Wafer chamfering method, device and apparatus

Abstract

The present disclosure provides a wafer chamfering processing method and device, wherein the method comprises: obtaining chamfering processing control parameters of a wafer chamfering device and chamfering processing response information of a current wafer during chamfering processing; inputting the chamfering processing control parameters and the chamfering processing response information into an edge quality prediction model to obtain an edge quality prediction result of the current wafer; obtaining an edge quality actual measurement result of the current wafer, and controlling the wafer chamfering device to perform a chamfering processing control operation according to the deviation between the edge quality prediction result and the edge quality actual measurement result. By applying the embodiments of the present disclosure, a feedback relationship between processing data and quality detection results can be formed, and the device control operation to be performed subsequently can be determined accordingly, thereby guaranteeing the edge quality stability of the wafer chamfering processing process.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor testing technology, and in particular to a wafer chamfering process method, apparatus and equipment. Background Technology

[0002] Wafer chamfering is a process that uses grinding wheels to process the edges of a wafer to create rounded surfaces, bevels, or other predetermined contours. Wafer chamfering typically includes steps such as wafer mounting and alignment, grinding wheel positioning, rough grinding, fine grinding, cleaning, drying, and quality inspection.

[0003] During the wafer chamfering process, the processing quality of the wafer edges is affected by various factors such as the condition of the grinding wheel, the operating status of the equipment, and the processing environment. Existing wafer chamfering equipment typically performs chamfering according to pre-set process parameters and performs quality inspection on the wafer edges after processing is completed.

[0004] As the wear condition of the grinding wheel, the operating status of the equipment, or the processing environment changes, the actual processing result of the wafer edge may change at any time. Existing wafer chamfering methods cannot capture this change in a timely manner, making it difficult to determine whether to take appropriate equipment control measures, resulting in difficulty in consistently controlling the processing quality of the wafer edge. Summary of the Invention

[0005] This disclosure provides a wafer chamfering process method, apparatus, and equipment; it can establish a feedback relationship between processing data and quality inspection results, and determine the subsequent equipment control operations to be performed accordingly, thereby ensuring the edge quality stability of the wafer chamfering process.

[0006] The technical solution disclosed herein is implemented as follows: In a first aspect, this disclosure provides a wafer chamfering method, comprising: acquiring chamfering control parameters of a wafer chamfering device and chamfering response information of the current wafer during chamfering; inputting the chamfering control parameters and chamfering response information into an edge quality prediction model to obtain an edge quality prediction result of the current wafer; acquiring the actual edge quality measurement result of the current wafer, and controlling the wafer chamfering device to perform chamfering control operations based on the deviation between the edge quality prediction result and the actual edge quality measurement result.

[0007] Secondly, this disclosure provides a wafer chamfering apparatus, comprising: an acquisition module configured to acquire chamfering control parameters of a wafer chamfering device and chamfering response information of the current wafer during chamfering; a prediction module configured to input the chamfering control parameters and chamfering response information into an edge quality prediction model to obtain an edge quality prediction result for the current wafer; and a control module configured to acquire the measured edge quality result of the current wafer and control the wafer chamfering device to perform chamfering control operations based on the deviation between the edge quality prediction result and the measured edge quality result.

[0008] Thirdly, this disclosure provides a computing device including a memory and a processor, the memory being used to store executable instructions, and the processor executing the executable instructions stored in the memory to implement the steps of the wafer chamfering method of the first aspect.

[0009] Fourthly, this disclosure provides a computer storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the wafer chamfering method of the first aspect.

[0010] Fifthly, this disclosure provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the wafer chamfering method of the first aspect.

[0011] By applying the embodiments of this disclosure, chamfering control parameters, chamfering response information, edge quality prediction results, and actual edge quality measurement results are correlated. This allows for the prediction of the current wafer's edge quality using the current chamfering control parameters and response information. The predicted edge quality is then compared with the actual measured edge quality, establishing a feedback relationship between the processing data and the quality inspection results. The deviation between these two factors determines whether the processing parameters, equipment status, and prediction model match the actual processing process. Based on this, the subsequent equipment control operations to be performed are determined, ensuring the edge quality stability of the wafer chamfering process. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the system architecture of a wafer chamfering processing system provided in an embodiment of this disclosure.

[0013] Figure 2 This is a flowchart of a wafer chamfering process provided in an embodiment of the present disclosure.

[0014] Figure 3 This is a schematic diagram illustrating a process for controlling a wafer chamfering device to perform chamfering control operations, as provided in an embodiment of this disclosure.

[0015] Figure 4This is a schematic diagram illustrating a process for fine-tuning the model parameters of an edge quality model, as provided in an embodiment of this disclosure.

[0016] Figure 5 This is a flowchart illustrating the training process for training an edge quality model, as provided in an embodiment of this disclosure.

[0017] Figure 6 This is a schematic diagram of the structure of a wafer chamfering device provided in an embodiment of the present disclosure.

[0018] Figure 7 This is a schematic diagram of a process for continuously adjusting chamfering control parameters based on vibration amplitude, as provided in an embodiment of this disclosure.

[0019] Figure 8 This is a schematic diagram of the composition of a wafer chamfering processing apparatus provided in an embodiment of this disclosure.

[0020] Figure 9 This is a structural block diagram of a computing device provided in an embodiment of the present disclosure. Detailed Implementation

[0021] The technical solutions of this disclosure will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art should fall within the protection scope of this disclosure.

[0022] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure. The singular forms “a,” “the,” and “the” used in this disclosure are also intended to include the plural forms unless the context clearly indicates otherwise.

[0023] Furthermore, in the embodiments of this disclosure, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0024] To facilitate understanding of the technical solutions of the embodiments of this disclosure, the related technologies of the embodiments of this disclosure are described below. The following related technologies are optional solutions and can be combined with the technical solutions of the embodiments of this disclosure in any way, and they all fall within the protection scope of the embodiments of this disclosure.

[0025] Wafer chamfering is a process of removing material from the outer periphery of a wafer to form a predetermined profile. The predetermined profile of the wafer edge can be an arc profile, a bevel profile, a composite profile formed by a combination of arcs and bevels, or other profiles suitable for subsequent wafer manufacturing, handling, and packaging processes.

[0026] Wafer chamfering equipment can be equipped with grinding wheels and a wafer spindle for driving wafer rotation. During processing, the wafer is mounted on the wafer spindle, which drives the wafer to rotate. The grinding wheels move relative to the wafer edge and remove material from the wafer edge. The wafer chamfering process can include stages such as wafer mounting and alignment, grinding wheel positioning, rough grinding, fine grinding, cleaning, drying, and inspection. The rough grinding stage is used to remove more material and form a preliminary edge contour, while the fine grinding stage is used to refine the edge contour and reduce surface roughness.

[0027] During wafer chamfering, the processing results are influenced by a combination of factors, including processing parameters, grinding wheel condition, equipment response, and the processing environment. For example, as the grinding wheel wears down, the contact state between the grinding wheel and the wafer edge, the grinding force, and the material removal state may change. Changes in the processing chamber temperature, coolant temperature, or coolant flow rate may also alter the heat dissipation and chip removal conditions in the grinding area. Furthermore, changes in the vibration amplitude of the wafer spindle or grinding wheel can cause fluctuations in the relative position and contact load between the grinding wheel and the wafer edge.

[0028] Therefore, even if wafer chamfering equipment uses the same initial processing parameters, the edge quality obtained from different wafers may vary. Processing solely based on pre-set parameters is insufficient to reflect changes in the grinding wheel's state and the equipment's operating status; judging solely based on post-processing inspection results fails to fully utilize the processing response information generated during the process. When processing control, processing response, and post-processing inspection are separated, equipment control operations may suffer from insufficient judgment basis or delayed control timing.

[0029] The wafer chamfering method disclosed herein links chamfering control parameters, chamfering response information, edge quality prediction results, and actual edge quality measurement results. It can predict the edge quality of the current wafer using the chamfering control parameters and chamfering response information, and compare the predicted edge quality with the actual measured edge quality. Based on the deviation between the two, it determines whether the processing parameters, equipment status, and prediction model match the actual processing, and accordingly determines the subsequent equipment control operations to be performed.

[0030] Through the above processing, a feedback relationship can be established between processing data and quality inspection results. The data generated during the current wafer processing is no longer stored separately as equipment operation records, but rather a correspondence can be established with the measured edge quality after the current wafer processing is completed. This correspondence can be used to control the wafer chamfering equipment and also to subsequently update the edge quality prediction model, allowing the edge quality prediction model to gradually adapt to the processing characteristics exhibited by the specific wafer chamfering equipment, specific grinding wheel, and specific processing environment.

[0031] Please refer to Figure 1 , Figure 1 A schematic diagram of the system architecture of a wafer chamfering system is shown. The wafer chamfering system may include a computing device 100, a wafer chamfering device 200, and a quality inspection device 300. The wafer chamfering device 200 is used to chamfer the current wafer 210, the quality inspection device 300 is used to inspect the edge quality of the current wafer 210 after the chamfering is completed, and the computing device 100 is used to process the data provided by the wafer chamfering device 200 and the quality inspection device 300, and output corresponding control information to the wafer chamfering device 200.

[0032] The wafer chamfering apparatus 200 may include a grinding wheel 220, a wafer spindle 230, and a drive mechanism for driving the grinding wheel 220 and the wafer spindle 230. A wafer 210 may be mounted on the wafer spindle 230, which drives the wafer 210 to rotate about its wafer axis. The grinding wheel 220 may rotate about its axis and contact the outer peripheral edge of the wafer 210 to remove material from the edge of the wafer 210.

[0033] The computing device 100 can be communicatively connected to the wafer chamfering device 200. This communication connection can be wired or wireless. The computing device 100 can be located inside or outside the wafer chamfering device 200. The computing device 100 can be implemented by an industrial control computer of the wafer chamfering device 200, or by a server that communicates with multiple wafer chamfering devices.

[0034] The drive mechanism may specifically include a grinding wheel drive mechanism, a wafer spindle drive mechanism, a feed mechanism, a pressure control mechanism, and a coolant supply mechanism, etc. The quality inspection equipment 300 may include a contour inspection device, a surface roughness inspection device, an edge defect inspection device, or at least two of the above, and may also include other inspection devices. The quality inspection equipment 300 can inspect edge quality indicators such as chamfer width, contour deviation, chamfer radius error, fillet integrity, surface roughness, and edge chipping.

[0035] Among these parameters, chamfer width represents the dimension of the chamfered area formed after chamfering the wafer edge along a predetermined measurement direction. Profile deviation represents the deviation of the measured edge profile from the target edge profile. Chamfer radius error represents the difference between the measured chamfer radius and the target chamfer radius. Fillet integrity indicates whether the wafer edge fillets are continuous and whether there are any local defects. Surface roughness can be characterized by arithmetic mean roughness (Ra), maximum profile height (Rz), or other roughness parameters. Edge chipping can be characterized by the number of chipped edges, maximum chipping depth, maximum chipping width, or the proportion of the chipped area.

[0036] In some embodiments, the quality inspection equipment 300 can perform inspection immediately after the current wafer 210 is processed. In other embodiments, the current wafer 210 can enter the quality inspection equipment 300 after cleaning and drying. After obtaining the edge quality measurement results, the quality inspection equipment 300 can send the edge quality measurement results to the computing device 100.

[0037] Please refer to Figure 2 , Figure 2 A flowchart of a wafer chamfering process is shown. This method can be executed by a computing device 100 or by a controller configured in the wafer chamfering apparatus 200. Specifically, it includes steps S202-S206.

[0038] Step S202: Obtain the chamfering control parameters of the wafer chamfering equipment, as well as the chamfering response information of the current wafer during the chamfering process.

[0039] The chamfering control parameters are data used to control the wafer chamfering equipment 200 to perform chamfering operations. These parameters may include one or more of the following: grinding wheel speed, wafer speed, feed rate, downforce, and coolant flow rate.

[0040] The grinding wheel rotation speed represents the rotational speed of the grinding wheel 220 during chamfering, and its unit is revolutions per minute (RPM). The wafer rotation speed represents the speed at which the wafer spindle 230 drives the current wafer 210 to rotate. The feed rate represents the relative speed of the grinding wheel 220 and the current wafer 210 along the feed direction, and can also be characterized by a predetermined amount of material removed per unit time. The downforce represents the load applied by the grinding wheel 220 to the edge of the current wafer 210. The coolant flow rate represents the volume of coolant delivered to the grinding area per unit time.

[0041] The chamfering control parameters may include the set parameters used by the wafer chamfering equipment 200 when the current wafer 210 begins processing, or the actual control parameters adjusted during the processing of the current wafer 210. When the chamfering control parameters change with the processing stage, the computing device 100 can obtain the parameter values, parameter change curves, or parameter statistical characteristics corresponding to each processing stage.

[0042] For example, the computing device 100 can acquire the grinding wheel speed, wafer speed, feed rate, downforce, and coolant flow rate during the rough grinding stage, and the grinding wheel speed, wafer speed, feed rate, downforce, and coolant flow rate during the fine grinding stage. The computing device 100 can also acquire the average value, maximum value, minimum value, rate of change, and fluctuation range of each parameter throughout the entire processing period.

[0043] The chamfering response information is generated by the combined effects of the grinding wheel state, wafer state, equipment operating state, and processing environment during the chamfering process of the wafer chamfering equipment 200 in accordance with the chamfering control parameters. The chamfering response information reflects the processing state formed after the chamfering control parameters are applied to the actual equipment.

[0044] In some implementations, the chamfering response information may include one or more of the following: grinding wheel wear degree, grinding force, grinding force fluctuation characteristics, coolant temperature, machining cavity temperature, vibration amplitude, and spindle power consumption.

[0045] The wear of the grinding wheel can be characterized by the cumulative processing time, the cumulative number of wafers processed, the change in grinding wheel diameter, the number of dressing cycles, or the surface condition of the grinding wheel. Grinding force can be acquired by force sensors installed on the grinding wheel holder, wafer spindle, or workpiece stage. Grinding force can include three-dimensional force components Fx, Fy, and Fz, or it can include the resultant force calculated from these three-dimensional force components.

[0046] Grinding force fluctuation characteristics can include the average value, peak value, peak-to-peak value, standard deviation, rate of change, or frequency of fluctuation within a specific time period. Spindle power consumption can represent the power of the wafer spindle 230 or the grinding wheel spindle during processing, and its unit can be kilowatts. Increases or fluctuations in spindle power can be used to characterize changes in grinding load or equipment operating conditions.

[0047] The vibration amplitude can be acquired by a vibration sensing unit mounted on the wafer spindle 230. The vibration amplitude can represent the peak value, RMS value, peak-to-peak value, or other values ​​that characterize the vibration intensity of the vibration signal within a predetermined time period.

[0048] The computing device 100 can acquire chamfering response information in real time or near real time. Real-time acquisition can mean acquiring information continuously or at shorter sampling periods during the current wafer 210 processing. Near real-time acquisition can mean acquiring relevant information after a processing stage, after the current wafer 210 processing is completed, or after a predetermined number of data acquisitions are completed.

[0049] Step S204: Input the chamfering control parameters and chamfering response information into the edge quality prediction model to obtain the edge quality prediction result of the current wafer.

[0050] Edge quality prediction models are used to characterize the correlation between chamfering control parameters, chamfering response information, and the edge quality after wafer processing. Edge quality prediction models can be implemented using regression models. These regression models can fit the relationship between multiple input features and one or more output targets based on multiple sets of historical processing data.

[0051] In some implementations, the input features of the edge quality prediction model can be categorized into processing parameter features, grinding wheel condition features, environmental parameter features, and intermediate process features. Processing parameter features may include grinding wheel speed, wafer speed, feed rate, and downforce. Grinding wheel condition features may include grinding wheel wear degree, grinding force, and grinding force fluctuation characteristics. Environmental parameter features may include coolant flow rate, coolant temperature, machining cavity temperature, and vibration amplitude. Intermediate process features may include the three-dimensional grinding force components Fx, Fy, and Fz, as well as spindle power consumption.

[0052] The aforementioned categories are not intended to exclusively distinguish the sources of features. For example, coolant flow rate can be used as a chamfering control parameter set by the equipment controller, or the actual coolant flow rate fed back by the equipment can be used as chamfering response information. The computing device 100 can simultaneously acquire the set value and the measured value of the coolant flow rate, and determine the actual operating status of the cooling system based on the difference between the set value and the measured value.

[0053] The output targets of the edge quality prediction model can include the chamfer geometry quality and surface quality of the current wafer 210. Chamfer geometry quality can include profile deviation, chamfer width, chamfer radius error, and fillet integrity. Surface quality can include chamfer width, surface roughness, and edge chipping.

[0054] In some implementations, the edge quality prediction model can also output processing performance information such as material removal rate, processing time, and processing energy consumption. The material removal rate can represent the volume or mass of material removed from the current wafer 210 edge per unit time. The processing time can represent the time consumed to complete the chamfering of the current wafer 210. The processing energy consumption can be determined based on the power consumption of the grinding spindle, wafer spindle, cooling system, and other actuators.

[0055] Material removal rate, processing time, and processing energy consumption can serve as auxiliary predictive information to determine whether the processing is proceeding as expected. Predicted quality indicators in the edge quality prediction results can primarily include chamfer width, contour deviation, chamfer radius error, fillet integrity, surface roughness, and edge chipping. This avoids directly interpreting processing time or processing energy consumption as wafer edge quality.

[0056] Edge quality prediction results can include one or more predicted quality metrics. For example, edge quality prediction results can include predicted chamfer width, predicted surface roughness, and predicted edge chipping. Predicted edge chipping can be represented by predicted maximum chipping depth and predicted number of chipped edges.

[0057] The computing device 100 can continuously update the edge quality prediction results during the current wafer 210 processing. For example, the computing device 100 can update the model input as new chamfering response information is collected, so that the edge quality prediction results gradually reflect the actual processing of the current wafer 210. Alternatively, after the current wafer 210 processing is completed, the computing device 100 can input the chamfering control parameters and chamfering response information throughout the entire processing period into the edge quality prediction model to obtain the final edge quality prediction result.

[0058] Step S206: Obtain the measured edge quality results of the current wafer, and control the wafer chamfering equipment to perform chamfering control operations based on the deviation between the edge quality prediction results and the measured edge quality results.

[0059] Specifically, the measured edge quality results can be generated by the quality inspection equipment 300 after inspecting the current wafer 210. The measured quality indicators included in the measured edge quality results can correspond to the predicted quality indicators included in the edge quality prediction results.

[0060] For example, when the edge quality prediction results include predicted chamfer width, predicted surface roughness, and predicted edge chipping, the actual edge quality measurement results can include measured chamfer width, measured surface roughness, measured maximum edge chipping depth, and measured number of edge chips. The predicted quality indicators and the measured quality indicators can correspond to each other in terms of indicator meaning, units of measurement, and measurement location, so that the deviation between the two can be calculated.

[0061] The "deviation" referred to here can be the absolute deviation, relative deviation, or comprehensive deviation determined based on multiple test values ​​between the predicted quality indicator and the corresponding measured quality indicator. Different deviation calculation methods can be used for different quality indicators.

[0062] For example, for chamfer width, the absolute value of the difference between the predicted chamfer width and the measured chamfer width can be used as the deviation. For surface roughness, the ratio of the difference between the predicted roughness and the measured roughness to the measured roughness or the target roughness can be used as the relative deviation. For chipping, the predicted maximum chipping depth and the measured maximum chipping depth, as well as the predicted number of chipped edges and the measured number of chipped edges, can be compared separately.

[0063] The chamfering control operation is a subsequent action performed by the wafer chamfering equipment 200, based on the degree of matching between the predicted and measured results by the computing device 100. The chamfering control operation can be applied to the next wafer, or it can be applied to the current wafer 210 if the corresponding judgment criteria can be obtained in real time.

[0064] For example, chamfering control operations may include continuing to process the next wafer according to the currently used chamfering control parameters, adjusting the chamfering control parameters, stopping chamfering, or triggering the dressing of the grinding wheel 220. Different control operations correspond to different deviation states between the predicted results and the measured results.

[0065] The above method maps the control parameters used before or during the processing of the current wafer 210, the equipment response generated during the processing of the current wafer 210, and the measured quality after the processing of the current wafer 210 to the same wafer. Therefore, the computing device 100 can determine whether the prediction model accurately describes the current actual processing state, and whether the currently used control parameters are still suitable for the current equipment state and grinding wheel state.

[0066] When the predicted edge quality and the measured edge quality are close to each other, it indicates that the edge quality prediction model has good consistency with the actual result of the current processing, and the processing result formed by the current control parameters is within the expected range. At this time, the corresponding control parameters can continue to be used.

[0067] When there is a certain difference between the predicted edge quality and the measured edge quality, but the difference has not reached a serious abnormality level, the computing device 100 can adjust the chamfering processing control parameters to reduce the degree to which the subsequent wafer processing results deviate from the expected state.

[0068] When any edge quality indicator deviates significantly, this deviation may indicate a noticeable change in grinding wheel wear, equipment vibration, cooling status, or other processing conditions. Continuing processing in this situation may increase the risk of abnormal edge contours, rough surfaces, or chipped edges. The calculation device 100 can then stop the chamfering process and trigger grinding wheel dressing.

[0069] The chamfering response information enables the computing device 100 to understand the processing state formed after the control parameters are applied to the actual equipment. The edge quality prediction model enables the processing state to be converted into a predicted quality that can be compared with the detection results. The deviation between the predicted results and the measured results further forms the basis for the judgment of the wafer chamfering device 200.

[0070] Through the aforementioned data transmission relationships, equipment control no longer relies solely on preset processing parameters or a single post-processing detection result. Instead, it can simultaneously combine current processing parameters, current equipment response, and actual detection results to determine control operations. This improves the matching degree between equipment control operations and the actual processing state, and helps enhance the consistency of wafer edge quality during continuous processing.

[0071] Please refer to Figure 3 , Figure 3 The diagram illustrates the process of controlling a wafer chamfering device to perform chamfering operations based on the deviation between predicted and measured results, specifically including steps S302-S310. After obtaining the edge quality prediction results and the edge quality measured results, the deviation between each predicted quality index and the corresponding measured quality index can be determined, and the corresponding control branch can be selected from different preset control branches for the wafer chamfering device according to the range of each deviation.

[0072] In some implementations, preset tolerance ranges, partial consistency conditions, and complete inconsistency conditions can be set for different quality indicators. A preset tolerance range represents the allowable deviation range when there is a high degree of consistency between the predicted quality indicator and the corresponding measured quality indicator. A partial consistency condition indicates that there is a deviation between the predicted and measured results that requires parameter correction, but the processing has not yet reached the point of stopping. A complete inconsistency condition indicates that the deviation between the predicted and measured results has reached a point where processing needs to be stopped and the grinding wheel condition checked.

[0073] Step S302: Obtain the measured edge quality results of the current wafer and calculate the deviation between the predicted edge quality results and the measured edge quality results.

[0074] Step S304: Determine whether the deviation is within the preset tolerance range. If yes, proceed to step S306. If no, further determine whether the relationship between the deviation and the preset tolerance range satisfies the condition of partial consistency or complete inconsistency, and select to proceed to step S308 or step S310 based on the determination result.

[0075] Step S306: Control the wafer chamfering equipment to continue processing the next wafer according to the chamfering processing control parameters.

[0076] Specifically, in response to the deviations between the predicted quality indicators in the edge quality prediction results and the corresponding measured quality indicators in the actual edge quality measurement results, all of which are within the corresponding preset tolerance range, the computing device 100 controls the wafer chamfering device 200 to continue processing the next wafer according to the chamfering processing control parameters.

[0077] Among them, each predicted quality indicator refers to all predicted quality indicators involved in deviation judgment in the edge quality prediction results. The corresponding measured quality indicator refers to the detection indicator that has the same quality meaning as the corresponding predicted quality indicator. For example, the predicted chamfer width corresponds to the measured chamfer width, the predicted surface roughness corresponds to the measured surface roughness, and the predicted maximum chipping depth corresponds to the measured maximum chipping depth. "All within the corresponding preset tolerance range" means that each deviation involved in the judgment meets the consistency requirement corresponding to the quality indicator, rather than some quality indicators meeting the requirement.

[0078] For example, for chamfer width, a preset tolerance range can be set where the deviation between the predicted and measured values ​​is less than or equal to 3 micrometers. Here, the deviation can be the absolute value of the difference between the predicted and measured chamfer widths. Therefore, "deviation less than or equal to ±3 micrometers" can indicate that the difference between the two is in the range of -3 micrometers to +3 micrometers.

[0079] For surface roughness, a preset tolerance range can be set where the relative deviation between the predicted and measured values ​​is less than or equal to 5%. For example, the difference between the predicted Ra and the measured Ra can be divided by the measured Ra to obtain the relative roughness deviation.

[0080] For edge chipping, the maximum chipping depth can be less than or equal to 3 micrometers and the number of chipped edges can be less than or equal to 2. The edge chipping conditions can be set based on the comparison between the predicted and measured edge chipping conditions, or based on whether the measured edge chipping conditions are within the prediction range given by the edge quality prediction model.

[0081] When the chamfer width, surface roughness, and edge chipping conditions all meet the corresponding conditions, the computing device 100 can determine that there is a good matching relationship between the current prediction model, the current control parameters, and the actual processing state. The computing device 100 can maintain parameters such as grinding wheel speed, wafer speed, feed rate, downforce, and coolant flow rate, and control the wafer chamfering device 200 to process the next wafer.

[0082] Continuing to process the next wafer can mean that after the current wafer 210 has completed inspection and obtained the deviation judgment result, the wafer chamfering equipment 200 processes the next wafer according to the same chamfering processing control parameters as the current wafer 210. Alternatively, it can mean that the wafer chamfering equipment 200 has already processed subsequent wafers according to the current control parameters while waiting for the quality inspection results, and continues to maintain the corresponding parameters after obtaining the quality inspection results.

[0083] For example, during the processing of the first batch of test wafers or the initial batch of wafers, the wafer chamfering equipment 200 can continue operating after the previous wafer is sent to the quality inspection equipment 300, without needing to stop for a long time to wait for inspection. After the inspection results are returned to the computing equipment 100, if all deviations are within the corresponding tolerance range, the current control parameters can be maintained. This method can reduce the interruption of the processing flow caused by waiting for inspection results, and balance processing efficiency and quality control.

[0084] Step S308: Adjust the chamfering control parameters.

[0085] Specifically, in response to the fact that the deviations between each predicted quality index and the corresponding measured quality index do not meet the condition of complete inconsistency, and at least one predicted quality index and the corresponding measured quality index meet the condition of partial consistency, the chamfering processing control parameters are adjusted.

[0086] The partial consistency condition refers to a deviation that partially overlaps with a preset tolerance range but does not fall within the numerical range specified by the complete inconsistency condition. For example, if the chamfer width deviation is greater than 3 micrometers and less than or equal to 6 micrometers, the chamfer width can be considered to meet the partial consistency condition. Similarly, if the relative deviation of surface roughness is greater than 5% and less than or equal to 10%, the surface roughness can be considered to meet the partial consistency condition.

[0087] For edge chipping, if the edge chipping condition does not meet the consistency conditions of a maximum chipping depth less than or equal to 3 micrometers and a chipping number less than or equal to 2, but also does not meet the completely inconsistent conditions of a maximum chipping depth greater than 5 micrometers or a chipping number greater than or equal to 5, then the edge chipping condition can be determined as partially consistent. For example, if the maximum chipping depth is 4 micrometers, or the number of chipping edges is 3 or 4, it can be judged as partially consistent based on specific process requirements.

[0088] None of the deviations meet the complete inconsistency condition, thus excluding situations requiring immediate cessation of processing and overhaul. If at least one deviation meets the partial consistency condition, it indicates that a identifiable deviation has occurred in the current processing state. Continuing to maintain the original control parameters completely may cause subsequent processing results to deviate further from the expected state; therefore, correction can be made through parameter adjustments.

[0089] Adjusting the control parameters for chamfering can include adjusting one or more of the following: grinding wheel speed, wafer speed, feed rate, downforce, and coolant flow rate. The adjustment amount can be determined based on the type, direction, and degree of deviation, as well as the chamfering response information.

[0090] For example, when the measured chamfer width is greater than the predicted or target chamfer width, the feed rate can be reduced, the downforce decreased, or the processing time of the corresponding machining stage shortened. When the measured surface roughness is higher than the predicted surface roughness, the feed rate can be reduced, the grinding wheel speed adjusted, or the coolant flow rate increased. When the number of chipped edges increases or the maximum chipping depth increases, the relative feed rate between the grinding wheel and the wafer can be reduced, the spindle speed decreased, or the coolant flow rate adjusted.

[0091] The direction of the above parameter adjustments can be determined by considering the current wafer chamfering equipment 200's structure, grinding wheel type, and actual processing data. The computing device 100 can determine the direction and magnitude of parameter adjustments based on the correlation between parameter changes and edge quality changes recorded in historical training data.

[0092] In some implementations, the computing device 100 may initially use a smaller parameter adjustment range and then obtain the predicted and measured results again after processing the next wafer. If the deviation decreases, the adjusted chamfering control parameters can continue to be used; if the deviation does not decrease or increases further, the parameters can be adjusted again or the control branch for stopping processing and finishing can be entered.

[0093] In some implementations, the determination of whether to delay the shutdown can be based on the judgment results of multiple consecutive wafers. For example, if two wafers out of three consecutive wafers meet the partial consistency condition and the vibration characteristics change by more than 20% relative to the normal state, the computing device 100 can generate a delayed shutdown instruction for processing and grinding wheel dressing.

[0094] The vibration characteristics here can be the average value, peak value, RMS value, peak-to-peak value of the vibration amplitude, or the distribution characteristics of the vibration amplitude over different sampling periods. If the vibration spectrum is available, the vibration characteristics can also include the amplitude or energy of a specific frequency band. A change in vibration characteristics exceeding 20% ​​indicates that the relative difference between the current vibration characteristics and historical normal vibration characteristics exceeds 20%.

[0095] Delayed processing can indicate that the currently processed wafer is allowed to stop after completing a predetermined processing stage, or it can indicate that the next wafer will not be started after the current wafer is completed. Using the results of multiple consecutive wafers for judgment can reduce frequent downtime caused by fluctuations in the detection of a single wafer, and can also schedule grinding wheel dressing when deviations persist.

[0096] Step S310: Control the wafer chamfering equipment to stop the chamfering process and trigger the grinding wheel to be dressed.

[0097] Specifically, in response to the fact that the deviation between any predicted quality index in the edge quality prediction result and the corresponding measured quality index in the actual edge quality measurement result meets the condition of complete inconsistency, the computing device 100 controls the wafer chamfering device 200 to stop the chamfering process and triggers the grinding wheel 220 to be dressed.

[0098] The phrase "any one" indicates that if at least one of the multiple quality indicators involved in the judgment meets the condition of complete inconsistency, the process can be stopped and the finishing process can be initiated. For example, in the cases of chamfer width deviation, relative surface roughness deviation, and edge chipping, even if the other two are within the normal range, chamfering can be stopped as long as one of them meets the condition of complete inconsistency.

[0099] In some implementations, for chamfer width, a deviation greater than 6 micrometers between the predicted and measured values ​​can be set as a complete inconsistency condition. For surface roughness, a relative deviation greater than 10% between the predicted and measured values ​​can be set as a complete inconsistency condition. For chipping, a maximum chipping depth greater than 5 micrometers or a number of chipped edges greater than or equal to 5 can be set as a complete inconsistency condition.

[0100] The aforementioned thresholds can be adjusted based on wafer type, target chamfer profile, product process requirements, and the accuracy of the inspection equipment. For example, a smaller deviation threshold can be used for 3D NAND wafers, advanced logic chip wafers, or other wafers with high requirements for wafer edge quality. Different preset tolerance ranges and complete inconsistency conditions can also be set for wafers of different diameters, materials, or thicknesses.

[0101] When any quality indicator meets the condition of complete inconsistency, the result indicates that the current actual processing has deviated significantly from the predicted state. For example, the grinding wheel 220 may experience wear, blockage, localized damage, or changes in surface condition; the wafer spindle 230 may experience abnormal vibration; the coolant supply may change; or the actual grinding load may not match the set state.

[0102] The computing device 100 controls the wafer chamfering device 200 to stop the chamfering process, which can indicate stopping the feed motion between the grinding wheel 220 and the wafer, stopping the current chamfering process, or preventing new wafers from entering the processing position. While ensuring equipment safety, the grinding wheel 220 or the wafer spindle 230 can gradually reduce its speed according to the equipment shutdown procedure.

[0103] Triggering grinding wheel dressing can mean that the computing device 100 outputs a dressing command to the dressing mechanism of the wafer chamfering device 200, or it can mean generating a prompt message indicating that grinding wheel dressing is required and sending it to the operation terminal. When the wafer chamfering device 200 is equipped with an automatic dressing mechanism, the dressing mechanism can reshape and sharpen the working surface of the grinding wheel 220 according to the dressing command. In the case of manual or semi-automatic dressing, the operator can initiate the corresponding dressing program according to the prompt message.

[0104] Dressing can be used to remove blockages from the surface of the grinding wheel 220, adjust the shape of the grinding wheel, and restore the cutting state of the abrasive grains. After dressing, the grinding wheel status information can be reacquired, and the processing state after dressing can be verified by testing the wafer. If the verification results meet the reprocessing conditions, the wafer chamfering equipment 200 can resume wafer chamfering processing.

[0105] By stopping processing when any quality indicator reaches a point of complete inconsistency, the likelihood of continuing to process multiple wafers when the equipment or grinding wheel condition has already significantly changed can be reduced. Triggering grinding wheel dressing allows abnormal control operations to correspond to grinding wheel conditions that may cause quality deviations.

[0106] Therefore, during wafer chamfering, the control flow of the wafer chamfering equipment is divided into distinct control branches. When all deviations are within the preset tolerance range, the current control parameters are maintained and processing continues; when there is no complete inconsistency but at least one condition meets the partial consistency condition, the chamfering control parameters are adjusted; when any condition meets the complete inconsistency condition, the chamfering process is stopped and the grinding wheel dressing is triggered.

[0107] This control method translates different degrees of deviation between predicted and measured results into equipment control operations of varying intensities. When the deviation is small, unnecessary parameter changes are avoided; when the deviation is within the intermediate range, it is corrected through parameter adjustments; when the deviation is large, shutdown and wafer repair are used to limit the impact of abnormal processing conditions on subsequent wafers, thereby adapting equipment control operations to the degree of edge quality deviation.

[0108] Please refer to Figure 4 , Figure 4 The diagram illustrates the process of fine-tuning the model parameters of the edge quality model. Specifically, it includes steps S402-S404.

[0109] Based on the edge quality prediction and actual edge quality measurement results of the current wafer 210, the edge quality prediction model can be updated using the data generated by the current wafer 210. Therefore, the current wafer 210 not only serves as the object being processed, but its processing and inspection results can also form training data for subsequent model iterations.

[0110] Step S402: Associate the chamfering control parameters, chamfering response information, and edge quality measurement results into a set of training samples, and add the training samples to the sample dataset.

[0111] Here, association refers to establishing a correspondence between data belonging to the same wafer or the same processing task. For example, each wafer entering the wafer chamfering equipment 200 can be assigned a wafer identifier, and this wafer identifier can be associated with the processing batch, equipment identifier, grinding wheel identifier, and processing time. Data collected by the wafer chamfering equipment 200 during the processing of the current wafer 210 can carry the corresponding wafer identifier or processing task identifier, and the edge quality measurement results obtained by the quality inspection equipment 300 can also carry the same or mutually mapped identifiers.

[0112] Using the aforementioned identifiers, computing device 100 can determine whether a set of chamfering control parameters, a set of chamfering response information, and a set of measured edge quality results belong to the same wafer. Thus, the training samples can reflect the actual wafer edge quality obtained under specific control parameters and specific device response conditions.

[0113] The control parameters for chamfering can be represented in the training samples using either the original time series or parameter features extracted from the time series. For example, the processing of a wafer includes rough grinding and fine grinding stages. The training samples can record the grinding wheel speed, wafer speed, feed rate, downforce, and coolant flow rate for each stage. The training samples can also record the average, maximum, minimum, duration, or number of changes for each control parameter.

[0114] Chamfering response information can be represented using raw sampled data during processing or response features extracted from raw sampled data. For example, grinding force can be recorded as the time-varying data of triaxial grinding forces Fx, Fy, and Fz, or as the average grinding force, peak grinding force, peak-to-peak grinding force, and fluctuation degree. Vibration signals can be recorded as the time-varying data of vibration amplitude, or as the peak, average, or effective value of vibration amplitude. Spindle power can be recorded as a power curve, or as the average and maximum power during the processing stage.

[0115] The measured results of edge quality can be used as a supervised target for training samples. The measured results of edge quality can include one or more of the following: measured chamfer width, measured profile deviation, measured chamfer radius error, measured fillet integrity, measured surface roughness, measured maximum chipping depth, and measured number of chipped edges.

[0116] In some implementations, the material removal rate, actual processing time, and actual processing energy consumption can also be correlated with the training sample. This processing performance information can be used to train the auxiliary output of the edge quality prediction model, or to evaluate the performance of the chamfering control parameters recommended by the model in terms of processing efficiency and energy consumption.

[0117] The sample dataset can be stored in the local memory of the computing device 100 or in a server communicatively connected to the computing device 100. The sample dataset can be categorized according to wafer type, wafer size, grinding wheel type, wafer chamfering equipment, target chamfer profile, or processing batch. Different categories of data can be used to train different edge quality prediction models, or the same model can be applied to various processing conditions by using category labels as model input.

[0118] Step S404: In response to the accumulation of training samples in the sample dataset reaching the model iteration condition, fine-tune the model parameters of the edge quality prediction model based on the sample dataset.

[0119] Specifically, it determines whether the accumulated number of training samples in the sample dataset has reached the model iteration condition. If the accumulated number of training samples in the sample dataset has reached the model iteration condition, the model parameters of the edge quality prediction model are fine-tuned based on the sample dataset.

[0120] The model iteration condition is used to determine whether the newly accumulated data is sufficient to fine-tune the model parameters.

[0121] In some implementations, the model iteration condition can be that the number of new training samples added since the last model update reaches a preset number. For example, after each real-time feedback, 5 to 10 sets of new data can be accumulated, and the parameters of the regression model can be fine-tuned when the accumulated training samples reach the corresponding number.

[0122] For example, when the model iteration condition is set to add 5 sets of training samples, the computing device 100 can update the edge quality prediction model once after completing the processing, inspection, and data association of 5 wafers. When the model iteration condition is set to add 10 sets of training samples, the computing device 100 can update the model after accumulating 10 new training samples.

[0123] Model iteration conditions can also be set in conjunction with the quality of training samples. For example, when the new training samples include at least one set of data that meets the partial consistency condition, or include at least one set of data obtained after adjusting the chamfering control parameters, the computing device 100 can consider the new samples to be of high value for correcting the model and trigger model fine-tuning in advance.

[0124] In other implementations, model iteration conditions may include time conditions. For example, updates can be performed in near real-time after each wafer is inspected, or updates can be performed centrally after a processing batch is completed. The update frequency can be real-time or near real-time. After each wafer is processed and the measured results are obtained, the data of that wafer is added to the sample dataset, and the regression coefficients are updated based on the currently accumulated data.

[0125] Real-time updates mean that the model update begins as soon as the quality inspection results for the current wafer are returned, without waiting for the entire production batch to finish. Near-real-time updates mean that the update is completed after data from several wafers has been generated or within a time window that does not affect the equipment control response.

[0126] Model parameter fine-tuning refers to adjusting the parameters in an existing edge quality prediction model that describe the relationship between input features and output targets using newly added training samples. Model parameter fine-tuning can retain the relationships already learned by the existing model and use recent processing data to correct the model's description of the current grinding wheel state, current equipment state, and current processing environment.

[0127] For example, when the edge quality prediction model is a regression model, the model parameters can include regression coefficients, intercept terms, or parameters describing the combined relationships between different input features for each input feature. As the wear of the grinding wheel increases, the chamfer width or surface roughness corresponding to the same feed rate and downward pressure may gradually change. By fine-tuning the regression parameters using recent samples, the model output can be adaptively corrected as the grinding wheel condition changes.

[0128] Model fine-tuning can utilize all training samples in the dataset, or it can use recently added training samples and some historical training samples. Using all training samples maintains the model's description of the overall processing patterns; using more recent training samples improves the model's adaptability to the current grinding wheel and equipment state.

[0129] In some implementations, different levels of participation can be assigned to training samples obtained at different times. For example, more recent training samples can have higher weights in model fine-tuning, while earlier training samples can have lower weights. This approach helps the edge quality prediction model to quickly reflect changes in grinding wheel wear, cooling status, and equipment vibration that occur during the production process.

[0130] After fine-tuning the model parameters, the updated edge quality prediction model can be used to process subsequent wafer chamfering control parameters and chamfering response information. The prediction bias before and after the model update can also be compared. When the prediction bias of the updated model on validation data decreases, the updated model can be used for subsequent predictions; when the prediction bias does not decrease or increases, the unupdated model can be retained and more training samples can be waited for.

[0131] By correlating control parameters, response information, and measured quality as training samples, both the model's input data and output target originate from the actual wafer fabrication process. The edge quality prediction model can gradually approximate the actual processing characteristics of specific equipment and grinding wheels through continuously accumulated data. This feedback mechanism reduces the discrepancy between the theoretical model and the actual processing environment, allowing the parameter relationships described by the model to adjust according to actual production conditions.

[0132] Furthermore, the relationship between changes in processing parameters and grinding wheel failure time can be fitted. In some implementations, the trend of grinding wheel condition changing with the number of wafers processed can be determined based on the cumulative processing time of the grinding wheel, the cumulative number of wafers processed by the grinding wheel, the degree of grinding wheel wear, model prediction bias, and dressing records recorded in the training samples.

[0133] For example, as the cumulative number of milling operations increases, if the grinding force, vibration amplitude, and spindle power gradually increase, and the predicted deviation of the chamfer width or surface roughness also gradually increases, the time when the mill wheel needs to be overhauled or replaced can be estimated based on this trend. This estimation result can be called the mill wheel failure time prediction result or the mill wheel remaining service life prediction result.

[0134] The predicted failure time of the grinding wheel can be used as equipment maintenance information, rather than being directly used as an edge quality prediction result. The production management system can schedule grinding wheel repair, replacement, and spare parts preparation based on the remaining service life of the grinding wheel, thereby reducing processing interruptions caused by sudden deterioration of the grinding wheel's condition and helping to improve the utilization of the grinding wheel and reduce spare parts costs.

[0135] Please refer to Figure 5 , Figure 5 A flowchart illustrating the training process for an edge quality model is provided. Before continuously fine-tuning the model using actual data from the current wafer, an initial edge quality prediction model can be trained and calibrated to obtain an edge quality prediction model that meets predetermined prediction requirements. In some embodiments, the training process for the edge quality prediction model specifically includes steps S502-S510.

[0136] Step S502: Obtain the initial edge quality prediction model and obtain the test wafer set.

[0137] The initial edge quality prediction model can be a model pre-established based on historical chamfering processing data, or it can be a model established based on experimental data but not yet calibrated for the current wafer chamfering equipment 200. The test wafer can be a test wafer specifically used for model calibration, or it can be the first batch of production wafers selected under the condition of meeting production management requirements.

[0138] A test wafer set can include one test wafer or multiple test wafers. When multiple test wafers are included, each test wafer can use the same chamfering control parameters to verify the predictive stability of the model under the same conditions; each test wafer can also use different combinations of control parameters to verify the predictive ability of the model under different processing conditions.

[0139] Step S504: Obtain the sample chamfering control parameters used by the wafer chamfering equipment when processing the test wafer, and the sample chamfering response information of the test wafer during the chamfering process.

[0140] The term "sample" indicates that the corresponding parameters and information are used for model training or model calibration, and does not imply that its data structure is different from the chamfering control parameters and chamfering response information of the current wafer.

[0141] The control parameters for sample chamfering can include grinding wheel speed, wafer speed, feed rate, downforce, and coolant flow rate. To ensure that the test data covers different process conditions, multiple sets of parameter combinations can be set within the processing range allowed by the equipment.

[0142] For example, the grinding wheel speed, wafer speed, feed rate, or downward pressure can be varied between different test wafers, allowing the initial edge quality prediction model to learn the correlation between different combinations of processing parameters and edge quality. The parameter combinations should meet the safety requirements of the wafer chamfering equipment 200 and the processing requirements of the test wafers.

[0143] The chamfering response information of the sample can include the wear degree of the grinding wheel, the grinding force and its fluctuation characteristics, the coolant temperature, the temperature of the machining cavity, the vibration amplitude, the three-dimensional grinding forces Fx, Fy and Fz, and the spindle power consumption.

[0144] Step S506: Input the sample chamfering processing control parameters and sample chamfering processing response information into the initial edge quality prediction model to obtain the sample edge quality prediction results of the test wafer.

[0145] The sample edge quality prediction results can include predicted chamfer width, predicted contour deviation, predicted chamfer radius error, predicted fillet integrity, predicted surface roughness, and predicted edge chipping. Before the initial model is calibrated for the current device, the sample edge quality prediction results may differ from the actual detection results.

[0146] After the test wafer undergoes chamfering, it can be cleaned and dried, and the sample edge quality measurement results can be obtained by the quality inspection equipment 300. The sample edge quality measurement results can include various measured quality indicators corresponding to the sample edge quality prediction results.

[0147] Step S508: Obtain the measured results of the sample edge quality of the test wafer, and determine the test deviation between the predicted results of the sample edge quality and the measured results of the sample edge quality.

[0148] The test deviation can be determined separately for each quality indicator, or it can be a comprehensive test deviation based on multiple quality indicators. For example, the width test deviation can be determined between the predicted chamfer width and the actual chamfer width; the roughness test deviation can be determined between the predicted surface roughness and the actual surface roughness; and the edge chipping test deviation can be determined between the predicted edge chipping and the actual edge chipping.

[0149] The calibration tolerance range is used to determine whether the initial edge quality prediction model meets the predetermined requirements for the current wafer chamfering device 200. The calibration tolerance range can be the same as the aforementioned preset tolerance range, or it can be set separately according to the requirements of the model training phase.

[0150] Step S510: In response to the test deviation exceeding the calibration tolerance range, based on the sample chamfering processing control parameters, sample chamfering processing response information, and sample edge quality test results, adjust the model parameters of the initial edge quality prediction model until the test deviation does not exceed the calibration tolerance range, and obtain the trained edge quality prediction model.

[0151] After setting the model parameters for the initial edge quality prediction model, the existing test data can be reprocessed using the adjusted initial edge quality prediction model, or new test wafers can be processed to obtain new sample edge quality prediction results and actual sample edge quality results. The test deviation is then re-determined, and it is further determined whether the test deviation exceeds the calibration tolerance range.

[0152] The above process can be repeated until the test deviation does not exceed the calibration tolerance range. When the test deviation does not exceed the calibration tolerance range, the trained edge quality prediction model can be obtained.

[0153] Training completion indicates that the model has reached a state where it can be used for subsequent edge quality prediction under the current calibration data and requirements, and the model parameters can continue to iterate during subsequent processing. As new actual processing data is generated, the trained edge quality prediction model can still be used according to... Figure 4 The process shown is fine-tuned online.

[0154] In the actual production process, after the test wafers are processed, they can be sent to the quality inspection equipment 300 for testing. While waiting for the test results, if the equipment status and production management conditions permit, the wafer chamfering equipment 200 can continue to process subsequent wafers based on the model simulation results or current control parameters, without needing to remain shut down for an extended period.

[0155] Once the detection results are returned to the computing device 100, the computing device 100 compares the predicted sample edge quality with the measured sample edge quality. If the test deviation is within the calibration tolerance range, the model can be considered to have completed the calibration of the current stage, and the model can be used for subsequent production processing. If the test deviation exceeds the calibration tolerance range, new processing data and measured results can be added to the training data, and the model can be readjusted.

[0156] Using the above method, model training, wafer fabrication testing, quality inspection, and model parameter adjustment form a cycle. Test feedback allows the model built based on historical data to gradually approach the actual state of the current equipment, grinding wheel, and processing environment, and can reduce the occurrence of large deviations between model predictions and actual quality.

[0157] The initial edge quality prediction model can be constructed using multiple sets of initial training samples. First, multiple sets of initial training samples can be obtained, each set including historical chamfering control parameters, historical chamfering response information, and historical measured edge quality results obtained after wafer chamfering. Then, based on these multiple sets of initial training samples, the correlation between the historical chamfering control parameters, historical chamfering response information, and historical measured edge quality results is fitted to obtain the initial edge quality prediction model.

[0158] Historical chamfering control parameters are the control parameters used by the wafer chamfering equipment or similar equipment in historical processing tasks. Historical chamfering response information refers to the equipment and environmental responses generated during the processing of the corresponding historical wafer. Historical edge quality measurement results are the inspection results obtained after the chamfering process of the corresponding historical wafer is completed.

[0159] "History" is used to distinguish between data that existed before the initial model was built and data that is currently being processed; it does not require that the data was generated a long time ago. For example, data generated from a set of process experiments specifically conducted before the model was built can also be used as historical training data.

[0160] In some implementations, the initial modeling data may include 100 to 300 sets of experimental data covering different combinations of process parameters. Each set of experimental data may correspond to a test wafer or a chamfering task with complete data records.

[0161] To improve the coverage of the processing range by the initial training data, multiple sets of initial training samples can cover different combinations of grinding wheel speed, wafer speed, feed rate, downforce, and coolant flow rate. Multiple sets of initial training samples can also cover different grinding wheel wear stages, different coolant temperatures, different cavity temperatures, and different vibration amplitude states.

[0162] Fitting correlations refers to determining the correspondence between changes in input features and changes in output targets using initial training samples. In some implementations, the initial edge quality prediction model specifically employs a regression model.

[0163] Regression models can be built for a single edge quality metric or for multiple edge quality metrics simultaneously. For example, a width regression model can be built to predict chamfer width, a roughness regression model to predict surface roughness, and a chipping regression model to predict edge chipping. Multiple outputs can also be set within the same model, allowing it to simultaneously predict multiple quality metrics.

[0164] During the fitting process, grinding wheel speed, wafer speed, feed rate, downforce, grinding wheel wear degree, grinding force, coolant flow rate, coolant temperature, machining cavity temperature, vibration amplitude, three-dimensional grinding force, and spindle power consumption can be used as input features. Chamfer width, contour deviation, chamfer radius error, fillet integrity, surface roughness, and edge chipping can be used as output targets.

[0165] Material removal rate, processing time, and processing energy consumption can be used as additional output targets. In some embodiments, the computing device 100 can simultaneously predict edge quality and processing performance, and select a combination of parameters from multiple optional control parameter combinations that takes into account edge quality, processing time, and processing energy consumption.

[0166] The fitted initial edge quality prediction model can be used to simulate the impact of chamfering processing parameter variations on wafer edge quality. For example, computing device 100 can input a set of candidate control parameters and current device response information into the initial edge quality prediction model to obtain the corresponding predicted chamfer width, predicted surface roughness, and predicted edge chipping.

[0167] By changing the candidate control parameters, the computing device 100 can compare the prediction results corresponding to different parameter combinations and determine appropriate chamfering control parameters for the test wafer or subsequent wafers.

[0168] For example, the operator can input the target chamfer width, target surface roughness, or allowable edge chipping range. The computing device 100 can evaluate multiple sets of candidate control parameters based on the edge quality prediction model and output candidate parameters that match the target quality. The operator can also directly set a set of custom chamfering control parameters, and the computing device 100 can predict the edge quality that this set of parameters may result in.

[0169] Even when using custom parameters, the computing device 100 can still acquire chamfering response information during actual processing and compare the predicted results with the measured results after processing is completed. Therefore, the custom parameters remain integrated into the real-time feedback and quality verification process.

[0170] Please refer to Figure 6 The wafer chamfering apparatus 200 may include a grinding wheel 220, a wafer spindle 230, and a vibration sensing unit 240. A wafer 210 is mounted on the wafer spindle 230, which drives the wafer 210 to rotate. The grinding wheel 220 can contact the edge of the wafer 210 and perform chamfering.

[0171] The vibration sensing unit 240 is mounted on the wafer spindle 230. The vibration sensing unit 240 can be fixed to the shaft body, bearing housing, spindle housing, or other location capable of transmitting wafer spindle vibrations. Figure 6 In one of the structures shown, a vibration sensing unit 240 is disposed on the spindle portion near the wafer 210 mounting position to collect vibrations related to wafer rotation and grinding contact.

[0172] By mounting the vibration sensing unit 240 on the wafer spindle 230 used to drive the wafer rotation, the attenuation or superposition of vibration after transmission between different structural components can be reduced, so that the collected vibration information can more directly reflect the vibration state of the wafer spindle 230 and the current wafer 210 during the grinding process.

[0173] The vibration sensing unit 240 can convert mechanical vibration into an electrical signal. The controller of the computing device 100 or the wafer chamfering device 200 can determine the vibration amplitude based on the vibration signal output by the vibration sensing unit 240.

[0174] Vibration amplitude is used to characterize vibration intensity and can be determined by the peak value, effective value, peak-to-peak value, or other statistical values ​​within a predetermined time period of the vibration signal.

[0175] In some embodiments, the vibration sensing unit 240 can directly output the calculated vibration amplitude. In other embodiments, the vibration sensing unit 240 outputs the original vibration signal, the computing device 100 samples and processes the original vibration signal, and determines the corresponding vibration amplitude.

[0176] The computing device 100 can acquire the vibration signal of the current wafer 210 during the chamfering process through the vibration sensing unit 240, and determine the vibration amplitude based on the vibration signal.

[0177] Vibration signal acquisition can be performed throughout the entire chamfering process of the current wafer 210, or during the rough grinding stage, fine grinding stage, or a predetermined period of contact between the grinding wheel 220 and the current wafer 210. The computing device 100 can continuously acquire vibration signals according to a preset sampling period and calculate the vibration amplitude using a sliding time window.

[0178] Furthermore, it can be determined whether the vibration amplitude exceeds the preset normal range. The preset normal range can be determined based on historical vibration data obtained from the wafer chamfering equipment 200 under normal processing conditions.

[0179] For example, multiple sets of vibration amplitude values ​​can be collected when the grinding wheel is in normal condition, the wafer edge quality meets the requirements, and the equipment is not malfunctioning, and the normal range can be determined based on the corresponding data. Different preset normal ranges can be set for different processing stages, different grinding wheel speeds, or different wafer speeds.

[0180] A vibration amplitude exceeding the preset normal range indicates that the vibration amplitude is greater than the upper limit of the normal range. In some devices, an abnormally low vibration amplitude can also reflect sensor detachment or abnormal device operation. A vibration amplitude below the lower limit of the normal range can also be considered an abnormal condition.

[0181] In response to a vibration amplitude exceeding a preset normal range, the computing device 100 adjusts the chamfering processing control parameters and generates the adjusted chamfering processing control parameters. This vibration control process can occur before the current wafer 210 has been processed, thus enabling intervention in the current processing after detecting an abnormal vibration.

[0182] Adjusting the chamfering control parameters can include reducing the spindle speed, reducing the feed rate, and increasing the coolant flow rate. The spindle speed can be specified as either the grinding wheel spindle speed or the wafer spindle speed, depending on the control structure of the wafer chamfering equipment 200. When the equipment independently controls both the grinding wheel spindle and the wafer spindle, the speeds of the respective spindles can be recorded and adjusted separately.

[0183] When the vibration abnormality detection module outputs a warning signal, it can reduce the spindle speed by 5% to 15% of the current value, reduce the feed rate by 10% to 20% of the current value, and increase the coolant flow rate by 10% to 30% of the current value.

[0184] For example, if the current spindle speed is a predetermined speed V1, the adjusted spindle speed can be 85% to 95% of V1. If the current feed rate is F1, the adjusted feed rate can be 80% to 90% of F1. If the current coolant flow rate is Q1, the adjusted coolant flow rate can be 110% to 130% of Q1.

[0185] Reducing the spindle speed can decrease the dynamic fluctuations caused by high-speed contact between the grinding wheel and the wafer edge. Reducing the feed rate can decrease the amount of material removed per unit time and the variation in grinding load. Increasing the coolant flow rate can improve the cooling and chip removal conditions in the grinding zone. Coordinated adjustment of these three parameters allows for the simultaneous application of vibration anomaly treatment to relative motion speed, feed load, and cooling conditions.

[0186] The above parameter adjustment ratio is an example. The computing device 100 can select the appropriate adjustment ratio according to the degree to which the vibration amplitude exceeds the normal range. For example, when the vibration amplitude slightly exceeds the normal range, a smaller adjustment ratio within the corresponding range can be used; when the vibration amplitude exceeds the normal range significantly, a larger adjustment ratio can be used.

[0187] In some implementations, the adjustment method can be determined by combining the grinding force and spindle power. When the vibration amplitude, grinding force, and spindle power all increase, the spindle speed and feed rate can be reduced simultaneously, while the coolant flow rate can be increased. When the vibration amplitude increases but the grinding force changes little, the speed can be adjusted first, and the vibration status can be continuously observed.

[0188] By acquiring vibration signals during the current wafer 210 processing, the computing device 100 can obtain real-time information reflecting changes in the processing state. When the vibration amplitude exceeds the normal range, parameter adjustments can be made to intervene in the processing state before the edge quality detection results are formed, thus cooperating with batch-to-batch feedback control based on the deviation between predicted and measured results.

[0189] Control based on predicted and measured results can reflect the actual edge quality of the completed wafer and be used to control subsequent wafers; control based on vibration amplitude can respond to abnormal conditions during the current wafer processing. The two control levels utilize post-processing quality feedback and in-process equipment response, respectively, to form a control method covering both the processing process and the processing result.

[0190] Please refer to Figure 7 After generating the adjusted chamfering control parameters, the vibration amplitude can be continuously monitored. Figure 7 A schematic diagram of the process for continuously adjusting the chamfering control parameters based on the vibration amplitude is shown, specifically including steps S702-S706.

[0191] Step S702: Control the wafer chamfering equipment to continue chamfering the current wafer according to the adjusted chamfering processing control parameters, and re-acquire vibration signals within a preset time period to redetermine the vibration amplitude.

[0192] The preset time period provides an observation window for changes in the device's state after parameter adjustment. The preset time period can be 5 seconds. The computing device 100 can continuously collect vibration signals within 5 seconds after parameter adjustment is completed, and determine the vibration amplitude multiple times at the end of the time period or within the time period.

[0193] The preset time period can also be set according to the response speed, processing stage, and remaining processing time of the wafer chamfering equipment 200. For example, when the equipment can stabilize quickly after parameter adjustment, the preset time period can be shorter than 5 seconds; when a longer period of observation of vibration changes is required, the preset time period can be longer than 5 seconds.

[0194] Specifically, it determines whether the reacquired vibration amplitude has fallen back to the preset normal range. When the vibration amplitude falls back to the preset normal range, the currently adjusted chamfering control parameters can be maintained, and the chamfering of the current wafer 210 can continue to be completed.

[0195] Even after the current 210 wafer is processed, its edge quality can still be inspected using quality inspection equipment 300. The control parameters before and after parameter adjustment, vibration response, and the final measured edge quality results can be correlated as training samples. These training samples can be used to teach the model the relationship between abnormal vibration states, parameter adjustment methods, and the final edge quality.

[0196] Step S704: In response to the fact that the reacquired vibration amplitude has not fallen back to the preset normal range and the number of times the chamfering control parameters have been adjusted has not reached the preset upper limit, the chamfering control parameters are adjusted again.

[0197] Return to step S702. If the reacquired vibration amplitude does not fall back to the preset normal range, and the number of times the chamfering control parameters are adjusted has not reached the preset upper limit, the chamfering control parameters can be adjusted again, and the steps of continuing processing according to the adjusted chamfering control parameters and reacquiring vibration signals can be repeated.

[0198] The parameter adjustment count records the number of parameter adjustment rounds performed in response to the current vibration anomaly. When a vibration anomaly is first detected and parameters are adjusted, the parameter adjustment count is recorded as one. If the vibration does not return to normal after a preset time period and adjustments are made again, the parameter adjustment count is incremented by one.

[0199] Readjustments can be made based on the parameters adjusted previously. For example, if the spindle speed has been reduced, it can be reduced again by 5% to 15%; if the feed rate has been reduced, it can be reduced again by 10% to 20%; if the coolant flow rate has been increased, it can be increased again by 10% to 30%.

[0200] During multiple rounds of adjustments, the computing device 100 can set upper and lower limits for each parameter to ensure that the adjusted parameters are within the operating range allowed by the wafer chamfering device 200. For example, the spindle speed and feed rate should not be lower than the lower limit allowed for the corresponding processing stage, and the coolant flow rate should not be higher than the upper limit allowed by the cooling system.

[0201] The preset upper limit is used to limit the number of times parameter adjustments can be performed consecutively for the same vibration anomaly. The preset upper limit can be set according to the control response of the wafer chamfering equipment 200, the wafer processing stage, and product quality requirements. For example, the preset upper limit can be two, three, or other times.

[0202] Setting an upper limit on the number of parameter adjustments can prevent the continuous reduction of rotational speed and feed rate or the continuous increase of coolant flow rate when abnormal vibration persists. If the vibration does not recover after a limited number of adjustments, the abnormal vibration may not be caused by minor machining fluctuations that can be corrected by parameters, but may be related to the condition of the grinding wheel or the mechanical condition of the equipment.

[0203] Step S706: In response to the parameter adjustment number reaching the preset upper limit and the re-determined vibration amplitude still not falling back to the preset normal range, control the wafer chamfering equipment to stop chamfering processing and trigger the grinding wheel of the wafer chamfering equipment to be repaired.

[0204] Specifically, the feed of the grinding wheel 220 relative to the current wafer 210 can be stopped first, and then the speeds of the grinding wheel spindle and the wafer spindle can be reduced according to the equipment shutdown sequence. The vibration amplitude, grinding force, spindle power, control parameters, and current processing stage at the time of processing stop can also be recorded for subsequent equipment checks and model updates.

[0205] After the grinding wheel dressing is triggered, the working surface of the grinding wheel 220 can be dressed by the dressing mechanism. After dressing is completed, the wafer chamfering equipment 200 can use a test wafer or a subsequent wafer for verification processing and re-acquire vibration signals. If the vibration amplitude returns to the normal range and the edge quality detection results meet the requirements, production processing can be resumed.

[0206] thus, Figure 7 The process shown includes continuous control steps such as anomaly detection, initial parameter adjustment, vibration retesting, further parameter adjustment, and shutdown for maintenance. The degree of vibration anomaly can be further determined by the recovery status after multiple rounds of adjustments.

[0207] When vibration returns to the normal range after one or more parameter adjustments, it indicates that the abnormal state can be corrected through processing parameters. If vibration fails to recover after reaching the upper limit of adjustment times, possible abnormal grinding wheel conditions can be addressed by stopping the machine and performing a wheel overhaul, reducing the continuous impact of abnormal vibration on the wafer edge. Following this process, the vibration sensing unit 240 is installed on the wafer spindle 230, enabling timely acquisition of vibrations related to wafer rotation and grinding contact. The computing device 100 determines whether an abnormality exists based on the vibration amplitude and coordinates the reduction of spindle speed, feed rate, and increase of coolant flow. When parameter adjustments fail to restore vibration, processing is further stopped and a wheel overhaul is triggered.

[0208] This control link enables the equipment to proactively intervene when vibration intensifies but edge quality measurement results have not yet been formed, helping to reduce the risk of edge chipping caused by continuously increasing vibration, and can reduce the impact of edge chipping by more than 80%. This effect is affected by wafer material, grinding wheel type, initial parameter values, vibration threshold, and quality inspection aperture, and can be verified through specific test data.

[0209] By combining model prediction, actual measurement feedback, and vibration monitoring, the wafer chamfering equipment 200 can continue operating based on the model and processing response before the inspection results are returned. After the inspection results are returned, it can confirm, adjust, or stop processing based on the prediction deviation, and update the model with new data after each wafer is processed. This approach helps improve the continuity of the processing flow and reduces downtime caused by long waiting times for inspection results.

[0210] The processing parameter adjustments and downtime maintenance are performed by the computing device 100 based on preset judgment conditions, which can reduce reliance on operator experience. By identifying vibrations and quality deviations before the grinding wheel condition deteriorates significantly, it can also provide a basis for grinding wheel maintenance and replacement, which helps improve the manageability of the grinding wheel's service life and reduce production risks and spare parts costs.

[0211] Please refer to Figure 8 , Figure 8 This is a schematic diagram illustrating the composition of a wafer chamfering apparatus provided in this disclosure. The wafer chamfering apparatus 800 includes: Acquisition module 802: is configured to acquire chamfering control parameters of the wafer chamfering equipment, as well as chamfering response information of the current wafer during chamfering.

[0212] Prediction module 804: is configured to input chamfering control parameters and chamfering response information into the edge quality prediction model to obtain the edge quality prediction result of the current wafer; Control module 806: is configured to acquire the measured edge quality results of the current wafer, and control the wafer chamfering equipment to perform chamfering control operations based on the deviation between the edge quality prediction results and the measured edge quality results.

[0213] In some embodiments, the chamfering control operation includes continuing to process the next wafer according to the chamfering control parameters; the control module 806 is further configured to: in response to the deviation between each predicted quality index in the edge quality prediction result and the corresponding measured quality index in the edge quality measurement result being within the corresponding preset tolerance range, control the wafer chamfering equipment to continue processing the next wafer according to the chamfering control parameters.

[0214] In some embodiments, the chamfering control operation includes stopping the chamfering process and triggering the dressing of the grinding wheel of the wafer chamfering equipment; the control module 806 is further configured to: in response to the deviation between any predicted quality index in the edge quality prediction result and the corresponding measured quality index in the edge quality measurement result satisfying the condition of complete inconsistency, control the wafer chamfering equipment to stop the chamfering process and trigger the dressing of the grinding wheel.

[0215] In some embodiments, the chamfering control operation further includes adjusting the chamfering control parameters; the control module 806 is further configured to: adjust the chamfering control parameters in response to the fact that the deviations between each predicted quality index and the corresponding measured quality index do not meet the completely inconsistent condition and the deviation between at least one predicted quality index and the corresponding measured quality index meets the partially consistent condition.

[0216] In some embodiments, the wafer chamfering apparatus 800 further includes a first training module configured to: associate chamfering control parameters, chamfering response information and edge quality measurement results into a set of training samples, and add the training samples to a sample dataset; and fine-tune the model parameters of the edge quality prediction model based on the sample dataset in response to the number of accumulated training samples in the sample dataset reaching the model iteration condition.

[0217] In some embodiments, the wafer chamfering apparatus 800 further includes a second training module configured to: acquire an initial edge quality prediction model and acquire a test wafer group, the test wafer group including at least one test wafer to be processed; acquire sample chamfering control parameters used by the wafer chamfering equipment when processing the test wafer and sample chamfering response information of the test wafer during chamfering; input the sample chamfering control parameters and sample chamfering response information into the initial edge quality prediction model to obtain the sample edge quality prediction result of the test wafer; acquire the sample edge quality measurement result of the test wafer and determine the test deviation between the sample edge quality prediction result and the sample edge quality measurement result; in response to the test deviation exceeding the calibration tolerance range, adjust the model parameters of the initial edge quality prediction model based on the sample chamfering control parameters, sample chamfering response information and sample edge quality measurement result until the test deviation does not exceed the calibration tolerance range, and obtain the trained edge quality prediction model.

[0218] In some embodiments, the second training module is further configured to: acquire multiple sets of initial training samples, the initial training samples respectively including historical chamfering processing control parameters, historical chamfering processing response information and historical edge quality measurement results obtained after wafer chamfering processing; based on the initial training samples, fit the correlation between historical chamfering processing control parameters, historical chamfering processing response information and historical edge quality measurement results to obtain an initial edge quality prediction model.

[0219] In some embodiments, the wafer chamfering equipment is provided with a vibration sensing unit mounted on the wafer spindle for driving wafer rotation. The wafer chamfering processing apparatus 800 also includes an adjustment module configured to: acquire the vibration signal of the current wafer during chamfering processing through the vibration sensing unit, and determine the vibration amplitude based on the vibration signal; and adjust the chamfering processing control parameters in response to the vibration amplitude exceeding a preset normal range, thereby generating the adjusted chamfering processing control parameters.

[0220] In some embodiments, the adjustment module is further configured to: control the wafer chamfering equipment to continue chamfering the current wafer according to the adjusted chamfering control parameters, and re-acquire vibration signals within a preset time period to redetermine the vibration amplitude; in response to the re-acquired vibration amplitude not falling back to the preset normal range, and the number of parameter adjustments of the chamfering control parameters not reaching the preset upper limit, adjust the chamfering control parameters again and repeat the steps of continuing chamfering the current wafer according to the adjusted chamfering control parameters and re-acquiring vibration signals within the preset time period; in response to the number of parameter adjustments reaching the preset upper limit, and the redetermined vibration amplitude still not falling back to the preset normal range, control the wafer chamfering equipment to stop chamfering and trigger the grinding wheel of the wafer chamfering equipment to be dressed.

[0221] The above is a schematic scheme of a wafer chamfering processing apparatus provided in this disclosure. The technical solution of this wafer chamfering processing apparatus and the technical solution of the wafer chamfering processing method described above belong to the same concept. For details not described in detail in the technical solution of the wafer chamfering processing apparatus, please refer to the description of the technical solution of the wafer chamfering processing method described above.

[0222] Please refer to Figure 9 , Figure 9 This is a structural block diagram of a computing device provided in this disclosure. In some examples, the computing device 100 can be at least one of devices such as a smartphone, smartwatch, desktop computer, laptop, virtual reality terminal, augmented reality terminal, wireless terminal, and laptop computer. The computing device 100 has communication functions and can access wired or wireless networks. The computing device 100 can refer to one of multiple terminals, and those skilled in the art will understand that the number of such terminals can be more or less. In some examples, the computing device 100 can receive data based on the accessed wired or wireless network. It is understood that the computing device 100 undertakes the calculation and processing work of the technical solution of this disclosure, and this disclosure does not limit it in this respect.

[0223] like Figure 9 As shown, the computing device in this disclosure may include one or more of the following components: processor 1010 and memory 1020.

[0224] Optionally, the processor 1010 connects various parts within the computing device using various interfaces and lines. It executes various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 1020, and by calling data stored in the memory 1020. Optionally, the processor 1010 can be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 1010 can integrate one or more of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), Neural-network Processing Unit (NPU), and baseband chip. Specifically, the CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content displayed on the touchscreen; the NPU implements Artificial Intelligence (AI) functions; and the baseband chip handles wireless communication. It is understandable that the aforementioned baseband chip may not be integrated into the processor 1010, but may be implemented using a separate chip.

[0225] The memory 1020 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 1020 may include a non-transitory computer-readable storage medium. The memory 1020 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 1020 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the various method embodiments described above, etc.; the data storage area may store data created according to the use of the computing device, etc.

[0226] In addition, those skilled in the art will understand that the structure of the computing device shown in the above figures does not constitute a limitation on the computing device. The computing device may include more or fewer components than shown, or combine certain components, or have different component arrangements. For example, the computing device may also include a display screen, camera assembly, microphone, speaker, radio frequency circuit, input unit, sensors (such as accelerometer, angular velocity sensor, light sensor, etc.), audio circuit, Wi-Fi module, power supply, Bluetooth module, etc., which will not be described in detail here.

[0227] This disclosure also provides a computer-readable storage medium storing at least one instruction that is executed by a processor to implement the wafer chamfering method described in the above embodiments.

[0228] This disclosure also provides a computer program product including computer instructions stored in a computer-readable storage medium; a processor of a computing device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computing device to perform the wafer chamfering process described in the above embodiments.

[0229] Those skilled in the art will recognize that the functions described in this disclosure in one or more of the examples above can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transfer of a computer program from one place to another. Storage media can be any available medium accessible to a general-purpose or special-purpose computer.

[0230] It should be noted that the technical solutions described in this disclosure can be combined arbitrarily as long as they do not conflict.

[0231] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A wafer chamfering process, characterized in that, include: Obtain the chamfering control parameters of the wafer chamfering equipment, as well as the chamfering response information of the current wafer during the chamfering process; The chamfering control parameters and the chamfering response information are input into the edge quality prediction model to obtain the edge quality prediction result of the current wafer. The edge quality measurement result of the current wafer is obtained, and the wafer chamfering equipment is controlled to perform chamfering processing control operation based on the deviation between the edge quality prediction result and the edge quality measurement result.

2. The wafer chamfering method according to claim 1, characterized in that, The chamfering control operation includes continuing to process the next wafer according to the chamfering control parameters; The step of controlling the wafer chamfering equipment to perform chamfering processing control operations based on the deviation between the edge quality prediction result and the actual edge quality result includes: In response to the fact that the deviations between the predicted quality indicators in the edge quality prediction results and the corresponding measured quality indicators in the edge quality measurement results are all within the corresponding preset tolerance range, the wafer chamfering equipment is controlled to continue processing the next wafer according to the chamfering processing control parameters.

3. The wafer chamfering method according to claim 1, characterized in that, The chamfering process control operation includes stopping the chamfering process and triggering the dressing of the grinding wheel of the wafer chamfering equipment; If the deviation between any predicted quality index in the edge quality prediction results and the corresponding measured quality index in the edge quality measurement results meets the condition of complete inconsistency, the wafer chamfering equipment is controlled to stop chamfering and the grinding wheel is triggered to be dressed.

4. The wafer chamfering method according to claim 3, characterized in that, The chamfering control operation also includes adjusting the chamfering control parameters; The step of controlling the wafer chamfering equipment to perform chamfering processing control operations based on the deviation between the edge quality prediction result and the actual edge quality result further includes: In response to the fact that the deviations between each of the predicted quality indicators and the corresponding measured quality indicators do not meet the completely inconsistent condition, and at least one of the predicted quality indicators and the corresponding measured quality indicators meet the partially consistent condition, the chamfering control parameters are adjusted.

5. The wafer chamfering method according to claim 1, characterized in that, The method further includes: The chamfering control parameters, the chamfering response information, and the measured edge quality results are associated as a set of training samples, and the training samples are added to the sample dataset. In response to the accumulation of training samples in the sample dataset reaching the model iteration condition, the model parameters of the edge quality prediction model are fine-tuned based on the sample dataset.

6. The wafer chamfering method according to claim 1, characterized in that, The training process of the edge quality prediction model includes: An initial edge quality prediction model is obtained, and a test wafer set is obtained, the test wafer set including at least one test wafer to be processed; The sample chamfering control parameters used by the wafer chamfering equipment to process the test wafer and the sample chamfering response information of the test wafer during the chamfering process are obtained. The sample chamfering processing control parameters and the sample chamfering processing response information are input into the initial edge quality prediction model to obtain the sample edge quality prediction result of the test wafer; Obtain the measured results of the sample edge quality of the test wafer, and determine the test deviation between the predicted results of the sample edge quality and the measured results of the sample edge quality; In response to the test deviation exceeding the calibration tolerance range, the model parameters of the initial edge quality prediction model are adjusted based on the sample chamfering processing control parameters, the sample chamfering processing response information, and the sample edge quality measurement results, until the test deviation does not exceed the calibration tolerance range, thereby obtaining the trained edge quality prediction model.

7. The wafer chamfering method according to claim 6, characterized in that, The process of obtaining the initial edge quality prediction model includes: Multiple sets of initial training samples are obtained, including historical chamfering control parameters, historical chamfering response information, and historical edge quality measurement results obtained after wafer chamfering is completed. Based on the initial training samples, the correlation between the historical chamfering processing control parameters, the historical chamfering processing response information, and the historical edge quality measurement results is fitted to obtain the initial edge quality prediction model.

8. The wafer chamfering method according to claim 1, characterized in that, The wafer chamfering equipment is equipped with a vibration sensing unit mounted on the wafer spindle used to drive wafer rotation, and the method further includes: The vibration sensing unit collects the vibration signal of the current wafer during the chamfering process and determines the vibration amplitude based on the vibration signal. In response to the vibration amplitude exceeding the preset normal range, the chamfering control parameters are adjusted to generate the adjusted chamfering control parameters.

9. The wafer chamfering method according to claim 8, characterized in that, The method further includes: The wafer chamfering equipment is controlled to continue chamfering the current wafer according to the adjusted chamfering processing control parameters, and the vibration signal is re-acquired within a preset time period to redetermine the vibration amplitude. If the reacquired vibration amplitude does not fall back to the preset normal range and the number of times the chamfering control parameters are adjusted does not reach the preset upper limit, the chamfering control parameters are adjusted again and the steps of continuing to chamfer the current wafer according to the adjusted chamfering control parameters and reacquiring the vibration signal within the preset time period are repeated. In response to the parameter adjustment count reaching the preset upper limit and the re-determined vibration amplitude still not returning to the preset normal range, the wafer chamfering equipment is controlled to stop chamfering and the grinding wheel of the wafer chamfering equipment is triggered to be dressed.

10. A wafer chamfering processing apparatus, characterized in that, include: The acquisition module is configured to acquire the chamfering control parameters of the wafer chamfering equipment, as well as the chamfering response information of the current wafer during the chamfering process; The prediction module is configured to input the chamfering control parameters and the chamfering response information into the edge quality prediction model to obtain the edge quality prediction result of the current wafer; The control module is configured to acquire the measured edge quality results of the current wafer, and control the wafer chamfering equipment to perform chamfering control operations based on the deviation between the edge quality prediction results and the measured edge quality results.

11. A computing device, characterized in that, include: Memory, used to store executable instructions; A processor, when executing executable instructions stored in the memory, implements the wafer chamfering method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the wafer chamfering method as described in any one of claims 1-9.

Images