A method and system for quality inspection of flexible integrated circuit packaging substrate

Abstract

The present invention discloses a method and system for quality inspection of a flexible integrated circuit packaging substrate, which relates to the technical field of quality control and inspection. The main scheme is: obtaining pixel data of a flexible integrated circuit packaging substrate through a laser inspection device to clearly present the details of the substrate; performing preprocessing operations based on the obtained pixel data to improve the quality and availability of an image generated by the pixel data obtained based on laser inspection; building a convolutional neural network model to analyze the preprocessed pixel data, which can improve overall production efficiency and reduce labor costs and human errors; generating a quality inspection report based on defect probability values ​​and defect position coordinates to provide a clear and quantitative evaluation basis for the quality of the packaging substrate, which is convenient for tracking and tracing product quality.

Description

Technical Field

[0001] The present invention relates to the technical field of quality control and detection, and in particular to a method and system for detecting quality of a flexible integrated circuit packaging substrate. Background Art

[0002] In recent years, the integrated circuit industry has been growing rapidly, which has led to an increasing demand for quality inspection of integrated circuit packaging substrates to ensure the performance and reliability of integrated circuits. Traditional inspection methods mainly use manual visual inspection, but manual visual inspection has problems such as low inspection efficiency, slow inspection speed, only specific defect inspection, and inability to identify multiple defects at the same time, which makes it difficult to meet actual application needs. In order to improve product quality, reduce production costs, and shorten production cycles, a more efficient, accurate, and comprehensive quality inspection method and system is urgently needed. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] In view of the shortcomings of the prior art, the present invention provides a method and system for quality inspection of flexible integrated circuit packaging substrates, which obtains pixel data of the flexible integrated circuit packaging substrate through a laser detection device; performs preprocessing operations based on the acquired pixel data; builds a convolutional neural network model, analyzes the preprocessed pixel data, and calculates the defect probability value of the packaging substrate surface , and evaluate the quality level of the flexible integrated circuit packaging substrate. When the quality evaluation level of the flexible integrated circuit packaging substrate is unqualified, calculate the defect position coordinates , which solves the problems of low detection efficiency, slow detection speed, only being able to detect specific defects, and being unable to identify multiple types of appearance defects at the same time.

[0005] (II) Technical solution

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for quality inspection of a flexible integrated circuit package substrate comprises the following steps:

[0007] Acquiring pixel data of the flexible integrated circuit package substrate by means of a laser detection device;

[0008] Perform preprocessing operations based on the acquired pixel data;

[0009] Build a convolutional neural network model to analyze the preprocessed pixel data and calculate the defect probability value on the surface of the package substrate , and evaluate the quality level of the flexible integrated circuit packaging substrate. When the quality evaluation level of the flexible integrated circuit packaging substrate is unqualified, calculate the defect position coordinates ;

[0010] According to the defect probability value and defect location coordinates , generate a quality inspection report.

[0011] Preferably, in the above-mentioned flexible integrated circuit package substrate quality inspection method, the pixel data of the flexible integrated circuit package substrate at least includes pixel coordinates And the total number of pixels n; among them, get the pixel coordinates The calculation steps are:

[0012] The laser detection device rotates around the X axis at an angle of , the rotation angle around the Y axis is , the rotation angle around the Z axis is ; According to the rotation angle of the laser detection device around the X axis , Rotation angle around the Y axis , Rotation angle around the Z axis Calculate the rotation matrix , the formula based on is:

[0013] ;

[0014] in, The value range is ;

[0015] According to the rotation matrix The total rotation matrix R is calculated based on the formula:

[0016] ;

[0017] The coordinates of each point on the substrate in the substrate coordinate system are , the coordinates of the laser detection device in the substrate coordinate system are ; According to the coordinates of each point on the substrate in the substrate coordinate system , the coordinates of the laser detection device in the substrate coordinate system , the total rotation matrix R calculates the coordinates of each point on the substrate in the coordinate system of the laser detection device , the formula based on is:

[0018] ;

[0019] The pixel resolution of the image sensor of the laser detection device in the horizontal direction is M, and the pixel resolution in the vertical direction is N; the horizontal size of the image sensor is W, and the vertical size is H; the focal length of the optical system of the laser detection device is According to the pixel resolution M in the horizontal direction, the pixel resolution N in the vertical direction, the horizontal size W and the vertical size H of the image sensor of the laser detection device, the focal length of the optical system of the laser detection device , the coordinates of each point on the substrate in the coordinate system of the laser detection device Calculate the pixel coordinates of each point on the substrate , the formula based on is:

[0020] ;

[0021] in, is the row coordinate; is the column coordinate; the pixel resolution M is 1920; the pixel resolution N is 1080; the horizontal size W and the vertical size H range from 1 mm to 10 mm;

[0022] The pixel coordinates of each point on the substrate The data are collected into a data set to form the pixel coordinate data of the flexible integrated circuit package substrate.

[0023] Preferably, in the above-mentioned flexible integrated circuit package substrate quality inspection method, the method for obtaining the total number of pixels n is:

[0024] The scanning speed of the laser line of the laser detection device on the substrate is V, the pulse frequency of the laser is F, and the diameter of the laser spot is D;

[0025] Calculate the unit time according to the scanning speed V of the laser line on the substrate The distance L of the internal laser scanning is based on the formula:

[0026] ;

[0027] The number of scanned pixels n is calculated based on the laser scanning distance L, the laser pulse frequency F, and the diameter D of the laser spot. The formula is:

[0028] ;

[0029] The scanning speed V range is ; The pulse frequency F has a range of ; The diameter D of the laser spot ranges from 0.01 mm to 0.1 mm;

[0030] Preferably, in the above-mentioned flexible integrated circuit package substrate quality inspection method, the preprocessing operation at least includes image smoothing and histogram equalization; the steps of performing image smoothing are:

[0031] According to pixel coordinates Calculate the output value of the pixel after filtering , the formula based on is:

[0032] ;

[0033] in, It is located in Line The output value of the filtered pixel of the column; Pixel The original value of ; In pixels The neighborhood window centered at is in the range of .

[0034] Preferably, in the above-mentioned flexible integrated circuit package substrate quality inspection method, the formula for performing histogram equalization is:

[0035] ;

[0036] in, The gray level after equalization is The pixel value; H is the total number of gray levels, which is 256; The gray level is The number of pixels.

[0037] Preferably, in the above-mentioned flexible integrated circuit package substrate quality inspection method, analyzing the pre-processed pixel data at least includes: Convert it into a matrix, extract features and compress the preprocessed pixel data; the output value of the filtered pixel point The formula for conversion to matrix A is:

[0038] ;

[0039] in, is the number of rows of pixels after filtering, ranging from 1 to 512; The number of columns of pixels after filtering, ranging from 1 to 512.

[0040] Preferably, in the above-mentioned flexible integrated circuit package substrate quality inspection method, the calculation step of performing feature extraction on the pre-processed pixel data is:

[0041] The formula for convolutional layer feature extraction is:

[0042] ;

[0043] Among them, E is the original image; I is the image after histogram equalization;

[0044] ;

[0045] in, The convolutional layer The layer output feature map is at position The value at ; For the The bias term of the layer; For the The convolution kernel of the layer is at position The weight of ; For the Layer input image at position The value at .

[0046] Preferably, in the above-mentioned flexible integrated circuit package substrate quality inspection method, the calculation step of performing feature compression on the preprocessed pixel data is:

[0047] Calculate the output value after the pooling layer feature is compressed The formula is:

[0048] ;

[0049] in, The output feature map of the pooling layer is at position The value at is the pooling window size; For the Layer input image at position The value at .

[0050] Preferably, in the above-mentioned flexible integrated circuit package substrate quality inspection method, the defect probability value of the package substrate surface is calculated The method for evaluating the quality level of flexible integrated circuit packaging substrates is as follows:

[0051] Build a convolutional neural network model, which contains a variety of image data related to flexible integrated circuit packaging substrates. The image data contains a variety of normal and defective samples. The image data are labeled separately. Normal samples are labeled as defect-free, and defective samples are labeled as: line defects, hole defects, scratches and wear, metal plating defects, and solder mask defects.

[0052] After multiple convolution and pooling layers, the feature vector is obtained. ; The feature vector Input the fully connected layer to calculate the defect probability value of the package substrate surface , the formula based on is:

[0053] ;

[0054] in, Represents the probability of defect occurrence, ranging from 0 to 1. The larger it is, the greater the probability of defects; It is the feature vector extracted from the convolutional and pooling layers, which contains information related to defects in the image; is the weight vector of the fully connected layer; is the bias term; T is the transposed sign;

[0055] The quality level assessment method is as follows:

[0056] when When , it means that at least one of the defects occurs on the surface of the package substrate; when When , it means that there are no defects on the surface of the package substrate; the specific standards are as follows:

[0057] ;

[0058] When the quality assessment level is unqualified, calculate the defect location coordinates , the formula based on is:

[0059] ;

[0060] in, are weight vectors used to calculate U coordinates and V coordinates respectively; They are the bias terms when calculating U coordinates and V coordinates respectively.

[0061] The present invention also discloses a flexible integrated circuit packaging substrate quality inspection system, which is used to implement the above-mentioned quality inspection method, comprising:

[0062] An image acquisition module, used to obtain pixel data of the flexible integrated circuit package substrate through a laser detection device;

[0063] An image processing module, used for performing preprocessing operations according to the acquired pixel data;

[0064] The data analysis module is used to build a convolutional neural network model, analyze the preprocessed pixel data, and calculate the defect probability value of the package substrate surface. , and evaluate the quality level of the flexible integrated circuit packaging substrate. When the quality evaluation level of the flexible integrated circuit packaging substrate is unqualified, calculate the defect position coordinates ;

[0065] Report generation module for and defect location coordinates , generate a quality inspection report.

[0066] (III) Beneficial effects

[0067] (1) Obtaining pixel data of flexible integrated circuit package substrates through laser detection devices helps to generate high-resolution substrate images and clearly present the details of the substrates. Laser detection can flexibly respond to various complex shapes and bending conditions that may exist on flexible substrates to ensure data accuracy.

[0068] (2) Performing image smoothing based on the acquired pixel data can make the sharp edge transition in the image smoother, reduce the visual interference caused by edge mutations, and make it easier to observe and analyze the overall image features; performing histogram equalization can significantly improve the image contrast, so that the originally darker or brighter areas in the image can be displayed more clearly, which helps to detect possible subtle defects or uneven parts in the substrate;

[0069] (3) Building a convolutional neural network model to analyze the preprocessed pixel data can more accurately identify defects and improve the accuracy and reliability of detection. At the same time, it can reduce downtime and rework on the production line and improve overall production efficiency. It can replace some manual inspection work, reduce the demand for a large number of manpower, and reduce labor costs and human errors. By clarifying the location coordinates of the defects, it helps to quickly locate and repair the defects and reduce losses and waste in the production process.

[0070] (4) Generating quality inspection reports helps to analyze the causes and patterns of defects in detail, providing direction for improving production processes; it provides a clear and quantitative evaluation basis for the quality of packaging substrates, making it easier to track and trace product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 The present invention is a schematic diagram of the working steps of a method for detecting quality of a flexible integrated circuit packaging substrate. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0073] See also Figure 1The present invention provides a method for quality inspection of a flexible integrated circuit package substrate, comprising the following steps:

[0074] Step 1: Obtain pixel data of the flexible integrated circuit package substrate through a laser detection device;

[0075] Step 1 includes the following steps:

[0076] Step 101: Get pixel coordinates The calculation steps are:

[0077] According to the rotation angle of the laser detection device around the X axis , Rotation angle around the Y axis , Rotation angle around the Z axis Calculate the rotation matrix , the formula based on is:

[0078] ;

[0079] According to the rotation matrix The total rotation matrix R is calculated based on the formula:

[0080] ;

[0081] According to the coordinates of each point on the substrate in the substrate coordinate system , the coordinates of the laser detection device in the substrate coordinate system , the total rotation matrix R calculates the coordinates of each point on the substrate in the coordinate system of the laser detection device , the formula based on is:

[0082] ;

[0083] According to the pixel resolution M in the horizontal direction, the pixel resolution N in the vertical direction, the horizontal size W and the vertical size H of the image sensor of the laser detection device, the focal length of the optical system of the laser detection device , the coordinates of each point on the substrate in the coordinate system of the laser detection device Calculate the pixel coordinates of each point on the substrate , the formula based on is:

[0084] ;

[0085] It should be noted that obtaining pixel coordinates The principle is: first, the points in the substrate coordinate system need to be transformed into the coordinate system of the laser detection device through the rotation matrix; then, according to the principle of similar triangles in optical imaging, the three-dimensional coordinates are projected onto the two-dimensional plane of the image sensor, and finally the pixel coordinates are obtained. ;

[0086] in, The value range is , determined by the parameters of the motor that controls the rotation; The focal length of the optical system of the detection device is in the range of 1 mm to 20 mm, which is determined by the equipment specifications; the horizontal dimension W and the vertical dimension H are in the range of 1 mm to 10 mm, which are obtained from the actual size of the image sensor;

[0087] Step 102: Get the total number of pixels The method is:

[0088] The scanning speed of the laser line of the laser detection device on the substrate is V, the pulse frequency of the laser is F, and the diameter of the laser spot is D;

[0089] Calculate the unit time according to the scanning speed V of the laser line on the substrate The distance L of the internal laser scanning is based on the formula:

[0090] ;

[0091] According to the distance of laser scanning , laser pulse frequency , the diameter of the laser spot Calculate the number of scanned pixels , the formula based on is:

[0092] ;

[0093] Among them, the scanning speed The value range is , the value range is determined by the performance of the laser scanning equipment; the value range of the pulse frequency F is , the range is determined by the characteristics of the laser generator and obtained through the equipment parameters; the diameter D of the laser spot ranges from 0.01 mm to 0.1 mm, and the range is determined by the laser focusing system;

[0094] By combining the contents of step 101 to step 102:

[0095] Acquiring pixel data of the flexible integrated circuit packaging substrate through a laser detection device can provide sufficient information for subsequent data analysis and processing, which helps to better perform quality inspection; at the same time, the laser has high collimation and high monochromaticity, and can detect tiny defects and subtle structural changes, which is conducive to improving the accuracy of detection.

[0096] Step 2: Perform preprocessing operations based on the acquired pixel data;

[0097] Step 2 includes the following steps:

[0098] Step 201: According to pixel coordinates Calculate the output value of the pixel after filtering , the formula based on is:

[0099] ;

[0100] It should be noted that this formula is an adaptive median filtering algorithm, the principle is that for each pixel in the image , determine a domain window with it as the center , calculate the median of the pixel values ​​in the window as the output value of the pixel after filtering ;in, In pixels The neighborhood window centered on is selected according to the characteristics of the noise and the detail level of the image. The value range is ;

[0101] Step 202: The formula for performing histogram equalization is:

[0102] ;

[0103] It should be noted that this formula calculates the frequency of each gray level and sums them up, then multiplies them by the maximum gray level. , redistribute the grayscale values ​​to make the grayscale distribution of the image more uniform; among them, is the gray level index in the original image, and its value range is ;

[0104] By combining the contents of step 201 to step 202:

[0105] By performing image smoothing on the acquired pixel data, random noise in the image can be effectively reduced, making the image clearer and easier for subsequent analysis and processing; the sharp edge transitions in the image can be made smoother, reducing the visual interference caused by edge mutations, making it easier to observe and analyze the overall image features.

[0106] Step 3: Build a convolutional neural network model to analyze the preprocessed pixel data and calculate the defect probability value on the surface of the package substrate , and evaluate the quality level of the flexible integrated circuit packaging substrate. When the quality evaluation level of the flexible integrated circuit packaging substrate is unqualified, calculate the defect position coordinates ;

[0107] Step 3 includes the following steps:

[0108] Step 301: Output the filtered pixel value The formula for conversion to matrix A is:

[0109] ;

[0110] It should be noted that the construction of the matrix is ​​based on the position arrangement of the pixels in the image; the output value of each pixel after filtering is placed in the corresponding position of the matrix according to its row and column position in the image, and finally a two-dimensional data structure is formed;

[0111] Step 302: Calculation steps for feature extraction of preprocessed pixel data:

[0112] The formula for convolutional layer feature extraction is:

[0113] ;

[0114] It should be noted that Indicates edge feature detection in a specific direction. In the original image, the grayscale of the edge changes slowly. After histogram equalization, the grayscale difference increases, making the edge feature easier to detect in the convolution operation.

[0115] ;

[0116] It should be noted that this formula calculates the inner product between the local area of ​​the input image and the convolution kernel, and adds the bias to obtain each pixel value of the output feature map; For the The bias term of the layer depends on the training data and network structure and is learned through training; For the The convolution kernel of the layer is at position The weight at is obtained through training and the value range is ;

[0117] Step 303: Calculate the output value after pooling layer feature compression The formula is:

[0118] ;

[0119] It should be noted that this formula is passed in the specified area Take the maximum value to achieve feature compression; for each output position of the pooling layer , all from the corresponding upper layer input area Select the maximum value from all the values ​​in as the output value ;

[0120] Step 304: Calculate the defect probability value of the package substrate surface The method for evaluating the quality level of flexible integrated circuit packaging substrates is as follows:

[0121] Build a convolutional neural network model, which contains a variety of image data related to flexible integrated circuit packaging substrates. The image data contains a variety of normal and defective samples. The image data are labeled separately. Normal samples are labeled as defect-free, and defective samples are labeled as: line defects, hole defects, scratches and wear, metal plating defects, and solder mask defects.

[0122] After multiple convolution and pooling layers, the feature vector is obtained. ; The feature vector Input the fully connected layer to calculate the defect probability value of the package substrate surface , the formula based on is:

[0123] ;

[0124] It should be noted that this formula is based on the principle of logistic regression. The defect probability value of the package substrate surface is obtained by linearly combining the input features and weights through a nonlinear logistic function. ;

[0125] in, is the weight vector of the fully connected layer, which represents the importance of the corresponding element in the input feature vector to the final defect probability calculation, and its value range is 0.1 to 0.5; is a bias term used to adjust the result of probability calculation and is learned through training;

[0126] The quality level assessment method is as follows:

[0127] when When , it means that at least one of the defects occurs on the surface of the package substrate; when When , it means that there are no defects on the surface of the package substrate; the specific standards are as follows:

[0128] ;

[0129] Step 305: When the quality assessment level is unqualified, calculate the defect location coordinates , the formula based on is:

[0130] ;

[0131] It should be noted that this formula calculates the defect position coordinates by linear regression, according to the weight vector and the bias term Linear relationship modeling is performed to calculate the defect location coordinates by adjusting these parameters;

[0132] By combining the contents of step 301 to step 305:

[0133] By building a convolutional neural network model, the pre-processed pixel data is analyzed to calculate the defect probability value of the package substrate surface , which is conducive to extracting complex features in the image, thereby more accurately identifying defects and improving the accuracy and reliability of detection; when the quality assessment grade of the flexible integrated circuit package substrate is unqualified, the defect position coordinates are calculated It is helpful to understand the location information of defects and can help analyze possible problems in the production process, so as to optimize the process parameters in a targeted manner and reduce the occurrence rate of defects.

[0134] Step 4: Based on the defect probability value and defect location coordinates , generate quality inspection report;

[0135] Step 4 includes the following steps:

[0136] Step 401: Based on the defect probability value and defect location coordinates , analyze the causes and patterns of defects and understand the overall trend of product quality;

[0137] Step 402: Based on the statistical analysis of defects, specific improvement suggestions are proposed, and a detailed and accurate quality inspection report is generated.

[0138] By combining the contents of step 401 to step 402:

[0139] By using the defect probability value and defect location coordinates , generating quality inspection reports, helps to analyze the causes and patterns of defects in detail and provide direction for improving the production process.

[0140] On the other hand, the present invention also discloses a flexible integrated circuit packaging substrate quality inspection system, which is used to implement the above-mentioned quality inspection method, comprising:

[0141] An image acquisition module, used to obtain pixel data of the flexible integrated circuit package substrate through a laser detection device;

[0142] An image processing module, used for performing preprocessing operations according to the acquired pixel data;

[0143] Used to build a convolutional neural network model, analyze the preprocessed pixel data, and calculate the defect probability value of the package substrate surface , and evaluate the quality level of the flexible integrated circuit packaging substrate. When the quality evaluation level of the flexible integrated circuit packaging substrate is unqualified, calculate the defect position coordinates ;

[0144] Report generation module for and defect location coordinates , generate a quality inspection report.

[0145] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware or in combination with computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0146] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0147] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A method for quality inspection of a flexible integrated circuit package substrate, characterized in that: The following steps are involved: Acquiring pixel data of the flexible integrated circuit package substrate by means of a laser detection device; Perform preprocessing operations based on the acquired pixel data; Build a convolutional neural network model to analyze the preprocessed pixel data and calculate the defect probability value on the surface of the package substrate , and evaluate the quality level of the flexible integrated circuit packaging substrate. When the quality evaluation level of the flexible integrated circuit packaging substrate is unqualified, calculate the defect position coordinates ; According to the defect probability value and defect location coordinates , generate a quality inspection report.

2. A method for quality inspection of a flexible integrated circuit package substrate according to claim 1, characterized in that: The pixel data of the flexible integrated circuit package substrate includes at least pixel coordinates And the total number of pixels n; among them, get the pixel coordinates The calculation steps are: Laser detection device around The axis rotation angle is , the rotation angle around the Y axis is , the rotation angle around the Z axis is ; According to the laser detection device around Axis rotation angle ,around Axis rotation angle , Rotation angle around the Z axis Calculate the rotation matrix , the formula based on is: ; in, The value range is 0 360 ; According to the rotation matrix The total rotation matrix R is calculated based on the formula: ; The coordinates of each point on the substrate in the substrate coordinate system are , the coordinates of the laser detection device in the substrate coordinate system are ; According to the coordinates of each point on the substrate in the substrate coordinate system , the coordinates of the laser detection device in the substrate coordinate system , the total rotation matrix R calculates the coordinates of each point on the substrate in the laser detection device coordinate system , the formula based on is: ; The pixel resolution of the image sensor of the laser detection device in the horizontal direction is M, and the pixel resolution in the vertical direction is N; the horizontal size of the image sensor is W, and the vertical size is H; the focal length of the optical system of the laser detection device is According to the pixel resolution M in the horizontal direction, the pixel resolution N in the vertical direction, the horizontal size W and the vertical size H of the image sensor of the laser detection device, the focal length of the optical system of the laser detection device , the coordinates of each point on the substrate in the coordinate system of the laser detection device Calculate the pixel coordinates of each point on the substrate , the formula based on is: ; in, is the row coordinate; is the column coordinate; the pixel resolution M is 1920; the pixel resolution N is 1080; the horizontal size W and the vertical size H range from 1 mm to 10 mm; The pixel coordinates of each point on the substrate The data are collected into a data set to form the pixel coordinate data of the flexible integrated circuit package substrate.

3. A method for quality inspection of a flexible integrated circuit package substrate according to claim 2, characterized in that: Get the total number of pixels The method is: The scanning speed of the laser line of the laser detection device on the substrate is V, the pulse frequency of the laser is F, and the diameter of the laser spot is D; The distance L scanned by the laser per unit time t is calculated based on the scanning speed V of the laser line on the substrate. The formula is: ; The number of scanned pixels n is calculated based on the laser scanning distance L, the laser pulse frequency F, and the diameter D of the laser spot. The formula is: ; Among them, the scanning speed The value range is ; The pulse frequency F has a range of ; The diameter D of the laser spot ranges from 0.01 mm to 0.1 mm.

4. A method for quality inspection of a flexible integrated circuit package substrate according to claim 3, characterized in that: The preprocessing operation includes at least image smoothing and histogram equalization; the steps of image smoothing are: According to pixel coordinates Calculate the output value of the pixel after filtering , the formula based on is: ; in, It is located in Line The output value of the filtered pixel of the column; Pixel The original value of ; In pixels The neighborhood window centered at is in the range of .

5. A method for quality inspection of a flexible integrated circuit package substrate according to claim 4, characterized in that: The formula for histogram equalization is: ; in, is the pixel value of gray level k after equalization; H is the total number of gray levels, which is 256; is the number of pixels with gray level a.

6. A method for quality inspection of a flexible integrated circuit package substrate according to claim 5, characterized in that: The analysis of the pre-processed pixel data at least includes the output value of the filtered pixel point Convert into matrix, perform feature extraction and feature compression on the preprocessed pixel data; in The output value of the pixel after filtering Convert to matrix The formula is: ； Among them, m is the number of rows of pixels after filtering, and its value range is 1 to 512; r is the number of columns of pixels after filtering, and its value range is 1 to 512.

7. A method for quality inspection of a flexible integrated circuit package substrate according to claim 6, characterized in that: The computational steps for feature extraction of preprocessed pixel data are: The formula for convolutional layer feature extraction is: ； Among them, E is the original image; I is the image after histogram equalization; ; in, The convolutional layer The layer output feature map is at position The value at ; For the The bias term of the layer; For the The convolution kernel of the layer is at position The weight of ; For the Layer input image at position The value at .

8. A method for quality inspection of a flexible integrated circuit package substrate according to claim 7, characterized in that: The calculation steps for feature compression of preprocessed pixel data are: Calculate the output value after the pooling layer feature is compressed The formula is: ; in, The output feature map of the pooling layer is at position The value at is the pooling window size; For the Layer input image at position The value at .

9. A method for quality inspection of a flexible integrated circuit package substrate according to claim 8, characterized in that: Calculate the defect probability value of the package substrate surface The method for evaluating the quality level of flexible integrated circuit packaging substrates is as follows: Build a convolutional neural network model, which contains a variety of image data related to flexible integrated circuit packaging substrates. The image data contains a variety of normal and defective samples. The image data are labeled separately. Normal samples are labeled as defect-free, and defective samples are labeled as: line defects, hole defects, scratches and wear, metal plating defects, and solder mask defects. After multiple convolution and pooling layers, the feature vector is obtained. ; The feature vector Input the fully connected layer to calculate the defect probability value of the package substrate surface , the formula based on is: ; in, Represents the probability of defect occurrence, ranging from 0 to 1. The larger it is, the greater the probability of defects; w is the feature vector extracted from the convolutional layer and the pooling layer, which contains information related to defects in the image; is the weight vector of the fully connected layer; b is the bias term; T is the transposed sign; The quality level assessment method is as follows: when When , it means that at least one of the defects occurs on the surface of the package substrate; when When , it means that there are no defects on the surface of the package substrate; the specific standards are as follows: ; When the quality assessment level is unqualified, calculate the defect location coordinates , the formula based on is: ; in, are weight vectors used to calculate U coordinates and V coordinates respectively; They are the bias terms when calculating U coordinates and V coordinates respectively.

10. A quality inspection system for a flexible integrated circuit package substrate, used to implement the quality inspection method according to any one of claims 1 to 9, characterized in that: include: An image acquisition module, used to obtain pixel data of the flexible integrated circuit package substrate through a laser detection device; An image processing module, used for performing preprocessing operations according to the acquired pixel data; The data analysis module is used to build a convolutional neural network model, analyze the preprocessed pixel data, and calculate the defect probability value of the package substrate surface. , and evaluate the quality level of the flexible integrated circuit packaging substrate. When the quality evaluation level of the flexible integrated circuit packaging substrate is unqualified, calculate the defect position coordinates ; Report generation module for and defect location coordinates , generate a quality inspection report.

Images