A method and system for controlling the thickness of a semiconductor tape adhesive layer

Abstract

This invention belongs to the field of adhesive tape thickness control technology, specifically relating to a method and system for controlling the thickness of semiconductor adhesive tape. The method reconstructs a non-uniform set of measurement point coordinates, generates candidate thickness distribution surfaces using multiple interpolation algorithms in parallel, and uses the local gradient and dispersion of the measurement point coordinate set as input. A first fuzzy controller calculates the fusion weights of each candidate thickness distribution surface, and the weighted sum is used to obtain the adhesive layer thickness distribution model. A smoothness index is calculated based on entropy values. The deviation between the model and the target thickness is decomposed into zero-order and higher-order components. The amplitude of the zero-order component, the energy of the higher-order component, and the smoothness index are input into a second fuzzy controller, which outputs a first control signal and a second control signal. This invention can precisely control the adhesive layer thickness, ensuring its uniformity and guaranteeing that the adhesive layer thickness meets specifications.

Description

Technical Field

[0001] This invention belongs to the field of adhesive tape thickness control technology, specifically relating to a method and system for controlling the thickness of semiconductor adhesive tape. Background Technology

[0002] Semiconductor tapes, as a key auxiliary material in semiconductor wafer processing and packaging, directly determine the precision of downstream processes due to the uniformity of their adhesive layer thickness. For example, excessively thick adhesive layers in wafer dicing tape can lead to dicing misalignment, while excessively thinned tape on the back side can affect chip heat dissipation efficiency. Therefore, the industry typically controls their thickness tolerance at the micrometer level. To meet this stringent standard, online thickness control systems are widely used in semiconductor tape production. These systems use a scanning thickness measuring device to reciprocate in the direction perpendicular to the tape's running direction, acquiring real-time adhesive layer thickness data. Based on the deviation between the measured data and the target thickness, the actuators of the coating mechanism are adjusted, forming a closed-loop control to ensure that the adhesive layer thickness meets specifications.

[0003] However, because the probe performs lateral reciprocating scanning while the tape moves at high speed longitudinally, the superposition of these two movements results in a non-uniform, sparse distribution of the acquired measurement points in a zigzag pattern in space. Traditional processing methods typically use fixed interpolation or fitting algorithms to construct a thickness distribution model for the entire tape width. However, these methods have poor universality. When faced with complex thickness fluctuations or noise interference, algorithm mismatch may lead to insufficient model accuracy, artifacts, or overfitting, failing to accurately reflect the actual contour of the adhesive layer. This results in slow convergence and low accuracy in the control process. When the measurement data is sparse or the noise is high, reducing the smoothness and reliability of the constructed model, using high gain for fine-tuning of local contours will amplify the noise, causing frequent oscillations in the actuator and compromising stability. Summary of the Invention

[0004] In view of this, the present invention proposes a method and system for controlling the thickness of semiconductor tape adhesive layer, which solves the problems of poor universality and low model accuracy of the prior art when processing non-uniform sparse measurement data. It also solves the defects of high gain control that amplifies noise, causes system oscillation and destroys stability when the model confidence is low.

[0005] To address the above problems, the present invention proposes a technical solution for controlling the adhesive layer thickness of semiconductor tape:

[0006] A method for controlling the thickness of a semiconductor adhesive tape layer includes the following steps:

[0007] Thickness measurement data along the width direction of the tape is obtained, and the spatial coordinates of the measurement points are reconstructed by combining the tape running speed and scanning cycle to obtain a non-uniform set of measurement point coordinates.

[0008] For the non-uniform measurement point coordinate set, multiple interpolation algorithms are used in parallel to generate multiple candidate thickness distribution surfaces; the local gradient and dispersion of the measurement point coordinate set are used as the input of the first fuzzy controller to calculate the fusion weight for each candidate thickness distribution surface, and the weighted sum is performed to obtain the adhesive layer thickness distribution model; the smoothness index of the current adhesive layer thickness distribution model is calculated based on the entropy value of the fusion weight.

[0009] The deviation distribution between the adhesive layer thickness distribution model and the target thickness is calculated, and the deviation distribution is decomposed into a zero-order component representing the global average deviation and multiple higher-order components representing local contour fluctuations.

[0010] The amplitude of the zero-order component, the energy of the multiple higher-order components, and the smoothness index are used as inputs to the second fuzzy controller. The controller outputs a first control signal for regulating the coating actuator and a second control signal for regulating the contour adjustment mechanism, so as to adjust the global average thickness and local contour of the adhesive layer, respectively. When the smoothness index is higher than a preset threshold, the response gain of the second control signal to the local contour adjustment is reduced.

[0011] Furthermore, the parallel generation of multiple candidate thickness distribution surfaces using various interpolation algorithms includes:

[0012] The inverse distance weighted interpolation algorithm, the kriging interpolation algorithm, and the radial basis function interpolation algorithm are used in parallel to generate the first candidate thickness distribution surface, the second candidate thickness distribution surface, and the third candidate thickness distribution surface, respectively.

[0013] Furthermore, the method for determining the input of the first fuzzy controller is as follows:

[0014] Define a local neighborhood window centered on each measurement point;

[0015] Calculate the difference between the maximum and minimum values ​​of the thickness measurement data within the local neighborhood window, and use this difference as the local gradient of the measurement point;

[0016] Calculate the standard deviation of the Euclidean distance between all measurement points within the local neighborhood window and the geometric center of the local neighborhood window, and use this standard deviation as the local dispersion of the measurement points;

[0017] Calculate the average local gradient and the average local dispersion of all measurement points, and use the two average values ​​as inputs to the first fuzzy controller.

[0018] Furthermore, the method for obtaining the smoothness index is as follows:

[0019] The interpolation algorithms used are defined as follows: This kind The fusion weights of the candidate thickness distribution surfaces generated by the interpolation algorithms are respectively , , , ,and ;

[0020] Using Shannon's entropy formula Calculate the entropy of the set consisting of these fusion weights. ;

[0021] The entropy value As a smoothness index for the current adhesive layer thickness distribution model.

[0022] Furthermore, the method for decomposing the deviation distribution into a zero-order component representing the global average deviation and multiple higher-order components representing local contour fluctuations is as follows:

[0023] A one-dimensional discrete Fourier transform is performed on the deviation distribution data sequence along the center line of the tape width direction; the DC component obtained after the transform is taken as the zero-order component, and the amplitude represents the global average deviation; the remaining AC components are taken as the multiple higher-order components.

[0024] Furthermore, the method for obtaining the energy of the multiple higher-order components is as follows: the amplitudes of the multiple higher-order components in the Fourier transform frequency domain are squared and summed to obtain the total energy value representing the intensity of local contour fluctuations, and this total energy value is used as the energy of the multiple higher-order components.

[0025] Furthermore, when the smoothness index is higher than a preset threshold, the method for reducing the response gain of the second control signal to local contour adjustment is as follows:

[0026] Define the smoothness index as The original response gain is The adjusted new response gain is Set the preset threshold for smoothness as follows: When the calculated smoothness index > At that time, the original response gain The new response gain is adjusted by multiplying by a decay factor. The calculation formula is:

[0027] ;In the formula, The preset maximum smoothness index value, and < .

[0028] The technical solution of the semiconductor tape adhesive layer thickness control system proposed in this invention is as follows:

[0029] A semiconductor adhesive tape adhesive layer thickness control system includes the following modules:

[0030] The acquisition module is used to acquire thickness measurement data along the width direction of the tape, and reconstruct the spatial coordinates of the measurement points by combining the tape running speed and scanning cycle to obtain a non-uniform set of measurement point coordinates.

[0031] The calculation module is used to generate multiple candidate thickness distribution surfaces in parallel using multiple interpolation algorithms on the non-uniform measurement point coordinate set; using the local gradient and dispersion of the measurement point coordinate set as the input of the first fuzzy controller, it calculates the fusion weight for each candidate thickness distribution surface, and performs weighted summation to obtain the adhesive layer thickness distribution model; and calculates the smoothness index of the current adhesive layer thickness distribution model based on the entropy value of the fusion weight.

[0032] The decomposition module is used to calculate the deviation distribution between the adhesive layer thickness distribution model and the target thickness, and decompose the deviation distribution into a zero-order component representing the global average deviation and multiple higher-order components representing local contour fluctuations.

[0033] The adjustment module is used to take the amplitude of the zero-order component, the energy of the multiple higher-order components and the smoothness index as inputs to the second fuzzy controller, and output a first control signal for regulating the coating execution mechanism and a second control signal for regulating the contour adjustment mechanism, so as to adjust the global average thickness and local contour of the adhesive layer respectively; when the smoothness index is higher than a preset threshold, the response gain of the second control signal to the local contour adjustment is reduced.

[0034] Furthermore, the parallel generation of multiple candidate thickness distribution surfaces using various interpolation algorithms includes:

[0035] The inverse distance weighted interpolation algorithm, the kriging interpolation algorithm, and the radial basis function interpolation algorithm are used in parallel to generate the first candidate thickness distribution surface, the second candidate thickness distribution surface, and the third candidate thickness distribution surface, respectively.

[0036] Furthermore, the method for determining the input of the first fuzzy controller is as follows:

[0037] Define a local neighborhood window centered on each measurement point;

[0038] Calculate the difference between the maximum and minimum values ​​of the thickness measurement data within the local neighborhood window, and use this difference as the local gradient of the measurement point;

[0039] Calculate the standard deviation of the Euclidean distance between all measurement points within the local neighborhood window and the geometric center of the local neighborhood window, and use this standard deviation as the local dispersion of the measurement points;

[0040] Calculate the average local gradient and the average local dispersion of all measurement points, and use the two average values ​​as inputs to the first fuzzy controller.

[0041] Furthermore, the method for obtaining the smoothness index is as follows:

[0042] The interpolation algorithms used are defined as follows: This kind The fusion weights of the candidate thickness distribution surfaces generated by the interpolation algorithms are respectively , , , ,and ;

[0043] Using Shannon's entropy formula Calculate the entropy of the set consisting of these fusion weights. ;

[0044] The entropy value As a smoothness index for the current adhesive layer thickness distribution model.

[0045] Furthermore, the method for decomposing the deviation distribution into a zero-order component representing the global average deviation and multiple higher-order components representing local contour fluctuations is as follows:

[0046] A one-dimensional discrete Fourier transform is performed on the deviation distribution data sequence along the center line of the tape width direction; the DC component obtained after the transform is taken as the zero-order component, and the amplitude represents the global average deviation; the remaining AC components are taken as the multiple higher-order components.

[0047] Furthermore, the method for obtaining the energy of the multiple higher-order components is as follows: the amplitudes of the multiple higher-order components in the Fourier transform frequency domain are squared and summed to obtain the total energy value representing the intensity of local contour fluctuations, and this total energy value is used as the energy of the multiple higher-order components.

[0048] Furthermore, when the smoothness index is higher than a preset threshold, the method for reducing the response gain of the second control signal to local contour adjustment is as follows:

[0049] Define the smoothness index as The original response gain is The adjusted new response gain is Set the preset threshold for smoothness as follows: When the calculated smoothness index > At that time, the original response gain The new response gain is adjusted by multiplying by a decay factor. The calculation formula is:

[0050] ;In the formula, The preset maximum smoothness index value, and < .

[0051] The beneficial effects of this invention are as follows: By employing multiple interpolation algorithms in parallel and performing fuzzy weighted fusion, this invention overcomes the problem of model distortion when faced with complex contours or noise using fixed interpolation or fitting algorithms, thus improving the universality of modeling. Furthermore, this invention decomposes thickness deviation into a global average component and a local contour component, achieving decoupled control of the two. This ensures that global thickness adjustment and local contour correction do not interfere with each other, improving the overall response speed and final control accuracy. This invention also utilizes a model smoothness index to evaluate the quality of the constructed model and adjusts the intensity of control on the local contour based on the evaluation results. When the model reliability is low, it can suppress the excessive response of the control system to noise and model errors, enhancing the stability of the control system operation, avoiding ineffective oscillations of the actuator, and ensuring the uniformity of the adhesive layer thickness and that the adhesive layer thickness meets specifications. Attached Figure Description

[0052] Figure 1 This is a flowchart of the steps in the semiconductor tape adhesive layer thickness control method of the present invention;

[0053] Figure 2 This is a schematic diagram showing the Z-shaped distribution of measurement points on the tape surface. Detailed Implementation

[0054] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0055] Specific embodiments of the semiconductor tape adhesive layer thickness control method proposed in this invention:

[0056] like Figure 1 As shown, the method for controlling the adhesive layer thickness of semiconductor tape includes the following steps:

[0057] S1: Obtain thickness measurement data along the width direction of the tape, and reconstruct the spatial coordinates of the measurement points by combining the tape running speed and scanning cycle to obtain a non-uniform set of measurement point coordinates.

[0058] In this step, a laser thickness sensor mounted on a reciprocating scanning frame scans the tape along the width direction X, perpendicular to the machine direction, at a period T, while the tape moves at a constant speed V along the machine direction Y. At any given time t, the width coordinates of the laser thickness sensor probe are recorded. Machine orientation coordinates of the tape and the corresponding thickness value Thus, a set of spatial coordinate points is obtained. , , ,in, =V×t. Due to the superposition of the longitudinal movement of the tape and the transverse scanning movement of the laser thickness sensor probe, the obtained measurement points form a sparse, non-uniform grid distribution in a Z-shape on the tape surface, such as... Figure 2 As shown.

[0059] S2, For the non-uniform measurement point coordinate set, multiple candidate thickness distribution surfaces are generated in parallel using multiple interpolation algorithms; The local gradient and dispersion of the measurement point coordinate set are used as the input of the first fuzzy controller to calculate the fusion weight for each candidate thickness distribution surface, and the weighted sum is performed to obtain the adhesive layer thickness distribution model; The smoothness index of the current adhesive layer thickness distribution model is calculated based on the entropy value of the fusion weight.

[0060] This step involves simultaneously inputting the non-uniform set of measurement point coordinates obtained in S1 into multiple interpolation algorithm models with different characteristics. Different interpolation algorithms have varying effects on handling noise, sparsity, and fluctuations in the data. Each interpolation algorithm independently calculates and outputs a complete candidate thickness distribution surface covering the entire tape width. The purpose of this step is to avoid the limitations of a single interpolation algorithm by generating multiple models in parallel, providing diverse options for subsequent optimal fusion and improving the model's adaptability to complex working conditions.

[0061] Specifically, this embodiment employs inverse distance weighted interpolation, kriging interpolation, and radial basis function interpolation. The inverse distance weighted interpolation is sensitive to local data, the kriging interpolation provides a smooth, globally optimal estimate, and the radial basis function interpolation maintains stability when handling sparse data. The three interpolation modules are computed in parallel. The first interpolation module runs the inverse distance weighted interpolation, assuming that the thickness value at any point can be obtained by a weighted average of the thickness values ​​of neighboring measurement points, with the weight proportional to the reciprocal of the distance, thus generating a first candidate thickness distribution surface. The second interpolation module runs the kriging interpolation, which analyzes the spatial autocorrelation of thickness values ​​among these measurement points, generating a statistically optimal continuous surface, i.e., the second candidate thickness distribution surface. The third module receives the same number of measurement point data points and uses the radial basis function interpolation algorithm to calculate the thickness distribution surface. This algorithm fits the data points through a linear combination of a series of basis functions, making it particularly suitable for handling irregularly distributed samples, thereby generating a third candidate thickness distribution surface.

[0062] A local neighborhood window is defined centered on each measurement point. The difference between the maximum and minimum values ​​of the thickness measurement data within this local neighborhood window is calculated, and this difference is used as the local gradient of that measurement point. The standard deviation of the Euclidean distance between all measurement points within the local neighborhood window and the geometric center of the local neighborhood window is calculated, and this standard deviation is used as the local dispersion of that measurement point. The average of the local gradients and the average of the local dispersions of all measurement points are calculated as two values ​​representing the overall volatility and spatial uniformity of the data. The two average values ​​are used as the input to the first fuzzy controller. The first fuzzy controller has built-in fuzzy rules. If the local gradient is large and the local dispersion is high, the data is considered to be highly volatile and undersampled. In this case, interpolation algorithms with strong smoothing and noise resistance, such as Kriging interpolation, should be given higher weights. If the local gradient is small and the local dispersion is low, the data is considered to be stable and sufficiently sampled. In this case, high-fidelity interpolation algorithms, such as inverse distance weighted interpolation, can be given higher weights.

[0063] Define the first candidate thickness distribution surface as The second candidate thickness distribution surface is The third candidate thickness distribution surface is The controller outputs each candidate surface. , , The corresponding fusion weights are respectively , and Then the expression for the adhesive layer thickness distribution model is: ,and + + =1. Calculated using the Shannon entropy formula. The entropy value of the weight distribution is calculated and used as a smoothness index. A higher entropy value indicates that the fusion weights output by the first fuzzy controller and assigned to each candidate thickness distribution surface are closer to each other and more uniformly distributed. A uniform weight distribution means that the first fuzzy controller is unsure which interpolation algorithm is best, indicating that the quality and reliability of the currently generated adhesive layer thickness distribution model are poor. Conversely, a lower entropy value means that a certain weight is dominant, and the first fuzzy controller is very certain that a certain algorithm is most suitable, indicating good data quality and high model reliability.

[0064] Of course, in other embodiments, if multiple types of interpolation algorithms are used, let the number of types be . ,this The fusion weights of the candidate thickness distribution surfaces generated by the interpolation algorithms are respectively , , , ,and At this point, using the formula Calculate the entropy of the set consisting of these fusion weights. .

[0065] S3, calculate the deviation distribution between the adhesive layer thickness distribution model and the target thickness, and decompose the deviation distribution into a zero-order component representing the global average deviation and multiple higher-order components representing local contour fluctuations.

[0066] In this embodiment, a one-dimensional discrete Fourier transform is performed on the deviation distribution data sequence along the center line of the tape width direction; the DC component obtained after the transform is taken as the zero-order component, and the amplitude represents the global average deviation; the remaining AC components are taken as the multiple higher-order components.

[0067] Specifically, a contour line is extracted from the centerline along the width direction of the tape from the fused thickness distribution surface. Sixty-four points are sampled at equal intervals along this contour line, resulting in a data sequence containing 64 thickness deviation values. This data sequence represents the difference between the actual thickness contour of the tape centerline and the target thickness. A one-dimensional discrete Fourier transform is performed on the data sequence containing 64 points, resulting in a frequency domain sequence containing 64 complex numbers. The first complex number corresponds to a frequency of zero and is called the DC component. The amplitude of this DC component is equal to the average of the original 64 deviation values. For example, if the calculated amplitude is 0.015, it means that the overall average thickness along the tape centerline is 0.015 mm higher than the target thickness; this amplitude represents the zeroth-order component of the global average deviation. The remaining 63 complex numbers in the frequency domain sequence, excluding the first point, correspond to different non-zero frequencies; these are AC components, collectively forming multiple higher-order components. These higher-order components represent various periodic fluctuations along the contour line, such as wavy lines, depressions, or bulges—local shape features.

[0068] In other embodiments, the adhesive layer thickness distribution model is subtracted point by point from the set target thickness value to obtain the deviation distribution surface. The average value of this deviation distribution surface over the entire width direction is taken as the zero-order component, which represents whether the adhesive layer is generally too thick or too thin. Subtracting this zero-order component from the deviation distribution surface yields the remaining local profile deviation. This local profile deviation is then fitted and decomposed using orthogonal basis functions such as Legendre polynomials or Fourier series, resulting in first-order, second-order, third-order, and other components, which are the higher-order components. These components correspond to typical fluctuation patterns in the adhesive layer profile, such as linear tilt, U-shape or inverted U-shape, and W-shape.

[0069] S4, the amplitude of the zero-order component, the energy of the multiple higher-order components, and the smoothness index are used as inputs to the second fuzzy controller, and the outputs a first control signal for regulating the coating execution mechanism and a second control signal for regulating the contour adjustment mechanism, so as to adjust the global average thickness and local contour of the adhesive layer respectively; when the smoothness index is higher than a preset threshold, the response gain of the second control signal to the local contour adjustment is reduced.

[0070] In this step, the amplitudes of multiple higher-order components in the Fourier transform frequency domain are squared and summed to obtain the total energy value representing the severity of local contour fluctuations. This total energy value serves as the energy of the multiple higher-order components. The amplitude of the zero-order component, the energy of all higher-order components, and the smoothness index calculated above are input to the second fuzzy controller. This fuzzy controller outputs two control signals according to a preset rule base. Rule 1: If the amplitude of the zero-order component is large, the first control signal is output to adjust the opening of the coating die head as a whole, quickly eliminating the average thickness deviation. Rule 2: If the energy of the higher-order components is large, the second control signal is output to independently adjust the settings of multiple heating bolts or piezoelectric ceramic sheets on the die lip to counteract the U-shaped or W-shaped contour of the adhesive layer. The controller also considers the smoothness index and sets rule three based on it: if the smoothness index is higher than a preset threshold, it indicates that the current thickness model is of poor quality and has low reliability. In this case, even if the higher-order component energy is large, the second fuzzy controller will multiply the output second control signal by an attenuation coefficient less than 1, thereby reducing the control force on the local contour adjustment mechanism and avoiding over-adjustment of the control system based on an unreliable model, which could lead to oscillations. Conversely, if the smoothness index is low, normal gain is used for control.

[0071] In an optional embodiment, using the energy of the plurality of higher-order components and the smoothness index as inputs to the second fuzzy controller includes:

[0072] The amplitudes of the multiple higher-order components in the Fourier transform frequency domain are squared and summed to obtain the total energy value representing the severity of local contour fluctuations; the total energy value, the amplitude of the zero-order component, and the smoothness index S are used as the inputs of the second fuzzy controller.

[0073] In an optional embodiment, the method for reducing the response gain of the second control signal to local contour adjustment when the smoothness index is higher than a preset threshold is as follows:

[0074] Define the smoothness index as The original response gain is The adjusted new response gain is Set the preset threshold for smoothness as follows: When the calculated smoothness index > At that time, the original response gain The new response gain is adjusted by multiplying by a decay factor. The calculation formula is:

[0075] ;In the formula, The preset maximum smoothness index value, and < .

[0076] This invention improves the universality and accuracy of non-uniform measurement point modeling by using parallel multi-interpolation algorithm fuzzy weighted fusion, decomposes thickness deviation into global and local components to achieve decoupled control, and dynamically adjusts the local control response gain based on the model smoothness index to avoid actuator oscillation. This effectively improves the accuracy, response speed and system stability of semiconductor tape adhesive layer thickness control, and meets the industry's micron-level tolerance requirements.

[0077] Specific embodiments of the semiconductor tape adhesive layer thickness control system proposed in this invention:

[0078] The semiconductor tape adhesive layer thickness control system includes the following modules:

[0079] The acquisition module is used to acquire thickness measurement data along the width direction of the tape, and reconstruct the spatial coordinates of the measurement points by combining the tape running speed and scanning cycle to obtain a non-uniform set of measurement point coordinates.

[0080] The calculation module is used to generate multiple candidate thickness distribution surfaces in parallel using multiple interpolation algorithms on the non-uniform measurement point coordinate set; using the local gradient and dispersion of the measurement point coordinate set as the input of the first fuzzy controller, it calculates the fusion weight for each candidate thickness distribution surface, and performs weighted summation to obtain the adhesive layer thickness distribution model; and calculates the smoothness index of the current adhesive layer thickness distribution model based on the entropy value of the fusion weight.

[0081] The decomposition module is used to calculate the deviation distribution between the adhesive layer thickness distribution model and the target thickness, and decompose the deviation distribution into a zero-order component representing the global average deviation and multiple higher-order components representing local contour fluctuations.

[0082] The adjustment module is used to take the amplitude of the zero-order component, the energy of the multiple higher-order components and the smoothness index as inputs to the second fuzzy controller, and output a first control signal for regulating the coating execution mechanism and a second control signal for regulating the contour adjustment mechanism, so as to adjust the global average thickness and local contour of the adhesive layer respectively; when the smoothness index is higher than a preset threshold, the response gain of the second control signal to the local contour adjustment is reduced.

[0083] It should be noted that the processing procedures of the acquisition module, calculation module, decomposition module and adjustment module are the same as the working procedures of steps S1, S2, S3 and S4 in the above embodiment of the semiconductor tape adhesive layer thickness control method, and will not be described in detail here.

[0084] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method of controlling the thickness of a semiconductor tape adhesive layer, characterized by, The method comprises the following steps: obtaining thickness measurement data along the width direction of the adhesive tape, reconstructing the spatial coordinates of the measurement points by combining the running speed of the adhesive tape and the scanning period, and obtaining a non-uniform set of measurement point coordinates; using multiple interpolation algorithms in parallel to generate multiple candidate thickness distribution surfaces based on the non-uniform set of measurement point coordinates; taking the local gradient and the dispersion of the set of measurement point coordinates as inputs of a first fuzzy controller, calculating the fusion weight of each candidate thickness distribution surface, and performing weighted summation to obtain an adhesive layer thickness distribution model, and calculating the smoothness index of the current adhesive layer thickness distribution model based on the entropy value of the fusion weight; calculating the deviation distribution between the adhesive layer thickness distribution model and the target thickness, and decomposing the deviation distribution into a zero-order component representing the global average deviation and multiple high-order components representing local contour fluctuations; taking the amplitude of the zero-order component, the energy of the multiple high-order components, and the smoothness index as inputs of a second fuzzy controller, and outputting a first control signal for regulating the coating execution mechanism and a second control signal for regulating the contour adjustment mechanism to adjust the global average thickness and the local contour of the adhesive layer, respectively; when the smoothness index is higher than a preset threshold, reducing the response gain of the second control signal to local contour adjustment.

2. The method of claim 1, wherein the thickness of the adhesive layer is controlled by the amount of the adhesive applied to the adhesive layer. The method of using multiple interpolation algorithms in parallel to generate multiple candidate thickness distribution surfaces comprises: using inverse distance weighting interpolation algorithm, Kriging interpolation algorithm and radial basis function interpolation algorithm in parallel to generate first, second and third candidate thickness distribution surfaces, respectively.

3. The method of claim 1, wherein the thickness of the adhesive layer is controlled by the amount of the adhesive applied to the adhesive layer. The determination method of the inputs of the first fuzzy controller is: defining a local neighborhood window centered on each measurement point; calculating the difference between the maximum and minimum values of the thickness measurement data in the local neighborhood window, and taking the difference as the local gradient of the measurement point; calculating the standard deviation of the Euclidean distance between all measurement points in the local neighborhood window and the geometric center of the local neighborhood window, and taking the standard deviation as the local dispersion of the measurement point; calculating the average value of the local gradient and the average value of the local dispersion of all measurement points, and taking the two average values as inputs of the first fuzzy controller.

4. The method of claim 1, wherein the thickness of the adhesive layer is controlled by the amount of the adhesive applied to the adhesive layer. The acquisition method of the smoothness index is: The interpolation algorithm used is defined in total One of the The fusion weights of the candidate thickness distribution surfaces generated by the interpolation algorithm are respectively , , , , and ; Using the Shannon entropy formula Calculating the entropy value of the set of these fusion weights ; The entropy value as a smoothness index of the current glue layer thickness distribution model.

5. The method of claim 1, wherein the thickness of the adhesive layer is controlled by the amount of the adhesive applied to the adhesive layer. The method of decomposing the deviation distribution into a zero-order component representing the global average deviation and multiple high-order components representing local contour fluctuations is: performing one-dimensional discrete Fourier transform on the deviation distribution data sequence along the center line of the width direction of the adhesive tape; taking the direct current component obtained after transformation as the zero-order component, and the amplitude representing the global average deviation; taking the remaining alternating current components as the multiple high-order components.

6. The method of claim 5, wherein the step of applying a layer of adhesive to the carrier web comprises applying a layer of adhesive to the carrier web in a pattern that is substantially uniform across the carrier web. The acquisition method of the energy of the multiple high-order components is: squaring and summing the amplitudes of the multiple high-order components in the Fourier transform frequency domain to obtain a total energy value representing the intensity of local contour fluctuations, and taking the total energy value as the energy of the multiple high-order components.

7. The method of claim 4, wherein the step of applying the adhesive to the carrier web is performed by a roll-to-roll process. When the smoothness index is higher than a preset threshold, the method of reducing the response gain of the second control signal to local contour adjustment is: The smoothness index is defined as , the original response gain is , and the adjusted new response gain is ; The preset threshold of the smoothness is set as When the calculated smoothness index The original response gain is adjusted by multiplying an attenuation factor, and the calculation formula of the new response gain after adjustment is:​ ; in the formula, is a preset maximum smoothness index value, and .​ 8. A semiconductor tape adhesive layer thickness control system, characterized by, comprising the following modules: The acquisition module is configured to acquire thickness measurement data along the width direction of the adhesive tape, reconstruct spatial coordinates of measurement points by combining the running speed of the adhesive tape and a scanning period, and obtain a non-uniform set of measurement point coordinates. The calculation module is configured to generate a plurality of candidate thickness distribution surfaces in parallel by using a plurality of interpolation algorithms on the non-uniform set of measurement point coordinates, calculate a fusion weight for each candidate thickness distribution surface by taking local gradients and dispersion of the set of measurement point coordinates as inputs of a first fuzzy controller, and obtain a thickness distribution model of the adhesive layer by weighted summation, and calculate a smoothness index of the thickness distribution model of the adhesive layer based on an entropy value of the fusion weight. The decomposition module is configured to calculate a deviation distribution between the thickness distribution model of the adhesive layer and a target thickness, and decompose the deviation distribution into a zero-order component representing a global average deviation and a plurality of high-order components representing local profile fluctuations. The adjustment module is configured to take the amplitude of the zero-order component, the energy of the plurality of high-order components, and the smoothness index as inputs of a second fuzzy controller, output a first control signal for regulating a coating execution mechanism and a second control signal for regulating a profile adjustment mechanism, and adjust a global average thickness and a local profile of the adhesive layer, respectively, and reduce a response gain of the second control signal for local profile adjustment when the smoothness index is higher than a preset threshold.

9. The semiconductor tape adhesive layer thickness control system of claim 8, wherein, The plurality of candidate thickness distribution surfaces are generated in parallel by using a plurality of interpolation algorithms, including: The inverse distance weighting interpolation algorithm, the Kriging interpolation algorithm, and the radial basis function interpolation algorithm are used in parallel to generate a first candidate thickness distribution surface, a second candidate thickness distribution surface, and a third candidate thickness distribution surface, respectively.

10. The semiconductor tape adhesive layer thickness control system of claim 8, wherein, The determination method of the inputs of the first fuzzy controller is: A local neighborhood window is defined with each measurement point as the center. The difference between the maximum value and the minimum value of the thickness measurement data in the local neighborhood window is calculated, and the difference is taken as the local gradient of the measurement point. The standard deviation of the Euclidean distance between all measurement points in the local neighborhood window and the geometric center of the local neighborhood window is calculated, and the standard deviation is taken as the local dispersion of the measurement point. The average value of the local gradients and the average value of the local dispersions of all measurement points are calculated, and the two average values are taken as the inputs of the first fuzzy controller.

Images