Exposure machine position error compensation parameter prediction method

Abstract

The invention belongs to the technical field of exposure machines, and particularly relates to an exposure machine position error compensation parameter prediction method which comprises the following steps: acquiring coordinate data of an alignment point and an alignment lens of an exposure machine, and obtaining deviation values of the alignment point in the X direction and the Y direction; the deviation value dXj of the alignment center of the alignment platform in the X direction and the deviation value dYj of the alignment center of the alignment platform in the Y direction are calculated, j is 1 and 2, and an upper mask and a lower mask are represented; performing data conversion on the angle parameters according to the deviation value dXj of the alignment center of the alignment platform in the X direction and the deviation value dYi of the alignment center of the alignment platform in the Y direction to obtain an angle deviation dT between the alignment center points of the upper mask and the lower mask of the exposure machine and the alignment platform; error compensation is carried out, and the processed alignment point of the exposure machine and the alignment lens coordinate data obtained in real time are processed through the position error compensation parameter prediction model to obtain a compensation value of the alignment platform in the alignment process. The alignment time is shortened, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of exposure machines, and specifically relates to a method for predicting position error compensation parameters of an exposure machine. Background Art

[0002] Flexible film exposure equipment is mainly used in the exposure stage of printed circuit board (PCB) manufacturing. The equipment is a precision instrument used to make complex patterns and circuits on film substrates. It uses light radiation to transfer the specified pattern or circuit to the film surface. These patterns are usually used to manufacture flexible electronic devices, including smartphone screens, tablets, flexible displays and sensors. As an important part of circuit board manufacturing, flexible film exposure machines have encountered continuous technical challenges in production accuracy and efficiency. Therefore, how to improve the production accuracy and efficiency of exposure machines is the focus and difficulty of current research. The current prediction methods are generally divided into three categories: Taguchi method, single factor method and methods based on artificial intelligence and machine learning. These methods require not only rich practical experience, but also a large amount of experimental data accumulation and expert knowledge, and the parameter prediction effect fluctuates greatly. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method for predicting position error compensation parameters of an exposure machine, which is used to improve the prediction speed and accuracy of the position error compensation parameters of the exposure machine, thereby improving the accuracy and efficiency of alignment.

[0004] The present invention is achieved in this way.

[0005] A method for predicting position error compensation parameters of an exposure machine, the method comprising:

[0006] Obtain the alignment point and alignment lens coordinate data of the exposure machine, and preliminarily organize the data, perform coordinate system conversion on the acquired alignment point and alignment lens coordinate data of the exposure machine, and convert them into the alignment platform coordinate system, and make a difference between the coordinates before and after the conversion to obtain the deviation values ​​of the alignment point in the X direction and the Y direction;

[0007] The deviation value dX of the alignment platform in the X direction is calculated based on the deviation values ​​of the alignment point in the X and Y directions. j Deviation dY from the alignment center of the alignment platform in the Y direction j , j takes 1 and 2, indicating the upper mask and the lower mask;

[0008] According to the deviation value dX of the alignment center in the X direction of the alignment platform j Deviation dY from the alignment center of the alignment platform in the Y direction i The angle parameters are converted into data to obtain the angle deviation dT between the alignment center point of the upper mask and the lower mask of the exposure machine and the alignment platform;

[0009] fx, fy, ft are introduced to compensate for the errors of dX, dY and dT respectively, so that dX=dY=dT=0 is satisfied; fx is the compensation value of the X direction of the alignment platform during the alignment process, fy is the compensation value of the Y direction of the alignment platform during the alignment process, and ft is the compensation value of the rotation direction of the alignment platform during the alignment process;

[0010] A position error compensation parameter prediction model is set, and the error compensation parameter prediction model is a back propagation BP neural network optimized by a genetic algorithm. The optimal parameters are found through genetic algorithm optimization to obtain an optimized error compensation parameter prediction model, and the optimal parameters include: initial weight and threshold; in the genetic algorithm optimization process, the error compensation fx, fy, ft are used as input parameters, and the evaluation prediction accuracy is used as output parameters, and the output parameters include: the weighted sum of (|E1|-|E2|) and (|θ1|-|θ2|); wherein, the first position error |E1|: the square root of the square difference between the distances dX and dY of the target center point of the alignment platform before the alignment platform position and posture are adjusted, |θ1|: the angular deviation of the connection line of the positioning points before the alignment platform posture is adjusted, the second position error |E2|: the square root of the square difference between the distances dX and dY of the target center point of the alignment platform after the alignment platform position and posture are adjusted, |θ2|: the angular deviation of the connection line of the positioning points after the alignment platform posture is adjusted; the specific values ​​of the output parameters are obtained according to the data processing results of the alignment points in actual situations;

[0011] The optimized error compensation parameter prediction model is trained to obtain a trained error compensation parameter prediction model; the input data used in the training process is the processed alignment point and alignment lens coordinate data of the exposure machine, and the output data is the compensation value of the alignment platform during the alignment process;

[0012] The trained error compensation parameter prediction model is used to process the alignment point and alignment lens coordinate data of the exposure machine obtained in real time to obtain the compensation value of the alignment platform during the alignment process.

[0013] Furthermore, the conversion formula for converting the alignment point coordinates and the alignment lens coordinates to the alignment platform coordinate system is:

[0014] XT=XC i +XT i ·Cos(T i )-YT i ·Sin(T i ) (1)

[0015] YT=XC i +XT i .Sin(T i )+YT i .Cos(T i) (2)

[0016] XT, YT are the coordinates of the data extracted on the alignment platform of the exposure machine, XC i is the coordinate based on the point data, XT i , YT i is the coordinate of the data on the unconverted alignment platform, T i The alignment point deviation angle within the exposure machine alignment platform is obtained by converting the alignment point data into compensation parameters, determining the output parameters, and filtering and reducing the weight of the data, i = 1, 2, 3, 4.

[0017] Furthermore, the deviation value dX of the alignment center of the alignment platform in the X direction j The formula is:

[0018] dX j =(dx 1s +dx 2s +dx 3s +dx 4s ) / 4

[0019] dX j is the deviation value of the alignment center of the alignment platform in the X direction, dx is It is the deviation value of the four alignment points in the X direction of each mask layer, which comes from the coordinates of the alignment points converted to the alignment platform coordinate system.

[0020] Furthermore, the deviation value dY of the alignment center of the alignment platform in the Y direction j The formula is:

[0021] Yj Y j =(dy 1s +dy 2s +dy 3s +dy 4s ) / 4

[0022] dY j is the deviation value of the alignment center of the alignment platform in the Y direction, dY is It is the deviation value of the four alignment points of each mask plate in the Y direction.

[0023] Furthermore, the conversion formula of the angle deviation dT is:

[0024] dY and dX are dY j and dX j The system

[0025] Compared with the prior art, the present invention has the following beneficial effects: the position error compensation of the exposure machine is an effective way to improve the production accuracy of the exposure machine, which can not only improve the parameter selection effect under various circumstances, but also reduce the alignment time and improve the production efficiency. In terms of real-time prediction of compensation parameters based on BPNN optimized by GA, the alignment platform coordinates and the alignment lens coordinates are unified into the same coordinate system, and the required compensation parameters and error models are extracted through data processing. However, there is too much data and complex calculations, which makes it difficult to meet the needs of real-time adjustment of compensation parameters in various productions, thereby affecting the production accuracy and efficiency. It has been verified that the real-time prediction method of compensation parameters based on BPNN optimized by GA proposed in the present invention can effectively predict the position error compensation parameters; compared with the traditional method, it can not only improve the prediction speed, but also improve the production efficiency. This method does not require expert knowledge and rich experience. It can effectively increase the alignment effect, save production time and improve the alignment accuracy only through the processing of alignment platform data, which is of great significance to improving the production efficiency of the exposure machine and saving production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A structural diagram of a real-time prediction method for exposure machine position error compensation parameters based on a genetic algorithm optimized back propagation neural network provided in an embodiment of the present invention; Part (A) describes, from top to bottom, the exposure machine alignment camera coordinate conversion formula, the alignment point data processing method of the alignment platform compensation parameter, and the input and output parameters of the prediction model; Part (B) describes the overall algorithm flow of the prediction model; Part (C) describes the alignment platform rotating platform around an arbitrary point to control the feed amount formula of the motor;

[0027] Figure 2 The alignment platform structure and parameter selection principle diagram provided by the embodiment of the present invention; (a) partially describes the layout of the exposure machine alignment platform; (b) partially describes the actual alignment calculation of the alignment algorithm;

[0028] Figure 3 A comparison diagram of the alignment effect of the method provided in an embodiment of the present invention, based on the alignment process of the center of the circle to the ring during the alignment process;

[0029] Figure 4 An iteration diagram of the Mse value of the iteration result of the method provided in an embodiment of the present invention (a), and an iteration diagram of the Mse value of the alignment compensation effect using a traditional BP neural network (b). DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0031] See also Figure 1 Combination Figure 2 As shown, in order to achieve the exposure machine position error compensation parameter prediction effect based on BPNN optimized by GA, the present invention is carried out according to the following scheme steps:

[0032] Establish the rotation coordinate formula of the alignment platform, obtain the precise coordinates of the points in the platform after rotation, and obtain the conversion formula from the alignment point coordinates and the alignment lens coordinates to the alignment platform coordinate system:

[0033] XT=XC i +XT i ·Cos(T i )-YT i ·Sin(T i ) (1)

[0034] YT=XC i +XT i .Sin(T i )+YT i .Cos(T i ) (2)

[0035] XT, YT are the coordinates of the data extracted on the alignment platform of the exposure machine, XC i is the coordinate based on the point data, XT i , YT i is the coordinate of the data on the unconverted alignment platform, T i For the alignment point deviation angle within the exposure machine alignment platform, the obtained alignment point data is converted into compensation parameters, the output parameters are determined, and the data is screened and reduced in weight;

[0036] The coordinate data of the alignment point and alignment lens of the exposure machine are obtained, and the data are preliminarily sorted. The obtained coordinate data of the alignment point and alignment lens of the exposure machine are converted into coordinate system using formula (1) and formula (2). The coordinates before and after the conversion are subtracted to obtain the deviation values ​​of the alignment point in the X direction and the Y direction, and the converted data are respectively converted into data using formula (3), formula (4) and formula (5):

[0037] dX j =(dx 1s +dx 2s +dx 3s +dx 4s ) / 4 (3)

[0038] dX j is the deviation value of the alignment center of the alignment platform in the X direction, dx isis the deviation value of the four alignment points of each mask plate in the X direction, i=1, 2, 3, 4, which comes from the coordinates of the alignment points converted to the alignment platform coordinate system. In this embodiment, the comprehensive compensation value is selected as 31.2; j is 1 and 2, indicating the upper and lower mask plates;

[0039] The Y direction parameters of the alignment platform are adjusted and converted. The specific formula is as follows:

[0040] dY j =(dy 1s +dy 2s +dy 3s +dy 4s ) / 4 (4)

[0041] dY j is the deviation value of the alignment center of the alignment platform in the Y direction, dy is is the deviation value of the four alignment points of each mask plate in the Y direction, i=1, 2, 3, 4, and the comprehensive compensation value is selected as 3.8 in this embodiment;

[0042] The specific formula for data conversion of angle parameters is as follows:

[0043]

[0044] dT is the angle deviation between the alignment center point of the upper mask and the lower mask of the exposure machine and the alignment platform. The comprehensive compensation value calculated in this embodiment is 0.00029;

[0045] When and only when dX=dY=dT=0, the exposure machine has the best alignment effect. fx, fy, and ft are introduced to perform error compensation for dX, dY, and dT respectively. The ranges of fx, fy, and ft are [-15, 75], [-20, 20], and [-0.0006, 0.0017] respectively. In this embodiment, the corresponding values ​​are selected as 1.7, -4.6, and 0.00059 respectively, and then the output parameters of the position error compensation parameter prediction model are determined;

[0046] Selecting appropriate output parameters of the position error compensation parameter prediction model is the key to evaluating the prediction accuracy. The weighted sum of (|E1|-|E2|) and (|θ1|-|θ2|) is set as the output parameter of the position error compensation parameter prediction model. The first position error |E1| is the square root of the distance difference between dX and dY of the target center point of the alignment platform before the position and posture of the alignment platform is adjusted. |θ1| is the angular deviation of the connection line of the positioning points before the alignment platform posture is adjusted. The second position error |E2| is the square root of the distance difference between dX and dY of the target center point of the alignment platform after the position and posture of the alignment platform is adjusted. |θ2| is the angular deviation of the connection line of the positioning points after the alignment platform posture is adjusted. The specific values ​​of the output parameters are obtained according to the data processing results of the alignment points in the actual situation, and the experimental design and data processing are carried out through the data of the position error compensation parameter prediction model.

[0047] Determine the input parameters as fx, fy, and ft, optimize the X direction, Y direction, and rotation direction of the alignment platform respectively, and the output parameters are the weighted sum of (|E1(-|E2|), (|θ1|-|θ2|). The obtained output parameters are optimized by using a BP neural network combined with a genetic algorithm to optimize the error compensation parameter prediction model to find the optimal parameters, and the optimal parameters include: initial weights and thresholds. The data is divided into a training set and a test set by comparing the training results with the actual values, and then a suitable prediction model is determined; in the present invention, the processed alignment point coordinate data is divided into a training set and a verification set; the training set and the verification set are input into the error compensation parameter prediction model to calculate the real-time error compensation parameters, and the training is stopped until the number of iterations exceeds 1000 or the global optimal solution is reached, and the processed verification set is input into the trained error compensation parameter prediction model to obtain the prediction results, and then the specific parameter selection of the error compensation parameter prediction model is refined;

[0048] In the present invention, a back propagation BP neural network optimized by a genetic algorithm is constructed, and preliminary model parameter selection is performed, and the obtained data is divided into a test set and a training set; based on a real-time prediction method for position error compensation parameters of an exposure machine, the training set data is input into the model, and the prediction results are obtained; in the parameter selection process, the hidden layer parameter of the BP neural network is determined to be 7, the population size of the genetic algorithm is set to 50, the chromosome length is set to 7 according to the hidden layer data, the crossover rate and the selection rate are given according to the actual fitness value, and the mutation rate is selected as 0.01, and a suitable fitness value is given;

[0049] Y i =f sigmoid (∑a ji X l +θ j )

[0050] Y k =fsigmoid (∑b kn X n +α k )

[0051] Sigmoid is the activation function type, Xi is the input layer node vector, Yk is the hidden layer node vector, aji connects the weights of the input layer and the hidden layer, and bkn connects the weights of the hidden layer and the output layer k;

[0052] The selection process uses the roulette wheel method to select individuals with high fitness, the crossover operation increases diversity through single-point crossover, and the mutation operation introduces new mutations. These steps are iterated until the initial weights and thresholds of the BP neural network that meet the requirements of the exposure machine alignment platform are found;

[0053] The network model is parameterized and appropriate BP neural network parameters are configured. The maximum number of iterations is set to 1000, the learning rate is 0.03, the transfer function Tansig defines the input layer to the hidden layer, Pureline defines the transfer function of the hidden layer, and the hidden layer parameters are determined by testing to have 7 layers. The Powell-Beale conjugate gradient is used as the activation function, and the gradient descent principle is used to evaluate the test results. In order to accurately describe the selection of alignment error compensation parameters and the processing method of alignment coordinates, the optimal compensation parameter prediction method is clarified, and the input parameters and output parameters of the BP neural network in the genetic algorithm and iterative optimization parameter back propagation are shown in Table 1.

[0054] Table 1:

[0055]

[0056] According to the verification results of the error compensation parameter prediction model (BPGA prediction model), it can be obtained that the training set data R=0.89304, the test data R=0.89026, and the validation set data R=0.945. The test results confirm that the fitting effect is good, proving that the error compensation parameter prediction model has good robustness in the prediction of compensation parameters, and the prediction results tend to converge and maintain consistency with the training results; the training results and verification results of the position error compensation parameter prediction model are compared. In the data verification simulation process, 700 groups of existing data are selected for training, and 70 groups of data are randomly selected as the verification set. The obtained prediction model results are compared with the traditional back propagation neural network prediction results. Figure 4 As shown in Table 2;

[0057] Table 2:

[0058] method Mse R Gen BPGA Neural Network 0.015133 0.945 50 Traditional BP neural network 0.031252 0.79924 1000

[0059] Table 2 shows the comparison results of the BP neural network optimized by genetic algorithm and the traditional BP neural network in the prediction of position error compensation parameters, and compares them from Mse, R, and the number of iterations. The closer the Mse value is to 0, the higher the prediction accuracy of the model is. The closer the R value is to 1, the better the model fit is. The fewer the number of iterations, the better the model efficiency and prediction accuracy are. From Table 2, it can be seen that the prediction accuracy of BPGA is improved by 52% compared with the traditional BP neural network in terms of prediction accuracy. The regression and stability of model prediction are also better than BP neural network. The number of iterations also quickly reaches the convergence condition and does not fall into the local optimal solution. Therefore, the BP neural network optimized by genetic algorithm is superior to the traditional BP neural network in all aspects.

[0060] See also Figure 3 As shown, the alignment process of the circle center to the ring during the alignment process is superior to the existing technical solutions.

[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting exposure machine position error compensation parameters, characterized in that: The method includes: Obtain the alignment point and alignment lens coordinate data of the exposure machine, and preliminarily organize the data, perform coordinate system conversion on the acquired alignment point and alignment lens coordinate data of the exposure machine, and convert them into the alignment platform coordinate system, and make a difference between the coordinates before and after the conversion to obtain the deviation values ​​of the alignment point in the X direction and the Y direction; The deviation value dX of the alignment platform in the X direction is calculated based on the deviation values ​​of the alignment point in the X and Y directions. j Deviation dY from the alignment center of the alignment platform in the Y direction j , j takes 1 and 2, indicating the upper mask and the lower mask; According to the deviation value dX of the alignment center in the X direction of the alignment platform j Deviation dY from the alignment center of the alignment platform in the Y direction i The angle parameters are converted into data to obtain the angle deviation dT between the alignment center point of the upper mask and the lower mask of the exposure machine and the alignment platform; fx, fy, ft are introduced to compensate for the errors of dX, dY and dT respectively, so that dX=dY=dT=0 is satisfied; fx is the compensation value of the X direction of the alignment platform during the alignment process, fy is the compensation value of the Y direction of the alignment platform during the alignment process, and ft is the compensation value of the rotation direction of the alignment platform during the alignment process; A position error compensation parameter prediction model is set, and the error compensation parameter prediction model is a back propagation BP neural network optimized by a genetic algorithm. The optimal parameters are found through genetic algorithm optimization to obtain an optimized error compensation parameter prediction model, and the optimal parameters include: initial weight and threshold; in the genetic algorithm optimization process, the error compensation fx, fy, ft are used as input parameters, and the evaluation prediction accuracy is used as output parameters, and the output parameters include: the weighted sum of (|E1|-|E2|) and (|θ1|-|θ2|); wherein, the first position error |E1|: the square root of the square difference between the distances dX and dY of the target center point of the alignment platform before the alignment platform position and posture are adjusted, |θ1|: the angular deviation of the connection line of the positioning points before the alignment platform posture is adjusted, the second position error |E2|: the square root of the square difference between the distances dX and dY of the target center point of the alignment platform after the alignment platform position and posture are adjusted, |θ2|: the angular deviation of the connection line of the positioning points after the alignment platform posture is adjusted; the specific values ​​of the output parameters are obtained according to the data processing results of the alignment points in actual situations; The optimized error compensation parameter prediction model is trained to obtain a trained error compensation parameter prediction model; the input data used in the training process is the processed alignment point and alignment lens coordinate data of the exposure machine, and the output data is the compensation value of the alignment platform during the alignment process; The trained error compensation parameter prediction model is used to process the alignment point and alignment lens coordinate data of the exposure machine obtained in real time to obtain the compensation value of the alignment platform during the alignment process.

2. The method for predicting position error compensation parameters of an exposure machine according to claim 1, characterized in that: The conversion formula for converting the alignment point coordinates and the alignment lens coordinates to the alignment platform coordinate system is: XT=XC i +XT i .Cos(T i )-YT i .Sin(T i ) (1) YT=XC i +XT i .Sin(T i )+YT i .Cos(T i ) (2) XT, YT are the coordinates of the data extracted on the alignment platform of the exposure machine, XC i is the coordinate based on the point data, XT i , YT i is the coordinate of the data on the unconverted alignment platform, T i The alignment point deviation angle within the exposure machine alignment platform is obtained by converting the alignment point data into compensation parameters, determining the output parameters, and filtering and reducing the weight of the data, i = 1, 2, 3, 4.

3. The method for predicting position error compensation parameters of an exposure machine according to claim 1, characterized in that: Deviation value dx of the alignment center in the X direction of the alignment platform j The formula is: dX j =(dx 1s +dx 2s +dx 3s +dx 4s ) / 4 dX j is the deviation value of the alignment center of the alignment platform in the X direction, dx is It is the deviation value of the four alignment points in the X direction of each mask layer, which comes from the coordinates of the alignment points converted to the alignment platform coordinate system.

4. The method for predicting position error compensation parameters of an exposure machine according to claim 1, characterized in that: Deviation value dY of the alignment center of the alignment platform in the Y direction j The formula is: of j =(to 1s +to 2s +to 3s +to 4s ) / 4 d Y j is the deviation value of the alignment center of the alignment platform in the Y direction, dy is It is the deviation value of the four alignment points of each mask plate in the Y direction.

5. The method for predicting position error compensation parameters of an exposure machine according to claim 1, characterized in that: The conversion formula for the angular deviation dT is: dY and dX are dY j and dX j A general term for .

Images