A lithography machine error correction method and system

Abstract

This invention relates to the field of semiconductor lithography error correction technology, specifically a lithography machine error correction method and system. The method includes: deploying an interferometric sampling array to capture the initial wavefront phase distribution to obtain a reference wavefront field; using a thermal drift separation operator to remove the slowly varying drift component caused by environmental thermal effects; and using multidimensional convolution matching between the residual wavefront field and a nominal lens aberration feature library to identify static aberration patterns, generating an inverse aberration compensation driving signal. A time-varying trend predictor based on a Kalman filter framework is used to continuously estimate future thermal drift, which is then superimposed with the inverse aberration compensation driving signal to synthesize a composite correction instruction set. Based on this instruction set, the workpiece stage positioning system and the projection lens focusing system are jointly kinematically adjusted to offset the overlay error jointly introduced by aberrations and thermal deformation. This method can separate the thermal drift component, improve aberration recognition accuracy, and achieve feedforward thermal drift correction and dual-system coordinated adjustment.

Description

Technical Field

[0001] This invention relates to the field of semiconductor lithography error correction technology, and in particular to a lithography machine error correction method and system. Background Technology

[0002] Conventional lithography error correction technology acquires object aberration data through a wavefront detection device, compares the aberration data with preset aberration features, and generates a compensation drive signal. This correction only applies to the static aberrations of the projection lens. Thermal drift errors caused by environmental thermal effects are handled using fixed compensation parameters or post-event feedback correction. No dedicated separation mechanism is built for the thermal drift components in the wavefront field. Time-varying thermal drift correction relies on hysteresis feedback adjustment logic, and the correction command only acts on the focusing actuator of the projection lens.

[0003] Existing correction techniques fail to remove the slowly varying trend term introduced by thermal effects during wavefront data processing. The thermal drift component and static aberration component overlap, leading to a decrease in the accuracy of multidimensional convolution matching of the nominal lens aberration feature library and an inability to accurately identify the true static aberration pattern of the projection lens. Furthermore, the estimation of thermal drift does not employ the rolling estimation mode of the Kalman filter framework, resulting in a temporal deviation between the estimated result and the actual thermal drift. The correction mode using a single actuator cannot adapt to the coupled error pattern of static aberration and thermal deformation, limiting the effectiveness of overlay error suppression.

[0004] This invention aims to remove the slowly varying drift component caused by environmental thermal effects in the wavefront field, thus avoiding interference from thermal drift on static aberration identification. It requires rolling estimation of the thermal drift amount at future moments based on a Kalman filter framework. The inverse aberration compensation drive signal is then superimposed with the estimated thermal drift value to form a composite correction command. This composite correction command enables coordinated adjustment of the workpiece stage positioning system and the projection lens focusing system. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for error correction based on a lithography machine.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for error correction based on a lithography machine, comprising: An interferometric sampling array is deployed to capture the initial wavefront phase distribution in order to obtain a continuous and smooth reference wavefront field; The reference wavefront field is input to the thermal drift separation operator, which identifies the trend term in the wavefront field that changes linearly with temperature and then extracts the slowly varying drift component caused by the environmental thermal effect. The residual wavefront field after the slowly varying drift components are stripped is then subjected to multidimensional convolution matching with a pre-stored nominal lens aberration feature library to identify the static aberration patterns generated by the lithography machine projection lens. Based on the identified static aberration pattern, an inverse aberration compensation drive signal for the objective lens actuator is generated in the control system. While generating the inverse aberration compensation drive signal, the slow drift component is input to the time-varying trend predictor, which performs rolling estimation of the thermal drift amount at future times based on the Kalman filter framework. The inverse aberration compensation drive signal is superimposed with the estimated thermal drift value at the future time to synthesize a composite correction instruction set; Based on the composite correction instruction set, the workpiece stage positioning system and the projection lens focusing system of the lithography machine are jointly kinematically adjusted to counteract the overlay error caused by aberration and thermal deformation.

[0007] As a further aspect of the present invention, the deployment of the interferometric sampling array to capture the initial wavefront phase distribution to obtain a continuous and smooth reference wavefront field includes: A high-density distributed interferometric sampling array is deployed within the exposure field of the lithography machine to capture the initial wavefront phase distribution of the wafer plane; A full-field unwrapping operation is performed on the initial wavefront phase distribution to remove measurement noise caused by phase folding, resulting in a continuous and smooth reference wavefront field. Specifically, this includes: The gradient amplitude of the initial wavefront phase distribution is calculated to determine the location of potential singularities where the phase changes abruptly in the wavefront field; Using the potential singularity location as the boundary, the entire exposure field is divided into several non-overlapping local sub-regions; Within each of the local sub-regions, a smooth phase gradient curve is recovered using a least-squares fitting algorithm; By integrating and accumulating the phase value of each pixel along the phase gradient curve, the absolute phase value of the local sub-region is reconstructed. The absolute phase values ​​of all the local sub-regions are seamlessly stitched together, and the discontinuities at the stitching points are smoothed out to finally output the reference wavefront field.

[0008] As a further aspect of the present invention, the reference wavefront field is input to a thermal drift separation operator. The thermal drift separation operator identifies the trend term in the wavefront field that changes linearly with temperature, and then extracts the slowly varying drift component caused by environmental thermal effects, including: Several key optical paths that run through the entire exposure field are extracted from the reference wavefront field as characteristic optical paths, and the characteristic optical paths correspond to specific rows or columns in the interference sampling array; Acquire multi-point temperature sensing data corresponding to the characteristic optical path, and the temperature sensing data is strictly synchronized with the acquisition time of the reference wavefront field; A multiple linear regression model is constructed, wherein the phase value of the characteristic optical path is used as the dependent variable and the multi-point temperature sensing data is used as the independent variable. Solve for the regression coefficients of the multiple linear regression model, whereby the regression coefficients characterize the wavefront distortion caused by a unit temperature change; Using the obtained regression coefficients, an inverse scaling operation is performed on the reference wavefront field to subtract the phase values ​​affected by temperature from the reference wavefront field, thereby obtaining the slowly varying drift component.

[0009] As a further aspect of the present invention, the residual wavefront field after the slowly varying drift component has been stripped is subjected to multidimensional convolution matching with a pre-stored nominal lens aberration feature library to identify the static aberration patterns generated by the lithography machine projection lens, including: A two-dimensional Fourier transform is performed on the residual wavefront field to convert it from the spatial domain to the frequency domain, resulting in a spectral distribution map. The spectrum distribution map is matched with the feature frequency vector of each preset aberration type in the nominal lens aberration feature library through multidimensional convolution, and the correlation coefficient is calculated. Select a predefined aberration type whose correlation coefficient exceeds a preset threshold value, and define it as a candidate aberration set; All aberration patterns in the candidate aberration set are weighted and superimposed, with the weights determined by their respective correlation coefficients, to reconstruct an approximate aberration residual field, which is the identified static aberration pattern.

[0010] As a further aspect of the present invention, based on the identified static aberration pattern, an inverse aberration compensation drive signal for the objective lens actuator is generated in the control system, including: The static aberration pattern is mapped to a virtual Zernike polynomial coefficient space to obtain a set of Zernike coefficients characterizing the aberration magnitude. The sensitivity matrix of the objective lens actuator is retrieved, and the sensitivity matrix describes the mapping relationship between the actuator displacement and the change in the Zernike coefficient; By using matrix inversion, a set of actuator displacement adjustments required to offset the Zernike coefficients is calculated; The actuator displacement adjustment is converted into a control voltage signal, and the control voltage signal is subjected to amplitude limiting and smoothing filtering to form the inverse aberration compensation drive signal.

[0011] As a further aspect of the present invention, the slowly varying drift component is input to a time-varying trend predictor, which performs a rolling estimate of the thermal drift amount at future times based on a Kalman filter framework, including: Establish the state-space equation of the slowly varying drift component, taking the wavefront distortion at the current moment as the system state and the fluctuation of ambient temperature as the system noise; Initialize the state estimate and covariance matrix of the Kalman filter, wherein the covariance matrix reflects the uncertainty of the initial state estimate; In each new measurement cycle, the latest slowly drifting component data is received and input as an observation into the Kalman filter; The prediction step of the Kalman filter is executed, and the prior state at the current time is extrapolated based on the state estimate at the previous time step. Next, the Kalman filter update step is executed, and the prior state is corrected using the observation value to obtain the posterior state estimate, which is the thermal drift estimate for the future time.

[0012] As a further aspect of the present invention, the inverse aberration compensation driving signal is superimposed with the estimated thermal drift value at the future time to synthesize a composite correction instruction set, including: The inverse aberration compensation drive signal is decomposed into a defocus term acting on the objective lens axis and an astigmatic term acting on the objective lens radial direction. The estimated thermal drift at the future moment is also decomposed into axial thermal expansion displacement and radial thermal deformation displacement. The total axial displacement correction is obtained by algebraically adding the defocus term in the axial direction to the thermal expansion displacement in the axial direction. The radial astigmatism term and the radial thermal deformation displacement are vector-synthesized to obtain the total radial deformation correction amount; The adjusted total axial displacement correction, the total radial deformation correction, and the higher-order aberration terms not involved in the superposition are packaged together into the composite correction instruction set.

[0013] As a further aspect of the present invention, based on the composite correction instruction set, joint kinematic adjustments are made to the workpiece stage positioning system and the projection lens focusing system of the lithography machine, including: The composite correction instruction set is analyzed to separate the focal length adjustment instruction that needs to be executed by the projection lens focusing system and the plane tilt compensation instruction that needs to be executed by the workpiece stage positioning system; The focus adjustment command is converted into the number of rotations of the stepper motor of the projection lens focusing system; The planar tilt compensation command is converted into the displacement increment of the six-degree-of-freedom drive axes in the workpiece stage positioning system; According to the preset execution priority, the position of the workpiece stage positioning system is first locked, then the projection lens focusing system is driven to complete the coarse adjustment of the focal length, and finally the workpiece stage positioning system is finely moved to eliminate the remaining tilt error.

[0014] As a further aspect of the present invention, after generating the inverse aberration compensation drive signal, a closed-loop fine-tuning step based on process feedback is also included: After completing an adjustment based on the composite correction instruction set, immediately initiate a short-exposure marking pattern printing; The marking pattern is scanned using an online optical inspection subsystem to extract the actual edge placement error value; The actual edge placement error value is compared with the theoretical design value to calculate the current residual error vector; The residual error vector is fed back to the thermal drift separation operator and the multidimensional convolution matching stage as a correction bias for the next iteration calculation, so as to correct the deviation of the nominal lens aberration feature library under the current working conditions. The step of using an online optical inspection subsystem to scan the marked pattern and extract the actual edge placement error value includes: High-resolution bright-field and dark-field images of the marked pattern are acquired. The bright-field images are used to identify the pattern outline, and the dark-field images are used to identify subsurface defects. Perform subpixel-level edge detection on the bright-field image to determine the actual pixel coordinates of the pattern edges; The design layout of the marking pattern is invoked to obtain the theoretical edge coordinates of the marking pattern; The actual pixel coordinates of the same edge are registered with the theoretical edge coordinates to eliminate projection deviation caused by the image acquisition angle. Calculate the minimum root mean square distance between the two sets of coordinates after registration, use the minimum root mean square distance as the edge placement error value, and record its direction information.

[0015] As a further aspect of the present invention, the present invention also includes a lithography machine error correction system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the lithography machine error correction method described above.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The thermal drift separation operator identifies the linearly temperature-dependent trend term in the wavefront field, extracting the slowly varying drift component caused by environmental thermal effects from the reference wavefront field. The residual wavefront field after the extraction operation is free of thermal drift-related trend noise. When the residual wavefront field performs multidimensional convolution matching with a pre-stored nominal lens aberration feature library, the matching process only calculates the true aberration features of the projection lens. The identification result of the static aberration pattern is not affected by the superposition of thermal drift components. The identified static aberration pattern matches the actual aberration state of the projection lens. The inverse aberration compensation drive signal generated based on this static aberration pattern can accurately correspond to the compensation requirements of the lens assembly actuator. The output parameters of the drive signal are adapted to the aberration compensation dimension of the lens.

[0017] A time-varying trend predictor based on a Kalman filter framework performs rolling estimations of thermal drift at future time points, using slowly varying drift components as input data. The estimated thermal drift output is synchronized with the actual trend of thermal drift change. The inverse aberration compensation drive signal is superimposed with the estimated thermal drift value at future time points to form a composite correction instruction set. This composite correction instruction set synchronously acts on the workpiece stage positioning system and the projection lens focusing system of the lithography machine. The kinematic adjustment actions of the two systems are executed collaboratively according to the parameters of the instruction set. Errors introduced by static aberrations and overlay errors introduced by thermal deformation are synchronously canceled during the adjustment process. The execution timing of the correction action matches the generation timing of the errors, and the correction effect covers all sources of error from static aberrations and time-varying thermal drift. The overlay error cancellation range encompasses the coupled influence of object aberrations and thermal deformation. Attached Figure Description

[0018] Figure 1 This is a flowchart of a photolithography machine error correction method according to the present invention; Figure 2 A flowchart for removing slowly varying drift components; Figure 3 A Kalman filter prediction analysis diagram of thermal drift in a lithography machine; Figure 4 Timing analysis diagram for kinematic adjustment of lithography machine; Figure 5 This is a convergence analysis diagram for closed-loop correction of edge placement error in a lithography machine. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 An interferometric sampling array is deployed within the exposure field of the lithography machine to capture the initial wavefront phase distribution on the wafer plane. This initial wavefront phase distribution is input to a processing system to obtain a continuous and smooth reference wavefront field. The reference wavefront field is then input to a thermal drift separation operator, which identifies the trend term in the wavefront field that varies linearly with temperature, thereby removing the slowly varying drift component caused by environmental thermal effects. After removing the slowly varying drift component, a residual wavefront field is obtained. This residual wavefront field is then multidimensionally convolved and matched with a pre-existing nominal lens aberration feature library in the system to identify the static aberration pattern generated by the lithography machine's projection lens. The control system generates an inverse aberration compensation drive signal for the lens assembly actuators based on the identified static aberration pattern. Simultaneously, the slowly varying drift component output from the thermal drift separation operator is input to a time-varying trend predictor, which performs a rolling estimate of the thermal drift amount at future times based on a Kalman filter framework. The generated inverse aberration compensation drive signal is superimposed with the estimated thermal drift at future moments to form a composite correction instruction set. Finally, based on this composite correction instruction set, the workpiece stage positioning system and the projection lens focusing system of the lithography machine are jointly kinematically adjusted to counteract the overlay error introduced by both aberration and thermal deformation.

[0022] In one embodiment of the present invention, a high-density distributed interferometric sampling array is deployed within the exposure field of a lithography machine to capture the initial wavefront phase distribution of the wafer plane. A full-field unwrapping operation is performed on the captured initial wavefront phase distribution to eliminate measurement noise caused by phase folding, resulting in a continuous and smooth reference wavefront field. This unwrapping operation includes the following steps: calculating the gradient magnitude of the initial wavefront phase distribution to determine the potential singularity locations where phase abruptly occurs in the wavefront field. Using the identified potential singularity locations as boundaries, the entire exposure field is divided into several non-overlapping local sub-regions. Within each local sub-region, a smooth phase gradient curve is reconstructed using a least-squares fitting algorithm. The phase value of each pixel is integrated and accumulated along the reconstructed phase gradient curve to reconstruct the absolute phase value of the local sub-region. After reconstructing all local sub-regions, the absolute phase values ​​of all local sub-regions are seamlessly stitched together, and any discontinuities at the stitching points are smoothed to finally output the required reference wavefront field.

[0023] In specific implementation, a high-density distributed interferometric sampling array is deployed within the exposure field of the lithography machine to capture the initial wavefront phase distribution of the wafer plane. The initial wavefront phase distribution is a two-dimensional matrix containing phase folding noise. A full-field unwrapping operation is performed on the initial wavefront phase distribution to remove measurement noise. The full-field unwrapping operation includes calculating the gradient magnitude of the initial wavefront phase distribution. The gradient magnitude calculation is based on the phase difference between adjacent pixels, and the potential singularity locations in the wavefront field where phase abrupt changes are determined through calculation. The entire exposure field is divided into several non-overlapping local sub-regions with the potential singularity locations as boundaries. Each local sub-region does not contain phase singularities. In some embodiments, the shape of the local sub-regions is rectangular or polygonal, and the division is based on the connected domain boundaries formed by the potential singularity locations. Within each local sub-region, a smooth phase gradient curve is recovered using a least-squares fitting algorithm. The least-squares fitting algorithm achieves this by minimizing the sum of squared errors between the gradient measurement value and the fitted curve. Its objective function is expressed as: , in: This represents the phase gradient measurement value at the i-th sampling point within a local sub-region. This represents the parameter coordinates of the sampling point along the integration path. and These are the fitting coefficients. This represents the total number of sampling points within a local sub-region. The phase value of each pixel is integrated and accumulated along the recovered phase gradient curve. The integration starts from a selected reference point within the local sub-region, and the absolute phase value of the local sub-region is reconstructed by multiplying the accumulated phase gradient value by the step size. It can be understood that the integration path covers all pixels within the local sub-region, and the integration process maintains path continuity. After reconstructing the absolute phase values ​​of all local sub-regions, the absolute phase values ​​of all local sub-regions are seamlessly stitched together. Seamless stitching is achieved by matching the phase values ​​on the boundaries of adjacent local sub-regions, and discontinuities at the stitching points are smoothed using a weighted average algorithm to fuse the phase values ​​on both sides of the boundary. In specific implementations, the weight coefficient of the smoothing process is inversely proportional to the distance from the pixel to the boundary. Optionally, the weighted average algorithm can use linear interpolation or bilinear interpolation methods. Finally, a continuous and smooth reference wavefront is output, which is used for subsequent thermal drift separation and aberration recognition steps. In some embodiments, the capture frequency of the initial wavefront phase distribution is synchronized with the lithography machine exposure cycle to ensure the timeliness of the reference wavefront. It is understandable that the execution of the full-field unwrapping operation relies on dedicated image processing hardware, which parallelizes gradient calculation, region partitioning, and curve fitting.

[0024] In one embodiment of the present invention, see [reference] Figure 2 Several key optical paths traversing the entire exposure field are extracted from the reference wavefront field as characteristic optical paths. These characteristic optical paths correspond to specific rows or columns in the interferometric sampling array. Multi-point temperature sensing data corresponding to these characteristic optical paths are acquired, with the acquisition time of this temperature sensing data strictly synchronized with the acquisition time of the reference wavefront field. A multiple linear regression model is constructed, where the phase value of the characteristic optical paths is the dependent variable, and the multi-point temperature sensing data is the independent variable. The regression coefficients of this multiple linear regression model are solved, and these regression coefficients characterize the wavefront distortion caused by a unit temperature change. Using the obtained regression coefficients, an inverse scaling operation is performed on the reference wavefront field to subtract the phase values ​​affected by temperature from the reference wavefront field, thereby obtaining the slowly varying drift component caused by the environmental thermal effect. After removing the slowly varying drift component, the residual wavefront field is obtained. A two-dimensional Fourier transform is performed on the residual wavefront field to transform it from the spatial domain to the frequency domain, obtaining the spectral distribution map. Multidimensional convolutional matching is achieved through a sliding window convolution operation. It performs a two-dimensional convolution between the spectral distribution map and the feature frequency vector of each preset aberration type in the nominal lens aberration feature library. The convolution output is a correlation coefficient matrix. The convolution operation formula is: , in, Indicates the location The convolution result at the point, For the spectrum distribution plot on coordinates The amplitude at that point, For the first Feature frequency vectors of a preset aberration type in coordinates The amplitude at that point, and These represent the number of rows and columns of the spectral distribution map, respectively. All preset aberration types in the nominal lens aberration feature library are traversed to obtain the correlation coefficient matrix corresponding to each type. The maximum value in the matrix is ​​selected as the correlation coefficient between that type and the residual wavefront field. A correlation coefficient threshold of 0.7 is set, and preset aberration types with correlation coefficients exceeding the threshold are selected to form a candidate aberration set. Each aberration mode in the candidate aberration set is weighted and superimposed, with the weight being the correlation coefficient corresponding to that aberration mode. The superposition formula is: , Where A(x,y) is the reconstructed aberration residual field. w represents the total number of aberration types in the candidate aberration set. t The correlation coefficient weight for the t-th aberration type is given by [the weight of the correlation coefficient]. For the first The feature frequency vectors of aberration types in spatial coordinates The amplitude at that point. The reconstructed aberration residual field is the identified static aberration pattern.

[0025] In practical implementation, several key optical paths traversing the entire exposure field are extracted from the continuous and smooth reference wavefront field as characteristic optical paths. These characteristic optical paths correspond to specific rows or columns in the interferometric sampling array. Multi-point temperature sensing data corresponding to the characteristic optical paths are acquired. The temperature sensing data is strictly synchronized with the acquisition time of the reference wavefront field and is provided by a sensor array distributed around the lithography machine cavity and projection lens. A multiple linear regression model is constructed, using the phase value of the characteristic optical paths as the dependent variable and the multi-point temperature sensing data as the independent variable. The regression coefficients of the multiple linear regression model are solved; these coefficients characterize the wavefront distortion caused by a unit temperature change. In practical implementation, the expression for the multiple linear regression model can be expressed as: , in: Indicates the first The phase value of each characteristic optical path, Indicates the relationship with the first The first feature optical path synchronously acquired Temperature sensor data, This is the total number of temperature sensing points. It is a constant term. It is the first The regression coefficients corresponding to each temperature variable It is the residual term. Solving for the regression coefficients of a multiple linear regression model. This is achieved using the least squares method. The obtained regression coefficients are then used. A reverse-phase scaling operation is performed on the reference wavefront field. This operation involves subtracting the portion predicted by temperature changes from the phase value of each pixel in the reference wavefront field, thus obtaining the slowly varying drift component caused by environmental thermal effects. After stripping the slowly varying drift component, the residual wavefront field is obtained. In some embodiments, the selection of the feature optical path covers the center, edge, and diagonal directions of the exposure field. A two-dimensional Fourier transform is performed on the residual wavefront field, which transforms the residual wavefront field from the spatial domain to the frequency domain, resulting in a spectral distribution map. A pre-stored nominal lens aberration feature library is traversed. This library stores multiple preset aberration types, and the feature frequency vector and phase modulation vector corresponding to each preset aberration type are extracted. The correlation coefficient between the spectral distribution map and the feature frequency vector of each preset aberration type is calculated. The correlation coefficient is used to quantify the degree of matching between the spectral distribution map and the preset aberration features in the frequency domain. It can be understood that the correlation coefficient can be calculated using the Pearson correlation coefficient or cross-correlation algorithm. Preset aberration types with correlation coefficients exceeding a preset threshold are selected and defined as the candidate aberration set. Optionally, the preset threshold value is set based on historical calibration data. All aberration patterns in the candidate aberration set are weighted and superimposed, with the weights determined by their respective correlation coefficients, to reconstruct an approximate aberration residual field, which is the identified static aberration pattern. In some embodiments, the weighted superposition process is linear superposition, and the amplitude of each aberration pattern is determined by normalizing the inner product of its corresponding characteristic frequency vector and the spectral distribution map. It can be understood that the nominal lens aberration feature library is measured and established using an interferometer before the lithography machine leaves the factory.

[0026] In one embodiment of the invention, the identified static aberration pattern is mapped to a virtual Zernike polynomial coefficient space to obtain a set of Zernike coefficients characterizing the magnitude of the aberration. The sensitivity matrix of the objective lens actuator is retrieved, which describes the mapping relationship between actuator displacement and Zernike coefficient changes. Using matrix inversion, a set of actuator displacement adjustments required to compensate for the Zernike coefficients is calculated. This actuator displacement adjustment is converted into a control voltage signal, and the control voltage signal is subjected to amplitude limiting and smoothing filtering to form an inverse aberration compensation drive signal. Simultaneously with generating the inverse aberration compensation drive signal, the slowly varying drift component is input to a time-varying trend predictor. The state-space equation for the slowly varying drift component is established, using the current wavefront distortion as the system state and ambient temperature fluctuations as system noise. The state estimate and covariance matrix of the Kalman filter are initialized; this covariance matrix reflects the uncertainty of the initial state estimate. At each new measurement cycle, the latest slowly varying drift component data is received and input as an observation into the Kalman filter. The Kalman filter performs a prediction step, extrapolating the prior state at the current time based on the state estimate from the previous time step. Immediately following, the Kalman filter performs an update step, correcting the prior state using the observations to obtain a posterior state estimate. This posterior state estimate is the estimate of the heat drift at future time steps.

[0027] In practice, the identified static aberration patterns are mapped to a virtual Zernike polynomial coefficient space. This mapping is achieved by solving the inner product of the static aberration pattern wavefront and a set of orthogonal Zernike polynomial basis functions, yielding a set of Zernike coefficients characterizing the magnitude of the aberrations. The sensitivity matrix of the objective lens actuator is retrieved. This matrix describes the mapping relationship between actuator displacement and Zernike coefficient changes. The sensitivity matrix of the objective lens actuator is obtained through interferometric calibration after assembly on the lithography machine. A set of actuator displacement adjustments required to counteract the Zernike coefficients is calculated using matrix inversion. Matrix inversion is performed on the sensitivity matrix of the objective lens actuator; a pseudo-inversion is used when the sensitivity matrix is ​​not a square matrix. The actuator displacement adjustments are converted into control voltage signals. The conversion relationship between the control voltage signal and the actuator displacement adjustments is determined by the actuator's voltage-displacement response curve. The control voltage signal undergoes amplitude limiting and smoothing filtering. Amplitude limiting ensures the control voltage signal does not exceed the safe operating range of the actuator drive circuit, while smoothing filtering uses a low-pass digital filter to eliminate high-frequency noise, ultimately forming the inverse aberration compensation drive signal. In some embodiments, the upper and lower limits of the amplitude limiting are set according to the maximum stroke and linear operating range of the objective lens actuator. Simultaneously with generating the inverse aberration compensation drive signal, a slowly varying drift component is input to a time-varying trend predictor. The slowly varying drift component is a time-varying wavefront distortion sequence. A state-space equation for the slowly varying drift component is established, using the current wavefront distortion as the system state and ambient temperature fluctuations as system noise. The state estimate and covariance matrix of the Kalman filter are initialized. The covariance matrix reflects the uncertainty of the initial state estimate, which can be set as a zero vector. In each new measurement cycle, the latest slowly varying drift component data is received and input as an observation into the Kalman filter. The prediction step of the Kalman filter is executed, extrapolating the prior state at the current moment based on the state estimate from the previous moment. The state extrapolation equation for the prediction step of the Kalman filter can be expressed as: , in: Indicates in Time based Prior state estimation based on time information. yes The state transition matrix at time step 1 describes the dynamic change of the system state from the previous time step to the current time step. yes The posterior state estimate at time step [1] is then obtained. Following this, the Kalman filter update step is executed, using the observed values ​​to correct the prior state and obtain the posterior state estimate. The Kalman filter update step includes calculating the Kalman gain, updating the state estimate, and updating the estimated covariance. This posterior state estimate is the estimated thermal drift for future time steps and is used to superimpose it with the inverse aberration compensation drive signal. The state transition matrix can be understood as... The Kalman filter is constructed based on a simplified model of the thermal drift physics process. Optionally, the covariance matrices of the system noise and observation noise need to be set during Kalman filter initialization. In some embodiments, the iteration frequency of the Kalman filter is consistent with the frequency at which the interferometric sampling array captures wavefront phase data. It is understood that the Kalman filter is implemented in an embedded real-time system to ensure the timeliness of the predictions.

[0028] See Figure 3 This is a Kalman filter prediction analysis chart of the thermal drift of a lithography machine, showing a time-series comparison between the actual measured values ​​and the Kalman filter prediction values. Its core purpose is to evaluate the accuracy and reliability of the thermal drift prediction algorithm. The actual value shows a monotonically increasing trend, gradually increasing from approximately 0 nm to about 12 nm. The actual value exhibits significant high-frequency fluctuations, while the predicted value smooths out these disturbances, demonstrating the noise reduction and trend-following capabilities of the Kalman filter. The two curves largely overlap, with only minor deviations at local fluctuations, indicating high prediction accuracy. The small local deviations suggest that the filter's system noise and observation noise covariance parameters are reasonably set, and the prediction uncertainty is controllable. The predicted value accurately follows the slow upward trend of the actual value, proving that the state-space model's simplification of the thermal drift physical process is effective.

[0029] In one embodiment of the present invention, the inverse aberration compensation drive signal is decomposed into a defocus term acting on the objective lens axis and an astigmatic term acting on the objective lens radially. Similarly, the estimated thermal drift at future time moments is decomposed into axial thermal expansion displacement and radial thermal deformation displacement. The axial defocus term and the axial thermal expansion displacement are algebraically added to obtain the total axial displacement correction. The radial astigmatic term and the radial thermal deformation displacement are vector-synthesized to obtain the total radial deformation correction. The total radial deformation correction needs to be decomposed into components in the radial X and Y directions, as shown in the following formula: , Among them, R x R is the radial X-direction deformation correction amount. y R is the radial deformation correction amount in the Y direction. total This represents the total radial deformation correction amount. The direction angle of radial deformation is determined by the vector angle between the astigmatism term and the thermal deformation displacement. When packaging the composite correction instruction set, R... x With R yThese are incorporated into the instruction set as radial X and Y direction deformation corrections, respectively, and together with axial displacement corrections and higher-order aberration corrections, they form a structured data array.

[0030] The adjusted total axial displacement correction, total radial deformation correction, and un-superimposed higher-order aberration terms are packaged together into a composite correction instruction set. The higher-order aberration terms are those with an order greater than or equal to 7 in the Zernike polynomials, corresponding to higher-order aberration modes such as coma Z7, Z8, and spherical aberration Z9. The mapping relationship between Zernike coefficients and actuator displacement adjustment is achieved through the sensitivity matrix of the objective lens actuator; the specific conversion formula is as follows: , in, This represents the actuator displacement adjustment vector corresponding to higher-order aberrations. This is a higher-order aberration sensitivity submatrix, describing the mapping relationship between the changes in higher-order Zernike coefficients and actuator displacement. The vector of higher-order Zernike coefficients is identified. The correction commands corresponding to higher-order aberrations are obtained by converting the actuator displacement adjustment amount into a control voltage signal, which is then incorporated into the composite correction command set after amplitude limiting and smoothing filtering.

[0031] Based on the composite correction instruction set, joint kinematic adjustments are performed on the workpiece stage positioning system and the projection lens focusing system of the lithography machine. The composite correction instruction set is analyzed to separate the focus adjustment command to be executed by the projection lens focusing system and the plane tilt compensation command to be executed by the workpiece stage positioning system. The focus adjustment command is converted into the number of rotations of the stepper motor in the projection lens focusing system. The plane tilt compensation command is converted into the displacement increments of the six-degree-of-freedom drive axes in the workpiece stage positioning system. According to the preset execution priority, the position of the workpiece stage positioning system is first locked, then the projection lens focusing system is driven to complete the coarse focus adjustment, and finally the workpiece stage positioning system is finely adjusted to eliminate any remaining tilt errors.

[0032] In practice, the inverse aberration compensation drive signal is decomposed into a defocus term acting on the objective lens axis and an astigmatic term acting on the objective lens radially. The decomposition process is based on the contribution ratio of the drive signal to the coefficients of the defocus and astigmatic terms in the Zernike polynomial aberration mode. Similarly, the estimated thermal drift at future time points is decomposed into axial thermal expansion displacement and radial thermal deformation displacement. This decomposition is based on the fitted components of the estimated thermal drift in the spatial wavefront distribution corresponding to axial translation and radial stretching. In practice, the coefficients of the defocus and astigmatic terms can be directly extracted from the Zernike coefficient vector mapped back from the inverse aberration compensation drive signal. The axial defocus term and the axial thermal expansion displacement are algebraically added together, meaning the values ​​of the two terms are directly summed to obtain the total axial displacement correction. The radial astigmatic term and the radial thermal deformation displacement are vector synthesized, taking into account the vector attributes of the astigmatic term and thermal deformation displacement in both direction and magnitude. The synthesized result is the total radial deformation correction. In some embodiments, vector synthesis is achieved by adding the astigmatic term and thermal deformation displacement, which are decomposed into components in the X and Y directions, respectively. It can be understood that the formula for vector synthesis can be expressed as an addition operation of two two-dimensional vectors, such as the total radial deformation correction. The calculation method is as follows: , in: This represents the deformation vector corresponding to the radial astigmatism term decomposed from the inverse aberration compensation drive signal. This represents the deformation vector corresponding to the radial thermal deformation displacement decomposed from the estimated thermal drift at future time points. The adjusted total axial displacement correction, the total radial deformation correction, and the un-superimposed higher-order aberration terms are packaged together into a composite correction instruction set, which is typically a structured data array. Optionally, the un-superimposed higher-order aberration terms include correction instructions corresponding to higher-order Zernike coefficients such as coma and spherical aberration. Refer to Table 1 for an example of the contents of a composite correction instruction set. In some embodiments, the composite correction instruction set employs a specific encoding format to ensure accuracy in transmission and resolution.

[0033] Table 1: Contents of the Composite Correction Instruction Set Package

[0034] The workpiece stage positioning system and projection lens focusing system of the lithography machine are jointly kinematically adjusted according to the composite correction instruction set. The composite correction instruction set is parsed to separate the focal length adjustment instruction to be executed by the projection lens focusing system and the plane tilt compensation instruction to be executed by the workpiece stage positioning system. The parsing process is based on preset field identifiers in the instruction encoding. The focal length adjustment instruction is converted into the number of rotations of the stepper motor in the projection lens focusing system, based on the calibration relationship between the number of pulses per rotation of the stepper motor and the focal length change corresponding to each pulse. The plane tilt compensation instruction is converted into the displacement increments of the six degrees of freedom drive axes in the workpiece stage positioning system. The displacement increments of the six degrees of freedom drive axes include translation along the X, Y, and Z axes and rotation about the X, Y, and Z axes. It can be understood that plane tilt compensation mainly involves rotational compensation about the X and Y axes and translational compensation about the Z axis. According to the preset execution priority, the position of the workpiece stage positioning system is first locked. The locking operation is achieved by switching the position loop controller of the workpiece stage positioning system to hold mode. Then, the projection lens focusing system is driven to complete the coarse adjustment of the focal length. The coarse adjustment of the focal length is performed by the stepper motor executing the axial displacement correction amount in the composite correction instruction set. Finally, the workpiece stage positioning system is micro-adjusted to eliminate the remaining tilt error. The micro-adjustment operation calculates and executes the micro-adjustment amount of the workpiece stage in multiple degrees of freedom based on the radial deformation correction amount and higher-order aberration compensation requirements in the composite correction instruction set. In some embodiments, the execution priority and motion timing are controlled by the state machine logic in the motion controller. Optionally, the micro-adjustment amount of the workpiece stage positioning system needs to take into account the change in the relative position between the lens and the workpiece stage caused by the action of the projection lens focusing system.

[0035] See Figure 4 This is a kinematics adjustment timing analysis diagram of a lithography machine, showing the time-varying curves of the adjustment amounts in three directions of the six degrees of freedom of the objective lens focusing and the stage under the composite correction command set of the lithography machine, reflecting the dynamic execution process of error correction. All adjustment amounts complete their dynamic response within 0–6 seconds, gradually converging to 0 nm after 6 seconds, indicating that the correction process is basically completed within 6 seconds. The objective lens focusing response is the fastest, reaching a peak of 1.2 nm at approximately 1.5 seconds, and is the main action in this correction. The objective lens focusing adjustment amount is the largest, indicating that axial aberration / thermal drift is the core error source in this correction. The stage Z-direction adjustment peak is approximately 0.8 nm, which, together with the objective lens focusing, constitutes the axial correction component. The stage X-direction peak is approximately 0.5 nm, and there is a negative adjustment in the Y-direction, used to compensate for planar tilt error. All adjustment amounts eventually converge to 0, indicating that the composite correction command set successfully offsets the overlay error introduced by aberration and thermal deformation.

[0036] In one embodiment of the present invention, after generating the inverse aberration compensation drive signal, a closed-loop fine-tuning step based on process feedback is performed. Immediately after completing an adjustment based on a composite correction instruction set, a short-exposure marking pattern printing is initiated. An online optical inspection subsystem scans the printed marking pattern to extract the actual edge placement error value. The extracted actual edge placement error value is compared with the theoretical design value to calculate the current residual error vector. This residual error vector is fed back to the thermal drift separation operator and the multidimensional convolution matching stage as a correction bias for the next iteration calculation, correcting the deviation of the nominal lens aberration feature library under the current operating conditions. Specifically, scanning the marking pattern and extracting the actual edge placement error value using the online optical inspection subsystem includes: acquiring high-resolution bright-field and dark-field images of the marking pattern; the bright-field image is used to identify the pattern outline, and the dark-field image is used to identify subsurface defects. Subpixel-level edge detection is performed on the bright-field image to determine the actual pixel coordinates of the pattern edges. The design layout of the marking pattern is retrieved to obtain the theoretical edge coordinates of the marking pattern. The actual pixel coordinates of the same edge are registered with the theoretical edge coordinates to eliminate projection deviation caused by the image acquisition angle. The minimum root mean square distance between the two sets of coordinates after registration is calculated, and this minimum root mean square distance is used as the edge placement error value, and its direction information is recorded.

[0037] In specific implementation, after generating the inverse aberration compensation drive signal, a closed-loop fine-tuning step based on process feedback is executed. Immediately after completing an adjustment based on a composite correction instruction set, a short-exposure marking pattern printing is initiated. The energy and duration of the short exposure are set to form a clear but incompletely developed marking pattern outline on the photoresist. The marking pattern is scanned using an online optical inspection subsystem, which includes a high-resolution camera and a specific wavelength illumination source. The actual edge placement error value is extracted, representing the deviation between the actual position of the pattern edge and the theoretical design position. The actual edge placement error value is compared with the theoretical design value, obtained from the marking pattern design layout database, to calculate the current residual error vector. The residual error vector is fed back to the thermal drift separation operator and the multidimensional convolution matching stage as a correction bias for the next iteration calculation. This correction bias is used to correct the deviation of the nominal lens aberration feature library under the current operating conditions. In some embodiments, the feedback of the residual error vector is achieved by modifying the constant term of the multiple linear regression model in the thermal drift separation operator and the weight coefficients of the feature vectors in the nominal lens aberration feature library. It is understandable that the closed-loop fine-tuning step is performed once after each exposure field or every few exposure fields. Scanning the marker pattern and extracting the actual edge placement error value using the online optical inspection subsystem involves several specific operations: acquiring high-resolution bright-field and dark-field images of the marker pattern; the high-resolution bright-field image is used to identify the pattern outline, and the dark-field image is used to identify subsurface defects. Subpixel-level edge detection is performed on the high-resolution bright-field image; the subpixel-level edge detection algorithm is based on gray-level gradients or image moments to determine the actual pixel coordinates of the pattern edges. The design layout of the marker pattern is called to obtain the theoretical edge coordinates of the marker pattern. The actual pixel coordinates and theoretical edge coordinates of the same edge are registered. Coordinate registration is achieved by finding rigid body transformations (translation and rotation) between the two sets of coordinate points to eliminate projection deviations caused by the image acquisition angle. The minimum root mean square distance between the two sets of coordinates after registration is calculated, and the minimum root mean square distance is used as the edge placement error value, and its direction information is recorded. (Minimum root mean square distance) The calculation formula is: , in: Indicates the first The actual pixel coordinates after registration Indicates the corresponding first A theoretical edge coordinate, This represents the total number of coordinate points involved in the calculation. In some embodiments, the orientation information is obtained by calculating the mean values ​​of the components of the error vector in the X and Y directions. Optionally, the scanning field of view of the online optical inspection subsystem covers multiple feature locations of the marked pattern. It can be understood that the edge placement error value and its orientation information together constitute the residual error vector, which is used to inversely correct the aberration and thermal drift model.

[0038] See Figure 5 This is a convergence analysis chart of the closed-loop correction of edge placement error in a lithography machine. It shows the trend of edge placement error with the number of iterations during the closed-loop fine-tuning process based on process feedback, used to evaluate the convergence performance of the correction algorithm. In the 0th iteration, the edge placement error is 5.2nm, which is the initial error level before correction. The error decreases monotonically with the number of iterations, with a faster rate of decrease in the early stages (0–4 iterations) and a gradual slowdown in the later stages (5–9 iterations), consistent with the convergence law of typical closed-loop correction. In the 1st iteration, the error drops to 3.8nm, a reduction of approximately 26.9%; in the 4th iteration, the error drops to 1.8nm, a cumulative reduction of approximately 65.4%. In the 7th iteration, the error drops to 1.0nm for the first time, reaching the convergence threshold, indicating that the closed-loop correction algorithm meets the accuracy requirements after 7 iterations. In the 9th iteration, the error further decreases to 0.8nm, below the convergence threshold, demonstrating the algorithm's continuous optimization capability.

[0039] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for error correction based on a photolithography machine, characterized in that, The method includes: An interferometric sampling array is deployed to capture the initial wavefront phase distribution in order to obtain a continuous and smooth reference wavefront field; The reference wavefront field is input to the thermal drift separation operator, which identifies the trend term in the wavefront field that changes linearly with temperature and then extracts the slowly varying drift component caused by the environmental thermal effect. The residual wavefront field after the slowly varying drift components are stripped is then subjected to multidimensional convolution matching with a pre-stored nominal lens aberration feature library to identify the static aberration patterns generated by the lithography machine projection lens. Based on the identified static aberration pattern, an inverse aberration compensation drive signal for the objective lens actuator is generated in the control system. While generating the inverse aberration compensation drive signal, the slow drift component is input to the time-varying trend predictor, which performs rolling estimation of the thermal drift amount at future times based on the Kalman filter framework. The inverse aberration compensation drive signal is superimposed with the estimated thermal drift value at the future time to synthesize a composite correction instruction set; Based on the composite correction instruction set, the workpiece stage positioning system and the projection lens focusing system of the lithography machine are jointly kinematically adjusted to counteract the overlay error caused by aberration and thermal deformation.

2. The lithography machine error correction method as described in claim 1, characterized in that, The deployment of the interferometric sampling array captures the initial wavefront phase distribution to obtain a continuous and smooth reference wavefront field, including: A high-density distributed interferometric sampling array is deployed within the exposure field of the lithography machine to capture the initial wavefront phase distribution of the wafer plane; A full-field unwrapping operation is performed on the initial wavefront phase distribution to remove measurement noise caused by phase folding, resulting in a continuous and smooth reference wavefront field. Specifically, this includes: The gradient amplitude of the initial wavefront phase distribution is calculated to determine the location of potential singularities where the phase changes abruptly in the wavefront field; Using the potential singularity location as the boundary, the entire exposure field is divided into several non-overlapping local sub-regions; Within each of the local sub-regions, a smooth phase gradient curve is recovered using a least-squares fitting algorithm; By integrating and accumulating the phase value of each pixel along the phase gradient curve, the absolute phase value of the local sub-region is reconstructed. The absolute phase values ​​of all the local sub-regions are seamlessly stitched together, and the discontinuities at the stitching points are smoothed out to finally output the reference wavefront field.

3. The lithography machine error correction method as described in claim 2, characterized in that, The reference wavefront field is input to a thermal drift separation operator, which identifies the trend term in the wavefront field that changes linearly with temperature, and then extracts the slowly varying drift component caused by environmental thermal effects, including: Several key optical paths that run through the entire exposure field are extracted from the reference wavefront field as characteristic optical paths, and the characteristic optical paths correspond to specific rows or columns in the interference sampling array; Acquire multi-point temperature sensing data corresponding to the characteristic optical path, and the temperature sensing data is strictly synchronized with the acquisition time of the reference wavefront field; A multiple linear regression model is constructed, wherein the phase value of the characteristic optical path is used as the dependent variable and the multi-point temperature sensing data is used as the independent variable. Solve for the regression coefficients of the multiple linear regression model, whereby the regression coefficients characterize the wavefront distortion caused by a unit temperature change; Using the obtained regression coefficients, an inverse scaling operation is performed on the reference wavefront field to subtract the phase values ​​affected by temperature from the reference wavefront field, thereby obtaining the slowly varying drift component.

4. The lithography machine error correction method as described in claim 3, characterized in that, The residual wavefront field, after stripping the slowly varying drift components, is subjected to multidimensional convolution matching with a pre-stored nominal lens aberration feature library to identify static aberration patterns generated by the lithography machine projection lens, including: A two-dimensional Fourier transform is performed on the residual wavefront field to convert it from the spatial domain to the frequency domain, resulting in a spectral distribution map. The spectrum distribution map is matched with the feature frequency vector of each preset aberration type in the nominal lens aberration feature library through multidimensional convolution, and the correlation coefficient is calculated. Select a predefined aberration type whose correlation coefficient exceeds a preset threshold value, and define it as a candidate aberration set; All aberration patterns in the candidate aberration set are weighted and superimposed, with the weights determined by their respective correlation coefficients, to reconstruct an approximate aberration residual field, which is the identified static aberration pattern.

5. The lithography machine error correction method as described in claim 4, characterized in that, Based on the identified static aberration pattern, an inverse aberration compensation drive signal for the objective lens actuator is generated in the control system, including: The static aberration pattern is mapped to a virtual Zernike polynomial coefficient space to obtain a set of Zernike coefficients characterizing the aberration magnitude. The sensitivity matrix of the objective lens actuator is retrieved, and the sensitivity matrix describes the mapping relationship between the actuator displacement and the change in the Zernike coefficient; By using matrix inversion, a set of actuator displacement adjustments required to offset the Zernike coefficients is calculated; The actuator displacement adjustment is converted into a control voltage signal, and the control voltage signal is subjected to amplitude limiting and smoothing filtering to form the inverse aberration compensation drive signal.

6. The lithography machine error correction method as described in claim 5, characterized in that, The slowly varying drift component is input into a time-varying trend predictor, which performs a rolling estimate of the thermal drift at future times based on a Kalman filter framework, including: Establish the state-space equation of the slowly varying drift component, taking the wavefront distortion at the current moment as the system state and the fluctuation of ambient temperature as the system noise; Initialize the state estimate and covariance matrix of the Kalman filter, wherein the covariance matrix reflects the uncertainty of the initial state estimate; In each new measurement cycle, the latest slowly drifting component data is received and input as an observation into the Kalman filter; The prediction step of the Kalman filter is executed, and the prior state at the current time is extrapolated based on the state estimate at the previous time step. Next, the Kalman filter update step is executed, and the prior state is corrected using the observation value to obtain the posterior state estimate, which is the thermal drift estimate for the future time.

7. The lithography machine error correction method as described in claim 6, characterized in that, The inverse aberration compensation drive signal is superimposed with the estimated thermal drift value at the future time to synthesize a composite correction instruction set, including: The inverse aberration compensation drive signal is decomposed into a defocus term acting on the objective lens axis and an astigmatic term acting on the objective lens radial direction. The estimated thermal drift at the future moment is also decomposed into axial thermal expansion displacement and radial thermal deformation displacement. The total axial displacement correction is obtained by algebraically adding the defocus term in the axial direction to the thermal expansion displacement in the axial direction. The radial astigmatism term and the radial thermal deformation displacement are vector-synthesized to obtain the total radial deformation correction amount; The adjusted total axial displacement correction, the total radial deformation correction, and the higher-order aberration terms not involved in the superposition are packaged together into the composite correction instruction set.

8. The lithography machine error correction method as described in claim 7, characterized in that, Based on the aforementioned composite correction instruction set, joint kinematic adjustments are made to the workpiece stage positioning system and the projection lens focusing system of the lithography machine, including: The composite correction instruction set is analyzed to separate the focal length adjustment instruction that needs to be executed by the projection lens focusing system and the plane tilt compensation instruction that needs to be executed by the workpiece stage positioning system; The focus adjustment command is converted into the number of rotations of the stepper motor of the projection lens focusing system; The planar tilt compensation command is converted into the displacement increment of the six-degree-of-freedom drive axes in the workpiece stage positioning system; According to the preset execution priority, the position of the workpiece stage positioning system is first locked, then the projection lens focusing system is driven to complete the coarse adjustment of the focal length, and finally the workpiece stage positioning system is finely moved to eliminate the remaining tilt error.

9. The lithography machine error correction method as described in claim 8, characterized in that, After generating the inverse aberration compensation drive signal, a closed-loop fine-tuning step based on process feedback is also included: After completing an adjustment based on the composite correction instruction set, immediately initiate a short-exposure marking pattern printing; The marking pattern is scanned using an online optical inspection subsystem to extract the actual edge placement error value; The actual edge placement error value is compared with the theoretical design value to calculate the current residual error vector; The residual error vector is fed back to the thermal drift separation operator and the multidimensional convolution matching stage as a correction bias for the next iteration calculation, so as to correct the deviation of the nominal lens aberration feature library under the current working conditions. The step of using an online optical inspection subsystem to scan the marked pattern and extract the actual edge placement error value includes: High-resolution bright-field and dark-field images of the marked pattern are acquired. The bright-field images are used to identify the pattern outline, and the dark-field images are used to identify subsurface defects. Perform subpixel-level edge detection on the bright-field image to determine the actual pixel coordinates of the pattern edges; The design layout of the marking pattern is invoked to obtain the theoretical edge coordinates of the marking pattern; The actual pixel coordinates of the same edge are registered with the theoretical edge coordinates to eliminate projection deviation caused by the image acquisition angle. Calculate the minimum root mean square distance between the two sets of coordinates after registration, use the minimum root mean square distance as the edge placement error value, and record its direction information.

10. A lithography machine error correction system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the lithography machine error correction method according to any one of claims 1 to 9.

Images