Spectroscopic on-line monitoring method for uniformity of coating thickness during deposition of aluminized film

Abstract

This application discloses a method for online spectral monitoring of coating thickness uniformity during aluminum thin film deposition. The method includes: step S1, spectral data acquisition; step S2, data preprocessing and feature extraction; step S3, optical constant and average thickness inversion; step S4, construction of a thickness-color mapping model; and step S5, thickness uniformity inversion and output. This method displays the inverted thickness distribution data in real-time as a curve on a human-machine interface, intuitively presenting the thickness uniformity along the glass width direction. It can also automatically trigger audible and visual alarms based on preset process control limits. Furthermore, this method supports transmitting thickness distribution data to the production line central controller via an industrial communication protocol. Combined with the process parameter-thickness response model, it automatically adjusts sputtering power, magnetic field strength, or process gas flow rate to form a closed-loop negative feedback control, improving the automation level of the production process and the stability of product quality.

Description

Technical Field

[0001] This application belongs to the field of online quality detection and control technology in the production process of aluminum-coated thin films, specifically involving a method for online spectral monitoring of coating thickness uniformity during the deposition process of aluminum-coated thin films. Background Technology

[0002] Metallized films are widely used in packaging materials, decorative films, optical films, and electronic devices due to their excellent metallic luster, barrier properties, and reflective characteristics. In the magnetron sputtering metallization process, the uniformity of the coating thickness directly determines the optical performance (such as reflectivity and color difference) and batch consistency of the product, and is one of the core indicators for measuring coating quality.

[0003] Traditional methods for measuring coating thickness often employ offline sampling, such as step profile measurement, gravimetric analysis, or offline spectral analysis. These methods require sampling and testing after the production line has stopped or the product has rolled off the line, resulting in significant detection lag. They cannot reflect thickness fluctuations during the production process in real time, making it difficult to guide process control promptly and easily leading to batch quality defects.

[0004] In recent years, some production lines have attempted to introduce online spectral measurement devices. However, existing solutions mostly employ a single-probe point-by-point scanning method, using guide rails and stepper motors to drive the probe laterally to collect data. However, for continuously moving coated glass substrates, this method has two inherent drawbacks: first, the single scan cycle is relatively long, making it difficult to meet the real-time requirements of high-speed production lines; second, there is a time difference between measurement points, meaning the obtained data cannot reflect the lateral thickness distribution at the same physical moment, thus introducing systematic errors and affecting the accuracy of uniformity evaluation.

[0005] Furthermore, existing online monitoring methods mostly rely on intuitive comparisons of reflectance spectra, lacking effective means to establish a deterministic mapping relationship between spectral information and physical thickness. Since changes in the thickness of the aluminum film at the nanoscale primarily cause alterations in the phase difference of reflected light, resulting in subtle shifts in the reflected color of the film surface, traditional methods struggle to accurately deduce the absolute thickness value from spectral data, let alone achieve high-precision thickness distribution inversion along the width direction.

[0006] To address the aforementioned issues, there is an urgent need for a method that can achieve high synchronization, high precision, and real-time online monitoring of coating thickness uniformity during the deposition of aluminum thin films. This would overcome the time lag and system errors of traditional offline detection and single-point scanning methods, and provide reliable data support for closed-loop process control. Summary of the Invention

[0007] This application provides a method for online spectral monitoring of coating thickness uniformity during aluminum thin film deposition, aiming to solve the problems existing in the prior art.

[0008] A method for online spectral monitoring of coating thickness uniformity during aluminum thin film deposition, the method comprising: Step S1, Spectral Data Acquisition: Using a fixed multi-channel spectral measurement system, the reflectance spectra of the film surface at multiple measurement points are simultaneously acquired at the exit of the aluminum-coated thin film production line to construct the first dataset. Where i is the measurement point number, Wavelength; Step S2, Data Preprocessing and Feature Extraction: Based on the first dataset Calculate the chromaticity coordinates of each measurement point in the CIEL*a*b* color space, and extract the blue-yellow axis eigenvalues ​​that are sensitive to thickness variations. Construct a second dataset Simultaneously, all spectra in the first dataset are averaged to generate an average reflectance spectrum. ; Step S3, Optical Constants and Average Thickness Inversion: Based on the average reflectance spectrum An optical constant model combined with a genetic algorithm was used to invert the intrinsic optical constants and average thickness of the aluminum film. ; Step S4: Construct a thickness-color mapping model: Based on the intrinsic optical constants, construct a thickness-color mapping relationship. This is used to describe the correspondence between the thickness of the aluminum film layer and its theoretical blue-yellow axis eigenvalues; Step S5, Thickness Uniformity Inversion and Output: Based on the measured values ​​in the second dataset B Values, utilizing the thickness-color mapping relationship The coating thickness at each measurement point was obtained through inversion. Generate a third dataset It outputs the distribution of coating thickness along the width direction.

[0009] Optionally, in step S1, the fixed multi-channel spectral measurement system includes multiple independent measurement probes evenly arranged along the width of the glass. All probes are synchronously triggered to acquire data when the glass reaches a stable measurement position, ensuring that the acquired spectral data has temporal consistency.

[0010] Optionally, in step S2, the blue-yellow axis eigenvalues The chromaticity was calculated based on the CIE15:2004 standard chromaticity calculation method, combined with the D65 standard illuminant and the CIE1931 standard chromaticity observer function.

[0011] Optionally, in step S3, the optical constant model is an extended Cauchy model, used to describe the refractive index of the aluminum film layer. and extinction coefficient The relationship between wavelength and the genetic algorithm is analyzed; the root mean square error between the theoretical spectrum and the measured average spectrum is used as the fitness function for optimization.

[0012] Optionally, in step S4, constructing the thickness-color mapping relationship includes the following sub-steps: With the average thickness Define the thickness analysis range centered on the target area. ; Discrete thickness values ​​are generated within the specified interval using a set step size. ; Based on the intrinsic optical constants, the theoretical reflection spectrum corresponding to each discrete thickness is calculated using the admittance matrix method; Based on the CIE standard colorimetric calculation method, extract the theoretical blue-yellow axis eigenvalues ​​corresponding to each theoretical spectrum. ; Construction Thickness and Theory Lookup tables or continuous functions between .

[0013] Optionally, in step S4, the continuous function The interpolation method is used to construct the interpolation, and self-consistency verification is performed after construction to ensure that the interpolation error is less than a preset threshold.

[0014] Optionally, in step S5, the thickness of each measurement point is inverted. The methods include: In high-speed online monitoring mode, rapid inversion is achieved through a lookup table combined with linear interpolation; In high-precision analysis mode, by solving continuous functions The thickness value is obtained by numerical solution.

[0015] Optionally, step S5 further includes transferring the third dataset. The thickness uniformity is determined and alarms are triggered in real time on the human-machine interface in the form of a curve graph, based on preset process control limits.

[0016] Optionally, step S5 further includes transmitting the third dataset to the production line central controller via an industrial communication protocol to achieve closed-loop feedback control of the coating thickness, including automatically adjusting the sputtering power, magnetic field strength, or process gas flow rate according to the thickness distribution deviation.

[0017] Optionally, the intrinsic optical constants obtained in step S3 are used as fixed inputs in step S4 to construct a thickness-color mapping model, ensuring the physical consistency and repeatability of the model across different batches.

[0018] Compared with the prior art, this application has at least the following beneficial effects: This application installs a fixed multi-channel spectral measurement system at the production line exit. When the glass reaches a stable measurement position, all probes are simultaneously triggered to acquire spectra. This ensures that the acquired data reflects the film surface reflection characteristics of the same transverse line on the glass at the same physical moment, improving the spatiotemporal consistency of the data and providing a high-precision raw data foundation for subsequent thickness uniformity analysis.

[0019] This application selects the blue-yellow axis eigenvalue, which is most sensitive to thickness variations. Using these parameters for inversion, and combining an optical constant model optimized by a genetic algorithm with the admittance matrix method, a thickness-color mapping relationship was constructed. The mapping model is based on the intrinsic optical properties of the aluminum film, has clear physical meaning and good repeatability, and ensures the accuracy and reliability of the thickness inversion results.

[0020] In step S5, this method employs two thickness inversion methods: in high-speed online monitoring mode, a lookup table combined with linear interpolation is used to quickly invert the thickness, meeting the response speed requirements of real-time control of the production line; in high-precision analysis mode, the thickness is inverted by solving a continuous function. The numerical solution yields more accurate thickness values, making it suitable for offline calibration or process optimization scenarios. The flexible switching between the two modes allows this method to achieve optimal performance in various application scenarios.

[0021] This method displays the inverted thickness distribution data in real time as a curve on the human-machine interface, intuitively showing the thickness uniformity along the glass width direction. It can also automatically trigger audible and visual alarms according to preset process control limits. This method supports the transmission of thickness distribution data to the production line central controller through industrial communication protocols. Combined with the process parameter-thickness response model, it automatically adjusts the sputtering power, magnetic field strength, or process gas flow rate to form a closed-loop negative feedback control, which improves the automation level of the production process and the stability of product quality. Attached Figure Description

[0022] Figure 1 A flowchart of the online spectral monitoring method for the uniformity of coating thickness during the aluminum thin film deposition process provided in this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0024] The online spectral monitoring method for the uniformity of coating thickness during the deposition of aluminum thin films provided in this application includes the following steps: S1. Spectral Data Acquisition: In the magnetron sputtering aluminum plating process, the uniformity of the coating thickness directly affects the optical performance and color difference consistency of the product. Traditional offline detection methods have significant lag and cannot meet the requirements of real-time process control. Currently, most online spectral measurement devices employ a single-probe point-by-point scanning method, using a guide rail and stepper motor to drive the probe laterally to collect data. However, for continuously moving coated glass substrates, this method presents two main problems: firstly, the measurement cycle is long; secondly, there is a time difference between each measurement point, causing the acquired data to fail to reflect the lateral thickness distribution at the same moment, thus introducing systematic errors. To address the aforementioned issues, this embodiment designs and installs a fixed multi-channel online spectral measurement system at the outlet of the aluminum plating production line, in the atmospheric environment after the product leaves the vacuum chamber. This system features 24 independent measurement probes evenly arranged along the width of the glass, forming a fixed array. This allows for synchronous data acquisition at the same physical moment, avoiding the systemic errors introduced by time differences in traditional scanning methods. The fixed multi-channel online spectroscopic measurement system consists of 24 independent measurement probes evenly arranged along the width of the glass, forming a fixed array. To ensure the accuracy and repeatability of the measurement data, the optical path of each probe is designed with a perpendicular incident / reflected light method. The light source section adopts a composite light source design, which combines the ultraviolet-visible-near-infrared continuous spectrum through a Y-shaped optical fiber and then guides it into the measurement optical path to cover the target analysis wavelength range. The reflected light is transmitted via optical fiber to the corresponding multi-channel spectrometer (e.g., using multiple CCD array spectrometers with different wavelength ranges for spectroscopic detection). To eliminate the influence of glass substrate jitter or positional deviation on the measurement, a first position sensor is installed on the support in front of the probe array, and a second position sensor is installed on the support behind it. Only when both position sensors simultaneously detect the coated glass under test does the system determine that the glass has moved to a stable measurement position. At this time, all 24 probes are triggered to synchronously perform a spectral acquisition, thereby ensuring that the acquired data reflects the film surface reflection characteristics of the same transverse line on the glass at the same physical moment. Actual production line testing showed that, under conditions where the glass running speed did not exceed 8 m / min, the total time for acquiring spectral data from 24 points (wavelength range 380 nm-780 nm, wavelength sampling interval better than 2.5 nm) in a single run was less than 260 milliseconds. The obtained 24 film surface reflectance spectra constitute the first dataset. ,in Number the measurement points. The wavelength is specified. This dataset provides a highly synchronized and accurate raw data foundation for subsequent refractive index inversion, average spectrum calculation, and thickness uniformity analysis. S2. Data Preprocessing and Feature Extraction: Based on the first dataset constructed in step S1 The process involves spectral data preprocessing and feature parameter extraction. First, wavelength calibration and dark noise subtraction are performed on the raw film reflectance spectral data from all 24 measurement points to ensure metrological consistency across all channels. Then, following the standard colorimetric calculation method specified in the CIE15:2004 technical report, each spectrum is preprocessed and its features are extracted. By performing a weighted integral with the spectral power distribution of the CIE standard illuminant D65 and the tristimulus value function of the CIE 1931 standard colorimetric observer, the coordinates of the measurement point in the CIEL*a*b* uniform color space are calculated. , , Since variations in the aluminum film thickness at the nanoscale primarily cause changes in the phase difference of reflected light, which in turn manifests as a blue-yellow axis shift in the reflected color of the film surface, i.e. This value exhibits the highest sensitivity to this type of thickness perturbation; therefore, this method selects... The values ​​are used as feature parameters for thickness inversion. The 24 obtained feature values ​​constitute the second dataset. This is used for thickness interpolation inversion in the subsequent step S5; Meanwhile, to reduce the impact of single-point measurement noise and local defects on the optical constant inversion, an equal-weighted arithmetic mean was performed on all 24 spectra to generate a characteristic spectrum that can characterize the average optical response of the current glass batch coating. Its calculation expression is: The average spectrum will be used as the target spectrum for optimization by the genetic algorithm in step S3, to invert the intrinsic optical constants and average thickness of the aluminum film, thereby constructing a deterministic mapping relationship between thickness and color feature values; S3. Optical Constants and Average Thickness Inversion: To accurately describe the optical response of the aluminum film in the visible light band, this method uses a complex refractive index dispersion model to characterize its optical constants. Since the refractive index *n* and extinction coefficient *k* of the metal thin film both vary with wavelength and have a significant impact on the amplitude and phase of the film's reflection spectrum, an extended Cauchy model is introduced to describe their dispersion relationship:

[0025] In the formula Wavelength (unit: nm) , , , , , These are the model parameters to be optimized. This model can fit the dispersion characteristics of metal thin films well in the visible light range, and the physical meaning of the parameters is clear, facilitating optimization using a genetic algorithm. The average reflectance spectrum calculated in step S2 To determine the target spectrum, a forward calculation model is constructed based on the thin-film optical admittance matrix method. The aluminum film is considered a homogeneous isotropic thin film, and the substrate is float glass (whose optical constants are known and fixed). Given the film thickness *d* and the six dispersion parameters mentioned above, the corresponding theoretical film surface reflectance spectrum can be calculated. The fitness function is defined as the root mean square error (RMSE) between the theoretical spectrum and the target spectrum.

[0026] Where N is the number of spectral sampling points. The smaller this value, the closer the parameter combination is to the true value; Subsequently, a real-number encoded genetic algorithm was used for global optimization, with parameter settings determined based on a balance between thin-film optical properties and computational efficiency. The population size was set to 60, with 100 generations, and a crossover probability of 0.8 and a mutation probability of 0.05. To accelerate convergence, the two best individuals were retained in each generation. The initial parameter range was determined with reference to the physical properties of aluminum materials and production experience: the search range for refractive index n was set to 0.5–2.0, considering typical values ​​of aluminum films in the visible light range; the search range for extinction coefficient k was 4.0–8.0, covering the actual variation range of magnetron sputtered aluminum films; and the search range for film thickness was set to 20–200 nm, meeting the conventional thickness requirements of industrial aluminum-coated films. In each generation, the admittance matrix was calculated for each individual to obtain the theoretical spectrum, and its RMSE value was calculated. After 100 iterations of optimization, the algorithm converges to the global optimum and outputs the optimal parameter set. and the corresponding average thickness For example, in a certain actual measurement, the average thickness was calculated. =78.3nm, Cauchy model parameters =1.13, =0.25, =0.08, =5.67, =0.34, =0.09, all parameters are within the expected physical range. Substituting the inverted optical constants into the admittance matrix model to recalculate the reflection spectrum, the RMSE with the measured spectrum is only 0.023, verifying the reliability of the inversion results. This set of optimal optical constants will be used as fixed inputs for constructing the thickness-color mapping model in step S4; S4. Construct a thickness-color mapping model: based on the optimal Cauchy dispersion parameters obtained by the genetic algorithm optimization in step S3. , , and extinction coefficient parameters , , This is used as a fixed input for the intrinsic optical constants of the aluminum film layer in the visible light band. Considering that the thickness fluctuation of the aluminum coating layer in actual production is usually controlled within ±5% of the set value, in order to ensure that the mapping model can fully cover the possible thickness deviation range, the average thickness obtained in step S3 is used. Centered on the thickness, extend 5% to both sides as the thickness analysis interval, i.e., set the lower limit of thickness. upper limit Meanwhile, to ensure the accuracy of subsequent interpolation and inversion, a series of discrete thickness values ​​were generated at equal intervals with a step size of 0.2 nm within this range. The step size is selected based on the following: at this step size, the root mean square difference of the theoretical reflectance spectrum corresponding to adjacent thicknesses in the range of 380 to 780 nm is less than 0.1%, which is sufficient to meet the accuracy requirements of industrial online detection. For each discrete thickness value The theoretical reflectance spectrum corresponding to each thickness value was calculated one by one using the admittance matrix method (also known as the transfer matrix method) in thin-film optics. Specifically, the aluminized glass sample to be analyzed was regarded as an optical system composed of three media: the incident medium was air (refractive index...). =1.0, extinction coefficient =0), the middle layer is an aluminum film layer (thickness is 0), The optical constants are obtained from the optimal Cauchy model parameters in step S3. The substrate is float glass (the optical constants are known and stable; in this embodiment, the refractive index is taken as the optimal parameter). =1.52, extinction coefficient =0, this value is calibrated based on the batch measured values ​​of the glass substrates used in production and is used as a fixed input). In terms of the interface model, the aluminum film layer and air, and the aluminum film layer and glass substrate are all considered to be in ideal optical contact. The influence of interface roughness, oxide transition layer or interface diffusion effect on reflection characteristics is not considered. This simplification provides sufficient engineering accuracy for the uniformity analysis of nanoscale aluminum film layers in the visible light band. The calculation parameters are set as follows: the incident light wavelength range is 380–780 nm, which is completely consistent with the acquisition range of the online spectral measurement system in step S1; the wavelength sampling interval is set to 2.5 nm to ensure point-to-point matching between the theoretical spectrum and the measured spectrum in subsequent error calculations; the incident light angle is set to perpendicular incidence (0°), consistent with the optical path design of the online measurement probe; polarization effects are not considered, and the reflectance is the average reflectance of natural light. Under the above settings, for each discrete thickness… Based on the admittance recursive formula of thin-film optics, and utilizing thin-film optics theory, the reflection characteristics of aluminum films of different thicknesses can be predicted by calculating the optical interactions between the aluminum film layer and the substrate / air interface. In practical implementation, the admittance matrix method is used to model the three-layer optical system (air-aluminum film-glass). This method can accurately calculate the film surface reflectivity under perpendicular incidence conditions. It closely matches the actual measurement conditions; This calculation process applies to each thickness value. Each operation is performed independently, ultimately generating a theoretical spectral library covering the entire thickness analysis range. , where M is the total number of discrete thickness points. This spectral library serves as input data for subsequent extraction of theoretical color feature values ​​and construction of thickness-color mapping relationships, ensuring that the physical basis of the mapping model is highly consistent with the measured conditions; After obtaining the theoretical reflectance spectrum, the colorimetric calculation method (D65 illuminator, 10° observer) was used exactly the same as in step S2, according to CIE 15:2004 standard, to calculate the colorimetric value for each thickness. Corresponding theoretical color feature value Given The value is most sensitive to changes in the thickness of the aluminum film layer; this method only extracts... As a mapping feature quantity, the thickness is thus obtained. With theory One-to-one correspondence sequence of values ; Furthermore, for all discrete thickness points calculated in step S4... and its corresponding theories Values ​​are used to construct a thickness-color mapping relationship that can be quickly queried. Specifically, the thickness sequence is... Corresponding theory The values ​​are stored in system memory as a two-dimensional data table, forming an offline calibration lookup table. This lookup table contains two columns of data: the first column is the discrete thickness value (step size 0.2 nm), and the second column is the theoretical thickness value calculated using the admittance matrix method and obtained through CIE chromaticity transformation. The lookup table uses a sequential storage data structure, which facilitates the subsequent step S5 based on actual measurements. The value is subjected to binary search and linear interpolation; Considering thickness - The relation exhibits good continuity and monotonicity within local intervals. To further improve query accuracy and eliminate quantization errors caused by discretization, this method also employs cubic spline interpolation to construct a continuous function. The basic principle of cubic spline interpolation is to construct a piecewise cubic polynomial between adjacent discrete points, ensuring the continuity of the function itself and its first and second derivatives at the nodes, thus obtaining a smooth interpolation curve. Compared with linear interpolation, cubic spline interpolation can more accurately reflect thickness. The nonlinear characteristics of the relationship, especially in regions where the curvature of the curve changes greatly (such as near the extreme points of interference), have higher approximation accuracy; To ensure the accuracy and reliability of the constructed mapping model, this method includes a self-consistency verification step. The specific steps are as follows: Randomly select several thickness values ​​from the thickness analysis interval that were not involved in the model construction (e.g., select thickness values...). Using the optimal optical constants obtained in step S3, the theoretical reflection spectrum is calculated using the admittance matrix method, and the corresponding theoretical spectrum is extracted. Values. Simultaneously, these random thickness values ​​are substituted into the constructed continuous function. In the process, the interpolation result is calculated. Define absolute deviation The verification criterion is that the absolute deviation of all randomly sampled points is less than 0.02 chromaticity units. This threshold matches the repeatability accuracy of the online spectral measurement system, ensuring that the error introduced by the mapping model is negligible. After verification, the mapping model is confirmed as the core link of this method and is used in the subsequent step S5 to invert the coating thickness distribution based on the measured color feature values. S5. Thickness Uniformity Inversion and Output: Based on the thickness-color mapping relationship constructed in step S4. This step converts the measured color feature values ​​into physical thickness values, thereby obtaining the thickness distribution of the aluminum coating along the width of the glass. Specifically: First, the second dataset obtained from step S2 The measured values ​​of each measurement point are read sequentially. Value. Because the thickness-color mapping model constructed in step S4 contains both discrete lookup tables and continuous functions. This embodiment employs two query methods based on actual application scenarios: when the system is in high-speed online monitoring mode, to meet real-time requirements, a lookup table combined with linear interpolation is preferentially used for fast inversion; when the system is in offline calibration or high-precision analysis mode, a continuous function is called. Iterative solutions are performed to obtain more accurate thickness values; For the lookup table pattern, the specific operation process is as follows: For a given actual test... The value is used to perform a binary search algorithm in the thickness-color lookup table to quickly locate the value that satisfies the requirement. Two adjacent nodes and Considering the thickness step size in the lookup table is 0.2 nm, the adjacent nodes... The difference is typically less than 0.1 chromaticity units, and within this range, the thickness... The relationship exhibits good linearity, therefore a linear interpolation formula is used to calculate the corresponding thickness value:

[0027] For continuous function modes, the equations are solved directly. Newton's iteration method or the bisection method are used in the thickness range The internal search finds a numerical solution that meets the accuracy requirements, and the iteration terminates when the thickness change is less than 0.01 nm or the residual is less than 10. -4 One colorimetric unit; The above inversion operation was performed sequentially on all 24 measurement points to obtain a set of thickness values. This constitutes the third dataset. This dataset represents the thickness distribution of the aluminum coating along the width direction of the current glass sample. Taking actual data as an example: assuming the measurement was taken at a certain measurement point... =7.2, the adjacent node in the lookup table is =78.0nm corresponds =7.1, =78.6nm corresponds =7.3, substituting into the linear interpolation formula, we get... =78.0+(7.2-7.1) / (7.3-7.1)*(78.6-78.0)=78.3nm; To visually represent the thickness uniformity, this method displays the third dataset in real-time as a curve on the human-computer interface: the horizontal axis represents the measurement point number (corresponding to the position in the glass width direction), and the vertical axis represents the inverted thickness value. The interface also indicates the target thickness and upper and lower control limits set by the process (e.g., target thickness 78.0 nm, control limit ±3%, i.e., 75.7–80.3 nm). When the thickness deviation at any measurement point exceeds the control limit, the system automatically triggers an audible and visual alarm and highlights the abnormal point on the interface. Furthermore, this method supports data interaction with the production line control system to achieve closed-loop feedback control of the coating thickness. Specifically, the third dataset obtained in step S5... Transmitted in real time to the central controller (typically a programmable logic controller (PLC) or distributed control system (DCS) of the coating production line via an industrial Ethernet interface, according to standard Modbus TCP / IP or Profinet communication protocols. The data transmission format uses floating-point arrays, and the data is updated after each glass sample acquisition and inversion. The data refresh cycle is synchronized with the production cycle, typically 10–30 seconds per batch. After receiving the thickness distribution data, the central controller executes a preset uniformity evaluation algorithm: calculating the statistical characteristic values ​​of the thickness at 24 measurement points, including the maximum value. Minimum value ,average value and standard deviation When the thickness value at any measurement point Deviation from target thickness If the thickness exceeds a preset threshold (e.g., ±3%) and three consecutive batches exceed the range, the system determines that there is a systematic thickness deviation in the area and automatically triggers a closed-loop adjustment program. The adjustment program generates corresponding control commands based on a pre-established process parameter-thickness response model. For example, in a magnetron sputtering aluminum plating production line, the thickness distribution along the width direction is mainly affected by factors such as the magnetic field strength distribution of the cathode target, the local process gas flow rate, and the cathode power distribution. If the thickness in the left-side region of the glass (corresponding to measurement points 1-6) is consistently too low, the control system will perform one or a combination of the following operations: 1) Increase the sputtering power of the cathode corresponding to the left region proportionally, with the increment step set to 1% to 2% of the current power. After each adjustment, wait for one control cycle (usually 3 to 5 production batches) to observe the thickness response. 2) Adjust the magnetic field generating device of the left cathode. By changing the relative position of the magnet and the target surface or the excitation current, the magnetic field strength in this area is enhanced, thereby increasing the local sputtering rate. 3) If the production line adopts a multi-stage gas distribution structure, the flow rate of the process gas (argon) in the left area can be appropriately increased to enhance the plasma density in that area; The above adjustment process is a closed-loop negative feedback control: the central controller continuously receives a new batch of thickness distribution data, compares it with the target value, and if the deviation converges, maintains the current parameters; if the deviation still exists, it continues to fine-tune until the thickness of all measurement points returns to the control limits. All parameter changes during the adjustment process are recorded in the system log and displayed in real time on the operation interface for process personnel to monitor.

[0028] 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.

Claims

1. A method for online spectral monitoring of coating thickness uniformity during aluminum thin film deposition, characterized in that, The method includes: Step S1, Spectral Data Acquisition: Using a fixed multi-channel spectral measurement system, the reflectance spectra of the film surface at multiple measurement points are simultaneously acquired at the exit of the aluminum-coated thin film production line to construct the first dataset. Where i is the measurement point number, Wavelength; Step S2, Data Preprocessing and Feature Extraction: Based on the first dataset Calculate the chromaticity coordinates of each measurement point in the CIEL*a*b* color space, and extract the blue-yellow axis eigenvalues ​​that are sensitive to thickness variations. Construct a second dataset Simultaneously, all spectra in the first dataset are averaged to generate an average reflectance spectrum. ; Step S3, Optical Constants and Average Thickness Inversion: Based on the average reflectance spectrum An optical constant model combined with a genetic algorithm was used to invert the intrinsic optical constants and average thickness of the aluminum film. ; Step S4: Construct a thickness-color mapping model: Based on the intrinsic optical constants, construct a thickness-color mapping relationship. This is used to describe the correspondence between the thickness of the aluminum film layer and its theoretical blue-yellow axis eigenvalues; Step S5, Thickness Uniformity Inversion and Output: Based on the measured values ​​in the second dataset B Values, utilizing the thickness-color mapping relationship The coating thickness at each measurement point was obtained through inversion. Generate a third dataset It outputs the distribution of coating thickness along the width direction.

2. The online spectral monitoring method for the uniformity of coating thickness during aluminum thin film deposition according to claim 1, characterized in that, In step S1, the fixed multi-channel spectral measurement system includes multiple independent measurement probes evenly arranged along the width of the glass. All probes are synchronously triggered to acquire data when the glass reaches a stable measurement position, ensuring that the acquired spectral data has temporal consistency.

3. The online spectral monitoring method for the uniformity of coating thickness during aluminum thin film deposition according to claim 1, characterized in that, In step S2, the blue-yellow axis eigenvalues The chromaticity was calculated based on the CIE15:2004 standard chromaticity calculation method, combined with the D65 standard illuminant and the CIE1931 standard chromaticity observer function.

4. The online spectral monitoring method for the uniformity of coating thickness during aluminum thin film deposition according to claim 1, characterized in that, In step S3, the optical constant model is the extended Cauchy model, used to describe the refractive index of the aluminum film. and extinction coefficient The relationship between wavelength and the genetic algorithm is analyzed; the root mean square error between the theoretical spectrum and the measured average spectrum is used as the fitness function for optimization.

5. The online spectral monitoring method for the uniformity of coating thickness during aluminum thin film deposition according to claim 1, characterized in that, In step S4, constructing the thickness-color mapping relationship includes the following sub-steps: With the average thickness The thickness analysis range is set with the center as the center. ; Discrete thickness values ​​are generated within the specified interval using a set step size. ; Based on the intrinsic optical constants, the theoretical reflection spectrum corresponding to each discrete thickness is calculated using the admittance matrix method; Based on the CIE standard colorimetric calculation method, extract the theoretical blue-yellow axis eigenvalues ​​corresponding to each theoretical spectrum. ; Construction Thickness and Theory Lookup tables or continuous functions between .

6. The online spectral monitoring method for the uniformity of coating thickness during aluminum thin film deposition according to claim 5, characterized in that, In step S4, the continuous function The interpolation method is used to construct the interpolation, and self-consistency verification is performed after construction to ensure that the interpolation error is less than a preset threshold.

7. The online spectral monitoring method for the uniformity of coating thickness during aluminum thin film deposition according to claim 1, characterized in that, In step S5, the thickness of each measurement point is inverted. The methods include: In high-speed online monitoring mode, rapid inversion is achieved through a lookup table combined with linear interpolation; In high-precision analysis mode, by solving continuous functions The thickness value is obtained by numerical solution.

8. The online spectral monitoring method for the uniformity of coating thickness during aluminum thin film deposition according to claim 1, characterized in that, Step S5 also includes transferring the third dataset. The thickness uniformity is determined and alarms are triggered in real time on the human-machine interface in the form of a curve graph, based on preset process control limits.

9. The online spectral monitoring method for the uniformity of coating thickness during aluminum thin film deposition according to claim 1, characterized in that, Step S5 further includes transmitting the third dataset to the production line central controller via an industrial communication protocol to achieve closed-loop feedback control of the coating thickness, including automatically adjusting the sputtering power, magnetic field strength, or process gas flow rate based on the thickness distribution deviation.

10. The method for online spectral monitoring of coating thickness uniformity during aluminum thin film deposition according to claim 1, characterized in that, The intrinsic optical constants obtained in step S3 are used as fixed inputs in step S4 to construct a thickness-color mapping model, ensuring the physical consistency and repeatability of the model across different batches.

Images