Precision optimization method and system for reel-to-reel high-precision projection type exposure machine

Abstract

The invention relates to the technical field of photoetching exposure, and discloses a precision optimization method and system for a reel-to-reel high-precision projection type exposure machine, and the method comprises the steps: obtaining microscopic depth data, material vibration data and an exposure overlapping region; performing energy distribution analysis according to the microscopic depth data to obtain exposure energy distribution; performing frequency domain analysis according to the material vibration data to obtain vibration frequency domain characteristics; performing correlation calculation according to the exposure energy distribution and the vibration frequency domain characteristics to obtain a correlation influence matrix; performing energy accumulation analysis on the exposure overlapping area to obtain an overlapping correction factor; and performing exposure parameter optimization according to the overlap correction factor and the correlation influence matrix, and updating control parameters of the exposure machine. The method has the following effect that the working precision of the reel-to-reel projection type exposure machine can be improved.

Description

Technical Field

[0001] The present invention relates to the field of lithography exposure technology, and particularly to a method and system for optimizing the accuracy of a roll-to-roll high-precision projection exposure machine. Background Art

[0002] With the rapid development of the electronics industry, especially the increasing demand for flexible electronic devices and high-density interconnect circuit boards, the application of roll-to-roll (R2R) high-precision projection exposure machines has become particularly crucial. Such equipment is mainly used to accurately transfer fine patterns onto flexible substrates, such as materials like polyimide, during the manufacturing process. Its process requirements are extremely high, involving nanoscale precision control. The traditional batch processing method can no longer meet the needs of large-scale production, and the R2R technology has become the mainstream choice in the industry due to its advantages of continuity and high efficiency. In this context, how to optimize the accuracy of R2R high-precision projection exposure machines has become an important issue for improving product quality and production efficiency.

[0003] In an existing technology, to improve the accuracy of roll-to-roll high-precision projection exposure machines, a method combining multi-step calibration and real-time feedback control is adopted. First, during the equipment installation stage, initial calibration is carried out using a high-precision laser interferometer to ensure that the parallelism and perpendicularity errors between each moving axis are within the allowable range. Then, during the actual operation process, a sensor network integrated within the system is used to monitor the substrate position, speed, and the distance between the exposure head and the substrate in real time, and these data are fed back to the control system. Based on the collected data, the control system can dynamically adjust the working parameters of each component to compensate for the accuracy deviation caused by factors such as temperature changes and mechanical wear. In addition, modern R2R exposure machines are also equipped with advanced image processing software, which can preprocess the designed pattern before exposure to remove defects or noise that affect the final imaging quality, thereby further improving the accuracy of pattern transfer.

[0004] In the traditional exposure process, it is assumed that the substrate surface is completely flat and rigid, which enables the optical system to be designed based on a fixed focal length. However, in reality, flexible substrates not only have microscopic undulations on the surface due to uneven thickness, but also undergo tensile or compressive deformations under the action of tension. When the exposure head attempts to focus on such a non-ideal surface, its preset focal length parameters cannot adapt to this dynamic change, resulting in blurred or distorted images. Secondly, temperature changes during the processing of the material also cause thermal expansion or contraction effects, further exacerbating the deformation degree of the material surface, and ultimately leading to the problem of insufficient accuracy in roll-to-roll projection exposure machines. Summary of the Invention

[0005] The present invention provides a method and system for optimizing the accuracy of a roll-to-roll high-precision projection exposure machine to improve the working accuracy of the roll-to-roll projection exposure machine.

[0006] In a first aspect, to solve the above technical problems, the present invention provides a method for optimizing the accuracy of a roll-to-roll high-precision projection exposure machine, including:

[0007] Obtaining microscopic depth data, material vibration data, and an exposure overlap area;

[0008] Performing energy distribution analysis based on the microscopic depth data to obtain an exposure energy distribution;

[0009] Performing frequency domain analysis based on the material vibration data to obtain vibration frequency domain characteristics;

[0010] Performing correlation calculation based on the exposure energy distribution and the vibration frequency domain characteristics to obtain a correlation influence matrix;

[0011] Performing energy accumulation analysis on the exposure overlap area to obtain an overlap correction factor;

[0012] Optimizing exposure parameters based on the overlap correction factor and the correlation influence matrix, and updating the exposure machine control parameters.

[0013] In an optional implementation manner, the performing energy distribution analysis based on the microscopic depth data to obtain an exposure energy distribution includes:

[0014] Dividing the microscopic depth data into grid cells with a preset spacing;

[0015] Performing numerical differential calculation based on the grid cells and the microscopic depth data to obtain a depth change amplitude;

[0016] Performing second derivative calculation based on the grid cells and the microscopic depth data to obtain a depth change frequency;

[0017] Performing weighted fusion based on the depth change amplitude and the depth change frequency to obtain a depth gradient value;

[0018] Inputting the depth gradient value and pre-stored exposure parameters into a preset energy compensation model, and outputting an exposure energy result;

[0019] Performing distribution mapping based on the exposure energy result to obtain an exposure energy distribution.

[0020] In an optional implementation manner, the training process of the energy compensation model includes:

[0021] Taking historical microscopic depth data and historical exposure parameters as inputs, and historical exposure results as outputs, divide the training set and validation set through normalization processing;

[0022] Initialize the parameters of the multi-layer perceptron model;

[0023] Optimize the parameters through iterative training;

[0024] After the loss of the validation set converges or reaches the maximum number of iterations, if the error of the test set is lower than the preset threshold, save the model parameters as the final energy compensation model.

[0025] In an alternative embodiment, the frequency domain analysis based on the material vibration data to obtain vibration frequency domain features includes:

[0026] Perform wavelet denoising on the material vibration data to obtain denoised vibration data;

[0027] Perform wavelet transform on the denoised vibration data to obtain vibration frequency domain data;

[0028] Perform dominant frequency identification based on the vibration frequency domain data to obtain the vibration dominant frequency;

[0029] Perform high-frequency interval identification based on the vibration frequency domain data to obtain the vibration high-frequency interval;

[0030] Perform normalization fusion based on the vibration dominant frequency and the vibration high-frequency interval to obtain vibration frequency domain features.

[0031] In an alternative embodiment, the correlation calculation based on the exposure energy distribution and the vibration frequency domain features to obtain the correlation influence matrix includes:

[0032] Perform block processing on the vibration dominant frequency and vibration high-frequency interval extracted from the vibration frequency domain features to form a frequency vector;

[0033] Extract the exposure energy results of each grid cell based on the exposure energy distribution;

[0034] Calculate the correlation coefficient based on the frequency vector and the exposure energy results to obtain the vibration-exposure correlation;

[0035] Perform normalization processing on the vibration-exposure correlation to obtain the normalized correlation;

[0036] Construct the correlation influence matrix according to the normalized correlation according to the correspondence between the grid cells and the frequency vector.

[0037] In an alternative embodiment, the energy accumulation analysis of the exposure overlapping region to obtain the overlap correction factor includes:

[0038] Integral calculation is performed according to the exposure overlap region and the exposure energy distribution to obtain the cumulative exposure energy;

[0039] Correction factor calculation is performed according to the cumulative exposure energy to obtain the attenuation factor and the enhancement factor;

[0040] When the cumulative exposure energy is less than a preset energy threshold, the enhancement factor is taken as the overlap correction factor;

[0041] When the cumulative exposure energy is greater than or equal to the energy threshold, the attenuation factor is taken as the overlap correction factor.

[0042] In an alternative embodiment, the exposure parameter optimization according to the overlap correction factor and the correlation influence matrix to update the exposure machine control parameters includes:

[0043] Weighted fusion is performed according to the overlap correction factor and the correlation influence matrix to obtain a corrected influence matrix;

[0044] Particle swarm parameters are initialized according to the corrected influence matrix;

[0045] When the number of iterations reaches a preset upper limit or the fitness function of the particles meets the constraint conditions, the iteration is completed and the exposure machine control parameters are output.

[0046] In a second aspect, the present invention provides an accuracy optimization system for a roll-to-roll high-precision projection exposure machine, including:

[0047] A data acquisition module for acquiring microscopic depth data, material vibration data, and exposure overlap regions;

[0048] An energy distribution module for performing energy distribution analysis according to the microscopic depth data to obtain an exposure energy distribution;

[0049] A vibration analysis module for performing frequency domain analysis according to the material vibration data to obtain vibration frequency domain characteristics;

[0050] A correlation calculation module for performing correlation calculation according to the exposure energy distribution and the vibration frequency domain characteristics to obtain a correlation influence matrix;

[0051] An overlap correction module for performing energy accumulation analysis on the exposure overlap region to obtain an overlap correction factor;

[0052] A parameter update module for optimizing exposure parameters according to the overlap correction factor and the correlation influence matrix and updating the exposure machine control parameters.

[0053] In a third aspect, the present invention further provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the accuracy optimization method of the roll-to-roll high-precision projection exposure machine described in any one of the above is implemented.

[0054] In a fourth aspect, the present invention further provides a computer-readable storage medium, which includes a stored computer program. When the computer program runs, the device where the computer-readable storage medium is located is controlled to execute the accuracy optimization method of the roll-to-roll high-precision projection exposure machine described in any one of the above.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] (1) The process of obtaining the microscopic depth data, material vibration data, and exposure overlapping area ensures a comprehensive understanding of the material surface characteristics and processing environment. Through precise data acquisition means, including high-resolution sensors and image processing techniques, the subtle changes and vibration conditions on the material surface can be effectively captured, and the overlapping area during the exposure process can be identified. These high-quality data provide a solid foundation for subsequent analysis, improving the reliability and accuracy of the entire system.

[0057] (2) According to the microscopic depth data, energy distribution analysis is performed to obtain the exposure energy distribution. This step uses advanced algorithm models to extract the energy requirement information at different positions from the microscopic depth data, realizing the precise calculation of the exposure energy distribution. This method can not only adapt to complex surface morphologies but also effectively avoid quality problems caused by uneven energy, improving the consistency and quality of the final pattern.

[0058] (3) According to the material vibration data, frequency domain analysis is performed to obtain the vibration frequency domain characteristics. Through the frequency domain analysis of the material vibration data, the system can identify different frequency components of the vibration and their influence on the exposure process. This analysis method based on spectral characteristics helps to optimize the exposure parameters, reduce the interference caused by vibration, and improve the fineness and stability of the pattern.

[0059] (4) According to the exposure energy distribution and the vibration frequency domain characteristics, correlation calculation is performed to obtain the correlation influence matrix. By combining the exposure energy distribution and the vibration frequency domain characteristics, a correlation analysis method is used to generate the correlation influence matrix, revealing the interaction relationship between the two. This quantitative analysis provides a scientific basis for adjusting the exposure parameters, enhancing the adaptability and flexibility of the system.

[0060] (5) Perform energy accumulation analysis on the exposure overlap region to obtain an overlap correction factor. By analyzing the energy accumulation in the exposure overlap region in detail, the system can determine an overlap correction factor to compensate for the energy accumulation effect caused by multiple exposures. This method ensures the exposure uniformity within the overlap region, prevents overexposure or underexposure, and guarantees the consistency and integrity of the image.

[0061] (6) Optimize the exposure parameters according to the overlap correction factor and the correlation influence matrix, and update the exposure machine control parameters. The last step integrates the results of all previous steps, optimizes and adjusts the exposure parameters through an intelligent algorithm, and generates updated exposure machine control parameters. This process not only considers the influence of the overlap correction factor and the correlation influence matrix, but also ensures the optimization of the overall exposure strategy, significantly improves the accuracy and efficiency of the exposure process, and guarantees high-quality manufacturing results. Description of the Drawings

[0062] Figure 1 It is a schematic flowchart of the accuracy optimization method for a roll-to-roll high-precision projection exposure machine provided by the first embodiment of the present invention;

[0063] Figure 2 It is a schematic structural diagram of the accuracy optimization system for a roll-to-roll high-precision projection exposure machine provided by the second embodiment of the present invention. Detailed Embodiments

[0064] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0065] Referring to Figure 1 , the first embodiment of the present invention provides a method for optimizing the accuracy of a roll-to-roll high-precision projection exposure machine, including the following steps:

[0066] S11. Obtain microscopic depth data, material vibration data, and exposure overlap region;

[0067] S12. Perform energy distribution analysis based on the microscopic depth data to obtain the exposure energy distribution;

[0068] S13. Perform frequency domain analysis based on the material vibration data to obtain the vibration frequency domain characteristics;

[0069] S14. Perform correlation calculation based on the exposure energy distribution and the vibration frequency domain characteristics to obtain the correlation influence matrix;

[0070] S15. Perform energy accumulation analysis on the exposure overlap region to obtain an overlap correction factor;

[0071] S16. Optimize the exposure parameters according to the overlap correction factor and the correlation influence matrix, and update the exposure machine control parameters.

[0072] In step S11, obtain the microscopic depth data, material vibration data, and exposure overlap region.

[0073] In one implementation, synchronous acquisition of three types of key data is achieved through an integrated multi-source sensing system: The microscopic depth data is measured by a white-light interference scanner with a 5-nm step accuracy for three-dimensional topography measurement, generating depth information of 4096 sampling points per square centimeter and storing it as a 16-bit floating-point three-dimensional matrix; the material vibration data is captured by triaxial MEMS accelerometers distributed at the drive roller at a sampling rate of 100 kHz, and the vibration transmission path is verified by combining a laser Doppler vibrometer. The original time-domain data is stored as a structured array containing frequency-amplitude-phase parameters after FFT conversion; the exposure overlap region is captured by a high-speed linear array CCD (5000 fps) triggered by an encoder to obtain the real-time position of the moving substrate. Combining the stepping error predicted by the feedforward motion control model, the coordinates of the overlap region are mapped into a two-dimensional probability distribution map with a deviation range of ±15 μm. The three types of data are associated and stored in an industrial-grade time-series database through a unified timestamp, and a spatial coordinate transformation matrix is established to achieve cross-sensor data fusion.

[0074] In a specific embodiment, taking the exposure of a flexible circuit board as an example: When the substrate passes through the exposure station at a speed of 0.2 m / s, the white-light interference scanner scans the surface of the PI film at a 5-μm interval, and a 0.32-μm concave topography is measured in a certain area (stored as a three-dimensional matrix element of X

[128]

[128] =0.32); at the same time, the triaxial MEMS sensor at the drive roller detects an abnormal vibration of 85 Hz / 0.15 g in the X-axis direction (stored as a structure of {frequency: 85.3 Hz, amplitude: 0.147 g, phase: 32°} after FFT conversion). The vibration is verified by a laser Doppler vibrometer and found to cause an amplitude amplification effect of 0.8 μm; the high-speed linear array CCD captures a +0.125-μm offset of the N+1th frame exposure area compared to the theoretical position in real time. After the three types of data are unified in coordinates through the spatial transformation matrix, an associated data set with microsecond-level synchronization accuracy is formed in the time-series database, laying a foundation for subsequent vibration-topography coupling analysis.

[0075] In step S12, perform energy distribution analysis according to the microscopic depth data to obtain the exposure energy distribution.

[0076] In one implementation, the microscopic depth data is divided into grid cells with a preset spacing; numerical differentiation is performed based on the grid cells and the microscopic depth data to obtain the depth change amplitude; second-order derivative calculation is performed based on the grid cells and the microscopic depth data to obtain the depth change frequency; weighted fusion is performed based on the depth change amplitude and the depth change frequency to obtain the depth gradient value; the depth gradient value and the pre-stored exposure parameters are input into a preset energy compensation model to output the exposure energy result; and distribution mapping is performed based on the exposure energy result to obtain the exposure energy distribution.

[0077] It should be noted that before performing the second-order derivative calculation, high-speed smoothing processing is performed on the microscopic depth data to reduce noise interference.

[0078] It should be noted that in energy distribution analysis, first, the three-dimensional microscopic depth data collected by a high-precision white light interferometer is intelligently grid-divided according to process characteristics. For example, for a 50μm×50μm square grid cell (covering approximately 20 sampling points) of a flexible circuit board substrate, when a steep edge similar to a microcrack is detected in a certain area, the system will automatically encrypt the grid to 25μm×25μm to improve the analysis accuracy.

[0079] It should be noted that the fluctuation amplitude of each grid cell is quantified by calculating the maximum depth difference in the four adjacent directions (up, down, left, and right). For example, when the east-west depth difference in a certain polyimide film area reaches 3.2μm, the gradient value in this direction will be marked as a red warning level; at the same time, the severity of the depth change is captured through second-order differential operations. For example, when a periodic fluctuation of 0.05μm per micrometer spacing is detected in a certain 0.5mm 2 area, the system will identify it as a high-frequency tremor characteristic area.

[0080] It should be noted that after fusing the gradient amplitude and the fluctuation frequency in a weight ratio of 6:4, a quantization heat map including the topography complexity classification is generated (such as the flat area is marked as level 1 and the wrinkled area is marked as level 5). This index and parameters such as the spot diameter and depth of focus tolerance of the exposure machine are jointly input into an energy compensation model calibrated based on one hundred thousand sets of experimental data.

[0081] In one implementation, the training process of the energy compensation model includes:

[0082] Using historical microscopic depth data and historical exposure parameters as inputs and historical exposure results as outputs, the training set and the validation set are divided through normalization processing; the parameters of the multi-layer perceptron model are initialized; the parameters are optimized through iterative training; after the loss of the validation set converges or reaches the maximum number of iterations, if the error of the test set is lower than the preset threshold, the model parameters are saved as the final energy compensation model.

[0083] In one embodiment, the training process of the energy compensation model includes: firstly, multi-dimensional preprocessing of the collected historical data is performed, and the three-dimensional microscopic depth data (including surface gradient, curvature and other features) is aligned with the corresponding exposure parameters (such as light source wavelength, mask transmittance, scanning speed) in time and space, and after eliminating the dimension difference through Z-score standardization, the training set, validation set and test set are divided according to the ratio of 7:2:1; then a multi-layer perceptron model including 3 hidden layers is constructed, in which the input layer is designed to have 1024 nodes to accommodate the topological structure parameters extracted by feature engineering, the LeakyReLU activation function is used to prevent the gradient from disappearing, and the network weight is set by the Xavier initialization method; in the iterative training stage, an adaptive moment estimation (Ad The optimizer is used to optimize the model. The initial learning rate is set to 0.001 and the cosine annealing strategy is used for dynamic adjustment. The mean square error loss between the predicted exposure dose and the actual development line width is calculated by forward propagation in each training cycle. At the same time, the L2 regularization term (λ=0.01) is added to suppress overfitting. When the validation set loss decreases by less than 1e-4 for 5 consecutive epochs or reaches the preset maximum number of iterations of 500, the model performance is evaluated using an independent test set. If key indicators (such as energy compensation error rate ≤3%, line width control accuracy ±1.5μm compliance rate ≥98%) meet the preset thresholds, the optimal model parameters are fixed and stored as a compensation engine that can be called by the production environment. At the same time, an online update mechanism is established to continuously incorporate newly collected process data to achieve model iterative optimization.

[0084] In step S13, frequency domain analysis is performed based on the material vibration data to obtain vibration frequency domain characteristics.

[0085] In one implementation, wavelet denoising is performed on the material vibration data to obtain denoised vibration data;

[0086] Performing wavelet transformation on the noise reduction vibration data to obtain vibration frequency domain data;

[0087] Performing main frequency identification according to the vibration frequency domain data to obtain the main frequency of vibration;

[0088] Perform high-frequency interval identification according to the vibration frequency domain data to obtain a vibration high-frequency interval;

[0089] The vibration main frequency and the vibration high-frequency interval are normalized and fused to obtain the vibration frequency domain feature.

[0090] It should be noted that the vibration frequency domain feature extraction process is realized through a multi-level signal processing method: First, wavelet denoising is performed on the original material vibration data. The multi-scale characteristics of the wavelet basis function are used to separate the effective vibration signal from the environmental noise, suppress the high-frequency noise components, and retain the low-frequency main vibration waveform reflecting the material characteristics. Subsequently, wavelet packet transform is performed on the denoised vibration data to decompose the time-domain waveform into multiple sub-bands containing different frequency components, and a spectral distribution map is drawn by calculating the energy proportion of each sub-band. In the main frequency identification stage, the sliding window energy integration method is used to screen out the frequency band with an energy concentration exceeding a set threshold (such as more than 60% of the total frequency band energy) as the dominant vibration frequency range. For high-frequency feature extraction, the coefficient of variation of the instantaneous amplitude of each sub-band is statistically analyzed to identify the high-frequency sensitive area where the amplitude fluctuates violently and the frequency is more than 3 times higher than the fundamental resonance frequency. Finally, the two indicators of the main frequency energy proportion and the high-frequency region fluctuation intensity are normalized and weighted and fused. The main frequency weight coefficient is dynamically adjusted according to the fatigue characteristics of the material (for example, 0.7 for metal materials and 0.5 for composite materials) to form a three-dimensional frequency domain feature vector representing the dynamic response of the material, providing multi-dimensional vibration mode parameters for subsequent fatigue life prediction.

[0091] In one embodiment, examples of identifying the high-frequency sensitive area by statistically analyzing the coefficient of variation include: When a composite material blade operates, a vibration signal with a fundamental resonance frequency of 500 Hz is generated. The denoised vibration data is decomposed into 16 sub-bands through wavelet packet transform. The center frequency of the sub-band numbered 12 is 1800 Hz (more than 3 times the fundamental frequency). The sub-band is analyzed in segments: Taking a 10-second vibration waveform and dividing it into 100 time segments, the degree of amplitude fluctuation within each segment (that is, the proportion of the difference between the maximum and minimum values to the average value) is calculated respectively. It is found that when the blade rotates to a specific angle, the amplitude volatility of this frequency band suddenly increases from the normal 15% to 65%, and the amplitude difference between adjacent segments reaches more than 3 times the normal state. Combining the energy distribution characteristics corresponding to this frequency band (accounting for 28% of the total high-frequency energy), the system determines that the 1800 Hz area is a high-frequency sensitive area. Actual inspection finds that there is a hidden crack damage about 3 cm long at the corresponding position, verifying the effectiveness of this method.

[0092] It should be noted that the coefficient of variation is a relative indicator in statistics used to measure the degree of data dispersion. Its core function is to eliminate the dimension difference and realize the volatility comparison of data sets with different scales or units. Specifically, this indicator characterizes the stability of the data distribution by calculating the ratio of the standard deviation to the average value. For example, the average charging time of a certain type of battery is 2 hours, and the standard deviation is 0.3 hours, then the coefficient of variation is 15%, indicating a 15% relative fluctuation in the charging time; while the average charging time of another battery is 8 hours, and the standard deviation is 0.8 hours, its coefficient of variation is 10%, indicating that although the absolute deviation of the latter is larger, its relative stability is better.

[0093] In step S14, a correlation calculation is performed based on the exposure energy distribution and the vibration frequency domain characteristics to obtain an association influence matrix.

[0094] In one implementation, the main vibration frequency and the high-frequency vibration interval extracted from the vibration frequency domain characteristics are block-processed to form a frequency vector;

[0095] Based on the exposure energy distribution, the exposure energy result of each grid cell is extracted;

[0096] A correlation coefficient calculation is performed based on the frequency vector and the exposure energy result to obtain the vibration-exposure correlation;

[0097] The vibration-exposure correlation is normalized to obtain a normalized correlation;

[0098] According to the normalized correlation and the correspondence between the grid cell and the frequency vector, an association influence matrix is constructed.

[0099] It should be noted that the construction of the correlation influence matrix is achieved through multi-dimensional data fusion: First, the main vibration frequency range and the high-frequency sensitive region are divided into several frequency blocks according to a preset bandwidth (such as every 50 Hz as an analysis unit), forming a multi-dimensional frequency vector containing parameters such as energy proportion and fluctuation intensity; at the same time, the material surface is divided into millimeter-level grids, and the cumulative exposure energy value of each grid point during the laser processing is extracted. Subsequently, a sliding window matching strategy is adopted to perform time-series alignment of the vibration feature sequence corresponding to each frequency block with the exposure energy distribution of the spatial grid, and the initial correlation value is obtained by calculating the synchronization degree of the change trends of the two (for example, whether the sudden increase in vibration energy is accompanied by a sharp drop in exposure energy). Then, the range normalization method is used to map the correlation to the interval from 0 to 1 to eliminate the influence of different dimensions, and a dynamic correction coefficient is set according to the material type (such as increasing the weight of the high-frequency region by 0.3 for metal materials). Finally, the pairwise correlations between 256 grid units and 32 frequency blocks are arranged as a matrix according to the spatial-frequency correspondence relationship, where the rows represent the grid positions and the columns represent the frequency components. The regions where the matrix element values are greater than 0.8 are identified as high-correlation risk regions, providing a quantitative basis for optimizing the laser processing parameters.

[0100] It should be noted that the sliding window matching strategy realizes time-series alignment by dynamically intercepting data segments, and its core lies in setting an analysis window with a fixed time length and gradually shifting it. For example, in laser welding quality monitoring, it is assumed that it is necessary to perform correlation analysis on the infrared thermal imager data (recording the temperature change of the material) at 1000 frames per second and the 200 Hz acceleration signal collected by the vibration sensor. The system sets a window width of 500 milliseconds and slides step by step at a step size of 100 milliseconds. When the window covers the interval from 3.2 seconds to 3.7 seconds, calculate the synchronization of the temperature rise slope within the window (such as increasing from 850 °C to 920 °C) and the increase in vibration energy (such as increasing from 0.5 to 1.2). If the coincidence degree of the change trends of the two exceeds the threshold (such as 80%), it is determined that there is an abnormal coupling of the process parameters during this period. This method of comparing in time segments can effectively capture transient correlation characteristics and avoid masking local anomalies in the overall correlation calculation.

[0101] It should be noted that the range normalization method eliminates the dimension difference through data range compression. The specific operation includes two steps: determining the extreme value boundary and linear scaling. For example, in the detection of lithography materials, the range normalization method realizes multi-parameter comparison by unifying the dimension. Taking three parameters of surface roughness (0.8 - 2.4 nanometers), assembly pressure (50 - 200 micronewtons), and laser welding duration (2 - 8 femtoseconds) as an example: First, calculate the range of each parameter. The range of surface roughness is 1.6 nanometers (2.4 - 0.8), the range of assembly pressure is 150 micronewtons (200 - 50), and the range of welding duration is 6 femtoseconds (8 - 2). Subsequently, subtract the minimum value from the measured value and divide by the range. For example, when the detected assembly pressure is 125 micronewtons, the normalized value is (125 - 50) / 150 = 0.5; if the welding duration is 5 femtoseconds, the normalized value is (5 - 2) / 6 ≈ 0.5. Through this method, parameters with different physical dimensions can be converted into dimensionless values within the range of 0 - 1, enabling each parameter to be within the same magnitude range. This kind of standardization process is crucial in the scenario of multi-sensor data fusion. For example, synchronously inputting normalized parameters such as vibration displacement (nanometer level), temperature (degree Celsius), and drive current (milliamperes) into a neural network model can effectively improve the accuracy of fault diagnosis of lithography equipment.

[0102] In step S15, energy accumulation analysis is performed on the exposure overlap region to obtain an overlap correction factor.

[0103] In one implementation, integral calculation is performed according to the exposure overlap region and the exposure energy distribution to obtain the cumulative exposure energy;

[0104] Correction factor calculation is performed according to the cumulative exposure energy to obtain an attenuation factor and an enhancement factor;

[0105] When the cumulative exposure energy is less than a preset energy threshold, take the enhancement factor as the overlap correction factor;

[0106] When the cumulative exposure energy is greater than or equal to the energy threshold, take the attenuation factor as the overlap correction factor.

[0107] It should be noted that the calculation of the overlap correction factor is realized through a piecewise linear model, and the parameters are calibrated through experiments: First, the energy density of the overlapping area of adjacent light spots (for example, the crescent area formed by the 40% overlap of circular light spots with a diameter of 0.1 mm) is integrated. The specific method is to divide this area into micron-level grids. Each grid point calculates the single-point energy (200 mW × 0.00005 s = 0.00001 J) according to the laser power (such as 200 mW) and the dwell time (such as 50 microseconds), and then accumulates the energy values of all 5000 grid points to obtain the total cumulative exposure energy (for example, 0.01 × 5000 = 0.050 J). For example: In the lithography process, a piecewise function is designed according to different parts of the overlapping area when calculating the correction factor. For example, when two circular light spots with a diameter of 0.2 mm are exposed with 50% overlap, it is actually measured that the energy in the central area reaches 1.5 times the theoretical value due to the superposition of double light spots, while the energy in the edge area only reaches 0.8 times the theoretical value due to energy attenuation. The piecewise model means that two linear functions with different slopes are used for the parts greater than one and less than one. Specifically, multiplying 1 / 1.5 by the experimentally calibrated attenuation slope gives 0.7, and multiplying 1 / 0.8 by the experimentally calibrated enhancement slope and then subtracting 1 gives 1.2. The specific value of the attenuation slope is 1.05, and the specific value of the enhancement slope is 1.76.

[0108] It should be noted that when the cumulative energy exceeds the threshold (for example, the critical value set in the processing of metal mask plates), using an attenuation factor (such as 0.8 times energy compensation) can effectively suppress heat accumulation. For example, on a continuously wound polyimide film, if the energy in a certain area reaches 450 mJ / cm 2 (50 mJ above the threshold) due to light spot overlap, attenuation can avoid micron-level deformation caused by thermal expansion of the substrate, which is crucial for maintaining the accuracy of a 10-μm line-width circuit. When the energy is below the threshold, an enhancement factor (such as 1.2 times energy compensation) can ensure sufficient reaction. Taking the manufacture of a flexible OLED display panel as an example, if the energy in a certain area is only 350 mJ / cm 2 (below the 400 mJ threshold) due to the jitter of the coil, the enhancement compensation can prevent nano-level residual defects from appearing after development, which is decisive for achieving sub-pixel-level evaporation alignment accuracy.

[0109] In step S16, the exposure parameters are optimized according to the overlap correction factor and the correlation influence matrix, and the control parameters of the exposure machine are updated.

[0110] In one implementation, weighted fusion is performed according to the overlap correction factor and the correlation influence matrix to obtain a corrected influence matrix;

[0111] The particle swarm parameters are initialized according to the corrected influence matrix;

[0112] When the number of iterations reaches the preset upper limit or the fitness function of the particle meets the constraint conditions, the iteration is completed, and the exposure machine control parameters are output.

[0113] It should be noted that in the optimization of exposure parameters, the system first dynamically weights and fuses the overlap correction factor (such as the 0.85 attenuation coefficient of metal parts) with the correlation influence matrix (the mutual influence weights of parameters including scanning speed, laser power, etc.): 60% weight is assigned to the correction factor for heat-sensitive materials (such as polyimide film), while 70% weight of the matrix data is allocated to the parameters sensitive to motion accuracy (such as galvanometer acceleration) to generate a correction influence matrix with process adaptability. Based on the guidance range of this matrix (such as the scanning speed is limited to 0.01 - 0.5 m / s, and the laser power window is set to 10 - 100 mW), a particle swarm composed of 200 groups of parameter combinations is initialized, and each particle contains parameter values randomly distributed within the limited range (such as particle A has a set speed of 0.23 m / s, power of 180 mW, and overlap compensation rate of 92%). During the iteration process, the exposure effect score corresponding to each particle is calculated in each round - by real-time simulation to calculate the alignment accuracy (such as a deviation of ±1.2 μm) and the heat influence score (such as a substrate temperature rise of 28 °C) under this parameter, and then a comprehensive score is synthesized according to a 6:4 weight. When it is detected that a certain particle reaches the preset standard (such as the score of the best particle in 5 consecutive generations exceeds 95 points) or 50 generations of iteration are completed, the historical optimal parameter combination (such as the combination of speed 3.1 m / s + power 220 mW + compensation 88%) is selected as the final control parameter. During the optimization process, a simulated annealing mechanism is particularly introduced. When it is detected that the best solution has not been updated for 10 consecutive generations, the parameter search range is automatically expanded by 10% to avoid local optimal traps. For example, in the OLED panel manufacturing scenario, this strategy successfully reduces the exposure positioning error from ±2.5 μm to ±1.1 μm, while reducing the heat accumulation per unit area by 40%.

[0114] In one embodiment, the weighted fusion of the overlap correction factor and the correlation influence matrix needs to dynamically adjust the weight ratio according to material characteristics and process requirements. For example, in the flexible circuit board manufacturing scenario, if the system detects that the cumulative energy in a certain polyimide substrate area is 380 mJ / cm 2If it is lower than the 400 mJ threshold, a 1.2-fold enhancement factor is enabled. At this time, the enhancement factor and the parameter weights in the associated influence matrix are hierarchically fused: for the key parameters of the photosensitive reaction (such as laser power), an adjustment weight of 70% is assigned to the enhancement factor, and the 40% weight of this parameter in the original matrix is correspondingly increased to 68% (that is, 40% × 70% superimposed with a 1.2-fold enhancement effect); for the parameters sensitive to motion stability (such as the web traction speed), 80% of the original 60% weight in the matrix is maintained, and only a 20% enhancement effect is applied. After normalization, the laser power weight in the generated new matrix is increased to 52%, the scanning speed weight is adjusted to 38%, and at the same time, a 10% dynamic compensation weight is introduced to offset the influence of web jitter. This intelligent weight distribution mechanism enables the full reaction of the photoresist to be ensured through power enhancement while maintaining a stable running speed of 0.8 m / s during the manufacturing of 5-μm line-width circuits, and finally achieving an exposure alignment accuracy of ±1.3 μm.

[0115] In summary, the present invention discloses a method and system for optimizing the accuracy of a roll-to-roll high-precision projection exposure machine, aiming to improve the accuracy during the exposure process through a series of precise data acquisition and analysis steps. This method first involves obtaining microscopic depth data, material vibration data, and exposure overlap regions, which ensures a comprehensive understanding of the substrate surface characteristics and its processing environment. Specifically, a white-light interference scanner is used for three-dimensional topography measurement to obtain microscopic depth data, a three-axis MEMS accelerometer is combined with a laser Doppler vibrometer to capture material vibration data, and a high-speed linear array CCD is responsible for determining the position of the exposure overlap region. Based on these data, the next step is to perform energy distribution analysis according to the microscopic depth data to obtain the exposure energy distribution. In this process, the microscopic depth data is divided into grid cells with a preset spacing, and the depth change amplitude and frequency are calculated through numerical differentiation, and finally a depth gradient value is formed. Then, using a pre-trained energy compensation model, combined with the depth gradient value and historical exposure parameters, the exposure energy result is output and mapped into an exposure energy distribution map.

[0116] In dealing with material vibration, this method uses frequency-domain analysis technology to extract vibration frequency-domain features from the original vibration data. This process includes steps such as wavelet denoising, wavelet transform, and dominant frequency identification. Finally, the vibration frequency-domain features are generated by normalizing and fusing the vibration dominant frequency and the vibration high-frequency interval. Next, the construction of the associated influence matrix depends on the correlation calculation of the exposure energy distribution and the vibration frequency-domain features. This step not only considers the influence of vibration on the exposure effect but also reveals the interaction relationship between the two. By performing block processing on the vibration frequency-domain features to form a frequency vector and extracting the exposure energy result of each grid cell based on the exposure energy distribution, the vibration-exposure correlation can be calculated, and then the associated influence matrix can be constructed.

[0117] In addition, in order to cope with the energy accumulation effect caused by multiple exposures, it is necessary to perform energy accumulation analysis on the exposure overlap area to determine the overlap correction factor. This involves calculating the cumulative exposure energy by integration and selecting an appropriate attenuation or enhancement factor as the overlap correction factor based on its relationship with the preset energy threshold. The last step is to optimize and update the exposure parameters based on all the above information, namely the overlap correction factor and the correlation influence matrix. By weighted fusion of these two key factors, a corrected influence matrix is generated, and the parameters are initialized using the particle swarm algorithm until specific conditions are met, thereby completing the update process of the exposure machine control parameters and achieving an improvement in working accuracy.

[0118] Referring to Figure 2 , the second embodiment of the present invention provides an accuracy optimization system for a roll-to-roll high-precision projection exposure machine, including:

[0119] A data acquisition module for acquiring microscopic depth data, material vibration data, and exposure overlap areas;

[0120] An energy distribution module for performing energy distribution analysis based on the microscopic depth data to obtain the exposure energy distribution;

[0121] A vibration analysis module for performing frequency domain analysis based on the material vibration data to obtain vibration frequency domain characteristics;

[0122] A correlation calculation module for performing correlation calculation based on the exposure energy distribution and the vibration frequency domain characteristics to obtain a correlation influence matrix;

[0123] An overlap correction module for performing energy accumulation analysis on the exposure overlap area to obtain an overlap correction factor;

[0124] A parameter update module for optimizing the exposure parameters based on the overlap correction factor and the correlation influence matrix and updating the exposure machine control parameters.

[0125] It should be noted that the accuracy optimization system for a roll-to-roll high-precision projection exposure machine provided in the embodiment of the present invention is used to execute all the process steps of the accuracy optimization method for a roll-to-roll high-precision projection exposure machine in the above embodiment, and their working principles and beneficial effects correspond one by one, so they will not be elaborated here.

[0126] The embodiment of the present invention also provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a data acquisition program. When the processor executes the computer program, it implements the steps in the embodiments of the above-mentioned accuracy optimization method for a roll-to-roll high-precision projection exposure machine, such as Figure 1Step S11 shown above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the data acquisition module.

[0127] Exemplarily, the computer program may be divided into one or more modules / units. The one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the electronic device.

[0128] The electronic device may be a computing device such as a desktop computer, a notebook, a palm computer, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device. It may include more or fewer components than the above, or combine some components, or different components. For example, the electronic device may further include input / output devices, network access devices, a bus, etc.

[0129] The so-called processor may be a central processing unit (CPU), or may also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the electronic device and connects various parts of the entire electronic device using various interfaces and lines.

[0130] The memory can be used to store the computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory, and by invoking the data stored in the memory, the processor realizes various functions of the electronic device. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0131] Among them, if the modules / units integrated in the electronic device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be realized. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, an executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disc, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0132] It should be noted that the device embodiments described above are merely illustrative. 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 to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0133] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. In particular, for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for optimizing the precision of a roll-to-roll high-precision projection exposure machine, characterized in that: include: Acquire microscopic depth data, material vibration data, and exposure overlap areas; Performing energy distribution analysis according to the microscopic depth data to obtain exposure energy distribution; Perform frequency domain analysis on the material vibration data to obtain vibration frequency domain characteristics; Perform correlation calculation based on the exposure energy distribution and the vibration frequency domain characteristics to obtain a correlation influence matrix; Performing energy accumulation analysis on the exposure overlap area to obtain an overlap correction factor; Exposure parameters are optimized according to the overlap correction factor and the correlation influence matrix, and exposure machine control parameters are updated.

2. The precision optimization method of a roll-to-roll high-precision projection exposure machine according to claim 1, characterized in that: The performing energy distribution analysis according to the microscopic depth data to obtain exposure energy distribution includes: Dividing the microscopic depth data into grid units with a preset spacing; Performing numerical differential calculation based on the grid unit and the microscopic depth data to obtain a depth variation amplitude; Performing second-order derivative calculation based on the grid unit and the microscopic depth data to obtain a depth variation frequency; Perform weighted fusion according to the depth change amplitude and the depth change frequency to obtain a depth gradient value; Inputting the depth gradient value and pre-stored exposure parameters into a preset energy compensation model, and outputting an exposure energy result; Distribution mapping is performed according to the exposure energy result to obtain exposure energy distribution.

3. The precision optimization method of a roll-to-roll high-precision projection exposure machine according to claim 2, characterized in that: The training process of the energy compensation model includes: Taking historical micro-depth data and historical exposure parameters as input and historical exposure results as output, the training set and validation set are divided through normalization processing; Initialize the multi-layer perceptron model parameters; Optimize parameters through iterative training; After the validation set loss converges or reaches the maximum number of iterations, if the test set error is lower than the preset threshold, the model parameters are saved as the final energy compensation model.

4. The precision optimization method of a roll-to-roll high-precision projection exposure machine according to claim 1, characterized in that: The performing frequency domain analysis according to the material vibration data to obtain vibration frequency domain characteristics includes: Performing wavelet denoising on the material vibration data to obtain denoised vibration data; Performing wavelet transformation on the noise reduction vibration data to obtain vibration frequency domain data; Performing main frequency identification according to the vibration frequency domain data to obtain the main frequency of vibration; Perform high-frequency interval identification according to the vibration frequency domain data to obtain a vibration high-frequency interval; The vibration main frequency and the vibration high-frequency interval are normalized and fused to obtain the vibration frequency domain feature.

5. The method for optimizing the precision of a roll-to-roll high-precision projection exposure machine according to claim 1, characterized in that: The correlation calculation is performed according to the exposure energy distribution and the vibration frequency domain characteristics to obtain a correlation influence matrix, including: The vibration main frequency and the vibration high frequency interval extracted from the vibration frequency domain feature are processed in blocks to form a frequency vector; Extracting an exposure energy result of each grid unit based on the exposure energy distribution; Calculate the correlation coefficient according to the frequency vector and the exposure energy result to obtain the vibration exposure correlation; Normalizing the vibration exposure correlation to obtain a normalized correlation; A correlation influence matrix is ​​constructed according to the normalized correlation and the corresponding relationship between the grid units and the frequency vectors.

6. The precision optimization method of a roll-to-roll high-precision projection exposure machine according to claim 1, characterized in that: The energy accumulation analysis of the exposure overlap area to obtain the overlap correction factor includes: Performing integral calculation according to the exposure overlap area and the exposure energy distribution to obtain cumulative exposure energy; Calculate the correction factor according to the accumulated exposure energy to obtain an attenuation factor and an enhancement factor; When the cumulative exposure energy is less than a preset energy threshold, taking the enhancement factor as an overlap correction factor; When the accumulated exposure energy is greater than or equal to the energy threshold, the attenuation factor is taken as the overlap correction factor.

7. The precision optimization method of a roll-to-roll high-precision projection exposure machine according to claim 1, characterized in that: The step of optimizing exposure parameters according to the overlap correction factor and the correlation influence matrix and updating exposure machine control parameters includes: Perform weighted fusion according to the overlap correction factor and the correlation influence matrix to obtain a correction influence matrix; Initializing particle swarm parameters according to the modified influence matrix; When the number of iterations reaches the preset upper limit or the fitness function of the particle meets the constraint conditions, the iteration is completed and the exposure machine control parameters are output.

8. A precision optimization system for a roll-to-roll high-precision projection exposure machine, characterized in that: include: A data acquisition module for acquiring microscopic depth data, material vibration data, and exposure overlap areas; An energy distribution module, used to perform energy distribution analysis according to the microscopic depth data to obtain exposure energy distribution; A vibration analysis module, used to perform frequency domain analysis based on the material vibration data to obtain vibration frequency domain characteristics; A correlation calculation module, used for performing correlation calculation according to the exposure energy distribution and the vibration frequency domain characteristics to obtain a correlation influence matrix; An overlap correction module, used for performing energy accumulation analysis on the exposure overlap area to obtain an overlap correction factor; A parameter updating module is used to optimize the exposure parameters according to the overlap correction factor and the correlation influence matrix, and update the exposure machine control parameters.

9. An electronic device, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements the accuracy optimization method of the roll-to-roll high-precision projection exposure machine as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the accuracy optimization method of the roll-to-roll high-precision projection exposure machine as described in any one of claims 1 to 7.

Images