original detection signal is segmented into the independent signal sequences based on the trigger signal, then the material type is determined according to the independent signal sequences, the corresponding optical model is selected according to the material type, and finally the temperature detection value and the reflectivity detection value with high accuracy are obtained through decoupling according to the optical model. Accurate analysis of a single wafer signal is realized, and process repeatability and stability are greatly improved.

Description

Technical Field

[0001] This application relates to the field of semiconductor testing technology, and in particular to a data testing method and system, testing device, computer equipment and storage medium. Background Technology

[0002] Metal-organic chemical vapor deposition (MOCVD) equipment is the core equipment for fabricating epitaxial wafers of compound semiconductors (such as GaAs and GaN). Within its typical planetary reaction chamber, a main graphite disk revolves under the drive of a motor, while multiple "satellite disks" mounted on it rotate on their own axes driven by gas flow. Once the chamber is heated to a certain temperature, chemical gases are introduced, triggering a chemical reaction that deposits the thin film layer required for semiconductor devices onto the wafer surface. This combined "revolution + rotation" motion aims to achieve uniformity in temperature and gas distribution within the reaction chamber, ensuring consistent film thickness and composition on each wafer carrier disk.

[0003] During the pre-epitaxy process setup, standard and auxiliary wafers are typically placed in the equipment to determine if the equipment cavity is operating normally. When multiple satellite disks contain wafers of different materials, such as germanium (Ge), gallium arsenide (GaAs), and sapphire substrates, the emissivity and reflectivity of the wafers with different materials vary significantly.

[0004] Therefore, while planetary turntables offer benefits in terms of process uniformity, they also present an inherent challenge for optical inspection: the coupling between the measurement field of view and the motion trajectory. Fixed optical detectors (such as infrared pyrometers or thermal imagers) have a fixed measurement field of view. When revolution and rotation occur simultaneously, the optical signal received by the detector at any given moment is the sum of the radiated energy from all wafer surfaces within its field of view. In the case of heterogeneous substrates, this signal is a dynamically changing mixture of radiation from substrates with different emissivity and temperatures. Traditional techniques typically treat this mixed signal as a whole and invert the temperature based on an "average" or "equivalent" emissivity. This method completely ignores the physical nature of different materials, resulting in a temperature value without clear physical meaning; or it simply averages the signal over a period of time, but the average is still a mixture weighted by various materials, failing to extract information from individual materials. This leads to inaccurate temperature and reflectivity values, failing to truly represent the process state of any specific wafer. Summary of the Invention

[0005] The purpose of this application is to provide a data detection method and system, detection device, computer equipment and storage medium to overcome the problem that the detection signal obtained in the traditional technology is a mixed signal, which leads to inaccurate temperature value and reflectivity value.

[0006] In a first aspect, this application proposes a data detection method, the method comprising: acquiring detection signals on the surface of a wafer; wherein each wafer is disposed within a planetary reaction cavity, and the detection signals include spectral reflectance signals and thermal radiation signals; cutting the spectral reflectance signals and the thermal radiation signals respectively based on a trigger signal to obtain a first signal sequence and a second signal sequence; wherein one trigger signal represents one rotation of a graphite disk within the planetary reaction cavity; processing the first signal sequence using a pre-trained classification model to obtain the material type on each wafer; querying the optical model corresponding to the material type from a preset database based on the material type; and iteratively optimizing the optical model, the first signal sequence, and the second signal sequence using a decoupling algorithm to obtain temperature detection values ​​and reflectance detection values.

[0007] In one embodiment, the step of cutting the spectral reflection signal and the thermal radiation signal based on the trigger signal to obtain a first signal sequence and a second signal sequence includes: for each trigger signal, determining the time interval for each satellite disk to reach a preset position, with the trigger signal as the time starting point; taking multiple time points within the time interval, setting a time window with each time point as the center, and cutting the signal segment corresponding to each time window from the spectral reflection signal and the thermal radiation signal; the width of the time window is sufficient to ensure that the detection signals of all wafers on the satellite disk are completely captured; and splicing the continuously acquired signal segments belonging to the same satellite disk in chronological order to obtain the first signal sequence and the second signal sequence for each satellite disk.

[0008] In one embodiment, the classification model includes a convolutional network, a support vector machine network, and a random forest network; the step of processing the first signal sequence using a pre-trained classification model to obtain the material type on each wafer includes: extracting feature vectors characterizing the optical features of different materials from the first signal sequence; inputting the feature vectors into the convolutional network, the support vector machine network, and the random forest network respectively to obtain a first classification result, a second classification result, and a third classification result; performing a weighted summation on the first classification result, the second classification result, and the third classification result, and determining the material type based on the summation result.

[0009] In one embodiment, the feature vector includes a time-domain feature vector and a frequency-domain feature vector; the step of extracting feature vectors characterizing the optical features of different materials from the first signal sequence includes: cutting the first signal sequence to obtain a sub-signal sequence for each wafer on a satellite disk; extracting a time-domain feature vector based on each sub-signal sequence, the time-domain feature vector including at least one of average value, signal strength, and quantity; performing a Fourier transform on each sub-signal sequence to extract its dominant frequency component, obtaining a frequency-domain feature vector, the frequency-domain feature vector including growth rate and / or film thickness.

[0010] In one embodiment, the step of cutting the first signal sequence to obtain a sub-signal sequence for each wafer on a satellite disk includes: detecting pulse peaks in the first signal sequence; wherein one pulse peak corresponds to one wafer; determining the start and end points of each pulse peak; cutting the first signal sequence based on the start and end points of each pulse peak to obtain multiple signal segments; calculating the signal intensity of each signal segment and classifying each signal segment into different categories according to the signal intensity; and splicing the continuously acquired signal segments belonging to the same category in chronological order to obtain a sub-signal sequence for each wafer.

[0011] In one embodiment, the feature vector further includes a model feature vector; the step of extracting feature vectors representing the optical characteristics of different materials from the first signal sequence includes: calculating the similarity between the curve of the first signal sequence changing over time and the theoretical prediction curves of each candidate material in a preset database, to obtain multiple initial correlation coefficients, and each of the initial correlation coefficients constitutes the model feature vector.

[0012] In one embodiment, the optical model is used to characterize the optical constants as a function of wavelength and temperature, the optical constants including refractive index and extinction coefficient; the step of iteratively optimizing the optical model, the first signal sequence, and the second signal sequence using a decoupled algorithm to obtain temperature detection values ​​and reflectivity detection values ​​includes: using a pre-constructed physical forward model, based on the optical constant function of the material, generating predicted data under given temperature and surface state parameters; wherein, the predicted data includes theoretical reflectivity and theoretical radiation intensity; the surface state parameters include at least one of emissivity correction coefficient, surface roughness parameter, and oxide layer influence parameter; progressively calculating the residual between the predicted data and the measured data, the measured data including measured reflectivity and measured radiation intensity, wherein the measured reflectivity is obtained based on the sub-signal sequence, and the measured radiation intensity is obtained based on the second signal sequence; using a decoupled alternating optimization algorithm, iteratively optimizing the temperature and surface state parameters alternately, and using a nonlinear least squares algorithm to minimize the residual function until convergence; using the temperature at model convergence as the temperature detection value, and correcting the measured reflectivity based on the surface state parameters at convergence to obtain the reflectivity detection value.

[0013] In one embodiment, the method further includes: displaying the temperature and reflectivity values ​​of each wafer as curves over time, and generating an alarm signal when the curves exceed the process range. Secondly, this application proposes a data detection system, the system comprising: an acquisition module for acquiring detection signals from the surface of a wafer; wherein each wafer is disposed within a planetary reaction cavity, and the detection signals include spectral reflectance signals and thermal radiation signals; a signal segmentation module for segmenting the spectral reflectance signals and the thermal radiation signals respectively based on a trigger signal to obtain a first signal sequence and a second signal sequence; wherein one trigger signal represents one rotation of a graphite disk within the planetary reaction cavity; a processing module for processing each of the first signal sequences using a pre-trained classification model to obtain the material type on each wafer; based on the material type, querying a preset database to obtain the optical model corresponding to the material type; and using a decoupling algorithm to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain temperature detection values ​​and reflectance detection values.

[0014] Thirdly, this application also provides a detection device, comprising: an optical measurement module for emitting a light beam to each wafer surface within a planetary reaction chamber and collecting spectral reflection signals from each wafer surface; a temperature measurement module for detecting thermal radiation signals from each wafer surface; a signal line connected to a spindle encoder or motor controller of a planetary rotary table for transmitting trigger signals; wherein one trigger signal represents one rotation of a graphite disk within the planetary reaction chamber; and a data detection system as described in the second aspect for periodically processing the spectral reflection signals and the thermal radiation signals according to the trigger signals to obtain temperature detection values ​​and reflectivity detection values.

[0015] Fourthly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method steps of the first aspect.

[0016] Fifthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the method steps of the first aspect.

[0017] The above-mentioned data detection methods, systems, detection devices, computer equipment, and storage media have at least the following advantages: This application acquires detection signals from the surfaces of each wafer within a planetary reaction chamber. These signals include spectral reflectance and thermal radiation signals. Based on a trigger signal, the spectral reflectance and thermal radiation signals are segmented to obtain a first signal sequence and a second signal sequence. Each trigger signal represents one rotation of a graphite disk within the planetary reaction chamber. A pre-trained classification model is used to process the first signal sequence to determine the material type on each wafer. Based on the material type, an optical model corresponding to that material type is retrieved from a pre-defined database. A decoupling algorithm is used to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain temperature and reflectance detection values. By employing this approach, this application segments the original detection signal into individual signal sequences based on the trigger signal, determines the material type based on these individual sequences, selects the corresponding optical model based on the material type, and finally decouples the optical model to obtain highly accurate temperature and reflectance detection values. This achieves precise analysis of individual wafer signals, significantly improving process repeatability and stability. Attached Figure Description

[0018] Figure 1 This is a structural block diagram of the detection device in one embodiment; Figure 2 This is a flowchart illustrating a data detection method in one embodiment; Figure 3This is a flowchart illustrating the step of cutting and detecting data in one embodiment; Figure 4 This is a flowchart illustrating the steps for obtaining the material type in one embodiment; Figure 5 This is a flowchart illustrating the feature vector extraction step in one embodiment; Figure 6 This is a flowchart illustrating the steps of the decoupling algorithm in one embodiment; Figure 7 This is a flowchart illustrating a data detection method in one embodiment. Figure 8 Here is a block diagram of the data detection system in one embodiment; Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0019] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0020] Some exemplary embodiments of this application have been described for illustrative purposes. It should be understood that this application may be implemented in other ways not specifically shown in the accompanying drawings.

[0021] Please see Figure 1 Optionally, in one embodiment, this application provides a detection device disposed above a planetary reaction chamber 1. The planetary reaction chamber 1 contains a rotatable graphite disk 11, on which multiple rotatable star-shaped disks 12 are disposed. Each star-shaped disk 12 has multiple wafer carrier disks 13 for placing wafers. The graphite disk 11 is rotated by a motor, and each star-shaped disk 12 is rotated by airflow. After the chamber is heated to a certain temperature, a chemical gas is introduced, causing a chemical reaction that deposits on the wafer surface, thereby growing the thin film layer required for semiconductor devices. Specifically, in this embodiment, at least one standard wafer and multiple auxiliary wafers are placed on the wafer carrier disks 13. By detecting the temperature and reflectivity of the standard wafer surface during the adjustment process before the start of epitaxial growth, the normal operation of the planetary reaction chamber 1 is determined. For ease of description, the term "wafer" will be used to refer to the standard wafer and auxiliary wafers below.

[0022] The detection device includes: optical measurement module 2, temperature measurement module 3, signal line 4, and data detection system 5.

[0023] The optical measurement module 2 is located above the planetary reaction cavity 1 and is used to emit light beams to the surface of each wafer in the planetary reaction cavity 1 and collect the spectral reflection signals of each wafer surface.

[0024] The optical measurement module 2 includes a transmitting unit and a receiving unit.

[0025] The transmitting unit includes at least one LED light source for generating a measurement beam; and a transmitting optical fiber for guiding the measurement beam to the probe. The beam, after being collimated and focused by the probe, is projected onto the rotating wafer surface through the observation window of the planetary reaction cavity 1.

[0026] The receiving unit includes a receiving optical fiber for collecting the light beam reflected from the wafer surface; and at least one APD detector for converting the received reflected light signal into an analog electrical signal.

[0027] Temperature measurement module 3 is used to detect the thermal radiation signal received on the surface of each wafer. For example, temperature measurement module 3 may be a pyrometer.

[0028] Signal line 4, one end of which is connected to the spindle encoder or motor controller of the planetary rotary table machine, and the other end of which is connected to the data detection system 5, is used to transmit trigger signals to the data detection system 5; wherein, one trigger signal represents one rotation of the graphite disk 11 in the planetary reaction chamber 1. The trigger signal is strictly synchronized with the rotation angle of the graphite disk 11, ensuring that the data acquisition and the physical position of the graphite disk 11 are locked without deviation.

[0029] The data detection system 5 is used to periodically process the spectral reflection signal and the thermal radiation signal according to the trigger signal to obtain the temperature detection value and the reflectance detection value.

[0030] Specifically, the data inspection system 5 receives detection signals from the wafer surface; these detection signals include spectral reflection signals and thermal radiation signals; based on a trigger signal, the spectral reflection signals and thermal radiation signals are cut to obtain a first signal sequence and a second signal sequence; a pre-trained classification model is used to process the first signal sequence to obtain the material type on each wafer; based on the material type, the corresponding optical model is retrieved from a preset database; a decoupling algorithm is used to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain temperature detection values ​​and reflectivity detection values.

[0031] The aforementioned detection device includes an optical measurement module for collecting spectral reflectance signals, a temperature measurement module for collecting thermal radiation signals, a signal line for real-time transmission of trigger signals, and a data detection system for periodically processing the spectral reflectance and thermal radiation signals based on the trigger signals to obtain temperature and reflectance detection values. Using this scheme, this application achieves accurate analysis of signals from a single wafer based on hardware triggering and software processing, significantly improving process repeatability and stability.

[0032] Please see Figure 2 In one embodiment, this application provides a data detection method, the method comprising: Step 202: Acquire detection signals from the wafer surface; wherein each wafer is disposed in a planetary reaction chamber, and the detection signals include spectral reflection signals and thermal radiation signals.

[0033] Specifically, each wafer is placed on a wafer carrier disk and arranged in a set position according to the testing requirements.

[0034] It should be noted that the above detection signal has been converted into a digital electrical signal.

[0035] Furthermore, given that the detection signal is the sum of the radiated energy from all wafer surfaces within the field of view of the optical detector, treating this mixed signal as a whole will lead to inaccurate monitoring, specifically manifested as follows: Temperature Inversion Inaccuracies: The fundamental formula for radiometric thermometry (Planck's law of blackbody radiation) requires the emissivity of the object to be known in order to calculate the true temperature from the measured radiation intensity. Different materials have vastly different emissivities; for example, Ge may have an emissivity higher than 0.6 at high temperatures, while sapphire may have an emissivity lower than 0.3. This means that for the same true temperature, they exhibit drastically different radiation intensities. Processing the mixed signals with a single emissivity will cause the inverted temperature to deviate significantly from the true temperature of all substrates.

[0036] Reflectivity / emissivity measurement distortion: When measuring film thickness and reflectivity using laser reflectometers or interferometry, interference from background thermal radiation and signal source confusion can distort the measured values. Without distinguishing the signal source, it is impossible to accurately determine the true reflectivity changes of each material surface, thus hindering precise monitoring of the film's growth rate and surface morphology.

[0037] Misjudgment of process status: Inaccurate temperature and reflectivity data cannot accurately represent the process status of any specific wafer. Equipment engineers may make misjudgments based on this erroneous data, assuming that the equipment cavity is operating normally or abnormally. This could lead to starting formal production when the equipment is in poor condition, resulting in the scrapping of a large number of products; or performing unnecessary maintenance and adjustments when the equipment is in normal condition, reducing equipment utilization.

[0038] Step 204: Based on the trigger signal, the spectral reflection signal and the thermal radiation signal are cut to obtain the first signal sequence and the second signal sequence; wherein, one trigger signal represents the graphite disk in the planetary reaction chamber rotating once.

[0039] Specifically, in the original detection signal, the spectral reflection signal and the thermal radiation signal are mixed detection signals of all wafers. When the wafer surface materials are different, the emissivity and reflectivity of wafers made of different materials vary greatly. If the traditional method of treating the acquired optical signals as a mixed whole or simply taking the average value is used, it will lead to serious deviations in the temperature and reflectivity retrieved.

[0040] Based on this, this application sets a hardware trigger signal to correlate the timing of data processing with the rotation angle of the graphite disk. A trigger signal is generated for each full rotation of the graphite disk. Using this trigger signal as the starting point, the timing relationship is determined according to the position of each star-shaped disk, thereby achieving the segmentation of the original detection data.

[0041] Step 206: Process the first signal sequence using a pre-trained classification model to obtain the material type on each wafer; based on the material type, query the optical model corresponding to the material type from a preset database.

[0042] Specifically, the classification model is used to infer the probability of the material type on the wafer by analyzing the reflectance spectral signal. It should be understood that the classification model is trained based on historical detection data, historical temperatures, and historical reflectance.

[0043] The pre-defined database is set in the process formulation based on the generation plan. Each satellite disk corresponds to a candidate material list, which lists all material types that can be placed on that satellite disk. For example, the list of satellite disks includes Ge, GaAs, and Sappire. Furthermore, the pre-defined database also includes material names, wavelength ranges, temperature ranges, refractive index (n) datasets, extinction coefficient (k) datasets, initial values ​​or ranges of surface state parameters, and pre-set theoretical reflectivity and thermal radiation models. Among them, the surface state parameters include at least one of emissivity correction coefficients, surface roughness parameters, and oxide layer influence parameters.

[0044] Step 208: The optical model, the first signal sequence, and the second signal sequence are iteratively optimized using a decoupling algorithm to obtain the temperature detection value and the reflectivity detection value.

[0045] Specifically, decoupling algorithms refer to separating two or more physically coupled physical quantities that jointly affect the final measurement signal, thereby enabling the independent and accurate determination of the value of each quantity. In this embodiment, spectral reflectance signals and thermal radiation signals are processed simultaneously, and the optimal temperature (T) and true reflectance (R) values ​​are solved through an iterative optimization algorithm, thereby eliminating errors caused by material differences and thin-film interference effects.

[0046] The aforementioned data detection method acquires detection signals from the surfaces of each wafer within a planetary reaction chamber. These signals include spectral reflectance and thermal radiation signals. Based on a trigger signal, the spectral reflectance and thermal radiation signals are segmented to obtain a first signal sequence and a second signal sequence. Each trigger signal represents one rotation of a graphite disk within the planetary reaction chamber. A pre-trained classification model is used to process the first signal sequence to determine the material type on each wafer. Based on the material type, an optical model corresponding to that material type is retrieved from a pre-defined database. A decoupling algorithm is used to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain temperature and reflectance detection values. Using this approach, this application segments the original detection signal into individual signal sequences based on the trigger signal, then determines the material type based on these individual signal sequences, selects the corresponding optical model based on the material type, and finally decouples the optical model to obtain highly accurate temperature and reflectance detection values. This achieves precise analysis of individual wafer signals, significantly improving process repeatability and stability.

[0047] Please see Figure 3 Optionally, based on the trigger signal, the spectral reflection signal and the thermal radiation signal are respectively cut to obtain a first signal sequence and a second signal sequence, including: Step 302: For each trigger signal, using the trigger signal as the time starting point, determine the time interval for each satellite disk to reach the preset position.

[0048] Step 304: Select multiple time points within the time interval, set a time window centered on each time point, and cut out the signal segment corresponding to each time window from the spectral reflection signal and thermal radiation signal; the width of the time window is sufficient to ensure that the detection signals of all wafers on the satellite disk are completely captured.

[0049] Step 306: The continuously acquired signal segments belonging to the same satellite disk are spliced ​​together in chronological order to obtain the first signal sequence and the second signal sequence corresponding to each satellite disk.

[0050] Specifically, the position of each satellite disk on the graphite disk is fixed, and the angular offset of each satellite disk relative to the spindle can be obtained based on the mechanical structure parameters of the turntable. A 360° rotation of the graphite disk generates a pulse signal, which serves as the trigger signal. Using the trigger signal as the starting point, the time interval for each satellite disk to reach its preset position can be calculated based on its angular offset and real-time rotational speed. For example, in this embodiment, the probe is positioned at the physical center of the reaction chamber, and the preset position refers to the position where the center of the satellite disk is directly below the probe. It should be understood that as each satellite disk reaches its preset position with the rotation of the graphite disk, the wafer on it will sequentially rotate to below the probe within the time interval due to the satellite disk's rotation. By setting time points within the time interval and establishing a time window centered on these time points, signal segments corresponding to each time window can be obtained from the original detection signal. These signal segments are then spliced ​​together in chronological order to obtain the first and second signal sequences for each satellite disk. For example, the width of the time window should be greater than the satellite disk diameter divided by the linear velocity.

[0051] For example, if the original spectral reflection signal S(t), the original thermal radiation signal X(t), and the number of satellite disks are 5, then 5 independent, long-time sequence first signal sequences S1(t), S2(t), S3(t), S4(t), S5(t) and second signal sequences X1(t), X2(t), X3(t), X4(t), X5(t) are finally cut out.

[0052] By adopting the above scheme, this application separates the original, mixed signals into independent signals for each satellite disk, which facilitates subsequent material analysis and coupling algorithm processing.

[0053] Please see Figure 4 Optionally, when the classification model includes convolutional networks, support vector machine networks, and random forest networks, a pre-trained classification model is used to process each signal sequence to obtain the material type on each wafer, including: Step 402: Extract feature vectors representing the optical characteristics of different materials from the first signal sequence.

[0054] Step 404: Input the feature vectors into the convolutional network, support vector machine network, and random forest network respectively to obtain the first classification result, the second classification result, and the third classification result.

[0055] Step 406: Perform a weighted summation of the first classification result, the second classification result, and the third classification result, and determine the material type based on the summation result.

[0056] Specifically, the classification model in this embodiment employs a parallel model approach. The convolutional network uses a CNN (Convolutional Neural Network) structure, which is a one-dimensional convolutional neural network containing three convolutional layers (kernel sizes of 16, 8, and 4 respectively). Each convolutional layer is followed by a ReLU activation function and a max-pooling layer, and finally, a fully connected layer outputs the initial probability distribution P_cnn for each class. The support vector machine network uses an SVM (Support Vector Machine) with radial basis function (RBF) kernels. The input is a multi-dimensional feature vector extracted from the original signal, and the output is a probability estimate P_svm. The random forest network constructs a random forest containing 100 decision trees. The input features are the same as those of the SVM branches, and the output is a probability estimate P_rf.

[0057] Furthermore, the input data for the classification model is a preprocessed time series of optical signals of at least one wavelength, forming a multi-channel input data format. Each sample contains multiple data points, and the number of data points is the rotation speed / number of star disks * sampling frequency.

[0058] Furthermore, the probabilities mentioned above are weighted and summed to obtain the summation result: P_final = α×P_cnn + β×P_svm + γ×P_rf, where α, β, and γ are weight coefficients. These weight coefficients can be determined through grid search optimization; for example, in this embodiment, the values ​​are α=0.5, β=0.3, and γ=0.2.

[0059] Optionally, if there are multiple summation results, the material type corresponding to the summation result with the largest value is selected as the matching material for the current satellite disk.

[0060] Optionally, this application also sets a preset threshold. If the summation result is less than the preset threshold, an alarm signal is output, indicating "material cannot be identified", and the software is allowed to intervene and check to prevent mismatch from causing all subsequent calculations to fail.

[0061] It should be noted that the classification model was trained based on historical testing data and historical process formulas.

[0062] Using the above approach, CNN excels at extracting local features, SVM is suitable for high-dimensional space classification, and random forest has excellent anti-overfitting capabilities. By weighted fusion of these multiple models, the overall model performance is optimized, enabling accurate identification of wafers made of different materials.

[0063] Please see Figure 5 Optionally, when the feature vector includes time-domain feature vectors and frequency-domain feature vectors, feature vectors characterizing the optical features of different materials are extracted from the first signal sequence, including: Step 502: Cut the first signal sequence to obtain a sub-signal sequence for each wafer on a satellite disk.

[0064] Step 504: Based on each sub-signal sequence, extract the time-domain feature vector. The time-domain feature vector includes at least one of the average value, signal strength, and quantity.

[0065] Step 506: Perform Fourier transform on each sub-signal sequence to extract its main frequency component and obtain frequency domain feature vector, which includes growth rate and / or film thickness.

[0066] Specifically, this application performs a second segmentation of the spectral reflection signal, that is, after first segmenting to obtain the optical signal time series of each satellite disk, it further segments to obtain the optical signal time series of each wafer on the satellite disk.

[0067] Time-domain features characterize the amplitude of a signal as it changes over time, enabling the model to learn the signal's intuitive statistical properties and temporal dynamics. Frequency-domain features characterize the frequency components of a signal and the energy distribution of each frequency component, allowing the model to identify periodic components and resonant frequencies that are difficult to detect in the time domain.

[0068] By using time-domain and frequency-domain feature vectors, the aforementioned CNN can automatically learn the local dependencies between features and discover deeper feature combinations; SVM can ensure the optimality of the classification boundary when the feature dimension is high, prevent overfitting, and improve the stability and generalization ability of the model; Random Forest can determine which time-domain or frequency-domain features contribute the most to classification by constructing multiple decision trees and voting, providing guidance for feature optimization. The above ensemble characteristics make the final decision more robust.

[0069] Optionally, the first signal sequence is cut to obtain a sub-signal sequence for each wafer on a satellite disk, including: Detect pulse peaks in the first signal sequence; where each pulse peak corresponds to one wafer; determine the start and end points of each pulse peak; based on the start and end points of each pulse peak, cut the first signal sequence to obtain multiple signal segments; calculate the signal intensity of each signal segment, and classify each signal segment into different categories according to the signal intensity; splice the continuously acquired signal segments belonging to the same category in chronological order to obtain the sub-signal sequence of each wafer.

[0070] For example, taking a satellite disk with three wafers as an example, the first signal sequence corresponding to the satellite disk is S1(t). Within the time window of the satellite disk, due to the rotation of the satellite disk, the three wafers on it will pass under the probe in sequence. Therefore, at least three pulse peaks will appear in the S1(t) signal. One pulse peak represents the probe scanning a wafer, and the number of pulse peaks is related to the rotation speed of the satellite disk. The faster the rotation speed, the more pulse peaks there will be.

[0071] Because different materials have different reflectivities, the pulse peak heights differ between different wafers. For example, the pulse peak height of Sapphire material is shorter than that of GaAs material.

[0072] By using a peak detection algorithm or setting a dynamic threshold, the start and end points of the above three pulse peaks are found, thereby cutting S1(t) into at least three shorter signal segments, each corresponding to the entire process of a wafer being scanned by the probe.

[0073] It should be understood that even on the same satellite disk, the signal strength of different wafers may vary due to slight differences in their surface condition or position. By calculating the signal strength of each sub-signal sequence, signal segments from the same physical wafer can be categorized and spliced, thereby generating an independent sub-signal sequence for each wafer. Using this approach, the temporal continuity and statistical redundancy of optical timing curves can be utilized to effectively improve the accuracy, stability, and robustness of subsequent coupling algorithms.

[0074] Furthermore, this application also pre-stores the fixed physical position sequence of each wafer on each satellite disk. After acquiring each pulse peak, the time sequence of the pulse peaks is compared with the rotation sequence of the wafers based on the characteristics of the pulse peaks to avoid mismatch.

[0075] Optionally, if the feature vector also includes a model feature vector, feature vectors characterizing the optical features of different materials are extracted from the first signal sequence, including: The similarity between the curve of the first signal sequence changing over time and the theoretical prediction curve of each candidate material in the preset database is calculated to obtain multiple initial correlation coefficients, and each initial correlation coefficient constitutes the model feature vector.

[0076] Specifically, the initial correlation coefficient characterizes the similarity between the two. By converting time series data into similarity scores with each material theory model, it can provide input features with clear physical meaning for subsequent classification algorithms.

[0077] Please see Figure 6Optionally, the optical model is used to characterize the optical constants as a function of wavelength and temperature, and the optical constants include refractive index and extinction coefficient; a decoupling algorithm is used to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain the temperature detection value and the reflectance detection value, including: Step 602: Using a pre-built physical forward model, based on the optical constant function of the material, predictive data is generated under given temperature and surface state parameters. The predicted data includes theoretical reflectivity and theoretical radiation intensity; the surface state parameters include at least one of emissivity correction coefficient, surface roughness parameter, and oxide layer influence parameter.

[0078] Step 604: Calculate the residuals between the predicted data and the measured data step by step. The measured data includes the measured reflectivity and the measured radiation intensity. The measured reflectivity is obtained based on the sub-signal sequence, and the measured radiation intensity is obtained based on the second signal sequence.

[0079] Step 606: A decoupled alternating optimization algorithm is used to iteratively optimize the temperature and surface state parameters, and a nonlinear least squares algorithm is used to minimize the residual function until convergence.

[0080] Step 608: Use the temperature at which the model converges as the temperature detection value, and correct the measured reflectivity based on the surface state parameters at the time of convergence to obtain the reflectivity detection value.

[0081] Specifically, once the material type is determined, the optical model corresponding to that material type can be retrieved from the preset database. The expression of the optical model is n, k = f(λ, T), where n represents the refractive index, k represents the extinction coefficient, λ represents the wavelength, and T represents the temperature. This optical model is used to characterize the optical constants as a function of wavelength and temperature, that is, the refractive index n and extinction coefficient k of different substrates and thin film materials at different wavelengths and temperatures.

[0082] It should be noted that temperature is one of the core physical quantities to be solved in this scheme, as it directly affects the optical constants (n, k) and thermal radiation intensity of the material. The goal of optimizing the temperature is to find the physical true value that best matches both the spectral reflectance signal and the thermal radiation signal.

[0083] Surface state parameters are compensation terms or correction factors used to describe and quantify the differences between the actual optical properties of a wafer surface and an ideal smooth, clean surface. They are also used to compensate for the influence of wafer surface optical properties on temperature and reflectivity measurements. The purpose of optimizing surface state parameters is to compensate for the systematic errors introduced by non-ideal surface states in temperature and reflectivity measurements, thereby making the solutions for core physical quantities (temperature and reflectivity) more accurate.

[0084] Optionally, to improve data accuracy, the aforementioned sub-signal sequence and second signal sequence can be obtained based on multiple trigger signals. That is, after each trigger signal arrives, the spectral reflectance signal is cut twice to obtain the sub-signal sequence of each wafer, and the thermal radiation signal is cut once to obtain the second signal sequence of each satellite disk. After multiple trigger signals, the sub-signal sequences of each wafer are spliced ​​together in chronological order to obtain the first time sequence combination; the second signal sequences of each satellite disk are spliced ​​together in chronological order to obtain the second time sequence combination; and the measured reflectivity and measured radiation intensity are calculated based on the first time sequence combination and the second time sequence combination.

[0085] Optionally, this application further processes the sub-signal sequences of each wafer into a first time-series curve, and combines the second time-series sequences of each satellite disk into a second time-series curve. The first and second time-series curves are used as input data for a decoupling algorithm. The goal of the decoupling algorithm is to find a unique temperature value that can simultaneously and optimally explain both curves. This is not achieved through direct differentiation, but rather through a physical model-driven "decoupling iterative optimization" process. The system simultaneously analyzes the unique reflectivity curve of each wafer and the thermal radiation curve of each satellite disk, and through continuous assumptions, corrections, and re-assumptions, ultimately finds the unique temperature value and surface state parameters that can perfectly explain both curves simultaneously.

[0086] Optionally, to further improve data accuracy, this application further segments the second signal sequence to obtain a thermal radiation subsequence for each wafer, and then processes the thermal radiation subsequence of each wafer into a second time-series curve. Thus, the data input to the coupling algorithm becomes a unique reflectivity curve and a unique thermal radiation curve for each wafer. For example, the secondary segmentation of the second signal sequence includes: pre-determining the transit time of each wafer based on the wafer size and rotation speed; performing denoising and baseline correction on the second signal sequence; calculating the envelope or integral energy; and, based on preset transit time constraints and intensity thresholds, subdividing the second signal sequence into radiation sub-windows corresponding to each wafer to obtain a thermal radiation subsequence for each wafer.

[0087] The physical forward model is based on optical constant functions and introduces surface state parameters to compensate for actual surface effects. Given temperature and surface state parameters, theoretical reflectivity and theoretical radiation intensity consistent with the detection link can be generated.

[0088] Furthermore, during the alternating iterative optimization process, the surface state parameters related to reflectivity are optimized while the temperature is fixed; then the surface state parameters are fixed while the temperature is optimized, and so on, alternating iteratively, using a nonlinear least squares algorithm to minimize the residual function until convergence.

[0089] Repeat the above steps, using a classification model to process the sub-signal sequence of each wafer to obtain its material type, and then using a decoupling algorithm for iterative optimization to finally obtain the detection value and reflectivity detection value of each wafer.

[0090] The aforementioned data detection method, through signal separation algorithms and machine learning models, enables independent and precise real-time analysis of the temperature and reflectivity of each wafer when different material wafers are placed on a planetary turntable, solving the problem of indistinguishable mixed signals in traditional technologies. Furthermore, this application employs a weighted fusion integrated learning framework, combined with an online learning mechanism, enabling the system to adapt to process changes and the introduction of new materials, continuously optimizing identification accuracy and improving long-term accuracy. Further, the combination of a multivariate decoupling algorithm and an optical constant database improves the in-situ monitoring accuracy of mixed material wafers. The aforementioned intelligent automatic material identification and parameter analysis reduce reliance on operator expertise, improve process repeatability and stability, and facilitate standardized production.

[0091] Optionally, the above data detection method further includes: Displays the temperature and reflectivity values ​​of each wafer as a function of time, and generates an alarm signal when the changes exceed the process range.

[0092] Specifically, this application also provides users with temperature and reflectivity values ​​of each wafer over time through a human-machine interface, which helps users quickly identify anomalies, trigger correction or interlocking in advance, and shorten processing delay.

[0093] Please see Figure 7 The following is combined Figure 7 The data detection method of this application is described in detail: This application first obtains the detection signal on the wafer surface, which includes spectral reflection signal and thermal radiation signal.

[0094] Furthermore, in response to a received trigger signal, this application determines the time interval for each satellite disk to reach a preset position, using the trigger signal as the starting point. Multiple time points are selected within this time interval, and a time window is set centered on each time point. Signal segments corresponding to each time window are obtained from the spectral reflectance signal and the thermal radiation signal. The width of the time window ensures that the detection signals of all wafers on the satellite disk are completely captured. The continuously acquired signal segments belonging to the same satellite disk are spliced ​​together in chronological order to obtain a first signal sequence and a second signal sequence corresponding to each satellite disk.

[0095] Furthermore, this application detects pulse peaks in the first signal sequence; wherein one pulse peak corresponds to one wafer; the start and end points of each pulse peak are determined; based on the start and end points of each pulse peak, the first signal sequence is segmented to obtain multiple signal segments; the signal intensity of each signal segment is calculated, and each signal segment is classified into different categories according to the signal intensity; the continuously acquired signal segments belonging to the same category are spliced ​​together in chronological order to obtain a sub-signal sequence for each wafer. Based on the sub-signal sequence, feature vectors are extracted, including time-domain feature vectors, frequency-domain feature vectors, and model feature vectors. The feature vectors are input into a convolutional network, a support vector machine network, and a random forest network, respectively, and weighted summation is performed. Material matching is performed based on the summation result. If the optimal match is obtained, the material type is determined; otherwise, the classification result is marked and an alarm is triggered.

[0096] Furthermore, this application retrieves an optical model from a pre-defined database based on the material type, employs a physical forward model, and generates predicted data based on the optical constant function of the material under given temperature and surface state parameters. It then progressively calculates the residuals between the predicted and measured data, uses a decoupled alternating optimization algorithm to iteratively optimize the temperature and surface state parameters, and minimizes the residual function using a nonlinear least squares algorithm until convergence. The temperature at which the model converges is used as the temperature detection value, and the measured reflectivity is corrected based on the surface state parameters at convergence to obtain the reflectivity detection value.

[0097] Furthermore, after completing the temperature and reflectivity detection values ​​for one wafer, the above feature extraction, material classification, and decoupling algorithms are repeated to finally obtain the temperature and reflectivity detection values ​​for all wafers.

[0098] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0099] Based on the same inventive concept, this application also provides a data detection system. This device is applicable to the above-described data detection method. The solution provided by this device is similar to the solution described in the above-described method. Therefore, the specific limitations of one or more device embodiments provided below can be found in the limitations of the method above, and will not be repeated here.

[0100] Please see Figure 8 In one embodiment, this application provides a data detection system, including: an acquisition module, a signal segmentation module, and a processing module. The acquisition module is used to acquire detection signals from the wafer surface; wherein each wafer is set in a planetary reaction chamber, and the detection signals include spectral reflection signals and thermal radiation signals.

[0101] The signal segmentation module is used to cut the spectral reflection signal and the thermal radiation signal respectively based on the trigger signal to obtain the first signal sequence and the second signal sequence; wherein, a trigger signal represents the graphite disk in the planetary reaction chamber rotating once.

[0102] The processing module is used to process the first signal sequence using a pre-trained classification model to obtain the material type on each wafer; based on the material type, it queries the preset database to obtain the optical model corresponding to the material type; and uses a decoupling algorithm to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain the temperature detection value and the reflectivity detection value.

[0103] Optionally, the signal segmentation module segments the spectral reflection signal and the thermal radiation signal based on the trigger signal to obtain a first signal sequence and a second signal sequence, including: for each trigger signal, determining the time interval for each satellite disk to reach a preset position, taking the trigger signal as the time starting point; taking multiple time points within the time interval, setting a time window with each time point as the center, and segmenting the signal segment corresponding to each time window from the spectral reflection signal and the thermal radiation signal; the width of the time window is sufficient to ensure that the detection signals of all wafers on the satellite disk are completely captured; and splicing the continuously acquired signal segments belonging to the same satellite disk in chronological order to obtain the first signal sequence and the second signal sequence corresponding to each satellite disk.

[0104] Optionally, when the classification model includes a convolutional network, a support vector machine network, and a random forest network, the processing module uses a pre-trained classification model to process each signal sequence to obtain the material type on each wafer, including: extracting feature vectors representing the optical characteristics of different materials from the first signal sequence; inputting the feature vectors into the convolutional network, the support vector machine network, and the random forest network respectively to obtain the first classification result, the second classification result, and the third classification result; performing a weighted summation on the first classification result, the second classification result, and the third classification result, and determining the material type based on the summation result.

[0105] Optionally, when the feature vector includes a time-domain feature vector and a frequency-domain feature vector, the processing module extracts feature vectors characterizing the optical features of different materials from the first signal sequence, including: cutting the first signal sequence to obtain a sub-signal sequence for each wafer on a satellite disk; extracting a time-domain feature vector based on each sub-signal sequence, wherein the time-domain feature vector includes at least one of the average value, signal intensity, and quantity; performing a Fourier transform on each sub-signal sequence to extract its dominant frequency component, thereby obtaining a frequency-domain feature vector, wherein the frequency-domain feature vector includes the growth rate and / or film thickness.

[0106] Optionally, the processing module segments the first signal sequence to obtain a sub-signal sequence for each wafer on a satellite disk, including: detecting pulse peaks in the first signal sequence; wherein one pulse peak corresponds to one wafer; determining the start and end points of each pulse peak; segmenting the first signal sequence based on the start and end points of each pulse peak to obtain multiple signal segments; calculating the signal intensity of each signal segment and classifying each signal segment into different categories according to the signal intensity; and splicing the continuously acquired signal segments belonging to the same category in chronological order to obtain a sub-signal sequence for each wafer.

[0107] Optionally, if the feature vector also includes a model feature vector, the processing module extracts feature vectors representing the optical characteristics of different materials from the first signal sequence, including: calculating the similarity between the curve of the first signal sequence changing over time and the theoretical prediction curve of each candidate material in the preset database, to obtain multiple initial correlation coefficients, and each initial correlation coefficient constitutes a model feature vector.

[0108] Optionally, the optical model is used to characterize the optical constants as a function of wavelength and temperature, where the optical constants include refractive index and extinction coefficient. The processing module employs a decoupled algorithm to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain temperature and reflectivity detection values. This includes: using a pre-built physical forward model, based on the material's optical constant function, generating predicted data under given temperature and surface state parameters. The predicted data includes theoretical reflectivity and theoretical radiative intensity; the surface state parameters include at least one of an emissivity correction coefficient, surface roughness parameters, and oxide layer influence parameters; progressively calculating the residuals between the predicted and measured data, where the measured data includes measured reflectivity and measured radiative intensity, with the measured reflectivity obtained based on a sub-signal sequence and the measured radiative intensity obtained based on a second signal sequence; using a decoupled alternating optimization algorithm to iteratively optimize the temperature and surface state parameters, and using a nonlinear least squares algorithm to minimize the residual function until convergence; using the temperature at model convergence as the temperature detection value, and correcting the measured reflectivity based on the surface state parameters at convergence to obtain the reflectivity detection value.

[0109] Optionally, the above data detection system also includes a display module.

[0110] The display module is used to display the temperature and reflectivity values ​​of each wafer over time, and to generate an alarm signal when the change curves exceed the process range.

[0111] The aforementioned data detection system, through signal separation algorithms and machine learning models, achieves independent and precise real-time analysis of the temperature and reflectivity of each wafer when different material wafers are placed on a planetary turntable, solving the problem of indistinguishable mixed signals in traditional technologies. Furthermore, this application employs a weighted fusion integrated learning framework, combined with an online learning mechanism, enabling the system to adapt to process changes and the introduction of new materials, continuously optimizing identification accuracy and improving long-term accuracy. Furthermore, the combination of a multivariate decoupling algorithm and an optical constant database improves the in-situ monitoring accuracy of mixed material wafers. The aforementioned intelligent automatic material identification and parameter analysis reduce reliance on operator expertise, improve process repeatability and stability, and facilitate standardized production. Each module in the aforementioned data detection system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0112] In one feasible embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the aforementioned data detection method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0113] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0114] In one feasible embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method steps in the data detection method described above.

[0115] In one feasible embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method steps in the data detection method described above.

[0116] In one feasible embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the method steps in the data detection method described above.

[0117] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0118] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A data detection method, characterized in that, The method includes: Acquire detection signals from the surface of a wafer; wherein each wafer is disposed within a planetary reaction chamber, and the detection signals include spectral reflectance signals and thermal radiation signals; Based on the trigger signal, the spectral reflection signal and the thermal radiation signal are respectively cut to obtain a first signal sequence and a second signal sequence; wherein, one trigger signal represents one rotation of the graphite disk in the planetary reaction chamber; The first signal sequence is processed using a pre-trained classification model to obtain the material type on each wafer; Based on the material type, the optical model corresponding to the material type is obtained by querying a preset database; A decoupling algorithm is used to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain the temperature detection value and the reflectivity detection value.

2. The method according to claim 1, characterized in that, The step of cutting the spectral reflection signal and the thermal radiation signal based on the trigger signal to obtain a first signal sequence and a second signal sequence includes: For each trigger signal, the time interval for each satellite disk to reach the preset position is determined, using the trigger signal as the time starting point. Multiple time points are selected within the time interval, and a time window is set with each time point as the center. The signal segment corresponding to each time window is obtained by cutting from the spectral reflection signal and the thermal radiation signal. The width of the time window is sufficient to ensure that the detection signals of all wafers on the satellite disk are completely captured. The signal segments collected continuously and belonging to the same satellite disk are spliced ​​together in chronological order to obtain the first signal sequence and the second signal sequence for each satellite disk.

3. The method according to claim 1, characterized in that, The classification model includes convolutional networks, support vector machine networks, and random forest networks; The process of using a pre-trained classification model to process the first signal sequence to obtain the material type on each wafer includes: Extract feature vectors characterizing the optical properties of different materials from the first signal sequence; The feature vectors are input into the convolutional network, the support vector machine network, and the random forest network respectively to obtain the first classification result, the second classification result, and the third classification result. The first classification result, the second classification result, and the third classification result are weighted and summed, and the material type is determined based on the summation result.

4. The method according to claim 3, characterized in that, The feature vector includes a time-domain feature vector and a frequency-domain feature vector; The step of extracting feature vectors characterizing the optical features of different materials from the first signal sequence includes: Cut the first signal sequence to obtain a sub-signal sequence for each wafer on a satellite disk; Based on each of the sub-signal sequences, a time-domain feature vector is extracted, wherein the time-domain feature vector includes at least one of the average value, signal strength, and quantity; Perform a Fourier transform on each sub-signal sequence to extract its main frequency component and obtain a frequency domain feature vector, which includes the growth rate and / or film thickness.

5. The method according to claim 4, characterized in that, The step of cutting the first signal sequence to obtain a sub-signal sequence for each wafer on a satellite disk includes: Detect pulse peaks in the first signal sequence; wherein, one pulse peak corresponds to one wafer; Determine the start and end points of each pulse peak; Based on the start and end points of each pulse peak, the first signal sequence is cut to obtain multiple signal segments; Calculate the signal strength of each signal segment and classify the signal segments into different categories based on the signal strength; By splicing together continuously acquired signal segments belonging to the same category in chronological order, a sub-signal sequence for each wafer is obtained.

6. The method according to claim 4, characterized in that, The feature vector also includes the model feature vector; The step of extracting feature vectors characterizing the optical features of different materials from the first signal sequence further includes: The similarity between the curve of the first signal sequence changing over time and the theoretical prediction curve of each candidate material in the preset database is calculated to obtain multiple initial correlation coefficients, and each of the initial correlation coefficients constitutes the model feature vector.

7. The method according to claim 4, characterized in that, The optical model is used to characterize the optical constants as a function of wavelength and temperature, and the optical constants include refractive index and extinction coefficient; The step of using a decoupling algorithm to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain temperature detection values ​​and reflectivity detection values ​​includes: Using a pre-built physical forward model, based on the optical constant function of the material, predictive data is generated under given temperature and surface state parameters; wherein, the predictive data includes theoretical reflectivity and theoretical radiation intensity; the surface state parameters include at least one of emissivity correction coefficient, surface roughness parameter and oxide layer influence parameter; The residual between the predicted data and the measured data is calculated step by step. The measured data includes the measured reflectance and the measured radiation intensity. The measured reflectance is obtained based on the sub-signal sequence, and the measured radiation intensity is obtained based on the second signal sequence. A decoupled alternating optimization algorithm is adopted to iteratively optimize temperature and surface state parameters, and a nonlinear least squares algorithm is used to minimize the residual function until convergence. The temperature at which the model converges is used as the temperature detection value, and the measured reflectivity is corrected based on the surface state parameters at the time of convergence to obtain the reflectivity detection value.

8. The method according to claim 1, characterized in that, The method further includes: The system displays the temperature and reflectivity values ​​of each wafer as a function of time, and generates an alarm signal when the curves exceed the process range.

9. A data detection system, characterized in that, The system includes: An acquisition module is used to acquire detection signals from the surface of a wafer; wherein each wafer is disposed in a planetary reaction chamber, and the detection signals include spectral reflectance signals and thermal radiation signals; The signal segmentation module is used to segment the spectral reflection signal and the thermal radiation signal respectively based on a trigger signal to obtain a first signal sequence and a second signal sequence; wherein, one trigger signal represents one rotation of the graphite disk in the planetary reaction chamber; The processing module is used to process each of the first signal sequences using a pre-trained classification model to obtain the material type on each wafer; based on the material type, it queries a preset database to obtain the optical model corresponding to the material type; and uses a decoupling algorithm to iteratively optimize the optical model, the first signal sequence, and the second signal sequence to obtain the temperature detection value and the reflectivity detection value.

10. A detection device, characterized in that, include: The optical measurement module is used to emit a light beam to the surface of each wafer in the planetary reaction cavity and collect the spectral reflection signal of each wafer surface; The temperature measurement module is used to detect the thermal radiation signals received on the surface of each wafer. A signal line is connected to the spindle encoder or motor controller of the planetary rotary table machine for transmitting trigger signals; wherein, one trigger signal represents one rotation of the graphite disk in the planetary reaction chamber; The data detection system as described in claim 9 is used to periodically process the spectral reflection signal and the thermal radiation signal according to a trigger signal to obtain temperature detection value and reflectance detection value.

11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-8.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-8.

Images