Wafer epitaxial film thickness measurement method and system based on trajectory self-adaptation

Abstract

The present application proposes a wafer epitaxial film thickness measurement method and system based on trajectory adaptation, belonging to the field of measurement and testing and semiconductor manufacturing technology. The measurement method includes wafer feeding and preliminary transfer steps, wafer positioning and correction compensation steps, measurement path planning steps, film thickness data acquisition and processing steps, data comparison and trend report generation steps, and abnormal alarm triggering steps. The measurement system includes feeding and transferring units, positioning and compensation units, measurement path planning units, film thickness data acquisition and processing units, data comparison and trend report generation units, and abnormal alarm triggering units. The positioning and compensation unit includes a correction mechanism and a compensation mechanism. The present application improves the precision and efficiency of wafer epitaxial film thickness measurement through trajectory adaptive planning and edge enhancement algorithm; combined with cloud collaborative analysis, it realizes real-time abnormal alarm and trend prediction, which helps to improve the yield and intelligent level of semiconductor manufacturing.

Description

Technical Field

[0001] This invention belongs to the field of measurement and testing and semiconductor manufacturing technology, and particularly relates to a method and system for measuring wafer epitaxial film thickness based on trajectory adaptation. Background Technology

[0002] During semiconductor wafer manufacturing, different epitaxial layers need to be formed, and the thickness of each epitaxial layer varies. The thickness accuracy of the epitaxial layers directly determines device performance and production yield. Especially as process nodes advance to 3nm mass production and 2nm development stages, film thickness control needs to meet the stringent standard of within ±10% of the target value. The measurement accuracy of some critical layers even approaches the atomic limit of 0.2nm. Precise detection and control of epitaxial film thickness has become a key link in ensuring the performance of core structures such as high-k metal gates and silicon photonic devices.

[0003] Traditional epitaxial film thickness measurement techniques rely on spectral reflectance and elliptic polarization methods to invert thickness information by analyzing the interference signal of reflected light from the thin film. However, existing technologies suffer from multiple bottlenecks: First, measurement accuracy is limited by interference factors. Chinese invention patent application CN202411201588.7 points out that existing equipment generally lacks dedicated calibration mechanisms. Positional deviations during wafer transfer directly lead to measurement errors, and calibration methods involving moving the measuring instrument reduce the stability of the results. Simultaneously, wafer warping causes angular shifts in the reflected beam, and edge diffraction interference can all cause spectral signal distortion, affecting data accuracy. Second, measurement efficiency and adaptability are insufficient. Traditional equipment relies on fixed measurement paths, requiring multiple displacements of the XY module to achieve multi-point detection, which is cumbersome and time-consuming, making it difficult to meet the high-efficiency requirements of mass production lines. Furthermore, it lacks adaptive adjustment capabilities for wafers of different sizes, limiting its versatility. Third, data processing and feedback are lagging. Most systems can only output data from a single measurement. As disclosed in Chinese invention patent application CN202410560289.8, existing methods suffer from systemic errors due to insufficient spectral resolution and lack of multi-dimensional data comparison. Furthermore, they are unable to predict film thickness trends and respond to anomalies in real time.

[0004] While there are improvement ideas such as hybrid metrology and in-situ metrology, they face many obstacles in practice: hybrid metrology needs to solve the problems of equipment interface adaptation and algorithm integration, which is costly and has high compatibility risks; in-situ metrology is affected by process environment interference, and its accuracy is lower than that of independent equipment, making it difficult to meet the requirements of advanced processes. The manual measurement process further exacerbates the error risk. From wafer handling to data interpretation, manual operations reduce efficiency and introduce human bias.

[0005] Against this backdrop, there is an urgent need to develop a measurement technology that combines high precision, adaptability, and intelligence. By solving core issues such as positioning correction, path optimization, signal noise reduction, and data collaboration, this technology can overcome existing technical bottlenecks and provide a reliable film thickness monitoring solution for advanced semiconductor processes. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes a method and system for measuring wafer epitaxial film thickness based on trajectory adaptation.

[0007] In a first aspect of the invention, a method for measuring wafer epitaxial film thickness based on trajectory adaptation is proposed;

[0008] The method includes the following steps:

[0009] Wafer loading and initial transfer: After loading the wafer to be measured, the wafer is transferred to the calibration mechanism through a preset turnover mechanism;

[0010] Wafer positioning and calibration compensation: The positioning component of the calibration mechanism performs initial positioning of the wafer transferred to the calibration mechanism, and drives the calibration stage to move based on the three-dimensional coordinate compensation value to complete the wafer calibration compensation;

[0011] Measurement path planning: The wafer that has completed the calibration and compensation is transferred to the measurement mechanism. The trajectory adaptive algorithm built into the measurement mechanism automatically plans an optimized measurement path based on the actual size of the current wafer, and controls the measurement stage to move along the optimized measurement path.

[0012] Film thickness data acquisition and processing: During the movement of the measurement stage, the spectrometer is activated to acquire the raw measurement data of the wafer epitaxial film, and the raw measurement data is transmitted to the edge enhancement algorithm module in real time. The edge enhancement algorithm module eliminates the edge diffraction interference in the raw measurement data to obtain effective measurement data.

[0013] Data comparison and trend report generation: The effective measurement data is transmitted to the cloud-based collaborative analysis system. The cloud-based collaborative analysis system calls a preset historical database to compare and analyze the effective measurement data with the epitaxial film thickness data of the same type of wafer in the historical database. Based on the comparison results, a wafer epitaxial film thickness trend prediction report is generated.

[0014] Anomaly alarm triggering: The cloud-based collaborative analysis system determines in real time whether there are any anomalies in the valid measurement data during the comparison and analysis process. If the data is determined to be abnormal, the system will trigger the local preset audible and visual alarm device to issue an alarm prompt.

[0015] The wafer positioning and correction compensation steps include:

[0016] During the initial positioning, the sensor group is activated to collect the reference position data of the wafer and transmit the reference position data to the AI ​​vision processing module.

[0017] The AI ​​vision processing module calls a preset deep learning model to identify the notches or positioning edges of the wafer, dynamically generates three-dimensional coordinate compensation values ​​based on the identification results, and then drives the calibration stage to perform the calibration compensation of the wafer based on the three-dimensional coordinate compensation values.

[0018] In the film thickness data acquisition and processing step, while the spectrometer acquires the original measurement data of the wafer epitaxial film, the X-ray reflectivity measurement (XRR) device is simultaneously activated to acquire the density data and roughness data of the epitaxial film. The density data, roughness data and the original measurement data are then transmitted together to the edge enhancement algorithm module, and the effective measurement data is corrected through multi-dimensional data fusion.

[0019] In the wafer positioning and correction compensation step, when the sensor group collects wafer reference position data, it simultaneously collects wafer warpage data and transmits it to the AI ​​vision processing module; the AI ​​vision processing module performs collaborative calculations on the warpage data and the three-dimensional coordinate compensation value.

[0020] When the wafer to be measured is a GaN epitaxial film wafer, in the film thickness data acquisition and processing step, the integral intensity of the 0002 diffraction peak of the GaN epitaxial film and the integral intensity of the 0006 diffraction peak of the sapphire substrate are additionally acquired by a diffraction intensity measurement device, the intensity ratio of the two is calculated, and the intensity ratio is incorporated into the edge enhancement algorithm module as a correction parameter.

[0021] In the film thickness data acquisition and processing step, when the measuring stage moves along the optimized measurement path, the high-resolution X-ray diffraction device is simultaneously activated to acquire the composition data, defect data, and strain data of the epitaxial film. The composition data, defect data, and strain data are transmitted to the cloud-based collaborative analysis system. The cloud-based collaborative analysis system combines this data with the effective measurement data to supplement and correct the film thickness trend prediction report.

[0022] In a second aspect of the present invention, to implement the method described in the first aspect, a wafer epitaxial film thickness measurement system based on trajectory adaptation is also proposed. The measurement system includes a loading and transfer unit, a positioning and compensation unit, a measurement path planning unit, a film thickness data acquisition and processing unit, a data comparison and trend report generation unit, and an abnormal alarm triggering unit. The positioning and compensation unit includes a calibration mechanism and a compensation mechanism.

[0023] After the loading and transfer unit loads the wafer to be measured, it transfers the wafer to the calibration mechanism through a preset turnover mechanism.

[0024] The positioning and compensation unit performs preliminary positioning of the wafer transferred to the correction mechanism through the positioning component of the correction mechanism. The compensation mechanism drives the correction stage to move based on the three-dimensional coordinate compensation value to complete the correction and compensation of the wafer.

[0025] The measurement path planning unit transfers the wafer that has completed the correction and compensation to the measurement mechanism. The trajectory adaptive algorithm built into the measurement mechanism automatically plans an optimized measurement path based on the actual size of the current wafer and controls the measurement stage to move along the optimized measurement path.

[0026] During the movement of the measurement stage, the film thickness data acquisition and processing unit starts the spectrometer to acquire the original measurement data of the wafer epitaxial film and transmits the original measurement data to the edge enhancement algorithm module in real time. The edge enhancement algorithm module eliminates the edge diffraction interference in the original measurement data to obtain effective measurement data.

[0027] The data comparison and trend report generation unit transmits the effective measurement data to the cloud collaborative analysis system. The cloud collaborative analysis system calls the preset historical database to compare and analyze the effective measurement data with the epitaxial film thickness data of the same type of wafer in the historical database, and generates a wafer epitaxial film thickness trend prediction report based on the comparison results.

[0028] The cloud-based collaborative analysis system determines in real time whether there are any anomalies in the valid measurement data during the comparison and analysis process. If the data is determined to be abnormal, the anomaly alarm triggering unit triggers the locally preset audible and visual alarm device to issue an alarm prompt.

[0029] The positioning and compensation unit also includes an AI vision processing module and an X-ray reflectivity measurement (XRR) device;

[0030] When the positioning component of the calibration mechanism performs initial positioning, it simultaneously activates the sensor group to collect the reference position data of the wafer and transmits the reference position data to the AI ​​vision processing module.

[0031] The AI ​​vision processing module calls a preset deep learning model to identify the notch or positioning edge of the wafer. The compensation mechanism dynamically generates a three-dimensional coordinate compensation value based on the identification result, and then drives the calibration table of the calibration mechanism to perform the calibration compensation of the wafer based on the three-dimensional coordinate compensation value.

[0032] Furthermore, when the sensor group collects wafer reference position data, it simultaneously collects wafer warpage data and transmits it to the AI ​​vision processing module; the AI ​​vision processing module performs collaborative calculations on the warpage data and the three-dimensional coordinate compensation value.

[0033] As the measuring stage moves along the optimized measuring path, the high-resolution X-ray diffraction device is simultaneously activated to collect the composition data, defect data, and strain data of the epitaxial film. The composition data, defect data, and strain data are then transmitted to the cloud-based collaborative analysis system. The cloud-based collaborative analysis system combines this data with the effective measurement data to supplement and correct the film thickness trend prediction report.

[0034] In a third aspect of the invention, a visualization measurement device is also proposed, the visualization measurement device including a human-computer interaction visualization interface, through which a wafer epitaxial film thickness measurement method based on trajectory adaptation as described in the first aspect is executed, and a wafer epitaxial film thickness trend prediction report is displayed on the human-computer interaction visualization interface.

[0035] Compared with the prior art, the present invention has at least the following advantages:

[0036] First, by employing a dual approach of "positioning correction + edge enhancement algorithm," the technical solution of this invention can significantly improve the accuracy of film thickness measurement. In the positioning stage, a dedicated correction and compensation mechanism is set up. The wafer transferred to the calibration stage is initially positioned, and then the calibration stage is driven based on the three-dimensional coordinate compensation value, precisely eliminating positional deviations during wafer transfer and avoiding measurement errors caused by the lack of a correction mechanism in traditional equipment. Simultaneously, in the data acquisition stage, the edge enhancement algorithm module processes the raw data acquired by the spectrometer in real time, specifically eliminating diffraction interference at the wafer edges, ensuring the accuracy of effective measurement data, and solving the pain point of spectral signal distortion affecting results in traditional technologies.

[0037] Secondly, the technical solution proposed in this invention also relies on a "trajectory adaptive algorithm + flexible transfer mechanism" to achieve efficient measurement of wafers of different sizes. After the wafer has been calibrated and compensated, it is transferred to the measurement mechanism. The trajectory adaptive algorithm built into the measurement mechanism will automatically plan an optimized measurement path based on the actual size of the current wafer, without the need for manual adjustment of the fixed path. This design breaks the limitations of traditional equipment that relies on multiple displacements of the XY module and can only adapt to wafers of specific sizes. It reduces the ineffective movement of the measurement stage, shortens the time of a single measurement, and is compatible with multiple wafer specifications, greatly improving the equipment's versatility and perfectly matching the dual requirements of semiconductor mass production lines for measurement efficiency and adaptability.

[0038] Finally, this invention utilizes a "cloud-based collaborative analysis system + real-time alarm device" to construct a fully intelligent data processing and anomaly response system. After effective measurement data is transmitted to the cloud-based collaborative analysis system, the system calls upon a preset historical database to compare and analyze the current data with film thickness data for similar wafers. This not only generates film thickness trend prediction reports, providing data support for production process optimization, but also determines in real-time whether the data is abnormal. Once an anomaly is detected, a local audible and visual alarm device is immediately triggered, solving the problems of slow data processing and inability to provide real-time warnings in traditional systems. This helps staff quickly intervene and make adjustments, reducing defective products and ensuring wafer production yield.

[0039] Further specific advantages and implementation principles of the present invention will be further detailed in the specific embodiments section in conjunction with the accompanying drawings. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a schematic diagram of the main execution flow of a wafer epitaxial film thickness measurement method based on trajectory adaptation according to an embodiment of the present invention;

[0042] Figure 2 yes Figure 1 A schematic diagram illustrating the specific steps of the wafer epitaxial film thickness measurement method based on trajectory adaptation;

[0043] Figure 3 This is a schematic diagram of the hardware unit composition of a wafer epitaxial film thickness measurement system based on trajectory adaptation according to an embodiment of the present invention;

[0044] Figure 4 yes Figure 3 A schematic diagram of some substructures of the trajectory-adaptive wafer epitaxial film thickness measurement system. Detailed Implementation

[0045] In the specific embodiments of this application, if the embodiments of the relevant technical solutions involve user-related data, then when the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0046] It should be clarified that the specific embodiments of the present invention belong to the fields of measurement and testing and semiconductor manufacturing technology.

[0047] Before introducing specific embodiments of the present invention, the existing technology and related problems of the present invention will be introduced first, so as to lead to the technical solution of the present invention, so as to better understand the improvement advantages and inventiveness of the present application.

[0048] In the field of semiconductor wafer epitaxial film thickness measurement and semiconductor manufacturing, existing technologies mainly revolve around spectral reflectance method and elliptic polarization method to build measurement systems. For example, the "Semiconductor Wafer Film Thickness Measurement Device" disclosed in Chinese invention patent application CN202411201588.7 completes wafer film thickness data acquisition by cooperating with a fixed measurement stage and spectral detection module.

[0049] However, this type of technology has obvious shortcomings in practical applications:

[0050] First, the positioning accuracy is insufficient. Existing equipment mostly relies on manual positioning or simple mechanical limiting, lacking a dedicated calibration and compensation mechanism. After the wafer is transferred by the turnover mechanism, the position is prone to shift, resulting in misalignment between the measurement probe and the wafer detection area, producing a measurement error of more than ±10%, which cannot meet the high precision requirements of advanced processes.

[0051] Secondly, the measurement path adaptability is poor. Traditional equipment uses preset fixed paths. For example, the detection trajectory set for 8-inch wafers cannot be adapted to 12-inch wafers. It is necessary to manually readjust the parameters and adjust the module movement trajectory, which is not only time-consuming (more than 30 minutes for a single adjustment), but also results in a large number of invalid movements of the measurement stage due to insufficient path optimization, reducing measurement efficiency and making it difficult to meet the detection requirements of mass production lines of "more than 30 wafers per hour".

[0052] Third, data processing and feedback are lagging. For example, the "Wafer Film Thickness Measurement System" described in Chinese invention patent application CNCN202410560289.8 can only perform basic filtering on single measurement data, lacking a collaborative comparison mechanism with historical data, and thus unable to identify film thickness change trends. Furthermore, anomaly detection relies on manual review, taking an average of 2 hours from data acquisition to anomaly detection, during which a large number of defective wafers are generated, severely impacting production yield. In addition, existing technologies generally lack dedicated processing algorithms for wafer edge diffraction interference; interference signals mixed in the raw data can cause spectral curve distortion, further reducing data accuracy. The aforementioned problems collectively restrict the accuracy, efficiency, and intelligence level of semiconductor wafer epitaxial film thickness measurement and manufacturing. There is an urgent need for an innovative technical solution that can simultaneously solve positioning deviation, path adaptation, data collaboration, and interference elimination. This invention is a systematic improvement solution proposed to address the above pain points.

[0053] Figure 1The main steps of an embodiment of the method of the present invention are shown in the diagram. Figure 1 The method mainly includes wafer loading and initial transfer steps, wafer positioning and correction compensation steps, measurement path planning steps, film thickness data acquisition and processing steps, data comparison and trend report generation steps, and abnormal alarm triggering steps.

[0054] exist Figure 1 On this basis, Figure 2 Further show Figure 1 A schematic diagram illustrating the specific steps of the wafer epitaxial film thickness measurement method based on trajectory adaptation.

[0055] Combination Figures 1-2 A specific embodiment of the wafer epitaxial film thickness measurement method based on trajectory adaptation of the present invention is summarized as follows:

[0056] Wafer loading and initial transfer: After loading the wafer to be measured, the wafer is transferred to the calibration mechanism through a preset turnover mechanism;

[0057] Wafer positioning and calibration compensation: The positioning component of the calibration mechanism performs initial positioning of the wafer transferred to the calibration mechanism, and drives the calibration stage to move based on the three-dimensional coordinate compensation value to complete the wafer calibration compensation;

[0058] Measurement path planning: The wafer that has completed the calibration and compensation is transferred to the measurement mechanism. The trajectory adaptive algorithm built into the measurement mechanism automatically plans an optimized measurement path based on the actual size of the current wafer, and controls the measurement stage to move along the optimized measurement path.

[0059] Film thickness data acquisition and processing: During the movement of the measurement stage, the spectrometer is activated to acquire the raw measurement data of the wafer epitaxial film, and the raw measurement data is transmitted to the edge enhancement algorithm module in real time. The edge enhancement algorithm module eliminates the edge diffraction interference in the raw measurement data to obtain effective measurement data.

[0060] Data comparison and trend report generation: The effective measurement data is transmitted to the cloud-based collaborative analysis system. The cloud-based collaborative analysis system calls a preset historical database to compare and analyze the effective measurement data with the epitaxial film thickness data of the same type of wafer in the historical database. Based on the comparison results, a wafer epitaxial film thickness trend prediction report is generated.

[0061] Anomaly alarm triggering: The cloud-based collaborative analysis system determines in real time whether there are any anomalies in the valid measurement data during the comparison and analysis process. If the data is determined to be abnormal, the system will trigger the local preset audible and visual alarm device to issue an alarm prompt.

[0062] Next, to better describe the above method embodiments, this section will elaborate on each step. The elaborated description will incorporate specific implementation structures, unit names, state parameters, such as specific values ​​and specific sensor types. It is understood that these specific examples are merely illustrative and not exhaustive, nor do they constitute a specific limitation on the technical solution of this invention. In actual manufacturing, those skilled in the art can select other sensor models and configure other types of values / parameters according to actual conditions.

[0063] First, the method begins with the wafer loading and preliminary transfer steps: after loading the wafer to be measured, the wafer is transferred to the calibration mechanism through a preset turnover mechanism.

[0064] In one specific embodiment, this step is completed using a fully automated wafer loading unit, which includes a wafer cassette storage rack, a robotic arm gripping mechanism, and a vision positioning component. The wafer cassette storage rack has layered slots inside, accommodating two mainstream wafer cassette sizes: 12-inch (300mm diameter) and 8-inch (200mm diameter). The inner walls of the slots are lined with a polytetrafluoroethylene (PTFE) buffer layer to prevent direct contact between the wafers and metal, thus avoiding scratches.

[0065] The robotic arm uses a SCARA horizontal multi-joint robot, with a vacuum suction cup assembly (containing three 20mm diameter silicone suction cups) at its end. The vacuum level of the suction cups is monitored in real time by a negative pressure sensor. For example, when the vacuum level drops below -75kPa, the system automatically triggers an alarm to prevent the wafer from falling off. The vision positioning assembly consists of two industrial cameras and matching optical lenses, installed directly above and to the side of the robotic arm's gripping station. The upper camera is used to identify the layer number and center coordinates of the wafer within the wafer cassette, while the side camera is used to detect the flatness of the wafer edge, ensuring that the gripping position deviation does not exceed ±0.1mm.

[0066] During loading, the system first reads the model and batch information of the wafers in the wafer box through the radio frequency identification (RFID) module of the wafer box storage rack. Then, based on the coordinates fed back by the vision positioning component, the robotic arm drives the vacuum suction cup to move directly above the target wafer, descends to a distance of 2mm from the wafer surface and starts vacuum adsorption. After confirming that the adsorption is stable (the negative pressure sensor shows that the vacuum level meets the standard), the robotic arm rises to a safe height (e.g., 50mm from the top of the wafer box) and then moves horizontally to the receiving platform of the turnover mechanism to complete the loading operation.

[0067] The transfer mechanism uses a synchronous belt drive conveyor line with a length of 1.2m. The effective conveying width can be adaptively adjusted by adjusting baffles on both sides to accommodate the transfer needs of wafers of different sizes. The surface of the conveyor line is covered with an anti-static polyurethane belt (2mm thick, coefficient of friction 0.3), and three sets of guide wheels (400mm apart) are installed under the belt to ensure that the belt runout does not exceed ±0.05mm.

[0068] The conveyor drive motor is a servo motor equipped with a planetary reducer. The motor speed is fed back in real time by a pulse encoder (1000 lines resolution). The system sets the conveying speed according to the wafer specifications. For example, the speed is 50 mm / s when transporting 12-inch wafers and 40 mm / s when transporting 8-inch wafers to avoid the wafers slipping on the belt due to excessive speed.

[0069] Three photoelectric sensors are installed along the conveyor line, located at the loading and receiving position, the intermediate detection position, and the alignment mechanism docking position: the sensor at the loading and receiving position confirms that the robotic arm has placed the wafer stably on the conveyor line; the sensor at the intermediate detection position confirms the wafer's position again. For example, if a wafer offset exceeding ±0.5mm is detected, the conveyor line immediately stops and awaits manual adjustment; the sensor at the alignment mechanism docking position triggers the conveyor line to decelerate. When the wafer is 100mm away from the alignment mechanism's receiving table, the conveyor speed drops to 10mm / s until the wafer fully reaches the alignment mechanism's positioning position, at which point the conveyor line stops, completing the initial transfer.

[0070] Next, the wafer positioning and calibration compensation steps are performed: the positioning component of the calibration mechanism performs initial positioning of the wafer transferred to the calibration mechanism, and drives the calibration stage to move based on the three-dimensional coordinate compensation value to complete the wafer calibration compensation.

[0071] Specifically, the wafer positioning and correction compensation steps include:

[0072] During the initial positioning, the sensor group is activated to collect the reference position data of the wafer and transmit the reference position data to the AI ​​vision processing module.

[0073] The AI ​​vision processing module calls a preset deep learning model to identify the notches or positioning edges of the wafer, dynamically generates three-dimensional coordinate compensation values ​​based on the identification results, and then drives the calibration stage to perform the calibration compensation of the wafer based on the three-dimensional coordinate compensation values.

[0074] Overall, the wafer positioning and calibration compensation steps are the core steps to eliminate wafer transfer deviations and ensure the accuracy of subsequent measurements. Through the collaborative process of "sensor group data acquisition + AI vision processing + calibration table precision action", the precise adjustment of the wafer's three-dimensional orientation is achieved.

[0075] In one specific embodiment, the positioning component of the calibration mechanism is equipped with multiple types of sensor groups, including 3 sets of laser displacement sensors, 2 sets of industrial CCD cameras and 1 set of pressure sensors, forming a multi-dimensional data acquisition system of "optics + mechanics".

[0076] Among them, the laser displacement sensor has a measurement range of 0~300mm and an accuracy of ±0.003mm. It is installed on the left front, right rear and center positions above the calibration table. The left front and right rear sensors are distributed at a 90° angle to collect the height data of the upper surface of the wafer, and the center sensor is used to capture the planar position deviation of the wafer center area.

[0077] The industrial CCD camera has a resolution of 5 megapixels and is equipped with an 8mm fixed-focus lens. It is mounted on both sides of the calibration stage to capture the notch on the edge of the wafer (such as the notch on a 12-inch wafer) or the positioning edge (such as the flat surface on an 8-inch wafer) to obtain the angular offset data of the wafer.

[0078] The pressure sensor is installed below the bearing surface of the calibration stage, with a range of 0~10N and an accuracy of ±0.01N. It is used to monitor the pressure distribution when the wafer is placed and to determine whether the wafer is stably attached to the calibration stage.

[0079] After the transfer mechanism moves the wafer to the calibration table, the system simultaneously activates the sensor group: the laser displacement sensor collects height data at 100 points on the upper surface of the wafer at a sampling frequency of 500Hz, generating a flatness curve of the wafer; the industrial CCD camera, assisted by a ring light source (wavelength 650nm, illuminance 1000lux), captures 20 consecutive images of the wafer edge to ensure complete capture of notch or positioning edge features; the pressure sensor provides real-time feedback on the contact pressure between the wafer and the calibration table. When the pressure value stabilizes at 3~5N and the fluctuation is less than ±0.1N, it is determined that the wafer is placed stably, the sensor group stops data acquisition, and all reference position data (height data, image data, pressure data) are packaged and transmitted to the AI ​​vision processing module.

[0080] The AI ​​vision processing module is mounted on an industrial control computer and incorporates a wafer feature recognition algorithm trained on a ResNet-50 deep learning model. This model was trained using 100,000 sets of wafer image samples of different specifications and orientations, achieving an accuracy of 99.8% in notch / positioning edge recognition, with a recognition time of less than 0.5 seconds (using an industrial control computer with multiple GPUs).

[0081] After receiving the reference position data, the AI ​​vision processing module first preprocesses the image captured by the industrial CCD camera: Gaussian filtering (convolution kernel size 5×5) is used to eliminate image noise, and then the Canny edge detection algorithm (threshold range 50~150) is used to extract the wafer edge contour. Subsequently, a deep learning model is called to locate the notch or positioning edge in the contour and calculate the current angular offset of the wafer (such as the angle deviation between the notch and the preset reference line). At the same time, the module performs fitting analysis on the height data collected by the laser displacement sensor, generates the spatial plane equation of the wafer through the least squares method, and compares it with the preset standard plane (calibration stage reference plane) to obtain the height deviation of the wafer in the Z-axis direction (such as the height difference between the highest and lowest points) and the planar position deviation in the X and Y-axis directions (such as the coordinate deviation between the center of the circle and the center of the calibration stage).

[0082] Based on the above analysis, the AI ​​vision processing module dynamically generates three-dimensional coordinate compensation values ​​according to the principle of "angle priority, height second, and plane fine-tuning": the angle compensation value ranges from -5° to +5° (accuracy ±0.001°), used to adjust the wafer's rotation attitude; the Z-axis height compensation value ranges from -0.5mm to +0.5mm (accuracy ±0.0005mm), used to adjust the lifting height of the calibration stage; and the X and Y-axis plane compensation values ​​range from -2mm to +2mm (accuracy ±0.001mm), used to adjust the horizontal displacement of the calibration stage. After the compensation values ​​are generated, the module transmits the data in real time to the calibration stage's drive control system via EtherCAT industrial Ethernet (transmission rate 100Mbps).

[0083] The calibration table adopts a three-dimensional precision adjustment platform, which consists of four motion axes: X-axis, Y-axis, Z-axis and θ-axis (rotation axis). Each axis is equipped with a servo motor (model: Yaskawa SGMAV-01ADA61) and a ball screw (lead 1mm, accuracy C3 level), and is equipped with a grating ruler (resolution 0.1μm) for closed-loop position control to ensure motion accuracy.

[0084] After receiving the three-dimensional coordinate compensation value, the drive control system drives the calibration table to operate according to the preset action sequence:

[0085] First, the θ-axis servo motor drives the calibration stage to rotate according to the angle compensation value until the wafer notch or positioning edge is aligned with the baseline. The grating ruler provides real-time feedback on the rotation angle, and the rotation stops when the deviation is less than ±0.001°. Second, the Z-axis servo motor drives the calibration stage to rise and fall according to the height compensation value, adjusting the upper surface of the wafer to the standard measurement height. The laser displacement sensor monitors the height change in real time, and the rising and falling stops when the height deviation is less than ±0.0005mm. Finally, the X-axis and Y-axis servo motors drive the calibration stage to move horizontally according to the plane compensation value, aligning the wafer center with the calibration stage center. The industrial CCD camera captures the wafer position in real time, and the movement stops when the plane deviation is less than ±0.001mm.

[0086] After calibration, the system starts the sensor group to collect data a second time. The AI ​​vision processing module verifies the newly collected data. If all deviations are within the preset allowable range (e.g., angle deviation ≤ 0.002°, height deviation ≤ 0.001mm, plane deviation ≤ 0.002mm), the calibration is deemed qualified and the next measurement stage begins. If the deviation exceeds the allowable range, the system is triggered to recalibrate, repeating up to 3 times. If it is still unqualified, an audible and visual alarm is triggered, prompting staff to check the equipment or wafer status.

[0087] In a more preferred embodiment, when the sensor group acquires wafer reference position data, it simultaneously acquires wafer warpage data and transmits it to the AI ​​vision processing module; the AI ​​vision processing module performs collaborative calculations on the warpage data and the three-dimensional coordinate compensation value.

[0088] At this point, based on the original sensor group, four new laser triangulation sensors (using Keyence LK-G5000 series, measurement range 0~50mm, linear accuracy ±0.001mm, sampling frequency 1000Hz) are added to form a "7-point distributed sampling array" with the original three laser displacement sensors.

[0089] The four newly added sensors are installed on the left rear, right front, left upper, and right lower positions above the calibration platform, respectively, in a regular hexagonal distribution with the original center sensor. The spacing between adjacent sensors is equal (100mm for 12-inch wafers; 67mm for 8-inch wafers) to ensure coverage of the effective area of ​​the wafer (90% of the diameter).

[0090] Once the transfer mechanism moves the wafer to the calibration table and the pressure sensor determines that the wafer is stably bonded, the system simultaneously activates all seven sets of laser displacement sensors: the original three sets of sensors collect wafer surface height data at a sampling frequency of 500Hz, and the newly added four sets of laser triangulation displacement sensors collect real-time height values ​​at corresponding points at a high-frequency sampling frequency of 1000Hz. Each set of sensors continuously collects 200 data points, and after removing the maximum and minimum values, the average value is taken to obtain accurate height data for the seven sampling points. Simultaneously, an industrial CCD camera captures a 360° panoramic image of the wafer edge (captured from 12 angles, with three images taken from each angle). Image stitching technology is used to generate a complete wafer edge contour, assisting in determining the distribution of warpage (such as central depression, edge upturn, etc.).

[0091] All warp-related data (average height of 7 sampling points, edge contour deformation features) and the original baseline data (planar position deviation, angle offset, pressure distribution) are synchronously transmitted to the AI ​​vision processing module via a dual-channel data transmission protocol (EtherCAT+UDP). The transmission delay is controlled within 5ms to ensure data timing consistency and avoid calculation errors caused by asynchronous data transmission.

[0092] After receiving the warpage data, the AI ​​vision processing module first constructs a wafer warpage model: based on the height data of 7 sampling points, it uses a quadratic surface fitting algorithm (formula: z=ax²+by²+cxy+dx+ey+f, where x and y are the wafer plane coordinates, and z is the height value) to calculate the warpage surface equation of the wafer surface, thereby obtaining the maximum warpage (the height difference between the highest and lowest points) and the warpage center coordinates (the geometric center of the warped surface). Simultaneously, through edge contour deformation features, it determines the warpage type (e.g., "bowl-shaped" warpage: low center, high edge; "arch-shaped" warpage: high center, low edge; "S-shaped" warpage: alternating concave and convex areas in local regions) and marks severely warped areas (areas with warpage > 50μm).

[0093] Subsequently, the module initiates a collaborative calculation process: First, it compares the X and Y axis plane compensation values ​​in the 3D coordinate compensation with the warp center coordinates. If the deviation between the warp center and the wafer geometric center is >0.1mm, the plane compensation value is adjusted so that the calibration stage prioritizes moving the warp center to the center area of ​​the calibration stage, avoiding contact between the probe and the severely warped area during subsequent measurements due to warp center offset. Second, for the Z-axis height compensation value, the compensation strategy is adjusted in conjunction with the maximum warp amount: if it is a "bowl-shaped" warp and the maximum warp amount is 30~50... The Z-axis compensation reference is adjusted from "average wafer height" to "warping center height" and a dynamic correction of ±5μm is added to the compensation value to ensure that the distance between the measurement probe and the wafer surface is always within the optimal measurement range of the spectrometer (200±10μm). If the warping is "S-shaped" and the local warping is >50μm, the module will mark the area as "non-measurement area" and automatically avoid it in subsequent measurement path planning. At the same time, the angle compensation value is adjusted so that the wafer rotates to the area with gentle warping and faces the initial position of the measurement probe.

[0094] After collaborative calculation, the module generates a combined 3D coordinate compensation value consisting of a "basic compensation value + warpage correction amount," with the warpage correction amount controlled to an accuracy of ±0.003mm. It also simultaneously outputs a warpage analysis report (including warpage type, maximum warpage amount, warpage center coordinates, and non-measurement area location), which is transmitted to the calibration stage drive control system and subsequent measurement path planning unit. During operation, the calibration stage dynamically adjusts the speed of each axis based on the warpage correction amount: in the adjustment phase corresponding to severely warped areas, the servo motor speed is reduced by 30% (e.g., from 5mm / s to 3.5mm / s), and position data is fed back via a grating ruler at a frequency of 2000Hz to ensure smooth calibration and prevent calibration stage overload or wafer damage caused by wafer warpage.

[0095] After calibration, in addition to routine deviation verification, the system can also collect warpage data again using the newly added laser triangular displacement sensor. By comparing the change in the maximum warpage before and after calibration, if the deviation between the warpage center and the calibration platform center is ≤0.05mm and the maximum warpage in the measurement area is ≤30μm, the warpage adaptation calibration is deemed qualified. If it does not meet the standard, the system will recalculate based on the new warpage data, adjust the compensation value, and calibrate again until the measurement requirements are met, further ensuring the accuracy of subsequent film thickness measurement data.

[0096] Based on this, the measurement path planning step is initiated: the wafer that has completed the calibration and compensation is transferred to the measurement mechanism, and the trajectory adaptive algorithm built into the measurement mechanism automatically plans the optimized measurement path according to the actual size of the current wafer, and controls the measurement stage to move along the optimized measurement path.

[0097] In a specific example, after the calibration compensation is completed, the system first confirms the wafer status a second time using the pressure sensor and laser displacement sensor of the calibration stage (warpage center deviation ≤ 0.05mm, warpage amount in the measurement area ≤ 30μm). After confirmation, the wafer transfer robotic arm (using the ABBIRB1200 series, with repeatability ±0.01mm) is activated. The end of the robotic arm is equipped with customized vacuum chucks (4 silicone chucks with a diameter of 15mm, arranged in a square, suitable for gripping the edges of wafers of different sizes). The chuck spacing is automatically adjusted according to the wafer size data (12 inches / 8 inches) transmitted by the AI ​​vision processing module (12-inch wafer spacing is set to 80mm, and 8-inch wafer spacing is set to 55mm).

[0098] During the transfer process, the robotic arm uses a vision guidance system (composed of two Baslerac A1920-40gm industrial cameras installed on both sides of the measurement mechanism entrance) to capture the wafer's position in real time and dynamically adjust its trajectory: after picking up the wafer from the calibration table, it first rises to a safe height (200mm from the equipment table surface), then moves horizontally to directly above the measurement platform of the measurement mechanism. During the descent, the alignment deviation between the wafer center and the measurement platform center is controlled within ±0.02mm using vision guidance. The measurement platform surface is covered with an anti-static rubber pad (3mm thick, coefficient of friction 0.4) and has three built-in electromagnetic adsorption devices. When the wafer contacts the measurement platform, the electromagnetic adsorption devices are immediately activated (the adsorption force is set to 8~10N, automatically adjusted according to the wafer thickness) to ensure that the wafer does not shift during the measurement process. After the transfer is completed, the robotic arm resets, waiting for the next operation.

[0099] The trajectory adaptive algorithm built into the measuring mechanism runs on the embedded industrial controller. After the algorithm is started, it first receives multi-dimensional input data, including wafer basic parameters (size, model, number of preset measurement points), pre-calibration data (three-dimensional coordinate compensation results, warpage analysis report), and equipment parameters (effective measurement range of spectrometer probe is 3mm×3mm, maximum movement speed of measuring stage is 10mm / s).

[0100] The core computational process of the algorithm consists of three steps:

[0101] Basic path framework generation: The measurement area boundary is determined according to the wafer size (the measurement radius is set to 145mm for 12-inch wafers and 95mm for 8-inch wafers, both avoiding the 5mm invalid area at the edge). The measurement points are preset in a "ring-shaped distributed" layout: 36 measurement points are set for 12-inch wafers (divided into 6 rings, 6 points in each ring, with a spacing of 25mm between adjacent rings), and 24 measurement points are set for 8-inch wafers (divided into 4 rings, 6 points in each ring, with a spacing of 20mm between adjacent rings), forming the basic path framework.

[0102] Warp region avoidance optimization: Call the "non-measurement area" data in the warp analysis report (such as the local area with warp > 50μm in S-shaped warp), and use the polygon collision detection algorithm to determine whether the measurement point in the base path falls into the non-measurement area. If there is an overlap, the position of the point is automatically adjusted (offset 3~5mm along the tangent of the ring to ensure that it is still in the same ring and far away from the warp area after the offset). At the same time, the ring completely covered by the warp area is deleted, and the measurement points are redistributed to other rings to ensure that the total number of measurement points remains unchanged.

[0103] Motion trajectory smoothing optimization: Based on the adjusted coordinates of the measurement points, the motion trajectory of the measuring stage is generated using a Bezier curve interpolation algorithm, so that the motion path between adjacent measurement points is a smooth curve (radius of curvature ≥ 10 mm), avoiding vibration of the measuring stage caused by right-angle turns; at the same time, the motion speed is dynamically adjusted according to the distance between measurement points: when the distance between adjacent points is > 20 mm, the speed is set to 8~10 mm / s, and when the distance is ≤ 20 mm, the speed is reduced to 4~6 mm / s, ensuring that the spectrometer has enough time to collect data (data collection time for each measurement point is 0.5 seconds). After the algorithm completes its calculations, it generates an optimized path file containing "measurement point coordinates, motion speed curve, and path priority," which is then transmitted to the measurement station drive system via an RS485 serial port (115200bps baud rate). The entire calculation process takes ≤2 seconds, meeting the high-efficiency measurement requirements of mass production lines.

[0104] The measuring table adopts an XY axis precision motion platform (stroke: X-axis 300mm, Y-axis 300mm, positioning accuracy ±0.005mm, repeatability ±0.002mm). Each axis is equipped with a servo motor (model: Panasonic A6 series MSMF012L1U2M) and a ball screw (lead 2mm, accuracy C5 grade). After receiving the optimized path file, the drive system executes the motion commands in the order of "inner circle first, then outer circle, clockwise direction".

[0105] During the movement, the system ensures accuracy through closed-loop feedback control: the X and Y axes of the measuring platform are equipped with grating rulers (0.1μm resolution) to provide real-time feedback of the current position coordinates and compare them with the target coordinates in the path file. If the deviation is >0.003mm, the drive system immediately adjusts the output torque of the servo motor to correct the position deviation. At the same time, the vibration sensor built into the measuring platform monitors the vibration amplitude in real time. When the vibration is >0.01mm, the movement speed is automatically reduced by 20% until the vibration returns to normal. When the measuring stage moves to each measuring point, the system triggers the spectrometer to collect data. After the data collection is completed, an audio-visual prompt is given (a green indicator light illuminates and a buzzer sounds briefly). The system then continues to move along the optimized path to the next measuring point until all measuring points have completed data collection. The measuring stage then automatically resets to its initial position (X=0mm, Y=0mm) and waits for the next wafer to be transferred.

[0106] Next, the formal process of film thickness data acquisition and processing begins: During the movement of the measurement stage, the spectrometer is activated to acquire the raw measurement data of the wafer epitaxial film, and the raw measurement data is transmitted in real time to the edge enhancement algorithm module. The edge enhancement algorithm module eliminates edge diffraction interference in the raw measurement data to obtain effective measurement data.

[0107] Preferably, in the film thickness data acquisition and processing step, while the spectrometer acquires the original measurement data of the wafer epitaxial film, the X-ray reflectivity measurement (XRR) device is simultaneously activated to acquire the density data and roughness data of the epitaxial film, and the density data, roughness data and the original measurement data are transmitted together to the edge enhancement algorithm module to correct the effective measurement data through multi-dimensional data fusion.

[0108] Understandably, in actual execution, XRR scans typically take several minutes, and there will be a certain delay (generally only 50ms) in the "synchronous" completion of the spectrometer at each point. Therefore, "synchronous start of X-ray reflectivity measurement (XRR) device" here can be understood as parallel start.

[0109] In one specific embodiment, when the measuring stage moves along the optimized path, the system synchronously triggers the spectrometer to start acquisition via IO signal. Its probe is fixed 150mm above the measuring stage by a mechanical bracket, and the probe's field of view is set to 10° to ensure that the measurement area and the preset measurement point on the measuring stage's movement trajectory are precisely coincident (the spot diameter is 3mm, matching the area of ​​the measurement point).

[0110] Based on the optical properties of the epitaxial film material (such as silicon epitaxial film and gallium nitride epitaxial film), the system presets different acquisition parameters: for silicon epitaxial film, the integration time is set to 50ms, and the average value is taken from 3 samples to avoid signal oversaturation; for gallium nitride epitaxial film, due to its lower reflectivity, the integration time is adjusted to 100ms, and the average value is taken from 5 samples to improve signal strength. During the acquisition process, the spectrometer records the reflectance spectral data of each measurement point in real time (including the correspondence between wavelength and reflectance intensity), forming raw measurement data in CSV file format, containing fields such as "measurement point number, acquisition time, wavelength sequence, and reflectance intensity sequence".

[0111] The raw data is transmitted in real time to the edge enhancement algorithm module via gigabit Ethernet at a transmission rate of ≥100Mbps. Once the data acquisition of each measurement point is completed, a data transmission is triggered immediately, with the transmission delay controlled within 100ms to ensure synchronization with the movement rhythm of the measurement station and avoid data accumulation.

[0112] The edge enhancement algorithm module is deployed on an industrial control computer (which can share hardware with the AI ​​vision processing module). It adopts an edge interference cancellation algorithm based on wavelet transform, and the core process consists of three steps:

[0113] Raw data preprocessing: Baseline correction is performed on the received reflectance spectral data. Baseline drift caused by the spectrometer's own dark current and ambient light interference is eliminated by polynomial fitting (using a 5th-order polynomial). At the same time, noise filtering is performed by using Gaussian filtering (3×3 convolution kernel size) to smooth high-frequency noise and retain effective spectral features.

[0114] Edge diffraction interference identification: Extract the "edge feature segment" in the preprocessed spectral data - when the measurement point is close to the edge of the wafer (5~10mm away from the edge), the reflection spectrum will show periodic fluctuations (diffraction fringes). The algorithm calculates the first derivative of the spectrum, locates the wavelength range where the absolute value of the derivative increases sharply (i.e., the diffraction interference range), and marks the abnormal values ​​of reflection intensity in this range.

[0115] Interference signal reconstruction and elimination: For the marked diffraction interference region, the spectral data of the adjacent non-interference region is used to reconstruct the reflection intensity value of the region by linear interpolation to replace the original outlier value; at the same time, an "edge distance correction coefficient" is introduced - the closer to the wafer edge, the larger the correction coefficient (range 1.05~1.2) to ensure that the reconstructed data matches the spectral characteristics corresponding to the actual film thickness.

[0116] After the algorithm is processed, it outputs valid measurement data containing "corrected wavelength-reflection intensity sequence, interference cancellation rate (≥90%), and data reliability (≥95%)", which is stored in a local database (MySQL database, data read / write speed ≥100MB / s) and backed up to the temporary data area of ​​the cloud collaborative analysis system. In a more preferred example, a Bruker D8DISCOVER X-ray reflectivity measurement device (XRR) can be added to the conventional procedure. Its X-ray source is CuKα rays (wavelength 0.154nm, power 50kV×100mA), and the detector is a two-dimensional semiconductor detector (resolution 1024×1024 pixels). The device is installed above the measurement stage at a 45° angle with the spectrometer probe to ensure that the measurement area completely overlaps with the spectrometer (together covering a 3mm×3mm measurement point).

[0117] When the measuring stage moves to each measuring point, the system synchronously triggers the spectrometer and XRR device: the XRR device collects X-ray reflection signals at a scanning speed of 0.01° / s within the incident angle range of 0.1°~5°, and calculates the density data of the epitaxial film by analyzing the position and intensity of the reflection peak (measurement accuracy ±0.01g / cm³); at the same time, the surface roughness data of the epitaxial film is retrieved by analyzing the full width at half maximum (FWHM) of the reflection peak (measurement range 0.1~10nm, accuracy ±0.05nm).

[0118] The density and roughness data collected by the XRR device are stored in XML format, containing information such as "measurement point number, density value, roughness value, and measurement error range". The data is transmitted in real time to the edge enhancement algorithm module via Fibre Channel (FC) with a transmission delay of ≤50ms, ensuring the temporal consistency with the original data from the spectrometer (timestamp error ≤10ms).

[0119] After receiving the density and roughness data, the edge enhancement algorithm module starts the weighted fusion-based correction model. The specific process is as follows:

[0120] Data correlation analysis: Establishing a correlation model between spectral reflectance intensity and density and roughness—when the epitaxial film density increases, the characteristic peak position of the reflectance spectrum shifts towards shorter wavelengths (the shift is positively correlated with the density increment); when the roughness increases, the intensity of the characteristic peak of the reflectance spectrum decreases (the decrease is positively correlated with the roughness increment). The algorithm fits historical data using the least squares method to obtain a correlation coefficient matrix (e.g., density correlation coefficient k1 = 0.02 nm / (g / cm³), roughness correlation coefficient k2 = 0.05 (1 / nm)). Correction calculation: Substitute the density data and roughness data of the current measurement point into the correlation model to calculate the theoretical correction amount of the spectral reflection intensity: If the actual density is higher than the standard density (such as the standard density of silicon epitaxial film 2.33g / cm³), then calculate the wavelength correction amount that needs to be shifted to the short-wavelength direction according to k1; if the actual roughness is higher than the standard roughness (such as the preset standard value 0.5nm), then calculate the reflection intensity correction amount that needs to be increased according to k2.

[0121] Secondary correction of effective data: The calculated wavelength correction and intensity correction are superimposed on the effective measurement data output by the edge enhancement algorithm to complete the secondary correction. For example, if the measurement point density of a silicon epitaxial film is 2.35 g / cm³ (higher than the standard of 0.02 g / cm³), the wavelength needs to be shifted by 0.02 × 0.02 = 0.0004 nm; if the roughness is 0.6 nm (higher than the standard of 0.1 nm), the reflection intensity needs to be increased by 0.05 × 0.1 = 0.005.

[0122] After correction, the "final valid measurement data" is output, which includes complete spectral information (corrected wavelength-reflection intensity), density value, roughness value, and the reliability of the data fusion (≥98%) required for film thickness calculation. This data will serve as the core basis for subsequent data comparison and trend analysis, and will be simultaneously stored in the local database and the official data area of ​​the cloud-based collaborative analysis system, replacing the temporary backup data.

[0123] As a specific example, when the wafer to be measured is a GaN epitaxial wafer, in the film thickness data acquisition and processing step, the integrated intensity of the 0002 diffraction peak of the GaN epitaxial film and the integrated intensity of the 0006 diffraction peak of the sapphire substrate are additionally acquired using a diffraction intensity measurement device. The intensity ratio R between the two is calculated, and this intensity ratio is incorporated as a correction parameter into the edge enhancement algorithm module. This ratio R is positively correlated with the film crystal quality. When R is lower than a preset threshold (e.g., 0.8), the system determines that the optical characteristics of this region deviate from the standard model, and the edge diffraction interference may be more complex. At this time, the edge enhancement algorithm will automatically increase the number of wavelet decomposition layers (e.g., from 3 layers to 4 layers) or decrease the threshold (e.g., from 0.1 to 0.08) to enhance the suppression capability of non-ideal interference.

[0124] As an example, the intensity ratio R can also be used as a weight or adjustment factor in the edge enhancement algorithm. For example:

[0125] If the R value is high (indicating good crystal quality), the algorithm judges that the optical model of the region is relatively standard and the characteristics of edge diffraction interference are relatively fixed. Therefore, a set of standard interference suppression coefficients is adopted.

[0126] If the R value is low (indicating poor crystal quality or thin film), the algorithm determines that the optical properties of the region may deviate from the standard model, and the edge interference may be more severe or have different morphology. Therefore, the algorithm parameters will be dynamically adjusted, such as increasing the intensity of smoothing or changing the threshold of edge detection, to more effectively filter out non-ideal interference.

[0127] After processing by the edge enhancement algorithm, the edge diffraction interference in the original spectrum is significantly weakened or eliminated, resulting in a relatively smooth and realistic corrected spectral curve.

[0128] The system fits this corrected spectral curve to a preset optical model (such as the Forouhi-Bloomer model or the Cauchy model). The fitting process retrieves the epitaxial film thickness and optical constants at the measurement point. This final film thickness value is the valid measurement data for that point.

[0129] The next step is data comparison and trend report generation: the effective measurement data is transmitted to the cloud collaborative analysis system, the cloud collaborative analysis system calls the preset historical database, compares and analyzes the effective measurement data with the epitaxial film thickness data of the same type of wafer in the historical database, and generates a wafer epitaxial film thickness trend prediction report based on the comparison results.

[0130] In one specific embodiment, the "final valid measurement data" generated in the film thickness data acquisition and processing step is uploaded to the cloud-based collaborative analysis system via an encrypted transmission protocol. During transmission, the system employs a "block transmission + verification retransmission" mechanism: the data is split into 100KB blocks, each block is accompanied by an MD5 checksum, and the cloud verifies the data integrity using the checksum. If data corruption is detected, local retransmission is triggered to ensure a data transmission success rate of ≥99.99%.

[0131] The cloud-based collaborative analysis system's data receiving module listens for transmission requests in real time. Upon receiving the data, it automatically parses the data format and extracts key information, including basic wafer information (model, batch, production process), film thickness-related data (corrected spectral information, density value, roughness value), and measurement process information (number of measurement points, data reliability). This information is then categorized and stored in the system's "real-time data cache," while generating a unique data identifier (containing wafer ID + measurement timestamp) for easy traceability.

[0132] The cloud-based collaborative analysis system calls the historical database retrieval module to filter historical epitaxial film thickness data for similar wafers based on the current wafer model, batch, and production process (by default, it retrieves 500 sets of historical data within the last 3 months under the same process parameters; the retrieval range can be adjusted via system parameters). The historical database adopts a "hierarchical storage architecture": frequently accessed recent data is stored in an in-memory database (response time ≤ 10ms), while infrequently accessed long-term data is stored in a distributed file system, ensuring efficient data retrieval.

[0133] The comparative analysis is divided into three dimensions:

[0134] Single-point data comparison: The calculated film thickness value of each measurement point on the current wafer (derived from the corrected spectral information) is compared with the mean and standard deviation of the film thickness at the corresponding measurement points in the historical data. The deviation rate between the current data and the historical mean is calculated. If the deviation rate is ≤ ±3%, it is judged as "normal fluctuation"; if the deviation rate is between ±3% and ±5%, it is marked as "slight abnormality"; if the deviation rate is > ±5%, it is marked as "significant abnormality".

[0135] Overall distribution comparison: Using statistical methods (such as normality test and analysis of variance), compare the distribution characteristics of the current wafer thickness data at all measurement points with the distribution characteristics of historical data. For example, calculate the ratio of the variance of the current data to the variance of the historical data. If the ratio is in the range of 0.8 to 1.2, it indicates that the film thickness uniformity is comparable to the historical level; if the ratio is >1.2, it indicates that the film thickness uniformity of the current wafer has decreased.

[0136] Key parameter correlation comparison: The density and roughness values ​​in the current data are compared with the corresponding parameters in the historical data to analyze the correlation between film thickness deviation and these parameters. For example, if the current average film thickness is higher than the historical average, and the density value is also higher than the historical average, the system will initially determine that the film thickness deviation may be related to the improvement of film density, providing direction for subsequent process analysis.

[0137] Based on the comparative analysis results, the report generation module of the cloud-based collaborative analysis system automatically generates the "Wafer Epitaxial Film Thickness Trend Prediction Report," which includes three core parts:

[0138] Current measurement results summary: The film thickness, density, and roughness values ​​at each measurement point on the wafer are presented in tabular form, with the location of abnormal measurement points and the reasons for the deviations marked (e.g., "Measurement point 12: film thickness deviation rate 4.2%, slight abnormality, density value higher than the historical average by 0.02 g / cm³"). The distribution trend of film thickness on the wafer surface is shown through a line graph.

[0139] Historical comparison analysis conclusions: Use bar charts to compare the current data with historical data for mean film thickness, variance, mean density, and mean roughness to clarify the quality of the current measurement results relative to historical levels; use pie charts to count the percentage of abnormal measurement points. If the percentage is <5%, it is judged as "overall qualified"; if the percentage is between 5% and 10%, it is judged as "needs attention"; if the percentage is >10%, it is judged as "needs rectification".

[0140] Trend prediction and suggestions: Based on the film thickness data of similar wafers over the past month (fitted by a linear regression model), the system predicts the range of change of the average film thickness in the next week. If the trend shows that the average film thickness is gradually deviating from the standard value, the system will combine historical process adjustment records to give targeted suggestions (such as "It is recommended to reduce the epitaxial growth temperature by 2℃. Referring to the process adjustment effect on X month X day, 2024, it is expected that the average film thickness can return to the standard range").

[0141] Once generated, the report is automatically converted to PDF format and pushed to the designated email addresses of production managers and process engineers via system email. At the same time, it is archived in the "Report Management Module" of the cloud system, supporting online viewing, downloading, and comparison with historical reports.

[0142] Preferably, in the film thickness data acquisition and processing step, when the measuring stage moves along the optimized measuring path, the high-resolution X-ray diffraction device is simultaneously activated to acquire the composition data, defect data, and strain data of the epitaxial film. The composition data, defect data, and strain data are then transmitted to the cloud-based collaborative analysis system. The cloud-based collaborative analysis system combines this data with the effective measurement data to supplement and correct the film thickness trend prediction report.

[0143] Specifically, as the measuring stage moves along the optimized path, the system synchronously activates the high-resolution X-ray diffraction device, which keeps pace with the movement of the measuring stage: when the measuring stage pauses at each measuring point to collect spectral data, the diffraction device immediately scans that measuring point to collect the composition data of the epitaxial film (such as the atomic percentage of each element in the multi-component compound epitaxial film), defect data (such as dislocation density and vacancy concentration), and strain data (such as in-plane strain and volumetric strain).

[0144] After collection, these data are transmitted in real time to the cloud-based collaborative analysis system via an industrial IoT gateway. During the transmission process, they are associated with valid measurement data: the composition, defect, and strain data are bound to the film thickness, density, and roughness data of the corresponding measurement points through "measurement point number + timestamp" to form a "single measurement point multi-dimensional dataset", which is stored in the "comprehensive data area" in the cloud to provide a data basis for subsequent report corrections.

[0145] The report correction module of the cloud-based collaborative analysis system calls upon the "single measurement point multi-dimensional dataset" to supplement and correct the initially generated trend prediction report from three levels:

[0146] Precise identification of the cause of the anomaly: If the preliminary report marks a measurement point as "significantly abnormal film thickness (deviation rate 6%)", supplement the analysis of the component data and defect data of that point: If it is found that the atomic percentage of a certain element at that point deviates from the standard value by 5% and the dislocation density is higher than the historical average by 20%, then add "Speculated cause of the anomaly: segregation of film components leads to abnormal growth rate, accompanied by an increase in defects, affecting film thickness uniformity" to the report, replacing the original statement "the cause of the anomaly is yet to be investigated".

[0147] Trend prediction model optimization: Strain data is introduced into the trend prediction model. If historical data shows that "when the in-plane strain of the epitaxial film increases, the average film thickness of subsequent batches of wafers will decrease", and the in-plane strain in the current data is 3% higher than the average of the previous month, then the original prediction conclusion is revised: "The average film thickness is expected to fluctuate by ±1% in the next week" is adjusted to "The average film thickness is expected to decrease by 1.5%~2% in the next week, and the epitaxial growth pressure parameters need to be adjusted in advance", and historical data charts on the relationship between strain and film thickness are added as support.

[0148] Process recommendations are refined: Process recommendations are optimized based on component data. If the initial report recommends "adjusting the growth temperature", but after supplementing the component data it is found that "the current proportion of a certain element in the film layer is too low, while the solubility of this element increases with increasing temperature", then the recommendation is refined to "it is recommended to increase the growth temperature by 3°C and extend the growth time by 10 seconds, which will both increase the proportion of the target element to the standard range and avoid excessive increase in film thickness". The application cases and effects of this adjustment scheme in historical data are marked (e.g., "This scheme was adopted on [Date] 2024, and the component qualification rate increased to 98%, and the film thickness deviation rate decreased to within 2%").

[0149] The revised report includes a new "Multi-dimensional Feature Analysis" section. The system uses different colors to highlight the revised content (original content in black, revised and supplemented content in blue) to help users clearly identify the corrections. At the same time, a "Revision Explanation Document" is generated, which records in detail the data sources and analysis logic on which the corrections were based, ensuring the report's traceability and credibility.

[0150] Finally, the abnormal alarm triggering step is implemented: the cloud-based collaborative analysis system judges in real time whether there are any abnormalities in the valid measurement data during the comparison and analysis process. If the data is determined to be abnormal, the local preset audible and visual alarm device is triggered to provide an alarm prompt.

[0151] Taking the aforementioned GaN epitaxial wafer as an example, the cloud-based collaborative analysis system uses historical GaN epitaxial wafer thickness data from the database as a benchmark, sets a ±3% deviation threshold (adjustable as needed), and compares the current valid film thickness data in real time. If a single measurement point exceeds the threshold, or if three consecutive adjacent measurement points show a deviation exceeding 1.5%, it is determined to be a data anomaly. The system immediately sends an alarm command to the local control terminal, triggering a preset audible and visual alarm device. After the device is activated, a red warning light flashes at a frequency of 1Hz, while a buzzer emits an intermittent alarm sound (decibels ≥80dB), and an abnormal data pop-up window appears on the terminal screen, displaying the coordinates of the abnormal measurement point and the deviation value, facilitating quick location and handling by staff.

[0152] The above is based on Figures 1-2 The illustrated method embodiments provide a detailed description of implementation methods. See below for further details. Figures 3-4 , corresponding to the system (product, device) implementation examples.

[0153] Figure 3 This is a schematic diagram of the hardware unit composition of a wafer epitaxial film thickness measurement system based on trajectory adaptation according to an embodiment of the present invention.

[0154] exist Figure 3 The measurement system includes a feeding and transfer unit, a positioning and compensation unit, a measurement path planning unit, a film thickness data acquisition and processing unit, a data comparison and trend report generation unit, and an abnormal alarm triggering unit. The positioning and compensation unit includes a calibration mechanism and a compensation mechanism.

[0155] Figure 4 yes Figure 3 A schematic diagram of some substructures of the trajectory-adaptive wafer epitaxial film thickness measurement system.

[0156] exist Figure 4 In the process, after the loading and transfer unit loads the wafer to be measured, it transfers the wafer to the calibration mechanism through a preset turnover mechanism.

[0157] The positioning and compensation unit performs preliminary positioning of the wafer transferred to the correction mechanism through the positioning component of the correction mechanism. The compensation mechanism drives the correction stage to move based on the three-dimensional coordinate compensation value to complete the correction and compensation of the wafer.

[0158] The measurement path planning unit transfers the wafer that has completed the correction and compensation to the measurement mechanism. The trajectory adaptive algorithm built into the measurement mechanism automatically plans an optimized measurement path based on the actual size of the current wafer and controls the measurement stage to move along the optimized measurement path.

[0159] During the movement of the measurement stage, the film thickness data acquisition and processing unit starts the spectrometer to acquire the original measurement data of the wafer epitaxial film and transmits the original measurement data to the edge enhancement algorithm module in real time. The edge enhancement algorithm module eliminates the edge diffraction interference in the original measurement data to obtain effective measurement data.

[0160] The data comparison and trend report generation unit transmits the effective measurement data to the cloud collaborative analysis system. The cloud collaborative analysis system calls the preset historical database to compare and analyze the effective measurement data with the epitaxial film thickness data of the same type of wafer in the historical database, and generates a wafer epitaxial film thickness trend prediction report based on the comparison results.

[0161] The cloud-based collaborative analysis system determines in real time whether there are any anomalies in the valid measurement data during the comparison and analysis process. If the data is determined to be abnormal, the anomaly alarm triggering unit triggers the locally preset audible and visual alarm device to issue an alarm prompt.

[0162] The positioning and compensation unit also includes an AI vision processing module and an X-ray reflectivity measurement (XRR) device;

[0163] When the positioning component of the calibration mechanism performs initial positioning, it simultaneously activates the sensor group to collect the reference position data of the wafer and transmits the reference position data to the AI ​​vision processing module.

[0164] The AI ​​vision processing module calls a preset deep learning model to identify the notch or positioning edge of the wafer. The compensation mechanism dynamically generates a three-dimensional coordinate compensation value based on the identification result, and then drives the calibration table of the correction mechanism to perform the correction compensation of the wafer based on the three-dimensional coordinate compensation value.

[0165] Furthermore, when the sensor group collects wafer reference position data, it simultaneously collects wafer warpage data and transmits it to the AI ​​vision processing module; the AI ​​vision processing module performs collaborative calculations on the warpage data and the three-dimensional coordinate compensation value.

[0166] As the measuring stage moves along the optimized measuring path, the high-resolution X-ray diffraction device is simultaneously activated to collect the composition data, defect data, and strain data of the epitaxial film. The composition data, defect data, and strain data are then transmitted to the cloud-based collaborative analysis system. The cloud-based collaborative analysis system combines this data with the effective measurement data to supplement and correct the film thickness trend prediction report.

[0167] Although not shown in the accompanying drawings, further embodiments propose a visual measurement device including a human-computer interactive visual interface, through which the visual measurement device performs... Figure 1 or Figure 2The method for measuring wafer epitaxial film thickness based on trajectory adaptation is described above, and a wafer epitaxial film thickness trend prediction report is displayed on the human-computer interactive visualization interface.

[0168] Although not shown in the accompanying drawings, a preferred and further embodiment of the product may also be an electronic device comprising: a memory and one or more processors. The memory stores one or more application programs adapted to be executed by the one or more processors, as described above, a trajectory-adaptive wafer epitaxial film thickness measurement method.

[0169] Although not shown in the accompanying drawings, further embodiments also include a computer-readable storage medium storing a computer program that, when executed, implements the steps of the aforementioned trajectory-adaptive wafer epitaxial film thickness measurement method.

[0170] It is understood that the system, product, equipment, and media implementation examples and method implementations correspond to each other and can be referenced by each other, and their principles are similar or the same, so they will not be elaborated again.

[0171] This invention improves the accuracy and efficiency of wafer epitaxial film thickness measurement through trajectory adaptive planning and edge enhancement algorithms; combined with cloud-based collaborative analysis, it enables real-time anomaly alarms and trend prediction, which helps improve the yield and intelligence level of semiconductor manufacturing. Its specific advantages are reflected in:

[0172] I. Positioning accuracy is significantly improved, and measurement errors are greatly reduced.

[0173] Existing technologies, lacking dedicated calibration and compensation mechanisms, are prone to positional shifts after wafer transfer, resulting in measurement errors exceeding ±10%. This invention addresses this issue through an innovative wafer positioning and calibration compensation process: after the wafer is transferred to the calibration mechanism, its positioning component performs initial positioning. This component, employing a combination of a high-precision vision sensor and a laser displacement sensor, accurately captures the wafer's edge contour and surface height information, establishing a realistic three-dimensional coordinate model of the wafer. Subsequently, the system calculates the three-dimensional coordinate compensation value based on the model, driving the calibration stage to perform translational adjustments in the X and Y axes and height compensation in the Z axis, ensuring complete alignment between the wafer detection area and the measurement probe. Actual testing shows that this calibration compensation mechanism can control wafer positional shifts within ±0.01mm, reducing the corresponding measurement error to below ±1%, far exceeding existing technologies. This fully meets the stringent requirements of advanced processes for high-precision measurement, laying a solid foundation for the accuracy of subsequent film thickness data acquisition.

[0174] II. Measurement path adaptive adaptation improves measurement efficiency and mass production compatibility.

[0175] Traditional equipment uses preset fixed paths, requiring manual parameter adjustments for different wafer sizes. Each adjustment takes over 30 minutes and involves a significant amount of ineffective movement, making it difficult to meet the mass production line's testing requirements of "more than 30 wafers per hour." This invention achieves a breakthrough in the measurement path planning step by leveraging a trajectory adaptive algorithm built into the measurement mechanism. This algorithm first obtains the actual diameter, thickness, and other dimensional parameters of the current wafer through the measurement mechanism's size detection unit (such as an infrared ranging sensor). Then, combined with the preset detection point density requirements on the wafer surface (e.g., 2 detection points per square centimeter), it automatically plans an optimized measurement path. For example, for an 8-inch wafer, the algorithm plans an efficient detection trajectory with the wafer center as the origin and a spiral distribution. When switching to a 12-inch wafer, no manual intervention is required; the algorithm can recalculate and generate an adapted, enlarged spiral trajectory within 10 seconds, automatically avoiding ineffective detection areas at the wafer edge and reducing the ineffective movement distance of the measurement stage. Practical application verification shows that after adopting this adaptive path planning, the measurement time of a single wafer is shortened to less than 2 minutes, the measurement efficiency is improved by more than 40%, which fully meets the testing rhythm of the mass production line. At the same time, no manual debugging is required, which reduces the workload of operators and improves the equipment's compatibility with wafers of different sizes.

[0176] III. Collaborative and intelligent data processing for accurate identification of film thickness trends

[0177] Existing technologies can only perform basic filtering on single measurement data, lacking collaborative comparison with historical data, and anomaly detection relies on manual review, which takes up to 2 hours and easily produces a large number of defective wafers. This invention constructs a cloud-based collaborative analysis system in the data processing stage, achieving comprehensive improvements: In the film thickness data acquisition and processing step, after the system starts the spectrometer to collect raw data, it does not simply filter it, but instead uses an edge enhancement algorithm module to specifically eliminate edge diffraction interference—this module uses wavelet transform algorithm to perform multi-scale decomposition of the raw spectral data, accurately separating and removing diffraction interference signals to obtain high-purity, effective measurement data. Subsequently, the valid data is transmitted to a cloud-based collaborative analysis system. The system accesses a pre-set historical database (containing nearly 12 months of film thickness data for similar wafers) and uses big data comparison algorithms (such as a neural network-based trend analysis model) to perform a deep comparison between the current data and historical data. This not only quickly identifies minute trends in film thickness (such as a monthly increase or decrease of 0.5%), but also automatically generates a detailed wafer epitaxial film thickness trend prediction report based on the comparison results. The report includes key parameters such as the average film thickness, standard deviation, and slope of change, providing data support for production process adjustments. This collaborative and intelligent data processing model reduces the time for film thickness trend identification from several hours in existing technologies to real-time synchronization, significantly improving the controllability of the production process.

[0178] IV. Real-time anomaly alarms reduce the generation of defective wafers.

[0179] Existing technologies suffer from delayed anomaly detection, averaging 2 hours from data acquisition to anomaly detection, during which a large number of defective wafers are generated, severely impacting production yield. This invention constructs a highly efficient real-time alarm mechanism in the anomaly alarm triggering step: When performing data comparison and analysis, the cloud-based collaborative analysis system does not wait for all data acquisition to complete before making a judgment, but instead monitors the effective measurement data of each measuring point in real time. The system presets dual anomaly judgment criteria: one is the deviation threshold between the data of a single measuring point and the historical average data at the same location (e.g., ±3%), and the other is the fluctuation threshold of three consecutive adjacent measuring point data (e.g., deviation exceeding 1.5%). If either criterion is met, the system immediately determines the data to be abnormal. Subsequently, the cloud-based collaborative analysis system sends an alarm command to the local control terminal within 1 second, triggering a preset audible and visual alarm device—a red warning light flashes at a frequency of 1Hz, a buzzer emits an intermittent alarm sound ≥80dB, and simultaneously, an anomaly data pop-up window appears on the terminal screen, clearly displaying the coordinates of the abnormal measuring point (e.g., X: 50mm, Y: 30mm) and the deviation value (e.g., +4.2%). Staff can locate the abnormal position and suspend production within 30 seconds, preventing the continuous generation of defective wafers. Statistics show that after adopting this real-time alarm mechanism, the number of defective wafers has decreased by more than 70%, significantly improving production yield and reducing production costs for enterprises.

[0180] In this section, the present invention provides several method or product embodiments. Each method or product embodiment constitutes an independent technical solution and makes at least one contribution to the prior art, capable of solving one or more technical problems mentioned in the background art, and having one or more of the aforementioned outstanding effects or advantages.

[0181] However, it is understood that not every embodiment is required to solve all the technical problems mentioned in the background or achieve all the technical effects. For each individual embodiment, as long as it makes at least one contribution to the prior art and achieves an improved technical effect, and has outstanding substantive features or significant progress, the technical solution constituted by that individual embodiment should possess novelty and inventiveness in the sense of patent law.

[0182] Other technologies, principles, algorithms, or models not elaborated in detail in this application can be found in the prior art.

[0183] The foregoing has shown and described the method embodiments and systems of the present invention, but it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for measuring wafer epitaxial film thickness based on trajectory adaptation, characterized in that, Includes the following steps: Wafer loading and initial transfer: After loading the wafer to be measured, the wafer is transferred to the calibration mechanism through a preset turnover mechanism; Wafer positioning and calibration compensation: The positioning component of the calibration mechanism performs initial positioning of the wafer transferred to the calibration mechanism, and drives the calibration stage to move based on the three-dimensional coordinate compensation value to complete the wafer calibration compensation; Measurement path planning: The wafer that has completed the calibration and compensation is transferred to the measurement mechanism. The trajectory adaptive algorithm built into the measurement mechanism automatically plans an optimized measurement path based on the actual size of the current wafer, and controls the measurement stage to move along the optimized measurement path. Film thickness data acquisition and processing: During the movement of the measurement stage, the spectrometer is activated to acquire the raw measurement data of the wafer epitaxial film, and the raw measurement data is transmitted to the edge enhancement algorithm module in real time. The edge enhancement algorithm module eliminates the edge diffraction interference in the raw measurement data to obtain effective measurement data. Data comparison and trend report generation: The effective measurement data is transmitted to the cloud-based collaborative analysis system. The cloud-based collaborative analysis system calls a preset historical database to compare and analyze the effective measurement data with the epitaxial film thickness data of the same type of wafer in the historical database. Based on the comparison results, a wafer epitaxial film thickness trend prediction report is generated. Anomaly alarm triggering: The cloud-based collaborative analysis system determines in real time whether there are any anomalies in the valid measurement data during the comparison and analysis process. If the data is determined to be abnormal, the system will trigger the local preset audible and visual alarm device to issue an alarm prompt.

2. The wafer epitaxial film thickness measurement method based on trajectory adaptation as described in claim 1, characterized in that, The wafer positioning and correction compensation steps include: During the initial positioning, the sensor group is activated to collect the reference position data of the wafer and transmit the reference position data to the AI ​​vision processing module. The AI ​​vision processing module calls a preset deep learning model to identify the notches or positioning edges of the wafer, dynamically generates three-dimensional coordinate compensation values ​​based on the identification results, and then drives the calibration stage to perform the calibration compensation of the wafer based on the three-dimensional coordinate compensation values.

3. The wafer epitaxial film thickness measurement method based on trajectory adaptation as described in claim 1, characterized in that, In the film thickness data acquisition and processing step, while the spectrometer acquires the original measurement data of the wafer epitaxial film, the X-ray reflectivity measurement (XRR) device is simultaneously activated to acquire the density data and roughness data of the epitaxial film. The density data, roughness data and the original measurement data are then transmitted together to the edge enhancement algorithm module, and the effective measurement data is corrected through multi-dimensional data fusion.

4. The wafer epitaxial film thickness measurement method based on trajectory adaptation as described in claim 1, characterized in that, In the wafer positioning and correction compensation step, when the sensor group collects wafer reference position data, it simultaneously collects wafer warpage data and transmits it to the AI ​​vision processing module; the AI ​​vision processing module performs collaborative calculations on the warpage data and the three-dimensional coordinate compensation value.

5. The wafer epitaxial film thickness measurement method based on trajectory adaptation as described in claim 1, characterized in that, When the wafer to be measured is a GaN epitaxial film wafer, in the film thickness data acquisition and processing step, the integral intensity of the 0002 diffraction peak of the GaN epitaxial film and the integral intensity of the 0006 diffraction peak of the sapphire substrate are additionally acquired by a diffraction intensity measurement device, the intensity ratio of the two is calculated, and the intensity ratio is incorporated into the edge enhancement algorithm module as a correction parameter.

6. The wafer epitaxial film thickness measurement method based on trajectory adaptation as described in claim 1, characterized in that, In the film thickness data acquisition and processing step, when the measuring stage moves along the optimized measurement path, the high-resolution X-ray diffraction device is simultaneously activated to acquire the composition data, defect data, and strain data of the epitaxial film. The composition data, defect data, and strain data are transmitted to the cloud-based collaborative analysis system. The cloud-based collaborative analysis system combines the composition data, defect data, and strain data with the effective measurement data to supplement and correct the film thickness trend prediction report.

7. A wafer epitaxial film thickness measurement system based on trajectory adaptation, the measurement system comprising a loading and transfer unit, a positioning and compensation unit, a measurement path planning unit, a film thickness data acquisition and processing unit, a data comparison and trend report generation unit, and an abnormal alarm triggering unit, wherein the positioning and compensation unit comprises a calibration mechanism and a compensation mechanism; Its features are: After the loading and transfer unit loads the wafer to be measured, it transfers the wafer to the calibration mechanism through a preset turnover mechanism. The positioning and compensation unit performs preliminary positioning of the wafer transferred to the correction mechanism through the positioning component of the correction mechanism. The compensation mechanism drives the correction stage to move based on the three-dimensional coordinate compensation value to complete the correction and compensation of the wafer. The measurement path planning unit transfers the wafer that has completed the correction and compensation to the measurement mechanism. The trajectory adaptive algorithm built into the measurement mechanism automatically plans an optimized measurement path based on the actual size of the current wafer and controls the measurement stage to move along the optimized measurement path. During the movement of the measurement stage, the film thickness data acquisition and processing unit starts the spectrometer to acquire the original measurement data of the wafer epitaxial film and transmits the original measurement data to the edge enhancement algorithm module in real time. The edge enhancement algorithm module eliminates the edge diffraction interference in the original measurement data to obtain effective measurement data. The data comparison and trend report generation unit transmits the effective measurement data to the cloud collaborative analysis system. The cloud collaborative analysis system calls the preset historical database to compare and analyze the effective measurement data with the epitaxial film thickness data of the same type of wafer in the historical database, and generates a wafer epitaxial film thickness trend prediction report based on the comparison results. The cloud-based collaborative analysis system determines in real time whether there are any anomalies in the valid measurement data during the comparison and analysis process. If the data is determined to be abnormal, the anomaly alarm triggering unit triggers the locally preset audible and visual alarm device to issue an alarm prompt.

8. The wafer epitaxial film thickness measurement system based on trajectory adaptation as described in claim 7, characterized in that, The positioning and compensation unit also includes an AI vision processing module and an X-ray reflectivity measurement (XRR) device; When the positioning component of the calibration mechanism performs initial positioning, it simultaneously activates the sensor group to collect the reference position data of the wafer and transmits the reference position data to the AI ​​vision processing module. The AI ​​vision processing module calls a preset deep learning model to identify the notch or positioning edge of the wafer. The compensation mechanism dynamically generates a three-dimensional coordinate compensation value based on the identification result, and then drives the calibration table of the correction mechanism to perform the correction compensation of the wafer based on the three-dimensional coordinate compensation value. Furthermore, when the sensor group collects wafer reference position data, it simultaneously collects wafer warpage data and transmits it to the AI ​​vision processing module; the AI ​​vision processing module performs collaborative calculations on the warpage data and the three-dimensional coordinate compensation value.

9. The wafer epitaxial film thickness measurement system based on trajectory adaptation as described in claim 7, characterized in that, As the measuring stage moves along the optimized measuring path, the high-resolution X-ray diffraction device is simultaneously activated to collect the composition data, defect data, and strain data of the epitaxial film. The composition data, defect data, and strain data are then transmitted to the cloud-based collaborative analysis system. The cloud-based collaborative analysis system combines the composition data, defect data, and strain data with the effective measurement data to supplement and correct the film thickness trend prediction report.

10. A visualization measurement device, the visualization measurement device comprising a human-computer interaction visualization interface, wherein the visualization measurement device performs a wafer epitaxial film thickness measurement method based on trajectory adaptation as described in any one of claims 1-6, and displays a wafer epitaxial film thickness trend prediction report on the human-computer interaction visualization interface.

Images