Wafer loading deviation correction method and device

Abstract

The invention discloses a wafer loading deviation rectifying method and device, belongs to the field of wafer transmission, and is applied to simf, efem and sort wafer transmission equipment.The method comprises the steps that firstly, gradienter posture data and lug signals of a wafer loading groove are collected in real time through a multi-source sensing assembly, and whether wafer installation is abnormal or not is judged through a wafer channel three-dimensional posture sensing algorithm; if yes, the gradient of the equipment is calculated through a multi-source sensing dynamic deviation fitting algorithm, and then the gradient is converted into the number of motor pulses by loading a dynamic twin calibration model; a motor is driven to adjust the position of the wafer according to the pulse number, updating data are collected in real time by means of an SMIF loading precise control data analysis platform during adjustment, and the steps of abnormity judgment, deviation calculation and adjustment are repeated until wafer installation reaches the standard. According to the method, multi-source data are fused, dynamic deviation correction is achieved through real-time feedback and repeated verification, follow-up misinformation and wafer damage are effectively avoided, and stable operation of equipment and wafer machining quality are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of wafer manufacturing technology, and in particular to a wafer loading and correction method and apparatus. Background Technology

[0002] In wafer manufacturing, wafer transport equipment such as SIMF, EFEM, and Sort are key devices for achieving automated wafer transport and processing. The precise loading of wafers during transport directly impacts subsequent processing quality and wafer yield. Currently, these devices require wafers to be stably placed in wafer loading slots, but their loading status is susceptible to factors such as equipment vibration, ambient temperature changes, and mechanical wear, leading to abnormal wafer mounting. Failure to detect and correct these abnormalities in a timely manner can not only cause false alarms during subsequent equipment operation but also result in damage such as wafer collisions and scratches due to improper loading, causing economic losses. Therefore, the industry urgently needs a method that can perceive the wafer loading status in real time, accurately calculate equipment deviations, and efficiently complete correction to ensure the stable operation of wafer transport equipment and wafer processing quality. A wafer loading correction method and device are proposed based on this need, achieving dynamic correction of wafer loading through multi-algorithm collaboration and multi-unit cooperation.

[0003] Existing technologies have significant shortcomings in wafer loading correction. On the one hand, existing methods rely heavily on single sensor data for anomaly detection and deviation calculation, failing to effectively integrate and analyze attitude data collected by the level and signal data collected by the tab recognition module. This results in low accuracy in detecting wafer mounting anomalies, and the calculated device tilt angle has a large error compared to the actual deviation, failing to provide accurate data support for subsequent correction. On the other hand, existing correction processes lack dynamic calibration and real-time feedback mechanisms. After converting the motor pulse count based on the deviation, only a single fixed adjustment can be performed. It is impossible to collect and update data in real time during the adjustment process and repeatedly verify the adjustment effect, which easily leads to inadequate or excessive adjustment. It is difficult to ensure that the final wafer loading state meets the preset standard, and it cannot effectively avoid subsequent false alarms and wafer damage. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a wafer loading and correction method and apparatus.

[0005] This invention provides a wafer loading correction method, comprising the following steps: S1, using the SMIF loading precision control data analysis platform to call multi-source sensing components to collect in real time the three-dimensional attitude data output by the level instrument configured on the wafer loading slot in the wafer transfer equipment and the corresponding convex signal generated by the wafer convex identification module. The three-dimensional attitude data includes tilt angle data in the X, Y, and Z axes of the loading slot, and the convex signal includes convex position coordinates and convex contour feature data; S2, inputting the convex signal collected in S1 into a three-dimensional pose perception algorithm for the wafer path. The algorithm compares the convex position coordinates with preset standard convex coordinates and, combined with the edge fitting results of the convex contour feature data, determines whether there is an installation abnormality on the wafer in the wafer loading slot; S3, when S2 determines that an installation abnormality exists, inputting the three-dimensional attitude data of the level instrument collected in S1 into a multi-source sensing dynamic deviation fitting algorithm. The algorithm performs time-series correlation analysis on the tilt angle data at different acquisition times, removes abnormal fluctuation data, and calculates the equipment tilt angle of the wafer loading slot in the current working state. The tilt angle includes the overall tilt angle and tilt direction vector of the loading slot; S4, based on the equipment tilt angle obtained in S3, the loading dynamic twin calibration model is called to calculate the deviation between the equipment tilt angle parameter and the preset standard attitude parameter of the loading slot in the model. Through the attitude-pulse mapping logic built into the model, the tilt angle deviation value is converted into the number of motor pulses used to drive the adjustment motor of the loading slot. The number of motor pulses includes the number of adjustment pulses in the X-axis direction, the number of adjustment pulses in the Y-axis direction, and the number of adjustment pulses in the Z-axis direction; S5, the number of motor pulses in each direction converted in S4 is transmitted to the motor control unit of the SMIF machine. The motor control unit drives the corresponding adjustment motor to operate according to the number of pulses in each direction, thereby driving the wafer loading slot to perform multi-dimensional attitude adjustment to change the position of the abnormally installed wafer in the slot; S6, during the adjustment process in S5, the SMIF loading precision control data analysis platform collects the updated three-dimensional attitude data of the level and the updated convex signal of the convex recognition module in real time, and repeats the operations of S2-S5 until the three-dimensional pose perception algorithm of the crystal channel determines that the wafer installation status meets the preset standard. Furthermore, the expression for the crystal channel three-dimensional pose sensing algorithm is: ,in, This represents the wafer mounting tolerance value. Let be the weighting coefficient of the i-th group of convex signals. Let be the coordinates of the position of the i-th group of convex pieces. To preset standard convex coordinates, This represents the actual fitted length of the convex profile feature data. The preset standard convex profile length is given by n, where n is the number of convex signal groups collected.

[0006] Furthermore, the expression for the multi-source sensing dynamic deviation fitting algorithm is: ,in, The value is the equipment tilt angle, and m is the number of data sets collected from the level gauge. For the j-th group of level data Tilt angles of the X, Y, and Z axes. These are the tilt weighting coefficients in the X, Y, and Z axis directions. This is the time-series decay coefficient. Let j be the time of data collection for the j-th group. This is the average of all data collection times.

[0007] Furthermore, the expression for the loaded dynamic twin calibration model is: ,in, Let k be the number of pulses for the k-th adjustment motor, where k corresponds to the motor in the X-axis, Y-axis, and Z-axis directions, respectively. Let be the pulse coefficient of the k-th motor. Let the standard tilt angle be in the k-th direction. Let be the standard deviation of the tilt data in the k-th direction. To preset the standard deviation, is the dynamic calibration coefficient in the k-th direction.

[0008] Furthermore, the SMIF loading precision control data analysis platform employs a data fusion formula during the data acquisition process: ,in, The merged data values For data fusion weights, Collect data for the level. For acquiring data of the convex plate signal, This represents the error value of the level data. This represents the error value of the convex plate signal data.

[0009] Furthermore, in S4, the motor pulse count conversion process also employs a compensation formula: ,in, The number of motor pulses after compensation. Let be the tilt compensation coefficient for the k-th motor. The historical data compensation coefficient for the k-th motor is... Let be the tilt angle in the k-th direction of the j-th group of level data.

[0010] Further, step S3 includes the following sub-steps: S31 Extracting the three-dimensional attitude data of the level instrument collected in S1 from the database of the SMIF-loaded precision control data analysis platform, and sorting the data according to the acquisition time sequence to obtain a time-series data sequence; S32 Inputting the time-series data sequence into the preprocessing module of the multi-source sensor dynamic deviation fitting algorithm, and the module performs outlier detection on the data, removing outliers that exceed the preset tilt angle range. Data is marked as outlier and removed, while valid data is retained. S33 assigns corresponding weight coefficients to the X, Y, and Z axes based on the importance of the tilt angle in each direction for the retained valid data. S34 calculates the weighted skew value corresponding to each valid data point; S34 introduces a time-series decay coefficient based on the difference between the data acquisition time and the average acquisition time. The weighted tilt values ​​are corrected, and the final tilt angle is calculated by averaging the results. .

[0011] Further, S4 includes the following sub-steps: S41 retrieves preset standard attitude parameters of the loading slot from the parameter library of the loading dynamic twin calibration model, which include standard tilt angles in the X-axis, Y-axis, and Z-axis directions. and standard deviation S42 will obtain the equipment tilt angle from S3. The difference between the tilt angle and the standard tilt angle in each direction is calculated to obtain the tilt deviation value in each direction; S43 adjusts the motor model and performance parameters according to each direction and calls the corresponding pulse coefficient. and dynamic calibration coefficient Simultaneously calculate the standard deviation of the current tilt data. S44 substitutes the tilt deviation value, pulse coefficient, standard deviation, and dynamic calibration coefficient into the expression of the loading dynamic twin calibration model to calculate the initial pulse number of the adjustment motor in each direction. .

[0012] Further, S5 includes the following sub-steps: S51 The initial pulse count of each direction motor obtained in S4 is transmitted to the motor control unit of the SMIF machine. The control unit performs format conversion on the pulse count, converting the digital signal into a control signal recognizable by the motor; S52 The motor control unit sends the converted control signal to the corresponding motor... axis, axis, The axis adjustment motor sends a drive command, which includes parameters such as the number of pulses, operating speed, and operating direction. After receiving the drive command, the adjustment motor starts to operate according to the command parameters. Through the transmission mechanism, it drives the wafer loading slot to adjust its posture along the corresponding direction. During the adjustment process, the operating status signal is fed back in real time. The motor control unit receives the operating status signal fed back by the motor and confirms whether the motor has completed the operation according to the preset number of pulses. If it has not completed the operation, it continues to send drive commands until the number of motor operating pulses reaches the preset value.

[0013] A wafer loading correction device includes: a multi-source sensor data acquisition unit connected to a level and bump recognition module on the wafer loading slot, used to acquire real-time three-dimensional attitude data of the level and bump signals, and transmit the acquired data to a SMIF loading precision control data analysis unit; an SMIF loading precision control data analysis unit connected to both the multi-source sensor data acquisition unit and a crystal lane three-dimensional pose perception algorithm processing unit, used to store, sort, and preprocess the acquired data, and send the processed data to the corresponding algorithm processing units; and a crystal lane three-dimensional pose perception algorithm processing unit connected to both the SMIF loading precision control data analysis unit and a multi-source sensor dynamic deviation fitting algorithm processing unit, used to receive the processed bump signals and determine the wafer mounting status, and if an abnormality is detected... The system triggers the operation of the multi-source sensor dynamic deviation fitting algorithm processing unit. This unit, connected to the crystal lane 3D pose perception algorithm processing unit and the loading dynamic twin calibration model calculation unit, receives level data and calculates the device tilt, transmitting the tilt data to the loading dynamic twin calibration model calculation unit. The loading dynamic twin calibration model calculation unit, connected to the multi-source sensor dynamic deviation fitting algorithm processing unit and the motor control drive unit, calculates the number of motor pulses based on the device tilt and transmits the pulse count to the motor control drive unit. The motor control drive unit, connected to the loading dynamic twin calibration model calculation unit and the wafer loading slot adjustment motor, receives the motor pulse count and drives the adjustment motor to adjust the wafer loading slot's attitude.

[0014] Beneficial Effects: This application uses multi-source sensing components to collect level gauge attitude data and cam signal in real time. A three-dimensional positional perception algorithm accurately determines wafer mounting anomalies. A multi-source sensing dynamic deviation fitting algorithm calculates the equipment tilt, and a dynamic twin calibration model is used to convert the motor pulse count, driving the motor to adjust the wafer position. Simultaneously, the SMIF loading precision control data analysis platform provides real-time feedback and repeated verification, achieving dynamic correction. This method integrates two types of data: level gauge and cam signal. Algorithm processing improves the accuracy of anomaly detection and tilt calculation, providing precise data support for correction. Addressing the shortcomings of existing technologies, such as lack of dynamic calibration and real-time feedback, and the tendency for single adjustments to be incomplete or excessive, this method collects and updates data in real time during the adjustment process, repeatedly executing anomaly detection, deviation calculation, and motor adjustment steps until the wafer mounting meets standards. This avoids false alarms and wafer damage, ensuring stable equipment operation and processing quality. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method steps of the present invention; Figure 2 This is a diagram showing the unit composition of the device of the present invention. Detailed Implementation

[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1 As shown, a wafer loading and correction method includes the following steps: S1. The SMIF loading precision control data analysis platform calls the multi-source sensing components to collect in real time the three-dimensional attitude data output by the level instrument configured on the wafer loading slot in the wafer transfer equipment and the corresponding wafer bump recognition module generated by the bump signal. The three-dimensional attitude data includes the tilt angle data of the loading slot in the X-axis, Y-axis and Z-axis directions, and the bump signal includes the bump position coordinates and the bump contour feature data. Specifically, the implementation process of S1 first involves connecting and debugging the multi-source sensing components with the SMIF loading precision control data analysis platform to ensure a stable data transmission rate of over 100Mbps and a latency of less than 50ms. The collected three-dimensional attitude data from the level gauge includes tilt angle data in the X, Y, and Z axes, with the measurement range for each direction set to -5° to +5° and a measurement accuracy of 0.001°. The acquisition frequency is set to 10Hz to ensure real-time capture of minute attitude changes in the loading slot. The convex signal is acquired using an industrial camera with a resolution of 1280×720 in conjunction with a laser positioning component. This includes the convex position coordinates (coordinate accuracy controlled within 0.01mm) and convex contour feature data (contour extraction error not exceeding 0.005mm). During acquisition, it is necessary to ensure that the vertical distance between the camera and the loading slot is fixed at 300mm and the illumination intensity is maintained at 500-800 lux to avoid ambient light interference with data accuracy. This step involves simultaneously collecting two types of key data, providing comprehensive and accurate raw data support for subsequent anomaly detection and deviation calculation, thus avoiding errors in subsequent processing due to missing or insufficient data.

[0018] S2. Input the convex signal collected in S1 into the three-dimensional pose perception algorithm of the crystal channel. The algorithm compares the convex position coordinates with the preset standard convex coordinates. Combined with the edge fitting results of the convex contour feature data, it determines whether there is an installation abnormality of the wafer on the wafer loading slot. Specifically, S2 first imports the convex signal acquired by S1 into the processing module of the three-dimensional pose perception algorithm of the crystal slot according to the time sequence. The algorithm first performs coordinate transformation on the convex position coordinates, converting the coordinates in the camera coordinate system to the coordinates in the loading slot coordinate system. During the transformation process, preset coordinate transformation parameters (including translation and rotation angles, with the translation error controlled within 0.002mm and the rotation angle error not exceeding 0.001°) need to be introduced. Subsequently, the algorithm compares the transformed convex position coordinates with the preset standard convex coordinates (the standard coordinates need to be pre-entered into the system according to the specifications of different wafer models, such as the standard convex coordinates of a 300mm diameter wafer set to (150mm, 0mm, 0mm)) and calculates the coordinate deviation value; at the same time, it performs edge fitting on the convex contour feature data, uses the least squares method to fit the contour curve, and calculates the ratio of the actual fitted length to the preset standard convex contour length (the standard length is set according to the wafer convex design specifications, such as 10mm). The algorithm combines the coordinate deviation value and the contour length ratio to set a judgment threshold (the coordinate deviation threshold is 0.05mm, and the contour length ratio threshold is 0.95-1.05). When the coordinate deviation exceeds the threshold or the contour length ratio exceeds the set range, the wafer is judged to have an installation abnormality. This step ensures the accuracy of the abnormality judgment through multi-dimensional data comparison and precise threshold setting, avoiding misjudgment or omission.

[0019] S3. When S2 determines that there is an installation abnormality, the three-dimensional attitude data of the level instrument collected by S1 is input into the multi-source sensor dynamic deviation fitting algorithm. The algorithm performs time-series correlation analysis on the tilt angle data at different collection times, removes abnormal fluctuation data, and calculates the equipment tilt of the wafer loading slot in the current working state. The equipment tilt includes the overall tilt angle of the loading slot and the tilt direction vector. Specifically, S3 starts after detecting an installation anomaly. First, it extracts the three-dimensional attitude data of the level instrument collected by S1 from the database of the SMIF loading precision control data analysis platform. The extracted data must contain at least 20 consecutive data sets (corresponding to a 2-second acquisition duration) to ensure data representativeness. Then, the data is input into the multi-source sensor dynamic deviation fitting algorithm. The algorithm first detects outliers using the 3σ criterion (σ is the data standard deviation, calculated based on at least 50 sets of historical normal data, typically not exceeding 0.002°). Data exceeding the range of [mean - 3σ, mean + 3σ] are marked as outliers and removed, retaining only valid data. Next, based on the force characteristics of the loading slot in each direction and the wafer mounting requirements, weight coefficients are assigned to the X, Y, and Z axes (e.g., X-axis weight 0.4, Y-axis weight 0.3, Z-axis weight 0.3; the weight coefficients must be determined by...). (Determined through more than 100 calibration experiments), the weighted tilt value of each valid data point is calculated; then, combined with the difference between the data acquisition time and the average acquisition time, a time-series decay coefficient (set to 0.8 to ensure that the data closer to the current moment has a greater weight) is introduced to correct the weighted tilt value. Finally, the average of all corrected weighted tilt values ​​is taken to obtain the device tilt angle (the calculation result is retained to 4 decimal places, with an accuracy of 0.0001°). This step, through data filtering, weighted calculation, and time-series correction, effectively reduces the impact of environmental interference and data fluctuations on the tilt angle calculation, and improves the calculation accuracy.

[0020] S4. Based on the equipment tilt angle obtained in S3, call the loading dynamic twin calibration model, calculate the deviation between the equipment tilt angle parameter and the preset loading slot standard attitude parameter in the model, and convert the tilt angle deviation value into the number of motor pulses used to drive the loading slot adjustment motor through the built-in attitude-pulse mapping logic of the model. The number of motor pulses includes the number of adjustment pulses in the X-axis direction, the number of adjustment pulses in the Y-axis direction, and the number of adjustment pulses in the Z-axis direction. Specifically, S4 first calls the parameter library of the loading dynamic twin calibration model to extract the preset standard attitude parameters of the loading slot. The standard tilt angle in each direction is set to 0° (allowable error ±0.001°), and the standard deviation is set to 0.001° (determined based on the equipment's factory calibration data). Then, the difference between the equipment tilt angle obtained in S3 and the standard tilt angle in each direction is calculated to obtain the tilt deviation value in each direction (the deviation value is retained to 4 decimal places). At the same time, the pulse coefficient of the motor in each direction is determined according to the model of the adjustment motor (e.g., stepper motor, step angle set to 1.8°, microstepping multiple of 16) (e.g., the pulse coefficient of the X-axis motor is 500 pulses / °, determined by the technical parameters provided by the motor manufacturer and actual debugging). Combined with the equipment operating conditions (e.g., temperature 25±2℃, humidity 40%-60%), the dynamic calibration coefficient is adjusted (set to 1.02 to compensate for the influence of environmental factors on motor operation), and the standard deviation of the current tilt data is calculated (calculated based on the valid data retained in S3). Finally, the tilt deviation value, pulse coefficient, standard deviation, and dynamic calibration coefficient are substituted into the model for calculation to obtain the number of pulses of the motor in each direction (the number of pulses is rounded to the nearest integer to ensure that the motor can execute accurately). This step ensures the accuracy of the motor pulse number conversion by introducing multiple sets of calibration parameters and working condition compensation coefficients, providing a scientific basis for subsequent adjustments.

[0021] S5. The number of motor pulses in each direction calculated in S4 is transmitted to the motor control unit of the SMIF machine. The motor control unit drives the corresponding adjustment motor to operate according to the number of pulses in each direction, thereby driving the wafer loading slot to perform multi-dimensional attitude adjustment to change the position of the abnormally installed wafer in the slot. Specifically, at the start of step S5, the number of motor pulses in each direction obtained in step S4 (e.g., 1500 pulses for the X-axis, 800 pulses for the Y-axis, and 500 pulses for the Z-axis) is transmitted to the motor control unit of the SMIF machine. The control unit receives the data using the RS485 communication protocol, with a transmission baud rate set to 115200bps to ensure stable data transmission. The control unit first converts the pulse count from decimal digital signals to binary control signals that the motor can recognize. Simultaneously, it sets the operating speed (e.g., 500 pulses / second to avoid excessive speed causing vibration in the loading slot) and operating direction (determined based on the tilt deviation direction; for example, if the X-axis deviation is positive, the motor is set to rotate forward). The control unit then sends drive commands to the corresponding adjustment motors, with the command sending interval controlled within 10ms to prevent synchronization errors caused by command delays. After receiving the command, the motor drives the wafer loading slot to move along the corresponding direction via a ball screw transmission mechanism. During the movement, the motor sends an operating status signal (including the current pulse count and operating status) to the control unit every 100 pulses. The control unit receives feedback signals in real time. When the actual number of motor pulses reaches the preset value, it sends a stop command. If the value is not reached, it continues to send drive commands until the number of pulses matches. This step ensures that the loading slot is adjusted according to the preset parameters through precise signal conversion and real-time feedback control, avoiding over-adjustment or under-adjustment.

[0022] S6. During the adjustment process in S5, the SMIF loading precision control data analysis platform collects the updated three-dimensional posture data of the level and the updated convex signal of the convex recognition module in real time. Repeat the operation of S2-S5 until the crystal three-dimensional pose perception algorithm determines that the wafer mounting status meets the preset standard. Specifically, S6 is initiated synchronously during the S5 adjustment process. The SMIF-loaded precision control data analysis platform collects updated 3D posture data from the level gauge at a frequency of 10Hz in real time (collection parameters are consistent with S1 to ensure data comparability) and updated convex signals from the convex recognition module (maintaining the original acquisition accuracy and environmental parameters). The newly collected data must be transmitted to the corresponding algorithm processing module within 100ms, repeating the anomaly judgment process of S2 (using the same judgment threshold and coordinate transformation parameters). If the wafer mounting is still determined to be abnormal, the tilt calculation of S3 (data processing logic is completely consistent with S3) and the pulse number conversion of S4 are continued (the parameter library remains unchanged). Then, the adjustment operation of S5 is performed according to the new pulse number. This cycle continues until the 3D pose perception algorithm of the crystal chute determines that the wafer mounting status meets the preset standard (coordinate deviation ≤ 0.05mm and contour length ratio within the range of 0.95-1.05). At this time, the platform records the final adjustment parameters and the time to meet the standard, and stops the cycle. This step, through real-time data acquisition and cyclic verification mechanisms, ensures that the wafer loading status ultimately meets the standards, thus completely resolving installation anomalies.

[0023] Preferably, the expression for the crystal channel three-dimensional pose sensing algorithm is: ,in, This represents the wafer mounting tolerance value. Let be the weighting coefficient of the i-th group of convex signals. Let be the coordinates of the position of the i-th group of convex pieces. To preset standard convex coordinates, This represents the actual fitted length of the convex profile feature data. The preset standard convex profile length is given by n, where n is the number of convex signal groups collected.

[0024] Specifically, the 3D pose sensing algorithm for wafers calculates wafer mounting deviation values ​​using multi-dimensional data, providing a quantitative basis for anomaly detection. During implementation, the number of convex signal groups participating in the calculation is first determined, typically set to 10 continuously acquired signals to balance computational efficiency and result stability. The weighting coefficient of each convex signal group needs to be adjusted based on the environmental stability during signal acquisition. If the illumination intensity is stable at 500-800 lux and there is no vibration interference, the weighting coefficient is set to 0.1; if slight interference exists, the weighting coefficient is lowered to 0.08 to ensure that interference signals have less impact on the results. The convex position coordinates need to undergo coordinate system transformation first. The difference between the transformed actual coordinates and the preset standard coordinates (determined according to the wafer model, e.g., the standard coordinates for a 200mm diameter wafer are (100mm, 0mm, 0mm)) is calculated, and then combined with the ratio of the actual fitted length of the convex contour to the standard length (e.g., 8mm), and substituted into the algorithm for calculation. The least squares method is used for convex profile fitting, with the fitting error controlled within 0.005mm. The standard length needs to be confirmed through wafer design drawings and entered into the system in advance. This algorithm, through weight allocation and multi-parameter fusion, enables the calculated wafer mounting deviation value to reach an accuracy of 0.001mm, accurately reflecting the actual wafer mounting status and avoiding errors caused by single-parameter judgments. This provides reliable data support for whether to initiate correction in the future.

[0025] Preferably, the expression for the multi-source sensing dynamic deviation fitting algorithm is: ,in, The value is the equipment tilt angle, and m is the number of data sets collected from the level gauge. For the j-th group of level data Tilt angles of the X, Y, and Z axes. These are the tilt weighting coefficients in the X, Y, and Z axis directions. This is the time-series decay coefficient. Let j be the time of data collection for the j-th group. This is the average of all data collection times.

[0026] Specifically, a multi-source sensor dynamic deviation fitting algorithm is used to accurately extract the equipment tilt angle from the level data and eliminate interference from abnormal data. During implementation, the number of levels to be collected is first determined, typically 30 consecutive sets (corresponding to a 3-second acquisition time) to ensure statistical representativeness. The weighting coefficients for tilt angles in each direction are set based on the structural characteristics of the loading slot. The X-axis, bearing the main weight of the wafer, has a weighting coefficient of 0.45, while the Y-axis and Z-axis have weights of 0.3 and 0.25 respectively. These values ​​are determined through more than 150 no-load and full-load calibration experiments to ensure accurate reflection of the impact of tilt in each direction on wafer mounting. The time-series attenuation coefficient is set to 0.85. This value is derived from the data fluctuation patterns during equipment operation; data acquired closer to the current moment has a greater weight after attenuation, reducing the impact of errors caused by environmental changes in early data. Outlier detection employs the 3σ criterion, where the standard deviation is calculated based on 100 sets of level gauge data continuously collected during normal equipment operation, typically controlled within 0.002°. Data exceeding the range of [mean - 3σ, mean + 3σ] are considered outliers and discarded. Through the above parameter settings and data processing flow, the algorithm calculates the equipment tilt accuracy to 0.0001°, accurately reflecting the current posture of the loading tank and providing precise deviation data for subsequent pulse count conversion.

[0027] Preferably, the expression for loading the dynamic twin calibration model is: ,in, Let k be the number of pulses for the k-th adjustment motor, where k corresponds to the motor in the X-axis, Y-axis, and Z-axis directions, respectively. Let be the pulse coefficient of the k-th motor. Let the standard tilt angle be in the k-th direction. Let be the standard deviation of the tilt data in the k-th direction. To preset the standard deviation, is the dynamic calibration coefficient in the k-th direction.

[0028] Specifically, a dynamic twin calibration model is installed to convert the equipment tilt angle into the number of pulses that the motor can execute, achieving a precise correspondence between deviation and adjustment action. Before implementation, the standard tilt angle and standard deviation for each direction need to be preset in the model parameter library. The standard tilt angle is set to 0° (allowing ±0.001° factory error), and the standard deviation is determined to be 0.0015° based on the data from 200 calibration experiments before the equipment leaves the factory, ensuring that the parameters conform to the initial state of the equipment. The motor pulse coefficient needs to be determined according to the motor model. For example, if a stepper motor with a step angle of 1.8° and a microstepping factor of 32 is selected, the X-axis pulse coefficient is set to 600 pulses / °, and the Y-axis and Z-axis pulse coefficients are set to 550 pulses / ° due to the smaller load. These values ​​need to be verified through motor stall experiments and actual load tests to ensure that the motor can generate an accurate displacement for each corresponding number of pulses received. The dynamic calibration coefficient needs to be adjusted based on real-time operating conditions. When the ambient temperature is between 23-27℃ and the humidity is between 40%-60%, it is set to 1.03. If the temperature exceeds this range, the coefficient is adjusted to 1.05 to compensate for the impact of temperature changes on the motor's operating accuracy. The standard deviation of the current tilt data is calculated based on the valid data retained in weight 3 to ensure that it reflects the dispersion of the current data. The error of the motor pulse count calculated by the model is controlled within ±1 pulse, which can guarantee the accuracy of the motor driving the loading slot adjustment and avoid over- or under-adjustment due to inaccurate pulse count.

[0029] Preferably, the SMIF-loaded precision control data analysis platform uses a data fusion formula during the data acquisition process: ,in, The merged data values For data fusion weights, Collect data for the level. For acquiring data of the convex plate signal, This represents the error value of the level data. This represents the error value of the convex plate signal data.

[0030] Specifically, SMIF incorporates a data fusion formula from a precision control data analysis platform to integrate data from both the level and convex signal sources, improving data reliability and subsequent processing efficiency. During implementation, the data fusion weights are dynamically adjusted based on the acquisition accuracy of the two data types. If the level measurement accuracy reaches 0.001° and the convex signal coordinate accuracy is 0.01mm, the weight is set to 0.6 to ensure a larger proportion of the more accurate level data. If the convex signal accuracy is improved to 0.005mm through laser-assisted positioning, the weight is adjusted to 0.5 to achieve balanced fusion of the two data types. Level data undergoes outlier removal and smoothing, with the processed data error controlled within 0.0005°. Convex signal data requires contour fitting and coordinate transformation to ensure a consistent data format. The error value of the level data is determined based on the equipment calibration report and is typically 0.001°; the error value of the convex signal is calculated through multiple repeated acquisitions and is set to 0.005mm. During the data fusion process, the platform needs to monitor the acquisition status of both types of data in real time. If any type of data is interrupted or the error exceeds the threshold, the weight is automatically adjusted. For example, if the level indicator data is abnormal, the weight is reduced to 0.3, while the weight of the convex signal is increased to 0.7, ensuring that the fused data still reflects the actual status of the equipment. Through data fusion, the stability of the platform's output data is improved by more than 40%, providing more reliable input for subsequent algorithm processing and reducing the impact of single data anomalies on the overall process.

[0031] Preferably, in step S4, the motor pulse count conversion process also employs a compensation formula: ,in, The number of motor pulses after compensation. Let be the tilt compensation coefficient for the k-th motor. The historical data compensation coefficient for the k-th motor is... Let be the tilt angle in the k-th direction of the j-th group of level data.

[0032] Specifically, the motor pulse count compensation formula is used to correct the base pulse count and eliminate adjustment errors caused by various factors during equipment operation. During implementation, the tilt compensation coefficient needs to be set according to the equipment tilt range. When the tilt is between 0° and 0.5°, the compensation coefficients for the X, Y, and Z axes are all set to 0.02; when the tilt exceeds 0.5°, the compensation coefficient is adjusted to 0.03. This value is obtained through adjustment experiments of the equipment under different tilt states to ensure that it can offset the impact of increased tilt angle on the motor's operating accuracy. The historical data compensation coefficient needs to be calculated based on the equipment's adjustment records for the past 30 days. If the average deviation between the actual and theoretical displacement of the motor in historical adjustments is 0.002mm, the compensation coefficient is set to 0.015; if the deviation exceeds 0.003mm, the coefficient is adjusted to 0.02 to correct for mechanical wear errors caused by long-term motor operation. During calculation, historical tilt angle data in each direction of weight 3 needs to be extracted, typically selecting the last 50 sets of valid data to ensure that the historical data reflects the recent operating status of the equipment. The compensated motor pulse count needs to be verified. By comparing the actual tilt change of the motor after operation with the theoretical change, it is ensured that the error after compensation is controlled within 0.001mm. The application of this compensation formula improves the motor adjustment accuracy by 30%, effectively solving the problem of inaccurate basic pulse count caused by equipment wear and environmental changes, and ensuring that the wafer can accurately reach the standard position after adjustment.

[0033] Preferably, step S3 includes the following sub-steps: S31 Extracting the three-dimensional attitude data of the level instrument collected in S1 from the database of the SMIF loaded with the precision control data analysis platform, and sorting the data according to the acquisition time sequence to obtain a time-series data sequence; S32 Inputting the time-series data sequence into the preprocessing module of the multi-source sensor dynamic deviation fitting algorithm, and the module performs outlier detection on the data, removing outliers that exceed the preset tilt angle range. Data is marked as outlier and removed, while valid data is retained. S33 assigns corresponding weight coefficients to the X, Y, and Z axes based on the importance of the tilt angle in each direction for the retained valid data. S34 calculates the weighted skew value corresponding to each valid data point; S34 introduces a time-series decay coefficient based on the difference between the data acquisition time and the average acquisition time. The weighted tilt values ​​are corrected, and the final tilt angle is calculated by averaging the results. .

[0034] Specifically, the step-by-step implementation process of S3 involves: S31 extracting the three-dimensional attitude data of the level instrument collected in S1 from the database of the precision control data analysis platform loaded in SMIF. During extraction, 20 consecutive data sets (corresponding to a 2-second acquisition duration and a 10Hz acquisition frequency) need to be selected to ensure the data covers the complete attitude change cycle. After extraction, the data is sorted in ascending order according to the acquisition timestamp to form a time-series data sequence. The sorting error is controlled within 10ms to prevent time sequence errors from causing subsequent analysis biases. S32 inputting the time-series data sequence into the multi-source sensor dynamic deviation fitting algorithm preprocessing module uses the 3σ criterion for outlier detection. The standard deviation is calculated based on 100 consecutive historical data sets during normal equipment operation, typically 0.002°. The preset tilt angle range is set to [-0.5°, 0.5°]. Data exceeding this range or deviating from the mean by more than 3σ are marked as outliers and removed. The amount of valid data retained needs to be... At least 15 sets of data are used to avoid abnormal data interfering with the calculation results. S33 assigns weight coefficients to the X, Y, and Z axes based on the structural characteristics of the loading slot and the wafer load requirements. The X-axis, which directly bears the wafer weight, is set to 0.4, while the Y and Z axes are each set to 0.3. These values ​​have been verified through 100 no-load and full-load calibration experiments to ensure accurate reflection of the impact of tilt in each direction on wafer mounting. The weighted tilt value for each set of valid data is then calculated according to these weights. S34 introduces a time-lapse coefficient of 0.8 (set based on the timeliness of data during equipment operation; data closer to the current moment has a higher weight). The weighted tilt value is corrected for time-lapse by combining the difference between the data acquisition time of each set and the average acquisition time of all data. Finally, the arithmetic mean of all corrected weighted tilt values ​​is taken to obtain the equipment tilt angle with an accuracy of 0.0001°, providing accurate deviation data for subsequent pulse count conversion.

[0035] Preferably, step S4 includes the following sub-steps: S41 retrieves preset standard attitude parameters of the loading slot from the parameter library of the loading dynamic twin calibration model, which include standard tilt angles in the X-axis, Y-axis, and Z-axis directions. and standard deviation S42 will obtain the equipment tilt angle from S3. The difference between the tilt angle and the standard tilt angle in each direction is calculated to obtain the tilt deviation value in each direction; S43 adjusts the motor model and performance parameters according to each direction and calls the corresponding pulse coefficient. and dynamic calibration coefficient Simultaneously calculate the standard deviation of the current tilt data. S44 substitutes the tilt deviation value, pulse coefficient, standard deviation, and dynamic calibration coefficient into the expression of the loading dynamic twin calibration model to calculate the initial pulse number of the adjustment motor in each direction. .

[0036] Specifically, the step-by-step implementation process of S4 involves: S41 retrieving preset standard attitude parameters of the loading slot from the loading dynamic twin calibration model parameter library. These parameters include the standard tilt angles and standard deviations in the X, Y, and Z axes. The standard tilt angle is set to 0° (allowing ±0.001° of factory calibration error), and the standard deviation is 0.0015° (determined based on 200 no-load calibration test data before the equipment leaves the factory). During the retrieval, the completeness and validity of the parameters must be confirmed through a parameter verification mechanism. If any parameters are missing or incorrect, they are automatically retrieved from the backup parameter library to ensure accuracy. S42 calculates the difference between the equipment tilt angle calculated in S3 and the standard tilt angle of each axis to obtain the tilt deviation values ​​for the X, Y, and Z axes. The calculation retains four decimal places, with accuracy controlled within 0.0001° to avoid the difference calculation error affecting the subsequent pulse number conversion accuracy. S43 determines the key parameters based on the model and specifications of the adjustment motor, such as selecting… A stepper motor with a step angle of 1.8° and a microstepping factor of 16 is used. The X-axis motor pulse coefficient is set to 500 pulses / °, while the Y-axis and Z-axis motors are set to 450 pulses / ° due to their smaller loads. These values ​​have been verified through motor stall experiments and actual load tests to ensure that the pulse count accurately corresponds to the actual displacement of the motor. At the same time, the standard deviation of the current tilt data is calculated based on the valid data retained in S3, with a sample size of no less than 15 sets. Then, the dynamic calibration coefficient is determined to be 1.02 based on the real-time environmental conditions (temperature 25±2℃, humidity 40%-60%). When the temperature exceeds this range, it is adjusted to 1.05 to compensate for the impact of environmental factors on the motor's operating accuracy. S44 substitutes the tilt deviation value of each axis, the corresponding pulse coefficient, the current standard deviation, and the dynamic calibration coefficient into the loading dynamic twin calibration model. The initial pulse count of the motor adjustment for each axis is obtained through the built-in calculation logic of the model. The calculation result is rounded to an integer, and the error is controlled within ±1 pulse to ensure that the motor can perform adjustment actions according to the precise pulse count.

[0037] Preferably, step S5 includes the following sub-steps: S51 The initial pulse count of each direction motor obtained in S4 is transmitted to the motor control unit of the SMIF machine. The control unit performs format conversion on the pulse count, converting the digital signal into a control signal recognizable by the motor; S52 The motor control unit sends the converted control signal to the corresponding motor control unit. axis, axis, The axis adjustment motor sends a drive command, which includes parameters such as the number of pulses, operating speed, and operating direction. After receiving the drive command, the adjustment motor starts to operate according to the command parameters. Through the transmission mechanism, it drives the wafer loading slot to adjust its posture along the corresponding direction. During the adjustment process, the operating status signal is fed back in real time. The motor control unit receives the operating status signal fed back by the motor and confirms whether the motor has completed the operation according to the preset number of pulses. If it has not completed the operation, it continues to send drive commands until the number of motor operating pulses reaches the preset value.

[0038] Specifically, the step-by-step implementation process of S5 is as follows: S51 transmits the initial pulse counts of each axis motor obtained in S4 to the motor control unit of the SMIF machine via the industrial communication interface. The transmission uses the RS485 communication protocol, with a baud rate set to 115200bps and a transmission delay controlled within 50ms to prevent data transmission lag from causing asynchronous multi-axis adjustments. After receiving the data, the motor control unit first verifies the rationality of the pulse count, confirming that the pulse count is within the maximum allowable pulse range of the motor (100-5000 pulses). After the verification is passed, the decimal digital pulse signal is converted into a binary control signal that the motor can recognize. The conversion error must be less than 0.1% to ensure the accuracy of the control signal. In S52, the motor control unit generates drive commands based on the converted control signals. The commands include the number of pulses required for motor operation, the operating speed (set to 500 pulses / second to avoid excessive speed causing vibration of the loading slot that affects wafer stability), and the operating direction (determined based on the tilt deviation direction of each axis; for example, if the X-axis deviation is positive, the motor rotates forward; if the deviation is negative, the motor rotates forward). (Negative response reverses) After the instruction is generated, it is stored in the local cache for easy fault tracing and data review. Then, drive instructions are sent to the corresponding motors at 10ms intervals to prevent instruction congestion from causing motor response delays. After receiving the drive instructions, the S53 adjustment motors drive the wafer loading slots to move in the corresponding directions through the ball screw transmission mechanism. During the movement, the motors feed back an operating status signal to the motor control unit every 100 pulses. The signal includes parameters such as the current number of pulses operated, motor operating voltage, and operating current. The feedback frequency is consistent with the instruction sending frequency to ensure that the control unit can monitor the motor operating status in real time. The S54 motor control unit compares the actual number of operating pulses with the preset number of pulses. If the difference is within ±1 pulse, the adjustment is considered complete, and a stop instruction is sent. If the difference exceeds 5 pulses, the adjustment is considered abnormal, and the drive instruction is resent for correction until the actual number of operating pulses matches the preset value. This ensures that the loading slots complete the attitude adjustment according to the preset parameters and avoids over- or under-adjustment affecting the wafer mounting accuracy.

[0039] The 3D wafer position and orientation sensing algorithm is the core algorithm used in this application for accurately determining the wafer mounting status. Essentially, it identifies mounting anomalies in the wafer within the loading slot through multi-dimensional data fusion and quantization calculation. To implement this algorithm, the following steps are required: first, an industrial camera and laser positioning components are used to acquire bump signals, including bump position coordinates (accuracy controlled within 0.01mm) and bump contour feature data (contour extraction error not exceeding 0.005mm). Then, the acquired bump position coordinates are converted from the camera coordinate system to the loading slot coordinate system (conversion parameter error controlled within 0.002mm and 0.001°). Subsequently, the algorithm compares the transformed coordinates with preset standard convex coordinates (set according to the wafer model, such as (150mm, 0mm, 0mm) for a 300mm diameter wafer) to calculate the deviation value. Simultaneously, it performs least-squares fitting on the convex contour data, calculating the ratio of the actual fitted length to the standard contour length (e.g., 10mm). Finally, combining the coordinate deviation value with the contour length ratio, and applying preset thresholds (coordinate deviation threshold 0.05mm, contour length ratio range 0.95-1.05), it determines whether the installation is abnormal. This algorithm replaces traditional single-dimensional anomaly detection methods, achieving accurate identification of the wafer installation status. Its significance lies in avoiding invalid or uncorrected situations caused by misjudgments or omissions, providing a reliable starting basis for subsequent correction processes, and ensuring the stable operation of the wafer transmission equipment.

[0040] The multi-source sensor dynamic deviation fitting algorithm is a key algorithm for extracting accurate equipment tilt angles from level data, aiming to eliminate the influence of data fluctuations and environmental interference on tilt angle calculations. In its implementation, the algorithm first acquires three-dimensional attitude data of the level (including tilt angles along the X, Y, and Z axes, with a measurement accuracy of 0.001° and a sampling frequency of 10Hz) from the data acquisition terminal. Twenty consecutive sets of valid data (corresponding to 2-second durations) are selected and sorted by time. Next, the 3σ criterion (based on the standard deviation calculated from 100 sets of historical normal data, typically 0.002°) is used to eliminate outlier data exceeding the range of [mean - 3σ, mean + 3σ], retaining at least 15 sets of valid data. Then, weighting coefficients for each axis are assigned according to the force characteristics of the loading tank (0.4 for X-axis, 0.3 for Y-axis, and 0.3 for Z-axis, determined through 100 calibration experiments). The weighted tilt value for each set of data is calculated, and a time-series decay coefficient of 0.8 (set according to the data timeliness rules) is introduced to correct the weighted tilt value. Finally, the mean value is taken to obtain the equipment tilt angle with an accuracy of 0.0001°. The algorithm transforms raw sensor data into a quantitative deviation value that accurately reflects the device's posture. Its significance lies in providing accurate data support for subsequent motor pulse count conversion, avoiding insufficient correction accuracy due to tilt calculation errors, and reducing the risk of wafer damage.

[0041] The dynamic twin calibration model is the core model connecting the equipment tilt angle and the motor adjustment action. By establishing a mapping relationship between deviation and execution parameters, precise correction control is achieved. To implement this model, standard attitude parameters must first be preset in the parameter library, including the standard tilt angle of each axis (0°, with an allowable error of ±0.001°) and standard deviation (0.0015°, determined based on 200 factory calibration data). At the same time, the pulse coefficients of different motor models (e.g., for a motor with a step angle of 1.8° and a microstepping factor of 16, 500 pulses / ° for the X-axis and 450 pulses / ° for the Y-axis) and dynamic calibration coefficients (1.02 under normal conditions and 1.05 under abnormal temperature conditions) must be stored. During model execution, the deviation between the device tilt angle and the standard tilt angle is first calculated. Combined with the standard deviation of the current tilt data (calculated based on 15 sets of valid data), the deviation value, pulse coefficient, standard deviation, and calibration coefficient are substituted into the built-in logic to calculate the initial pulse count for each axis motor (error ±1 pulse). The pulse count can also be corrected using a compensation formula (introducing a tilt compensation coefficient of 0.02-0.03 and a historical data compensation coefficient of 0.015-0.02). The model's function is to transform the abstract tilt deviation into specific pulse parameters that the motor can execute. Its significance lies in achieving precise matching between the deviation and the adjustment action, avoiding over- or under-adjustment, ensuring the wafer quickly returns to the standard position, and improving equipment operating efficiency.

[0042] The SMIF-loaded precision control data analysis platform is the core data processing and scheduling hub of this invention, undertaking the functions of data acquisition, storage, transmission, and collaborative scheduling. Implementing this platform requires establishing multi-source data access interfaces, connecting with a level (transmission rate 100Mbps, latency within 50ms), a convex recognition module (1280×720 resolution camera), and a motor control unit (RS485 protocol, 115200bps baud rate) to acquire 3D posture data, convex signals, and motor status signals in real time. The platform has a built-in database storing historical data and standard parameters, supporting data sorting by time and preliminary outlier filtering (based on a preset range [-0.5°, 0.5°]). It also has a data distribution function, sending processed convex signals to the crystal 3D pose perception algorithm, level data to the multi-source sensor dynamic deviation fitting algorithm, and then transmitting motor pulse counts to the control unit. Furthermore, it can collect and update data in real time during adjustment, driving the process loop until the target is met. The platform's function is to enable data flow and collaborative work among various technical modules. Its significance lies in integrating dispersed sensing and control links to form a closed-loop correction system, avoiding inefficiency or data gaps caused by independent operation of each module, ensuring the smoothness and efficiency of the entire correction process, and improving the automation and intelligence level of wafer transmission equipment.

[0043] like Figure 2As shown, a wafer loading correction device includes: a multi-source sensor data acquisition unit connected to a level and bump recognition module on the wafer loading slot, used to acquire real-time three-dimensional attitude data of the level and bump signals, and transmit the acquired data to an SMIF loading precision control data analysis unit; an SMIF loading precision control data analysis unit connected to the multi-source sensor data acquisition unit and a crystal lane three-dimensional pose perception algorithm processing unit, used to store, sort, and preprocess the acquired data, and send the processed data to the corresponding algorithm processing units; and a crystal lane three-dimensional pose perception algorithm processing unit connected to the SMIF loading precision control data analysis unit and a multi-source sensor dynamic deviation fitting algorithm processing unit, used to receive the processed bump signals and determine the wafer mounting status, and if an abnormality is detected... This triggers the operation of the multi-source sensor dynamic deviation fitting algorithm processing unit. This unit, connected to the crystal 3D pose perception algorithm processing unit and the loading dynamic twin calibration model calculation unit, receives level data and calculates the device tilt, transmitting the tilt data to the loading dynamic twin calibration model calculation unit. The loading dynamic twin calibration model calculation unit, connected to the multi-source sensor dynamic deviation fitting algorithm processing unit and the motor control drive unit, calculates the number of motor pulses based on the device tilt and transmits the pulse count to the motor control drive unit. The motor control drive unit, connected to the loading dynamic twin calibration model calculation unit and the wafer loading slot adjustment motor, receives the motor pulse count and drives the adjustment motor to adjust the wafer loading slot's attitude.

[0044] A wafer loading correction method and apparatus are disclosed, which can comprehensively optimize the wafer loading correction effect and overcome the shortcomings of existing technologies. Its advantages are: First, comprehensive and real-time data acquisition, relying on multi-source sensing components to synchronously acquire level gauge attitude data and tab signals, providing a rich data foundation for subsequent processing; second, accurate anomaly detection and deviation calculation, using a crystal 3D pose perception algorithm and a multi-source sensing dynamic deviation fitting algorithm to accurately identify wafer mounting anomalies and precisely calculate equipment tilt; third, a scientific and controllable adjustment process, using a loading dynamic twin calibration model to calculate motor pulse counts to ensure reasonable adjustment parameters, while relying on the SMIF loading precision control data analysis platform to achieve real-time feedback and repeated verification. To address the shortcomings of existing technologies that rely on single sensor data, leading to large errors in judgment and calculation, this method integrates two types of key data and processes them with professional algorithms, significantly improving the accuracy of judgment and calculation, and providing reliable data support for correction. Regarding the lack of dynamic calibration and real-time feedback in existing technologies, and the tendency for single adjustments to be inadequate or excessive, this method continuously collects and updates data during the adjustment process, repeatedly executing anomaly judgment, deviation calculation, and motor adjustment procedures until the wafer installation meets the standards, effectively avoiding false alarms and wafer damage, and ensuring stable equipment operation and wafer processing quality.

[0045] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A wafer loading and bias correction method, characterized in that, Includes the following steps: S1. The SMIF loading precision control data analysis platform calls the multi-source sensing components to collect in real time the three-dimensional attitude data output by the level instrument configured on the wafer loading slot in the wafer transfer equipment and the corresponding wafer bump recognition module generated by the bump signal. The three-dimensional attitude data includes the tilt angle data of the loading slot in the X-axis, Y-axis and Z-axis directions, and the bump signal includes the bump position coordinates and the bump contour feature data. S2. Input the convex signal collected in S1 into the three-dimensional pose perception algorithm of the crystal channel. The algorithm compares the convex position coordinates with the preset standard convex coordinates. Combined with the edge fitting results of the convex contour feature data, it determines whether there is an installation abnormality of the wafer on the wafer loading slot. S3. When S2 determines that there is an installation abnormality, the three-dimensional attitude data of the level instrument collected by S1 is input into the multi-source sensor dynamic deviation fitting algorithm. The algorithm performs time-series correlation analysis on the tilt angle data at different collection times, removes abnormal fluctuation data, and calculates the equipment tilt of the wafer loading slot in the current working state. The equipment tilt includes the overall tilt angle of the loading slot and the tilt direction vector. S4. Based on the equipment tilt angle obtained in S3, call the loading dynamic twin calibration model, calculate the deviation between the equipment tilt angle parameter and the preset standard attitude parameter of the loading slot in the model, and convert the tilt angle deviation value into the number of motor pulses used to drive the loading slot adjustment motor through the built-in attitude-pulse mapping logic of the model. The number of motor pulses includes the number of adjustment pulses in the X-axis direction, the number of adjustment pulses in the Y-axis direction, and the number of adjustment pulses in the Z-axis direction.

2. The wafer loading and correction method according to claim 1, characterized in that, The method also includes: S5. The number of motor pulses in each direction calculated in S4 is transmitted to the motor control unit of the SMIF machine. The motor control unit drives the corresponding adjustment motor to operate according to the number of pulses in each direction, thereby driving the wafer loading slot to perform multi-dimensional attitude adjustment to change the position of the abnormally installed wafer in the slot. S6. During the adjustment process in S5, the SMIF loading precision control data analysis platform collects the updated three-dimensional attitude data of the level and the updated convex signal of the convex recognition module in real time. The operation of S2-S5 is repeated until the three-dimensional position perception algorithm of the crystal channel determines that the wafer installation status meets the preset standard. The expression for the crystal channel 3D pose sensing algorithm is: ,in, This represents the wafer mounting tolerance value. Let be the weighting coefficient of the i-th group of convex signals. Let be the coordinates of the position of the i-th group of convex pieces. To preset standard convex coordinates, This represents the actual fitted length of the convex profile feature data. The preset standard convex profile length is given by n, where n is the number of convex signal groups collected.

3. The wafer loading and correction method according to claim 1, characterized in that, The expression for the multi-source sensing dynamic deviation fitting algorithm is: ,in, The value is the equipment tilt angle, and m is the number of data sets collected from the level gauge. For the j-th group of level data Tilt angles of the X, Y, and Z axes. These are the tilt weighting coefficients in the X, Y, and Z axis directions. This is the time-series decay coefficient. Let j be the time of data collection for the j-th group. This is the average of all data collection times.

4. The wafer loading and correction method according to claim 1, characterized in that, The expression for the loaded dynamic twin calibration model is: ,in, Let k be the number of pulses for the k-th adjustment motor, where k corresponds to the motor in the X-axis, Y-axis, and Z-axis directions, respectively. Let be the pulse coefficient of the k-th motor. Let the standard tilt angle be in the k-th direction. Let be the standard deviation of the tilt data in the k-th direction. To preset the standard deviation, is the dynamic calibration coefficient in the k-th direction.

5. The wafer loading and correction method according to claim 1, characterized in that, The SMIF-loaded precision control data analysis platform employs a data fusion formula during data acquisition: ,in, The merged data values For data fusion weights, Collect data for the level. For acquiring data of the convex plate signal, This represents the error value of the level data. This represents the error value of the convex plate signal data.

6. The wafer loading and correction method according to claim 1, characterized in that, In step S4, the motor pulse count conversion process also employs a compensation formula: ,in, The number of motor pulses after compensation. Let be the tilt compensation coefficient for the k-th motor. The historical data compensation coefficient for the k-th motor is... Let be the tilt angle in the k-th direction of the j-th group of level data.

7. The wafer loading and correction method according to claim 1, characterized in that, S3 includes the following steps: S31 Extract the three-dimensional attitude data of the level instrument collected in S1 from the database of the SMIF-loaded precision control data analysis platform, and sort the data according to the acquisition time sequence to obtain a time-series data sequence; S32 Input the time-series data sequence into the preprocessing module of the multi-source sensor dynamic deviation fitting algorithm, and the module performs outlier detection on the data, removing outliers that exceed the preset tilt angle range. Data is marked as outlier and removed, while valid data is retained. S33 assigns corresponding weight coefficients to the X, Y, and Z axes based on the importance of the tilt angle in each direction for the retained valid data. S34 calculates the weighted skew value corresponding to each valid data point; S34 introduces a time-series decay coefficient based on the difference between the data acquisition time and the average acquisition time. The weighted tilt values ​​are corrected, and the final tilt angle is calculated by averaging the results. .

8. The wafer loading and correction method according to claim 1, characterized in that, S4 includes the following sub-steps: S41 retrieves preset standard attitude parameters of the loading slot from the parameter library of the loading dynamic twin calibration model. These parameters include standard tilt angles in the X-axis, Y-axis, and Z-axis directions. and standard deviation ; S42 obtains the equipment tilt angle from S3. The difference between the tilt angle and the standard tilt angle in each direction is calculated to obtain the tilt deviation value in each direction; S43 adjusts the motor model and performance parameters according to each direction, and calls the corresponding pulse coefficient. and dynamic calibration coefficient Simultaneously calculate the standard deviation of the current tilt data. S44 substitutes the tilt deviation value, pulse coefficient, standard deviation, and dynamic calibration coefficient into the expression of the loading dynamic twin calibration model to calculate the initial pulse number of the adjustment motor in each direction. .

9. The wafer loading and correction method according to claim 2, characterized in that, S5 includes the following sub-steps: S51 The initial pulse counts of the motors in each direction obtained in S4 are transmitted to the motor control unit of the SMIF machine. The control unit performs format conversion on the pulse counts, converting the digital signals into control signals that the motors can recognize; S52 The motor control unit sends the converted control signals to the corresponding motors... axis, axis, The axis adjustment motor sends a drive command, which includes parameters such as the number of pulses, operating speed, and operating direction. After receiving the drive command, the adjustment motor starts to operate according to the command parameters. Through the transmission mechanism, it drives the wafer loading slot to adjust its posture along the corresponding direction. During the adjustment process, the operating status signal is fed back in real time. The motor control unit receives the operating status signal fed back by the motor and confirms whether the motor has completed the operation according to the preset number of pulses. If it has not completed the operation, it continues to send drive commands until the number of motor operating pulses reaches the preset value.

10. A wafer loading and correction device, characterized in that, include: A multi-source sensor data acquisition unit is connected to the level and bump recognition module on the wafer loading slot. It is used to acquire the three-dimensional attitude data of the level and the bump signal in real time, and transmit the acquired data to the SMIF loading precision control data analysis unit. The SMIF is equipped with a precision control data analysis unit, which is connected to the multi-source sensor data acquisition unit and the crystal channel three-dimensional pose perception algorithm processing unit. It is used to store, sort and preprocess the acquired data, and send the processed data to the corresponding algorithm processing units. The three-dimensional pose perception algorithm processing unit of the crystal channel is connected to the SMIF loading precision control data analysis unit and the multi-source sensor dynamic deviation fitting algorithm processing unit. It is used to receive the processed convex signal and determine the wafer mounting status. If the determination is abnormal, the multi-source sensor dynamic deviation fitting algorithm processing unit is triggered to work. The multi-source sensing dynamic deviation fitting algorithm processing unit is connected to the crystal channel three-dimensional pose perception algorithm processing unit and the loading dynamic twin calibration model calculation unit. It is used to receive level data and calculate the device tilt, and transmit the tilt data to the loading dynamic twin calibration model calculation unit. The system is equipped with a dynamic twin calibration model calculation unit, which is connected to a multi-source sensor dynamic deviation fitting algorithm processing unit and a motor control drive unit. This unit is used to calculate the number of motor pulses based on the equipment tilt and transmit the number of pulses to the motor control drive unit. The motor control drive unit is connected to the dynamic twin calibration model calculation unit and the adjustment motor of the wafer loading slot. It is used to receive the number of motor pulses and drive the adjustment motor to perform attitude adjustment of the wafer loading slot.

Images