A focusing control method and device for wafer topography detection

Abstract

The application provides a focusing control method and device for wafer topography detection. The method comprises: controlling a wafer to be detected to move relative to an image acquisition device according to a preset detection path, the direction of relative movement being in a plane parallel to the reference plane of a stage carrying the wafer to be detected, and the wafer to be detected being divided into multiple detection areas according to the detection path; judging whether a current detection area meets a preset condition, and performing a primary capture operation or a predicted guide reset operation on the current detection area according to the judgment result; and processing the focusing tasks of the first detection and the subsequent detection in modes, establishing an area focusing reference through the first detection, and providing priori for the subsequent detection, so that repeated large-range axial search on all areas is avoided, and the detection efficiency in large-area, multi-area and continuous detection processes can be significantly improved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor device inspection technology, and in particular to a focusing control method and apparatus for wafer morphology inspection. Background Technology

[0002] In wafer inspection equipment, the probe typically needs to move between multiple different inspection areas to perform morphological inspection, geometric measurement, or defect-related inspection of the wafer surface. To ensure inspection clarity and measurement accuracy, the system needs to be positioned near a suitable focal plane when reaching each inspection area.

[0003] Existing solutions typically employ independent point focusing or perform a large-scale axial search for each region. Independent point focusing refers to building an independent auxiliary autofocus system based on single-point ranging, separate from the main measurement optical path. It usually consists of an independent laser source, a projection optical path at a specific angle, an independent receiving sensor, and a dedicated micro-motion drive displacement stage. The most common implementation methods are oblique laser triangulation or critical angle methods. The system obliquely projects a laser beam onto a specific "point" on the wafer surface. When the wafer surface shifts vertically, the position of the reflected beam spot on the receiving sensor shifts. The system calculates the displacement of this spot, converts it into an electrical signal (focus error signal), and then drives a motor for feedback correction, ensuring that the point remains in focus. Independent point focusing has an extremely fast response speed, typically reaching the kHz level, and provides very real-time tracking and focusing for a single point, but it cannot perceive information about the "surface." If even a tiny localized dust particle, scratch, or circuit pattern shift exists at that point, focusing will fail. Furthermore, as it's additional hardware, it imposes a significant burden on optical path alignment, mechanical mounting, spatial coordinate registration, and long-term temperature drift calibration. Larger-range axial search focusing is achieved through the following steps: First, the search range is determined. Due to potential micrometer-level or even millimeter-level mechanical errors in the initial placement of the wafer on the stage, the system sets a safe axial range greater than these mechanical errors. Then, axial scanning and image acquisition drive a high-precision axial micro-motion mechanism, causing the probe or sample stage to move continuously or in steps vertically within the set axial range. During this movement, the main camera or sensor continuously captures a series of wafer surface images at different heights at a very high frame rate. Each acquired frame then undergoes real-time digital signal processing, using specific mathematical algorithms to evaluate the image's edge sharpness and high-frequency components. When the image is out of focus, the image is blurry, with fewer high-frequency signals and a low evaluation value. When the image approaches the focal plane, details are clear, contrast is high, and the evaluation value rises rapidly. Finally, after completing the large-scale axial scan, the system obtains a bell-shaped curve representing the "evaluation value - Z-axis position". By finding the maximum point of this curve, the system can accurately locate the Z-axis coordinate of the optimal focal plane, and then drive the motor to retract and precisely stop the probe at that position. While a large-scale axial search is robust, it results in repeated large-scale searches in each region, making it difficult to meet the production efficiency requirements of large-area, multi-area, and continuous inspection scenarios.

[0004] Patent CN122085477A proposes a dual Z-axis autofocus system and method based on depth vision. This scheme can handle coarse and fine adjustments using the large and small Z-axis respectively. However, it primarily plans the focus position of each region based on the depth map of the target obtained from a single pre-scan, without considering continuous detection of multiple target regions on the wafer. If continuous region detection is performed, coarse and fine adjustments are still required for each region, failing to meet the detection efficiency requirements for large-area, multi-area, and continuous detection scenarios. Furthermore, the local fine adjustment of the above scheme relies on image sharpness evaluation values. When the wafer surface texture is weak, the pattern is highly periodic, the reflectivity changes significantly, or there are local defects, the sharpness curve may be affected by the pattern and lighting conditions, resulting in low accuracy. Summary of the Invention

[0005] This invention provides a focusing control method and apparatus for wafer morphology inspection, which can effectively simplify the focusing process in multi-region wafer inspection scenarios and meet the inspection efficiency requirements in large-area, multi-region, and continuous inspection scenarios.

[0006] A focusing control method for wafer morphology inspection, comprising: The wafer to be inspected is controlled to move relative to the image acquisition device along a preset detection path. The direction of the relative movement is located in a plane parallel to the reference plane of the stage that carries the wafer to be inspected. The wafer to be inspected is divided into multiple detection areas according to the detection path. Determine whether the current detection area meets the preset conditions, and perform an initial capture operation or a prediction-guided reset operation on the current detection area based on the determination result; The initial capture operation includes: controlling the image acquisition device to perform an initial scan and a secondary scan of the current detection area along a direction perpendicular to the surface of the wafer to be inspected, and calculating the first area focusing reference; The predictive guided reset operation includes: calculating the predicted focus reference position of the current detection area based on the first regional focus reference or the historical regional focus reference in the historical prior queue, resetting and confirming the predicted focus reference position, and performing morphology detection on the wafer to be detected based on the predicted focus reference position after successful reset confirmation.

[0007] Furthermore, if the current detection area is the first detection area or the historical prior queue is empty, an initial capture operation is performed; if the current detection area is not the first detection area and the historical prior queue is not empty, a prediction-guided reset operation is performed.

[0008] Furthermore, an initial scan and a second scan are performed on the current detection area to calculate the first regional focusing reference, including: Within the first preset axial window, the image acquisition device is controlled to acquire a sequence of images of the first region at a first preset step size; Feature calculation is performed based on the image sequence of the first region to obtain coarse capture features; Based on the coarse capture features, threshold judgment and verification are performed to determine the initial detection frame; The position corresponding to the previous frame of the initial detection frame is used as the detection start position for a second scan to calculate the first region focusing reference.

[0009] Furthermore, feature calculation is performed based on the image sequence of the first region to obtain coarse capture features, including: Calculate the absolute value of the pixel difference between the current first region image and the previous frame's first region image, and sum them up to obtain the local energy of the current first region image. Calculate the absolute value of the pixel difference between the current first region image and the first region image of the first frame, and form an absolute value signal array of the current first region image; A sliding window is used to perform low-pass filtering on the absolute value signal data to obtain a weak localization feature array; Summing the values ​​in the weak localization feature array yields the lightweight envelope value of the current first region image; The local energy and lightweight envelope value are used as coarse capture features of the current first region image.

[0010] Further, based on the coarse capture features, threshold judgment and verification are performed to determine the initial detection frame, including: The first region image with local energy greater than a preset energy threshold and light envelope value greater than a preset envelope threshold is used as the frame to be confirmed. The position corresponding to the frame to be confirmed is taken as the confirmation start position, and a verification image is acquired with a second preset step size, which is smaller than the first preset step size. Calculate the first local energy of the verification image and the frame to be confirmed, and the second local energy of the verification image and the first region image of the previous frame of the frame to be confirmed; When both the first local energy and the second local energy are greater than a preset energy threshold, the frame to be confirmed is determined to be the initial detection frame.

[0011] Furthermore, the position corresponding to the previous frame of the initial detection frame is used as the detection start position for a second scan to calculate the region focusing reference, including: Starting from the detection start position, the second region image sequence is acquired with a third preset step size and height reconstruction is performed to obtain the three-dimensional coordinate set of the effective pixels in the second region image sequence. Height values ​​are extracted from the three-dimensional coordinate set of the second region image sequence. The median or height-weighted average of the height values ​​is calculated based on the height values, and the median or height-weighted average of the height values ​​is used as the representative height of the region. Based on the height value, a first-order plane fitting is performed to obtain local plane parameters; The region's representative height and local planar parameters are used as the first region's focusing reference. The third preset step size is smaller than the first preset step size.

[0012] Further, the predicted focus reference position of the current detection area is calculated based on the first regional focus reference or the historical regional focus reference in the historical prior queue, including: Obtain the center physical coordinates of the previous detection area and the current detection area, and calculate the relative translational displacement vector between the previous detection area and the current detection area; If the previous detection area of ​​the current detection area performed an initial capture operation, the predicted focus reference position of the current detection area is calculated based on the area representative height and local plane parameters in the first area focus reference and the relative translation displacement vector; If the previous detection area of ​​the current detection area did not perform the initial capture operation, the historical area focusing reference is obtained from the historical prior queue and weighted averaged to obtain the weighted average area focusing reference. The predicted focusing reference position of the current detection area is calculated based on the area representative height and local plane parameters in the weighted average area focusing reference and the relative translation displacement vector.

[0013] Furthermore, the predicted focus reference position of the current detection area is calculated based on the historical area focus benchmarks in the historical prior queue, including: Establish a polynomial global topographic trend surface model with respect to height and center physical coordinates; Obtain the physical coordinates of the center of the historical detection area within the preset sliding window, and obtain the corresponding historical area focusing reference from the historical prior queue. Substitute the area representative height and the physical coordinates of the center of the historical detection area in the historical area focusing reference into the polynomial global topography trend surface model, perform quadratic surface fitting based on the least squares method, and solve the coefficients in the polynomial global topography trend surface model. Obtain the center physical coordinates of the distant detection area that is farther from the current detection area than a preset distance, and substitute them into the polynomial global topography trend surface model for solution to obtain the predicted focus reference position of the current detection area.

[0014] Furthermore, the predicted focus reference position is reset and confirmed, including: The image acquisition device is moved to the predicted focus reference position and the image sequence of the third region is acquired within the second preset axial window with a fourth preset step size. Extract the region of interest from the third region image sequence and calculate the local energy of the region of interest; Find the position and height corresponding to the frame with the maximum local energy as the actual focus reference value; If the absolute value of the difference between the actual focus reference value and the predicted focus reference position is less than a preset threshold, then the reset is determined to be successful.

[0015] Furthermore, topography detection is performed based on the predicted focus reference position, including: The image acquisition device is controlled to move to the predicted focus reference position, and the image sequence of the fourth region is acquired with a third preset step size; Three-dimensional height reconstruction is performed based on the image sequence of the fourth region to obtain the three-dimensional reconstruction result of the current detection region. The regional focus reference for the current detection area is calculated based on the 3D reconstruction results of the current detection area and stored in the historical prior queue.

[0016] Furthermore, if the absolute value of the difference between the actual focus reference value and the predicted focus reference position is greater than or equal to a preset threshold and less than a preset safety value, then the second preset axial window is expanded to re-acquire the third region image sequence and the reset confirmation is performed again. If the reset confirmation fails again, or if the root mean square height residual of the 3D reconstruction result exceeds the preset residual threshold, the third preset step size is reduced, the third region image is acquired for the third time, and the reset confirmation is performed for the third time. If the current detection area has not undergone initial capture operation, the current detection area is divided into multiple sub-regions. Based on the 3D reconstruction results, the high continuity of the multiple sub-regions is calculated to determine the high-confidence sub-regions and calculate the proportion of the high-confidence sub-regions. If the third reset confirmation fails, or if the proportion of the high-confidence sub-region is lower than the preset proportion threshold, then the initial capture operation is performed on the current detection area.

[0017] A focusing control device for wafer morphology inspection, comprising: The motion control module is used to control the movement of the wafer under test relative to the image acquisition device along a preset detection path. The direction of the relative movement is located in a plane parallel to the reference plane of the stage that carries the wafer under test. The wafer under test is divided into multiple detection areas according to the detection path. The judgment module is used to determine whether the current detection area meets the preset conditions, and to perform an initial capture operation or a prediction-guided reset operation on the current detection area based on the judgment result. The initial capture module is used to perform the initial capture operation, including: controlling the image acquisition device to perform an initial scan and a secondary scan of the current detection area along a direction perpendicular to the surface of the wafer to be inspected, and calculating the first area focus reference; The guided reset module is used to perform a predictive guided reset operation, including: calculating the predicted focus reference position of the current detection area based on the first regional focus reference or the historical regional focus reference in the historical prior queue, resetting and confirming the predicted focus reference position, and performing morphology detection on the wafer to be detected based on the predicted focus reference position after successful reset confirmation.

[0018] Furthermore, if the current detection area is the first detection area or the historical prior queue is empty, an initial capture operation is performed; if the current detection area is not the first detection area and the historical prior queue is not empty, a prediction-guided reset operation is performed.

[0019] Furthermore, the initial acquisition module performs an initial scan and a second scan of the current detection area to calculate the first region focusing reference, including: Within the first preset axial window, the image acquisition device is controlled to acquire a sequence of images of the first region at a first preset step size; Feature calculation is performed based on the image sequence of the first region to obtain coarse capture features; Based on the coarse capture features, threshold judgment and verification are performed to determine the initial detection frame; The position corresponding to the previous frame of the initial detection frame is used as the detection start position for a second scan to calculate the first region focusing reference.

[0020] Furthermore, the initial capture module performs feature calculations based on the image sequence of the first region to obtain coarse capture features, including: Calculate the absolute value of the pixel difference between the current first region image and the previous frame's first region image, and sum them up to obtain the local energy of the current first region image. Calculate the absolute value of the pixel difference between the current first region image and the first region image of the first frame, and form an absolute value signal array of the current first region image; A sliding window is used to perform low-pass filtering on the absolute value signal data to obtain a weak localization feature array; Summing the values ​​in the weak localization feature array yields the lightweight envelope value of the current first region image; The local energy and lightweight envelope value are used as coarse capture features of the current first region image.

[0021] Further, the initial capture module performs threshold judgment and verification based on the coarse capture features to determine the initial detection frame, including: The first region image with local energy greater than a preset energy threshold and light envelope value greater than a preset envelope threshold is used as the frame to be confirmed. The position corresponding to the frame to be confirmed is taken as the confirmation start position, and a verification image is acquired with a second preset step size, which is smaller than the first preset step size. Calculate the first local energy of the verification image and the frame to be confirmed, and the second local energy of the verification image and the first region image of the previous frame of the frame to be confirmed; When both the first local energy and the second local energy are greater than a preset energy threshold, the frame to be confirmed is determined to be the initial detection frame.

[0022] Furthermore, the initial capture module uses the position corresponding to the previous frame of the initial detection frame as the detection start position for a second scan, and calculates the first region focusing reference, including: Starting from the detection start position, the second region image sequence is acquired with a third preset step size and height reconstruction is performed to obtain the three-dimensional coordinate set of the effective pixels in the second region image sequence. Height values ​​are extracted from the three-dimensional coordinate set of the second region image sequence. The median or height-weighted average of the height values ​​is calculated based on the height values, and the median or height-weighted average of the height values ​​is used as the representative height of the region. Based on the height value, a first-order plane fitting is performed to obtain local plane parameters; The region's representative height and local planar parameters are used as the first region's focusing reference. The third preset step size is smaller than the first preset step size.

[0023] Furthermore, the guidance reset module calculates the predicted focus reference position of the current detection area based on the first regional focus reference or the historical regional focus reference in the historical prior queue, including: Obtain the center physical coordinates of the previous detection area and the current detection area, and calculate the relative translational displacement vector between the previous detection area and the current detection area; If the previous detection area of ​​the current detection area performed an initial capture operation, the predicted focus reference position of the current detection area is calculated based on the area representative height and local plane parameters in the first area focus reference and the relative translation displacement vector; If the previous detection area of ​​the current detection area did not perform the initial capture operation, the historical area focusing reference is obtained from the historical prior queue and weighted averaged to obtain the weighted average area focusing reference. The predicted focusing reference position of the current detection area is calculated based on the area representative height and local plane parameters in the weighted average area focusing reference and the relative translation displacement vector.

[0024] Furthermore, the guidance reset module calculates the predicted focus reference position of the current detection area based on the historical area focus benchmark in the historical prior queue, including: Establish a polynomial global topographic trend surface model with respect to height and center physical coordinates; Obtain the physical coordinates of the center of the historical detection area within the preset sliding window, and obtain the corresponding historical area focusing reference from the historical prior queue. Substitute the area representative height and the physical coordinates of the center of the historical detection area in the historical area focusing reference into the polynomial global topography trend surface model, perform quadratic surface fitting based on the least squares method, and solve the coefficients in the polynomial global topography trend surface model. Obtain the center physical coordinates of the distant detection area that is farther from the current detection area than a preset distance, and substitute them into the polynomial global topography trend surface model for solution to obtain the predicted focus reference position of the current detection area.

[0025] Furthermore, the guide reset module performs a reset confirmation of the predicted focus reference position, including: The image acquisition device is moved to the predicted focus reference position and the image sequence of the third region is acquired within the second preset axial window with a fourth preset step size. Extract the region of interest from the third region image sequence and calculate the local energy of the region of interest; Find the position and height corresponding to the frame with the maximum local energy as the actual focus reference value; If the absolute value of the difference between the actual focus reference value and the predicted focus reference position is less than a preset threshold, then the reset is determined to be successful.

[0026] Furthermore, the fine-focus module performs topography detection based on the predicted focus reference position, including: The image acquisition device is controlled to move to the predicted focus reference position, and the image sequence of the fourth region is acquired with a third preset step size; Three-dimensional height reconstruction is performed based on the image sequence of the fourth region to obtain the three-dimensional reconstruction result of the current detection region. Calculate the regional focusing reference of the current detection area based on the 3D reconstruction results of the current detection area and store it in the historical prior queue; The fourth preset step size is smaller than the first preset step size.

[0027] Furthermore, if the absolute value of the difference between the actual focus reference value and the predicted focus reference position is greater than or equal to a preset threshold and less than a preset safety value, then the second preset axial window is expanded to re-acquire the third region image sequence and the reset confirmation is performed again. If the reset confirmation fails again, or if the root mean square height residual of the 3D reconstruction result exceeds the preset residual threshold, the third preset step size is reduced, the third region image is acquired for the third time, and the reset confirmation is performed for the third time. If the current detection area has not undergone initial capture operation, the current detection area is divided into multiple sub-regions. Based on the 3D reconstruction results, the high continuity of the multiple sub-regions is calculated to determine the high-confidence sub-regions and calculate the proportion of the high-confidence sub-regions. If the third reset confirmation fails, or if the proportion of the high-confidence sub-region is lower than the preset proportion threshold, then the initial capture operation is performed on the current detection area.

[0028] The focusing control method and apparatus for wafer morphology inspection provided by the present invention have at least the following beneficial effects: (1) The focusing tasks of the first detection and subsequent detection are processed in different modes. The regional focusing benchmark is established through the first detection to provide prior information for subsequent detection. This avoids repeatedly performing large-scale axial searches in all regions, which can significantly improve the detection efficiency in large-area, multi-area, and continuous detection processes. (2) The prior knowledge of the surface shape of the preceding region guides the rapid reset of the subsequent region, so that the system does not need to start from zero to find the position in each region. Furthermore, since the prediction focus reference comes from the extrapolation of the region's representative height and / or local planar parameters, the reset results of the subsequent region have a clearer source of technical data. (3) Based on the area surface information rather than single point value, the focus reference is generated. It can improve the robustness to edge structure, local anomalies and reflection fluctuations by using the area representative height parameter, local plane fitting and high confidence sub-region screening, and reduce the probability of misfocusing caused by single point distortion. Local fine adjustment does not depend on sharpness and has higher accuracy. (4) The same surface measurement link simultaneously serves topography detection, focus reference generation and subsequent prediction update, reducing the independent point-type focus module and the resulting opto-mechanical integration and calibration burden, and reducing the coordinate registration requirements between the independent focus link and the topography detection link. (5) It has the ability to roll back from abnormal situations, and can automatically expand the search or recapture when a failure or local anomaly is predicted, thus balancing speed and reliability. Attached Figure Description

[0029] Figure 1 This is a flowchart of one embodiment of the focusing control method for wafer morphology detection provided by the present invention.

[0030] Figure 2 This is a schematic diagram illustrating the relationship between the image acquisition device and the wafer to be inspected in the focusing control method for wafer morphology inspection provided by the present invention.

[0031] Figure 3 This is a schematic diagram of the effective envelope width of a white light interferometric image for the focusing control method for wafer morphology detection provided by the present invention.

[0032] Figure 4 This is a schematic diagram illustrating the initial capture operation process in the focusing control method for wafer morphology detection provided by the present invention.

[0033] Figure 5This is a schematic diagram illustrating the reset confirmation process in the focusing control method for wafer morphology detection provided by the present invention.

[0034] Figure 6 This is a schematic diagram illustrating the morphology detection process in the focusing control method for wafer morphology detection provided by the present invention.

[0035] Figure 7 This is a schematic diagram of one embodiment of the focusing control device for wafer morphology detection provided by the present invention. Detailed Implementation

[0036] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0037] refer to Figure 1 In some embodiments, a focusing control method for wafer morphology inspection is provided, comprising: S1. Control the wafer to be inspected to move relative to the image acquisition device along a preset detection path. The direction of the relative movement is located in a plane parallel to the reference plane of the stage carrying the wafer to be inspected. The wafer to be inspected is divided into multiple detection areas according to the detection path. S2. Determine whether the current detection area meets the preset conditions, and perform an initial capture operation or a prediction-guided reset operation on the current detection area based on the judgment result. S3. The initial capture operation includes: controlling the image acquisition device to perform an initial scan and a secondary scan of the current detection area along a direction perpendicular to the surface of the wafer to be inspected, and calculating the first area focusing reference; S4. The predicted guided reset operation includes: calculating the predicted focus reference position of the current detection area based on the first regional focus reference or the historical regional focus reference in the historical prior queue, resetting and confirming the predicted focus reference position, and performing morphology detection on the wafer to be detected based on the predicted focus reference position after successful reset confirmation.

[0038] Specifically, refer to Figure 2 In step S1, the wafer 2 to be inspected is placed horizontally and fixed on the stage. The stage is a multi-degree-of-freedom motion platform, which can drive the wafer to be inspected to perform step-by-step or continuous translational motion in a horizontal two-dimensional plane (i.e., the XY direction) according to a preset "S" shaped trajectory, so that the image acquisition device 1 can sequentially align with different detection areas.

[0039] Let the size of the wafer 2 to be inspected be N mm * M mm, the effective size of a single area acquired by the image acquisition device 1 be m rows * n columns, and the resolution of the image acquisition device be dpi mm / pixel. Then the total number of inspection areas num that need to be acquired is: (1) The image acquisition device is vertically suspended above the stage and integrates an area array camera, an optical imaging link, and a motion control mechanism perpendicular to the surface of the wafer to be inspected (i.e., the axial direction). To balance the large macroscopic search range and microscopic nanoscale measurement accuracy required for focusing, the image acquisition device incorporates both a large-stroke axial actuator and a small-stroke precision actuator along the axial direction.

[0040] Large-stroke axial actuator: It has a millimeter-level motion stroke, and its adjustment range corresponds to the first preset axial window. It is used to perform large-scale surface search and capture in the initial capture operation, macroscopic coarse positioning and reset of the subsequent area, and wide-window recapture in abnormal backtracking mode.

[0041] Small-stroke precision actuator: It has a micron-level motion stroke and a nanometer-level motion resolution. Its adjustment range corresponds to the second preset axial window. It is superimposed on the motion end of the large-stroke axial actuator or independently drives the core optical element. It is used to perform local interference signal confirmation and / or fine axial micro-motion scanning after the initial capture operation is completed, so as to achieve high-precision three-dimensional topography reconstruction.

[0042] Through a pre-defined mapping relationship, the displacement of the large-stroke axial actuator and the small-stroke precision actuator are unified into the same axial focusing reference coordinate system, realizing hierarchical closed-loop coordination at the physical level.

[0043] In the actual image acquisition process, the overall collaborative workflow is as follows: First, the wafer to be inspected is fixed on the stage. For the first inspection area, the stage remains horizontal and stationary. The image acquisition device utilizes a large-stroke axial actuator and a small-stroke precision actuator in a layered, coordinated manner to perform a large-scale, graded scan along the axis perpendicular to the wafer surface and continuously acquire a sequence of two-dimensional images. After completing the data reconstruction of the current inspection area, the image acquisition device is reset axially to the designated safe range or predicted starting position. Subsequently, the stage moves stepwise in the horizontal plane, moving the wafer to be inspected and aligning the image acquisition device with the next inspection area. During this process, the overall movement trajectory of the image acquisition device relative to the wafer on the horizontal surface is planned as an "S" shape. Upon reaching the subsequent inspection area, the stage remains stationary. Based on the predicted reference generated from the previous sequence, the image acquisition device performs a rapid scan along the axis within a small stroke range to confirm and acquire images, and then performs an axial reset again. The system cycles through "horizontal stepping adjustment of the stage - axial focusing and scanning of the image acquisition device and image acquisition - axial reset" in this manner until the continuous detection of multiple areas of the entire wafer to be inspected is completed.

[0044] Furthermore, in step S2, the preset conditions are whether the current detection area is the first detection area and whether the historical prior queue is empty. When the current detection area is the first detection area or the historical prior queue is empty, the initial capture operation is performed; when the current detection area is not the first detection area and the historical prior queue is not empty, the prediction-guided reset operation is performed.

[0045] Further, refer to Figure 4 In step S3, an initial capture operation is performed, controlling the image acquisition device to perform an initial scan and a secondary scan of the current detection area along a direction perpendicular to the surface of the wafer to be inspected, and calculating the first region focusing reference, including: S31. Within the first preset axial window, control the image acquisition device to acquire the first region image sequence at a first preset step size; S32. Perform feature calculation based on the image sequence of the first region to obtain coarse capture features; S33. Based on the coarse capture features, perform threshold judgment and verification to determine the initial detection frame; S34. Take the position corresponding to the previous frame of the initial detection frame as the detection start position and perform a second scan to calculate the first region focusing reference.

[0046] Specifically, in step S31, a large-scale scan is performed on the current detection area within the first preset axial window, and a first region image sequence of the detection area is obtained by using white light interferometry or other methods. Each first region image in the first region image sequence records the intensity of each pixel at the current axial position. Taking white light interferometry as an example, when the probe scans along the axial direction, the light intensity of each pixel will change with the axial position to form an interference envelope curve.

[0047] Initially, due to the lack of focus, a large axial distance is maintained between the image acquisition device and the wafer under inspection, resulting in the image acquisition device being out of focus. However, when the image acquisition device begins to perform a large-scale scan along the axial direction, the surface of the wafer under inspection will enter the depth of field region near the focal plane of the image acquisition device, and the image acquisition device will capture a clear image. If a white light interference image acquisition device is used, black and white interference fringes will appear on the image.

[0048] When the current detection area is in the initial capture operation, a relatively large initial distance is set between the image acquisition device and the surface of the wafer to be detected. Then, the image acquisition device performs a large-scale scan of the current detection area within the first preset axial window with a specific first preset step size and along the direction perpendicular to the surface of the wafer to be detected (i.e., the axial direction).

[0049] The first preset step size is larger than the formal topography detection step size (the third preset step size). The formal topography detection step size is strictly determined by the system's optical hardware parameters and the final measurement accuracy requirements, and is a known constant during system initialization. Taking white light interferometry or phase-shifting interferometry as an example, in order to meet the mathematical orthogonality requirements of phase-shifting demodulation algorithms (such as the five-step phase-shifting method), the formal topography detection step size is usually precisely set to 1 / 8 of the effective light source center wavelength (for example, if the center wavelength λ = 600nm, then the formal topography detection step size is fixed at 75nm), thereby ensuring nanometer-level 3D reconstruction accuracy. To ensure that the effective envelope width is not exceeded during the initial acquisition operation, such as... Figure 3 As shown, space is left for the confirmation step size, and the first preset step size must be less than half of the effective envelope width.

[0050] Further, in step S32, feature calculation is performed based on the first region image sequence to obtain coarse capture features, including: S321. Calculate the absolute value of the pixel difference between the current first region image and the previous frame's first region image and sum them up as the local energy of the current first region image. S322. Calculate the absolute value of the pixel difference between the current first region image and the first region image of the first frame, and form an absolute value signal array of the current first region image; S323. A sliding window is used to perform low-pass filtering on the absolute value signal data to obtain a weak localization feature array; S324. Sum the values ​​in the weak localization feature array to obtain the lightweight envelope value of the current first region image; S325. The local energy and lightweight envelope value are used as coarse capture features of the current first region image.

[0051] Specifically, in step S321, the local energy is the sum of the absolute values ​​of the pixel differences between two adjacent frames of the first region image. Let the current first region image be I(k), where I(k) is a two-dimensional array of size m rows * n columns, and the pixel value of the point with coordinates (x, y) on I(k) is I(x, y, k). The previous frame of the first region image is I(k-1), where I(k-1) is a two-dimensional array of size m * n, and the pixel value of the point with coordinates (x, y) on I(k-1) is I(x, y, k-1), where k is the scan frame number. The formula for calculating the local energy E(k) of the current first region image is: (2) Where (x, y) are pixel coordinates. When approaching the surface, surface texture or coherent fringes appear, the high-frequency content of the image increases, and the local energy will increase dramatically, which is recognized as capturing surface features.

[0052] Further, in step S322, the original white light interference signal is superimposed on a strong ambient background light. In order to accurately extract the energy of the interference fringes later, since the first frame of the first region image I(0) captured by coarse acquisition is definitely not within the white light interference signal, the first frame of the first region image I(0) is taken as the DC component and subtracted from the original signal, so that the interference fringes can oscillate perfectly around the zero mark. The absolute value of the signal after the mean is removed is taken, and all the negative valleys are flipped to positive values ​​to prepare for envelope extraction, thus obtaining the absolute value signal array I of the current first region image. abs (k): (3) In step S323, a simplified sliding window is used to perform low-pass filtering on the absolute value signal array. A sliding window with a width of W is set, and the absolute value signal array I is filtered vertically. abs (k) Perform local averaging. This step flattens out the dense, spike-like absolute value stripes, merging them into a gentle, hill-like curve, resulting in the weak localization feature array A(k) of the current first region image: (4) in: (5) In step S324, the values ​​in the weak localization feature array are summed to obtain the lightweight envelope value H(k) of the current first region image; (6) Coarse capture features include local energy E(k) and lightweight envelope value H(k).

[0053] Further, in step S33, threshold judgment and verification are performed based on the coarse capture features to determine the initial detection frame, including: S331. The first region image with local energy greater than a preset energy threshold and light envelope value greater than a preset envelope threshold is taken as the frame to be confirmed. S332. Take the position corresponding to the frame to be confirmed as the confirmation start position, and collect a frame of verification image with a second preset step size. The second preset step size is less than the first preset step size and greater than the third preset step size. S333. Calculate the first local energy of the verification image and the frame to be confirmed, and the second local energy of the verification image and the first region image of the previous frame of the frame to be confirmed. S334. When both the first local energy and the second local energy are greater than the preset energy threshold, the frame to be confirmed is determined to be the initial detection frame.

[0054] Specifically, in step S331, when the local energy E(k) is detected to be greater than the preset energy threshold and the light envelope value H(k) is greater than the preset envelope threshold, it is considered that the first region image has obvious coarse capture features, the frame number of the first region image is recorded as k0, and it is considered as a frame to be confirmed. It is initially considered that the frame has entered the effective envelope width.

[0055] In step S332, the position corresponding to the frame to be confirmed k0 is taken as the confirmation start position, and a verification image k0+1 is acquired with a second preset step size that is smaller than the first preset step size, wherein the second preset step size is larger than the formal morphology detection step size (the third preset step size).

[0056] In step S33, the first local energy E1(k0+1) of the verification image and the frame to be confirmed, and the second local energy E2(k0+1) of the verification image and the first region image k0-1 of the previous frame of the frame to be confirmed are calculated: (7) (8) In step S34, if both the first local energy and the second local energy are greater than the preset energy threshold, then the k0th frame is considered to have entered the effective envelope width, and the frame to be confirmed is taken as the initial detection frame; otherwise, the camera continues to acquire images with a coarse acquisition step size and simultaneously calculates coarse acquisition features.

[0057] Further, in step S34, the position corresponding to the previous frame of the initial detection frame is used as the detection start position for a second scan to calculate the region focusing reference, including: S341. Starting from the detection start position, acquire the second region image sequence with a third preset step size and perform height reconstruction, and obtain the three-dimensional coordinate set of the effective pixels of the second region image sequence. S342. Extract height values ​​from the three-dimensional coordinate set of the second region image sequence, calculate the median or height-weighted average of the height values ​​based on the height values, and use the median or height-weighted average of the height values ​​as the representative height of the region; S343. Perform first-order plane fitting based on the height value to obtain local plane parameters; S344. Use the region's representative height and local plane parameters as the first region's focusing reference.

[0058] The third preset step size is smaller than the first preset step size.

[0059] In step S341, the image acquisition device returns to the position of the k0-1 frame, that is, the position corresponding to the previous frame of the initial detection frame is taken as the detection start position, and a second scan is performed to acquire the second region image sequence with the third preset step size. Taking white light interference as an example, when scanning along the axis, the light intensity of each pixel will change with the axial position to form an interference envelope curve. By finding the axial mechanical coordinates corresponding to the peak position of the interference envelope of the pixel, the height value corresponding to the pixel can be calculated, and then the three-dimensional coordinate set (x, y, z) of the effective pixels of the second region image sequence can be obtained; the third preset step size is the formal morphology detection step size.

[0060] The set of three-dimensional coordinates of the effective pixels in the second region image sequence obtained in step S341 is the three-dimensional reconstruction result of the current detection region.

[0061] Further, in step S342, it is assumed that the set of three-dimensional coordinates of the second region image sequence is... Where K is the total number of valid pixels in the current target area whose height values ​​were successfully reconstructed. Two methods are used to obtain the representative height of the area: Option 1: Select the median height value.

[0062] If the median height of all valid reconstructed pixels is selected as the representative height z of the region. rep The calculation formula is as follows: (9) Specifically, if the height values ​​of all valid pixels are... Rearrange them in ascending order to obtain an ascending sequence. ,satisfy Then the piecewise explicit expression for the median z_{rep} is: (10) Option 2: Select the height-weighted average.

[0063] If the height-weighted average value is selected as the representative height z of the region rep The calculation formula is as follows: (11) Among them, w i Let w be the weight coefficient for the i-th pixel. In one specific embodiment, this weight coefficient can be positively correlated with the underlying confidence index or coherent signal modulation of the pixel, used to naturally filter out noise with low signal-to-noise ratio. Specifically, when w i When = 1, the above formula degenerates into the arithmetic mean formula: (12) Furthermore, in step S343, this embodiment provides a method for obtaining local plane parameters (A, B, C), and the objective equation for first-order plane fitting is: (13) To solve for the optimal coefficient vector Construct the target residual sum of squares function using the least squares method. To bring it to a minimum: (14) According to the necessary condition for finding the extrema of a multivariable function, let the partial derivatives of F with respect to parameters A, B, and C be equal to 0: (15) Expanding and rearranging the above equations, we obtain the normal equations for the unknown local plane parameters (A, B, C): (16) In engineering calculations, the explicit analytical solution of the local plane parameters (A, B, C) can be obtained directly by matrix inversion. The solved (A, B) represent the local physical slopes of the region in the X and Y directions, respectively, and C is the intercept of the fitted plane in the Z-axis direction.

[0064] The region represents the height z rep Together with at least some of the local plane parameters (A, B, C), they serve as the first regional focusing reference, providing prior information for subsequent detection of the region.

[0065] Furthermore, in step S4, two methods are provided to calculate the predicted focus reference position of the current detection area.

[0066] Option 1: First-order linear extrapolation scheme based on historical regions.

[0067] This solution is suitable for scenarios where the local continuity of the wafer inspection path is relatively strong and the height change between adjacent inspection areas is relatively gentle.

[0068] Specifically, the predicted focus reference position of the current detection area is calculated based on the first regional focus reference or the historical regional focus reference in the historical prior queue, including: S41. Obtain the center physical coordinates of the previous detection area and the current detection area, and calculate the relative translational displacement vector between the previous detection area and the current detection area; S42. If the previous detection area of ​​the current detection area performed an initial capture operation, the predicted focus reference position of the current detection area is calculated based on the area representative height and local plane parameters in the first area focus reference and the relative translation displacement vector. S43. If the previous detection area of ​​the current detection area did not perform an initial capture operation, then the historical area focusing reference is obtained from the historical prior queue and weighted averaged to obtain the weighted average area focusing reference. The predicted focusing reference position of the current detection area is calculated based on the area representative height and local plane parameters in the weighted average area focusing reference, as well as the relative translation displacement vector. The weights can be determined based on the root mean square residual height of the corresponding detection area's 3D reconstruction result.

[0069] Specifically, let the physical coordinates of the center of the previous adjacent detection area be (x1, y1), and the physical coordinates of the center of the current detection area be (x2, y2). The relative planar displacement vector between the centers of the two areas in physical space is: (17) (18) Furthermore, if the previous detection area performed an initial capture operation, the region representative height and local plane parameters in the first region focusing reference are obtained. If the previous detection area did not perform an initial capture operation, historical region focusing references are obtained from the historical prior queue and weighted averaged to obtain a weighted average region focusing reference. The local plane parameters in the weighted average region focusing reference are (A1, B1, C1), where (A1, B1) are the local slopes in the X and Y directions, respectively, and the region representative height z in the weighted average region focusing reference is... rep As a reference term, with local plane parameters As a correction term, the predicted focus reference position Z at the center of the current detection area is calculated using the first-order differential formula. predict : (19) The predicted focus reference position Z calculated in this way predict As the predicted center location of the current detection area.

[0070] Option 2: Path-level fast reset solution based on multi-region trend model.

[0071] This solution is suitable for scenarios where large-size wafers experience macroscopic warping (such as basin-shaped or bow-shaped deformation) due to internal stress or mechanical clamping.

[0072] Specifically, the predicted focus reference position of the current detection area is calculated based on the historical area focus benchmarks in the historical prior queue, including: S4A: Establish a polynomial global topographic trend surface model with respect to height and center physical coordinates; S4B: Obtain the physical coordinates of the center of the historical detection area within the preset sliding window, and obtain the corresponding historical area focusing reference from the historical prior queue. Substitute the area representative height and the physical coordinates of the center of the historical detection area in the historical area focusing reference into the polynomial global morphology trend surface model, perform quadratic surface fitting based on the least squares method, and solve the coefficients in the polynomial global morphology trend surface model. S4C: Obtain the center physical coordinates of the distant detection area that is farther from the current detection area than a preset distance, and substitute them into the polynomial global topography trend surface model for solution to obtain the predicted focus reference position of the current detection area.

[0073] Specifically, as the "S"-shaped detection path advances, the system maintains a preset sliding window of length K in memory, continuously accumulating the center physical coordinates (x, y, x) of the K most recently successfully detected historical detection areas. i ,y i ) and its corresponding region representing height z i (in ).

[0074] Construct a polynomial global topographic trend surface model: (20) Using these K sets of historical data, the coefficient matrix is ​​dynamically solved and updated in real time through least-squares quadratic surface fitting. .

[0075] Obtain the center physical coordinates (x, y) of the distant detection area that is at a distance greater than a preset distance from the current detection area. M ,y M Substituting these values ​​into the dynamically updated trend surface model equations, the predicted focus reference position Z of the current detection area can be directly calculated. predict : ;(twenty one) Further, refer to Figure 5 The predicted focus reference position is reset and confirmed, including: S44. Control the image acquisition device to move to the predicted focus reference position, and acquire the image sequence of the third region within the second preset axial window with a fourth preset step size. S45. Extract the region of interest from the third region image sequence and calculate the local energy of the region of interest; S46. Find the position height corresponding to the frame with the maximum local energy as the actual focus reference value; S47. If the absolute value of the difference between the actual focus reference value and the predicted focus reference position is less than a preset threshold, then the reset is determined to be successful.

[0076] The fourth preset step size is greater than the third preset step size, but less than the first preset step size.

[0077] Specifically, during system initialization, the displacements of different actuators, such as the Z-axis adjustment mechanism of the image acquisition device and the Z-axis adjustment mechanism of the stage, are unified into the same focusing reference coordinate system through a pre-calibrated mapping relationship.

[0078] The system first drives the large-stroke axial actuator to quickly move the current detection area to near the predicted focus reference position, ensuring that the remaining axial mechanical cumulative error falls completely within the second preset axial window of the small-stroke precision actuator. At this point, the large-stroke axial actuator is kept stationary, while the small-stroke precision actuator is positioned within a preset optimal working range.

[0079] Within the second preset axial window corresponding to the adjustment range of the small-stroke precision actuator, the small-stroke precision actuator performs a fine axial scan with a short stroke and a small number of samples around the predicted focus reference position. At this time, the trigger frequency of the image acquisition device or the third preset step size can typically be larger than the actual topography detection step size. During the movement, a very small number of images, ranging from a few to dozens, of the third region are continuously captured.

[0080] At this point, instead of traversing all millions of pixels in the image, one or several sparse regions of interest are directly extracted from the third region image. For continuously acquired third region image sequences, the sum of the absolute values ​​of the pixel differences between adjacent frames of the third region image sequence within the region of interest is calculated in real time, which is the local energy mentioned earlier. As the small-stroke mechanism moves slightly, the surface of the wafer to be inspected passes through the focal plane. When the local energy within the region of interest reaches its maximum value, the position height corresponding to that frame image is calculated, i.e., the small-stroke Z-axis mechanical coordinate, which is directly used as the actual focusing reference value Z. actual_rep This hierarchical coordination strategy can effectively avoid the problem of missed signal capture caused by excessive minimum effective step of large-stroke axial actuators, repeated positioning errors, or dynamic residual vibration.

[0081] If the actual focus reference value Z in the current detection area actual_rep With the predicted focus reference position Z predict The absolute deviation is within the preset threshold Z limit Within, that is, satisfied: ;(twenty two) The system then determines that the rapid reset of the current target area was successful.

[0082] Further, refer to Figure 6 In step S4, topography detection is performed based on the predicted focus reference position, including: S48. Control the image acquisition device to move to the predicted focus reference position, and acquire the image sequence of the fourth region with a third preset step size; S49. Perform three-dimensional height reconstruction based on the image sequence of the fourth region to obtain the three-dimensional reconstruction result of the current detection region; S410. Calculate the regional focusing reference of the current detection area based on the three-dimensional reconstruction results of the current detection area, and store it in the historical prior queue. The fourth preset step size is smaller than the first preset step size.

[0083] Specifically, after a successful reset, the image acquisition device is controlled by a small-stroke precision actuator to start the formal morphological detection stage. The small-stroke precision actuator is controlled by a fourth preset step size that meets the final accuracy requirements to perform fine axial scanning and image acquisition, obtain a high-precision fourth region image sequence, and reconstruct the three-dimensional height information of the current detection area to obtain the three-dimensional reconstruction result of the current detection area, that is, to obtain the three-dimensional coordinate set of the effective pixels in the entire field of view.

[0084] The 3D reconstruction results of the fourth region image sequence are reused, and the region focusing reference of the current detection region is calculated using steps S342 to S344 and stored in the historical prior queue.

[0085] Furthermore, if the reset fails, that is, if the absolute value of the difference between the actual focus reference value and the predicted focus reference position is greater than or equal to the preset threshold and less than the preset safety value, the second preset axial window is expanded to acquire the third region image sequence again and the reset is confirmed again. If the reset confirmation fails again, or if the root mean square height residual of the 3D reconstruction result exceeds the preset residual threshold, it indicates that the wafer surface morphology is too complex or the measurement noise is too large. Reduce the third preset step size, acquire the third region image for the third time, and perform a third reset confirmation. If the current detection area has not undergone initial capture operation, the current detection area is divided into multiple sub-regions. Based on the 3D reconstruction results, the high continuity of the multiple sub-regions is calculated to determine the high-confidence sub-regions and calculate the proportion of the high-confidence sub-regions. If the third reset confirmation fails, or if the proportion of the high-confidence sub-region is lower than the preset proportion threshold, then the initial capture operation is performed on the current detection area.

[0086] Among them, the height continuity calculation can be the variance of the height difference between adjacent pixels, and the sub-regions with a variance less than a set threshold and a continuous and unbroken surface are selected as "high confidence sub-regions".

[0087] The current detection area cannot extract any valid high-confidence sub-regions, meaning the proportion of high-confidence sub-regions that pass the screening is lower than a preset proportion threshold. hour: ;(twenty three) This means that the current detection area may be located at a wafer edge fracture, or there may be local strong edge interference and severe defects, causing the prior knowledge to completely fail. At this point, the system completely abandons the current prediction-guided reset operation and forcibly re-enters the initial capture operation for the current detection area. The image acquisition device returns to a more distant initial distance and re-executes the adaptive scan within the first preset axial window to re-establish the area focusing reference from zero.

[0088] If the root mean square height residual of the 3D reconstruction result of the current detection area exceeds the preset residual threshold, it indicates that the surface morphology of the current detection area is very complex or has a lot of noise. On the one hand, the third preset step size is reduced to acquire the third area image for the third time and the third reset confirmation is performed. On the other hand, when it is used as the historical area focusing reference in the future, its weight is reduced in step S43.

[0089] refer to Figure 7 In some embodiments, a focusing control device for wafer morphology inspection is provided, comprising: The motion control module 201 is used to control the wafer to be inspected to move relative to the image acquisition device along a preset detection path. The direction of the relative movement is located in a plane parallel to the reference plane of the stage that carries the wafer to be inspected. The wafer to be inspected is divided into multiple detection areas according to the detection path. The judgment module 202 is used to determine whether the current detection area meets the preset conditions, and to perform an initial capture operation or a prediction-guided reset operation on the current detection area based on the judgment result. The initial capture module 203 is used to perform the initial capture operation, including: controlling the image acquisition device to perform an initial scan and a secondary scan of the current detection area along a direction perpendicular to the surface of the wafer to be inspected, and calculating the first area focusing reference; The guided reset module 204 is used to perform a predictive guided reset operation, including: calculating the predicted focus reference position of the current detection area based on the first regional focus reference or the historical regional focus reference in the historical prior queue, resetting and confirming the predicted focus reference position, and performing morphology detection on the wafer to be detected based on the predicted focus reference position after successful reset confirmation.

[0090] Furthermore, if the current detection area is the first detection area or the historical prior queue is empty, an initial capture operation is performed; if the current detection area is not the first detection area and the historical prior queue is not empty, a prediction-guided reset operation is performed.

[0091] Furthermore, the initial acquisition module 203 performs an initial scan and a secondary scan of the current detection area, and calculates the first regional focusing reference, including: Within the first preset axial window, the image acquisition device is controlled to acquire a sequence of images of the first region at a first preset step size; Feature calculation is performed based on the image sequence of the first region to obtain coarse capture features; Based on the coarse capture features, threshold judgment and verification are performed to determine the initial detection frame; The position corresponding to the previous frame of the initial detection frame is used as the detection start position for a second scan to calculate the first region focusing reference.

[0092] Furthermore, the initial capture module 203 performs feature calculations based on the first region image sequence to obtain coarse capture features, including: Calculate the absolute value of the pixel difference between the current first region image and the previous frame's first region image, and sum them up to obtain the local energy of the current first region image. Calculate the absolute value of the pixel difference between the current first region image and the first region image of the first frame, and form an absolute value signal array of the current first region image; A sliding window is used to perform low-pass filtering on the absolute value signal data to obtain a weak localization feature array; Summing the values ​​in the weak localization feature array yields the lightweight envelope value of the current first region image; The local energy and lightweight envelope value are used as coarse capture features of the current first region image.

[0093] Further, the initial capture module 203 performs threshold judgment and verification based on the coarse capture features to determine the initial detection frame, including: The first region image with local energy greater than a preset energy threshold and light envelope value greater than a preset envelope threshold is used as the frame to be confirmed. The position corresponding to the frame to be confirmed is taken as the confirmation start position, and a verification image is acquired with a second preset step size, which is smaller than the first preset step size. Calculate the first local energy of the verification image and the frame to be confirmed, and the second local energy of the verification image and the first region image of the previous frame of the frame to be confirmed; When both the first local energy and the second local energy are greater than a preset energy threshold, the frame to be confirmed is determined to be the initial detection frame.

[0094] Furthermore, the initial capture module uses the position corresponding to the previous frame of the initial detection frame as the detection start position for a second scan, and calculates the first region focusing reference, including: Starting from the detection start position, the second region image sequence is acquired with a third preset step size and height reconstruction is performed to obtain the three-dimensional coordinate set of the effective pixels in the second region image sequence. Height values ​​are extracted from the three-dimensional coordinate set of the second region image sequence. The median or height-weighted average of the height values ​​is calculated based on the height values, and the median or height-weighted average of the height values ​​is used as the representative height of the region. Based on the height value, a first-order plane fitting is performed to obtain local plane parameters; The region's representative height and local planar parameters are used as the first region's focusing reference. The third preset step size is smaller than the first preset step size.

[0095] Furthermore, the guide reset module 204 calculates the predicted focus reference position of the current detection area based on the first regional focus reference or the historical regional focus reference in the historical prior queue, including: Obtain the center physical coordinates of the previous detection area and the current detection area, and calculate the relative translational displacement vector between the previous detection area and the current detection area; If the previous detection area of ​​the current detection area performed an initial capture operation, the predicted focus reference position of the current detection area is calculated based on the area representative height and local plane parameters in the first area focus reference and the relative translation displacement vector; If the previous detection area of ​​the current detection area did not perform the initial capture operation, the historical area focusing reference is obtained from the historical prior queue and weighted averaged to obtain the weighted average area focusing reference. The predicted focusing reference position of the current detection area is calculated based on the area representative height and local plane parameters in the weighted average area focusing reference and the relative translation displacement vector.

[0096] Furthermore, the guide reset module 204 calculates the predicted focus reference position of the current detection area based on the historical area focus reference in the historical prior queue, including: Establish a polynomial global topographic trend surface model with respect to height and center physical coordinates; Obtain the physical coordinates of the center of the historical detection area within the preset sliding window, and obtain the corresponding historical area focusing reference from the historical prior queue. Substitute the area representative height and the physical coordinates of the center of the historical detection area in the historical area focusing reference into the polynomial global topography trend surface model, perform quadratic surface fitting based on the least squares method, and solve the coefficients in the polynomial global topography trend surface model. Obtain the center physical coordinates of the distant detection area that is farther from the current detection area than a preset distance, and substitute them into the polynomial global topography trend surface model for solution to obtain the predicted focus reference position of the current detection area.

[0097] Furthermore, the guide reset module 204 resets and confirms the predicted focus reference position, including: The image acquisition device is moved to the predicted focus reference position and the image sequence of the third region is acquired within the second preset axial window with a fourth preset step size. Extract the region of interest from the third region image sequence and calculate the local energy of the region of interest; Find the position and height corresponding to the frame with the maximum local energy as the actual focus reference value; If the absolute value of the difference between the actual focus reference value and the predicted focus reference position is less than a preset threshold, then the reset is determined to be successful.

[0098] Furthermore, the guide reset module 204 performs topography detection based on the predicted focus reference position, including: The image acquisition device is controlled to move to the predicted focus reference position, and the image sequence of the fourth region is acquired with a third preset step size; Three-dimensional height reconstruction is performed based on the image sequence of the fourth region to obtain the three-dimensional reconstruction result of the current detection region. Calculate the regional focusing reference of the current detection area based on the 3D reconstruction results of the current detection area and store it in the historical prior queue; The fourth preset step size is smaller than the first preset step size.

[0099] Furthermore, if the absolute value of the difference between the actual focus reference value and the predicted focus reference position is greater than or equal to a preset threshold and less than a preset safety value, then the second preset axial window is expanded to re-acquire the third region image sequence and the reset confirmation is performed again. If the reset confirmation fails again, or if the root mean square height residual of the 3D reconstruction result exceeds the preset residual threshold, the third preset step size is reduced, the third region image is acquired for the third time, and the reset confirmation is performed for the third time. If the current detection area has not undergone initial capture operation, the current detection area is divided into multiple sub-regions. Based on the 3D reconstruction results, the high continuity of the multiple sub-regions is calculated to determine the high-confidence sub-regions and calculate the proportion of the high-confidence sub-regions. If the third reset confirmation fails, or if the proportion of the high-confidence sub-region is lower than the preset proportion threshold, then the initial capture operation is performed on the current detection area.

[0100] The focusing control method and apparatus for wafer morphology inspection provided in the above embodiments have at least the following beneficial effects: (1) The focusing tasks of the first detection and subsequent detection are processed in different modes. The regional focusing benchmark is established through the first detection to provide prior information for subsequent detection. This avoids repeatedly performing large-scale axial searches in all regions, which can significantly improve the detection efficiency in large-area, multi-area, and continuous detection processes. (2) The prior knowledge of the surface shape of the preceding region guides the rapid reset of the subsequent region, so that the system does not need to start from zero to find the position in each region. Furthermore, since the prediction focus reference comes from the extrapolation of the region's representative height and / or local planar parameters, the reset results of the subsequent region have a clearer source of technical data. (3) Based on the area surface information rather than single point value, the focus reference is generated. It can improve the robustness to edge structure, local anomalies and reflection fluctuations by using the area representative height parameter, local plane fitting and high confidence sub-region screening, and reduce the probability of misfocusing caused by single point distortion. Local fine adjustment does not depend on sharpness and has higher accuracy. (3) The same surface measurement link simultaneously serves topography detection, focus reference generation and subsequent prediction update, reducing the independent point-type focus module and the resulting opto-mechanical integration and calibration burden, and reducing the coordinate registration requirements between the independent focus link and the topography detection link. (5) It has the ability to roll back from abnormal situations, and can automatically expand the search or recapture when a failure or local anomaly is predicted, thus balancing speed and reliability.

[0101] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A focusing control method for wafer morphology inspection, characterized in that, include: The wafer to be inspected is controlled to move relative to the image acquisition device along a preset detection path. The direction of the relative movement is located in a plane parallel to the reference plane of the stage that carries the wafer to be inspected. The wafer to be inspected is divided into multiple detection areas according to the detection path. Determine whether the current detection area meets the preset conditions, and perform an initial capture operation or a prediction-guided reset operation on the current detection area based on the determination result; The initial capture operation includes: controlling the image acquisition device to perform an initial scan and a secondary scan of the current detection area along a direction perpendicular to the surface of the wafer to be inspected, and calculating the first area focusing reference; The predictive guided reset operation includes: calculating the predicted focus reference position of the current detection area based on the first regional focus reference or the historical regional focus reference in the historical prior queue, resetting and confirming the predicted focus reference position, and performing morphology detection on the wafer to be detected based on the predicted focus reference position after successful reset confirmation.

2. The method according to claim 1, characterized in that, If the current detection area is the first detection area or the historical prior queue is empty, perform the initial capture operation; if the current detection area is not the first detection area and the historical prior queue is not empty, perform the prediction-guided reset operation.

3. The method according to claim 1, characterized in that, Perform initial and secondary scans on the current detection area, and calculate the first regional focusing reference, including: Within the first preset axial window, the image acquisition device is controlled to acquire a sequence of images of the first region at a first preset step size; Feature calculation is performed based on the image sequence of the first region to obtain coarse capture features; Based on the coarse capture features, threshold judgment and verification are performed to determine the initial detection frame; The position corresponding to the previous frame of the initial detection frame is used as the detection start position for a second scan to calculate the first region focusing reference.

4. The method according to claim 3, characterized in that, Feature calculation is performed based on the image sequence of the first region to obtain coarse capture features, including: Calculate the absolute value of the pixel difference between the current first region image and the previous frame's first region image, and sum them up to obtain the local energy of the current first region image. Calculate the absolute value of the pixel difference between the current first region image and the first region image of the first frame, and form an absolute value signal array of the current first region image; A sliding window is used to perform low-pass filtering on the absolute value signal data to obtain a weak localization feature array; Summing the values ​​in the weak localization feature array yields the lightweight envelope value of the current first region image; The local energy and lightweight envelope value are used as coarse capture features of the current first region image.

5. The method according to claim 4, characterized in that, Based on the coarse capture features, threshold judgment and verification are performed to determine the initial detection frame, including: The first region image with local energy greater than a preset energy threshold and light envelope value greater than a preset envelope threshold is used as the frame to be confirmed. The position corresponding to the frame to be confirmed is taken as the confirmation start position, and a verification image is acquired with a second preset step size, which is smaller than the first preset step size. Calculate the first local energy of the verification image and the frame to be confirmed, and the second local energy of the verification image and the first region image of the previous frame of the frame to be confirmed; When both the first local energy and the second local energy are greater than a preset energy threshold, the frame to be confirmed is determined to be the initial detection frame.

6. The method according to claim 3, characterized in that, The position corresponding to the previous frame of the initial detection frame is used as the detection start position for a second scan to calculate the region focusing reference, including: Starting from the detection start position, the second region image sequence is acquired with a third preset step size and height reconstruction is performed to obtain the three-dimensional coordinate set of the effective pixels in the second region image sequence. Height values ​​are extracted from the three-dimensional coordinate set of the second region image sequence. The median or height-weighted average of the height values ​​is calculated based on the height values, and the median or height-weighted average of the height values ​​is used as the representative height of the region. Based on the height value, a first-order plane fitting is performed to obtain local plane parameters; The region's representative height and local planar parameters are used as the first region's focusing reference. The third preset step size is smaller than the first preset step size.

7. The method according to claim 6, characterized in that, The predicted focus reference position of the current detection area is calculated based on the first regional focus reference or the historical regional focus reference in the historical prior queue, including: Obtain the center physical coordinates of the previous detection area and the current detection area, and calculate the relative translational displacement vector between the previous detection area and the current detection area; If the previous detection area of ​​the current detection area performed an initial capture operation, the predicted focus reference position of the current detection area is calculated based on the area representative height and local plane parameters in the first area focus reference and the relative translation displacement vector; If the previous detection area of ​​the current detection area did not perform the initial capture operation, the historical area focusing reference is obtained from the historical prior queue and weighted averaged to obtain the weighted average area focusing reference. The predicted focusing reference position of the current detection area is calculated based on the area representative height and local plane parameters in the weighted average area focusing reference and the relative translation displacement vector.

8. The method according to claim 6, characterized in that, The predicted focus reference position of the current detection area is calculated based on the historical area focus benchmark in the historical prior queue, including: Establish a polynomial global topographic trend surface model with respect to height and center physical coordinates; Obtain the physical coordinates of the center of the historical detection area within the preset sliding window, and obtain the corresponding historical area focusing reference from the historical prior queue. Substitute the area representative height and the physical coordinates of the center of the historical detection area in the historical area focusing reference into the polynomial global topography trend surface model, perform quadratic surface fitting based on the least squares method, and solve the coefficients in the polynomial global topography trend surface model. Obtain the center physical coordinates of the distant detection area that is farther from the current detection area than a preset distance, and substitute them into the polynomial global topography trend surface model for solution to obtain the predicted focus reference position of the current detection area.

9. The method according to claim 1, characterized in that, Reset and confirm the predicted focus reference position, including: The image acquisition device is moved to the predicted focus reference position and the image sequence of the third region is acquired within the second preset axial window with a fourth preset step size. Extract the region of interest from the third region image sequence and calculate the local energy of the region of interest; Find the position and height corresponding to the frame with the maximum local energy as the actual focus reference value; If the absolute value of the difference between the actual focus reference value and the predicted focus reference position is less than a preset threshold, then the reset is determined to be successful.

10. The method according to claim 6, characterized in that, Shape detection is performed based on the predicted focus reference position, including: The image acquisition device is controlled to move to the predicted focus reference position, and the image sequence of the fourth region is acquired with a third preset step size; Three-dimensional height reconstruction is performed based on the image sequence of the fourth region to obtain the three-dimensional reconstruction result of the current detection region. The regional focus reference for the current detection area is calculated based on the 3D reconstruction results of the current detection area and stored in the historical prior queue.

11. The method according to claim 9, characterized in that, If the absolute value of the difference between the actual focus reference value and the predicted focus reference position is greater than or equal to a preset threshold and less than a preset safety value, then the second preset axial window is expanded to acquire the third region image sequence again and the reset confirmation is performed again. If the reset confirmation fails again, or if the root mean square height residual of the 3D reconstruction result exceeds the preset residual threshold, the third preset step size is reduced, the third region image is acquired for the third time, and the reset confirmation is performed for the third time. If the current detection area has not undergone initial capture operation, the current detection area is divided into multiple sub-regions. Based on the 3D reconstruction results, the high continuity of the multiple sub-regions is calculated to determine the high-confidence sub-regions and calculate the proportion of the high-confidence sub-regions. If the third reset confirmation fails, or if the proportion of the high-confidence sub-region is lower than the preset proportion threshold, then the initial capture operation is performed on the current detection area.

12. A focusing control device for wafer morphology inspection, characterized in that, include: The motion control module is used to control the movement of the wafer under test relative to the image acquisition device along a preset detection path. The direction of the relative movement is located in a plane parallel to the reference plane of the stage that carries the wafer under test. The wafer under test is divided into multiple detection areas according to the detection path. The judgment module is used to determine whether the current detection area meets the preset conditions, and to perform an initial capture operation or a prediction-guided reset operation on the current detection area based on the judgment result. The initial capture module is used to perform the initial capture operation, including: controlling the image acquisition device to perform an initial scan and a secondary scan of the current detection area along a direction perpendicular to the surface of the wafer to be inspected, and calculating the first area focus reference; The guided reset module is used to perform a predictive guided reset operation, including: calculating the predicted focus reference position of the current detection area based on the first regional focus reference or the historical regional focus reference in the historical prior queue, resetting and confirming the predicted focus reference position, and performing morphology detection on the wafer to be detected based on the predicted focus reference position after successful reset confirmation.

Images