Method, apparatus and medium for processing a silicon wafer

Abstract

This disclosure provides a method, apparatus, and medium for processing silicon wafers. The processing method, applied to a processing equipment, includes: acquiring specification information of the silicon wafer to be processed in response to a tool switching command; identifying and determining a target processing tool from a tool repository based on the specification information, wherein the tool repository includes at least two different types of processing tools; switching the current processing tool currently installed at the working position of the processing equipment to the target processing tool; generating target processing parameters adapted to the specification information and the target processing tool based on a historical parameter database associated with the specification information; and controlling the processing equipment to process the silicon wafer to be processed based on the target processing parameters. This method, through automatic tool replacement and automatic generation of target processing parameters driven by historical parameters, reduces manual intervention and downtime during specification switching, lowers parameter debugging errors, and improves production continuity and processing consistency.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor manufacturing technology, and more specifically, to a method, apparatus and medium for processing silicon wafers. Background Technology

[0002] In the semiconductor processing field, such as in the secondary chamfering process of silicon wafers, processing equipment often needs to frequently switch between batches of different product specifications to handle workpieces of different sizes, requiring switching between different types of processing tools. To ensure processing quality and efficiency, the processing equipment must be able to quickly change grinding wheels when specifications change, accurately set processing parameters, and maintain stable process conditions during batch processing.

[0003] Currently, these goals are typically achieved through manually operated processes. Operators must manually disassemble, install, and calibrate the grinding wheels, and rely on experience to determine processing parameters through repeated trial runs. This approach results in long downtime for tool changes, low efficiency and accuracy in parameter adjustments, and impacts production continuity, yield, and output efficiency. Summary of the Invention

[0004] In view of this, the present invention aims to provide a silicon wafer processing method, equipment and medium to solve the problems of long wheel changing time and low parameter debugging efficiency and consistency caused by manual specification switching in the prior art.

[0005] This invention provides a method for processing silicon wafers, applied to processing equipment, the method comprising: In response to a tool switching command, the specification information of the silicon wafer to be processed is obtained, wherein the tool switching command is generated by the manufacturing execution system or human-machine interface based on the specification change of the silicon wafer to be processed; Based on the specification information, identify and determine the target machining tool from the tool repository, wherein the tool repository includes at least two different models of machining tools; Switch the current machining tool currently installed at the working position of the machining equipment to the target machining tool; Based on a historical parameter database associated with specification information, target machining parameters are generated that are compatible with the specification information and the target machining tool. The processing equipment is controlled to process the silicon wafers based on the target processing parameters.

[0006] In one embodiment, identifying and determining the target machining tool from a tool repository based on specification information includes: Based on specification information, obtain identifier matching data associated with the target machining tool; Obtain the unique electronic identification of all processing tools in the tool repository; All unique electronic identifiers are compared with the identifier matching data to identify the target processing tool; Based on the identification results, the location information of the target processing tool in the tool repository is determined, and the corresponding grabbing and positioning information is generated.

[0007] In one embodiment, switching the current machining tool currently installed at the working position of the machining equipment to the target machining tool includes: Based on a preset disassembly force threshold, the current processing tool is removed. The preset disassembly force threshold is the maximum allowable disassembly force set during the disassembly process to avoid damage to the processing equipment and the current processing tool. Clean the work area based on preset cleaning rules; Based on the captured location information, target processing tools are captured; Obtain the pose deviation between the target machining tool and the working position; Based on the positional deviation and the preset installation force threshold, the target machining tool is installed into the working position without damage. The preset installation force threshold is the target installation force set during the installation process to ensure that the target machining tool and the working position reach a predetermined fit.

[0008] In one embodiment, based on a historical parameter database associated with specification information, target machining parameters adapted to the specification information and the target machining tool are generated, including: The target historical parameter set is obtained from the historical parameter database. The target historical parameter set is a combination of historically successful processing parameters that match the specification information and the target processing tool, and is used as the initial iteration point of the iterative optimization algorithm. Test parameters are automatically generated based on the target's historical parameter set through an iterative optimization algorithm. Based on the test parameters, the trial processing quality data is obtained. The trial processing quality data is the comprehensive processing quality data obtained after the processing equipment performs trial processing based on the test parameters. Based on the processing quality data, the optimization direction of the iterative optimization algorithm is corrected. When the iterative optimization algorithm meets the convergence condition, the optimal combination of processing parameters under the current working condition is output as the target processing parameters.

[0009] In one embodiment, the iterative optimization algorithm includes a Bayesian optimization algorithm, and when the number of iterations of the Bayesian optimization algorithm reaches a preset number, it is determined that the convergence condition is met.

[0010] In one embodiment, in controlling the processing equipment to process the silicon wafer based on target processing parameters, the process includes: an adaptive compensation processing procedure, wherein the adaptive compensation processing procedure includes: Acquire real-time processing parameters, which are data continuously collected during the processing to characterize the current actual processing status of the processing equipment; The real-time processing parameters and the target processing parameters are input into the preset prediction and decision model. Through inference calculation, the predicted deviation trend of the real-time processing parameters relative to the target processing parameters and the corresponding optimal compensation strategy are obtained. When a trend deviation in real-time processing parameters is predicted but has not yet exceeded the preset specification range, the adjustable process parameters of the processing equipment are adjusted based on the optimal compensation strategy to make the real-time processing parameters converge to the target value of the target processing parameters.

[0011] In one embodiment, the predictive decision model is a deep Q-network model pre-trained offline using a historical parameter database, and the processing method further includes: The data of the compensation events and the corresponding compensation effects are sent back to the historical parameter database to serve as training samples for subsequent offline training, thereby continuously optimizing the deep Q-network model.

[0012] In one embodiment, after switching the current machining tool currently installed at the working position of the machining equipment to the target machining tool, the method further includes: determining whether the target machining tool is in a machinable state, wherein determining whether the target machining tool is in a machinable state includes: Obtain dynamic balance test values ​​and runout test values. The dynamic balance test values ​​are used to characterize the residual imbalance after the target machining tool is installed, and the runout test values ​​are used to characterize the radial runout and / or axial runout after the target machining tool is installed. Based on preset rules, it is determined whether the dynamic balance detection value and the runout detection value meet their respective preset requirements. When both the dynamic balance detection and the runout detection meet their respective preset requirements, the target machining tool is confirmed to be in a machinable state.

[0013] In one embodiment, the processing method further includes: If the dynamic balance test value does not meet the corresponding preset requirements, the dynamic balance fine-tuning program will be executed automatically, and then the dynamic balance test value will be re-acquired and judged again until the dynamic balance test value meets the corresponding preset requirements. The dynamic balance fine-tuning program is used to eliminate or compensate for residual imbalance by adaptively adjusting the preset counterweight structure on the target machining tool. If the fluctuation detection value does not meet the corresponding preset requirements, a correction action is performed or a reinstallation process is triggered. Then, the fluctuation detection value is obtained again for re-judgment until the fluctuation detection value meets the corresponding preset requirements.

[0014] In another aspect, the present invention provides a processing apparatus, which includes a processor and a memory, wherein the processor is used to execute instructions stored in the memory to implement the processing method in the above embodiments.

[0015] In another aspect, the present invention provides a computer storage medium storing at least one instruction, which is executed by a processor to implement the processing method in the above embodiments.

[0016] Compared with related technologies, the silicon wafer processing method provided by this invention has the following advantages: The silicon wafer processing method provided by this invention automatically acquires specification information in response to tool switching commands, automatically identifies the target processing tool from the tool repository and performs the switch, and simultaneously generates target processing parameters adapted to the specification information and the target processing tool based on a historical parameter database. This solves the problems of long changeover times and error-prone parameter settings caused by manual downtime for tool replacement and manual parameter adjustment in existing solutions. This solution avoids long downtime caused by manual searching and disassembly, achieving efficient automatic tool replacement; at the same time, it replaces trial-and-error debugging based on experience, allowing parameter settings to quickly converge to reasonable values, thereby effectively improving changeover efficiency, reducing human error, and ensuring consistent processing quality. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart illustrating a silicon wafer processing method according to an embodiment of the present disclosure.

[0018] Figure 2 This is a partial flowchart illustrating a silicon wafer processing method according to an embodiment of the present disclosure.

[0019] Figure 3 This is a partial flowchart illustrating a silicon wafer processing method according to an embodiment of the present disclosure.

[0020] Figure 4 This is a partial flowchart illustrating a silicon wafer processing method according to an embodiment of the present disclosure.

[0021] Figure 5 This is a partial flowchart illustrating a silicon wafer processing method according to an embodiment of the present disclosure.

[0022] Figure 6 This is a schematic flowchart illustrating a silicon wafer processing method according to an embodiment of the present disclosure.

[0023] Figure 7 This is a partial flowchart illustrating a silicon wafer processing method according to an embodiment of the present disclosure.

[0024] Figure 8This is a schematic flowchart illustrating a silicon wafer processing method according to an embodiment of the present disclosure.

[0025] Figure 9 This is a schematic block diagram of a silicon wafer processing apparatus provided in one embodiment of the present disclosure.

[0026] Figure 10 This is a schematic block diagram of another silicon wafer processing apparatus provided in one embodiment of the present disclosure.

[0027] Figure 11 This is a schematic diagram of a processing device provided in one embodiment of the present disclosure. Detailed Implementation

[0028] The technical solutions of the embodiments of this disclosure will be clearly and completely described below 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 without creative effort are within the scope of protection of this disclosure.

[0029] In semiconductor manufacturing, edge grinding 2 (EG2) is a precision machining process. Its core purpose is to perform a second, finer grinding on the edges of the silicon wafer after previous cutting and initial beveling, precisely controlling the final dimensions of the wafer, such as beveling width, groove depth, and surface roughness. This process is crucial for preventing damage to the silicon wafer due to edge chipping or stress concentration during subsequent processes such as transfer, photolithography, and thin film deposition, thereby improving chip yield and reliability.

[0030] In actual production, due to the diverse requirements of downstream chip manufacturing processes for silicon wafer specifications such as diameter, thickness, and chamfer profile, a secondary chamfering machine typically needs to be capable of processing silicon wafers of various specifications. For example, the processing equipment may need to frequently switch between processing standard thickness silicon wafers and ultra-thin silicon wafers, or between processing T-shaped and R-shaped profiles.

[0031] In current practical applications, to cope with such frequent specification changes, the industry generally adopts the practice of having operators manually change grinding wheels and adjust parameters according to the specifications of the silicon wafers to be processed. Specifically, after the processing equipment is shut down, the operator first removes the currently installed grinding wheel, then finds and retrieves the corresponding specification grinding wheel from the tool rack or storage area, manually installs it onto the spindle, and calibrates it. After installation, the operator manually inputs or modifies processing parameters such as cooling water flow rate, feed speed, and vacuum pressure into the equipment controller based on the process card or personal experience, and then performs trial processing to verify whether the parameters are suitable. If the trial processing results do not meet the quality requirements, the parameters are further adjusted and trial processing is repeated until it is qualified. This solution is widely used mainly because it has relatively low requirements for the degree of equipment automation and can adapt to the production needs of changes in product specifications.

[0032] However, when silicon wafer specifications change frequently and there are many processing specifications, the above solution does not perform ideally in actual production. A fundamental contradiction lies in the fact that this solution relies on manual operation and trial-and-error debugging to adapt to the flexibility of multi-specification processing. However, the low efficiency of manual operation and the repeated iterations of the debugging process make it difficult to balance the time spent on changeover with the consistency of parameter settings. For example, when switching from a T-300 type grinding wheel to an R type grinding wheel, the entire process of shutdown, disassembly, location, installation, and calibration is usually time-consuming. Subsequent parameter debugging requires multiple trial processing operations, which not only wastes time and materials but may also lead to defects in the first piece or small batch processing due to parameter deviations during the debugging stage. Specifically, this manifests as low changeover efficiency and large fluctuations in processing quality.

[0033] In view of the above problems, one embodiment of the present invention provides a method for processing silicon wafers. This method is applied to processing equipment and can be executed by computer equipment (e.g., a server or user terminal). Optionally, the computer equipment can be communicatively connected to the processing equipment or integrated into the processing equipment. Figure 1 As shown, the processing method for this silicon wafer may include the following: 100. In response to a tool switching command, obtain the specification information of the silicon wafer to be processed, wherein the tool switching command is generated by the manufacturing execution system or human-machine interface based on the specification change of the silicon wafer to be processed.

[0034] For example, the tool switching command is triggered when the specifications of the silicon wafer to be processed change. When the upper-level Manufacturing Execution System (MES) determines that the next batch of silicon wafers to be processed needs to be of different specifications based on the production scheduling plan, the system will automatically identify the specification change of the silicon wafer to be processed and generate a corresponding tool switching command based on this, which will be sent to the control system of this processing equipment. As another implementation, when the operator learns through the Human-Machine Interface (HMI) that a change in production specifications is required, they can also confirm the specification change on the interface, thereby triggering the software to generate a tool switching command. After responding to and receiving the command, the control system of the processing equipment will parse and obtain the specification information of the silicon wafer to be processed corresponding to the changed specifications, providing a basis for subsequent automatic tool replacement and parameter configuration.

[0035] Optionally, specification information refers to data used to describe the type of silicon wafer product to be processed and the processing requirements. It may include product model codes, such as T-300, T-395, and R-type; silicon wafer geometric parameters, such as diameter and thickness; target chamfer profile parameters, such as chamfer width, groove depth, and material grade; or combinations thereof.

[0036] It is understood that step 100 provides the most fundamental input condition for the processing method of this embodiment, that is, it determines what specifications of product to be processed, and all subsequent automated operations are based on this information. The method of generating tool switching instructions is not limited to the above method; it can also be automatically triggered by the production scheduling module inside the processing equipment according to a preset plan.

[0037] 200. Based on the specifications, identify and determine the target machining tool from the tool repository, wherein the tool repository includes at least two different models of machining tools.

[0038] Optionally, the tool storage unit includes any storage device capable of accommodating at least two different types of machining tools and supporting automatic identification and retrieval. For example, the tool storage unit includes, but is not limited to, a rotary multi-station grinding wheel storage bin, a linear tool array, an automatic cutter head, or a combination thereof; each storage location may be integrated with identification and positioning components, such as an RFID reader, a visual positioning target, etc.

[0039] Optionally, processing tools refer to consumables or components used to physically process silicon wafers, such as grinding wheels, cutting tools, polishing heads, or combinations thereof; the tools have a unique identifier and can be automatically identified and replaced.

[0040] Following the example above, after obtaining the specification information from step 100, the control system sends a query command to the tool repository to identify and determine the target machining tool. This identification process utilizes the unique identifier corresponding to each machining tool and preset identifier matching data, thereby ensuring that the target machining tool is correctly selected from at least two different models of machining tools. Since this step does not require manual searching and identification, the risk of human error in installation is avoided.

[0041] It is understandable that the storage locations of the aforementioned tool storage units can have active temperature control and vibration isolation functions to ensure that the spare grinding wheels are in optimal standby condition; in addition to being used for the EG2 secondary chamfering process, the processing equipment can also be used for other semiconductor wafer processing processes that require frequent changes of grinding wheels or tools; the processing tools can be different types of grinding wheels, including but not limited to T-300 type grinding wheels, T-395 type grinding wheels and R type grinding wheels, etc.

[0042] 300. Switch the current machining tool currently installed at the working position of the machining equipment to the target machining tool.

[0043] Following the example above, after determining the target machining tool, the control system controls the changing mechanism to perform an automatic changing process, unloading the current machining tool currently installed at the working position and returning it to an empty slot in the tool storage container, or placing it directly in a predetermined position, and then grabbing the target machining tool and installing it at the working position.

[0044] Optionally, the working position is the part on the spindle of the machining equipment used to mount the current machining tool and perform grinding.

[0045] Optionally, the changing mechanism may include a multi-axis robotic arm with an end effector integrated with a force sensor and a vision alignment system for performing the disassembly and installation of machining tools.

[0046] Understandably, the control system can be an industrial computer or an embedded controller, which communicates with the various actuators of the processing equipment to receive instructions, process data, and control the execution of various actions. Besides using a robotic arm, the changing mechanism can also employ other automated actuators such as a tool changer or a gantry robot.

[0047] 400. Based on the historical parameter database associated with the specification information, generate target machining parameters that are compatible with the specification information and the target machining tool.

[0048] Optionally, a historical parameter database refers to a data set that stores the processing parameters used in previous successful processing processes and their relationship with product specifications and processing tools. For example, a historical parameter database includes, but is not limited to, a recipe database, a recipe library, a related data table, or a combination thereof; this database is used to support parameter matching and model training.

[0049] Optionally, the target processing parameters refer to the set of equipment operating parameters generated or invoked to complete the current silicon wafer processing task. For example, the target processing parameters include, but are not limited to, cooling water flow rate, vacuum pressure, spindle speed, feed rate, or combinations thereof; the parameters should be compatible with the specification information and the target processing tool.

[0050] Following the example above, after automatically switching grinding wheels, the system generates target processing parameters based on a historical parameter database associated with the current specifications. This database pre-constructs a mapping relationship between product specifications and formulas. The system automatically matches and calls the corresponding historical successful parameters according to the information of the product to be processed, obtaining the optimal combination of processing parameters under the current working conditions as the target processing parameters. Through the above fully automated changeover and parameter self-adjustment process, changeover time is reduced to minutes, the adjustment process requires no manual intervention, and a qualified first piece of product can be obtained after one adjustment, thereby eliminating manual adjustment errors and ensuring the consistency of processing quality.

[0051] 500. Control the processing equipment to process the silicon wafers to be processed based on the target processing parameters.

[0052] Using the example above, the processing equipment performs grinding or chamfering on the silicon wafer to be processed under the control of the target processing parameters.

[0053] Based on the above embodiments, this disclosure describes a specific implementation of identifying and determining a target machining tool from a tool repository according to specification information, so as to improve the accuracy of tool identification.

[0054] In some implementations, such as Figure 2 As shown, a specific implementation of step 200, "identifying and determining the target machining tool from the tool repository based on specification information," may include the following: 210. Based on specification information, obtain the identifier matching data associated with the target machining tool.

[0055] Optionally, the identification matching data refers to the data used to compare with electronic identifications in the tool repository to determine the target tool. For example, it may include, but is not limited to, the RFID code corresponding to the target tool model, a specification mapping table, a pre-stored pairing table, or a combination thereof.

[0056] 220. Obtain the unique electronic identification of all processing tools in the tool repository.

[0057] Optionally, a unique electronic identity refers to an electronic information carrier that enables each processing tool to be uniquely identified and distinguished by the system, such as, but not limited to, RFID tags, QR codes, barcodes, magnetic codes, near-field communication chips, or combinations thereof.

[0058] Understandably, a unique electronic identifier can also be a magnetic code bar or an optical feature pattern, and the corresponding identification method will change accordingly.

[0059] 230. Compare all unique electronic identifiers with the identifier matching data to identify the target processing tool.

[0060] Following the example above, after acquiring the unique electronic identification of all processing tools in the tool storage repository (e.g., RFID tag codes) in step 120, the control system iteratively compares these identifications one by one with the identification matching data associated with the target tool model obtained in step 110 (e.g., the RFID code corresponding to that model). When the electronic identification of a processing tool completely matches the identification matching data, the system determines that the tool is the target processing tool that meets the current processing specifications. This comparison process is executed at high speed by the control unit's processor, exhibiting high efficiency and accuracy.

[0061] Understandably, the core logic of this comparison mechanism is to match unknown information with known information. For example, it matches the identifier of the target tool with the identifier of the tools in the library, rather than blindly searching for the target tool in the library. If multiple tools of the same model exist in the library, the system will accurately distinguish them based on the tool's unique identifier, such as a globally unique RFID serial number, or by combining preset priority rules, such as the fewest recent uses or the longest remaining lifespan, to further determine the unique target tool. Furthermore, the location identification method for the tool storage library can not be limited to electronic tags; it can also combine mechanical positioning pins or laser ranging to determine the location of the tools in the storage area.

[0062] 240. Based on the identification results, determine the location information of the target processing tool in the tool repository and generate the corresponding grabbing and positioning information.

[0063] Optionally, the grasping and positioning information refers to the spatial location data required to guide the changing mechanism to accurately obtain the target tool, such as including but not limited to the three-dimensional coordinates of the storage location, the attitude angle, the grasping point offset, or a combination thereof; this information is generated based on the identification results of the tool storage location and the tool's identity.

[0064] For example, after obtaining the specification information from step 100, the control system first queries the internal mapping relationship based on this information to determine the model information of the target processing tool that matches the specification, such as an R-type grinding wheel. Then, based on this model information, the system retrieves the associated identification matching data, such as the RFID code of the corresponding grinding wheel model, from the database. The control system then sends a query command to the tool storage warehouse, where RFID readers read the unique electronic identification of all processing tools stored in all storage locations. The control system then iterates through and compares all the read identifications with the identification matching data. When the electronic identification of a processing tool matches the identification matching data, the tool is identified as the target processing tool. Finally, based on the identification result, the control system obtains the storage location number and spatial coordinates of the target processing tool, generating grabbing positioning information for subsequent grabbing.

[0065] In this way, the processing method can accurately locate the target processing tool through comparison and instantly obtain its storage location number and specific spatial coordinates within that location. The changing mechanism can then accurately grasp the tool based on this grabbing and positioning information. Compared to manually checking nameplates and searching for tools, this method eliminates the possibility of human error and significantly shortens the identification and positioning time, laying the foundation for the reliability of fully automated changing.

[0066] In some implementations, such as Figure 3 As shown, the specific implementation of "switching the current machining tool currently installed at the working position of the machining equipment to the target machining tool" in step 300 may include the following: 310. Based on a preset disassembly force threshold, remove the current processing tool. The preset disassembly force threshold is the maximum allowable disassembly force set during the disassembly process to avoid damage to the processing equipment and the current processing tool.

[0067] Optionally, the preset disassembly force threshold can be determined in advance based on factors such as the interface type between the machining tool and the spindle, the preload design, and the material strength.

[0068] For example, when a machining tool is locked using a tapered fit and pull studs, the disassembly process requires first applying a certain axial tensile force to overcome the locking force, followed by applying a loosening torque. To avoid overload damage to the spindle bearings or tool shank, a dynamic upper limit is set for the disassembly force threshold. Force sensors at the end of the changing mechanism monitor forces and torques in all directions in real time. Once the disassembly force exceeds this threshold, the current action is immediately stopped or the posture is adjusted to complete the disassembly smoothly.

[0069] 320. Clean the work area based on preset cleaning rules.

[0070] Optionally, high-pressure gas purging or vacuum cleaning can be used to remove shavings, cutting fluid residue, etc. from the working area, such as the interface surface of the spindle, to avoid impurities affecting the installation accuracy.

[0071] 330. Based on the captured location information, capture target processing tools.

[0072] 340. Obtain the pose deviation between the target machining tool and the working position.

[0073] Using the example above, after the system controls the changing mechanism to grasp the target processing tool, it obtains the pose deviation between the target processing tool and the working position through the vision alignment system, specifically including translational deviation and angular deviation.

[0074] 350. Based on the positional deviation and the preset installation force threshold, the target machining tool is installed into the working position without damage. The preset installation force threshold is the target installation force set during the installation process to ensure that the target machining tool and the working position reach a predetermined fit state.

[0075] Using the example above, the system adjusts the orientation of the changing mechanism based on the deviation, compensates for it, and then performs the installation action. Through these measures, the changing process not only achieves automation without human intervention but also reduces the risk of mechanical damage to an extremely low level. High-precision repeatability can be achieved without manual calibration after installation.

[0076] For example, based on the gripping and positioning information generated in step 200, the control system issues a switching command to the changing mechanism, such as a multi-axis robotic arm. This step specifically includes a series of ordered sub-steps: First, a flexible disassembly is performed. The changing mechanism moves to the working position, such as the spindle interface, and uses a force-controlled compliant algorithm integrated at its end to slowly and smoothly remove the current machining tool, with a preset disassembly force threshold as the upper limit. During this process, the contact force is monitored in real time, and if it exceeds the threshold, the posture is automatically adjusted to prevent damage.

[0077] After the old tools are removed, the control system controls cleaning devices, such as high-pressure gas nozzles, to blow or vacuum the spindle interface to remove residual shavings and coolant, ensuring the cleanliness of the interface for subsequent installations.

[0078] After cleaning, the changing mechanism accurately retrieves the target processing tool from the tool storage repository based on the grasping and positioning information.

[0079] After the object is grasped, the micro-vision alignment system at the end of the changing mechanism immediately works, accurately measuring the pose deviation between the target processing tool and / or the spindle interface by identifying the visual target on the target processing tool and / or the spindle interface.

[0080] Finally, based on the aforementioned positional deviation, the control system adjusts the posture of the changing mechanism and uses force sensors to perform compliant installation through force-position hybrid control with a preset installation force threshold as the target. This ensures that the target machining tool is installed onto the spindle interface without damage and with high precision, achieving the predetermined fit, such as a tight fit between the conical surfaces.

[0081] It is understandable that step 300 is not simply switching the old machining tool to the new target machining tool, but rather a precision mechatronics process integrating force sensing, vision, and compliance control technologies. It solves the two major problems commonly found in traditional manual tool changeovers: susceptibility to damage and poor precision, laying a solid hardware foundation for subsequent precision machining. Furthermore, this embodiment does not limit the specific values ​​of the preset disassembly force threshold and the preset installation force threshold; these can be selected based on the design strength of the spindle interface and the tool type, etc., and will not be elaborated upon here.

[0082] In some implementations, such as Figure 4 As shown, the specific implementation of "generating target machining parameters adapted to the specification information and the target machining tool based on the historical parameter database associated with the specification information" in step 400 may include the following: 410. Obtain the target historical parameter set from the historical parameter database. The target historical parameter set is a combination of historically successful processing parameters that match the specification information and the target processing tool, and serves as the initial iteration point for the iterative optimization algorithm.

[0083] Optionally, the target historical parameter set may include core process parameters such as machining tool rotation speed, feed rate, coolant flow rate, vacuum pressure, grinding depth or transverse feed rate, as well as corresponding processing quality evaluation indicators, such as silicon wafer surface roughness, dimensional tolerance, and edge chipping rate. This historical success data is accumulated from previous mass production records and has high reference value.

[0084] 420. Based on the target historical parameter set, test parameters are automatically generated through an iterative optimization algorithm.

[0085] Continuing with the example above, the iterative optimization algorithm uses the target historical parameter set from step 410 as the starting search region, which avoids a large number of invalid attempts caused by starting the search from random points and significantly speeds up the convergence. In each iteration, the algorithm generates one or more sets of test parameters based on the current surrogate model.

[0086] In at least one embodiment, the iterative optimization algorithm includes a Bayesian optimization algorithm. When the number of iterations of the Bayesian optimization algorithm reaches a preset number, it is determined that a convergence condition is met. A convergence condition refers to the criteria used to determine whether the iterative optimization algorithm can terminate the search and output the current optimal parameters. For example, a convergence condition includes, but is not limited to, the number of iterations reaching a preset number, the change in the objective function value being less than a threshold, or a combination thereof.

[0087] For example, the Bayesian optimization algorithm uses the historical successful parameter set as the initial sampling point and aims to achieve the highest comprehensive evaluation of processing quality, which can obtain a better parameter combination in a very short number of iterations.

[0088] For example, the preset number of iterations for the Bayesian optimization algorithm can be set according to the actual production cycle requirements and the size of the parameter search space, and is generally set between 3 and 10. For example, the preset number of iterations can take the following specific values: 3, 4, 5, 6, 7, 8, 9, and 10.

[0089] Understandably, besides Bayesian optimization algorithms, iterative optimization algorithms can also be replaced by genetic algorithms. This is suitable for scenarios with a huge parameter search space and multiple local optima in the objective function, and is appropriate for scenarios where the first-piece debugging time requirement is not so stringent. Additionally, iterative optimization algorithms can also use particle swarm optimization, simulated annealing, and other algorithms.

[0090] 430. Based on the test parameters, obtain trial processing quality data, wherein the trial processing quality data is the comprehensive processing quality data obtained after the processing equipment performs trial processing based on the test parameters.

[0091] Following the example above, the system controls the processing equipment to perform trial processing and measure quality indicators, generating trial processing quality data. This new data is then used to update the surrogate model, re-evaluate the potential optimal solution in the parameter space, and determine the next set of test parameters. This process is repeated until convergence is achieved.

[0092] 440. Based on the processing quality data, correct the optimization direction of the iterative optimization algorithm. When the iterative optimization algorithm meets the convergence condition, output the optimal combination of processing parameters under the current working condition as the target processing parameters.

[0093] Understandably, the scheme disclosed in step 400, which uses a historical parameter database combined with an iterative optimization algorithm to obtain target processing parameters, replaces the traditional approach of relying on personal experience and blind trial and error for debugging. This transforms the parameter debugging process from a high-cost, low-consistency activity into an efficient, deterministic, and high-quality automated process.

[0094] For example, after the tool replacement is successfully completed in step 300, the control system immediately enters the intelligent parameter debugging stage.

[0095] First, the control system accesses its historical parameter database and, based on the specification information obtained in step 100 and the target machining tool identified in step 200, retrieves the matching parameter combinations that have been successfully machined in the past, as the target historical parameter set. This parameter set includes core process parameters such as spindle speed, feed rate, cooling water flow rate, and vacuum pressure, and serves as the initial iteration point for the iterative optimization algorithm.

[0096] Next, the intelligent debugging engine in the control system, for example, runs a Bayesian optimization algorithm starting from the initial iteration point to automatically generate the first set of test parameters and sends them to the processing equipment, instructing it to perform a trial processing.

[0097] After trial processing is completed, the online measurement system on the equipment, such as a profilometer, immediately provides feedback on the trial processing quality data, such as the surface roughness of the workpiece, chamfer width, and groove depth. The control system feeds this new data back to the iterative optimization algorithm, which then modifies its internal probabilistic surrogate model based on the feedback and generates a new set of better test parameters for processing. This step is repeated until the algorithm meets a predetermined convergence condition, such as reaching a preset number of iterations, such as 5. After convergence, the algorithm outputs the optimal parameter combination under the current operating conditions, which is the target processing parameter.

[0098] For example, if the preset number of times is set to 5, then from the first trial processing to the final output of the target processing parameters, only 5 trial processing and iterations need to be performed in total.

[0099] In step 500, the processing equipment initiates batch processing operations under the control of the target processing parameters. However, during continuous processing, factors such as equipment temperature changes, tool wear, and environmental fluctuations may cause the processing status to drift, resulting in the actual processing parameters deviating from the preset target values. If such drifts are not detected and corrected in a timely manner, they may lead to batch-specific quality defects, thereby affecting product yield and production efficiency.

[0100] To address the aforementioned issues, a passive monitoring approach is typically employed, where the machine is only stopped and manually adjusted when processing parameters exceed preset specifications. This method struggles to effectively intervene in the early stages of parameter drift, failing to prevent the drift from escalating into out-of-tolerance issues and thus preserving a batch of defective products.

[0101] Based on this, in some embodiments, an adaptive compensation processing procedure is integrated into step 500, "in controlling the processing equipment to process the silicon wafer to be processed based on the target processing parameters." For example... Figure 5 As shown, the specific implementation of the "adaptive compensation processing procedure" in step 500 may include the following: 510. Obtain real-time processing parameters, wherein the real-time processing parameters are data continuously collected during the processing process to characterize the current actual processing status of the processing equipment.

[0102] For example, the control system continuously collects data, i.e. real-time processing parameters, that characterizes the current actual processing status of the processing equipment through a sensor network deployed on the processing equipment.

[0103] Optionally, real-time machining parameters may include spindle load current, grinding force, coolant temperature, actual feed rate, and machining dimensions measured online, such as chamfer width and groove depth. These parameters are acquired in real time and transmitted to the control unit via a sensor network.

[0104] 520. Input the real-time processing parameters and the target processing parameters into the preset prediction decision model, and obtain the predicted deviation trend of the real-time processing parameters relative to the target processing parameters and the corresponding optimal compensation strategy through inference calculation.

[0105] Following the example above, the control system inputs the real-time machining parameters acquired in step 510 and the target machining parameters generated in step 400 into a preset predictive decision model. This model uses its internally learned system dynamics to infer and predict future deviations in the machining parameters. For example, when the model detects a gradual increase in spindle current accompanied by a slight increase in the grinding zone temperature, it can predict that the chamfer dimensions will gradually increase.

[0106] Optionally, the preset predictive decision model refers to a computational model that infers the future parameter deviation trend based on real-time processed data and target parameters and outputs a compensation strategy. This model can be pre-trained offline in advance and fine-tuned online during processing.

[0107] 530. When it is predicted that the real-time processing parameters will deviate from the preset specification range, the adjustable process parameters of the processing equipment will be adjusted based on the optimal compensation strategy so that the real-time processing parameters converge to the target value of the target processing parameters.

[0108] Continuing with the example above, before the parameters exceed the preset upper and lower limits, the model outputs the corresponding optimal compensation strategy, such as fine-tuning the feed rate or adjusting the coolant flow rate. The control system then issues commands based on this strategy to actively correct the adjustable process parameters of the machining equipment, thereby pulling the drifting real-time machining parameters back to near the target value. This approach transforms the passive control mode of traditional solutions, which involves stopping the machine after machining parameters exceed tolerances, into predictive active control that proactively maintains the stability of machining parameters during the machining process. This reduces scrap and downtime losses caused by tolerances, making it particularly suitable for long-term, continuous batch production scenarios.

[0109] For example, after the target processing parameters are output in step 400, the control system sends these parameters to the PLC of the processing equipment, officially starting the batch chamfering operation of the current batch of silicon wafers to be processed. During batch processing, this method also integrates and executes an adaptive compensation step to proactively and preventively maintain the stability of processing quality.

[0110] Specifically, the adaptive compensation step includes: during the machining process, continuously collecting real-time machining parameters through various sensors, which reflect the current actual machining status such as spindle load, grinding force, and online measurement dimensions.

[0111] Subsequently, the control system inputs these real-time parameters, along with the target processing parameters generated in step 400, into a predictive decision model, such as a deep Q-network model pre-trained offline using a historical parameter database. This model, through inference calculations, predicts the deviation trend of the real-time parameters relative to the target parameters and generates the corresponding optimal compensation strategy.

[0112] When the model successfully predicts that the real-time parameters will deviate from the preset specification range, but have not yet exceeded the preset specification range, for example, the size is still within ±5μm of the target value, the control system immediately issues a command to adjust the adjustable process parameters of the processing equipment according to the compensation strategy, such as fine-tuning the feed speed or cooling flow rate, thereby pulling the outward drifting real-time parameters back and converging them towards the target value, realizing active control of real-time adjustment and optimization of real-time parameters during the processing.

[0113] For example, during continuous processing, if the model detects that a real-time parameter, such as the groove depth, is slowly decreasing at a rate of 0.1 μm / piece, even though the current value is still 238 μm, which is within the specification range of 240 ± 5 μm, the model predicts that without intervention, it will exceed the lower limit of the specification after 10 pieces. In this case, the model automatically outputs a compensation strategy, such as increasing the Z-axis feed by 0.01 μm, and executes it immediately, thereby eliminating the potential problem before the deviation occurs.

[0114] Understandably, the specific implementation of the predictive decision-making model is not unique. In addition to the deep Q-network model, an adaptive fuzzy PID controller can also be used. This controller does not require a complex mathematical model, achieves online self-tuning through a fuzzy rule base, and has good dynamic response and anti-interference ability for nonlinear and time-varying grinding processes, while also having a lower implementation cost. The specific model selection can be flexibly determined based on the equipment's computing resources, the richness of historical data, and the control accuracy requirements.

[0115] In at least one embodiment of this disclosure, the predictive decision model is a deep Q-network model pre-trained offline using a historical parameter database. The processing method further includes: feeding back the data of the compensation events and the corresponding compensation effects to the historical parameter database as training samples for subsequent offline training, thereby continuously optimizing the deep Q-network model.

[0116] For example, in the offline pre-training phase, sequence data consisting of processing states, compensation actions, and compensation effect evaluations are extracted from historical processing data and used as training samples for initial learning. By analyzing these sequences, the model learns the mapping relationship between the current processing state and the optimal compensation action, thereby gaining preliminary prediction and decision-making capabilities. During online processing, the deep Q-network model selects the optimal compensation action based on the current state, such as the deviation or trend between real-time parameters and target parameters.

[0117] After performing the compensation action, the deep Q-network model continuously monitors the new state and actual effect, such as whether the processing parameters have been successfully adjusted back to the target range, and records this complete intervention process as a compensation event. The data for this compensation event, including at least the state before drift, the executed compensation action, and its effect, is fed back to the historical parameter database as training samples for subsequent offline training, used for retraining or incremental learning of the model. Through this closed-loop feedback mechanism, the model can continuously adapt to the time-varying characteristics of the device and environmental fluctuations, achieving self-learning and self-optimization, thereby gradually improving its prediction accuracy and compensation success rate as the processing continues.

[0118] As can be understood, a Deep Q-Network model refers to a reinforcement learning model that approximates an action-value function using a deep neural network, and is used to learn parameter compensation strategies from historical data. The input to this model may include device state and target recipe, and the output is a specific compensation action. For example, a Deep Q-Network model can be, but is not limited to, a DQN, Double DQN, or Dueling DQN model with convolutional or fully connected layers.

[0119] In at least one embodiment of this disclosure, after switching the current processing tool currently installed at the working position of the processing equipment to the target processing tool, the processing method further includes: determining whether the target processing tool is in a processable state.

[0120] Based on the above embodiments, in some embodiments, such as Figure 6 As shown, the processing method for this silicon wafer may include the following: 10. In response to a tool switching command, obtain the specification information of the silicon wafer to be processed, wherein the tool switching command is generated by the manufacturing execution system or human-machine interface based on the specification change of the silicon wafer to be processed.

[0121] 20. Based on the specifications, identify and determine the target machining tool from the tool repository, which includes at least two different models of machining tools.

[0122] 30. Switch the current machining tool currently installed at the working position of the machining equipment to the target machining tool.

[0123] 40. Determine whether the target machining tool is in a machinable state.

[0124] 50. Based on the historical parameter database associated with the specification information, generate target machining parameters that are compatible with the specification information and the target machining tool.

[0125] 60. After confirming that the target processing tool is in a processable state, control the processing equipment to process the silicon wafer to be processed based on the target processing parameters.

[0126] It should be noted that the specific implementation of this processing method is not limited to the above embodiments, and may also include other implementation methods. For example, the order of steps 40 and 50 may be interchanged. In addition, the specific implementation methods in steps 10 to 60 can refer to the above embodiments, and will not be repeated here.

[0127] Based on the above embodiments, in some embodiments, such as Figure 7 As shown, the specific implementation of step 40, "determining whether the target machining tool is in a machinable state," may include the following: 41. Obtain dynamic balance test values ​​and runout test values. The dynamic balance test values ​​are used to characterize the residual imbalance after the target machining tool is installed, and the runout test values ​​are used to characterize the radial runout and / or axial runout after the target machining tool is installed.

[0128] For example, dynamic balance test values ​​can be obtained using a vibration sensor mounted on the spindle or a dedicated dynamic balancing instrument, and are typically expressed as unbalance (in g·mm) or residual unbalance level. Runout test values ​​are obtained by measuring the change in the outer circle or end face of the tool relative to a reference axis using a non-contact displacement sensor, such as a laser displacement sensor, during tool rotation; radial runout is typically expressed in micrometers (μm).

[0129] Optionally, the dynamic balance test values ​​may include, but are not limited to, vibration amplitude, unbalanced mass moment, unbalanced angle, or combinations thereof.

[0130] Optionally, the runout detection value may include, but is not limited to, radial runout, axial runout, end face runout, or a combination thereof.

[0131] 42. Based on preset rules, determine whether the dynamic balance detection value and the runout detection value meet their respective preset requirements. When both the dynamic balance detection and the runout detection meet their respective preset requirements, confirm that the target machining tool is in a machinable state.

[0132] For example, when both dynamic balancing and runout testing meet their corresponding preset requirements, the target machining tool is confirmed to be in a machinable state. The preset requirements can be determined based on the machining accuracy level and tool specifications. For instance, for high-precision chamfering, the residual imbalance can be set to no more than 0.5 g·mm / kg, and the radial runout to no more than 5 μm. The system only allows subsequent trial or batch machining when both tests meet the requirements; otherwise, a correction procedure will be initiated. Setting appropriate thresholds can strike a balance between ensuring machining quality and avoiding unnecessary adjustments.

[0133] Understandably, setting overly strict preset requirements may lead to unnecessary multiple adjustments. Increasing non-machining time can balance machining quality and production efficiency by selecting an appropriate threshold based on actual machining accuracy requirements. If the dynamic balance test value or runout test value exceeds its respective upper limit, it may indicate that the tool is not installed correctly or that the tool itself has quality defects. Forcing machining will cause increased vibration, deterioration of surface quality, or even damage to the spindle.

[0134] Based on the above embodiments, in at least one embodiment of this disclosure, such as Figure 8 As shown, the processing method also includes the following: 42a. If the dynamic balance test value does not meet the corresponding preset requirements, the dynamic balance fine-tuning program is automatically executed, and then the dynamic balance test value is re-acquired for re-judgment until the dynamic balance test value meets the corresponding preset requirements. The dynamic balance fine-tuning program is used to eliminate or compensate for residual imbalance by adaptively adjusting the preset counterweight structure on the target machining tool.

[0135] Optionally, the preset counterweight structure can be multiple adjustable balance blocks installed at the end of the tool or circumferentially, or the mass distribution can be adjusted by removing a small amount of material. The dynamic balancing fine-tuning program automatically calculates the counterweight position and adjustment amount to be adjusted based on the magnitude and phase of the initially measured residual imbalance, drives the adjustment mechanism to execute, and then performs detection and verification. This closed-loop iteration continues until the requirements are met.

[0136] 42b. If the fluctuation detection value does not meet the corresponding preset requirements, a correction action is performed or a reinstallation process is triggered. Then, the fluctuation detection value is re-acquired and judged again until the fluctuation detection value meets the corresponding preset requirements.

[0137] Optionally, the calibration process may include readjusting the clamping force, lightly tapping the positioning surface, or re-executing the installation alignment procedure; if the calibration is still not satisfactory, a reinstallation process is triggered, which involves removing the target machining tool, cleaning it, and then reinstalling it.

[0138] Thus, after the target machining tool is installed in the working position, the above-mentioned dynamic balance test value and runout test value are used to detect and adjust the installation status of the target machining tool to ensure that the installation status of the target machining tool meets the machining requirements and avoids machining defects caused by installation deviation.

[0139] As can be seen from the above description, the silicon wafer processing method forms a complete closed-loop automated processing solution through tool automatic identification and positioning, force-controlled compliant replacement, intelligent parameter optimization based on historical data, predictive adaptive compensation, and automatic detection and fine-tuning after installation. It is especially suitable for the secondary chamfering scenario of silicon wafers with multiple varieties, small batches, and frequent production changes.

[0140] To more clearly demonstrate how the processing method disclosed herein operates and creates value in a real-world industrial environment, a dynamic description is provided using a typical scenario from processing T-300 type_A specification to R type_B specification in the EG2 secondary chamfering process of a semiconductor silicon wafer manufacturing company, where frequent product specification changes are required. In related technologies, each product changeover necessitates machine shutdown and manual grinding wheel replacement and parameter adjustment by skilled technicians, taking several hours in total. Furthermore, parameter settings are prone to deviation due to differences in human experience. Figure 8 As shown, the following is the specific execution flow after using the processing method of the present invention.

[0141] First, the specification switching instruction is received. Specifically, the Manufacturing Execution System (MES) issues a tool switching instruction to the processing equipment based on the production schedule, requiring the current T-300 grinding wheel to be replaced with an R-type grinding wheel, and R-type_B specification silicon wafers to be produced. The manufacturing unit then obtains the specification information of the silicon wafer to be processed, including the product code and chamfer contour parameters.

[0142] Secondly, the target grinding wheel is located and identified. Specifically, based on this specification information, the system automatically identifies the target processing tool from the rotary multi-station grinding wheel bin. The system retrieves the identification matching data associated with the R-type grinding wheel model, scans the electronic identification of all grinding wheels one by one using the RFID reader integrated into the storage location, accurately locates the storage location of the R-type grinding wheel through comparison, and calculates its spatial coordinates by combining with the visual positioning target to generate grab positioning information. This process is completed within seconds, fundamentally eliminating the risk of human error in loading and unloading.

[0143] Next, the robotic arm removes the current grinding wheel. Specifically, the robotic arm, equipped with a six-dimensional force sensor and a micro-vision alignment system at the system's control end, begins the automatic wheel-changing operation. The robotic arm first moves to the working position and removes the current T-300 grinding wheel in a force-controlled compliant manner based on a preset disassembly force threshold. The force sensor monitors the force in all directions in real time to ensure no damage to the spindle interface. The removed grinding wheel is then returned to its idle storage location.

[0144] Subsequently, the robotic arm installs the target grinding wheel. The robotic arm uses a high-pressure gas nozzle to clean the spindle interface. Then, based on the gripping positioning information, it grips the R-type grinding wheel, and the micro-vision system measures the positional deviation between the grinding wheel and the spindle interface in real time. The robotic arm adjusts its posture accordingly and performs force-controlled compliant installation under the control of a preset installation force threshold, ensuring the grinding wheel smoothly fits to the pre-tightened state and completing the installation without damage. The entire replacement process can be completed within 3 minutes. After installation, the system automatically performs dynamic balancing and runout detection on the target grinding wheel, confirming that it is in a machinable state before allowing it to proceed to the next process.

[0145] Next, after confirming that the installed target grinding wheel is ready for machining, the system enters the intelligent parameter debugging phase. Specifically, the control unit accesses the historical parameter database and retrieves the matching target historical parameter set based on the R-type grinding wheel model. This parameter set includes core process parameters such as spindle speed, feed rate, and coolant flow rate, and serves as the initial sampling point for the Bayesian optimization algorithm.

[0146] Subsequently, the intelligent debugging engine optimizes parameters based on historical data to obtain target processing parameters. Specifically, the engine aims for the optimal comprehensive processing quality evaluation value, generates the first set of test parameters based on a Gaussian process model, and controls the equipment to perform trial processing. The online measurement system immediately provides feedback on the processing quality data of this trial processing, and the algorithm updates the proxy model accordingly, corrects the optimization direction, and generates the next set of test parameters. This process is iterative. After a preset number of iterations, such as 5, the algorithm outputs the optimal parameter combination under the current working conditions and automatically distributes it. The system then executes the first piece processing and inspection to obtain key quality indicators.

[0147] Finally, the system automatically sends out the target machining parameters and executes the first piece machining. The system then makes judgments and cycles based on the inspection results. If the inspection is successful: the first piece is successfully debugged on the first attempt, and the system immediately enters the batch processing stage. During this process, the adaptive compensation system runs continuously, using a deep Q-network model to monitor processing parameters in real time, actively suppressing drift, and ensuring stable processing quality. Specifically, after the first piece is processed, all key quality indicators obtained by the online measurement system, such as chamfer width, groove depth, and surface roughness, are within the preset specification tolerance range. For example, the groove depth is within the target value ±5μm.

[0148] If the inspection fails: the system automatically returns to the intelligent debugging engine, and re-optimizes the parameters based on the data from the failed trial processing until the first piece passes the inspection.

[0149] Through the fully automated process described above, the processing method of this embodiment can achieve the following beneficial effects: the total time for tool switching and parameter debugging is reduced from several hours to about 5 minutes, reducing the dependence on highly skilled operators and effectively avoiding misassembly and parameter deviation caused by human error; during continuous processing, the system actively suppresses parameter drift, which significantly improves the dimensional consistency within the batch, reduces the product defect rate by more than 30%, and significantly improves the overall equipment efficiency (OEE).

[0150] Figure 9 The diagram shown is a block diagram of a silicon wafer processing apparatus according to an embodiment of the present invention. Figure 9 As shown, the silicon wafer processing apparatus 800 includes: The acquisition module 810 is used to acquire the specification information of the silicon wafer to be processed in response to the tool switching command, wherein the tool switching command is generated by the manufacturing execution system or human-machine interface based on the specification change of the silicon wafer to be processed; The identification module 820 is used to identify and determine the target machining tool from the tool repository based on specification information, wherein the tool repository includes at least two different models of machining tools; The switching module 830 is used to switch the current machining tool currently installed at the working position of the machining equipment to the target machining tool; The generation module 840 is used to generate target machining parameters that are compatible with the specification information and the target machining tool from a historical parameter database associated with the specification information. The control module 850 is used to control the processing equipment to process the silicon wafer to be processed based on the target processing parameters.

[0151] In some implementations, the identification module 820 is also used for: Based on specification information, obtain identifier matching data associated with the target machining tool; Obtain the unique electronic identification of all processing tools in the tool repository; All unique electronic identifiers are compared with the identifier matching data to identify the target processing tool; Based on the identification results, the location information of the target processing tool in the tool repository is determined, and the corresponding grabbing and positioning information is generated.

[0152] In some implementations, the switching module 830 is also used for: Based on a preset disassembly force threshold, the current processing tool is removed. The preset disassembly force threshold is the maximum allowable disassembly force set during the disassembly process to avoid damage to the processing equipment and the current processing tool. Clean the work area based on preset cleaning rules; Based on the captured location information, target processing tools are captured; Obtain the pose deviation between the target machining tool and the working position; Based on the positional deviation and the preset installation force threshold, the target machining tool is installed into the working position without damage. The preset installation force threshold is the target installation force set during the installation process to ensure that the target machining tool and the working position reach a predetermined fit.

[0153] In some implementations, the generation module 840 is further configured to: The target historical parameter set is obtained from the historical parameter database. The target historical parameter set is a combination of historically successful processing parameters that match the specification information and the target processing tool, and is used as the initial iteration point of the iterative optimization algorithm. Test parameters are automatically generated based on the target's historical parameter set through an iterative optimization algorithm. Based on the test parameters, the trial processing quality data is obtained. The trial processing quality data is the comprehensive processing quality data obtained after the processing equipment performs trial processing based on the test parameters. Based on the processing quality data, the optimization direction of the iterative optimization algorithm is corrected. When the iterative optimization algorithm meets the convergence condition, the optimal combination of processing parameters under the current working condition is output as the target processing parameters.

[0154] In some implementations, the control module 850 is also used for: Acquire real-time processing parameters, which are data continuously collected during the processing to characterize the current actual processing status of the processing equipment; The real-time processing parameters and the target processing parameters are input into the preset prediction and decision model. Through inference calculation, the predicted deviation trend of the real-time processing parameters relative to the target processing parameters and the corresponding optimal compensation strategy are obtained. When a trend deviation in real-time processing parameters is predicted but has not yet exceeded the preset specification range, the adjustable process parameters of the processing equipment are adjusted based on the optimal compensation strategy to make the real-time processing parameters converge to the target value of the target processing parameters.

[0155] In some implementations, such as Figure 10 As shown, the silicon wafer processing apparatus 800 further includes: The judgment module 860 is used to determine whether the target machining tool is in a machinable state.

[0156] In some implementations, the determination module 860 is further used for: Obtain dynamic balance test values ​​and runout test values. The dynamic balance test values ​​are used to characterize the residual imbalance after the target machining tool is installed, and the runout test values ​​are used to characterize the radial runout and / or axial runout after the target machining tool is installed. Based on preset rules, it is determined whether the dynamic balance detection value and the runout detection value meet their respective preset requirements. When both the dynamic balance detection and the runout detection meet their respective preset requirements, the target machining tool is confirmed to be in a machinable state.

[0157] In some implementations, the determination module 860 is further used for: If the dynamic balance test value does not meet the corresponding preset requirements, the dynamic balance fine-tuning program will be executed automatically, and then the dynamic balance test value will be re-acquired and judged again until the dynamic balance test value meets the corresponding preset requirements. The dynamic balance fine-tuning program is used to eliminate or compensate for residual imbalance by adaptively adjusting the preset counterweight structure on the target machining tool. If the fluctuation detection value does not meet the corresponding preset requirements, a correction action is performed or a reinstallation process is triggered. Then, the fluctuation detection value is obtained again for re-judgment until the fluctuation detection value meets the corresponding preset requirements.

[0158] The specific implementation process of the functions and roles of each module in the above-mentioned device can be found in the corresponding steps of the silicon wafer processing method in the above embodiments, and will not be repeated here.

[0159] Figure 11 The diagram shown is a block diagram of a processing device 900 provided in an embodiment of the present invention.

[0160] like Figure 11 As shown, the processing equipment 900 includes a processor 910 and a memory 920. The processor 910 is used to execute instructions stored in the memory 920, such as an application program, to implement the silicon wafer processing method in the above embodiment.

[0161] In at least one embodiment of this invention, the processing apparatus 900 includes one or more processors 910, and the application programs stored in the memory 920 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processor 910 is configured to execute instructions to perform the silicon wafer processing method described in the above embodiments.

[0162] In at least one embodiment of this invention, the processing device 900 may further include a power supply component configured to perform power management of the processing device 200, a wired or wireless network interface configured to connect the processing device 900 to a network, and an input / output (I / O) interface. The processing device 900 may operate on an operating system stored in memory 920, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0163] The present invention also provides a computer storage medium storing at least one instruction, which is executed by the processor 910 of the above-described processing equipment 900 to implement the silicon wafer processing method in the above embodiments.

[0164] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0165] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0166] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0167] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0168] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0169] If the functionality is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program verification codes, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0170] Furthermore, it should be noted that the combination of the various technical features in this case is not limited to the combination methods described in the claims of this case or the combination methods described in the specific embodiments. All technical features described in this case can be freely combined or combined in any way, unless they contradict each other.

[0171] It should be noted that the above examples are merely specific embodiments of the present invention, and the present invention is obviously not limited to the above embodiments, with many similar variations. All modifications that can be directly derived or conceived by those skilled in the art from the content disclosed in this invention should fall within the protection scope of this invention.

[0172] It should be understood that the terms "first," "second," etc., mentioned in the embodiments of the present invention are merely for the purpose of more clearly describing the use of the technical solutions in the embodiments of the present invention, and are not intended to limit the scope of protection of the present invention.

[0173] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for processing silicon wafers, characterized in that, Applied to processing equipment, the processing method includes: In response to a tool switching command, the specification information of the silicon wafer to be processed is obtained, wherein the tool switching command is generated by the manufacturing execution system or human-machine interface based on the specification change of the silicon wafer to be processed; Based on the specifications, the target machining tool is identified and determined from the tool repository, wherein the tool repository includes at least two different models of machining tools; Switch the current processing tool currently installed at the working position of the processing equipment to the target processing tool; Based on a historical parameter database associated with the specification information, target processing parameters adapted to the specification information and the target processing tool are generated. The processing equipment is controlled to process the silicon wafer to be processed based on the target processing parameters.

2. The processing method according to claim 1, characterized in that, The step of identifying and determining the target machining tool from the tool repository based on the specification information includes: Based on the specification information, obtain the identifier matching data associated with the target processing tool; Obtain the unique electronic identification of all the processing tools in the tool repository; All the unique electronic identifiers are compared with the identifier matching data to identify the target processing tool; Based on the identification results, the location information of the target processing tool in the tool repository is determined, and corresponding grabbing and positioning information is generated.

3. The processing method according to claim 2, characterized in that, The step of switching the current machining tool currently installed at the working position of the machining equipment to the target machining tool includes: Based on a preset disassembly force threshold, the current processing tool is removed, wherein the preset disassembly force threshold is the maximum allowable disassembly force set during the disassembly process to avoid damage to the processing equipment and the current processing tool; The work area is cleaned based on preset cleaning rules; Based on the grasping and positioning information, the target processing tool is grasped; Obtain the pose deviation between the target machining tool and the working position; Based on the positional deviation and the preset installation force threshold, the target machining tool is installed into the working position without damage. The preset installation force threshold is a target installation force set during the installation process to ensure that the target machining tool and the working position reach a predetermined fit.

4. The processing method according to claim 1, characterized in that, The step of generating target processing parameters adapted to the specification information and the target processing tool based on a historical parameter database associated with the specification information includes: The target historical parameter set is obtained from the historical parameter database. The target historical parameter set is a combination of historically successful processing parameters that match the specification information and the target processing tool, and is used as the initial iteration point of the iterative optimization algorithm. Based on the target historical parameter set, test parameters are automatically generated through the iterative optimization algorithm; Based on the test parameters, trial processing quality data is obtained, wherein the trial processing quality data is the comprehensive processing quality data obtained by the processing equipment after performing trial processing based on the test parameters; Based on the processing quality data, the optimization direction of the iterative optimization algorithm is corrected. When the iterative optimization algorithm meets the convergence condition, the optimal combination of processing parameters under the current working condition is output as the target processing parameters.

5. The processing method according to claim 4, characterized in that, The iterative optimization algorithm includes a Bayesian optimization algorithm. When the number of iterations of the Bayesian optimization algorithm reaches a preset number, it is determined that the convergence condition is met.

6. The processing method according to claim 1, characterized in that, In the process of controlling the processing equipment to process the silicon wafer based on the target processing parameters, the method includes: an adaptive compensation processing process, wherein the adaptive compensation processing process includes: Acquire real-time processing parameters, wherein the real-time processing parameters are data continuously collected during the processing process to characterize the current actual processing status of the processing equipment; The real-time processing parameters and the target processing parameters are input into a preset prediction decision model. Through inference calculation, the predicted deviation trend of the real-time processing parameters relative to the target processing parameters and the corresponding optimal compensation strategy are obtained. When a trend deviation of the real-time processing parameters is predicted but has not yet exceeded the preset specification range, the adjustable process parameters of the processing equipment are adjusted based on the optimal compensation strategy so that the real-time processing parameters converge to the target value of the target processing parameters.

7. The processing method according to claim 6, characterized in that, The prediction and decision model is a deep Q-network model pre-trained offline using the historical parameter database, and the processing method further includes: The data of the compensation events and the corresponding compensation effects are sent back to the historical parameter database to serve as training samples for subsequent offline training, thereby continuously optimizing the deep Q-network model.

8. The processing method according to any one of claims 1 to 7, characterized in that, After switching the current processing tool currently installed at the working position of the processing equipment to the target processing tool, the method further includes: determining whether the target processing tool is in a processable state, wherein determining whether the target processing tool is in a processable state includes: Obtain dynamic balance detection values ​​and runout detection values, wherein the dynamic balance detection values ​​are used to characterize the residual imbalance after the target machining tool is installed, and the runout detection values ​​are used to characterize the radial runout and / or axial runout after the target machining tool is installed; Based on preset rules, it is determined whether the dynamic balance detection value and the runout detection value meet their respective preset requirements. When both the dynamic balance detection value and the runout detection value meet their respective preset requirements, it is confirmed that the target machining tool is in a machinable state.

9. The processing method according to claim 8, characterized in that, The processing method further includes: If the dynamic balance detection value does not meet the corresponding preset requirements, the dynamic balance fine-tuning program is automatically executed, and then the dynamic balance detection value is re-acquired for re-judgment until the dynamic balance detection value meets the corresponding preset requirements. The dynamic balance fine-tuning program is used to eliminate or compensate for the residual imbalance by adaptively adjusting the preset counterweight structure on the target machining tool. If the fluctuation detection value does not meet the corresponding preset requirements, a correction action is performed or a reinstallation process is triggered, and then the fluctuation detection value is obtained again for re-judgment until the fluctuation detection value meets the corresponding preset requirements.

10. A processing device, characterized in that, It includes a processor and a memory, the processor being configured to execute instructions stored in the memory to implement the processing method as described in any one of claims 1 to 9.

11. A computer storage medium, characterized in that, The storage medium stores at least one instruction, which is executed by a processor to implement the processing method as described in any one of claims 1 to 9.

Images