A compound optical waveguide master defect compensation photolithography system and method

Abstract

The present application relates to the technical field of optical waveguide master defect compensation lithography, and in particular to a composite optical waveguide master defect compensation lithography system and method, a nonlinear decoupling model of grating period-wavelength-Talbot distance is constructed by using Newton-Raphson iteration method, and a multi-physical field compensation matrix is generated by combining FDTD simulation correction dispersion effect. The positioning unit realizes sub-nanometer substrate positioning through homogeneous coordinate transformation and thermal drift PID control, the motion control module optimizes trajectory planning by using Fourier series filtering, and the optical field uniformity is maintained by real-time calibration of liquid crystal phase modulation and polarization state. The defect suppression unit fuses Fresnel diffraction integral and deep learning feature matching, and realizes the spatiotemporal collaborative suppression of high-frequency defects through a double averaging algorithm. The quality verification module establishes a closed-loop feedback network, dynamically optimizes parameter acquisition models by using machine learning, and effectively solves the technical contradiction that the defect suppression blind area and system complexity cannot be considered in the existing process.

Description

Technical Field

[0001] This invention relates to the field of optical waveguide master plate defect compensation lithography technology, and in particular to a composite optical waveguide master plate defect compensation lithography system and method. Background Technology

[0002] As a high-precision template for mass-producing optical waveguide structures, the optical waveguide master is made of etch-resistant material and carries the designed optical waveguide geometric pattern and refractive index distribution information, serving as the fundamental benchmark for large-scale chip manufacturing. Photolithography is a micro-nano processing method that uses optical imaging principles to transfer mask patterns to photosensitive materials, achieving patterning by controlling the light intensity distribution. The composite approach is manifested in the integration of photolithography processes with technologies such as step-by-step exposure, multi-wavelength light sources, or composite photosensitive material systems to meet the processing requirements of optical waveguide master for complex curvature and high aspect ratio structures. Defect compensation involves identifying defects such as pattern distortion, material residue, or etching deviation through optical detection during master photolithography and post-processing, and actively adjusting them using methods such as local etching correction, selective material deposition, or pattern proximity effect correction to make the defective areas approximate the design shape.

[0003] Existing composite optical waveguide master plate defect compensation photolithography systems have the following technical pain points: When the master plate has both periodic grating defects and irregular micro-damage, relying solely on the Talbot distance positioning mechanism can only achieve optical averaging repair for periodic defects, but lacks effective suppression capabilities for non-periodic defects such as randomly distributed nanoscale material distortions or sudden mechanical scratches; if an additional mechanical scanning compensation mechanism is added, dynamic focal length drift is easily caused by the coupling problem of multi-axis motion control precision, resulting in distortion of photoresist exposure energy distribution, and the system integration complexity increases exponentially, ultimately causing a negative correlation fluctuation between defect repair rate and mass production yield. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a composite optical waveguide master plate defect compensation lithography system and method. This invention solves the technical problem that in composite optical waveguide master plate defect compensation lithography systems, relying solely on Tyber distance positioning or mechanical scanning compensation results in a defect suppression blind zone, leading to insufficient high-frequency non-periodic defect repair rate. At the same time, the system integration complexity is multiplied, causing fluctuations in mass production yield.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows:

[0006] In a first aspect, the present invention provides a composite optical waveguide master plate defect compensation photolithography system, comprising: The data acquisition unit is used to acquire master design parameters and light source characteristic data; The parameter calculation unit is used to receive the design parameters and light source characteristic data output by the data acquisition unit, and generate positioning reference parameters by solving a set of nonlinear equations. The positioning unit is used to receive the positioning reference parameters output by the parameter calculation unit, convert the Tiber distance parameters into XYZ three-axis physical coordinate offsets through a preset coordinate transformation algorithm, and output the compensated positioning coordinates. The motion control unit is used to receive the compensated positioning coordinates output by the positioning unit, calculate and output motion trajectory data based on the physical coordinate offset; The light field control unit is used to receive motion trajectory data output by the motion control unit, optimize the light field distribution through light intensity uniformity calibration and dynamic polarization adjustment, and output the optimized light field parameters. The defect suppression unit is used to receive the optimized light field parameters output by the light field control unit, generate a dual average light intensity distribution based on the preset Fresnel diffraction integral calculation, suppress periodic and non-periodic defects, and output defect repair data including regional exposure control parameters. The quality verification unit is used to receive the defect repair data output by the defect suppression unit and generate a repair rate statistical report through optical detection and electrical performance testing. The data acquisition unit, parameter calculation unit, positioning unit, motion control unit, light field modulation unit, defect suppression unit, and quality verification unit are sequentially connected through standardized interfaces to form a closed-loop control system from data acquisition to quality verification.

[0007] Furthermore, in the composite optical waveguide master plate defect compensation photolithography system of the present invention, the data acquisition unit includes: The design data extraction module is used to read the raster period design value from the digital file of the master design drawing and output the raster period design value data. The spectral calibration module is used to acquire the spectral signal of the laser output light through a spectral analyzer, analyze the center wavelength value, and output wavelength measurement data. The data fusion module receives the grating period design value data output by the design data extraction module and the wavelength measurement data output by the spectral calibration module, and merges the grating period design value data and the wavelength measurement data into an original measurement dataset. It then performs noise filtering on the original measurement dataset using the least squares method and removes outlier data points with a standard deviation exceeding 3σ of the mean of the original measurement dataset, generating filtered data. The filtered data is then processed using Fourier transform to extract the grating period dominant frequency component. This component is compared with the received grating period design value data, and a deviation correction function is constructed based on the comparison result. Finally, based on the deviation correction function and the wavelength measurement data, a standardized period parameter set and wavelength calibration coefficients are generated and output.

[0008] Furthermore, in the composite optical waveguide master plate defect compensation photolithography system of the present invention, the parameter calculation unit includes: The nonlinear solution module receives the standardized periodic parameter set and wavelength calibration coefficients output by the data acquisition unit, and uses the Newton-Raphson iterative method to solve the nonlinear equations of grating period, wavelength and Talbot distance based on the standardized periodic parameter set and wavelength calibration coefficients, generating and outputting a numerical sequence of Talbot distance. The multiphysics compensation module receives the Talbot distance numerical sequence output by the nonlinear solution module, and calculates the distance offset caused by the grating periodic dispersion effect by combining the preset FDTD simulation model and the material dispersion coefficient; the distance offset is then added to the received Talbot distance numerical sequence to generate and output a multidimensional Talbot distance matrix. The redundancy generation module receives the multi-dimensional Tyber distance matrix output by the multi-physics compensation module, and superimposes a process tolerance random parameter conforming to a preset normal distribution on each element of the multi-dimensional Tyber distance matrix to generate a matrix with embedded tolerance parameters; the matrix with embedded tolerance parameters is ASCII encoded, and the encoded data is transmitted to the positioning unit.

[0009] Furthermore, in the composite optical waveguide master plate defect compensation photolithography system of the present invention, the positioning unit includes: The coordinate transformation module receives the ASCII encoded data output by the parameter calculation unit, decodes the data, and restores it to a multi-dimensional Talbot distance matrix. It uses a preset homogeneous coordinate transformation algorithm to extract Talbot distance parameters from the multi-dimensional Talbot distance matrix and converts the Talbot distance parameters into XYZ three-axis physical coordinate offsets including translation vectors and rotation matrices, generating and outputting absolute positioning reference data. The thermal drift compensation module receives the absolute positioning reference data output by the coordinate transformation module, and acquires the actual position data of the base on the XYZ axes in real time through an integrated interferometer; it performs differential calculation between the actual position data and the absolute positioning reference data to generate position deviation data; based on the position deviation data, it uses a PID controller with a preset proportional coefficient Kp=2.5, integral coefficient Ki=0.01, and derivative coefficient Kd=0.1 to dynamically calculate the adjustment amount of the piezoelectric ceramic driving voltage with an exponential decay function, and outputs the compensated positioning coordinates corrected based on the adjustment amount. The motion command compilation module receives the compensated positioning coordinates output by the thermal drift compensation module, converts the compensated positioning coordinates into a pulse sequence signal that matches the displacement resolution of the voice coil motor, and uses the EtherCAT bus and its distributed clock synchronization mechanism to time-align the pulse sequence signals corresponding to the XYZ axes, generating and outputting synchronous motion commands with a phase difference within ±10ns.

[0010] Furthermore, in the composite optical waveguide master plate defect compensation photolithography system of the present invention, the motion control unit includes: The trajectory prediction module receives synchronous motion commands output by the motion command compilation module of the positioning unit, and predicts the instantaneous position of the base in the future motion cycle based on the sinusoidal waveform driving signal included in the synchronous motion command, generating instantaneous position distribution data; performs Fourier series expansion on the instantaneous position distribution data, filters out harmonic components with frequencies higher than the preset cutoff frequency in the expansion result, and generates and outputs optimized motion trajectory parameters including the filtered fundamental phase information; The light field synchronization module receives the optimized motion trajectory parameters output by the trajectory prediction module, extracts the fundamental phase information from the optimized motion trajectory parameters, generates a corresponding phase modulation voltage signal based on the fundamental phase information, and applies the phase modulation voltage signal to the liquid crystal spatial light modulator to change the refractive index of each pixel on the liquid crystal spatial light modulator, thereby adjusting the wavefront phase of the transmitted light field, generating and outputting synchronized light field parameters. The polarization matching module receives the synchronized optical field parameters output by the optical field synchronization module, calculates the deviation between the current polarization state and the target circular polarization state by measuring the Stokes vector of the synchronized optical field parameters, and dynamically adjusts the driving voltage applied to the liquid crystal phase retardation film based on the deviation so that the polarization direction of the optical field transmitted through the liquid crystal phase retardation film is adjusted to the target circular polarization state, and generates and outputs polarization-consistent optical field parameters to the optical field control unit.

[0011] Furthermore, in the composite optical waveguide master plate defect compensation photolithography system of the present invention, the optical field modulation unit includes: The light intensity uniformity calibration module receives polarization-uniform light field parameters output by the polarization matching module of the motion control unit. It performs a two-dimensional scan of the actual light field characterized by the polarization-uniform light field parameters using a photodetector array to acquire light field intensity distribution data. It compares the intensity value of each pixel in the light field intensity distribution data with a preset light intensity uniformity threshold to identify low-intensity pixel regions with intensities below the threshold. It generates a grayscale compensation mapping table corresponding to the positions of the low-intensity pixel regions and drives a liquid crystal spatial light modulator based on the grayscale compensation mapping table to perform pixel-level light intensity tuning on the low-intensity regions, generating and outputting the light field parameters after light intensity uniformity calibration. The dynamic polarization adjustment module receives the light field parameters after light intensity uniformity calibration output by the light intensity uniformity calibration module, guides the light field through a λ / 4 waveplate, and rotates the polarization direction of the light field by 45 degrees; after the λ / 4 waveplate, the Stokes vector of the output light field is measured in real time by a Stokes vector analyzer; based on the measured Stokes vector, the purity deviation between the current polarization state and the target circular polarization state is calculated; based on the purity deviation, the driving voltage applied to the λ / 4 waveplate is dynamically adjusted through a closed-loop feedback circuit to generate and output polarization-uniform light field parameters; The edge gradient control module receives the polarization-consistent optical field parameters output by the dynamic polarization adjustment module, guides the optical field through the microlens array, and converts the input optical field with a Gaussian distribution into an output optical field with a flat-top distribution. Based on the Fresnel diffraction integral formula, it calculates the intensity gradient distribution of the output optical field in the edge region. It iteratively adjusts the spacing parameters of the lens elements in the microlens array through a gradient descent algorithm so that the calculated edge intensity gradient is lower than a preset gradient threshold, and generates and outputs the edge gradient-controlled optical field parameters to the defect suppression unit.

[0012] Furthermore, in the composite optical waveguide master plate defect compensation photolithography system of the present invention, the defect suppression unit includes: The diffraction integral calculation module receives the light field parameters after edge gradient control output by the edge gradient control module of the light field modulation unit, and performs Fresnel diffraction integral calculation on the light field wavefront represented by the light field parameters after edge gradient control based on the fast Fourier transform algorithm to generate an initial light intensity distribution map; performs spatial domain sliding window averaging on the spatial pixel array of the initial light intensity distribution map, and performs temporal domain averaging on multiple consecutive frames of the initial light intensity distribution map in the time series to generate and output a predicted light intensity distribution map. The defect identification module receives the predicted light intensity distribution map output by the diffraction integral calculation module, inputs the predicted light intensity distribution map into a pre-trained convolutional neural network model, and the convolutional neural network model extracts defect features from the image and outputs a defect region mask map; based on the defect region mask map, it calculates the pixel area ratio of the defect region as the defect repair rate, and based on a preset qualified pixel intensity range, it calculates the qualified pixel ratio as the yield rate; simultaneously, it extracts the pixel intensity values ​​of the region identified by the defect region mask map from the predicted light intensity distribution map, and generates and outputs defect intensity distribution data; The exposure threshold adjustment module receives defect intensity distribution data output by the defect identification module, calculates the required photoresist exposure energy compensation for each defect region based on the average pixel intensity of each defect region in the defect intensity distribution data, and dynamically calculates and outputs the exposure time adjustment value corresponding to each defect region using a PID controller with a preset proportional coefficient Kp=0.6, integral time Ti=0.5 seconds, and derivative time Td=0.1 seconds. The module integrates the exposure time adjustment values ​​of all regions to generate regional exposure control parameters, and sends these regional exposure control parameters to the light field control unit.

[0013] Furthermore, in the composite optical waveguide master plate defect compensation photolithography system of the present invention, the quality verification unit includes: The morphology detection module receives the regional exposure control parameters output by the exposure threshold adjustment module of the defect suppression unit. Based on the spatial coordinates of each exposure region defined by the regional exposure control parameters, it drives the stage to move the corresponding exposed region to the field of view of the optical microscope. The optical microscope is used to photograph the surface morphology of the exposed region, generating morphology image data. In the same exposed region, an ellipsometer is used to measure the ellipsoidal parameters of the region at a preset incident angle, and the refractive index distribution of the region is calculated based on the Shermeier equation to generate refractive index distribution data. The morphology image data and the refractive index distribution data are integrated to generate and output morphology and refractive index detection data. The refractive index analysis module receives morphology and refractive index detection data output by the morphology detection module. From the morphology image data, it extracts the contour of the actual defect region based on an image segmentation algorithm and calculates the area and depth of the actual defect region. Based on the area and depth, it calculates the actual defect repair rate. From the refractive index distribution data, it calculates the period deviation and wavelength drift between the actual refractive index value and the designed refractive index value. The actual defect repair rate, the period deviation, and the wavelength drift are used as input features and input to a pre-trained support vector machine model to establish a mathematical correlation model between the actual defect repair rate, the period deviation, and the wavelength drift. Based on the mathematical correlation model, it generates process parameter correction suggestions including material compensation and etching rate adjustment values, and outputs the process parameter correction suggestions. The data inversion module receives process parameter correction suggestions output by the refractive index analysis module, parses correction instructions for the standardized periodic parameter set and correction instructions for the laser center wavelength from the process parameter correction suggestions; encapsulates the correction instructions for the standardized periodic parameter set into a JSON format data packet and sends it to the data fusion module of the data acquisition unit via the HTTP protocol; converts the correction instructions for the laser center wavelength into temperature control parameter adjustment values ​​and writes them into the laser's temperature control register via the SPI interface to adjust the laser's output wavelength; after completing the instruction sending and register writing, it generates and outputs a closed-loop control feedback signal.

[0014] Furthermore, the composite optical waveguide master plate defect compensation photolithography system of the present invention further includes: The data benchmark establishment module receives the closed-loop control feedback signal output by the data inversion module of the quality verification unit. Under a constant temperature of 23±0.5℃ and a relative humidity of less than 40%, it jointly operates the data acquisition unit, parameter calculation unit, positioning unit, motion control unit, light field modulation unit, and defect suppression unit to collect process data throughout the entire process. It then uses Zernike polynomial fitting to fit the optical system aberrations to generate a standardized parameter set including 256 error compensation coefficients. The standardized parameter set is then output as the initial calibration parameters to the data fusion module of the data acquisition unit. The spatial positioning optimization module receives the compensated positioning coordinates output by the thermal drift compensation module of the positioning unit, and measures the displacement data of the substrate in real time along the XYZ axes and the six degrees of freedom of rotation around the axes using a dual-frequency laser interferometer. Simultaneously, it collects the motion acceleration data of the substrate using an accelerometer. A Kalman filter algorithm is used to fuse the displacement data and the motion acceleration data to generate more accurate fused position data. Based on the difference between the fused position data and the compensated positioning coordinates, a nanometer-level position correction is calculated, and the piezoelectric ceramic actuator is driven to adjust the substrate position according to the position correction, generating and outputting positioning accuracy verification data. The spatiotemporal synchronization control module receives the optimized motion trajectory parameters output by the trajectory prediction module of the motion control unit, extracts the fundamental phase information from the optimized motion trajectory parameters, generates a hardware clock signal that is strictly synchronized with the fundamental phase information using a field-programmable gate array (FPGA), and re-aligns and outputs XYZ three-axis motion commands based on the hardware clock signal through the distributed clock synchronization mechanism of the EtherCAT bus, so that the phase difference between the three-axis commands is controlled within ±5 nanoseconds. At the same time, the hardware clock signal is sent as a trigger signal to the light intensity uniformity calibration module of the light field control unit to synchronize the light field acquisition timing, generate and output spatiotemporal synchronization optimization parameters. The effect feedback module receives process parameter correction suggestions output by the refractive index analysis module of the quality verification unit, and parses the material compensation amount and etching rate adjustment value from the process parameter correction suggestions. Using the material compensation amount and etching rate adjustment value as environmental rewards, the module uses a Q-learning reinforcement learning algorithm to dynamically and iteratively optimize the weight coefficients of the noise filtering function used in the data fusion module of the data acquisition unit, and the convergence threshold of the Newton-Raphson iteration method in the nonlinear solution module of the parameter calculation unit. The optimized weight coefficients and convergence threshold are encapsulated as correction coefficients Kp, Kd, ​​and Ki, and fed back to the nonlinear solution module of the parameter calculation unit.

[0015] Secondly, the present invention provides a composite optical waveguide master plate defect compensation photolithography method, applied to the composite optical waveguide master plate defect compensation photolithography system as described above, comprising: Step 1: Obtain master design parameters and light source characteristic data; Step 2: Receive the design parameters and light source characteristic data, and generate positioning reference parameters by solving a system of nonlinear equations; Step 3: Receive the positioning reference parameters, convert the Tiber distance parameters into XYZ three-axis physical coordinate offsets using a preset coordinate transformation algorithm, and generate compensated positioning coordinates; Step 4: Calculate and generate motion trajectory data based on the offset between the compensated positioning coordinates and the physical coordinates; Step 5: Receive the motion trajectory data, optimize the light field distribution through light intensity uniformity calibration and dynamic polarization adjustment, and generate optimized light field parameters; Step 6: Receive the optimized light field parameters, generate a dual average light intensity distribution based on Fresnel diffraction integral calculation, suppress periodic and non-periodic defects, and generate defect repair data including regional exposure control parameters. Step 7: Receive the defect repair data and generate a repair rate statistical report through optical detection and electrical performance testing; Steps 1 to 7 are executed sequentially through standardized interfaces, forming a closed-loop control process from data acquisition to quality verification.

[0016] Beneficial effects of this invention: This invention significantly improves the overall performance of a waveguide master plate defect compensation lithography system by constructing a multi-dimensional collaborative optimization closed-loop control system. The system achieves nonlinear decoupling calculations of grating period, wavelength, and Talbot distance through deep coupling of the data acquisition unit and parameter calculation unit, effectively eliminating the cumulative impact of single parameter fluctuations on positioning accuracy. The positioning unit employs a composite algorithm combining homogeneous coordinate transformation and thermal drift compensation, combined with the nanometer-level control capability driven by piezoelectric ceramics, achieving sub-nanometer-level positioning accuracy in the XYZ axes and solving the dynamic focal length drift problem caused by existing mechanical scanning compensation. The motion control unit, through a dual optimization mechanism of Fourier series expansion and liquid crystal phase compensation, maintains optical field polarization consistency while suppressing motion blur, thus improving light intensity uniformity. The defect suppression unit innovatively integrates Fresnel diffraction integral calculation and deep learning feature matching, achieving spatial-temporal domain collaborative suppression of high-frequency defects through a dual averaging algorithm, resulting in a higher repair rate than existing methods. The closed-loop feedback mechanism established by the quality verification unit dynamically optimizes parameter acquisition and the Taibo calculation model through machine learning models, reducing the system response time to within 200ms and controlling the fluctuation range of mass production yield to within ±1.2%. All functional units form an organic whole through standardized interfaces, enabling the system to exhibit excellent adaptability in dealing with periodic grating defects and irregular microscopic damage, effectively solving the technical challenge of balancing defect suppression blind spots and system complexity in existing technologies. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0018] Figure 1This is a flowchart of a composite optical waveguide master plate defect compensation photolithography method according to the present invention. Detailed Implementation

[0019] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0020] In a first aspect, the present invention provides a composite optical waveguide master plate defect compensation photolithography system, comprising: The data acquisition unit is used to acquire master design parameters and light source characteristic data; The parameter calculation unit is used to receive the design parameters and light source characteristic data output by the data acquisition unit, and generate positioning reference parameters by solving a set of nonlinear equations. The positioning unit is used to receive the positioning reference parameters output by the parameter calculation unit, convert the Tiber distance parameters into XYZ three-axis physical coordinate offsets through a preset coordinate transformation algorithm, and output the compensated positioning coordinates. The motion control unit is used to receive the compensated positioning coordinates output by the positioning unit, calculate and output motion trajectory data based on the physical coordinate offset; The light field control unit is used to receive motion trajectory data output by the motion control unit, optimize the light field distribution through light intensity uniformity calibration and dynamic polarization adjustment, and output the optimized light field parameters. The defect suppression unit is used to receive the optimized light field parameters output by the light field control unit, generate a dual average light intensity distribution based on the preset Fresnel diffraction integral calculation, suppress periodic and non-periodic defects, and output defect repair data including regional exposure control parameters. The quality verification unit is used to receive the defect repair data output by the defect suppression unit and generate a repair rate statistical report through optical detection and electrical performance testing. The data acquisition unit, parameter calculation unit, positioning unit, motion control unit, light field modulation unit, defect suppression unit, and quality verification unit are sequentially connected through standardized interfaces to form a closed-loop control system from data acquisition to quality verification.

[0021] The system begins with a data acquisition unit that performs two tasks in parallel. The design data extraction module reads and outputs design values ​​for key geometric parameters such as grating period, linewidth, and duty cycle from digital files of the master design drawings, such as vector graphics files in Gerber or OASIS format using optical character recognition (OCR) technology. Simultaneously, the spectral calibration module acquires the spectral signal of the laser output light using a spectral analyzer, analyzes the center wavelength and spectral bandwidth, and outputs wavelength measurement data characterizing the light source. The data fusion module receives both types of data, merges them into a raw measurement dataset, and performs noise filtering using the least squares method to remove outliers that significantly deviate from the mean, generating filtered data. A Fourier transform is performed on the signal containing spatial periodic information represented by the filtered data, extracting the spatial frequency component with the highest amplitude in the frequency domain; its reciprocal is the dominant frequency component of the grating period. The extracted dominant frequency component of the grating period... The extracted grating period main frequency component is compared with the grating period design value initially read by the design data extraction module. The numerical differences between the two are analyzed, and a deviation correction function is constructed based on this difference to calibrate subsequent period parameters. Finally, based on this deviation correction function and the original wavelength measurement data, the data fusion module generates and outputs a standardized set of period parameters and wavelength calibration coefficients, providing accurate, calibrated, and denoised input for subsequent calculations. The design parameters include: grating period, linewidth, and duty cycle. The light source characteristic data includes: center wavelength and spectral bandwidth.

[0022] The parameter calculation unit receives a standardized periodic parameter set and wavelength calibration coefficients from the data acquisition unit. One of its core tasks is to solve a set of nonlinear equations concerning the grating period, operating wavelength, and Talbot distance. The nonlinear solution module within the parameter calculation unit uses the standardized periodic parameter set and wavelength calibration coefficients as input parameters and employs the Newton-Raphson iterative method to solve the nonlinear equations, generating a numerical sequence of Talbot distances. This sequence defines a series of theoretical positions where the light field produces self-images at the corresponding grating period and wavelength. However, in actual manufacturing processes, the dispersion effect of materials can cause deviations in the theoretical distances. Therefore, the multiphysics compensation module in the parameter calculation unit receives the numerical sequence of Talbot distances output by the nonlinear solution module and, combined with a preset finite-difference time-domain simulation model and the dispersion coefficients of the photoresist material, calculates the distance offset caused by the dispersion effect. Performing a matrix addition operation between the calculated distance offset and the original numerical sequence of Talbot distances generates a more accurate multidimensional Talbot distance matrix that includes compensation for physical effects. To further enhance system robustness, the redundancy generation module superimposes a pre-defined normal distribution of random process tolerance parameters onto each element of this matrix to simulate process fluctuations, generating a matrix with embedded tolerance parameters. This matrix, after ASCII encoding, forms a data stream that is transmitted to the positioning unit.

[0023] The coordinate transformation module of the positioning unit receives ASCII encoded data from the parameter calculation unit, decodes it, and restores it to a multi-dimensional Talbot distance matrix. The coordinate transformation module uses a preset homogeneous coordinate transformation algorithm to extract the Talbot distance parameters from the matrix and converts them into XYZ three-axis physical coordinate offsets, including three-dimensional translation vectors and rotation matrices, generating absolute positioning reference data to guide the motion system. During actual motion, environmental thermal drift and other factors can cause the actual position of the base to deviate from the theoretical reference. The thermal drift compensation module uses an integrated laser interferometer to collect the actual position data of the base on the XYZ axes in real time, and performs differential calculations between this actual position data and the absolute positioning reference data to generate real-time position deviation data. A PID controller with preset proportional, integral, and derivative coefficients dynamically calculates the required adjustment amount of the piezoelectric ceramic driving voltage based on this position deviation data using an exponential decay function, and outputs the compensated positioning coordinates after this dynamic correction. The motion command compilation module receives the compensated positioning coordinates output by the thermal drift compensation module and converts them into a pulse sequence signal that precisely matches the displacement resolution of the voice coil motor. Through the EtherCAT industrial Ethernet bus and its distributed clock synchronization mechanism, the motion instruction compilation module performs high-precision timing alignment of the pulse sequence signals corresponding to the three motion axes XYZ, generates and outputs synchronous motion instructions with phase difference control within an extremely narrow time window, thereby driving the actuator to achieve coordinated motion with nanometer-level precision.

[0024] The motion control unit receives synchronization motion commands from the positioning unit, and its internal processing aims to optimize motion and synchronize the light field. The trajectory prediction module, based on the sinusoidal drive signal contained in the synchronization motion commands, predicts the instantaneous position distribution of the substrate within a future complete motion cycle, generating instantaneous position distribution data. By performing a Fourier series expansion on this distribution data and filtering out harmonic components with frequencies higher than a preset cutoff frequency, the module generates and outputs optimized motion trajectory parameters, including the filtered fundamental phase information. The light field synchronization module extracts key fundamental phase information from the optimized motion trajectory parameters. Based on the extracted phase information, the module generates a corresponding phase modulation voltage signal and applies this voltage signal to the liquid crystal spatial light modulator. The voltage signal changes the refractive index of each pixel on the liquid crystal spatial light modulator, thereby adjusting the wavefront phase of the light field transmitted through the modulator, achieving strict synchronization between the light field distribution and mechanical motion, and generating and outputting synchronized light field parameters. To further ensure light field quality, the polarization matching module receives the synchronized light field parameters and calculates the deviation between the current polarization state and the target circular polarization state by measuring its Stokes vector. Based on this deviation, the module dynamically adjusts the driving voltage applied to the subsequent liquid crystal phase retardation film, so that the polarization direction of the light field transmitted through the retardation film is adjusted to the desired target circular polarization state, and finally generates and outputs polarization-consistent light field parameters to the light field control unit.

[0025] The light field control unit is responsible for fine-tuning the uniformity, polarization state, and spatial distribution of the input light field. The light intensity uniformity calibration module receives polarization-consistency light field parameters from the motion control unit. For the actual beam represented by these parameters, a photodetector array performs a two-dimensional scan to acquire light field intensity distribution data. The module compares the intensity value of each pixel in the distribution data with a preset light intensity uniformity threshold, identifying low-intensity pixel regions with intensities below the threshold. Subsequently, the module generates a grayscale compensation mapping table corresponding one-to-one with the locations of these low-intensity pixel regions. Based on this mapping table, it drives the liquid crystal spatial light modulator to perform pixel-level intensity modulation on the identified low-intensity regions, thereby generating and outputting the light field parameters after light intensity uniformity calibration. The dynamic polarization adjustment module receives these light field parameters and guides the light field through a λ / 4 waveplate, rotating the polarization direction of the light field by 45 degrees to generate or optimize a circular polarization state. After the λ / 4 waveplate, a Stokes vector analyzer measures the Stokes vector of the output light field in real time, calculating the purity deviation between the current polarization state and the target circular polarization state. Based on purity deviation, the closed-loop feedback circuit dynamically adjusts the driving voltage applied to the λ / 4 waveplate to maintain a stable circular polarization state, generating and outputting polarization-consistent optical field parameters. The edge gradient control module receives these parameters and guides the optical field through a microlens array, transforming the input optical field, which originally had a Gaussian distribution, into an output optical field with a flat-top distribution. After calculating the intensity gradient distribution of the output optical field in the edge region based on the Fresnel diffraction integral formula, the module iteratively adjusts the spacing parameters of the lens elements in the microlens array using a gradient descent algorithm, ensuring that the calculated edge intensity gradient is below a preset gradient threshold. Finally, it generates and outputs the edge gradient-controlled optical field parameters to the defect suppression unit.

[0026] The defect suppression unit analyzes and processes the optimized light field to predict and suppress defects. The diffraction integral calculation module receives the light field parameters after edge gradient control from the light field modulation unit. Based on the Fast Fourier Transform algorithm, the module performs Fresnel diffraction integral calculation on the wavefront of the light field represented by the light field parameters to simulate the propagation of light in the photoresist and generate an initial light intensity distribution map. To suppress noise, the module performs spatial domain sliding window averaging on the spatial pixel array of the initial light intensity distribution map and temporal domain averaging on multiple consecutive frames of the initial light intensity distribution map in the time series. This dual averaging process generates and outputs a clearer and more stable predicted light intensity distribution map. The defect recognition module receives the predicted light intensity distribution map and inputs it into a pre-trained convolutional neural network model. This convolutional neural network model automatically extracts defect features from the image and outputs a defect region mask map indicating the defect location. Based on the defect region mask map, the module calculates the pixel area ratio of the defect region as an indicator of the defect repair rate and statistically analyzes the percentage of qualified pixels based on a preset range of qualified pixel intensity as an indicator of the yield rate. Simultaneously, this module extracts the pixel intensity values ​​of the areas identified by the defect region mask image from the predicted light intensity distribution map, generating and outputting defect intensity distribution data. The exposure threshold adjustment module receives the defect intensity distribution data and calculates the required photoresist exposure energy compensation for each defect region based on the average pixel intensity of each defect region in the data. Based on the exposure energy compensation, a PID controller with preset specific control parameters dynamically calculates and outputs the exposure time adjustment value corresponding to each defect region. By integrating the exposure time adjustment values ​​of all regions, the module generates regional exposure control parameters and sends these control parameters to the light field control unit to perform differentiated energy injection in subsequent exposures.

[0027] The quality verification unit forms a closed-loop feedback loop for the system, detecting and evaluating the processing results. The morphology detection module receives regional exposure control parameters from the defect suppression unit. Based on the spatial coordinates of each exposure area defined by the control parameters, the module drives the precision stage to move the corresponding exposed area to the field of view of the optical microscope. The surface morphology of the exposed area is then photographed through the microscope, generating morphology image data. In the same area, the module uses an ellipsometer to measure the ellipsoidal parameters of the area at a preset incident angle and calculates the refractive index distribution of the area based on the Shermeier equation, generating refractive index distribution data. Integrating the morphology image data and refractive index distribution data, the module outputs morphology and refractive index detection data. The refractive index analysis module receives this detection data, extracts the contour of the actual defect area from the morphology image data using an image segmentation algorithm, calculates its area and depth, and then calculates the actual defect repair rate. Simultaneously, it calculates the period deviation and wavelength drift between the actual refractive index value and the designed refractive index value from the refractive index distribution data. Using the actual defect repair rate, cycle deviation, and wavelength drift as input features, a pre-trained support vector machine model can be established to model the mathematical relationship between the actual defect repair rate and the latter two. Based on this model, the module generates and outputs process parameter correction suggestions, including specific measures such as material compensation and etching rate adjustment values. The data inversion module receives the process parameter correction suggestions and parses out the correction instructions for the standardized cycle parameter set and the correction instructions for the laser center wavelength. The module encapsulates the correction instructions for the standardized cycle parameter set into a JSON format data packet and sends it back to the data fusion module of the data acquisition unit via HTTP protocol to update the initial parameters. Simultaneously, the correction instructions for the laser center wavelength are converted into temperature control parameter adjustment values ​​and written to the laser's temperature control register via the SPI interface, thereby adjusting the laser's output wavelength. After completing the instruction sending and register writing, the module generates and outputs a closed-loop control feedback signal, marking the end of one complete process cycle and the beginning of the next optimization cycle.

[0028] This invention sequentially connects the data acquisition unit, parameter calculation unit, positioning unit, motion control unit, optical field modulation unit, defect suppression unit, and quality verification unit through the aforementioned standardized interface. The data and command flow begins with initial parameter acquisition, undergoes physical quantity calculation, precise motion control, dynamic optical field modulation, and defect prediction and suppression, ultimately reaching process effect verification. Based on the verification results, correction commands are generated and fed back to the system front end. This process constitutes a closed-loop control system from data acquisition to quality verification, enabling the system to learn and optimize parameters based on the actual results of each process execution, thereby dynamically improving its ability to suppress periodic and non-periodic defects in the optical waveguide motherboard and enhancing process stability.

[0029] The data acquisition unit is responsible for providing calibrated and verified initial input for this invention. The design data extraction module begins by receiving digital files of the master design drawings, typically in standard industry formats such as Gerber or OASIS that describe the geometry of the photomask. The module analyzes the layers and graphic elements in the vector graphics file to identify the lines and spaces representing the raster structure, thereby reading the design values ​​for the raster period. It may also extract related geometric parameters such as linewidth and duty cycle to generate structured raster period design value data. This process provides the system with a theoretical design benchmark for the master template to be processed.

[0030] Another concurrent data stream is processed by the spectral calibration module. This module couples the output light emitted by the laser to a spectral analyzer via optical fiber. A grating or interferometer in the spectral analyzer disperses the incident light, and a CCD or photodiode array detector captures the intensity distribution at different wavelengths, forming a spectral signal. The module's analytical algorithms, such as finding the peaks or centroids of the spectral envelope, calculate the center wavelength of the laser from this spectral signal, outputting wavelength measurement data. This data characterizes the key characteristics of the light source used in actual processing, and its accuracy directly affects all subsequent wavelength-based optical calculations.

[0031] The data fusion module is the core processing unit of the data acquisition unit, receiving the output from the two parallel channels mentioned above. First, the data fusion module aligns and merges the grating period design value data and wavelength measurement data in time sequence, forming a raw measurement dataset that includes both design theory and actual measurements. To address unavoidable electrical noise or random fluctuations during measurement, the data fusion module uses the least squares method to filter the dataset, aiming to find an optimal curve to fit the data trend. Based on statistical principles, it removes outlier data points that deviate from the dataset mean by more than three standard deviations, thereby generating more reliable filtered data.

[0032] Performing a Fourier transform on the filtered data is a crucial step in extracting its frequency domain characteristics. The transform result converts the data from the time or spatial domain to the frequency domain, where the frequency component with the highest amplitude corresponds to the principal spatial frequency of the grating structure, and its reciprocal is the principal frequency component of the grating period. This period value "extracted" from the measured data may deviate from the theoretical design value initially "read" from the design file. Therefore, the module compares the principal frequency component of the grating period with the received grating period design value data, analyzing the differences between the two in terms of value and phase. Based on this comparison result, a mathematical model, namely the deviation correction function, can be constructed to describe the mapping or correction relationship between the design value and the actual achievable value.

[0033] Ultimately, the data fusion module integrates the constructed deviation correction function and the original wavelength measurement data. The deviation correction function is used to calibrate periodic parameters, while the wavelength measurement data directly provides the calibration benchmark for the light source. Through a series of calculations, the module outputs a standardized set of periodic parameters and wavelength calibration coefficients. The standardized set of periodic parameters is a set of grating geometry parameters that have been standardized to eliminate significant errors, while the wavelength calibration coefficients are scaling factors used to correct wavelength values ​​in subsequent calculations. These two outputs provide high-quality, calibrated input to the downstream parameter calculation unit, serving as the data starting point for the closed-loop control system.

[0034] The parameter calculation unit is responsible for converting the calibrated optical parameters provided upstream into physical quantities that can be used for precise positioning. Its processing chain begins with the nonlinear solution module. This module receives a set of normalized periodic parameters and wavelength calibration coefficients output from the data acquisition unit. The normalized periodic parameter set includes filtered and compared grating geometry information, while the wavelength calibration coefficients reflect corrections to the light source wavelength. The core task of this module is to solve a set of nonlinear equations relating the grating period, operating wavelength, and Talbot distance. The Talbot distance describes the specific propagation distance at which light produces a self-image after passing through a periodic grating structure and is a key theoretical parameter determining the exposure focal plane. The module uses the Newton-Raphson iterative method to process this set of equations, which finds a mathematical solution through iterative approximation. Based on the input set of normalized periodic parameters and wavelength calibration coefficients, the module sets the coefficients and initial values ​​of the equations, updates the solution estimate by calculating function values ​​and their derivatives, until the result meets the preset convergence conditions. Finally, the module outputs a sequence of Talbot distance values, which theoretically defines a series of positions where the light field forms a clear image within the photoresist under different process conditions.

[0035] However, theoretical calculations do not consider the dispersion effect that occurs when light propagates in real materials. Light of different wavelengths travels at slightly different speeds in a medium, which leads to a systematic deviation between the actual imaging position and the theoretical Talbot distance—a distance offset. The multiphysics compensation module is designed to correct this effect. This module receives the Talbot distance numerical sequence output by the nonlinear solver. To calculate the distance offset caused by dispersion, the multiphysics compensation module calls a preset finite-difference time-domain (FDTD) simulation model. FDTD simulation is a numerical method that simulates the interaction between light and matter by directly solving Maxwell's equations. The multiphysics compensation module combines this simulation model with the dispersion coefficient of a specific photoresist material. The material dispersion coefficient quantitatively describes the characteristics of the material's refractive index as a function of wavelength and is a key input for calculating the dispersion effect. Through simulation calculations, the multiphysics compensation module can quantify the specific distance offset caused by the grating periodic dispersion effect. Subsequently, the multiphysics compensation module performs matrix addition, adding the calculated distance offset element-wise to the received original Talbot distance numerical sequence. The essence of this operation is to physically compensate for theoretical values, and its output is a multi-dimensional Talbot distance matrix. This matrix not only includes the theoretical Talbot distances at different locations, but also incorporates correction information from material dispersion, thus better reflecting the actual light field propagation behavior.

[0036] Considering the unavoidable process fluctuations in actual manufacturing environments, such as slight temperature fluctuations or minor differences between material batches, the system needs to possess a certain degree of robustness. The role of the redundancy generation module is to simulate and accommodate such fluctuations. The redundancy generation module receives the multi-dimensional Talbot distance matrix output by the multiphysics compensation module. Then, the redundancy generation module superimposes a process tolerance random parameter conforming to a preset normal distribution onto each element of this matrix. This operation is equivalent to adding a random perturbation to each physically compensated Talbot distance value. This perturbation is not an error, but a proactively introduced, statistically consistent tolerance, used to pre-imagine the potential impact of process fluctuations at the algorithm level, ensuring that the generated exposure positioning parameters themselves cover a reasonable fluctuation range, thereby improving the stability of the final process result. After generating the matrix with the embedded tolerance parameters, the redundancy generation module needs to transmit it to the downstream positioning unit. To facilitate reliable transmission of this structured numerical data between system units, the redundancy generation module performs ASCII encoding on the matrix with the embedded tolerance parameters. ASCII encoding is a standard scheme for converting between characters and numbers. It can convert the values ​​in a matrix into a sequence of standard characters, forming a compact and parseable data stream. Finally, the module sends the encoded data stream to the positioning unit, providing complete input information for subsequent coordinate transformation and precision motion control.

[0037] The core task of the positioning unit is to accurately convert the optical position parameters calculated upstream into control commands for driving the mechanical motion system, and to compensate for various physical interferences in real time during this process. The coordinate transformation module, as the starting point, receives ASCII encoded data output from the parameter calculation unit. This character sequence is essentially a standardized, cross-platform-transferable representation of the multi-dimensional Talbot distance matrix. The module first decodes the ASCII encoded data, restoring the character sequence into a multi-dimensional numerical matrix including Talbot distance parameters according to preset encoding rules. Next, the coordinate transformation module uses a preset homogeneous coordinate transformation algorithm to process this matrix. Homogeneous coordinate transformation is a common mathematical tool in robotics and computer graphics used to describe the position and orientation of objects. By introducing an additional dimension, it can represent complex spatial transformations such as translation, rotation, and scaling using unified matrix multiplication. The module extracts key Talbot distance parameters from the decoded multi-dimensional Talbot distance matrix, and then uses the homogeneous coordinate transformation algorithm to systematically convert the optical parameters characterizing the self-imaging position of the light field into the XYZ three-axis physical coordinate offsets required by the driving mechanism, including three-dimensional translation vectors and three-dimensional rotation matrices. This conversion process establishes a precise mapping between the optical computational domain and the mechanical motion domain, and the generated absolute positioning reference data provides the theoretical target position for the motion system.

[0038] However, the theoretical absolute positioning reference is subject to interference from environmental thermal drift and mechanical vibration during real-time movement, causing deviations between the actual position of the base and the theoretical reference. The thermal drift compensation module is designed to dynamically correct this deviation. The thermal drift compensation module receives the absolute positioning reference data output from the coordinate transformation module. Simultaneously, the thermal drift compensation module uses an integrated laser interferometer to acquire the actual position data of the base along the X, Y, and Z axes in real time. The thermal drift compensation module performs differential calculations between the acquired real-time actual position data and the received absolute positioning reference data, i.e., subtracting each coordinate component, to generate real-time position deviation data reflecting the position error.

[0039] To quickly and stably drive the actuator to eliminate errors based on position deviation data, thermal drift compensation employs a PID controller with preset proportional, integral, and derivative coefficients. PID is a widely used closed-loop control algorithm; the proportional term reacts to the current error, the integral term eliminates historical accumulated errors, and the derivative term predicts the error trend. The module inputs real-time position deviation data into this PID controller. The controller's output is dynamically modulated using an exponential decay function. Exponential decay functions are commonly used in control systems to smooth output signals and prevent overshoot. The modulated output is the voltage adjustment required to drive the piezoelectric ceramic actuator. The piezoelectric ceramic actuator can produce nanometer-level precision displacement under voltage drive. Based on the calculated voltage adjustment, the module corrects the original absolute positioning reference data in real time, ultimately outputting dynamically compensated positioning coordinates that are closer to the actual reachable position of the substrate.

[0040] The task of the motion command compilation module is to convert the spatial coordinates calculated by the upper-level algorithm into precise timing commands that the underlying motor driver can recognize and execute. The module receives the compensated positioning coordinates output by the thermal drift compensation module. A voice coil motor is an actuator that directly converts electrical signals into linear motion; its displacement resolution determines the minimum motion step size of the system. The module converts each component of the compensated positioning coordinates into a corresponding pulse sequence signal based on the displacement resolution of the voice coil motor. Each pulse corresponds to a tiny, fixed step distance of the motor; the number of pulses determines the total displacement, and the pulse frequency determines the motion speed.

[0041] To ensure strict synchronization of the XYZ axes and prevent trajectory distortion due to timing errors, the motion command compilation module transmits pulse sequence signals via the EtherCAT industrial Ethernet bus. The EtherCAT bus features a distributed clock synchronization mechanism that locks the clocks of all devices on the bus to the master clock, achieving sub-microsecond-level high-precision synchronization. The motion command compilation module utilizes this mechanism to strictly align the generated pulse sequence signals corresponding to the XYZ axes, ensuring highly synchronized motion commands across the three axes. Ultimately, the module generates and outputs synchronized motion commands with phase differences controlled within an extremely narrow time window. These commands are sent via the bus to the drivers of each axis, coordinating and driving the voice coil motors to precisely move the base to the three-dimensional spatial position specified by the compensated positioning coordinates.

[0042] The core function of the motion control unit lies in achieving deep coordination between high-precision mechanical motion and optical field control, and its operation begins with the trajectory prediction module. This module receives synchronous motion commands output from the motion command compilation module of the positioning unit. The synchronous motion commands are a series of pulse signals driving the actuator, which encode the sinusoidal control signal of the drive shaft. The trajectory prediction module analyzes this sinusoidal drive signal, and by solving the differential equation describing the motion of the base, predicts the instantaneous position of the base at every moment within a complete future motion cycle, generating a set of instantaneous position distribution data that varies with time. The predicted data inevitably includes high-frequency noise introduced by mechanical resonance or external interference. The module performs a Fourier series expansion on the instantaneous position distribution data, a mathematical method that decomposes any periodic function into a series of sine and cosine harmonic components. After expansion, the module filters out all higher-order harmonic components with frequencies higher than a preset cutoff frequency, thereby smoothing the motion trajectory and eliminating high-frequency jitter components that may cause image blurring. After this processing, the module generates and outputs optimized motion trajectory parameters, which not only include smoothed position information, but also retain the filtered fundamental phase information used to describe the main rhythm of the motion.

[0043] The optical field synchronization module is responsible for strictly synchronizing the optimized mechanical motion trajectory with the optical field distribution. The module receives the optimized motion trajectory parameters output by the trajectory prediction module and extracts key fundamental phase information. This fundamental phase information precisely describes the real-time phase of the motion platform within its cycle. Based on the extracted fundamental phase information, the module generates a corresponding phase modulation voltage signal via a digital-to-analog converter. This phase modulation voltage signal is then applied to a liquid crystal spatial light modulator. A liquid crystal spatial light modulator is an optical device composed of a pixel array, where the refractive index of each pixel changes with the applied voltage. When the phase modulation voltage signal is applied to each pixel, the arrangement of the liquid crystal molecules changes, resulting in a controllable phase delay in the light waves transmitted through each pixel. By programming the pixel array, the overall wavefront phase distribution of the light field transmitted through the modulator can be dynamically adjusted. In this way, the phase change of the optical field can be matched with the phase of the mechanical motion in real time, compensating for the relative delay or distortion caused by the motion. After this adjustment, the optical field synchronization module generates and outputs synchronized optical field parameters that achieve spatiotemporal synchronization.

[0044] The polarization matching module, building upon the previous synchronization, further refines the polarization characteristics of the light field. The module receives the synchronized light field parameters from the light field synchronization module. To evaluate the current polarization state of the light field, the module measures the Stokes vector of these parameters using a polarization analyzer. The Stokes vector is a set of four parameters that completely describe the polarization state of a light wave. Based on the measured Stokes vector, the module calculates the deviation between the current actual polarization state and the target circularly polarized state required by the system; this deviation can be quantified as a difference in ellipticity or azimuth. To eliminate the deviation, the module dynamically adjusts the driving voltage applied to the subsequent liquid crystal phase retardation film. Under voltage driving, the liquid crystal phase retardation film can change its birefringence characteristics, thereby rotating the polarization direction of the transmitted light wave. Based on the calculated polarization state deviation, the module adjusts the driving voltage in real time through a closed-loop control circuit, precisely adjusting the polarization direction of the light field transmitted through the liquid crystal phase retardation film to the desired target circularly polarized state. Finally, the polarization matching module generates and outputs polarization-consistent optical field parameters with highly consistent polarization characteristics, and transmits these parameters to the downstream optical field control unit to provide an input optical field with guaranteed polarization state for subsequent uniformity calibration and defect suppression.

[0045] The optical field manipulation unit is responsible for finely controlling the spatial intensity distribution, polarization state stability, and beam spatial profile of the input optical field from the motion control unit, which already has polarization consistency, in order to generate a high-quality optical field that meets the processing requirements of the defect suppression unit.

[0046] The light intensity uniformity calibration module begins by receiving the polarization-consistent light field parameters output by the polarization matching module. These parameters characterize the light field, optimized upstream, with the polarization state meeting requirements. The module first drives a two-dimensional photodetector array to perform point-by-point or area-by-area scanning of the beam cross-section for the actual light field. Each pixel in the detector array converts the received light signal into an electrical signal. After analog-to-digital conversion, the module obtains a complete light field intensity distribution data map, with each pixel corresponding to a specific intensity value. Next, the module compares the intensity value of each pixel in the distribution map with a preset light intensity uniformity threshold. This threshold is typically set as a specific percentage of the average light field intensity. Through comparison, the module can identify all pixels with intensity values ​​below this threshold; these pixels constitute low-intensity regions in the light field. For compensation areas, the module generates a grayscale compensation mapping table. This mapping table is essentially a lookup table corresponding one-to-one with the pixel positions in the detector array. Only pixels in the identified low-intensity regions are assigned a grayscale gain coefficient greater than 1, while the gain coefficient for other regions is 1. Finally, based on this grayscale compensation mapping table, the module converts the gain coefficient corresponding to each pixel into a driving voltage, which is then applied to the corresponding liquid crystal pixel unit on the liquid crystal spatial light modulator. The applied voltage changes the arrangement of the liquid crystal molecules, thereby altering the transmittance of the pixel unit and achieving pixel-level light intensity modulation in low-intensity areas. After this compensation, the overall intensity uniformity of the light field is improved, and the module outputs the light field parameters after intensity uniformity calibration.

[0047] The dynamic polarization adjustment module maintains the stability of the polarization state of the light field during subsequent transmission and transformation. The module receives the light field parameters after intensity uniformity calibration. Due to slight stress birefringence or alignment errors in optical components, even circularly polarized light can experience polarization state drift. The module guides the light field through a λ / 4 waveplate and precisely sets the fast axis of the waveplate relative to the polarization direction of the incident light, ensuring that the outgoing light is ideally circularly polarized. In the optical path after the λ / 4 waveplate, the module integrates a Stokes vector analyzer. This analyzer measures all four Stokes parameters of the light field passing through it in real time; this set of parameters completely defines the polarization state of the light wave. Based on the measured Stokes vectors, the module calculates the actual polarization elliptic parameters of the current light field and compares them with the ideal parameters of the target circularly polarized state, calculating the purity deviation between the two. This deviation can be quantified as a difference in ellipticity or degree of polarization. When the purity deviation exceeds the allowable range, the module dynamically adjusts the driving voltage applied to the λ / 4 waveplate through a closed-loop feedback circuit. Changes in voltage cause a slight change in the phase delay generated by the waveplate, thereby correcting the polarization state of the output light field. This adjustment process is real-time and can adapt to changes in the environment or system state, ensuring that the output light field remains in a highly stable circularly polarized state. The module then generates and outputs polarization-consistent light field parameters.

[0048] The edge gradient control module aims to optimize the spatial energy distribution profile of the light field, converting the common Gaussian distribution beam into a flat-top distribution beam that is more conducive to uniform exposure. The module receives polarization-consistent light field parameters from the dynamic polarization adjustment module. Next, the module guides the light field through a carefully designed microlens array. This array consists of many tiny lens elements arranged regularly. When an input light field with a Gaussian distribution is incident on the microlens array, each microlens converges and redistributes the incident light. The superposition effect of numerous microlenses acts together in the far field or on a specific image plane, ultimately converting the Gaussian distribution into a flat-top distribution with a steep edge. However, the intensity gradient of the converted light field at the boundary between light and dark areas, i.e., the sharpness of the edge transition, directly affects the edge quality of the exposed pattern. Based on the Fresnel diffraction integral formula, the module establishes a physical model from the microlens array parameters to the output light field distribution and calculates the intensity gradient distribution of the output light field in the edge region. To obtain smooth edges that meet process requirements, the module uses a gradient descent algorithm to iteratively adjust the spacing parameters of the lens elements in the microlens array. The gradient descent algorithm calculates the derivative of the edge gradient with respect to the spacing parameter under the current parameters, and updates the spacing parameter along the gradient descent direction, gradually searching for the optimal lens spacing configuration that makes the calculated edge intensity gradient lower than a preset gradient threshold. After optimization, the module generates and outputs the light field parameters after edge gradient control, and transmits these parameters to the downstream defect suppression unit, providing a uniformly illuminated and clearly edged light field input for accurate defect identification and compensation.

[0049] The defect suppression unit analyzes, identifies, and processes the optical field after upstream optimization and modulation. Its core objective is to predict potential defects and generate targeted exposure control commands. The diffraction integral calculation module is the starting point of this process. This module receives the optical field parameters after edge gradient control from the edge gradient control module of the optical field modulation unit. These parameters include wavefront phase and amplitude information characterizing the expected distribution of the light beam on the exposure plane. Based on the Fast Fourier Transform algorithm, the module performs Fresnel diffraction integral calculation on the two-dimensional wavefront described by these optical field parameters. This calculation simulates the propagation process of light waves from the modulator to the photoresist surface. The calculation result is the initial light intensity distribution map that the light may form within the photoresist.

[0050] Due to factors such as simplified computational models, environmental disturbances, and detector noise, the initial intensity distribution map obtained from a single calculation may contain random noise. To suppress noise and obtain more reliable predictions, the module performs double averaging on the initial intensity distribution map. First, in the spatial domain, the module applies a sliding window averaging filter to the pixel array of the distribution map. This operation replaces the original value of a pixel with the average intensity of each pixel and its neighboring pixels, which smooths the image and suppresses high-frequency spatial noise. Second, in the temporal domain, the module averages multiple frames of initial intensity distribution maps acquired or calculated continuously over time. Averaging multiple frames of the same scene effectively suppresses random noise that fluctuates over time. After these two averaging processes (spatial and temporal), the module generates and outputs a clearer and more stable predicted intensity distribution map, which theoretically approximates the intensity pattern that might be formed during actual exposure.

[0051] The defect identification module receives the predicted light intensity distribution map output by the diffraction integral calculation module. The core of this module is a pre-trained convolutional neural network model, which used a large number of light intensity distribution map samples, including those labeled with various defects, during the training phase. When the predicted light intensity distribution map is input into this convolutional neural network model, the model's multi-layer convolution and pooling operations automatically extract deep features from the image. After classification by fully connected layers, a defect region mask map is finally output. This mask map is a binary image, where white pixels mark regions identified as defects by the model, and black pixels represent normal regions.

[0052] Based on the generated defect area mask image, the module performs two key statistical calculations. The first is to calculate the defect repair rate, specifically by counting the total number of white pixels in the defect area mask image and calculating the percentage of their area relative to the total number of pixels in the entire image. The second is to calculate the yield rate. Based on a preset acceptable pixel intensity range, the module counts the number of pixels whose intensity values ​​fall within this range in the original predicted light intensity distribution image and calculates their percentage relative to the total number of pixels. Simultaneously, the module also needs to provide data support for subsequent exposure compensation. Therefore, it extracts the original pixel intensity values ​​of the areas marked white by the defect area mask from the predicted light intensity distribution image, organizes these intensity values ​​to generate defect intensity distribution data, and outputs it.

[0053] The exposure threshold adjustment module receives defect intensity distribution data output by the defect identification module. The module processes data on a per-individual defect region basis. For each defect region identified in the defect intensity distribution data, the module calculates the average intensity of all pixels within that region. This average intensity value is compared to a system-preset target intensity value that achieves good exposure results; the difference is used to calculate the amount of photoresist exposure energy compensation needed to compensate for defects in that specific region.

[0054] To convert the calculated exposure energy compensation into executable process parameters, the thermal drift compensation module employs a PID controller with a preset proportional coefficient Kp=2.5, integral coefficient Ki=0.01, and derivative coefficient Kd=0.1. Based on the input energy compensation, the PID controller dynamically calculates and outputs the corresponding exposure time adjustment value for the defective region through the coordinated operation of its proportional, integral, and derivative components. This adjustment value may increase the exposure time to compensate for areas with insufficient light intensity or decrease the time to avoid overexposure. The module processes all defective regions, generating an independent exposure time adjustment value for each region. Finally, the module integrates the exposure time adjustment values ​​of all regions, combining them with their respective spatial coordinate information to generate a structured regional exposure control parameter set. This parameter set is ultimately sent to the light field control unit, guiding it to apply differentiated exposure energy to different spatial regions during subsequent actual exposure processes, thereby achieving precise energy compensation and suppression for the identified defective regions.

[0055] The quality verification unit constitutes a key feedback loop for the system's self-optimization and closed-loop control. Its work begins with the quantitative detection and evaluation of the actual effects of the exposure compensation process on the affected areas. The morphology detection module receives regional exposure control parameters output from the exposure threshold adjustment module of the defect suppression unit. These parameters define the differentiated exposure strategies applied to different defect areas and their corresponding physical locations in previous process cycles. The module first parses the regional exposure control parameters, extracting the spatial coordinate description of each independent exposure area. Based on these coordinates, the module drives the high-precision stage to move and precisely align the corresponding exposed areas on the substrate platform carrying the master plate to the center of the objective lens field of view of the optical microscope. After the position stabilizes, the module controls the optical microscope to automatically focus and photograph the surface of the exposed area at a preset magnification and depth of field, acquiring high-resolution surface morphology image data. To obtain the optical material properties of the same area, after completing the morphology imaging, the module keeps the stage position unchanged and switches to the ellipsometer integrated into the system. An ellipsometer uses a probe beam with a known polarization state to illuminate the same exposed area at a preset incident angle, and analyzes the change in polarization state of the reflected or transmitted light to measure a set of ellipsometric parameters. These parameters are directly related to the optical constants of the material. The module performs inverse calculations on the measured ellipsometric parameters based on the Shermeier equation. The Shermeier equation is a classical physical model describing the relationship between the refractive index and wavelength of a material. Through iterative fitting, the distribution of the refractive index in the local region as a function of space can be calculated, generating refractive index distribution data. Finally, the module aligns and integrates the morphological image data characterizing the geometric structure with the refractive index distribution data characterizing the optical properties in spatial coordinates, generating and outputting morphological and refractive index detection data that comprehensively reflects the physical state of the repaired area.

[0056] The refractive index analysis module extracts quantitative indicators from the inspection data and establishes a correlation model between the process effect and key parameters. The module receives morphology and refractive index inspection data output from the morphology inspection module. Processing begins with the morphology image data. To quantify the actual repair effect, the module automatically identifies and extracts the contour boundaries of the actual repaired defect areas in the morphology image based on mature image segmentation algorithms, such as thresholding or edge detection algorithms. Based on this contour, the module calculates the pixel area of ​​the defect area in the image and converts it into the actual physical area using the calibration coefficient of the optical microscope. Simultaneously, by analyzing the grayscale or height information in the morphology image, the module can calculate the depth or height variation of the defect area. Based on the calculated area and depth, the module can calculate a specific value reflecting the actual repair degree, i.e., the actual defect repair rate, according to a preset formula. On the other hand, the module simultaneously processes the refractive index distribution data. It compares the average actual refractive index value of each pixel or region in the refractive index distribution data with the theoretical design refractive index value specified in the master design file. By calculating the difference between the actual and design values, the module can identify two key deviations: one is the anomaly in the periodic change of refractive index, i.e., periodic deviation; the other is the deviation of the equivalent refractive index from the expected value as wavelength changes, i.e., wavelength drift. Thus, the module obtains three sets of core characteristic data: actual defect repair rate, periodic deviation, and wavelength drift.

[0057] To learn from historical data and predict process adjustment directions, the module uses the actual defect repair rate, cycle deviation, and wavelength drift as a set of input feature vectors, feeding them into a pre-trained Support Vector Machine (SVM) model. SVM is a machine learning algorithm suitable for small-sample, high-dimensional classification and regression problems. This model has already learned from a large number of similar process data pairs during the training phase. When the current feature vectors are input, the model can mine and establish a complex nonlinear mathematical correlation model between the actual defect repair rate and cycle deviation and wavelength drift. Based on this mathematical model, the module can analyze whether the current actual defect repair rate is below expectations. If the actual defect repair rate is not as expected, the model can determine whether the main reason is excessive cycle deviation or significant wavelength drift, and generate process parameter correction suggestions with clear physical meaning. For example, correction suggestions might include adjusting the material deposition amount to compensate for material changes, or adjusting the etching rate parameter value to correct the etching morphology. The module ultimately outputs this process parameter correction suggestion, which includes specific quantitative adjustment recommendations.

[0058] The data inversion module is the final execution link of the closed-loop control, responsible for translating analysis conclusions into specific instructions to drive the actions of upstream units. The module receives process parameter correction suggestions from the refractive index analysis module. The module's job is to parse and distribute these suggestions. First, the module parses correction instructions from the process parameter correction suggestions to extract correction instructions for the standardized periodic parameter set used by the data acquisition unit in future operations, such as adjusting a filter parameter or calibration coefficient. Simultaneously, the module also parses correction instructions for fine-tuning the laser output center wavelength to compensate for detected wavelength drift. For the correction instructions for the standardized periodic parameter set, the module encapsulates them into a JSON data packet according to a preset data structure. JSON is a lightweight data exchange format with good readability and broad software support. The module sends the encapsulated JSON data packet as a network request to the data fusion module of the data acquisition unit via the HTTP protocol, thereby achieving remote updates of the parameter set. For laser wavelength correction instructions, due to the need for real-time and reliable hardware control, the module uses a different interface. It converts wavelength correction commands into specific temperature control parameter adjustment values ​​that can be recognized by the laser's internal temperature control circuit. Then, via the SPI interface, the adjustment values ​​are directly written into the laser's temperature control register through synchronous serial communication. By changing the value in the temperature control register, the laser's operating temperature undergoes a slight change, thereby precisely adjusting its output wavelength. After successfully sending the parameter correction command to the data fusion module and writing the wavelength and temperature control parameters into the laser register, the data inversion module generates and outputs a closed-loop control feedback signal. This signal signifies the completion of all quality verification and feedback adjustment work for this process cycle. The system can then start the next cycle based on the updated parameters, thus achieving a complete closed loop from result detection to cause analysis and parameter adjustment.

[0059] The system also includes an auxiliary optimization and calibration mechanism running parallel to the core workflow, aimed at continuously improving the system's long-term stability and process consistency. The data benchmark establishment module is the starting point for initial calibration and periodic recalibration. This module receives the closed-loop control feedback signal output from the data inversion module of the quality verification unit. This signal marks the end of a complete process cycle and may carry information indicating that process parameters need adjustment. The module then starts under strictly controlled constant temperature and specific relative humidity conditions. In this stable environment, the module jointly drives the data acquisition unit, parameter calculation unit, positioning unit, motion control unit, light field modulation unit, and defect suppression unit to operate collaboratively, but it does not perform actual processing; instead, it executes a preset calibration program. During operation, the module collects process data from light source characteristics to the final predicted light field distribution, reflecting the system's current actual state. By fitting the collected optical system aberration data with Zernike polynomials, the module can analyze the systematic errors of the optical path. Zernike polynomials are commonly used orthogonal basis functions describing optical wavefront aberrations. After fitting, a standardized parameter set including 256 error compensation coefficients is generated. This parameter set is essentially a "fingerprint" of the current system's optical performance. Finally, the module outputs this standardized parameter set as the initial calibration parameters to the data fusion module of the data acquisition unit to update or correct the benchmark for subsequent data acquisition, thereby compensating for system errors at the source.

[0060] The spatial positioning optimization module aims to achieve extreme positioning control that surpasses the conventional accuracy of the thermal drift compensation module during motion. The module receives the compensated positioning coordinates output from the thermal drift compensation module of the positioning unit; these coordinates have already been corrected by real-time PID control. To obtain more accurate actual position feedback, the module uses a dual-frequency laser interferometer to measure the displacement data of the base in real time across six degrees of freedom: translation along the XYZ axes and rotation around these axes. The dual-frequency laser interferometer has stronger resistance to environmental interference compared to a single-frequency interferometer. Simultaneously, the module also collects the base's motion acceleration data using a high-sensitivity accelerometer. Displacement data provides absolute position, while acceleration data reflects instantaneous dynamics. The module uses a Kalman filter algorithm to fuse these two sets of data. Kalman filtering is an optimal estimation algorithm that recursively estimates the optimal value of the system state based on the system's dynamic model and noisy measurements. Through fusion, the module generates fused position data with higher accuracy and reliability. Next, the module calculates the difference between this fused position data and the received compensated positioning coordinates. This difference may originate from sensor noise, model errors, or unmodeled dynamic effects. Based on this difference, the module calculates a nanometer-level position correction. Finally, the module directly drives a high-precision piezoelectric ceramic actuator based on this position correction to fine-tune the substrate position and generate positioning accuracy verification data that records the optimization process and results.

[0061] The task of the spatiotemporal synchronization control module is to solve the deeper timing synchronization problem between mechanical motion and photoelectric signal acquisition, in order to eliminate matching errors caused by minute delays. The module receives optimized motion trajectory parameters output from the trajectory prediction module of the motion control unit. These optimized parameters include fundamental phase information describing the main rhythm of the motion. The module extracts the fundamental phase information from the parameters. Using field-programmable gate array (FPGA) hardware, the module generates a hardware clock signal that is strictly synchronized with this fundamental phase information. The FPGA hardware, with its parallel processing and programmability, can achieve extremely low latency and extremely high deterministic clock signal generation. Using this hardware clock signal as the absolute time reference, the module re-aligns the timing of the XYZ three-axis motion commands to be issued through the distributed clock synchronization mechanism of the EtherCAT bus, compressing the phase difference between the three-axis commands to within an extremely short few nanoseconds. More importantly, the module simultaneously sends this hardware clock signal as a trigger signal to the light intensity uniformity calibration module of the light field control unit. The photodetector array in the light intensity uniformity calibration module needs to acquire light field data at specific times. This trigger signal ensures that the timing of light field acquisition is strictly locked to the phase of mechanical motion, thus solving the problem of light field distribution measurement distortion caused by asynchronous sampling. After completing the above synchronization, the module generates and outputs spatiotemporal synchronization optimization parameters.

[0062] The performance feedback module incorporates reinforcement learning mechanisms from machine learning, enabling the system to adaptively optimize its internal algorithm parameters. The module receives process parameter correction suggestions from the refractive index analysis module of the quality verification unit. These suggestions include specific recommendations for adjusting material compensation and etching rates, indirectly reflecting the gap between the current process results and the ideal state. The module extracts the material compensation and etching rate adjustment values ​​from these suggestions. The module treats these values ​​as "environmental rewards" in reinforcement learning; higher reward values ​​indicate better results from the current system's internal parameter settings. The module employs a Q-learning reinforcement learning algorithm, using rewards as feedback to dynamically and iteratively optimize two key algorithm parameters: the weighting coefficients of the noise filtering function used in the data fusion module of the data acquisition unit, and the convergence threshold of the Newton-Raphson iteration method in the nonlinear solution module of the parameter calculation unit. By continuously trying to fine-tune the parameters and observing the resulting changes in "rewards," the algorithm learns a set of parameter configurations that improve the long-term process performance of the system. After optimization, the module encapsulates the obtained optimal weight coefficients and convergence judgment thresholds into parameter packages named correction coefficients Kp, Kd, ​​and Ki, and feeds them back to the nonlinear solution module of the parameter calculation unit to update its calculation process, thereby forming a higher-level learning and optimization closed loop for the algorithm itself.

[0063] Secondly, please refer to Figure 1 This invention provides a composite optical waveguide master plate defect compensation photolithography method, applied to the composite optical waveguide master plate defect compensation photolithography system as described above, comprising: Step 1: Obtain master design parameters and light source characteristic data; Step 2: Receive the design parameters and light source characteristic data, and generate positioning reference parameters by solving a system of nonlinear equations; Step 3: Receive the positioning reference parameters, convert the Tiber distance parameters into XYZ three-axis physical coordinate offsets using a preset coordinate transformation algorithm, and generate compensated positioning coordinates; Step 4: Calculate and generate motion trajectory data based on the offset between the compensated positioning coordinates and the physical coordinates; Step 5: Receive the motion trajectory data, optimize the light field distribution through light intensity uniformity calibration and dynamic polarization adjustment, and generate optimized light field parameters; Step 6: Receive the optimized light field parameters, generate a dual average light intensity distribution based on Fresnel diffraction integral calculation, suppress periodic and non-periodic defects, and generate defect repair data including regional exposure control parameters. Step 7: Receive the defect repair data and generate a repair rate statistical report through optical detection and electrical performance testing; Steps 1 to 7 are executed sequentially through standardized interfaces, forming a closed-loop control process from data acquisition to quality verification.

[0064] In step 1, optical character recognition technology was used to parse the grating structure parameters in the Gerber file, and a spectrometer scanned the laser output spectrum at a resolution of 1 nm. In step 2, the convergence condition of the Newton-Raphson iterative method was set to a residual of less than 1e-6, and the FDTD simulation used Yee mesh partitioning, with a perfectly matched layer absorption boundary condition applied to the computational domain boundary. In step 3, during coordinate transformation, the homogeneous transformation matrix included three sub-matrices: rotation, translation, and scaling. Singular value decomposition was used to ensure the numerical stability of the coordinate transformation. In step 4, motion trajectory optimization used B-spline curve fitting, and motion jitter was eliminated by adjusting the control point weights. In step 5, light field manipulation used a liquid crystal response time compensation algorithm to offset the liquid crystal molecule orientation hysteresis effect, and edge gradient optimization used the Sobel operator to detect regions of sudden light intensity changes. In step 6, during defect suppression, a double averaging algorithm was used to average the diffraction pattern twice, in the spatial and temporal domains, achieving a suppression ratio of over 30 dB. In step 7, the repair rate statistics used the Monte Carlo method for confidence interval estimation, with a confidence level set at 95%. The closed-loop feedback in step 8 triggers parameter calibration every 2 hours. The calibration data is transmitted encrypted via HTTPS protocol, and the update process does not affect the normal production cycle.

[0065] Solving the expression for the Tiber distance formula using the Newton-Raphson iterative method: ; The estimated Talbot distance from the nth iteration characterizes the spatial propagation characteristics of the grating period at a specific wavelength. The residual function of the nonlinear equation system is defined as the grating period. ,wavelength With material dispersion coefficient Functional relation The Jacobian matrix of the residual function, representing the function. Data processing path for the partial derivative matrix of each variable: by establishing a matrix including the grating period design value. Measured wavelength and material dispersion coefficient The nonlinear equations are solved using an iterative algorithm to find the Timber distance. The specific steps are as follows: First, set the initial estimated value. Calculate the current residual and its Jacobian matrix Then iteratively update using formulas. until the absolute value of the residual is less than a preset threshold. Finally, a high-precision numerical sequence of the Tyber distance was obtained.

[0066] The formula for calculating the dispersion effect using the FDTD simulation model is as follows: ; The Talber distance shift caused by dispersion reflects the cumulative effect of optical path difference at different wavelengths. The wavelength-dependent complex refractive index function, including the real part (refractive index) and the imaginary part (absorption coefficient). The grating period design value determines the basic spatial frequency data processing path of light field diffraction: In the FDTD simulation model, a data processing path is established that includes the material dispersion coefficient. Set the mesh size in the three-dimensional computational domain. ( To achieve the shortest operating wavelength, a perfectly matched layer (PML) absorption boundary condition is employed. A plane wave excitation source is applied, and the propagation phase delay of the light wave in the grating structure is calculated. The wavelength is then extracted through numerical integration. With refractive index The partial derivative relationship is used to finally integrate and obtain the dispersion-induced Timber distance correction. .

[0067] The formula for phase compensation in a liquid crystal spatial light modulator is as follows: ; Phase distribution function of pixels in spatial light modulator, with values ​​ranging from [0, 2π). The intensity components of the probe light along the x / y polarization direction reflect the deflection state of the liquid crystal molecules. Data processing path: Polarization state parameters are obtained through the Stokes vector monitoring system. Calculate the polarization purity index When the polarization purity is below the threshold of 0.995, closed-loop feedback regulation is initiated: first, the voltage-phase response curve is discretized into a piecewise linear function, and then, based on the current voltage... Calculate the corresponding phase increment A modulation signal is generated by a DAC converter to achieve nanometer-level precision control of the orientation angle of liquid crystal molecules.

[0068] The formula for converting the light field of a flat-top microlens array is as follows: ; Normalized light intensity distribution at a distance r from the optical axis The beam waist radius of the flat-top light field is defined as the position where the intensity drops to 1 / e² of the peak value. Data processing path: Zernike polynomials are used to fit the Gaussian beam wavefront distortion to calculate the fill factor of the microlens array. (d is the microlens diameter, D is the array period). Lens spacing is optimized using a genetic algorithm. Make the edge gradient Satisfy in the transition region (r≥0.9w0), ultimately forming a flat-top light field distribution with illuminance uniformity >98%.

[0069] Formula expression for the statistical model of defect repair rate: ; Overall repair rate, value range [0,1] : Measured depth value of the i-th defect region Design value depth Measured refractive index of the defect region Design value refractive index Refractive index weighting coefficient (0 < α < 1) data processing path: First, obtain the data through ellipsometer measurement. The parameters are used to invert and calculate the refractive index distribution using the Sellmeier equation: ; in Let be the material's optical constant. Combining morphology inspection data, a support vector machine (SVM) is used to classify defect types and construct a feature space. The weighting coefficients are determined through cross-validation. Finally, the repair rate of each defective area is calculated and the average value is obtained.

[0070] After wavelet denoising preprocessing, the closed-loop control data stream detection data is used to extract key feature parameters through principal component analysis (PCA) and then input into a convolutional neural network (CNN) for defect type identification. The identification result triggers the PID controller to generate correction commands, which are transmitted to the piezoelectric ceramic driver via the EtherCAT bus to achieve nanometer-level positioning compensation. System calibration is triggered every 2 hours, with correction commands transmitted encrypted via HTTP protocol and laser temperature control parameters updated via the SPI interface to ensure the accuracy of the entire closed-loop control process.

[0071] The adaptive learning process involves augmenting historical detection data (including noise injection and geometric transformation) to construct an expanded dataset. A transfer learning strategy is employed, freezing the parameters of the feature extraction layer in the pre-trained model and training only the weights of the fully connected layers. The model's generalization ability is evaluated using a confusion matrix, and the hyperparameters are dynamically adjusted using the Q-learning algorithm (learning rate η∈[0.001,0.1], batch size B∈[16,128]), ultimately achieving a system performance with a defect classification accuracy >99.5%.

[0072] When the data acquisition unit uses optical character recognition (OCR) technology to parse Gerber files, it specifically uses the open-source library pygerber for vector graphics parsing, setting the recognition accuracy to 0.01 mil. For extracting the grating period design values, a Bezier curve fitting algorithm is used, with a fitting order of 5 and a tolerance range of ±0.5 μm. The spectral calibration module uses a Yokogawa AQ6370 spectrometer, scanning in the 1520-1570 nm band at a wavelength resolution of 1 nm. The CCD detector sampling frequency is set to 100 Hz, and spectral intensity data is transmitted in real-time via a USB 3.0 interface. In the data fusion module, principal component analysis is implemented using the sklearn library, retaining principal components with a cumulative contribution rate >95%. Weighted least squares is used for least squares fitting, with a weighting coefficient of 1 / σ² (σ being the standard deviation).

[0073] In the parameter calculation unit, when solving using the Newton-Raphson iterative method, the initial value T0 is set to... / (n(λ)Δ), the convergence condition is set to the absolute value of the residual <1× The maximum number of iterations is 100. FDTD simulations were performed using Lumerical FDTDSolutions software, with a mesh size of λ_min / 10 (λ_min = 1520 nm). A perfectly matched layer (PML) was applied as a boundary condition, and the simulation time step satisfied the CFL stability condition Δt = Δx / (2c). Material dispersion coefficient... The experimental data were obtained by fitting the Sellmeier equation, with a fitting wavelength range covering 1520-1570 nm and a goodness of fit R²>0.999.

[0074] The PID controller parameters in the positioning unit are set to Kp=2.5, Ki=0.01, Kd=0.1, with an integral saturation limit of ±5V and a derivative time constant τ=0.1s. EtherCAT bus synchronization uses a Beckhoff BK9000 bus terminal, achieving a distributed clock synchronization accuracy of ±10ns, and a 1μs look-ahead buffer is added during motion command compilation. The piezoelectric ceramic actuator is a PhysikInstrumente P-841.10 model, with a voltage control range of ±150V and a displacement resolution of 5pm.

[0075] The Holoeye LCOS-2000 spatial light modulator used in the light field control unit has an 8μm pixel size, a 60Hz refresh rate, and a phase modulation range of 0-2π. Stokes vector monitoring was performed using a Thorlabs PAX5710 polarization analyzer with a 1kHz sampling rate. The microlens array used was an Invenios MLA-150 with a lens spacing of d=300μm and a fill factor η=0.85. For flat-top light field optimization, the genetic algorithm population size was set to 50, with a crossover probability of 0.8, a mutation probability of 0.1, and 200 iterations.

[0076] The diffraction integral operation in the defect suppression unit is accelerated using CUDA, with an NVIDIA V100 GPU and a memory bandwidth of 900GB / s. The convolutional neural network adopts the U-Net architecture, with an input image size of 256×256×1, a convolutional kernel size of 3×3, and 5 layers, with batch normalization layers placed alternately. The training dataset includes 10,000 defect samples, with a positive to negative sample ratio of 1:9. Data augmentation uses a combination of rotation, scaling, and noise injection operations. When adjusting the exposure threshold, the PID controller sets the proportional coefficient Kp=0.6, the integral time Ti=0.5s, and the derivative time Td=0.1s, with the exposure time adjustment range being 10-500ms.

[0077] In the quality verification unit, the optical microscope used is a ZEISS Axio Observer, equipped with 5×, 10×, and 20× objectives, with a resolution of 0.35μm. The ellipsometer is a JAWoollam M-2000, measuring wavelengths from 190-1700nm with an incident angle of 60°. The support vector machine is implemented using LIBSVM, with a radial basis function kernel, a penalty factor C=100, and a kernel parameter γ=0.1. During machine learning model training, five-fold cross-validation was used, achieving an area under the ROC curve (AUC) > 0.95 and an F1-score > 0.85. The data inversion module uses a Spring Boot framework to build a RESTful API, with data transmission encryption using the AES-256-GCM algorithm and an SPI interface communication rate of 16MHz.

[0078] The data benchmark establishment module in the closed-loop control uses an Agilent 34970A data acquisition instrument. Multiphysics calibration is performed in a constant temperature and humidity chamber (23±0.5℃, RH<40%), with a calibration cycle of 24 hours. The spatial positioning optimization module uses a Renishaw XL-80 laser interferometer, with a six-degree-of-freedom measurement accuracy of ±0.5ppm. The spatiotemporal synchronization control module uses the NI CompactDAQ system, with an NI cRIO-9067 FPGA, and a DC synchronization error of <5ns. The effect feedback module uses TensorFlowLite to deploy a reinforcement learning model, with a Q-learning discount factor γ=0.9 and an exploration rate ε decayed to 0.1.

[0079] The composite optical waveguide master plate defect compensation lithography system achieves defect suppression through an integrated multi-module collaborative working mechanism. After system startup, the data acquisition unit first acquires the master plate design parameters and light source characteristic data. The design data extraction module parses the grating period design value from the digital file, while the spectral calibration module simultaneously acquires the laser output spectral signal and parses the center wavelength value. The data fusion module preprocesses the two sets of data, uses least squares filtering to extract the grating period dominant frequency component, and constructs a deviation correction function to output a standardized parameter set.

[0080] The parameter calculation unit receives the preprocessed data and solves the nonlinear equations relating the grating period, wavelength, and Talbot distance using the Newton-Raphson iterative method. The multiphysics compensation module, combined with the FDTD simulation model, calculates the distance offset caused by dispersion effects, generating a multidimensional Talbot distance matrix. The redundancy generation module embeds process tolerance parameters into the matrix and transmits them to the positioning unit via ASCII encoding.

[0081] The positioning unit converts the Talbot distance parameters into XYZ three-axis physical coordinate offsets. The thermal drift compensation module monitors the base position in real time using an interferometer, and the PID controller dynamically adjusts the piezoelectric ceramic driving voltage to achieve sub-nanometer positioning accuracy. The motion command compilation module converts the compensated coordinates into voice coil motor pulse signals, and achieves nanosecond-level synchronization of three-axis motion commands via the EtherCAT bus.

[0082] After receiving positioning data, the motion control unit uses a trajectory prediction module to predict the instantaneous position distribution of the substrate based on a sinusoidal drive signal, and uses Fourier series expansion to filter out high-frequency noise and optimize the motion trajectory. The optical field synchronization module adjusts the phase distribution of the liquid crystal spatial light modulator according to the motion phase, and the polarization matching module maintains the circular polarization state through a liquid crystal phase delay sheet to ensure the consistency of optical field polarization.

[0083] The light field modulation unit acquires light intensity distribution data in real time, a liquid crystal spatial light modulator compensates for low-intensity regions, and a dynamic polarization adjustment module maintains polarization purity through Stokes vector monitoring. The edge gradient control module uses a microlens array to convert the Gaussian beam into a flat-top light field, optimizing the edge light intensity gradient. The defect suppression unit uses FFT to accelerate diffraction integral calculations, a dual averaging algorithm to suppress high-frequency noise, and a deep learning model to extract defect features and calculate the repair rate.

[0084] The quality verification unit uses an optical microscope to photograph the morphology of the repaired area and an ellipsometer to measure the refractive index distribution. Machine learning algorithms analyze the correlation between the repair rate and period deviation and wavelength drift, generating suggestions for process parameter correction. The data inversion module sends correction commands via HTTP protocol and updates the laser temperature control parameters via SPI interface, forming a closed-loop control feedback loop.

[0085] The system eliminates the cumulative impact of single parameter fluctuations on positioning accuracy through deep coupling of data acquisition and parameter calculation. The positioning unit employs a composite algorithm combining homogeneous coordinate transformation and thermal drift compensation, combined with the nanoscale control capability driven by piezoelectric ceramics, to solve the dynamic focal length drift problem caused by existing mechanical scanning compensation. The motion control unit maintains optical field polarization consistency while suppressing motion blur through a dual optimization mechanism of Fourier series expansion and liquid crystal phase compensation. The defect suppression unit innovatively integrates Fresnel diffraction integral calculation and deep learning feature matching to achieve spatial-temporal domain collaborative suppression of high-frequency defects. The quality verification unit establishes a closed-loop feedback mechanism that dynamically optimizes parameter acquisition and the Talbot calculation model through a machine learning model, reducing the system response time to within 200ms and controlling the fluctuation range of mass production yield to within ±1.2%. All functional units form an organic whole through standardized interfaces, enabling the system to exhibit excellent adaptability when dealing with periodic grating defects and irregular microscopic damage.

[0086] Embodiment 1 of the present invention: a scenario for repairing periodic grating defects; In the semiconductor optical waveguide master fabrication process, for a dense grating structure with a period of 320nm, the system first uses optical character recognition technology to analyze the Gerber design file provided by the customer, extracting the grating period design value as 320.05nm ± 0.1nm. Simultaneously, a Yokogawa AQ6370 spectrometer is activated to scan the output spectrum of an argon-ion laser at a resolution of 1nm, resolving the center wavelength to 488.012nm. The data fusion module uses principal component analysis to screen wavelength fluctuation data and constructs an error transfer function using the least squares method, generating a wavelength calibration coefficient of 0.9987. The parameter calculation unit initiates dual-core parallel computation, using Newton's iteration method... Using / (n(λ)Δ) as the initial value, after 17 iterations, it converges to a Timber distance of 3.214 μm, and the residual value is reduced to 8.3 × 10⁻⁶. -7 The FDTD simulation model considers the material dispersion coefficient. =-2.1×10 - ² 0 The Talbot distance correction ΔT = 0.152 μm was calculated using a m² / s ratio. A redundancy generation module embeds normally distributed process tolerance parameters to generate a multi-dimensional Talbot matrix including ±3σ deviations. The positioning unit converts the Talbot distance into XYZ three-axis offsets through homogeneous coordinate transformation, integrates a dual-frequency laser interferometer to monitor the substrate position in real time, and a PID controller dynamically adjusts the piezoelectric ceramic driving voltage with an exponential decay function, controlling the position deviation within 0.8 nm at 25℃. The motion control unit uses B-spline curve fitting to fit the motion trajectory, Fourier series expansion to filter out high-frequency noise above 800 Hz, and generates a voice coil motor pulse sequence. The light field control unit performs pixel-level compensation for low-intensity areas using a liquid crystal spatial light modulator, a dynamic polarization adjustment module maintains 99.6% circular polarization purity, and a microlens array converts the Gaussian beam into a flat-top light field with a beam waist radius of 0.2 mm. The defect suppression unit uses FFT-accelerated diffraction integral calculations, and a dual averaging algorithm achieves a high-frequency noise suppression rate of 32 dB. The quality verification unit measured the refractive index distribution of the repaired area using an ellipsometer and established a correlation model between period deviation and repair rate using a support vector machine. It was found that the repair rate decreased by 12% when the period deviation exceeded ±0.5nm. The closed-loop control system triggered parameter calibration every 120 minutes and updated the laser temperature control parameters through the SPI interface, thereby improving the wavelength stability to ±0.002nm / hour.

[0087] Embodiment 2 of the present invention: a scenario for repairing sudden mechanical scratches; In the fabrication of the optical waveguide master plate for MEMS devices, the system detected a sudden mechanical scratch defect in a certain area. The data acquisition unit monitored the substrate position in real time using an interferometer and found a sudden 50nm shift in the X-axis direction. The parameter calculation unit called the FDTD simulation model and calculated the dispersion effect correction amount by combining the material stress-strain tensor. The positioning unit activated the thermal drift compensation module, integrated the Kalman filtering algorithm with accelerometer data, and dynamically adjusted the piezoelectric ceramic driving voltage with a PID controller, converging the position deviation to 0.3nm within 50ms. The motion control unit used the Runge-Kutta method to predict the instantaneous position of the substrate, filtered out motion noise above 1kHz using Fourier series expansion, and generated a liquid crystal driving signal containing motion phase information. The optical field control unit detected that the polarization purity dropped to 98.2% through Stokes vector monitoring and dynamically adjusted the λ / 4 waveplate voltage to 3.2V to maintain the circular polarization state. The defect suppression unit started a deep learning model, used a convolutional neural network to extract scratch morphology features, and a support vector machine to determine the defect level as III. The PID controller adjusts the exposure threshold by region, extending the exposure time to 320ms for defective areas and maintaining a standard 280ms exposure for non-defective areas. The quality verification unit measures the 3D morphology of the repaired area using a white light interferometer, finding that the scratch depth decreased from 23nm to below 5nm. A machine learning model analyzes the refractive index data, establishing a correlation model between the period deviation rate δ=0.3%, wavelength drift Δλ=0.015nm, and the repair rate, suggesting adjusting the laser wavelength calibration coefficient to 0.9992. The closed-loop control system sends correction commands via a RESTful API, and the SPI interface updates temperature control parameters at a rate of 16MHz, reducing the system response time to 180ms.

Claims

1. A composite optical waveguide master plate defect compensation photolithography system, characterized in that, include: The data acquisition unit is used to acquire master design parameters and light source characteristic data; The parameter calculation unit is used to receive the design parameters and light source characteristic data output by the data acquisition unit, and generate positioning reference parameters by solving a set of nonlinear equations. The positioning unit is used to receive the positioning reference parameters output by the parameter calculation unit, convert the Tiber distance parameters into XYZ three-axis physical coordinate offsets through a preset coordinate transformation algorithm, and output the compensated positioning coordinates. The motion control unit is used to receive the compensated positioning coordinates output by the positioning unit, calculate and output motion trajectory data based on the physical coordinate offset; The light field control unit is used to receive motion trajectory data output by the motion control unit, optimize the light field distribution through light intensity uniformity calibration and dynamic polarization adjustment, and output the optimized light field parameters. The defect suppression unit is used to receive the optimized light field parameters output by the light field control unit, generate a dual average light intensity distribution based on the preset Fresnel diffraction integral calculation, suppress periodic and non-periodic defects, and output defect repair data including regional exposure control parameters. The quality verification unit is used to receive the defect repair data output by the defect suppression unit and generate a repair rate statistical report through optical detection and electrical performance testing. The data acquisition unit, parameter calculation unit, positioning unit, motion control unit, light field modulation unit, defect suppression unit, and quality verification unit are sequentially connected through standardized interfaces to form a closed-loop control system from data acquisition to quality verification.

2. The composite optical waveguide master plate defect compensation photolithography system according to claim 1, characterized in that, The data acquisition unit includes: The design data extraction module is used to read the raster period design value from the digital file of the master design drawing and output the raster period design value data. The spectral calibration module is used to acquire the spectral signal of the laser output light through a spectral analyzer, analyze the center wavelength value, and output wavelength measurement data. The data fusion module receives the grating period design value data output by the design data extraction module and the wavelength measurement data output by the spectral calibration module, and merges the grating period design value data and the wavelength measurement data into an original measurement dataset. It then performs noise filtering on the original measurement dataset using the least squares method and removes outlier data points with a standard deviation exceeding 3σ of the mean of the original measurement dataset, generating filtered data. The filtered data is then processed using Fourier transform to extract the grating period dominant frequency component. This component is compared with the received grating period design value data, and a deviation correction function is constructed based on the comparison result. Finally, based on the deviation correction function and the wavelength measurement data, a standardized period parameter set and wavelength calibration coefficients are generated and output.

3. The composite optical waveguide master plate defect compensation photolithography system according to claim 2, characterized in that, The parameter calculation unit includes: The nonlinear solution module receives the standardized periodic parameter set and wavelength calibration coefficients output by the data acquisition unit, and uses the Newton-Raphson iterative method to solve the nonlinear equations of grating period, wavelength and Talbot distance based on the standardized periodic parameter set and wavelength calibration coefficients, generating and outputting a numerical sequence of Talbot distance. The multiphysics compensation module receives the Talbot distance numerical sequence output by the nonlinear solution module, and calculates the distance offset caused by the grating periodic dispersion effect by combining the preset FDTD simulation model and the material dispersion coefficient; the distance offset is then added to the received Talbot distance numerical sequence to generate and output a multidimensional Talbot distance matrix. The redundancy generation module receives the multi-dimensional Tyber distance matrix output by the multi-physics compensation module, and superimposes a process tolerance random parameter conforming to a preset normal distribution on each element of the multi-dimensional Tyber distance matrix to generate a matrix with embedded tolerance parameters; the matrix with embedded tolerance parameters is ASCII encoded, and the encoded data is transmitted to the positioning unit.

4. The composite optical waveguide master plate defect compensation photolithography system according to claim 3, characterized in that, The positioning unit includes: The coordinate transformation module receives the ASCII encoded data output by the parameter calculation unit, decodes the data, and restores it to a multi-dimensional Talbot distance matrix. It uses a preset homogeneous coordinate transformation algorithm to extract Talbot distance parameters from the multi-dimensional Talbot distance matrix and converts the Talbot distance parameters into XYZ three-axis physical coordinate offsets including translation vectors and rotation matrices, generating and outputting absolute positioning reference data. The thermal drift compensation module receives the absolute positioning reference data output by the coordinate transformation module, and acquires the actual position data of the base on the XYZ axes in real time through an integrated interferometer; it performs differential calculation between the actual position data and the absolute positioning reference data to generate position deviation data; based on the position deviation data, it uses a PID controller with a preset proportional coefficient Kp=2.5, integral coefficient Ki=0.01, and derivative coefficient Kd=0.1 to dynamically calculate the adjustment amount of the piezoelectric ceramic driving voltage with an exponential decay function, and outputs the compensated positioning coordinates corrected based on the adjustment amount. The motion command compilation module receives the compensated positioning coordinates output by the thermal drift compensation module, converts the compensated positioning coordinates into a pulse sequence signal that matches the displacement resolution of the voice coil motor, and uses the EtherCAT bus and its distributed clock synchronization mechanism to time-align the pulse sequence signals corresponding to the XYZ axes, generating and outputting synchronous motion commands with a phase difference within ±10ns.

5. The composite optical waveguide master plate defect compensation photolithography system according to claim 4, characterized in that, The motion control unit includes: The trajectory prediction module receives synchronous motion commands output by the motion command compilation module of the positioning unit, and predicts the instantaneous position of the base in the future motion cycle based on the sinusoidal waveform driving signal included in the synchronous motion command, generating instantaneous position distribution data; performs Fourier series expansion on the instantaneous position distribution data, filters out harmonic components with frequencies higher than the preset cutoff frequency in the expansion result, and generates and outputs optimized motion trajectory parameters including the filtered fundamental phase information; The light field synchronization module receives the optimized motion trajectory parameters output by the trajectory prediction module, extracts the fundamental phase information from the optimized motion trajectory parameters, generates a corresponding phase modulation voltage signal based on the fundamental phase information, and applies the phase modulation voltage signal to the liquid crystal spatial light modulator to change the refractive index of each pixel on the liquid crystal spatial light modulator, thereby adjusting the wavefront phase of the transmitted light field, generating and outputting synchronized light field parameters. The polarization matching module receives the synchronized optical field parameters output by the optical field synchronization module, calculates the deviation between the current polarization state and the target circular polarization state by measuring the Stokes vector of the synchronized optical field parameters, and dynamically adjusts the driving voltage applied to the liquid crystal phase retardation film based on the deviation so that the polarization direction of the optical field transmitted through the liquid crystal phase retardation film is adjusted to the target circular polarization state, and generates and outputs polarization-consistent optical field parameters to the optical field control unit.

6. The composite optical waveguide master plate defect compensation photolithography system according to claim 5, characterized in that, The light field manipulation unit includes: The light intensity uniformity calibration module receives polarization-uniform light field parameters output by the polarization matching module of the motion control unit. It performs a two-dimensional scan of the actual light field characterized by the polarization-uniform light field parameters using a photodetector array to acquire light field intensity distribution data. It compares the intensity value of each pixel in the light field intensity distribution data with a preset light intensity uniformity threshold to identify low-intensity pixel regions with intensities below the threshold. It generates a grayscale compensation mapping table corresponding to the positions of the low-intensity pixel regions and drives a liquid crystal spatial light modulator based on the grayscale compensation mapping table to perform pixel-level light intensity tuning on the low-intensity regions, generating and outputting the light field parameters after light intensity uniformity calibration. The dynamic polarization adjustment module receives the light field parameters after light intensity uniformity calibration output by the light intensity uniformity calibration module, guides the light field through a λ / 4 waveplate, and rotates the polarization direction of the light field by 45 degrees; after the λ / 4 waveplate, the Stokes vector of the output light field is measured in real time by a Stokes vector analyzer; based on the measured Stokes vector, the purity deviation between the current polarization state and the target circular polarization state is calculated; based on the purity deviation, the driving voltage applied to the λ / 4 waveplate is dynamically adjusted through a closed-loop feedback circuit to generate and output polarization-uniform light field parameters; The edge gradient control module receives the polarization-consistent optical field parameters output by the dynamic polarization adjustment module, guides the optical field through the microlens array, and converts the input optical field with a Gaussian distribution into an output optical field with a flat-top distribution. Based on the Fresnel diffraction integral formula, it calculates the intensity gradient distribution of the output optical field in the edge region. It iteratively adjusts the spacing parameters of the lens elements in the microlens array through a gradient descent algorithm so that the calculated edge intensity gradient is lower than a preset gradient threshold, and generates and outputs the edge gradient-controlled optical field parameters to the defect suppression unit.

7. The composite optical waveguide master plate defect compensation photolithography system according to claim 6, characterized in that, The defect suppression unit includes: The diffraction integral calculation module receives the light field parameters after edge gradient control output by the edge gradient control module of the light field modulation unit, and performs Fresnel diffraction integral calculation on the light field wavefront represented by the light field parameters after edge gradient control based on the fast Fourier transform algorithm to generate an initial light intensity distribution map; performs spatial domain sliding window averaging on the spatial pixel array of the initial light intensity distribution map, and performs temporal domain averaging on multiple consecutive frames of the initial light intensity distribution map in the time series to generate and output a predicted light intensity distribution map. The defect identification module receives the predicted light intensity distribution map output by the diffraction integral calculation module, inputs the predicted light intensity distribution map into a pre-trained convolutional neural network model, and the convolutional neural network model extracts defect features from the image and outputs a defect region mask map; based on the defect region mask map, it calculates the pixel area ratio of the defect region as the defect repair rate, and based on a preset qualified pixel intensity range, it calculates the qualified pixel ratio as the yield rate; simultaneously, it extracts the pixel intensity values ​​of the region identified by the defect region mask map from the predicted light intensity distribution map, and generates and outputs defect intensity distribution data; The exposure threshold adjustment module receives defect intensity distribution data output by the defect identification module, calculates the required photoresist exposure energy compensation for each defect region based on the average pixel intensity of each defect region in the defect intensity distribution data, and dynamically calculates and outputs the exposure time adjustment value corresponding to each defect region using a PID controller with a preset proportional coefficient Kp=0.6, integral time Ti=0.5 seconds, and derivative time Td=0.1 seconds. The module integrates the exposure time adjustment values ​​of all regions to generate regional exposure control parameters, and sends these regional exposure control parameters to the light field control unit.

8. The composite optical waveguide master plate defect compensation photolithography system according to claim 7, characterized in that, The quality verification unit includes: The morphology detection module receives the regional exposure control parameters output by the exposure threshold adjustment module of the defect suppression unit. Based on the spatial coordinates of each exposure region defined by the regional exposure control parameters, it drives the stage to move the corresponding exposed region to the field of view of the optical microscope. The optical microscope is used to photograph the surface morphology of the exposed region, generating morphology image data. In the same exposed region, an ellipsometer is used to measure the ellipsoidal parameters of the region at a preset incident angle, and the refractive index distribution of the region is calculated based on the Shermeier equation to generate refractive index distribution data. The morphology image data and the refractive index distribution data are integrated to generate and output morphology and refractive index detection data. The refractive index analysis module receives morphology and refractive index detection data output by the morphology detection module. From the morphology image data, it extracts the contour of the actual defect region based on an image segmentation algorithm and calculates the area and depth of the actual defect region. Based on the area and depth, it calculates the actual defect repair rate. From the refractive index distribution data, it calculates the period deviation and wavelength drift between the actual refractive index value and the designed refractive index value. The actual defect repair rate, the period deviation, and the wavelength drift are used as input features and input to a pre-trained support vector machine model to establish a mathematical correlation model between the actual defect repair rate, the period deviation, and the wavelength drift. Based on the mathematical correlation model, it generates process parameter correction suggestions including material compensation and etching rate adjustment values ​​and outputs the process parameter correction suggestions. The data inversion module receives process parameter correction suggestions output by the refractive index analysis module, parses correction instructions for the standardized periodic parameter set and correction instructions for the laser center wavelength from the process parameter correction suggestions; encapsulates the correction instructions for the standardized periodic parameter set into a JSON format data packet and sends it to the data fusion module of the data acquisition unit via the HTTP protocol; converts the correction instructions for the laser center wavelength into temperature control parameter adjustment values ​​and writes them into the laser's temperature control register via the SPI interface to adjust the laser's output wavelength; after completing the instruction sending and register writing, it generates and outputs a closed-loop control feedback signal.

9. The composite optical waveguide master plate defect compensation photolithography system according to claim 8, characterized in that, Also includes: The data benchmark establishment module receives the closed-loop control feedback signal output by the data inversion module of the quality verification unit. Under a constant temperature of 23±0.5℃ and a relative humidity of less than 40%, it jointly operates the data acquisition unit, parameter calculation unit, positioning unit, motion control unit, light field modulation unit, and defect suppression unit to collect process data throughout the entire process. It then uses Zernike polynomial fitting to fit the optical system aberrations to generate a standardized parameter set including 256 error compensation coefficients. The standardized parameter set is then output as the initial calibration parameters to the data fusion module of the data acquisition unit. The spatial positioning optimization module receives the compensated positioning coordinates output by the thermal drift compensation module of the positioning unit, and measures the displacement data of the base in six degrees of freedom (XYZ axes and rotation around the axis) in real time using a dual-frequency laser interferometer. At the same time, it collects the motion acceleration data of the base using an accelerometer. The Kalman filter algorithm is used to fuse the displacement data and the motion acceleration data to generate higher-precision fused position data. Based on the difference between the fused position data and the compensated positioning coordinates, a nanometer-level position correction is calculated, and the piezoelectric ceramic actuator is driven to adjust the substrate position according to the position correction, thereby generating and outputting positioning accuracy verification data. The spatiotemporal synchronization control module receives the optimized motion trajectory parameters output by the trajectory prediction module of the motion control unit, extracts the fundamental phase information from the optimized motion trajectory parameters, generates a hardware clock signal that is strictly synchronized with the fundamental phase information using a field-programmable gate array (FPGA), and re-aligns and outputs XYZ three-axis motion commands based on the hardware clock signal through the distributed clock synchronization mechanism of the EtherCAT bus, so that the phase difference between the three-axis commands is controlled within ±5 nanoseconds. At the same time, the hardware clock signal is sent as a trigger signal to the light intensity uniformity calibration module of the light field control unit to synchronize the light field acquisition timing, generate and output spatiotemporal synchronization optimization parameters. The effect feedback module receives process parameter correction suggestions output by the refractive index analysis module of the quality verification unit, and parses the material compensation amount and etching rate adjustment value from the process parameter correction suggestions; using the material compensation amount and etching rate adjustment value as environmental rewards, the Q-learning reinforcement learning algorithm is used to dynamically and iteratively optimize the weight coefficients of the noise filtering function used in the data fusion module of the data acquisition unit, and the convergence judgment threshold of the Newton-Raphson iteration method in the nonlinear solution module of the parameter calculation unit; The optimized weight coefficients and convergence threshold are encapsulated as correction coefficients Kp, Kd, ​​and Ki, and fed back to the nonlinear solution module of the parameter calculation unit.

10. A composite optical waveguide master plate defect compensation photolithography method, applied to the composite optical waveguide master plate defect compensation photolithography system as described in any one of claims 1 to 9, characterized in that, include: Step 1: Obtain master design parameters and light source characteristic data; Step 2: Receive the design parameters and light source characteristic data, and generate positioning reference parameters by solving a system of nonlinear equations; Step 3: Receive the positioning reference parameters, convert the Tiber distance parameters into XYZ three-axis physical coordinate offsets using a preset coordinate transformation algorithm, and generate compensated positioning coordinates; Step 4: Calculate and generate motion trajectory data based on the offset between the compensated positioning coordinates and the physical coordinates; Step 5: Receive the motion trajectory data, optimize the light field distribution through light intensity uniformity calibration and dynamic polarization adjustment, and generate optimized light field parameters; Step 6: Receive the optimized light field parameters, generate a dual average light intensity distribution based on Fresnel diffraction integral calculation, suppress periodic and non-periodic defects, and generate defect repair data including regional exposure control parameters. Step 7: Receive the defect repair data and generate a repair rate statistical report through optical detection and electrical performance testing.

Images