Wafer on-line detection system

Abstract

The invention relates to the technical field of online detection, in particular to a wafer online detection system which comprises an edge calibration module, a probe switching module, a temperature sensing module, an optical detection module and a dynamic compensation module. According to the method, the coordinate error of the wafer edge area is meticulously analyzed, the offset rate data is established, and the periodic deviation value is compared with the periodic deviation value in the probe switching process, so that the probe switching precision is remarkably improved, the analysis of temperature sensing and optical reflection data is optimized, and an accurate spectrogram layer difference labeling graph is obtained; through combination of the regional temperature difference fluctuation map and the optical error map, dynamic compensation is carried out on optical detection and temperature detection of the wafer, the wafer online detection effect is greatly improved, finer positioning and identification of wafer surface defects are realized, the overall precision and efficiency of wafer online detection are improved, the risk of missing detection and misjudgment of the defects is reduced, and the wafer online detection efficiency is improved. And the automation level and the accuracy of the detection result in the detection process of the production line are further optimized.

Description

Technical Field

[0001] This application relates to the field of online inspection technology, and more particularly to wafer online inspection systems. Background Technology

[0002] The field of online inspection technology involves technical matters related to real-time monitoring and data acquisition of the condition of workpieces or materials during product manufacturing or processing. This includes detection signal acquisition, real-time data analysis, defect identification methods, physical quantity conversion and measurement methods, etc. It is widely used in multiple industrial processes such as electronic manufacturing, machining, and material forming. Among them, the wafer online inspection system refers to a system that uses imaging devices and image analysis technology to inspect the surface condition of wafers in the semiconductor manufacturing process, while the wafers are still on the production line. Typically, a high-speed linear array camera is used to acquire surface images during the wafer's movement. Then, the image data is compared and analyzed by statistical analysis of image grayscale distribution characteristics or wafer pattern matching to identify specific types of surface defects such as scratches, contamination, indentations, and particle deposition.

[0003] Current online wafer inspection technologies largely rely on high-speed cameras to capture surface images and use image grayscale distribution or pattern matching methods for defect identification. However, when faced with complex environmental conditions and changing wafer surface conditions, these technologies often struggle to cope with factors such as offset errors, inaccurate probe switching, and temperature fluctuations, which require high real-time performance. In particular, when there are minute deformations, temperature differences, or abnormal optical reflections on the wafer surface, existing technologies are prone to errors, leading to reduced defect identification accuracy and even missed detections, thus affecting the stability and accuracy of the inspection results. Summary of the Invention

[0004] In view of this, embodiments of this application provide an online wafer inspection system to address the shortcomings of existing technologies in handling complex wafer surface condition changes, which often struggle to cope with factors such as offset errors, inaccurate probe switching, and temperature fluctuations, which have high real-time requirements. In particular, when there are minute deformations, temperature differences, or abnormal optical reflections on the wafer surface, the inspection results of existing technologies are prone to errors, leading to reduced accuracy in defect identification and even missed detections, thus affecting the stability and accuracy of the inspection results.

[0005] A first aspect of this application provides a wafer in-circuit inspection system, comprising: The edge calibration module acquires the structural image of the edge region of the wafer to be inspected, identifies the coordinate group of the optical calibration pattern and the coordinate group of the temperature-sensitive array within the boundary of the embedded ring calibration area, calculates the overall offset value to establish a coordinate error distribution map, and compares the cumulative error with the total number of points in the region to obtain the offset rate data of the calibration area. Based on the calibration area offset rate data, the probe switching module switches the detection probe to the wafer calibration area, records the switching cycle position change curve, compares the cycle deviation with the spatial path length of the calibration area, determines whether the switching has caused hysteresis behavior, and obtains the probe switching accuracy index. The temperature sensing module activates the resistor array according to the probe switching accuracy index, collects the thermal resistance change value sequence of the calibration area, calculates the average thermal response speed of the node and performs response speed difference judgment. If it is greater than the response equilibrium limit, it is marked as temperature sensing uneven behavior, and the regional temperature difference fluctuation map is obtained. The optical detection module acquires optical reflection images of the marked areas based on the regional temperature difference fluctuation map, obtains the pixel gray-level difference matrix between the current layer and the reference calibration layer at the same coordinate point, locates the position of the point with the maximum absolute value of the pixel difference, and generates a spectral layer difference annotation map. The dynamic compensation module performs compensation image reconstruction on the original detection image region by region based on the correspondence between the spectral layer difference annotation map and the regional coordinates in the temperature difference fluctuation map of the region, and obtains the wafer online detection imaging result map.

[0006] The beneficial effects of this application embodiment compared with the prior art are as follows: by meticulously analyzing the coordinate error of the wafer edge region, establishing offset rate data, and comparing it with the periodic deviation during probe switching, the accuracy of probe switching is significantly improved. This optimizes the analysis of temperature sensing and optical reflection data, thereby obtaining an accurate spectral layer difference annotation map. By combining the regional temperature difference fluctuation map with the optical error map, dynamic compensation is performed on the optical inspection and temperature detection of the wafer, greatly improving the wafer online inspection effect, achieving more precise positioning and identification of wafer surface defects, improving the overall accuracy and efficiency of wafer online inspection, reducing the risk of missed defects and misjudgments, and further optimizing the automation level and accuracy of inspection results in the production line inspection process. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a system flowchart illustrating an application scenario of this application. Figure 2 This is a schematic diagram of the edge calibration module of the wafer online inspection system provided in this application embodiment; Figure 3 This is a schematic diagram of the probe switching module of the wafer online inspection system provided in this application embodiment; Figure 4 This is a schematic diagram of the temperature sensing module of the wafer online inspection system provided in this application embodiment; Figure 5 This is a schematic flowchart of the optical inspection module of the wafer in-circuit inspection system provided in this application embodiment; Figure 6 This is a schematic diagram of the dynamic compensation module structure of the wafer online inspection system provided in this application embodiment. Detailed Implementation

[0009] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and systems are omitted so as not to obscure the description of this application with unnecessary detail.

[0010] The wafer online inspection system according to embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0011] Figure 1 This is a schematic flowchart of the wafer online inspection system provided in an embodiment of this application. Figure 1 As shown, the wafer online inspection system includes: The edge calibration module acquires the structural image of the edge region of the wafer to be inspected, identifies the coordinate group of the optical calibration pattern and the coordinate group of the temperature-sensitive array within the boundary of the embedded ring calibration area, and performs point-by-point comparison with the actual layout template position by accumulating the coordinate difference, calculates the overall offset value of the pattern and establishes the coordinate error distribution map of the edge region, and obtains the offset rate data of the calibration area by comparing the cumulative error in the coordinate error distribution map with the total number of points in the region. Based on the calibration area offset rate data, the probe switching module drives the piezoelectric ceramic actuator to switch the detection probe to the wafer calibration area, records the position change curve within the switching cycle and compares it with the spatial path length of the calibration area, calculates the cycle deviation using the difference between the position change rate and the set path length, determines whether the switching has formed hysteresis behavior, and obtains the probe switching accuracy index. The temperature sensing module activates the silicon carbide substrate platinum resistance array according to the probe switching accuracy index, collects the thermal resistance change value sequence of each temperature sensing node in the calibration area per unit time, calculates the average thermal response rate of each node, and performs response rate difference judgment between nodes in different regions. If the difference is greater than the response equilibrium limit, the uneven temperature sensing behavior in the region is marked, and the regional temperature difference fluctuation spectrum is obtained. The optical detection module controls the micro spectrometer probe to enter the marked area to collect optical reflection images based on the marked area in the regional temperature difference fluctuation map. It obtains the pixel gray-level difference matrix between the current layer and the reference calibration layer at the same coordinate point, performs layer error clustering analysis based on the gray-level difference value range of each pixel, and locates and marks the position corresponding to the point with the largest absolute value of pixel difference in each class to generate a spectral layer difference annotation map. The dynamic compensation module sets the optical compensation coefficient and thermal compensation coefficient for each region as a mapping parameter set based on the correspondence between the spectral layer difference annotation map and the regional temperature difference fluctuation map. It then calls the image processing unit in the FPGA module to perform compensation image reconstruction on the original detection image region by region to obtain the wafer online detection imaging result map.

[0012] The high-speed switching mechanism achieves 0.1-second switching of the detection probe between the wafer working area and the calibration area through a piezoelectric ceramic driving unit; the temperature-sensitive array is composed of a platinum resistance film and a silicon carbide substrate, with a thermal response time ≤5ms; the dynamic compensation module includes an FPGA processor and a compensation parameter storage unit, and is configured with a temperature-optical dual-parameter coupling correction algorithm based on the least squares method; the end of the detection probe integrates a miniature spectrometer and an infrared temperature sensor, forming a closed feedback loop with the calibration area.

[0013] The calibration area offset data includes the coordinate set of the optical calibration pattern, the coordinate set of the temperature-sensitive array points, the cumulative value of the coordinate difference, the actual placement of the template, the overall offset value of the pattern, the coordinate error distribution map of the edge area, the cumulative amount of error, and the total number of points in the area; the probe switching accuracy indicators include the position change curve, the spatial path length of the calibration area, the position change rate, the set path length, the periodic deviation, and the hysteresis behavior record; the regional temperature difference fluctuation map includes the thermal resistance change value sequence, the average thermal response rate, the response rate difference, the temperature sensing non-uniformity behavior, and the response balance limit; the spectral layer difference annotation map includes the optical reflection image, the pixel gray-level difference matrix, the error clustering analysis results, the maximum pixel difference point, and the positioning annotation record; the wafer online inspection imaging result map includes the optical compensation coefficient, the thermal compensation coefficient, and the compensated reconstructed image.

[0014] According to the technical solution provided in the embodiments of this application, by meticulously analyzing the coordinate error of the wafer edge region, offset rate data is established and compared with the periodic deviation during probe switching, significantly improving the accuracy of probe switching. This optimizes the analysis of temperature sensing and optical reflection data, resulting in an accurate spectral layer difference annotation map. By combining the regional temperature difference fluctuation map with the optical error map, dynamic compensation is performed on the optical and temperature detection of the wafer, greatly improving the online wafer inspection effect, achieving more precise positioning and identification of wafer surface defects, improving the overall accuracy and efficiency of online wafer inspection, reducing the risk of missed defects and misjudgments, and further optimizing the automation level and accuracy of inspection results in the production line inspection process.

[0015] Please see Figure 2 The edge calibration module includes: The edge structure image extraction submodule acquires the edge region structure image of the wafer to be detected, collects embedded ring calibration area image data as image input source, performs boundary recognition operation on the embedded region in the image, and detects the pixel position of the pattern center point and the center coordinate of the temperature sensitive array by identifying the pixel distribution characteristics of the embedded optical pattern and temperature sensitive array points within the boundary of the ring calibration area, and performs structural normalization processing of the coordinate position to generate the initial coordinate set of the embedded pattern. The operation of acquiring the structural image of the edge region of the wafer to be inspected based on the edge calibration module first involves capturing the image of the wafer edge region by configuring an imaging device to obtain image data of the embedded annular calibration area. The annular calibration area is generally pre-embedded within a region with a radius of 120mm to 125mm at the wafer edge, containing regularly distributed optical calibration patterns and temperature-sensitive array points. In actual implementation, an industrial CCD camera with a resolution of 2048×2048 pixels is used for image acquisition. The grayscale range of the acquired image is 0 to 255. After normalization processing of the raw image data to eliminate brightness interference and unify contrast, the boundary is then identified by combining the features of the calibration area in the image. The grayscale threshold for region recognition is 190, which serves as the binarization segmentation benchmark. When the grayscale value of a pixel is greater than this threshold, it is determined to be a pixel at the boundary of the embedded region; otherwise, it is considered background. In a certain recognition operation, the edge recognition result of the calibration area shows that the center position of the outer edge circle is (x=1024, y=1024), with a radius of 1120 pixels. Based on the boundary coordinate information, feature extraction is performed on the optical patterns and array points within the calibration area. The set of center point coordinates is identified by judging the symmetry, density, and arrangement direction of the pixel set. In one extraction task, 64 optical patterns were detected. The centroid of each pattern is calculated as the center coordinate of the pattern. Taking the grayscale distribution of pattern (1,1) as an example, the centroid coordinates in the x-direction are: ; in For the first Line number The grayscale value of the column pixels and the size of the pattern pixel area are set to... Similarly, we can obtain The center pixel coordinates of the pattern are (25.52, 26.40). This operation is repeated to obtain the set of center coordinates of all patterns and array points. Then, the coordinates are normalized according to the image pixel ratio (1 pixel = 1.5μm) to obtain the set of pattern and point coordinates in millimeters. In this example, the actual coordinates of the pattern (1,1) are (0.0383mm, 0.0396mm). Finally, the initial coordinate set of the embedded pattern is generated.

[0016] Table 1. Results of Embedded Pattern Coordinate Extraction (Unit: pixels and millimeters) As shown in Table 1, by normalizing the grayscale centroid coordinates, the millimeter coordinates of the pattern center point can be obtained, providing basic positioning data for subsequent comparison with the template position.

[0017] The calibration coordinate difference generation submodule is based on the initial coordinate set of the embedded pattern and the standard coordinate set of the actual wafer layout template. It establishes a pairing relationship based on the point index of the two sets of coordinates in the same area. By calculating the coordinate difference of the corresponding points, it performs a coordinate difference accumulation operation on all matching points and divides the accumulated value by the number of points to obtain the mean error coordinate. It then establishes a two-dimensional coordinate difference matrix in the wafer edge area to obtain the regional coordinate error distribution map. To call the initial coordinate set of the embedded pattern and the standard coordinate set of the actual wafer layout template, the point correspondence must first be established. Patterns with the same number are paired one-to-one in the two coordinate sets. For example, pattern number (1, 1) is located at (0.0383mm, 0.0396mm) in the initial coordinate set and at (0.0400mm, 0.0400mm) in the template standard coordinate set. The offset is obtained by calculating the coordinate difference along the x-axis and y-axis for each pair of points. For pattern (1, 1), the x-axis difference is: ; Similarly, the difference along the y-axis is: ; During the coordinate difference accumulation process, the x-axis and y-axis offset values ​​of all patterns are added point by point. In this example, there are 64 pattern points, and the total difference matrix is ​​set as follows: ; The mean of the matrix is ​​calculated to obtain the overall average error value of the region. Assuming the total difference along the x-axis is -0.102 mm and the total difference along the y-axis is -0.058 mm among the 64 patterns, the mean error of the region is: ; ; The error is mapped to the position on the wafer edge image, and the offset direction and magnitude of the pattern are represented by heatmaps or vector arrows to form a coordinate error vector map in two-dimensional space, and a regional coordinate error distribution map is established.

[0018] The offset rate calculation submodule selects the number of points that exceed the set offset threshold as the cumulative error based on the coordinate difference of each point in the regional coordinate error distribution map. It then calculates the offset ratio of the error points in the calibration area by performing a ratio calculation with the total number of points in the regional coordinate error distribution map, and obtains the offset rate data of the calibration area. Based on the coordinate difference magnitude of each point in the regional coordinate error distribution map, Euclidean distance is calculated for the difference between the x-axis and y-axis of each point to determine whether it exceeds the set offset threshold. The offset distance calculation formula is as follows: ; Among them, when When the offset threshold is set to mm, this point is counted as the offset point. For example, if the difference between pattern (1, 1) is (-0.0017, -0.0004), then its offset distance is: ; Therefore, pattern (1, 1) is not an offset point. Assuming that out of the 64 points, 11 points have an offset distance exceeding 0.005 mm, the offset rate calculation formula is: ; According to the calculation, the offset rate is 17.19%, which means that 17.19% of the pattern points in the calibration area have shifted beyond the standard offset threshold. Thus, the offset rate data of the calibration area is obtained.

[0019] Please see Figure 3 The probe switching module includes: The probe target driving submodule defines the starting point and ending point of the probe switching path within the boundary of the wafer calibration area based on the offset rate data of the calibration area. It estimates the error center by using the offset rate data within the area, determines the initial driving direction and target path curve of the piezoelectric ceramic actuator, guides the piezoelectric ceramic actuator to complete the transfer action of the probe to the calibration area according to the target path, and records the execution cycle used for each step in the transfer, generating a probe switching path trajectory data set. Based on the offset rate data of the calibration area, the region with a high offset rate distribution in the calibration area on the wafer surface is used as the switching target point set of the piezoelectric ceramic actuator. In a certain detection process, the edge region with an offset rate exceeding 17% is selected as the high offset target area. First, the initial position of the probe is defined as the standby position on the outer ring of the wafer, and the target position is the high offset center point of the calibration area. By analyzing the weight of each coordinate point in the offset rate data matrix, the weighted centering method is used to calculate the coordinates of the target path endpoint. In this embodiment, the offset rate weight coefficient is selected as the ratio of the offset rate value of each coordinate point to the total offset rate. It is set that there are four key points in the center area of ​​a certain calibration area, with offset rates of 0.21, 0.18, 0.25 and 0.16, respectively. The total offset rate is 0.80, and the corresponding coordinates are (2.0, 3.0), (3.0, 3.0), (2.5, 2.5) and (2.0, 2.0), respectively. The target point coordinates are calculated according to the weighted centering method as follows: ; ; The coordinates of the end of the target path are obtained as (2.4375, 2.7375). Then, a probe switching path is established based on this target point, using a piezoelectric ceramic sheet structure as the driving component. The step distance is set to 0.05mm each time, the control cycle is 8ms, and the total path length is 3.5mm. The theoretical number of step cycles is 70. The system sets the step number marker range to 0 to 69. The control system records the current position coordinates of each step and compares them with the theoretical trajectory curve to build a path tracking table. The table records information such as the current coordinates, theoretical target point, step number, and step distance. Refer to the trajectory sampling data below.

[0020] Table 2 Probe Switching Path Trajectory Data Table As shown in Table 2, the probe will experience slight displacement during the stepping process. By gradually recording the position coordinates of all trajectory points, a complete switching path data structure can be constructed, forming a probe switching path trajectory data set.

[0021] The position change acquisition submodule calculates the linear motion rate of the probe in each cycle based on the probe switching path trajectory data group, using the step time and corresponding position coordinates, and obtains the velocity vector sequence in continuous cycles. Based on the difference comparison between the actual moving distance of the probe in each cycle and the set path length, it identifies abnormal pauses and return positions during path execution, filters out position change points that have abrupt changes and extracts the corresponding cycle number to obtain the probe motion change rate sequence. The probe switching path trajectory data group is invoked to calculate the motion rate of the probe's actual movement trajectory in each cycle. First, the actual coordinate difference between any two adjacent points is extracted from the trajectory table. Combined with the step cycle time interval of 8ms, the average rate is calculated. Taking the first to second cycle as an example, the actual positions are (0.049, 0.052) to (0.098, 0.101), and the corresponding displacement differences in the X and Y directions are 0.049mm and 0.049mm, respectively. The rate calculation is as follows: ; ; The rates of the remaining cycles are calculated in the same way to construct a sequence containing the rates of continuous cycle motion. In 70 consecutive cycles, each 8 cycles is a monitoring group. By comparing the rate stability within each group, it is determined whether there are any abrupt changes. If the rate change of any cycle compared to the preceding and following cycles exceeds the set reference value of 0.30 mm / s, the cycle number is recorded as an abrupt change cycle. In this embodiment, the rate of the 13th cycle is 9.2 mm / s, and the rates before and after are 8.6 mm / s and 8.4 mm / s, respectively. The fluctuation value of the 13th cycle is 0.6 mm / s, which is greater than the threshold and is considered an abnormal cycle, marked as 13. Other abrupt change cycles are counted to form a fluctuation trend table. The numbers of each abrupt change cycle and their rates are output, which are further used to judge abnormal backlash behavior and obtain the probe motion change rate sequence.

[0022] The period deviation determination submodule divides and determines the rate difference between any consecutive periods based on the probe motion change rate sequence. The speed change threshold is set to 0.25 mm / s. The period number in which the rate fluctuation exceeds the threshold is marked. Combined with the calibration area path length parameter, the distance difference between the offset path and the target path is calculated for these period segments to determine whether there is backlash behavior during the movement. If the period difference direction is opposite to the path direction and the distance difference is greater than the backlash threshold of 0.02 mm, it is considered as one backlash. All backlashes are accumulated to obtain the probe switching accuracy index. Based on the probe motion rate change sequence, difference calculations and path difference determinations are performed on all cycles marked as rate abrupt changes. First, the actual path coordinates corresponding to these cycles and the path values ​​of the cycles before and after them are extracted. It is then determined whether the direction of the difference is consistent with the target path direction. If the direction is opposite and the absolute value of the distance difference is greater than the backlash threshold of 0.02 mm, it is recorded as a backlash cycle. In the 13th cycle, the coordinates regress from (0.750, 0.751) to (0.700, 0.701), and the path difference is: ; Since the direction is reversed and the deviation is greater than 0.02mm, this cycle is determined to be a hysteresis cycle. All abrupt change cycles are processed. Assuming that a total of 5 hysteresis cycles are identified, accounting for 7.14% of the theoretical path step count of 70, this proportion is the switching error rate, reflecting the accuracy of the probe switching process. Further, by combining the path execution error distribution map, velocity sequence and abnormal point distribution density, a comprehensive index is constructed to obtain the probe switching accuracy index of this switching task.

[0023] Please see Figure 4 The temperature sensing module includes: The thermal node activation submodule is based on the probe switching accuracy index. According to the number of hysteresis cycles and the stability parameters of the switching path during the probe switching process, a trigger threshold is set. When the accuracy index value exceeds the upper limit of the set range of 0.08, the power channel of the silicon carbide substrate platinum resistance array is driven to perform periodic excitation operation on 128 distributed nodes in the calibration area. The resistance response value sequence is recorded at a period of 10ms and a matrix data structure is established. The resistance change value of each node per unit time is obtained and the response history sequence is constructed to generate a thermal resistance change value sequence matrix. Based on the probe switching accuracy index, a driving threshold is set. When the accuracy index value is greater than 0.08, the silicon carbide substrate platinum resistance array is activated. Each activation is based on a group of 32 nodes, with a total of 128 nodes controlled concurrently. After activation, the current excitation cycle of each node is 10ms, and the total measurement cycle is 1000ms. The system records the resistance change value every 10ms. In one test, the initial resistance of the thermistor with node number N15 is set to 108.3Ω. After activation, the resistance rises to 108.9Ω after 10ms and reaches 109.5Ω after 20ms. The resistance change of each node in each time period is sampled accordingly. The sample data structure is a two-dimensional matrix constructed with time and node dimensions as the two axes, with the node number as the main index and the time axis unit being ms increments.

[0024] Table 3. Sample examples of the thermal resistance variation value sequence matrix (unit: Ω) As shown in Table 3, each node corresponds to multiple resistance values ​​at various times. The sampling resolution is 0.1Ω. Each cycle is collected by the array controller and stored as a timestamped data block. During the recording process, the current input is kept constant, with an applied current of 1.0mA to ensure sampling stability. This process continues until all nodes complete 10 sets of time-series sampling tasks. After the collection is completed, the raw data is classified according to the node number and sorted by the time index to form a complete sequence matrix of thermal resistance change values.

[0025] The thermal response rate calculation submodule is based on the thermal resistance change value sequence matrix. It calculates the rate based on the resistance increase and duration of each node in a continuous sampling period. The response rate value is obtained by dividing the average resistance increase over 5 periods by the total duration. In a certain node, the resistance increases from 108.2Ω to 109.7Ω in 50ms, so the average response rate is 0.03Ω / ms. The 128 nodes are processed in sequence to generate a rate array with the node index number as the primary key, and the node thermal response rate sequence is obtained. The thermal resistance change sequence matrix is ​​used to calculate the rate of resistance increase for each node within a set time period. Five sets of continuous data are used as a sliding window for averaging. In the N15 node data, the resistance increases from 108.3Ω to 110.6Ω from 0ms to 40ms. Therefore, the average thermal response rate is: ; The remaining 127 nodes were processed in the same way to construct a complete thermal response rate array. In each processing step, if the node had less than 5 sets of sampled data, linear interpolation was used to complete the completion operation. Then, number mapping was performed on the rate array, binding each rate value to its corresponding node index number to generate a rate dictionary. Based on this rate dictionary, a node sorting sequence was constructed to identify the node with the highest and lowest rate and their rate difference for the next step of regional difference judgment. In this test, the highest response rate appeared in N27 at 0.063Ω / ms, and the lowest appeared in N49 at 0.025Ω / ms, with a difference of 0.038Ω / ms, which was lower than the judgment threshold. The current region was not marked as abnormal, and the response rate values ​​of all nodes were written to the storage unit to obtain the node thermal response rate sequence.

[0026] The temperature fluctuation determination submodule calculates the rate difference between adjacent regions within the calibration area based on the node thermal response rate sequence. Every four adjacent nodes constitute a sub-region unit. The difference between the maximum and minimum rates within each unit is calculated. If the difference in any sub-region exceeds the set response equilibrium limit of 0.06Ω / ms, the coordinates of the region are marked and the anomaly type is output. The boundary points of all anomaly regions are counted and superimposed onto the wafer region template to obtain the regional temperature fluctuation map. Based on the node thermal response rate sequence, the wafer calibration area is divided into 32 sub-blocks, each containing 4 adjacent nodes forming a two-dimensional grid in numerical order. The difference between the maximum and minimum rates of the 4 nodes within each sub-block is calculated as an indicator of the sub-block's response imbalance. If this difference exceeds the set response equilibrium limit of 0.06 Ω / ms, the sub-block is marked as an abnormal region. Taking sub-block Z12 as an example, it contains nodes N41~N44, with corresponding thermal response rates of 0.060, 0.065, 0.061, and 0.054 Ω / ms, respectively. The maximum value is 0.065, the minimum value is 0.054, and the difference is 0.011, which is lower than the set threshold and is not marked as abnormal. In another sub-region Z17, the node rates are 0.045, 0.103, 0.050, and 0.048 Ω / ms, and the difference between the maximum and minimum values ​​is 0.058, which is slightly lower than the threshold and is not marked. However, if the actual difference is 0.068, then region Z17 is immediately marked as a temperature difference abnormality region. By superimposing the boundary points of all marked regions onto the wafer template image, a thermal response imbalance visualization image is drawn, and a two-dimensional chromatographic thermogram output result is constructed to obtain the regional temperature difference fluctuation spectrum.

[0027] Please see Figure 5 The optical detection module includes: The image acquisition submodule controls the micro spectrometer probe to move to the center point coordinates of each region and perform a fan-shaped scan based on the abnormal areas marked in the regional temperature difference fluctuation spectrum. It acquires the reflected light intensity image of each coordinate point in the current layer and simultaneously calls the standard image frame of the reference calibration layer. It establishes a pixel correlation matrix by mapping the original layer and the calibration layer in a one-to-one correspondence of spatial coordinate points to obtain the coordinate mapping image group. Based on the marked abnormal areas in the regional temperature difference fluctuation map, the coordinate analysis module is called to extract the center point coordinates of all abnormal blocks according to the coordinate boundaries of the red-marked areas in the map, constructing a center point distribution array, which is then input into the probe control unit of the miniature spectrometer as the target path. During this process, the probe movement step interval is set to 0.1 mm, and the scanning path step size is limited to ±0.2 mm. The probe initiates a linear fan-shaped scanning mode at the center coordinate point of each target area, with a scanning angle range of ±30°. Simultaneously, the reflectance image of the current area layer is acquired during the scanning cycle, with an image acquisition frame rate set to 20 fps and an image size of 128×128 pixels. After acquisition, the reference calibration layer image at the corresponding coordinate point is also synchronously loaded by calling the historical database, establishing a one-to-one mapping table of pixels between the current layer and the reference layer at the same spatial coordinate points. This mapping table uses a two-dimensional index structure, with each coordinate pair containing the corresponding pixel block number in the current frame image and the calibration image, the grayscale matrix position identifier, and the position matching index. The following are some examples of image coordinate mapping data: Table 4. Partial sample data of coordinate mapping image group As shown in Table 4, a complete spatial mapping relationship is established for each acquisition point, including the reflection image frame number, coordinate point position and matching degree index. This structure is used to accurately assign the corresponding pixels in the subsequent difference matrix calculation, and finally generate a coordinate mapping image group.

[0028] The pixel difference matrix calculation submodule extracts the grayscale value of each pixel in the two layers based on the coordinate-mapped image group and performs point-by-point difference calculation. The grayscale value range is set to 0 to 255. If the grayscale value of the current layer is 198 and the grayscale value of the calibration layer is 162 at a certain coordinate point in an abnormal region, the corresponding difference is 36. The grayscale difference of all pixels is calculated in this way, using the formula: ; The composite difference magnitude of each pixel is calculated, a two-dimensional matrix structure is constructed, and a layer pixel gray-level difference matrix is ​​generated; where... Indicates the first Line number The result of the difference between the columns of pixels. The grayscale value of the current layer. To calibrate the grayscale values ​​of the layer, This represents the sum of the horizontal grayscale differences between four adjacent points. The current image frame and the calibration image frame in the coordinate mapping image group are called to extract the grayscale data of all pixels in the two-dimensional image matrix. The grayscale value range is limited to 0 to 255. For each pair of pixels at the same position, the grayscale difference is calculated. Taking the pixel at position (i, j) as an example, if the grayscale of the current image is 198 and the grayscale of the calibration image is 162, the difference is 36. Further considering the complexity of the local structure of the image, the gradient change of surrounding pixels is introduced to adjust the difference. The calculation is performed using a formula. In a set of actual data, if , , , , , The calculation process is as follows: ; This value is the composite grayscale difference of the current pixel. This operation expands the image matrix pixel by pixel to construct a two-dimensional numerical matrix with the same size as the original image, forming the difference image structure data, and obtaining the layer pixel grayscale difference matrix.

[0029] The formula's operational logic can be explained in three parts. First, the first term is the absolute value of the grayscale difference between corresponding pixels in the current image and the reference image. This is used to directly reflect the basic brightness difference between the current layer and the standard layer at this coordinate. Its purpose is to quantify the first-order change trend of pixel reflection characteristics. Secondly, the second item... This is the square root of the product of the grayscale values ​​of the current layer and the calibration layer. This term expresses the relative brightness coupling strength between the two grayscale values. Since grayscale values ​​are all non-negative integers, taking the square root after multiplication compresses the value to a scale more suitable for superimposing with the first term, thus reflecting the influence of areas with higher brightness matching on difference enhancement. Finally, the third term is the average grayscale difference between the current pixel and its four horizontally adjacent pixels. This item is used to compensate for high-frequency fluctuations or noise features in local images. If the gray level change of the current point is relatively large compared to the change of neighboring pixels, this value increases, forming a deduction item for the overall difference, thereby suppressing the interference of high-frequency errors on the difference map. Therefore, the combination of the three items is a logical structure of difference enhancement + matching enhancement - noise cancellation, which can effectively reflect the differences between layers of brightness shift, structure alignment and edge fluctuation.

[0030] The composite difference magnitude refers to the overall difference in grayscale levels between the current layer and the reference calibration layer at the same coordinate pixel point. This difference not only considers the absolute difference in grayscale values ​​of corresponding pixels in the two layers, but also introduces adjustment factors such as brightness coupling strength and local structural perturbation. This allows the index to comprehensively reflect the overall state of pixel brightness shift, region alignment, and local texture changes in optical imaging. Its physical meaning is the overall deviation of a pixel from the standard image in terms of structural stability and spectral consistency. The larger the value, the more obvious the optical anomaly or imaging structural perturbation at that location. The smaller the value, the stronger the consistency and continuity between the current layer and the calibration layer at that pixel point.

[0031] The difference annotation extraction submodule uses a fixed hierarchical interval to cluster the difference value range according to the layer pixel gray-level difference matrix. It divides the difference into 4 groups according to the size of the gray-level difference and extracts the coordinate value of the point with the largest absolute difference in each group. After mapping back to the original image space coordinates, it is marked with a red box and superimposed on the original reflection layer to create a spectral layer difference annotation map. Based on the layer pixel grayscale difference matrix, the difference range in the entire image is graded and divided into four groups: 0 to 63, 64 to 127, 128 to 191, and 192 to 255. Within each group, the maximum value is selected based on the absolute grayscale difference, and its (i, j) position index in the image is recorded. After obtaining the coordinates of the maximum difference point, the corresponding coordinate position in the original image is traced back according to the coordinate mapping relationship. A crosshair positioning graphic is used to perform the annotation operation. Each positioning icon is drawn with a red border on the layer above the image. The annotation size is fixed at 5×5 pixel rectangle, and the position accuracy is ±0.5 pixels. After each clustering and filtering, the system generates a set of positioning annotation coordinates. The overlay processing adopts a bitmap addition blending method. The layer fusion operation is performed on the bitmap rendering layer. After fusion, the marked areas in the image have a highlight display feature, generating a spectral layer difference annotation map.

[0032] Please see Figure 6 The dynamic compensation module includes: The compensation coefficient mapping submodule obtains the coordinate point information in the spectral layer difference annotation map and the regional temperature difference fluctuation map, constructs the regional coordinate mapping relationship structure between the two, sets multiple optical and thermal compensation coefficient configuration rules according to the joint combination of gray level difference distribution and temperature response rate interval, binds the parameter combination results of each region with the region number index, and generates a regional compensation parameter matrix. The coordinate information contained in the spectral layer difference annotation map and the regional temperature difference fluctuation map was obtained. All coordinate points in both were merged and matched according to the region number to construct a unified two-dimensional coordinate index table. The gray-level difference range and temperature difference response rate range of the layer corresponding to each region number were extracted. Based on this, optical-thermal dual-parameter combination conditions were established. The combination mapping rule was set as follows: the optical compensation coefficient was set to 0.85 for gray-level differences between 64 and 127, and the thermal compensation coefficient was set to 1.10 for temperature difference rates between 0.05 and 0.07 Ω / ms. For all regional coordinate points, the interval was judged and the corresponding compensation coefficient was selected. When there is a region with region number B07, the maximum gray-level difference of the layer is 121 and the temperature difference response rate is 0.063 Ω / ms, the compensation coefficient combination of 0.85 and 1.10 was selected according to the preset mapping rule and recorded as the compensation factor combination for this region. All regional compensation factors and region numbers were constructed into a two-dimensional structure matrix and arranged in ascending order of number. The output results are shown in Table 5. Table 5 Example of Regional Compensation Parameter Matrix As shown in Table 5, by classifying various physical quantities in the region into the compensation coefficient mapping relationship, a two-way index structure from the region number to the compensation factor group is established to obtain the regional compensation parameter matrix.

[0033] The image dual-parameter reconstruction submodule is based on the region compensation parameter matrix. For the gray-level distribution of each region in the original detection image, it extracts the corresponding optical compensation coefficient and thermal compensation coefficient according to the coordinate index. It performs correction processing according to the coupling relationship between the two parameters and the original pixel gray level, updates all pixels in the region, outputs the whole image compensation result, and obtains the compensated gray-level image matrix. The optical and thermal compensation coefficients corresponding to the region compensation parameter matrix are called and written into the FPGA image processing control buffer according to the region number. Before executing the image reconstruction task, the control unit reads the coordinates of the corresponding region pixel by pixel and extracts the initial gray value of the pixel in the original detection image. The region number of the corresponding coordinate is B07, its gray value is 185, and the corresponding compensation coefficients are 0.85 and 1.10. Then the system performs subtraction compensation according to the gray value × coefficient weight difference rule, adjusts the gray value of the current pixel to obtain the compensated gray value of 173. Subsequently, all pixels in the region are processed in sequence and the gray value matrix is ​​updated in real time. If the gray value exceeds the standard gray value upper limit of 255 after correction during the matrix update process, it is set to 255. If it is lower than 0, it is set to 0. After processing, the complete image compensation gray value matrix is ​​output and the matrix data stream is encapsulated in the region index structure to obtain the compensation gray value image matrix.

[0034] The wafer image output submodule performs a standard grayscale mapping operation on the pixel values ​​according to the compensation grayscale image matrix, uniformly adjusting the pixel values ​​to the range of 0 to 255. The parts exceeding the range are truncated and reduced. The data is written into the image buffer structure in row and column order and a standard 8-bit grayscale image data structure is generated. The entire image data frame is output and stored to obtain the wafer online inspection imaging result image. Based on the compensated grayscale image matrix, image numerical format normalization is performed, converting all pixel grayscale values ​​to 8-bit format and expressing them as integers. During normalization, grayscale upper and lower limits are set to 0 and 255, respectively. Pixels exceeding the range are truncated. Subsequently, the entire image is re-encoded into an image buffer frame format in row and column order and written to the image display output buffer channel in the FPGA. The buffer pointer is refreshed in the buffer area, and the current compensated image frame is output. The image data is converted into the standard output image format PNG by the video stream protocol conversion module and written to the image frame recording unit. This image structure contains a complete region number index, position coordinate mapping structure, and grayscale value matrix. Finally, a unified structure data file output result is generated to obtain the wafer online inspection imaging result image.

[0035] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0036] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A wafer online inspection system, characterized in that, The system includes: The edge calibration module acquires the structural image of the edge region of the wafer to be inspected, identifies the coordinate group of the optical calibration pattern and the coordinate group of the temperature-sensitive array within the boundary of the embedded ring calibration area, calculates the overall offset value to establish a coordinate error distribution map, and compares the cumulative error with the total number of points in the region to obtain the offset rate data of the calibration area. Based on the calibration area offset rate data, the probe switching module switches the detection probe to the wafer calibration area, records the switching cycle position change curve, compares the cycle deviation with the spatial path length of the calibration area, determines whether the switching has formed hysteresis behavior, and obtains the probe switching accuracy index. The temperature sensing module activates the resistor array according to the probe switching accuracy index, collects the thermal resistance change value sequence of the calibration area, calculates the average thermal response speed of the node and performs response speed difference judgment. If it is greater than the response equilibrium limit, it is marked as temperature sensing uneven behavior, and the regional temperature difference fluctuation map is obtained. The optical detection module acquires the optical reflection image of the marked area based on the marked area in the temperature difference fluctuation spectrum of the region, obtains the pixel gray level difference matrix between the current layer and the reference calibration layer at the same coordinate point, locates the position of the point with the maximum absolute value of the pixel difference, and generates a spectral layer difference annotation map. The dynamic compensation module performs compensation image reconstruction on the original detection image region by region based on the correspondence between the spectral layer difference annotation map and the regional coordinates in the regional temperature difference fluctuation map, and obtains the wafer online detection imaging result map.

2. The wafer online inspection system according to claim 1, characterized in that, The calibration area offset rate data includes the optical calibration pattern coordinate group, the temperature-sensitive array point coordinate group, the cumulative coordinate difference value, the actual template position, the overall pattern offset value, the edge area coordinate error distribution map, the cumulative error amount, and the total number of points in the area; the probe switching accuracy index includes the position change curve, the calibration area spatial path length, the position change rate, the set path length, the period deviation amount, and the hysteresis behavior record; the area temperature difference fluctuation spectrum includes the thermal resistance change value sequence, the average thermal response rate, the response rate difference, the temperature sensing unevenness behavior, and the response balance limit. The spectral layer difference annotation map includes optical reflection image, pixel gray-level difference matrix, error clustering analysis results, pixel difference maximum point, and location annotation record; The wafer online inspection imaging results include optical compensation coefficient, thermal compensation coefficient, and compensated reconstructed image.

3. The wafer online inspection system according to claim 1, characterized in that, The edge calibration module includes: The edge structure image extraction submodule acquires the edge region structure image of the wafer to be detected, collects embedded ring calibration area image data as image input source, performs boundary recognition operation on the embedded region in the image, and detects the pixel position of the pattern center point and the center coordinate of the temperature sensitive array by identifying the pixel distribution characteristics of the embedded optical pattern and temperature sensitive array points within the boundary of the ring calibration area, and performs structural normalization processing of the coordinate position to generate the initial coordinate set of the embedded pattern. The calibration coordinate difference generation submodule establishes a pairing relationship based on the initial coordinate set of the embedded pattern and the standard coordinate set of the actual wafer layout template, based on the point index of the two sets of coordinates in the same area. By calculating the coordinate difference of the corresponding points, the coordinate difference accumulation operation is performed on all matching points, and the accumulated value is divided by the number of points to obtain the mean error coordinate. A two-dimensional coordinate difference matrix is ​​established in the wafer edge area to obtain the regional coordinate error distribution map. The offset rate calculation submodule selects the number of points that exceed the set offset threshold as the cumulative error based on the coordinate difference of each point in the regional coordinate error distribution map. It then calculates the offset ratio of the error points in the calibration area by performing a ratio calculation with the total number of points in the regional coordinate error distribution map, and obtains the offset rate data of the calibration area.

4. The wafer online inspection system according to claim 1, characterized in that, The probe switching module includes: Based on the calibration area offset rate data, the probe target driving submodule defines the starting point and ending point of the probe switching path within the boundary of the wafer calibration area. It estimates the error center through the offset rate data within the area, determines the initial driving direction and target path curve of the piezoelectric ceramic actuator, guides the piezoelectric ceramic actuator to complete the probe transfer action to the calibration area according to the target path, and records the execution cycle used for each step in the transfer, generating a probe switching path trajectory data set. The position change acquisition submodule calculates the linear motion rate of the probe in each cycle using the step time and corresponding position coordinates based on the probe switching path trajectory data group, obtains the velocity vector sequence in continuous cycles, compares the difference between the actual moving distance of the probe in each cycle and the set path length, identifies abnormal pauses and return positions during path execution, filters out position change points that have abrupt changes and extracts the corresponding cycle number to obtain the probe motion change rate sequence. The period deviation determination submodule performs interval division and determination calculation on the rate difference between any consecutive periods based on the probe motion change rate sequence, marks the period number where the rate fluctuation amplitude exceeds the speed change threshold, and calculates the distance difference between the offset path and the target path for the corresponding period segment in combination with the calibration area path length parameter, determines whether there is backlash behavior during the movement, accumulates all backlash counts, and obtains the probe switching accuracy index.

5. The wafer online inspection system according to claim 4, characterized in that, The process of marking the period number in which the rate fluctuation amplitude of any period exceeds the threshold is as follows: the absolute value of the rate difference between two consecutive periods is compared with the speed change threshold. If the absolute value of the rate difference is greater than or equal to the speed change threshold, the corresponding period is marked as an abnormal period.

6. The wafer online inspection system according to claim 5, characterized in that, The process of determining whether there is backlash behavior during movement is as follows: In the abnormal cycle, if the angle between the movement direction of the current cycle and the direction of the target path curve is greater than 90 degrees, and the distance difference between the offset path and the target path is greater than the backlash threshold, then it is determined that a backlash has occurred.

7. The wafer online inspection system according to claim 1, characterized in that, The temperature sensing module includes: Based on the probe switching accuracy index, the thermal node activation submodule drives the power channel of the silicon carbide substrate platinum resistance array when the accuracy index value exceeds the trigger threshold, according to the hysteresis cycle number and switching path stability parameters presented during the probe switching process. It performs periodic excitation operations on the distributed nodes in the calibration area, records the resistance response value sequence and establishes a matrix data structure, obtains the resistance change value of each node per unit time and constructs the response history sequence, and generates a thermal resistance change value sequence matrix. The thermal response rate calculation submodule calculates the rate based on the thermal resistance change value sequence matrix and the resistance increase amplitude and duration of each node in the continuous sampling period. The response rate value is obtained by dividing the average resistance increase by the total duration. All nodes are processed in sequence to obtain a rate array with the node index number as the primary key, and the node thermal response rate sequence is obtained. The temperature fluctuation determination submodule calculates the rate difference between adjacent regions within the calibration area based on the node thermal response rate sequence. Every four adjacent nodes constitute a sub-region unit. The difference between the maximum and minimum rates within each unit is calculated. If the difference within any sub-region exceeds the set response equilibrium limit, the coordinates of that region are marked and the anomaly type is output. The boundary points of all anomaly regions are counted and superimposed onto the wafer region template to obtain the regional temperature fluctuation map.

8. The wafer online inspection system according to claim 1, characterized in that, The optical detection module includes: Based on the abnormal areas marked in the temperature difference fluctuation spectrum of the region, the image acquisition submodule controls the micro spectrometer probe to move to the center point coordinates of each region and unfolds a fan-shaped scan, acquires the reflected light intensity image of each coordinate point in the current layer and synchronously calls the standard image frame of the reference calibration layer, and establishes a pixel correlation matrix by using a one-to-one mapping method between the original layer and the calibration layer in spatial coordinate points to obtain the coordinate mapping image group. The pixel difference matrix calculation submodule extracts the gray value of each pixel in the two layers according to the coordinate mapping image group and performs point-by-point difference calculation. It sets the gray value range to calibrate the gray value of the layer, calculates the gray value difference of all pixels, calculates the composite difference amplitude of each pixel, constructs a two-dimensional matrix structure, and generates the layer pixel gray value difference matrix. The difference annotation extraction submodule performs clustering and grouping of the difference value range based on the pixel gray-level difference matrix of the layer using a fixed hierarchical interval. It divides the group according to the size of the gray-level difference and extracts the coordinate value of the point with the largest absolute difference in each group. After mapping back to the original image space coordinates, it marks the difference with a red box and superimposes it onto the original reflection layer to establish a spectral layer difference annotation map.

9. The wafer online inspection system according to claim 1, characterized in that, The dynamic compensation module includes: The compensation coefficient mapping submodule obtains the coordinate point information in the spectral layer difference annotation map and the regional temperature difference fluctuation map, constructs the mapping relationship structure between the two regional coordinates, sets multiple sets of optical and thermal compensation coefficient configuration rules according to the joint combination of gray level difference distribution and temperature response rate interval, binds the parameter combination results of each region with the region number index, and generates a regional compensation parameter matrix. The image dual-parameter reconstruction submodule, based on the region compensation parameter matrix, extracts the corresponding optical compensation coefficient and thermal compensation coefficient according to the coordinate index of each region in the original detection image, performs correction processing according to the coupling relationship between the two parameters and the original pixel gray level, updates all pixels in the region, outputs the whole image compensation result, and obtains the compensated gray level image matrix. The wafer image output submodule performs a standard grayscale mapping operation on the pixel values ​​according to the compensated grayscale image matrix, uniformly adjusting the pixel values ​​to the range of 0 to 255. The parts exceeding the range are truncated and reduced. The data is written into the image cache structure in row and column order and a standard grayscale image data structure is generated. The entire image data frame is output and stored to obtain the wafer online inspection imaging result image.

Images