Motion precision detection system and method based on coupling of multi-wavelength interference and moire fringe

Abstract

The application provides a motion precision detection system and method based on multi-wavelength interference and Moire fringe coupling, and relates to the field of optical measurement.The system comprises a Moire fringe alignment and calibration module, a plurality of degrees of freedom motion detection modules, wherein the X-axis translation motion detection module comprises an X-axis multi-wavelength light source unit for providing an X-axis multi-wavelength laser beam, an X-axis optical interference unit comprising an X-axis beam splitter and an X-axis mirror, the X-axis multi-wavelength laser beam is projected onto the X-axis mirror of the worktable in the X-axis direction, the reflected light is superimposed with the reference light on the X-axis beam splitter to form X-axis interference fringes, a signal acquisition module for simultaneously capturing signals, and a signal processing module for calculating the multi-degree of freedom motion parameters of the worktable based on the captured signals, and performing motion precision detection, which has the advantage of realizing high-precision measurement of the multi-degree of freedom motion of the worktable in a complex industrial environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical measurement, and particularly to a motion precision detection system and method based on multi-wavelength interference and Moire fringe coupling. BACKGROUND

[0002] In the semiconductor manufacturing process, the workpiece table is the core component that carries the wafer and performs precise motion, and its motion precision directly affects the resolution and yield of chip manufacturing. In key processes such as lithography, the workpiece table needs to have nanometer-level motion precision to ensure accurate alignment of the exposure pattern. However, the existing detection methods have the following shortcomings:

[0003] 1. Limitations of single-wavelength interferometer: Existing laser interferometers are usually based on a single-wavelength light source, which can achieve high-precision measurement under certain conditions, but the measurement range is limited by the wavelength, making it difficult to balance large range and high precision at the same time.

[0004] 2. Complexity of multi-degree-of-freedom detection: For X, Y, Z axis translation and pitch, yaw, roll, etc. of the workpiece table, the traditional detection system needs multiple independent devices to work together, with high system complexity and low detection efficiency.

[0005] 3. Influence of environmental interference: Existing optical interference systems are sensitive to environmental vibration, temperature fluctuations, and other disturbances, limiting their application in industrial environments.

[0006] 4. Low initial alignment efficiency: Existing systems require complex initial alignment at startup, which takes a long time and affects production efficiency.

[0007] Therefore, it is necessary to provide a motion precision detection system and method based on multi-wavelength interference and Moire fringe coupling, which can realize high-precision measurement of the multi-degree-of-freedom motion of the workpiece table in complex industrial environments, while balancing large measurement range and anti-interference ability. SUMMARY

[0008] The application provides a motion precision detection system based on multi-wavelength interference and Moire fringe coupling, comprising: a Moire fringe alignment and calibration module, including a reference grating, a measurement grating and a calibration photodetector, the reference grating is fixed on the reference surface of the workpiece table, the measurement grating is fixed on the moving part of the workpiece table, the Moire fringe alignment and wavelength calibration are completed through the periodic characteristics of the Moire fringe; X-axis translation motion detection module, Y-axis translation motion detection module, Z-axis translation motion detection module, pitch motion detection module, yaw detection module and roll motion detection module; the X-axis translation motion detection module comprises: an X-axis multi-wavelength light source unit for providing an X-axis multi-wavelength laser; an X-axis optical interference unit comprising an X-axis beam splitter and an X-axis mirror, the X-axis multi-wavelength laser is projected onto the X-axis mirror of the workpiece table along the X-axis direction, the reflected light and the reference light are superimposed on the X-axis beam splitter to form X-axis interference fringes; a signal acquisition module for simultaneously capturing Moire fringe signals, X-axis interference fringes, Y-axis interference fringes, Z-axis interference fringes, pitch interference fringes, yaw interference fringes and roll interference fringes; a signal processing module for calculating multi-degree-of-freedom motion parameters of the workpiece table based on the Moire fringe signals, the X-axis interference fringes, the Y-axis interference fringes, the Z-axis interference fringes, the pitch interference fringes, the yaw interference fringes and the roll interference fringes, and performing motion precision detection.

[0009] Further, the X-axis multi-wavelength light source unit comprises a 532nm laser, a 635nm laser and a 780nm laser.

[0010] Further, the pitch motion detection module comprises: a pitch multi-wavelength light source unit for providing two Z-axis multi-wavelength lasers; a pitch optical interference unit comprising a pitch beam splitter and a pitch mirror, the two Z-axis multi-wavelength lasers are projected onto the pitch mirrors at two different positions of the workpiece table along the Z-axis direction respectively, the two reflected lights and the reference light are superimposed on the X-axis beam splitter to form pitch interference fringes.

[0011] Further, the yaw detection module comprises: a yaw multi-wavelength light source unit for providing two X-axis multi-wavelength lasers; a yaw optical interference unit comprising a yaw beam splitter and a yaw mirror, the two X-axis multi-wavelength lasers are projected onto the yaw mirrors at two different positions of the workpiece table along the X-axis direction respectively, the two reflected lights and the reference light are superimposed on the yaw beam splitter to form yaw interference fringes.

[0012] Further, the roll motion detection module comprises: a roll multi-wavelength light source unit for providing two Y-axis multi-wavelength lasers; a roll optical interference unit comprising a roll beam splitter and a roll mirror, the two Y-axis multi-wavelength lasers are projected onto the roll mirrors at two different positions of the workpiece table along the Y-axis direction respectively, the two reflected lights and the reference light are superimposed on the roll beam splitter to form roll interference fringes.

[0013] Further, the signal processing module calculates the multi-degree-of-freedom motion parameters of the worktable based on the Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe, including: generating phase data corresponding to each degree of freedom based on the Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe; and performing fusion processing on the phase data of the multi-degree-of-freedom using a synchronous solving algorithm to calculate the multi-degree-of-freedom motion parameters of the worktable.

[0014] Further, the signal processing module generates phase data corresponding to each degree of freedom based on the Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe, including: performing denoising processing on the Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe; and generating phase data corresponding to each degree of freedom based on the denoised Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe.

[0015] Further, the signal processing module performs denoising processing on the Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe, including: using a wavelet transform-convolutional neural network hybrid denoising model to perform wavelet decomposition on the Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe, perform feature extraction on the wavelet-decomposed Moiré fringe signal, X-axis interference fringe, Y-axis interference fringe, Z-axis interference fringe, pitch interference fringe, yaw interference fringe, and roll interference fringe, respectively, dynamically allocate denoising weights through an attention mechanism, and retain effective phase information.

[0016] Further, the signal processing module generates phase data corresponding to each degree of freedom based on the denoised Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe, including: constructing effective phase information into a graph structure through a phase unwrapping algorithm based on a graph neural network, with nodes being phase sampling points and edges representing spatial correlation; and learning phase jump paths of the graph structure using a graph neural network to generate phase data corresponding to each degree of freedom.

[0017] The application provides a motion precision detection method based on coupling of multi-wavelength interference and Moire fringes, which is applied to the motion precision detection system based on coupling of multi-wavelength interference and Moire fringes, and comprises the following steps: completing Moire fringe alignment and wavelength calibration; simultaneously capturing Moire fringe signals, X-axis interference fringes, Y-axis interference fringes, Z-axis interference fringes, pitch interference fringes, yaw interference fringes and roll interference fringes; and calculating multi-degree-of-freedom motion parameters of a workpiece table based on the Moire fringe signals, the X-axis interference fringes, the Y-axis interference fringes, the Z-axis interference fringes, the pitch interference fringes, the yaw interference fringes and the roll interference fringes, and performing motion precision detection.

[0018] Compared with the prior art, the motion precision detection system and method based on coupling of multi-wavelength interference and Moire fringes provided by the application have at least the following beneficial effects:

[0019] The multi-wavelength light source unit is used to provide multi-wavelength laser, the optical interference unit is used to form interference fringes, and the multi-wavelength interference technology can be used to realize high-precision measurement. Through comprehensive processing of the interference fringe signals and the Moire fringe signals, the measurement range can be significantly expanded, more complex motion requirements can be met, and high precision and a large range can be considered.

[0020] The signal acquisition module can simultaneously capture interference fringe signals of each degree of freedom, and the signal processing module can calculate multi-degree-of-freedom motion parameters of the workpiece table based on the signals, so that real-time detection of six-degree-of-freedom motion of the workpiece table can be realized.

[0021] The Moire fringe signals have specific periodic characteristics, and have certain anti-interference ability to environmental interference.

[0022] The signal acquisition module simultaneously captures multiple signals, and the signal processing module combines the Moire fringe signals and multi-channel signal acquisition and filtering processing, so that the influence of environmental interference such as vibration and temperature change can be effectively suppressed.

[0023] The Moire fringe alignment and calibration module uses the periodic characteristics of the Moire fringes to complete Moire fringe alignment and wavelength calibration. The Moire fringes have high sensitivity and can quickly respond to position changes of the workpiece table. By using the high sensitivity of the Moire fringes, the system can quickly complete the initial alignment of the workpiece table, and the system startup efficiency is improved.

[0024] The modular design is adopted, including the Moire fringe alignment and calibration module, the multi-degree-of-freedom motion detection module, the signal acquisition module and the signal processing module, and the modules are independent of each other and work cooperatively. The modular design simplifies the device architecture, reduces the complexity and maintenance cost of the system, and the cooperative work of the modules improves the detection efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0025] The present specification will be further described in the manner of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same numbers represent the same structures, wherein:

[0026] Figure 1 is a module schematic diagram of a motion precision detection system based on multi-wavelength interference and Moiré fringe coupling according to some embodiments of the present specification;

[0027] Figure 2 is a flowchart schematic diagram of a motion precision detection method based on multi-wavelength interference and Moiré fringe coupling according to some embodiments of the present specification;

[0028] Figure 3 is a schematic diagram of denoising by using a wavelet transform-convolutional neural network hybrid denoising model according to some embodiments of the present specification;

[0029] Figure 4 is a schematic diagram of generating phase data corresponding to each degree of freedom according to some embodiments of the present specification. DETAILED DESCRIPTION

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present specification, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some examples or embodiments of the present specification, and for those skilled in the art, without paying creative labor, the present specification can also be applied to other similar scenarios according to these drawings. Unless it is obvious from the language environment or otherwise stated, the same reference numbers in the drawings represent the same structure or operation.

[0031] Figure 1 is a module schematic diagram of a motion precision detection system based on multi-wavelength interference and Moiré fringe coupling according to some embodiments of the present specification, as shown in Figure 1 The motion precision detection system based on multi-wavelength interference and Moiré fringe coupling includes a Moiré fringe alignment and calibration module, an X-axis translation motion detection module, a Y-axis translation motion detection module, a Z-axis translation motion detection module, a pitch motion detection module, a yaw detection module, a roll motion detection module, a signal acquisition module, and a signal processing module.

[0032] The Moiré fringe alignment and calibration module includes a reference grating, a measurement grating, and a calibration photodetector. The reference grating is fixed on the reference surface of the workpiece table, and the measurement grating is fixed on the moving part of the workpiece table. Through the periodic characteristics of the Moiré fringe, the Moiré fringe alignment and wavelength calibration are completed.

[0033] Specifically, the Moiré fringes are interference phenomena generated by the superposition of two periodic gratings (reference grating and measurement grating). When the relative position of the two gratings changes, the period and direction of the Moiré fringes will also change. The generation principle is as follows:

[0034] 1. Grating superposition: the periods of the reference grating and the measurement grating are d1 and d2 respectively, and when the two gratings are superimposed, Moiré fringes will be formed, and their period D is determined by the following formula:

[0035]

[0036] 2. Fringe movement: when the measurement grating moves with the workpiece table, the Moiré fringes will move with it. The movement direction of the Moiré fringes is perpendicular to the relative displacement direction of the gratings, and the movement distance is proportional to the grating displacement.

[0037] 3. Grating installation and adjustment: the reference grating is fixed on the reference surface of the workpiece table, and the measurement grating is fixed on the moving part of the workpiece table. Adjust the relative angle of the two gratings to ensure the clarity and contrast of the Moiré fringes.

[0038] 4. Fringe observation and alignment: observe the movement direction and period of the Moiré fringes through the photoelectric detector. According to the movement direction of the Moiré fringes, adjust the position of the workpiece table to achieve the initial alignment state.

[0039] 5. Alignment accuracy: the high sensitivity of the Moiré fringes makes the initial alignment accuracy reach the sub-micron level, significantly improving the system startup efficiency.

[0040] The X-axis translation motion detection module includes:

[0041] The X-axis multi-wavelength light source unit is used to provide an X-axis multi-wavelength laser beam. Through fiber coupling technology, multiple wavelength laser signals are introduced into the X-axis optical interference unit.

[0042] The X-axis optical interference unit includes an X-axis beam splitter and an X-axis mirror. The X-axis multi-wavelength laser is projected onto the X-axis mirror of the workpiece table along the X-axis direction. The reflected light and the reference light are superimposed on the X-axis beam splitter to form X-axis interference fringes.

[0043] Specifically, the measurement range of the motion detection module is closely related to the combination of laser wavelengths. Generally, longer wavelengths can provide a larger measurement range, while shorter wavelengths can provide higher precision. To achieve nanometer-level precision, at least one shorter wavelength (such as 532 nm or 405 nm) is usually required. When selecting the wavelength, the ratio between the wavelengths should be considered. Generally, the ratio between the wavelengths should be a non-integer ratio to avoid ambiguity in the unwrapping algorithm. The motion detection module eliminates the phase ambiguity problem in single-wavelength interference through multi-wavelength interference and wavelength unwrapping algorithm. The selection of wavelength combination should ensure the effectiveness and stability of the unwrapping algorithm.

[0044] For example only, the multi-wavelength laser can be:

[0045] Example 1: λ1: 532 nm (green laser), λ2: 635 nm (red laser), λ3: 780 nm (near-infrared laser).

[0046] Example 2: λ1: 405 nm (blue laser), λ2: 532 nm (green laser), λ3: 1064 nm (infrared laser).

[0047] Example 3: λ1: 450 nm (blue laser), λ2: 650 nm (red laser), λ3: 850 nm (near-infrared laser).

[0048] As preferred, the X-axis multi-wavelength light source unit includes a 532 nm laser, a 635 nm laser, and a 780 nm laser. 532 nm is a common green laser wavelength, with mature technology and high light source stability, suitable for high-precision measurement. 635 nm is a common red laser wavelength, and the light source is easy to obtain and has good stability. 780 nm is a near-infrared laser wavelength commonly used in industrial measurement, with strong anti-interference ability and high light source stability.

[0049] It can be understood that the multi-wavelength interference signal can suppress the influence of environmental interference (such as vibration and temperature fluctuation). For example, the combination of 532 nm and 635 nm has low sensitivity to environmental interference, and the introduction of 780 nm can further improve the anti-interference ability. Through filtering and amplification processing of the multi-wavelength signal, the signal-to-noise ratio can be further improved to ensure the stability and accuracy of the measurement.

[0050] When the worktable translates along the X-axis, the path length of the reflected light changes, causing the interference fringes to move. By detecting the phase change of the interference fringes.

[0051] The structure of the Y-axis translation motion detection module and the Z-axis translation motion detection module is consistent with that of the X-axis translation motion detection module, and will not be described here.

[0052] The pitch motion detection module comprises:

[0053] A pitch multi-wavelength light source unit is configured to provide two beams of Z-axis multi-wavelength laser light.

[0054] A pitch optical interference unit comprises a pitch beam splitter and a pitch mirror, and the two beams of Z-axis multi-wavelength laser light are projected onto the pitch mirror at two different positions of the worktable along the Z-axis direction, respectively, and the two beams of reflected light and the reference light are superimposed on the X-axis beam splitter to form pitch interference fringes.

[0055] When the worktable pitches, the path length difference of the two beams of reflected light changes, causing the interference fringes to move. The phase difference between the two beams of interference fringes is detected.

[0056] The yaw detection module comprises:

[0057] A yaw multi-wavelength light source unit is configured to provide two beams of X-axis multi-wavelength laser light.

[0058] A yaw optical interference unit comprises a yaw beam splitter and a yaw mirror, and the two beams of X-axis multi-wavelength laser light are projected onto the yaw mirror at two different positions of the worktable along the X-axis direction, respectively, and the two beams of reflected light and the reference light are superimposed on the yaw beam splitter to form yaw interference fringes.

[0059] When the worktable yaws, the path length difference of the two beams of reflected light changes, causing the interference fringes to move. The phase difference between the two beams of interference fringes is detected.

[0060] The roll motion detection module comprises:

[0061] A roll multi-wavelength light source unit is configured to provide two beams of Y-axis multi-wavelength laser light.

[0062] A roll optical interference unit comprises a roll beam splitter and a roll mirror, and the two beams of multi-wavelength laser light are projected onto the roll mirror at two different positions of the worktable along the Y-axis direction, respectively, and the two beams of reflected light and the reference light are superimposed on the roll beam splitter to form roll interference fringes.

[0063] When the worktable rolls, the path length difference of the two beams of reflected light changes, causing the interference fringes to move. The phase difference between the two beams of interference fringes is detected.

[0064] A signal acquisition module is configured to simultaneously capture the Moiré fringe signal, the X-axis interference fringes, the Y-axis interference fringes, the Z-axis interference fringes, the pitch interference fringes, the yaw interference fringes, and the roll interference fringes.

[0065] As preferred, the signal acquisition module can be equipped with a high-sensitivity photodetector, which can capture multiple wavelengths of interference signals simultaneously. The signal acquisition module is a key component of the multi-wavelength interference system, responsible for converting interference fringes into electrical signals and performing preprocessing to improve signal-to-noise ratio and measurement accuracy. If a silicon-based photodiode (SiPhotodiode) is selected, it is suitable for visible and near-infrared wavebands. A low-noise amplifier is used to amplify the weak electrical signals output by the photodetector while minimizing the noise introduced. The low-noise operational amplifier can be ADI's AD797 or TI's OPA1611.

[0066] The signal processing module is used to calculate the multi-degree-of-freedom motion parameters of the workpiece table based on the Moiré fringe signal, X-axis interference fringe, Y-axis interference fringe, Z-axis interference fringe, pitch interference fringe, yaw interference fringe, and roll interference fringe, and perform motion accuracy detection.

[0067] Specifically, the signal processing module can use a distributed processing architecture to independently solve the phase changes corresponding to each wavelength.

[0068] As preferred, the signal processing module calculates the multi-degree-of-freedom motion parameters of the workpiece table based on the Moiré fringe signal, X-axis interference fringe, Y-axis interference fringe, Z-axis interference fringe, pitch interference fringe, yaw interference fringe, and roll interference fringe, including:

[0069] Based on the Moiré fringe signal, X-axis interference fringe, Y-axis interference fringe, Z-axis interference fringe, pitch interference fringe, yaw interference fringe, and roll interference fringe, phase data corresponding to each degree of freedom is generated;

[0070] The multi-degree-of-freedom phase data is fused and processed using a synchronous solving algorithm to calculate the multi-degree-of-freedom motion parameters of the workpiece table.

[0071] As preferred, the signal processing module generates phase data corresponding to each degree of freedom based on the Moiré fringe signal, X-axis interference fringe, Y-axis interference fringe, Z-axis interference fringe, pitch interference fringe, yaw interference fringe, and roll interference fringe, including:

[0072] The Moiré fringe signal, X-axis interference fringe, Y-axis interference fringe, Z-axis interference fringe, pitch interference fringe, yaw interference fringe, and roll interference fringe are denoised;

[0073] Based on the denoised Moiré fringe signal, X-axis interference fringe, Y-axis interference fringe, Z-axis interference fringe, pitch interference fringe, yaw interference fringe, and roll interference fringe, phase data corresponding to each degree of freedom is generated.

[0074] Figure 3is a schematic diagram of denoising using a wavelet transform-convolutional neural network hybrid denoising model according to some embodiments of the present specification, as shown in Figure 3 As preferred, the signal processing module performs denoising processing on the Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe, including:

[0075] The wavelet transform-convolutional neural network hybrid denoising model is used to perform wavelet decomposition on the Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe, and the roll interference fringe, and to perform feature extraction on the wavelet-decomposed Moiré fringe signal, X-axis interference fringe, Y-axis interference fringe, Z-axis interference fringe, pitch interference fringe, yaw interference fringe, and roll interference fringe, respectively, to dynamically allocate denoising weights through an attention mechanism and to retain effective phase information.

[0076] Specifically, the wavelet transform-convolutional neural network hybrid denoising model can include:

[0077] 1. Multi-scale wavelet decomposition layer:

[0078] A Daubechies 9 wavelet basis is used for 3-layer decomposition to decompose the original signal into low-frequency components and high-frequency components. The Daubechies wavelet basis is a class of wavelet basis functions with compact support, orthogonality, and approximate symmetry. The Daubechies 9 wavelet basis (db9) has 9 vanishing moments, which means it can better capture the local features of the signal when processing the signal, and has good compression and denoising effect for signals with polynomial trend. The original signal is decomposed into components of different frequency ranges through 3-layer wavelet decomposition. Each decomposition further decomposes the low-frequency component of the previous layer into low-frequency and high-frequency components. After 3-layer decomposition, the original signal is decomposed into components of different frequency bands, including low-frequency approximation components and high-frequency detail components of different levels. Among them, the low-frequency component corresponds to the overall phase trend of the signal, which reflects the change law of the signal in a long time range. In signal processing, the low-frequency component usually contains the main information of the signal, such as the phase change caused by the large-scale displacement of the moving platform. However, due to the long period of low-frequency signals, traditional convolution operations may not be able to fully capture these long-period changes, which can easily lead to errors in phase unwrapping operations. Therefore, the goal of processing the low-frequency component is to expand the receptive field to better capture long-period phase changes and avoid phase unwrapping errors.

[0079] The high-frequency components contain noise and local discontinuities, such as phase jumps caused by motion stage vibration and random noise caused by environmental electromagnetic interference. Although these high-frequency information may have a negative impact on the overall quality of the signal, it also contains important local features, such as real phase jump edges. Therefore, the goal of processing high-frequency components is to extract high-frequency noise features while preserving real phase jump edges to improve the signal-to-noise ratio and accuracy of the signal.

[0080] For the low-frequency approximation component, a wide kernel convolution (5x1) combined with a dilated convolution (dilation=2) is used. The wide kernel convolution (5x1) means that the convolution kernel has a large width in the spatial dimension, which can increase the coverage of the convolution operation in space and thus expand the receptive field. The dilated convolution (dilation=2) introduces a hole in the convolution kernel, allowing the convolution kernel to skip some pixel points during calculation, further expanding the receptive field. By combining wide kernel convolution and dilated convolution, the receptive field can be effectively expanded without significantly increasing the computational load, better capturing long-period phase changes in the low-frequency component and preventing phase unwrapping errors caused by large-scale displacement of the motion stage.

[0081] For the high-frequency component, a Dense Residual Block structure is used to extract high-frequency noise features (such as random noise caused by environmental electromagnetic interference) while preserving real phase jump edges. The Dense Residual Block is a network structure in deep learning that uses dense connections and residual connections to allow the network to better learn the features of the input signal. In processing high-frequency components, the 3x1 convolution in the Dense Residual Block structure can effectively extract high-frequency noise features such as random noise caused by environmental electromagnetic interference. At the same time, the skip connection can directly pass the input signal to the subsequent layer, preserving useful information in the input signal such as real phase jump edges. This structure can extract noise features while preserving as much local feature information as possible, improving the effectiveness of signal processing.

[0082] Grouped Convolution is introduced to group input channels by physical degrees of freedom (such as X / Y / Z axes belonging to different groups), reducing parameter redundancy and enhancing motion parameter independence. Input channels are grouped by physical degrees of freedom, i.e., channels belonging to different physical degrees of freedom are assigned to different groups, and each group performs convolution operations independently. Through grouped convolution, the number of parameters in the convolution operation can be reduced, and parameter redundancy can be reduced. Because each group only processes information of the physical degree of freedom corresponding to the group, mixing of information between different physical degrees of freedom is avoided, thereby enhancing the independence of motion parameters. This allows the network to better learn features in different physical degrees of freedom, improving the accuracy and robustness of signal processing.

[0083] 2. Hybrid attention mechanism

[0084] Channel attention:

[0085] Global average pooling: The operation converts a two-dimensional feature map on each channel into a scalar value by performing global average pooling on the spatial dimensions, generating a channel description vector:

[0086]

[0087] where, is the scalar value converted from the two-dimensional feature map on the channel, T is the sequence length, is the data value of the c-th channel at time t.

[0088] Assuming the size of the feature map is C×H×W (C is the number of channels, H and W are the height and width of the feature map, respectively), after global average pooling, a vector of length C is obtained. Here the signal length T can be understood as when processing one-dimensional signal or sequence data, similarly, the features of each channel are aggregated to obtain a scalar value representing the channel. Through global average pooling, the spatial information is compressed into the channel dimension, so that the subsequent attention calculation can be based on the features of the entire channel, rather than being limited to local regions.

[0089] Two fully connected layers: Two fully connected layers are used to learn the channel weights. The first fully connected layer is nonlinearly transformed by the ReLU activation function (δ) to increase the expression ability of the network; the second fully connected layer maps the output to the [0,1] interval by the Sigmoid activation function (σ) to obtain the weight of each channel:

[0090]

[0091] where, is the weight of the channel, σ is the Sigmoid activation function, is the weight matrix of the first fully connected layer, is the weight matrix of the second fully connected layer, δ is the ReLU activation function.

[0092] Through the learning of the two fully connected layers, the network can automatically assign different weights to each channel according to the importance of the input features. For noise-dominant channels (such as random interference in the high frequency band), their weights will be suppressed; while for effective phase signal channels, their weights will be enhanced, thereby improving the effect of signal processing.

[0093] Spatial attention:

[0094] In the feature map X, the similarity between position i and global position j is calculated. Here Q, K are the query and key generated by 1x1 convolution. The 1x1 convolution can linearly transform the channel without changing the spatial size of the feature map, thereby generating the query and key vectors for similarity calculation. By calculating the dot product or other similarity measurement methods between the query vector and the key vector, the similarity value between position i and position j is obtained:

[0095]

[0096] wherein, is the similarity between position i and global position j in the feature map X, denotes the dot product operation of the query vector and the key vector .

[0097] By calculating the similarity between global positions, the spatial attention mechanism can capture the long-distance dependency between different positions in the feature map.

[0098] For phase signals with periodicity, such as fixed periods of Moire fringes, the spatial attention mechanism can enhance the weights between similar positions, making these periodic signals more coherent in the feature map. Because periodic signals have similar features at different positions, their similarity is high, and the spatial attention mechanism can enhance the weights of these positions, thereby improving the periodic feature extraction ability of the signal. Isolated noise points usually have large differences in signal features with surrounding positions, and their similarity is low. The spatial attention mechanism will suppress the weights of these isolated noise points, thereby reducing the impact of noise on signal processing.

[0099] In traditional attention mechanisms, channel attention and spatial attention are usually cascaded, that is, channel attention is calculated first, and then the channel attention weighted feature map is input into the spatial attention module for calculation. This cascading method may cause information loss during transmission and cannot fully utilize the information in the channel and spatial dimensions. Channel attention and spatial attention are parallelly fused:

[0100]

[0101] wherein, α is a learnable parameter, is the channel attention weighted feature map, For the spatial attention weighted feature map, Y is the final output feature map. Through the parallel fusion mode, the network can simultaneously consider the information of the channel and space dimensions, and adaptively adjust the weights of the channel attention and the spatial attention through the learnable parameter a. This way can make more full use of the information in the feature map, and improve the performance and robustness of signal processing.

[0102] Residual learning and wavelet reconstruction:

[0103] The phase information of the original signal is preserved through the residual connection, avoiding phase distortion caused by nonlinear operations in the denoising process. Wavelet reconstruction is the inverse process of wavelet decomposition. In the previous wavelet decomposition step, the original signal is decomposed into low-frequency components and high-frequency components. Wavelet reconstruction is to recombine these decomposed components according to certain rules to restore the approximate representation of the original signal. When reconstructing, a phase consistency constraint is adopted: L2 regularization is applied to the low-frequency component to ensure that the overall phase shift matches the actual displacement of the motion platform. Because the actual displacement of the motion platform is a relatively stable physical quantity, its corresponding phase change should also be relatively smooth and stable. Through L2 regularization, it can avoid excessive fluctuations in the low-frequency component during reconstruction, thereby ensuring the phase consistency of the reconstructed signal with the actual displacement of the motion platform.

[0104] Compared with the prior art, the wavelet transform-convolutional neural network hybrid denoising model has the performance advantages shown in Table 1.

[0105] Table 1

[0106] Index Traditional wavelet threshold method CNN denoising method Wavelet transform-convolutional neural network hybrid denoising model Phase jump reservation capability Poor Medium Excellent Multi-degree of freedom coupling processing No No Excellent Environmental noise suppression ratio (dB) 10-15 18-22 25-30 Real-time performance (ms / frame) 5 20 12

[0107] Compared with traditional FIR / IIR filters, the wavelet transform-convolutional neural network hybrid denoising model improves the suppression effect of non-stationary noise (such as mechanical vibration) by 40%, and improves the signal-to-noise ratio by more than 15dB.

[0108] As a preferred, the signal processing module generates phase data corresponding to each degree of freedom based on the denoised Moiré fringe signal, the X-axis interference fringe, the Y-axis interference fringe, the Z-axis interference fringe, the pitch interference fringe, the yaw interference fringe and the roll interference fringe, including:

[0109] Through the phase unwrapping algorithm based on the graph neural network, the effective phase information is constructed into a graph structure, the nodes are phase sampling points, and the edges represent spatial correlation.

[0110] Using the graph neural network, the phase jump path of the graph structure is learned to generate phase data corresponding to each degree of freedom, solving the cumulative error problem of the traditional Itoh algorithm under complex deformation. Within a 10μm displacement range, the phase unwrapping error is reduced from ±0.5rad of the traditional method to ±0.1rad.

[0111] Figure 4 is a schematic diagram of generating phase data corresponding to each degree of freedom according to some embodiments of the present specification, as shown in Figure 4 Specifically, as shown in the figure, the graph neural network automatically enhances the cross-jump edge connection through the graph attention mechanism in the phase jump area, adopts the residual learning strategy, and directly predicts the phase jump correction amount instead of the absolute phase value:

[0112]

[0113] wherein, is the continuous phase obtained after processing, is the phase jump correction amount, K is the integer jump number, is the wrapped phase.

[0114] The loss function of the graph neural network is a weighted sum function of the phase continuity loss, the gradient consistency loss and the loop path integral loss, wherein the phase continuity loss is:

[0115]

[0116] is the phase continuity loss, N is the number of samples, is the predicted unwrapped phase value for position i, is the true unwrapped phase value for position i.

[0117] Gradient consistency loss:

[0118]

[0119] wherein, is the gradient consistency loss, is the predicted unwrapped phase value for position i, is the predicted unwrapped phase value for position j, is the true unwrapped phase value for position i, is the true unwrapped phase value for position j.

[0120] Loop path integral loss:

[0121]

[0122] wherein, is the loop path integral loss, is the predicted unwrapped phase, is the gradient of the predicted unwrapped phase, dl is the infinitesimal length on the integral path C.

[0123] In some embodiments, the signal processing module may calculate the multi-degree-of-freedom motion parameters of the workpiece stage by:

[0124] 1. Translational motion solution: For X, Y, and Z axis translation, the displacement is calculated using the unwrapped phase information:

[0125]

[0126] : The displacement of the workpiece stage on one of the X, Y or Z axes, is the phase after unwrapping, is the equivalent wavelength.

[0127] According to the above formula, calculate the displacement of X, Y and Z axis respectively 、 、 .

[0128] 2. Rotational motion calculation: For pitch, yaw, and roll, the angle change is calculated by the phase difference of the dual-beam interference signal:

[0129]

[0130] in, is the phase difference of the dual-beam interference signal, and d is the distance between the dual beams.

[0131] Calculate the pitch angle according to the above formula , yaw angle , roll angle .

[0132] 3. Synchronous solution: Synchronize the solution results of translational motion and rotational motion to output six-degree-of-freedom motion parameters:

[0133] Six degrees of freedom parameters = [ , , , , , ].

[0134] In some embodiments, the motion accuracy detection system based on multi-wavelength interference and moiré fringe coupling may also include a result display module for providing a user-friendly graphical interface, real-time display of displacement curves, rotation angle changes and error trend graphs, and supporting the storage, export and report generation of measurement data.

[0135] As an example, a motion accuracy detection system based on multi-wavelength interference and moiré fringe coupling can detect the dynamic motion accuracy of the workpiece stage during production, ensuring high-precision wafer positioning.

[0136] Further, the motion precision detection system based on the coupling of multi-wavelength interference and Moiré fringes can acquire multi-degree-of-freedom motion data in high-precision mechanical motion experiments, and provide reliable support for research.

[0137] Figure 2 FIG. 1 is a flow diagram of a motion precision detection method based on the coupling of multi-wavelength interference and Moiré fringes according to some embodiments of the present specification. Figure 2 As shown in FIG. 1, the motion precision detection method based on the coupling of multi-wavelength interference and Moiré fringes can include the following steps:

[0138] aligning the Moiré fringes and calibrating the wavelengths;

[0139] simultaneously capturing Moiré fringes signals, X-axis interference fringes, Y-axis interference fringes, Z-axis interference fringes, pitch interference fringes, yaw interference fringes, and roll interference fringes;

[0140] calculating multi-degree-of-freedom motion parameters of the worktable based on the Moiré fringes signals, the X-axis interference fringes, the Y-axis interference fringes, the Z-axis interference fringes, the pitch interference fringes, the yaw interference fringes, and the roll interference fringes, and performing motion precision detection.

[0141] The motion precision detection method based on the coupling of multi-wavelength interference and Moiré fringes can be applied to the motion precision detection system based on the coupling of multi-wavelength interference and Moiré fringes described above, which will not be repeated here.

[0142] Finally, it should be understood that the embodiments described herein are merely intended to illustrate the principles of the embodiments of the present specification. Other variations can also be within the scope of the present specification. Therefore, as an example but not limitation, alternative configurations of the embodiments of the present specification can be considered consistent with the teachings of the present specification. Accordingly, the embodiments of the present specification are not limited to the embodiments explicitly introduced and described in the present specification.

Claims

1. A motion accuracy detection system based on multi-wavelength interference and moiré fringe coupling, characterized in that: include: The moiré fringe alignment and calibration module includes a reference grating, a measuring grating, and a calibration photodetector. The reference grating is fixed to the reference surface of the workpiece stage, and the measuring grating is fixed to the moving part of the workpiece stage. The moiré fringe alignment and wavelength calibration are completed by using the periodic characteristics of the moiré fringe. X-axis translation motion detection module, Y-axis translation motion detection module, Z-axis translation motion detection module, pitch motion detection module, yaw detection module and roll motion detection module; The X-axis translation motion detection module includes: An X-axis multi-wavelength light source unit is used to provide a beam of X-axis multi-wavelength laser light; The X-axis optical interference unit includes an X-axis beam splitter and an X-axis reflector. The X-axis multi-wavelength laser is projected onto the X-axis reflector of the workpiece stage along the X-axis direction. The reflected light and the reference light are superimposed on the X-axis beam splitter to form X-axis interference fringes. Signal acquisition module, used to simultaneously capture moiré fringe signals, X-axis interference fringes, Y-axis interference fringes, Z-axis interference fringes, pitch interference fringes, yaw interference fringes and roll interference fringes; The signal processing module is used to calculate the multi-degree-of-freedom motion parameters of the workpiece stage based on the moiré fringe signals, X-axis interference fringes, Y-axis interference fringes, Z-axis interference fringes, pitch interference fringes, yaw interference fringes, and roll interference fringes, and perform motion accuracy detection. Specifically, it includes: A wavelet transform-convolutional neural network hybrid denoising model is used to perform wavelet decomposition on the moiré fringe signals, X-axis interference fringes, Y-axis interference fringes, Z-axis interference fringes, pitch interference fringes, yaw interference fringes, and roll interference fringes. Feature extraction is then performed on the decomposed moiré fringe signals, X-axis interference fringes, Y-axis interference fringes, Z-axis interference fringes, pitch interference fringes, yaw interference fringes, and roll interference fringes, respectively. The denoising weights are dynamically assigned through an attention mechanism to retain effective phase information. Through the phase unwrapping algorithm based on graph neural network, the effective phase information is constructed into a graph structure, where the nodes are phase sampling points and the edges represent spatial correlation; Using graph neural networks, we learn the phase transition paths of graph structures and generate phase data corresponding to each degree of freedom. The synchronous solution algorithm is used to fuse the phase data of multiple degrees of freedom and calculate the multi-degree-of-freedom motion parameters of the workpiece stage; In the phase jump region, the graph neural network automatically enhances cross-jump edge connections through the graph attention mechanism and adopts a residual learning strategy to directly predict the phase jump correction amount: in, is the continuous phase obtained after processing, is the phase jump correction value, K is the integer jump number, is the winding phase; The loss function of the graph neural network is a weighted sum function of phase continuity loss, gradient consistency loss, and loop path integral loss, where the phase continuity loss is: is the phase continuity loss, N is the number of samples, is the unwrapped phase value predicted for position i, is the true unwrapped phase value at position i; Gradient consistency loss: in, is the gradient consistency loss, is the unwrapped phase value predicted for position i, is the predicted unwrapped phase value for position j, is the true unwrapped phase value at position i, is the true unwrapped phase value at position j; Loop path integral loss: in, is the loop path integral loss, is the predicted unwrapping phase, is the predicted gradient of the unwrapping phase, and dl is the length of the infinitesimal element on the integration path C.

2. The motion accuracy detection system based on multi-wavelength interference and moiré fringe coupling according to claim 1, characterized in that: The X-axis multi-wavelength light source unit includes a 532nm laser, a 635nm laser, and a 780nm laser.

3. The motion accuracy detection system based on multi-wavelength interference and moiré fringe coupling according to claim 2, characterized in that: The pitch motion detection module includes: A pitch multi-wavelength light source unit, used to provide two beams of Z-axis multi-wavelength lasers; The pitch optical interference unit includes a pitch beam splitter and a pitch reflector. It uses two Z-axis multi-wavelength laser beams to project onto the pitch reflectors at two different positions on the workpiece stage along the Z-axis direction. The two reflected light beams and the reference light are superimposed on the X-axis beam splitter to form pitch interference fringes.

4. The motion accuracy detection system based on multi-wavelength interference and moiré fringe coupling according to claim 1, characterized in that: The yaw detection module includes: A yaw multi-wavelength light source unit, used to provide two beams of X-axis multi-wavelength lasers; The yaw optical interference unit includes a yaw beam splitter and a yaw reflector. It uses two beams of X-axis multi-wavelength laser to project along the X-axis direction onto the yaw reflectors at two different positions on the workpiece stage. The two reflected light beams and the reference light are superimposed on the yaw beam splitter to form yaw interference fringes.

5. The motion accuracy detection system based on multi-wavelength interference and moiré fringe coupling according to claim 1, characterized in that: The rolling motion detection module includes: A rolling multi-wavelength light source unit, used to provide two beams of Y-axis multi-wavelength laser light; The rolling optical interference unit includes a rolling beam splitter and a rolling reflector. Two multi-wavelength laser beams are projected along the Y-axis direction onto the rolling reflector at two different positions on the workpiece stage. The two reflected light beams and the reference light are superimposed on the rolling beam splitter to form rolling interference fringes.

6. A motion accuracy detection method based on multi-wavelength interference and moiré fringe coupling, characterized in that: The motion accuracy detection system based on multi-wavelength interference and moiré fringe coupling as claimed in any one of claims 1 to 5 comprises: Complete moiré fringe alignment and wavelength calibration; Simultaneously capture moiré fringe signals, X-axis interference fringes, Y-axis interference fringes, Z-axis interference fringes, pitch interference fringes, yaw interference fringes and roll interference fringes; Based on the moiré fringe signals, X-axis interference fringes, Y-axis interference fringes, Z-axis interference fringes, pitch interference fringes, yaw interference fringes and roll interference fringes, the multi-degree-of-freedom motion parameters of the workpiece stage are calculated to perform motion accuracy detection.

Images