Edge chipping prevention polishing method and system suitable for ultra-thin wafers

Abstract

The application discloses an edge-crack-preventing grinding processing method and system suitable for ultra-thin wafers, and belongs to the technical field of grinding processing. The method comprises the following steps: obtaining current parameters of a wafer, calculating and generating a grinding control instruction, obtaining real-time parameters in a grinding operation process and generating a new grinding control instruction, gradually thinning the wafer to a target thickness, continuously monitoring the stress state of the wafer in the grinding operation process to determine the deformation of the wafer, generating a control instruction set, checking edge cracking and defects of the wafer according to the control instruction set, generating a rejection item, obtaining target processing equipment parameters, combining the rejection item to perform trimming processing on the wafer, recording processing parameters, and completing processing on the wafer. The grinding processing method of the application realizes the collaborative control of thickness thinning and stress stability in the grinding process, effectively inhibits the generation and extension of edge cracks, and significantly reduces the edge cracking failure rate.

Description

Technical Field

[0001] This application relates to the field of grinding and polishing technology, and specifically to a grinding and polishing method, system, electronic device, and computer-readable storage medium adapted to prevent edge breakage of ultrathin wafers. Background Technology

[0002] In the ultra-thin wafer back-side grinding process, a vacuum adsorption ring is typically used to hold the wafer to the front side, and then grinding equipment is used to grind the back side of the wafer, gradually thinning it to 60μm. However, for wafers with stepped edges, edge cracks are easily generated during the grinding process, and these cracks can easily extend into the wafer, causing product scrap. Currently, there is no effective grinding solution in the industry, and the wafer edge cracking defect rate remains high. Summary of the Invention

[0003] To address the technical problem that edge cracks are easily triggered during wafer grinding, and that these cracks can easily extend into the wafer, causing product scrap, this application aims to provide a method, system, electronic device, and computer-readable storage medium adapted to ultra-thin wafers for edge-crack prevention grinding, which can effectively reduce the risk of edge cracking during wafer grinding.

[0004] In a first aspect, the edge-fracture-resistant grinding process for ultra-thin wafers described in this application includes:

[0005] The current thickness and vacuum level of the wafer are obtained, and grinding control commands are calculated and generated. During the grinding operation, the real-time thickness and vacuum level of the wafer are obtained, and new grinding control commands are generated to gradually thin the wafer to the target thickness while keeping the adsorption force within the stable boundary range of the adsorption force throughout the process.

[0006] The grinding process continuously monitors the stress state of the wafer to generate an initial stress set, determines the wafer deformation based on the initial stress set, and generates a control instruction set.

[0007] The wafer is inspected for edge cracks and defects according to the control instruction set, and defect rejection items are generated.

[0008] Obtain the target processing equipment parameters, combine them with defect rejection criteria, perform grinding and trimming on the wafer, record the processing parameters, and complete the wafer processing.

[0009] Preferably, the steps of obtaining the current wafer thickness and vacuum level, calculating and generating grinding control commands, obtaining the real-time wafer thickness and vacuum level during the grinding operation, and generating new grinding control commands to gradually thin the wafer to the target thickness while maintaining the adsorption force within the stable adsorption force boundary range throughout the process, further include:

[0010] Obtain the wafer structure diagram, analyze the coordinate values ​​of the step surface, set the clamping position in the area outside the step surface in the wafer structure diagram, set the adsorption ring in the clamping position, and control the vacuum degree of the adsorption ring to complete the wafer fixation.

[0011] The predetermined value of the grinding spindle is obtained, and the vacuum degree of the adsorption ring is adjusted based on the predetermined value of the grinding spindle to make the wafer subjected to uniform force. After the wafer is fixed and the adsorption force is adjusted, the grinding control command is generated.

[0012] Preferably, the step of obtaining a predetermined value for the grinding spindle, adjusting the vacuum level of the adsorption ring based on the predetermined value to ensure uniform force on the wafer, and generating a grinding control command after completing wafer fixation and adsorption force adjustment includes:

[0013] Obtain the rotation speed and turning angle of the grinding spindle and the grinding surface information of the wafer from the predetermined values ​​to determine the target adsorption force;

[0014] Based on the target adsorption force, feature extraction is performed to obtain the adsorption force feature vector;

[0015] The adsorption force feature vector is compared with a preset threshold. If it is greater than the preset threshold, the first force value is determined.

[0016] Based on the first force value, a second force value is obtained through smoothing filtering.

[0017] Based on the second force value, combined with the rotation speed value and steering angle, the vacuum degree of the adsorption ring is updated.

[0018] Preferably, the step of acquiring the current wafer thickness and vacuum level, calculating and generating grinding control commands, acquiring the real-time wafer thickness and vacuum level during the grinding operation, and generating new grinding control commands to gradually thin the wafer to the target thickness while maintaining the adsorption force within the stable adsorption force boundary range throughout the process includes:

[0019] The current thickness and vacuum level of the wafer are obtained through sensors to obtain the first thickness data and the first adsorption force data.

[0020] Based on the first thickness data and the first adsorption force data, the initial pressing parameters are calculated to obtain the first grinding control command;

[0021] The ultra-thinning operation is performed according to the first grinding control command, and the real-time thickness value during the operation is obtained to obtain the second thickness data;

[0022] If the second thickness data is greater than the preset thickness value, the current decrease amount is calculated, and the second grinding control command is determined.

[0023] The real-time vacuum level during the execution of the second grinding control command is obtained by the sensor, and the second adsorption force data is obtained.

[0024] If the second adsorption force data deviates from the preset adsorption force stability boundary range, the vacuum pump speed is adjusted to obtain the third adsorption force data.

[0025] The grinding control signal for the next cycle is generated based on the third adsorption force data and the second thickness data to determine the target processing state.

[0026] Preferably, the grinding process continuously monitors the wafer's stress state to generate an initial stress set, determines the wafer deformation based on the initial stress set, and generates a control instruction set, including:

[0027] The force values ​​on the processing surface of the wafer and the adsorption force on the adsorption surface are obtained by sensors to obtain the initial force set of the adsorption surface.

[0028] Based on the initial force set, a smooth force set is obtained by mean filtering.

[0029] The stress value of the adsorption surface is calculated based on the smooth force set. If the stress value is greater than the preset stress threshold, the outline of the adsorption surface is extracted.

[0030] Based on the contour line, the contour line is processed to determine the deformation at the edge;

[0031] If the deformation exceeds the preset deformation threshold, the amount of cracks at the edge is calculated.

[0032] A set of control instructions is generated based on the amount of cracks at the edge.

[0033] Preferably, the step of performing edge cracking and defect detection on the wafer according to the control instruction set and generating defect rejection items includes:

[0034] According to the control instruction set, the first wafer image of the polished wafer is obtained, and the second wafer image is obtained after grayscale conversion.

[0035] Based on the second wafer image, the first pixel set is obtained by using threshold segmentation to extract the edge region.

[0036] Based on the first pixel set, the second pixel set is obtained through morphological closing operations.

[0037] Based on the second pixel set, a convolutional neural network model is used to identify the cracking point and obtain the first defect value;

[0038] If the first defect value is greater than the preset defect threshold, then the extension line corresponding to the first defect value is extracted to obtain the defect connected region.

[0039] The defect determination value is obtained by calculating the area of ​​the connected component of the defect;

[0040] If the defect judgment value is greater than the preset judgment threshold, the trimming position corresponding to the defect judgment value is marked to obtain the defect removal item.

[0041] Preferably, the step of acquiring the target processing equipment parameters, combining them with defect rejection items, performing grinding and trimming on the wafer, and recording the processing parameters to complete the wafer processing includes:

[0042] Based on the defect elimination items, obtain the equipment number of the target processing equipment and the corresponding parameter values, and generate a processing record table;

[0043] Based on the processing record table, a decision tree algorithm is used for classification to obtain the status code;

[0044] If the status code is greater than the preset status threshold, a processing instruction code is generated;

[0045] Grinding parameters are set using processing instruction codes to obtain the processing task number;

[0046] Perform a grinding operation on the corresponding material according to the processing task number to obtain the trimming and grinding degree;

[0047] Based on the degree of trimming and grinding, the processing results are classified and determined.

[0048] Secondly, the edge-fracture prevention grinding system adapted for ultra-thin wafers described in this application includes:

[0049] Grinding control module: acquires the current thickness and vacuum level of the wafer, calculates and generates grinding control commands, acquires the real-time thickness and vacuum level of the wafer during the grinding operation, and generates new grinding control commands to gradually thin the wafer to the target thickness while maintaining the adsorption force within the stable boundary range of the adsorption force throughout the process.

[0050] Force monitoring module: Continuously monitors the stress state of the wafer during the grinding process, generates an initial force set, determines the wafer deformation based on the initial force set, and generates a control instruction set.

[0051] Defect detection module: Based on the control instruction set, it performs edge cracking and defect detection on the wafer and generates defect rejection items.

[0052] Trimming module: Obtains the parameters of the target processing equipment, combines them with defect rejection items, performs grinding and trimming processing on the wafer, records the processing parameters, and completes the wafer processing.

[0053] Thirdly, this application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the edge-crushing grinding process method for adapting to ultra-thin wafers as described in the first aspect.

[0054] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the edge-crushing grinding process for adapting to ultra-thin wafers as described in the first aspect.

[0055] The beneficial effects of this application are as follows:

[0056] This application acquires real-time parameters and generates new grinding control commands during the grinding process to gradually thin the wafer to the target thickness while maintaining the adsorption force within the stable boundary range throughout the process. The wafer's stress state is continuously monitored during the grinding process to determine wafer deformation, edge cracking and defect detection are performed, and the wafer is ground and trimmed based on the target processing equipment parameters to complete the wafer processing. By implementing the above grinding processing method, this application achieves synergistic control of thickness reduction and stress stability during the grinding process, effectively suppressing the generation and propagation of edge cracks, significantly reducing the edge cracking defect rate, and fully meeting the high-precision, high-yield production requirements of ultra-thin wafers. Attached Figure Description

[0057] Figure 1 This is a flowchart of a grinding process for preventing edge breakage of ultrathin wafers as described in this application.

[0058] Figure 2 This is a schematic diagram of an edge-fragmentation grinding system adapted to ultra-thin wafers as described in this application.

[0059] Figure 3 This is a framework diagram of the terminal device provided in the embodiments of this application. Detailed Implementation

[0060] This application implements an edge-crack prevention grinding method adapted to ultra-thin wafers, which achieves synergistic control of thickness reduction and stress stability during the grinding process, effectively suppresses the generation and extension of edge cracks, significantly reduces the edge crack defect rate, and fully meets the production requirements of high precision and high yield of ultra-thin wafers.

[0061] Example 1

[0062] like Figure 1 As shown, one embodiment of this application provides a method for edge-crack prevention grinding processing adapted to ultra-thin wafers, including:

[0063] S101. Obtain the wafer structure diagram, analyze the coordinate values ​​of the step surface, and set up clamping positions in the area outside the step surface in the wafer structure diagram. Place an adsorption ring at the clamping position and control the vacuum level of the adsorption ring to fix the wafer. Traditional processes typically use large-size vacuum adsorption rings that match the outer diameter of the wafer, with the adsorption area covering the wafer edge. However, for wafers with step structures at the edges, due to uneven thickness distribution, the sudden change in vacuum suction in the edge step area can easily cause severe stress concentration, leading to edge cracks. This step, through position planning, ensures that the adsorption ring avoids the step structure, preventing the step surface from bearing vacuum adsorption pressure, thus eliminating the potential for step pressure at the clamping structure level.

[0064] S102. Obtain the predetermined values ​​of the grinding spindle, and adjust the vacuum level of the adsorption ring based on these values ​​to ensure uniform force on the wafer. After completing wafer fixation and adsorption force adjustment, generate grinding control commands. Obtain predetermined operating parameters such as the rotation speed and steering angle of the grinding spindle, and dynamically adjust the vacuum level of the adsorption ring according to the spindle's operating conditions to compensate for mechanical disturbances caused by rotation, ensuring uniform force on the wafer before grinding starts. This step calibrates the vacuum level based on actual grinding conditions, compensates for rotational mechanical disturbances, and ensures optimal wafer stress before grinding starts.

[0065] S103. Obtain the current wafer thickness and vacuum level, calculate and generate grinding control commands. During the grinding process, acquire the real-time wafer thickness and vacuum level, and generate new grinding control commands to gradually thin the wafer to the target thickness while maintaining the adsorption force within the stable adsorption force boundary range throughout the process. Collect the real-time wafer thickness and vacuum level of the adsorption ring to generate grinding control commands; continuously collect these two real-time parameters during the grinding process, iteratively update the grinding commands, and gradually grind the wafer to the target thickness. Maintain stable adsorption force throughout the grinding process, implementing dual closed-loop dynamic control of wafer thickness and adsorption force. In traditional processes, the vacuum level is generally not dynamically adjusted after being set. As the wafer is continuously thinned, its mechanical stiffness decreases, and even small fluctuations in adsorption force can cause stress impacts at the wafer edges, leading to cracks. This step maintains stable adsorption force control throughout the entire thinning process, dynamically maintaining the adsorption force within a reasonable range as the wafer thickness changes, avoiding dynamic stress caused by sudden changes in adsorption force during the grinding process. The stable adsorption force boundary range is ±5% fluctuation range based on the target adsorption force, gradually decreasing as the wafer thickness decreases.

[0066] S104. During the grinding process, the wafer's stress state is continuously monitored to generate an initial stress set. Based on this initial stress set, wafer deformation is determined, and a control command set is generated. The grinding process continuously collects stress data on the wafer's adsorption surface to obtain the initial stress set. This stress data is processed to calculate stress and edge deformation, identify risk areas with excessive stress and abnormal deformation, and output the control command set. This step further collects stress data, identifies early wafer deformation risks, identifies high-risk edge areas, and outputs the control command set to provide location guidance for subsequent defect detection.

[0067] S105. Based on the control instruction set, perform edge crack and defect detection on the wafer, generating defect rejection items. Receive the control instruction set output from S104, perform image defect detection on high-risk edge areas indicated by stress deformation, identify edge cracks, mark crack locations and extents, and generate defect rejection items containing trimming location information. This step relies on the risk guidance from the previous stress monitoring, focusing on high-risk edge areas for defect identification, accurately detecting early micro-cracks, and outputting trimming location information.

[0068] S106. Obtain the target processing equipment parameters, combine them with defect removal criteria, perform grinding and trimming on the wafer, and record the processing parameters to complete the wafer processing. For wafers with edge cracks, if no treatment is performed, the cracks will continue to propagate inwards and the wafer will be scrapped. This step involves directional trimming based on the defect location results to remove cracks and crack propagation channels; at the same time, process parameters are recorded to facilitate process iteration and traceability.

[0069] This embodiment presents an edge-crack prevention grinding method for ultra-thin wafers. Addressing the problem of edge crack initiation and inward propagation leading to wafer scrap in traditional ultra-thin wafer grinding with steps, this method couples steps S101-S106. S101 sets up a clamping position to avoid the step surface; S102 calibrates the vacuum level based on the grinding spindle's operating conditions, eliminating the root cause of step stress concentration; S103 uses a thickness-adhesion dual closed loop to maintain stable adhesion throughout the wafer's gradual ultra-thinning process, avoiding dynamic stress impacts during grinding; S104 directly collects stress deformation data to identify early risks, guiding S105 to perform defect detection on high-risk edge areas and accurately locate edge cracks; finally, S106 uses defect information to directionally trim and remove cracked areas. This entire solution integrates prevention, process monitoring, and post-process defect repair, effectively suppressing the generation and propagation of edge cracks in ultra-thin wafers, significantly reducing the edge crack defect rate, and meeting the high-precision, high-yield processing requirements of ultra-thin wafers. Compared to traditional methods that use auxiliary processes such as laser trimming to treat edge defects, this method does not require additional equipment investment. It simultaneously grinds and removes potential cracked areas during the grinding stage, making it suitable for the mass production needs of large-volume ultra-thin wafers.

[0070] Furthermore, S101 includes:

[0071] Obtain the structure diagram of the wafer and extract the edge region image of the wafer from the structure diagram.

[0072] The edge region image is analyzed to obtain the coordinate values ​​of the step surface in the edge region image.

[0073] The avoidance area is determined based on the coordinate values ​​of the step surface, and the area outside the avoidance area is defined as the adsorption area in the structural diagram.

[0074] Obtain the dimensions of the adsorption area and determine the clamping position based on the dimensions.

[0075] An adsorption ring is set at the clamping position, and an image of the adsorption surface between the adsorption ring and the wafer is obtained.

[0076] If a step surface exists in the adsorption surface image, the clamping position is redefined; if no step surface exists in the adsorption surface image, the positioning is complete.

[0077] After positioning is completed, the vacuum level of the adsorption ring is controlled to fix the wafer.

[0078] By implementing this step, a wafer structure diagram can be obtained through a high-resolution camera and the coordinates of the step surface can be accurately extracted. The clamping position of the adsorption ring can be set in an area outside the step surface to avoid the adsorption ring contacting the step structure and eliminate the structural risk of edge cracks caused by stress concentration at the step during the grinding process.

[0079] In a specific embodiment, a structural diagram of the wafer is obtained, and an image of the edge region of the wafer is extracted from the structural diagram.

[0080] An edge detection algorithm is used to analyze the edge region image to obtain the coordinate values ​​of the step surface in the edge region image.

[0081] The avoidance area is determined based on the coordinate values ​​of the step surface, and the area outside the avoidance area is defined as the adsorption area in the structural diagram.

[0082] Obtain the dimensions of the adsorption area and determine the clamping position based on the dimensions.

[0083] An adsorption ring is set at the clamping position, and an image of the adsorption surface between the adsorption ring and the wafer is obtained.

[0084] If a step surface exists in the adsorption surface image, the clamping position is redefined; if no step surface exists in the adsorption surface image, the positioning is complete.

[0085] After positioning is completed, the vacuum level of the adsorption ring is controlled to fix the wafer.

[0086] Specifically, in S101, after receiving the wafer clamping instruction, the system immediately starts the wafer structure analysis process. It acquires the wafer structure diagram through a high-resolution line scan camera, calls the Canny edge detection operator to extract the edge region information in the structure diagram, and obtains a local image of the wafer edge region. It performs sub-pixel level edge localization on the edge region image, extracts the boundary coordinate set of the step surface, and converts the image coordinates into the actual position coordinates in the world coordinate system through the camera calibration parameters, thereby obtaining the precise coordinate values ​​of the step surface in the wafer coordinate system.

[0087] The system marks the identified step positions on the structural diagram based on the step surface coordinates. After extending this step position 2mm towards the wafer center, it defines the avoidance area, and the area outside the avoidance area is defined as the effective adsorption area. The system acquires the dimensions of the effective adsorption area and determines the clamping coordinates of the adsorption ring based on these dimensions, ensuring that the adsorption surface of the adsorption ring completely falls within the effective adsorption area. The system controls a robotic arm to move the wafer onto the adsorption ring for pre-positioning. After pre-positioning, a contact image sensor acquires an image of the adsorption surface between the adsorption ring and the lower surface of the wafer. The system performs binarization processing on the adsorption surface image. The system uses a template matching algorithm to identify the edge contour features of a step structure within the contact area. During template matching, the system uses the step surface image captured from the structural diagram as the template image and the adsorption surface image as the search image. If a step surface feature is identified in the adsorption surface image, the clamping position is determined to be off. The system recalculates the clamping position coordinates based on the relative offset between the step surface and the edge of the adsorption ring and controls the robot arm to make fine adjustments to the position. The adsorption surface image is then re-acquired for secondary verification until no step surface feature is found in the adsorption surface image. If no step surface feature is identified in the adsorption surface image, the current clamping position is determined to be valid, and the positioning is completed.

[0088] After positioning is completed, the system starts the vacuum generator to apply vacuum adsorption force to the adsorption ring. The system gradually adjusts the vacuum level through a closed-loop PID control algorithm until a stable airtight contact is formed between the adsorption ring and the wafer, thus completing the wafer fixing operation.

[0089] Specifically, in S101, the template matching algorithm is as follows:

[0090] ;

[0091] in, The normalized cross-correlation matching coefficient is dimensionless.

[0092] The image of the stepped surface extracted from the structural diagram at the pixel point The grayscale value at that location;

[0093] This represents the average grayscale value of all pixels in the step surface image.

[0094] The corresponding region in the adsorption surface image at the pixel point The grayscale value at that location;

[0095] This represents the average grayscale value of the corresponding region in the adsorption surface image;

[0096] For the adsorption surface image in The offset in direction;

[0097] For the adsorption surface image in The offset in direction.

[0098] Furthermore, S102 includes:

[0099] Obtain the rotation speed and steering angle of the grinding spindle, as well as the grinding surface information of the wafer, from the predetermined values ​​to determine the target adsorption force.

[0100] Based on the target adsorption force, feature extraction is performed to obtain the adsorption force feature vector.

[0101] The adsorption force feature vector is compared with a preset threshold. If it is greater than the preset threshold, the first force value is determined.

[0102] Based on the first force value, a second force value is obtained through smoothing filtering.

[0103] Based on the second force value, combined with the rotation speed value and steering angle, the vacuum degree of the adsorption ring is updated.

[0104] Implementing this step allows for the establishment of a target adsorption force benchmark based on the predetermined value of the grinding spindle and the information of the grinding surface. The uniformity of the adsorption force distribution is intelligently determined and calibrated using a random forest algorithm, ensuring that the wafer is in a uniform stress state before grinding, thus avoiding localized stress anomalies caused by uneven initial stress.

[0105] In a specific embodiment, the rotational speed and steering angle of the grinding spindle and the grinding surface information of the wafer are obtained from the predetermined values ​​to determine the target adsorption force that needs to be applied to the wafer under the current working conditions.

[0106] Based on the target adsorption force, the adsorption surface image, clamping position coordinates, and vacuum degree setting value are constructed into a multi-dimensional feature vector. The random forest algorithm is used to extract features from the multi-dimensional feature vector to obtain the adsorption force feature vector.

[0107] The adsorption force feature vectors are weighted and aggregated to obtain the adsorption force feature value. If the adsorption force feature value is greater than the preset adsorption force safety threshold, the first force value is determined. If the adsorption force feature value is greater than the adsorption force safety threshold, it is determined that the current adsorption force distribution is uneven. The feature importance ranking of the random forest algorithm is reversed to locate the region with the largest pressure deviation on the adsorption surface, and the pressure deviation of this region is determined as the first force value.

[0108] Based on the first force value, a smoothing filter is applied to the first force value to obtain the second force value.

[0109] Based on the second force value, combined with the rotation speed and steering angle, the adsorption force compensation coefficient under the current working condition is calculated, and the vacuum degree of the control adsorption ring is updated according to the adsorption force compensation coefficient.

[0110] Specifically, in S102, after receiving the adsorption force calibration command, the system obtains the predetermined rotation speed and turning angle of the grinding spindle of the current production line, as well as the grinding surface information of the wafer. Combined with the clamping position coordinates and vacuum degree determined in S101, the system calculates the target adsorption force required to theoretically ensure that the wafer does not shift under the current working conditions through a mechanical model.

[0111] The system uses the target adsorption force as a reference baseline and constructs a multi-dimensional feature vector from the clamping position coordinates and vacuum degree of S101. The system uses a random forest algorithm to perform nonlinear mapping and dimensionality reduction aggregation on the multi-dimensional feature vector to extract features and obtain an adsorption force feature vector. The feature vector is a set of multi-dimensional results output by each decision tree in the random forest. The values ​​of each dimension reflect the degree of force dispersion of different adsorption regions (edge ​​region, center region, transition region) under the current feature combination.

[0112] The system obtains adsorption force feature values ​​by weighted aggregation of the adsorption force feature vectors. These feature values ​​are then compared with a preset adsorption force safety threshold, assuming the threshold is 0.35. If the adsorption force feature value is greater than 0.35, the current adsorption force distribution is determined to be uneven. The system then uses a reverse analysis of the random forest algorithm to rank the features by importance, locates the region with the largest pressure deviation on the adsorption surface, and determines the pressure deviation in that region as the first force value. If the adsorption force feature value is less than or equal to 0.35, the current adsorption force meets the requirements, the force distribution is uniform, and no calibration is required.

[0113] Based on the first force value, the system uses a moving average filtering algorithm to smooth it, filtering out high-frequency noise components introduced by vacuum pump pulsation or mechanical vibration, to obtain a stable and smooth second force value. Based on the second force value, combined with the current spindle speed and steering angle, the system calculates the total additional disturbance of centrifugal force and tangential force on the adsorption force under the current working condition, and converts the total additional disturbance into a global adsorption force compensation coefficient. Based on the global adsorption force compensation coefficient, the system generates a vacuum adjustment command to update the vacuum setting value of the adsorption coil.

[0114] Specifically, in S102, the random forest algorithm:

[0115] ;

[0116] in, This is the dimensionless feature vector of the adsorption force output by the random forest algorithm.

[0117] The number of decision trees in the random forest;

[0118] For the first Each decision tree is used to process the input feature vector. The predicted output value;

[0119] Let be the input multidimensional feature vector, where For the clamping position coordinates, The vacuum level set for S101, The target adsorption force.

[0120] Weighted aggregation formula:

[0121] ;

[0122] in, The characteristic value of the adsorption force is dimensionless;

[0123] The number of adsorption regions;

[0124] For the first The weight coefficients corresponding to each region;

[0125] Adsorption force eigenvector The Middle Each component value.

[0126] Furthermore, S103 includes:

[0127] The current thickness and vacuum level of the wafer are obtained by sensors, thus obtaining the first thickness data and the first adsorption force data.

[0128] Based on the first thickness data and the first adsorption force data, the initial pressure parameters are calculated to obtain the first grinding control command.

[0129] The ultra-thinning operation is performed according to the first grinding control command, and the real-time thickness value during the operation is obtained to obtain the second thickness data.

[0130] If the second thickness data is greater than the preset thickness value, the current decrease amount is calculated, and the second grinding control command is determined.

[0131] The real-time vacuum level during the execution of the second grinding control command is obtained by the sensor, and the second adsorption force data is obtained.

[0132] If the second adsorption force data deviates from the preset adsorption force stability boundary range, the vacuum pump speed is adjusted to obtain the third adsorption force data.

[0133] The grinding control signal for the next cycle is generated based on the third adsorption force data and the second thickness data to determine the target processing state.

[0134] Implementing this step enables real-time monitoring of wafer thickness changes and dynamic adjustment of pressure parameters. Combined with automatic adjustment of vacuum pump speed within the stable adsorption force boundary range, this achieves dual closed-loop control of thickness reduction and adsorption force stability throughout the entire process of gradually thinning the wafer to the target thickness, ensuring no sudden changes in suction force during the grinding process.

[0135] In a specific embodiment, the current thickness value and vacuum level of the wafer are obtained through sensors to obtain first thickness data and first adsorption force data.

[0136] Based on the first thickness data and the first adsorption force data, the initial downward pressure parameters of the grinding spindle are calculated, and the first grinding control command is obtained.

[0137] The grinding spindle is activated according to the first grinding control command to perform ultra-thinning operation on the machined surface, and the real-time thickness value during the operation is obtained to obtain the second thickness data.

[0138] If the second thickness data is greater than the preset thickness value, the current decrease amount is calculated based on the second thickness data, and the second grinding control command is determined.

[0139] The real-time vacuum level during the execution of the second grinding control command is obtained by the sensor, and the second adsorption force data is obtained.

[0140] If the second adsorption force data deviates from the preset adsorption force stability boundary range corresponding to the constant state, the vacuum pump speed is adjusted to obtain the third adsorption force data.

[0141] The grinding control signal for the next cycle is generated based on the third adsorption force data and the second thickness data to determine the target processing state.

[0142] Specifically, in S103, after receiving the grinding start command, the system obtains the current initial thickness value of the wafer through the laser displacement sensor, and at the same time obtains the current vacuum level of the adsorption ring through the vacuum pressure sensor, thus obtaining the first thickness data and the first adsorption force data. The system compares the first thickness data with the preset target thickness value, and calculates the initial pressing parameters of the grinding spindle in combination with the first adsorption force data, thereby generating the first grinding control command. The first grinding control command is used to control the grinding spindle to perform the initial pressing action, and to perform ultra-thinning processing on the back side of the wafer with the set initial grinding force.

[0143] During the grinding process, the system collects the wafer thickness value in real time through a laser displacement sensor to obtain second thickness data. If the second thickness data is greater than the preset thickness value, the system calculates the total amount of material removed and the remaining thickness based on the second thickness data, further calculates the thickness reduction in the current grinding cycle, dynamically adjusts the pressure parameter based on the reduction, and determines the second grinding control command. The second grinding control command is used to control the grinding spindle to continue the processing operation with the adjusted pressure parameter in subsequent grinding cycles.

[0144] During the execution of the second grinding control command, the system collects the vacuum level of the adsorption ring in real time through a vacuum pressure sensor to obtain the second adsorption force data. Based on the processing conditions where the grinding equipment and vacuum pump work together, the wafer's stress state is continuously stable, and the thickness decreases uniformly, a corresponding stable adsorption force boundary range is preset. The stable adsorption force boundary range is a fluctuation range of ±5% based on the target adsorption force. The stable adsorption force boundary range gradually decreases as the wafer thickness decreases. If the second adsorption force data deviates from the stable adsorption force boundary range, it is determined that the current adsorption force deviates from the stable state. The system calculates the vacuum pump speed adjustment amount based on the difference between the measured vacuum level and the stable adsorption force boundary range, and adjusts the vacuum pump speed through the frequency converter to restore the adsorption force to the stable adsorption force boundary range, thus obtaining the third adsorption force data. If the second adsorption force data does not deviate from the stable adsorption force boundary range, it is determined that the current adsorption force state is stable, and there is no need to adjust the vacuum pump speed; the second adsorption force data is directly used as the third adsorption force data.

[0145] The system generates a grinding control signal for the next grinding cycle based on the third adsorption force data and the second thickness data. The system outputs the grinding control signal to the grinding spindle and vacuum pump to determine the target processing state and continuously executes the above-mentioned thickness monitoring and adsorption force monitoring process until the wafer thickness reaches the target thickness value, thus completing the ultra-thin grinding process.

[0146] Furthermore, S104 includes:

[0147] By acquiring the force values ​​on the processing surface of the wafer and the adsorption force on the adsorption surface using sensors, the initial force set of the adsorption surface is obtained.

[0148] Based on the initial force set, a smooth force set is obtained by mean filtering.

[0149] The stress value of the adsorption surface is calculated based on the smooth force set. If the stress value is greater than the preset stress threshold, the contour line of the adsorption surface is extracted.

[0150] Based on the contour line, the contour line is processed to determine the deformation at the edge.

[0151] If the deformation exceeds the preset deformation threshold, the amount of cracks at the edge is calculated.

[0152] A set of control instructions is generated based on the amount of cracks at the edge.

[0153] Implementing this step enables continuous monitoring of the stress distribution on the adsorption surface and allows for stress analysis and deformation detection. It also enables real-time monitoring of stress concentration points and deformation anomalies, generating a set of control commands to provide precise guidance for subsequent accurate detection.

[0154] In a specific embodiment, the force value of the processing surface of the wafer and the adsorption force of the adsorption surface are obtained by the sensor to obtain the initial force set of the adsorption surface.

[0155] Based on the initial force set of the adsorption surface, a smooth force set is obtained by mean filtering.

[0156] For a smooth set of forces, calculate the stress value of the adsorption surface. If the stress value is greater than the preset stress threshold, extract the contour line of the adsorption surface.

[0157] Obtain the contour line of the adsorption surface, process the contour line, and determine the deformation at the edge.

[0158] If the deformation at the edge is greater than the preset deformation threshold, the amount of crack at the edge is calculated.

[0159] A set of control instructions is generated based on the amount of cracks at the edge.

[0160] Specifically, in S104, after receiving the force monitoring command, the system collects the force value of the wafer processing surface and the adsorption force value of the adsorption ring in real time through a distributed pressure sensor array to obtain the initial force set of the adsorption surface; the initial force set includes the normal force value and tangential force value at different positions of the processing surface, as well as the adsorption force distribution data of the adsorption ring.

[0161] The system applies mean filtering to the initial force set, setting the filter window size to 5×5 to filter out high-frequency interference components introduced by grinding vibration or sensor noise, thus obtaining a smooth force set. Based on this smooth force set, stress analysis is performed to obtain the stress distribution on the adsorption surface. If the stress concentration point is located at the edge of the adsorption surface and its stress value is greater than a preset stress threshold, the contour line of the adsorption surface is extracted. If the stress concentration point is located in a non-edge region and the stress value exceeds the threshold, it is determined to be an abnormal condition, generating a global alarm signal and triggering an adsorption force adjustment command. If the stress value is less than or equal to the stress threshold, the current stress level is determined to be within a safe range, and contour line extraction is unnecessary; the next cycle of force monitoring continues.

[0162] The system performs geometric feature analysis on the extracted edge contour line of the adsorption surface, calculates the displacement of each point on the contour line relative to the initial position, and determines the deformation at the wafer edge step surface. The deformation is compared with a preset deformation threshold. If the deformation is 5 μm, the step surface deformation is determined to be outside the safe range, posing a risk of crack initiation. The system calculates the crack quantity at the step surface edge based on the magnitude and distribution of the deformation. If the deformation is less than or equal to 5 μm, the deformation is determined to be within the safe range, and no crack quantity calculation is required; the system continues with the next cycle of deformation monitoring. The system generates a control command set based on the calculated crack quantity.

[0163] Specifically, in S104, the mean filtering formula is:

[0164] ;

[0165] in, Filtered pixels Smooth force value at the location, unit: N;

[0166] This represents the original force value at the corresponding location within the filter window, in N.

[0167] This is the size of the filter window. and All numbers are odd to ensure window center symmetry;

[0168] for These are the row and column indices in the data matrix, corresponding to the circumferential and radial positions of the adsorption surface, respectively.

[0169] , Within the window, relative to the center point The row offset and column offset.

[0170] Furthermore, S105 includes:

[0171] According to the control instruction set, the first wafer image of the polished wafer is obtained, and the second wafer image is obtained after grayscale conversion.

[0172] Based on the second wafer image, the edge region is extracted using threshold segmentation to obtain the first pixel set.

[0173] The second pixel set is obtained by morphological closing operations based on the first pixel set.

[0174] Based on the second pixel set, a convolutional neural network model is used to identify the crack points and obtain the first defect value.

[0175] If the first defect value is greater than the preset defect threshold, then the extension line corresponding to the first defect value is extracted to obtain the defect connected region.

[0176] The defect determination value is obtained by calculating the area of ​​the connected region of the defect.

[0177] If the defect judgment value is greater than the preset judgment threshold, the trimming position corresponding to the defect judgment value is marked to obtain the defect removal item.

[0178] Implementing this step enables high-precision imaging of the wafer edge region based on the control instruction set, intelligent identification of crack points by combining adaptive threshold segmentation and convolutional neural network model, and quantitative determination of defect severity based on the area of ​​connected regions, thereby achieving automated and accurate inspection of edge cracks and extended defects.

[0179] In a specific embodiment, a first wafer image of the polished wafer is obtained according to a set of control instructions, and a second wafer image is obtained after grayscale conversion of the first wafer image.

[0180] Based on the second wafer image, the edge regions of the second wafer image are extracted using threshold segmentation to obtain the first pixel set.

[0181] The second pixel set is obtained by processing the first pixel set through morphological closing operations.

[0182] Based on the second pixel set, a convolutional neural network model is used to identify the crack points in the second pixel set to obtain the first defect value.

[0183] If the first defect value is greater than the preset defect threshold, then the extension line corresponding to the first defect value is extracted to obtain the defect connected region.

[0184] The defect determination value is obtained by calculating the area of ​​the connected region of the defect.

[0185] If the defect judgment value is greater than the preset judgment threshold, the trimming position corresponding to the defect judgment value is marked to obtain the defect removal item.

[0186] Specifically, in S105, after receiving the defect investigation instruction, the system performs high-precision imaging of the edge area of ​​the polished wafer using a high-resolution industrial camera according to the control instruction set to generate a first wafer image; the first wafer image is then subjected to grayscale conversion processing to convert the RGB three-channel color image into a single-channel grayscale image to obtain a second wafer image.

[0187] Based on the second wafer image, the system uses an adaptive threshold segmentation algorithm to extract the wafer edge region. The adaptive threshold is dynamically calculated based on the average gray value and local variance of the edge region to separate the edge region from the background and obtain the first pixel set.

[0188] The system performs morphological closing operations on the first pixel set. The closing operation is a combination of dilation and erosion, with a structuring element size of 5×5 pixels. It fills the holes and gaps caused by small fractures in the edge area and connects the broken crack edges to obtain the second pixel set.

[0189] Based on the second pixel set, the system calls a convolutional neural network model to identify crack points. The convolutional neural network model is a ResNet-50 structure, with an input image size of 224×224 pixels and an output crack confidence score to obtain a first defect value. If the first defect value is greater than a preset defect threshold, it is determined that there is an actual crack at that location. The system extracts the crack extension line corresponding to the defect value and determines the continuous pixel area covered by the extension line as the defect connected region. If the first defect value is less than or equal to the defect threshold, it is determined that there is no actual crack, no further processing is required, and the output detection result is "no defect".

[0190] Based on the defective connected component, the system counts the number of pixels covered by the connected component, and converts the pixel area into the actual physical area by combining the image pixel size to obtain the defect judgment value. If the defect judgment value is greater than the preset judgment threshold, the system marks the trimming position parameter corresponding to the crack and generates a defect rejection item.

[0191] Specifically, in S105, the adaptive threshold segmentation algorithm is as follows:

[0192] Adaptive threshold calculation:

[0193] ;

[0194] in, For pixels The adaptive segmentation threshold at the point is dimensionless.

[0195] This is the average grayscale value of pixels within a local window, and is dimensionless.

[0196] is the standard deviation of pixel grayscale values ​​within a local window, and is dimensionless.

[0197] The sensitivity coefficient is dimensionless.

[0198] Binarization segmentation:

[0199] ;

[0200] in, For binarized pixels The grayscale value is dimensionless, with 255 representing the edge region and 0 representing the background;

[0201] For pixels in the second wafer image The original grayscale value is dimensionless;

[0202] For pixels The adaptive segmentation threshold at the location.

[0203] Furthermore, S106 includes:

[0204] Based on the defect rejection criteria, the equipment number of the target processing equipment and the corresponding parameter values ​​are obtained, and a processing record table is generated.

[0205] Based on the processing record table, a decision tree algorithm is used for classification to obtain the status code.

[0206] If the status code is greater than the preset status threshold, a processing instruction code is generated.

[0207] The grinding parameters are set by the processing instruction code to obtain the processing task number.

[0208] The grinding operation is performed on the corresponding material according to the processing task number to obtain the trimming and grinding degree.

[0209] Based on the degree of trimming and grinding, the processing results are classified and determined.

[0210] This step involves using a decision tree algorithm to fuse and classify defect information with equipment parameters, automatically generating processing instruction codes and scheduling the target processing equipment to perform trimming and grinding. The classification is based on the degree of grinding after trimming and the standard thickness of the finished product, forming a complete closed loop from defect detection to processing and result archiving.

[0211] In a specific embodiment, based on the defect rejection item, the equipment number of the target processing equipment and the parameter value corresponding to the equipment number are obtained, and a processing record table is generated.

[0212] Based on the processing record table, a decision tree algorithm is used for classification to obtain the status code.

[0213] If the status code is greater than the preset status threshold, a processing instruction code is generated based on the status code.

[0214] The grinding parameters are set by the processing instruction code to obtain the processing task number.

[0215] The grinding operation is performed on the corresponding material according to the processing task number to obtain the trimming and grinding degree.

[0216] Based on the degree of trimming and grinding, the processing results are classified and determined.

[0217] Specifically, in S106, after receiving the trimming instruction, the system obtains the equipment number of the target processing equipment and the current parameter value corresponding to the equipment number based on the trimming position parameter and defect area in the defect rejection item, binds the defect rejection item with the equipment information, and generates a processing record table.

[0218] Based on the processing record table, the system uses a decision tree algorithm for classification. The decision tree is a CART classification tree. The input features include defect area, defect location, cumulative equipment runtime, and current equipment status. The output is a status code.

[0219] The system compares the status code with a preset status threshold; if the status code is greater than the status threshold, it determines that the current material needs to be trimmed. The system generates a corresponding processing instruction code based on the status code, which includes operation parameters such as trimming depth, trimming angle, and target removal thickness.

[0220] The system sets the grinding parameters of the grinding spindle through the processing instruction code, determines the processing start position according to the trimming position parameters, and generates a unique processing task number for the current material.

[0221] The system performs trimming and grinding operations on the corresponding material according to the processing task number to remove edge crack areas. After trimming, the surface quality data and thickness data of the trimmed surface are obtained through a laser displacement sensor to obtain the trimming and grinding degree. The trimming and grinding degree includes the measured surface roughness value and the measured thickness value.

[0222] The system compares the trimming and grinding degree with the preset acceptance standard. If the measured surface roughness is less than or equal to 1.0 μm and the measured thickness is within ±5 μm of the finished product standard thickness, the material is deemed to be trimmed and marked as "acceptable". If the measured roughness is greater than 1.0 μm or the measured thickness exceeds 5 μm of the finished product standard thickness, it is deemed to be "rework required" and re-enters the trimming process. If the measured thickness is less than -5 μm of the finished product standard thickness, it is deemed to be "unacceptable". The final processing result set is determined. The processing result set includes material identification, measured values ​​of various processing indicators, and the final judgment conclusion, which are archived and traced by the production line quality management system.

[0223] Specifically, in S106, the decision tree algorithm includes two stages: offline training and online inference.

[0224] Offline training phase: The system uses historical processed data as training samples and recursively selects the optimal splitting features and splitting thresholds based on minimizing the mean squared error to construct a decision tree model. The node splitting criteria are as follows:

[0225] ;

[0226] in, For feature index;

[0227] This is the splitting threshold;

[0228] and These are the feature space regions corresponding to the left and right child nodes after the split, respectively;

[0229] and These are the output values ​​for the left and right child nodes, respectively.

[0230] For the first The input feature vector of each training sample;

[0231] For the first The true state code of each training sample is dimensionless.

[0232] Online inference phase: After training, the system inputs the input feature vector from the current material processing record table into the decision tree model, and routes downwards layer by layer from the root node. At each node, it determines whether to enter the left or right child node based on the split threshold, until the leaf node is reached; the split threshold determination is as follows:

[0233] ;

[0234] in, This is the index of the current node;

[0235] The optimal splitting feature index selected for the current node;

[0236] The optimal split threshold for the current node;

[0237] For feature vectors The Middle The specific values ​​that each feature can take.

[0238] The predicted state code stored in the leaf nodes is the state code itself, and the decision function is as follows:

[0239] ;

[0240] in, Status codes are dimensionless.

[0241] The input feature vector;

[0242] To determine the total number of tree nodes in the decision tree;

[0243] For the first The output values ​​of the leaf nodes;

[0244] For the first The feature space partitioning region corresponding to each leaf node;

[0245] For indicator functions, when Belongs to the region The value is 1 if the condition is met, and 0 otherwise.

[0246] Example 2

[0247] like Figure 2 As shown, an edge-fracture prevention grinding system adapted for ultra-thin wafers includes:

[0248] Wafer positioning module: acquires the wafer structure diagram, parses the coordinate values ​​of the step surface, sets the clamping position in the area outside the step surface in the wafer structure diagram, sets the adsorption ring in the clamping position, and controls the vacuum degree of the adsorption ring to complete the wafer fixation.

[0249] Adsorption calibration module: acquires the predetermined value of the grinding spindle, adjusts the vacuum degree of the adsorption ring based on the predetermined value of the grinding spindle to make the wafer subjected to uniform force, and generates grinding control commands after completing wafer fixation and adsorption force adjustment.

[0250] Grinding control module: acquires the current thickness and vacuum level of the wafer, calculates and generates grinding control commands, acquires the real-time thickness and vacuum level of the wafer during the grinding operation, and generates new grinding control commands to gradually thin the wafer to the target thickness while maintaining the adsorption force within the stable boundary range of the adsorption force throughout the process.

[0251] Force monitoring module: Continuously monitors the stress state of the wafer during the grinding process, generates an initial force set, determines the wafer deformation based on the initial force set, and generates a control instruction set.

[0252] Defect detection module: Based on the control instruction set, it performs edge cracking and defect detection on the wafer and generates defect rejection items.

[0253] Trimming module: Obtains the parameters of the target processing equipment, combines them with the defect rejection items, performs grinding and trimming processing on the wafer, records the processing parameters, and completes the wafer processing.

[0254] Example 3

[0255] like Figure 3 As shown, this embodiment provides a terminal device 30, including a processor 300 and a memory 301.

[0256] The memory 301 is used to store the computer program 302 and to transfer the computer program 302 to the processor 300.

[0257] The processor 300 is used to execute the steps in the above-described edge-fracture-resistant grinding process for adapting ultra-thin wafers according to the instructions in the computer program 302.

[0258] For example, computer program 302 may be divided into one or more modules / units; one or more modules / units are stored in memory 301 and executed by processor 300 to complete this application; one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of computer program 302 in terminal device 30.

[0259] Terminal device 30 may be a desktop computer, laptop, handheld computer, cloud server, or other computing device; terminal device 30 may include, but is not limited to, processor 300 and memory 301.

[0260] Those skilled in the art will understand that Figure 3 This is merely an example of terminal device 30 and does not constitute a limitation on terminal device 30. It may include more or fewer components than shown, or combine certain components, or different components. For example, terminal device 30 may also include input / output devices, network access devices, buses, etc.

[0261] The processor 300 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0262] The memory 301 can be an internal storage unit of the terminal device 30, such as a hard disk or memory of the terminal device 30; the memory 301 can also be an external storage device of the terminal device 30, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the terminal device 30.

[0263] Furthermore, the memory 301 may include both internal storage units and external storage devices of the terminal device 30; the memory 301 is used to store the computer program 302 and other programs and data required by the terminal device 30, and the memory 301 may also be used to temporarily store data that has been output or will be output.

[0264] Example 4

[0265] This embodiment provides a computer-readable storage medium storing a computer program; when the computer program is executed by a processor, it implements all the steps of the edge-crushing grinding process for adapting ultrathin wafers as described in Embodiment 1.

[0266] Computer-readable storage media include, but are not limited to: ROM, RAM, disk, optical disc, USB flash drive, solid-state drive, cloud storage media; the program can be run by being loaded into any electronic device.

[0267] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this application.

Claims

1. A method for edge-fracture prevention grinding processing adapted to ultra-thin wafers, characterized in that, include: The current thickness and vacuum level of the wafer are obtained, and grinding control commands are calculated and generated. During the grinding operation, the real-time thickness and vacuum level of the wafer are obtained, and new grinding control commands are generated to gradually thin the wafer to the target thickness while keeping the adsorption force within the stable boundary range of the adsorption force throughout the process. The grinding process continuously monitors the stress state of the wafer to generate an initial stress set, determines the wafer deformation based on the initial stress set, and generates a control instruction set. Based on the control instruction set, edge cracking and defect detection are performed on the wafer, and defect rejection items are generated. Obtain the target processing equipment parameters, combine them with defect rejection criteria, perform grinding and trimming on the wafer, record the processing parameters, and complete the wafer processing.

2. The edge-fracture-resistant grinding process for ultra-thin wafers according to claim 1, characterized in that, The process of acquiring the current wafer thickness and vacuum level, calculating and generating grinding control commands, acquiring the real-time wafer thickness and vacuum level during the grinding operation, and generating new grinding control commands to gradually thin the wafer to the target thickness while maintaining the adsorption force within the stable adsorption force boundary range throughout the process, also includes: Obtain the wafer structure diagram, analyze the coordinate values ​​of the step surface, set the clamping position in the area outside the step surface in the wafer structure diagram, set the adsorption ring in the clamping position, and control the vacuum degree of the adsorption ring to complete the wafer fixation. The predetermined value of the grinding spindle is obtained, and the vacuum degree of the adsorption ring is adjusted based on the predetermined value of the grinding spindle to make the wafer subjected to uniform force. After the wafer is fixed and the adsorption force is adjusted, the grinding control command is generated.

3. The edge-fracture-resistant grinding process for ultra-thin wafers according to claim 2, characterized in that, The process of obtaining a predetermined value for the grinding spindle and adjusting the vacuum level of the adsorption ring based on that predetermined value to ensure uniform force on the wafer includes: Obtain the rotation speed and turning angle of the grinding spindle and the grinding surface information of the wafer from the predetermined values ​​to determine the target adsorption force; Based on the target adsorption force, feature extraction is performed to obtain the adsorption force feature vector; The adsorption force feature vector is compared with a preset threshold. If it is greater than the preset threshold, the first force value is determined. Based on the first force value, a second force value is obtained through smoothing filtering. Based on the second force value, combined with the rotation speed value and steering angle, the vacuum degree of the adsorption ring is updated.

4. The edge-fracture prevention grinding method for ultra-thin wafers according to claim 1, characterized in that, The process of acquiring the current wafer thickness and vacuum level, calculating and generating grinding control commands, acquiring the real-time wafer thickness and vacuum level during the grinding operation, and generating new grinding control commands to gradually thin the wafer to the target thickness while maintaining the adsorption force within the stable adsorption force boundary range throughout the process includes: The current thickness and vacuum level of the wafer are obtained through sensors to obtain the first thickness data and the first adsorption force data. Based on the first thickness data and the first adsorption force data, the initial pressing parameters are calculated to obtain the first grinding control command; The ultra-thinning operation is performed according to the first grinding control command, and the real-time thickness value during the operation is obtained to obtain the second thickness data; If the second thickness data is greater than the preset thickness value, the current decrease amount is calculated, and the second grinding control command is determined. The real-time vacuum level during the execution of the second grinding control command is obtained by the sensor, and the second adsorption force data is obtained. If the second adsorption force data deviates from the preset adsorption force stability boundary range, the vacuum pump speed is adjusted to obtain the third adsorption force data. The grinding control signal for the next cycle is generated based on the third adsorption force data and the second thickness data to determine the target processing state.

5. The edge-fracture prevention grinding method for ultra-thin wafers according to claim 1, characterized in that, The grinding process continuously monitors the wafer's stress state to generate an initial stress set, determines the wafer deformation based on the initial stress set, and generates a control command set, including: The force values ​​on the processing surface of the wafer and the adsorption force on the adsorption surface are obtained by sensors to obtain the initial force set of the adsorption surface. Based on the initial force set, a smooth force set is obtained by mean filtering. The stress value of the adsorption surface is calculated based on the smooth force set. If the stress value is greater than the preset stress threshold, the outline of the adsorption surface is extracted. Based on the contour line, the contour line is processed to determine the deformation at the edge; If the deformation exceeds the preset deformation threshold, the amount of cracks at the edge is calculated. A set of control instructions is generated based on the amount of cracks at the edge.

6. The edge-fracture-resistant grinding method for ultra-thin wafers according to claim 1, characterized in that, The step of performing edge crack and defect detection on the wafer according to the control instruction set and generating defect rejection items includes: According to the control instruction set, the first wafer image of the polished wafer is obtained, and the second wafer image is obtained after grayscale conversion. Based on the second wafer image, the first pixel set is obtained by using threshold segmentation to extract the edge region. Based on the first pixel set, the second pixel set is obtained through morphological closing operations. Based on the second pixel set, a convolutional neural network model is used to identify the cracking point and obtain the first defect value; If the first defect value is greater than the preset defect threshold, then the extension line corresponding to the first defect value is extracted to obtain the defect connected region. The defect determination value is obtained by calculating the area of ​​the connected component of the defect; If the defect judgment value is greater than the preset judgment threshold, the trimming position corresponding to the defect judgment value is marked to obtain the defect removal item.

7. The edge-fracture-resistant grinding method for ultra-thin wafers according to claim 1, characterized in that, The process of acquiring target processing equipment parameters, combining them with defect removal criteria, performing grinding and trimming on the wafer, and recording processing parameters to complete the wafer processing includes: Based on the defect elimination items, obtain the equipment number of the target processing equipment and the corresponding parameter values, and generate a processing record table; Based on the processing record table, a decision tree algorithm is used for classification to obtain the status code; If the status code is greater than the preset status threshold, a processing instruction code is generated; Grinding parameters are set using processing instruction codes to obtain the processing task number; Perform a grinding operation on the corresponding material according to the processing task number to obtain the trimming and grinding degree; Based on the degree of trimming and grinding, the processing results are classified and determined.

8. A grinding and polishing system for preventing edge breakage of ultra-thin wafers, characterized in that, include: Grinding control module: acquires the current thickness and vacuum level of the wafer, calculates and generates grinding control commands, acquires the real-time thickness and vacuum level of the wafer during the grinding operation, and generates new grinding control commands to gradually thin the wafer to the target thickness while maintaining the adsorption force within the stable adsorption force boundary range throughout the process; Force monitoring module: Continuously monitors the stress state of the wafer during the grinding process, generates an initial force set, determines the wafer deformation based on the initial force set, and generates a control instruction set; Defect detection module: Based on the control instruction set, it performs edge cracking and defect detection on the wafer and generates defect rejection items; Trimming module: Obtains the parameters of the target processing equipment, combines them with defect rejection items, performs grinding and trimming processing on the wafer, records the processing parameters, and completes the wafer processing.

9. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the edge-fracture-resistant grinding process for adapting ultrathin wafers as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the edge-breakage grinding process for adapting ultrathin wafers as described in any one of claims 1 to 7.

Images