A semiconductor component cutting control method and system supporting autonomous deviation correction
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- KUNSHAN YUYUANHONG MECHANICAL & ELECTRICAL EQUIPMENT CO LTD
- Filing Date
- 2025-06-16
- Publication Date
- 2026-06-05
Smart Images
Figure CN120637217B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of spin correction control technology, and in particular to a method and system for controlling the cutting of semiconductor components that supports autonomous spin correction. Background Technology
[0002] In the current semiconductor component cutting and manufacturing process, to meet the dimensional control requirements of precision devices at the micron or even nanometer level, a combination of high-precision cutting equipment and CNC systems is commonly used to accurately process the geometry of the components. However, in practical applications, due to the interplay of various factors such as the material properties of the components themselves, the thermal effects during manufacturing, differences in equipment operating stability, and external vibrations, problems such as displacement deviations, structural deformations, or contour errors can easily occur during the cutting process. These errors are usually nonlinear and discontinuous, with strong coupling and difficulty in predicting them in advance, making traditional control methods that rely on fixed paths, static compensation, or preset tool compensation values inadequate for handling dynamic changes.
[0003] In summary, existing technologies suffer from technical problems such as displacement deviation, structural deformation, or contour error during the cutting process due to the combined effects of various factors, including component material properties, thermal effects, differences in equipment operational stability, and external vibrations. These issues further affect cutting accuracy and product quality. Summary of the Invention
[0004] The purpose of this application is to provide a semiconductor component cutting control method and system that supports autonomous correction, in order to solve the technical problems in the prior art that are prone to displacement deviation, structural deformation or contour error during the cutting process due to the combined effects of multiple factors such as component material characteristics, thermal effects, differences in equipment operating stability and external vibration, which further affect the cutting accuracy and product quality.
[0005] In view of the above problems, this application provides a semiconductor component cutting control method and system that supports autonomous correction.
[0006] In a first aspect, this application provides a semiconductor component cutting control method that supports autonomous correction, implemented through a semiconductor component cutting control system that supports autonomous correction. The method includes: performing semiconductor component state identification and cutting equipment initialization, and performing cutting stage positioning to obtain the cutting cycle; establishing an anomaly predictor, and deploying a correction plan for the cutting cycle based on the intrinsic state information of the anomaly predictor; performing real-time micro-correction control based on the correction plan, and compressing the predicted anomaly back to the allowable tolerance range to obtain the correction processing result.
[0007] Preferably, the semiconductor component cutting control method supporting autonomous correction further includes: acquiring the intrinsic state information by collecting the material properties of the semiconductor component, measuring its geometric dimensions, detecting its surface stress, and judging its structural stability; importing the target cutting index and the target cutting index value into the cutting equipment to obtain the cutting task requirements; and locating the cutting stage based on the intrinsic state information and the cutting task requirements to obtain the cutting cycle.
[0008] Preferably, the semiconductor component cutting control method supporting autonomous correction further includes: generating a control cycle with the intrinsic state information as the control starting point and the cutting task requirement as the control ending point; dividing the control cycle by a spatial dimension to obtain a spatially divided control cycle; dividing the control cycle by a time dimension to obtain a time-divided control cycle; and performing a weighted calculation on the spatially divided control cycle and the time-divided control cycle to obtain the cutting cycle, wherein the cutting cycle includes a state monitoring threshold obtained based on the intrinsic state information and the cutting task requirement.
[0009] Preferably, the semiconductor component cutting control method supporting autonomous correction further includes: extracting historical anomaly locations and historical deviation vectors from historical cutting data to obtain deviation features; performing supervised training based on the deviation features and historical cutting control data in the historical cutting data to establish the anomaly predictor, wherein the anomaly predictor and the cutting control interface have decision linkage, and the decision linkage is visualized on the cutting control interface.
[0010] Preferably, the semiconductor component cutting control method supporting autonomous deviation further includes: predicting deviations based on the intrinsic state information using the deviation predictor to obtain the predicted deviation; locating the predicted deviation position in the cutting cycle according to the predicted deviation position to obtain the predicted deviation stage and the corresponding deviation stage state monitoring threshold; and performing deviation correction simulation based on the predicted deviation vector within the deviation stage state monitoring threshold to generate the deviation correction plan.
[0011] Preferably, the semiconductor component cutting control method supporting autonomous correction further includes: if the number of correction plans exceeds the preset number of plans, setting a condition triggering mechanism to trigger the correction plans: static threshold triggering based on the real-time acquisition status of the semiconductor component; dynamic trend triggering when the predicted anomaly is continuously increasing; and multi-condition joint triggering based on auxiliary correction indicators and combined with the anomaly stage status monitoring threshold.
[0012] Preferably, the semiconductor component cutting control method supporting autonomous bias correction further includes: adjusting the predicted cutting control data corresponding to the predicted anomaly through the bias correction plan to generate an initial bias correction processing result; obtaining a real-time acquisition status based on the initial bias correction processing result; comparing the real-time acquisition status with the pre-bias correction status of the bias correction plan to obtain a bias correction status to be supplemented; presetting the allowable tolerance range of the bias correction plan, and obtaining the corresponding allowable tolerance bias correction execution result based on the allowable tolerance range; if the allowable tolerance bias correction status of the allowable tolerance bias correction execution result does not meet the bias correction status to be supplemented, performing the next stage of supplementary bias correction in the predicted anomaly stage until the bias correction status to be supplemented is met, and obtaining the bias correction processing result.
[0013] Preferably, the semiconductor component cutting control method supporting autonomous correction further includes: the micro-correction control includes automatic correction and human-machine interaction correction, wherein the automatic correction is based on the real-time acquisition status triggering a correction plan, and the human-machine interaction correction is based on the cutting control interface selecting fine-tuning parameters to obtain the correction processing result.
[0014] Preferably, the semiconductor component cutting control method supporting autonomous bias correction further includes: retraining the parameters of the mutation predictor, wherein the retraining process is based on incremental learning through a dynamic sampling window, and whether to update the mutation predictor is determined according to a preset stability threshold.
[0015] Secondly, this application also provides a semiconductor component cutting control system supporting autonomous correction, used to execute a semiconductor component cutting control method supporting autonomous correction as described in the first aspect, comprising: a cutting stage positioning module, used to perform semiconductor component state identification and cutting equipment initialization, and to perform cutting stage positioning to obtain a cutting cycle; a correction plan deployment module, used to establish an anomaly predictor, and to deploy a correction plan under the cutting cycle based on the anomaly predictor and intrinsic state information; and a correction processing result obtaining module, used to perform real-time micro-correction control based on the correction plan, and to compress the predicted anomaly back to the allowable tolerance range to obtain the correction processing result.
[0016] The technical solution provided in this application has at least the following technical effects or advantages: by realizing precise cutting control based on real-time data and dynamic adjustment, it achieves the technical effect of improving the intelligence, adaptability and accuracy of the cutting process.
[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0019] Figure 1 This is a flowchart illustrating a semiconductor component cutting control method that supports autonomous bias correction according to this application.
[0020] Figure 2 This is a schematic diagram of the structure of a semiconductor component cutting control system that supports autonomous correction according to this application.
[0021] Explanation of reference numerals in the attached diagram: 11 is the positioning module for the cutting stage, 12 is the deployment module for the correction plan, and 13 is the module for obtaining the correction processing result. Detailed Implementation
[0022] This application provides a semiconductor component cutting control method and system that supports autonomous correction. It solves the technical problem in existing technologies where the interplay of various factors, such as component material properties, thermal effects, differences in equipment operational stability, and external vibrations, leads to displacement deviations, structural deformations, or contour errors during the cutting process, further affecting cutting accuracy and product quality. The method achieves precise cutting control based on real-time data and dynamic adjustments, thereby improving the intelligence, adaptability, and accuracy of the cutting process.
[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.
[0024] Example 1, please refer to the appendix. Figure 1 This application provides a semiconductor component cutting control method that supports autonomous bias correction, applied to a semiconductor component cutting control system that supports autonomous bias correction, specifically including the following steps:
[0025] S1: Perform semiconductor component status identification and cutting equipment initialization, and perform cutting stage positioning to obtain the cutting cycle.
[0026] Specifically, before the cutting process begins, a comprehensive inspection and feature extraction are performed on the semiconductor components to be processed. This includes material properties such as crystal type and conductivity, geometric dimensions such as length, width, and thickness, surface stress such as thermal stress or residual stress distribution, and structural stability such as the degree of deformation of the component under minute vibrations. Cutting equipment initialization refers to starting the cutting device and setting its parameters, such as calibrating the initial position of the cutter, setting the laser focal length or track speed, and ensuring normal data communication between the sensors and the execution system. Cutting stage positioning refers to determining the various stages in the entire cutting process based on the aforementioned identification information and equipment parameters. These stages are divided into a pre-acceleration stage, a stabilization cutting stage, and a finishing stage. The duration and process parameters of each stage are directly related to the component's state. The cutting cycle is the total control time window or spatial path interval formed by the combination of stages, used to constrain subsequent path planning, error assessment, and the execution of corrective measures.
[0027] S2: Establish an anomaly predictor, and deploy a correction plan for the cutting cycle based on the intrinsic state information using the anomaly predictor.
[0028] Specifically, an anomaly predictor is established to identify potential deviations or anomalies in advance by analyzing the real-time status information of semiconductor components. By combining historical cutting data, real-time monitoring information, and physical models, the anomaly predictor can identify deviations caused by factors such as thermal expansion, material inhomogeneity, and equipment errors. For example, the predictor can predict geometric deformation caused by thermal expansion based on temperature sensor data and component thickness data. Based on the anomaly predictor's intrinsic state information, a correction plan is deployed during the cutting cycle. Throughout the cutting cycle, cutting parameters are adjusted in a timely manner according to the predictor's output to ensure that the component remains within the set tolerance range throughout the processing. The cutting cycle refers to the time window from the start to the completion of cutting.
[0029] S3: Based on the aforementioned correction plan, perform real-time micro-correction control and compress the predicted anomalies back to the allowable tolerance range to obtain the correction processing result.
[0030] Specifically, during the cutting process, cutting parameters are adjusted in real time according to the deployed correction plan to ensure that each operation is within a controllable range. For example, the correction plan may include fine-tuning the tool angle, adjusting the feed speed, or changing the laser focus position to address anticipated potential deviations. Real-time micro-correction control reacts rapidly at every moment to address sudden deviation changes during dynamic cutting, ensuring that accuracy is consistently maintained. After executing micro-correction control, it ensures that the cutting error does not exceed the preset allowable error range. If a predicted anomaly causes the deviation to exceed the tolerance range, the error is automatically corrected through micro-correction. After executing micro-correction control, the corrected result is output and compared with the target cutting size to ensure that the final product's accuracy meets design requirements.
[0031] Furthermore, this application also includes: acquiring the intrinsic state information by collecting the material properties of the semiconductor component, measuring its geometric dimensions, detecting its surface stress, and judging its structural stability; importing the target cutting index and the target cutting index value into the cutting equipment to obtain the cutting task requirements; and locating the cutting stage based on the intrinsic state information and the cutting task requirements to obtain the cutting cycle.
[0032] Specifically, intrinsic state information of semiconductor components is obtained through material property acquisition, geometric dimension measurement, surface stress detection, and structural stability assessment. Material property acquisition involves testing the basic properties of semiconductor components, such as composition, hardness, coefficient of thermal expansion, and conductivity, to understand the material's performance under different environments. Geometric dimension measurement involves accurately measuring the shape and size of the semiconductor component to ensure it meets design requirements. Surface stress detection measures the stress state on the component's surface to understand potential deformation trends during processing. Structural stability assessment analyzes the stability of the semiconductor component during processing and predicts whether it may unexpectedly deform or break during cutting.
[0033] Next, the specific requirements of the cutting task are input into the cutting equipment, such as target size, accuracy, tolerance range, and required cutting speed. Target cutting parameters include the thickness of the material being cut, edge smoothness, and accuracy requirements. The target cutting parameter value is a specific numerical definition of the parameter, for example, requiring the cutting error to be no more than ten micrometers.
[0034] The positioning and cutting stage refers to identifying the phases that require special attention throughout the cutting process, based on the intrinsic state of the semiconductor component and the requirements of the cutting task. For example, if the material has high hardness, the cutting speed or temperature may need to be adjusted to avoid material damage. The cutting cycle refers to the time required from the start to the completion of cutting, and this time is adjusted according to the performance of the cutting equipment, material properties, and task requirements.
[0035] Furthermore, this application also includes: generating a control cycle using the intrinsic state information as the control starting point and the trimming task requirement as the control ending point; dividing the control cycle by a spatial dimension to obtain a spatially divided control cycle; dividing the control cycle by a time dimension into equal-period segments to obtain a time-divided control cycle; and performing a weighted calculation on the spatially divided control cycle and the time-divided control cycle to obtain the trimming cycle, wherein the trimming cycle includes a state monitoring threshold obtained based on the intrinsic state information and the trimming task requirement.
[0036] Specifically, a control cycle is generated by using intrinsic state information as the starting point and the cutting task requirements as the control endpoint. During the cutting process, basic material information (such as hardness, stress, and dimensions) needs to be monitored and controlled as initial reference points, while the cutting task requirements (such as target dimensions and tolerance requirements) determine the final goal of the cutting process. The control cycle refers to the time required from the start to the completion of cutting; the entire cycle is generated based on the differences and correlations between these two key points. The control cycle may vary depending on the material properties and task requirements; for example, materials with higher hardness require a longer time to complete the cutting process.
[0037] The control cycle is divided into spatial control cycles by spatial dimensions. Spatial dimension division refers to segmenting the cutting process according to the spatial characteristics of the component (such as shape, cutting path, geometric features, etc.). Each spatial control cycle corresponds to a part of the component, which may be a specific area or a segment on the cutting path.
[0038] The control cycle is divided into equal periods by the time dimension, resulting in a time-divided control cycle. This time-division involves dividing the entire cutting process into several equal-length time intervals. Within each time interval, the cutting process is controlled according to set conditions such as speed and temperature to ensure that the operation at each point in time meets the accuracy requirements.
[0039] The trimming cycle is obtained by weighting the spatial and temporal control cycles. The trimming cycle includes a state monitoring threshold derived from intrinsic state information and trimming task requirements. Weighted calculation combines the spatial and temporal division results and calculates them proportionally to arrive at a comprehensive trimming cycle. Weights are allocated based on the influence of spatial division (e.g., the cutting characteristics of different regions) and temporal division (e.g., cutting speed, temperature control). Ultimately, the trimming cycle includes not only temporal and spatial arrangements but also a state monitoring threshold for real-time monitoring of whether the trimming process meets requirements.
[0040] Furthermore, this application also includes: extracting historical mutation locations and historical deviation vectors from historical cropping data to obtain deviation features; performing supervised training based on the deviation features and historical cropping control data in the historical cropping data to establish the mutation predictor, wherein the mutation predictor and the cropping control interface have decision linkage, and the cropping control interface is visualized based on the decision linkage.
[0041] Specifically, historical anomaly locations and historical deviation vectors are extracted from historical cutting data to obtain deviation characteristics. During the completed semiconductor component cutting process, the locations where errors occurred, i.e., anomaly locations, are identified from historical records, and the direction and magnitude of the errors, i.e., deviation vectors, are analyzed. Anomaly locations refer to the specific areas where the actual cutting process deviates from the ideal trajectory, while deviation vectors represent the numerical direction of the error, such as a deviation of 0.02 mm to the left or 0.05 mm upwards.
[0042] Next, a variation predictor is established through supervised training based on deviation features and historical cutting control data from historical cutting data. Supervised training is a machine learning method that uses training samples composed of input features and known results to teach the model the correspondence between features and results. Historical cutting control data provides the input parameters of the control system at that time, such as tool speed, feed rate, and laser power, while deviation features provide the representation of the output results. Through supervised training, a variation predictor can be established to predict in advance under which conditions errors are likely to occur during future cutting processes and to provide possible variation trends, thereby achieving early warning and proactive intervention.
[0043] Then, the anomaly predictor and the trimming control interface have decision-making linkage. Decision-making linkage means that the predictor's judgment result can directly affect the operating strategy of the trimming control system. When the predictor judges that a deviation may occur at a certain stage, it issues adjustment suggestions to the control system, such as reducing speed, changing trajectory, or initiating a correction program. The trimming control interface is a human-machine interaction platform used to manage the entire trimming process, displaying system operating status, control parameters, alarm information, etc.
[0044] Finally, based on the decision-making linkage, the system is visualized on the control interface, displayed in real time as graphics or data, making it easier for operators to understand the current status of the system and potential risks. For example, the control interface may display a region marked as a high-risk anomaly zone, or use a red warning bar to show the direction and degree of possible deviation from the current trajectory, thereby improving the operator's perception and response efficiency.
[0045] Furthermore, this application also includes: predicting the intrinsic state information based on the mutation predictor to obtain the predicted mutation; locating the predicted mutation position in the cutting cycle according to the predicted mutation position to obtain the predicted mutation stage and the corresponding mutation stage state monitoring threshold; and performing mutation correction simulation based on the prediction deviation vector within the mutation stage state monitoring threshold to generate the correction plan.
[0046] Specifically, based on the anomaly predictor, the intrinsic state information is analyzed to predict anomalies such as positional shifts, angular errors, or material deformation that may occur during the actual cutting process. The intrinsic state information includes features such as the component's material composition, geometric dimensions, surface stress, and structural integrity. Anomaly prediction refers to the model's prediction of potential problems during future cutting, such as predicting that a certain area will shift 0.04 mm to the right after heating.
[0047] Based on the predicted mutation location within the trimming cycle, the predicted mutation stage and its corresponding state monitoring threshold are obtained. The trimming cycle refers to the entire process from start to finish trimming, which is subdivided into several stages. The predicted mutation location is the component position where errors may occur, given by the predictor. Mapping this location to the trimming cycle in the time dimension determines the trimming stage. The state monitoring threshold is the error tolerance range set for this stage, used to determine whether the error in the actual trimming exceeds the acceptable range.
[0048] Based on the predicted deviation vector, anomaly correction simulation is performed within the state monitoring threshold during the anomaly stage to generate a correction plan. The predicted deviation vector is a specific numerical expression of the direction and magnitude of the abnormal behavior, such as the offset direction being the positive X-axis and the magnitude being 0.05 mm. Given the predicted deviation and the state monitoring threshold, simulations can be performed to determine whether different adjustment strategies can compress the prediction error to within the threshold, serving as anomaly correction simulation to find the optimal or feasible correction scheme. The final generated correction plan is a set of preset control commands or operating procedures, such as reducing the feed rate at a certain time point, fine-tuning the laser focus, and adjusting the trajectory curvature, used to actively control the deviation during the cutting process.
[0049] Furthermore, this application also includes: if the number of the correction plan exceeds the preset number of plans, a condition triggering mechanism is set to trigger the correction plan: static threshold triggering is performed based on the real-time acquisition status of the semiconductor component; dynamic trend triggering is performed when the predicted anomaly is continuously increasing; and multi-condition joint triggering is performed based on auxiliary correction indicators and combined with the anomaly stage status monitoring threshold.
[0050] Specifically, when the number of corrective action plans generated based on the prediction model exceeds the pre-set maximum allowable number, an automated mechanism is activated to determine whether further measures are needed. The pre-set number of plans is a maximum value; exceeding this value indicates that there may be problems in the trimming process that cannot be resolved by conventional means, thus requiring the activation of a triggering mechanism to ensure timely response.
[0051] Static threshold triggering is based on real-time data acquisition of semiconductor components. Static threshold triggering involves comparing real-time acquired data (such as temperature, stress, and speed) with pre-set fixed thresholds. Once the acquired real-time data exceeds the set threshold, a corrective action is triggered. For example, when the temperature exceeds 50 degrees Celsius, a plan might be activated to adjust the cutting speed to prevent excessive temperature from affecting the semiconductor components.
[0052] When a predicted anomaly shows continuous growth, dynamic trend triggering is implemented. Dynamic trend triggering involves judging the continuous growth trend of the anomaly based on the prediction model. When the prediction indicates that a deviation is gradually worsening and the rate of change exceeds the expected range, a contingency plan is triggered to address this trend. For example, if the deviation has gradually increased from 0.01 mm to 0.05 mm in the past few cutting processes without being effectively contained, judging based on the trend whether more urgent corrective measures are needed can better address the impact of long-term changes, especially when the deviation is gradually increasing, allowing for early identification of problems and automatic adjustments.
[0053] Based on auxiliary correction indicators, a multi-condition joint triggering mechanism is implemented by combining anomaly stage status monitoring thresholds. Auxiliary correction indicators are additional parameters related to cutting quality, such as cutting force, vibration frequency, and tool wear, providing more information to help determine whether correction is needed. The anomaly stage status monitoring threshold is the maximum allowable deviation within a specific cutting stage. A more comprehensive judgment is made by combining different factors. For example, if the cutting force exceeds a set threshold at a certain stage, and the temperature and speed also exceed limits, multiple indicators are combined to trigger the execution of the correction plan.
[0054] Furthermore, this application also includes: adjusting the prediction clipping control data corresponding to the predicted anomaly through the correction plan to generate an initial correction processing result; obtaining a real-time acquisition status based on the initial correction processing result; comparing the real-time acquisition status with the pre-correction status of the correction plan to obtain a supplementary correction status; pre-setting the allowable tolerance range of the correction plan, and obtaining the corresponding allowable tolerance correction execution result based on the allowable tolerance range; if the allowable tolerance correction status of the allowable tolerance correction execution result does not meet the supplementary correction status, performing the next stage of supplementary correction in the predicted anomaly stage until the supplementary correction status is met, and obtaining the correction processing result.
[0055] Specifically, the predictive cutting control data corresponding to the predicted anomalies is adjusted through a correction plan to generate initial correction results. After predicting potential anomalies in the semiconductor component, the control parameters during the cutting process are adjusted according to the pre-prepared correction plan. The control data may include the tool's trajectory, cutting speed, and laser focal position. By adjusting these parameters, it is expected that the predicted anomalies can be reduced or eliminated.
[0056] The real-time status is obtained based on the initial corrective action results. After the initial corrective action is executed, the current operating status is monitored in real time, including various indicators such as temperature, stress, speed, and vibration, and fed back to the control system to verify whether the preliminary corrective measures have achieved the expected results.
[0057] The real-time data acquisition status is compared with the pre-correction status in the correction plan to obtain the status to be supplemented for correction. By comparing the real-time data acquisition status with the preset ideal status (i.e., the pre-correction status), it is determined whether the current trimming is within an acceptable error range. If there is a deviation between the real-time data acquisition status and the preset correction status, further adjustments are required.
[0058] The permissible tolerance range of the pre-defined correction plan is used to obtain the corresponding permissible tolerance correction execution result. The tolerance range refers to the allowable error range, which specifies the maximum acceptable deviation value, and then calculates the correction result that meets the permissible tolerance.
[0059] If the allowable tolerance correction state of the execution result does not meet the supplementary correction state, the next stage of supplementary correction in the prediction anomaly stage is performed until the supplementary correction state is met, and the correction result is obtained. Supplementary correction measures are gradually implemented until the correction effect meets the preset target. For example, if the error after the first supplement still does not reach the tolerance value, further compensation is performed by adjusting the parameters in the next control cycle until the error is within the allowable range.
[0060] Furthermore, this application also includes: the micro-correction control includes automatic correction and human-computer interaction correction, the automatic correction is based on the real-time acquisition status triggering a correction plan, and the human-computer interaction correction is based on the selection of fine-tuning parameters through the trimming control interface to obtain the correction processing result.
[0061] Specifically, micro-calibration control includes automatic calibration and human-machine interface (HMI) calibration. This means that during the semiconductor component cutting process, the system employs two micro-calibration methods to ensure cutting accuracy and control stability. Micro-calibration control refers to making small, continuous adjustments to control parameters without interrupting the cutting process. Automatic calibration relies on the system's adaptive control capabilities, while HMI calibration requires operator intervention based on interface feedback. HMI calibration means that when automatic calibration cannot fully meet accuracy requirements or handle complex unforeseen situations, operators can intervene through the human-machine interface. The cutting control interface is a graphical control terminal that displays key data from the cutting process and allows for permissible operations. Operators can select appropriate fine-tuning parameters based on the displayed information.
[0062] Furthermore, this application also includes: retraining the parameters of the mutation predictor, wherein the retraining process is based on incremental learning using a dynamic sampling window, and whether to update the mutation predictor is determined according to a preset stability threshold.
[0063] Specifically, the parameters of the anomaly predictor are retrained. Building upon its initial learning capabilities, the original prediction model is continuously optimized and adjusted by acquiring new cutting data, making it more closely reflect the actual state of the current equipment and components. An anomaly predictor is an intelligent model used to predict potential deviations or anomalies during semiconductor cutting. It is constructed using neural networks, decision trees, or ensemble learning models to identify potential errors in advance, thereby assisting in making adjustment decisions.
[0064] The retraining process utilizes incremental learning based on a dynamic sampling window. A sliding or variable-length data window continuously collects cropped data from the most recent period and feeds it into the model as training samples. The dynamic sampling window automatically adjusts its size according to the rate of data change; for example, the window width is 100 seconds when the device is running stably, and expands to 300 seconds during periods of frequent anomalies to ensure the representativeness and timeliness of the sampling. Incremental learning emphasizes that the model continuously absorbs new data while retaining the original training results, gradually correcting its predictive ability for future states.
[0065] The decision to update the mutation predictor is based on a preset stability threshold. Certain evaluation metrics, such as error reduction rate, output volatility, or prediction accuracy, are used to determine if the new model possesses sufficient reliability. The stability threshold is a set of boundary conditions set to measure prediction performance. Otherwise, the new model is discarded, and the old model continues to be used to avoid error amplification or unstable behavior. Table 1 shows a partial record of the most recent mutation predictor retraining.
[0066] Table 1: Partial records of the most recent mutation predictor retraining
[0067]
[0068] In summary, the semiconductor component cutting control method supporting autonomous correction provided in this application has the following technical effects: by realizing precise cutting control based on real-time data and dynamic adjustment, it achieves the technical effect of improving the intelligence, adaptability and accuracy of the cutting process.
[0069] Example 2: Based on the same inventive concept as the semiconductor component cutting control method supporting autonomous bias correction in the foregoing examples, this application also provides a semiconductor component cutting control system supporting autonomous bias correction. Please refer to the appendix. Figure 2 The system includes: a cutting stage positioning module 11, used to perform semiconductor component state identification and cutting equipment initialization, and to perform cutting stage positioning to obtain the cutting cycle; a deviation correction plan deployment module 12, used to establish an anomaly predictor, and to deploy a deviation correction plan under the cutting cycle based on the anomaly predictor and intrinsic state information; and a deviation correction processing result obtaining module 13, used to perform real-time micro-correction control based on the deviation correction plan, and to compress the predicted anomaly back to the allowable tolerance range to obtain the deviation correction processing result.
[0070] Furthermore, the semiconductor component cutting control system that supports autonomous correction is also used to: obtain the intrinsic state information by acquiring the material properties of the semiconductor component, measuring its geometric dimensions, detecting its surface stress, and judging its structural stability; import the target cutting index and the target cutting index value into the cutting equipment to obtain the cutting task requirements; and locate the cutting stage based on the intrinsic state information and the cutting task requirements to obtain the cutting cycle.
[0071] Furthermore, the semiconductor component cutting control system supporting autonomous correction is also used to: generate a control cycle with the intrinsic state information as the control starting point and the cutting task requirement as the control ending point; divide the control cycle by a spatial dimension to obtain a spatially divided control cycle; divide the control cycle by a time dimension into equal periods to obtain a time-divided control cycle; and perform a weighted calculation on the spatially divided control cycle and the time-divided control cycle to obtain the cutting cycle, wherein the cutting cycle includes a state monitoring threshold obtained based on the intrinsic state information and the cutting task requirement.
[0072] Furthermore, the semiconductor component cutting control system that supports autonomous correction is also used to: extract historical anomaly locations and historical deviation vectors from historical cutting data to obtain deviation features; perform supervised training based on the deviation features and historical cutting control data in the historical cutting data to establish the anomaly predictor, wherein the anomaly predictor and the cutting control interface have decision linkage, and the decision linkage is visualized on the cutting control interface.
[0073] Furthermore, the semiconductor component cutting control system supporting autonomous deviation correction is also used for: predicting deviations based on the intrinsic state information using the deviation predictor to obtain the predicted deviations; locating the predicted deviation position in the cutting cycle according to the predicted deviation position to obtain the predicted deviation stage and the corresponding deviation stage state monitoring threshold; and performing deviation correction simulation based on the predicted deviation vector within the deviation stage state monitoring threshold to generate the deviation correction plan.
[0074] Furthermore, the semiconductor component cutting control system that supports autonomous correction is also used to: if the number of correction plans exceeds the preset number of plans, set a condition triggering mechanism to trigger the correction plans: perform static threshold triggering based on the real-time acquisition status of the semiconductor component; perform dynamic trend triggering when the predicted anomaly is continuously increasing; and perform multi-condition joint triggering based on auxiliary correction indicators and combined with the anomaly stage status monitoring threshold.
[0075] Furthermore, the semiconductor component cutting control system supporting autonomous correction is also used for: adjusting the predictive cutting control data corresponding to the predicted anomaly through the correction plan to generate an initial correction processing result; obtaining a real-time acquisition status based on the initial correction processing result; comparing the real-time acquisition status with the pre-correction status of the correction plan to obtain a supplementary correction status; presetting the allowable tolerance range of the correction plan, and obtaining the corresponding allowable tolerance correction execution result based on the allowable tolerance range; if the allowable tolerance correction status of the allowable tolerance correction execution result does not meet the supplementary correction status, performing the next stage of supplementary correction in the predicted anomaly stage until the supplementary correction status is met, and obtaining the correction processing result.
[0076] Furthermore, the semiconductor component cutting control system that supports autonomous correction is also used in the following ways: the micro-correction control includes automatic correction and human-machine interaction correction, wherein the automatic correction is based on the real-time acquisition status triggering a correction plan, and the human-machine interaction correction is based on the cutting control interface selecting fine-tuning parameters to obtain the correction processing result.
[0077] Furthermore, the semiconductor component cutting control system that supports autonomous correction is also used to: retrain the parameters of the mutation predictor, wherein the retraining process is based on incremental learning through a dynamic sampling window, and determines whether to update the mutation predictor according to a preset stability threshold.
[0078] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The semiconductor component cutting control method and specific examples supporting autonomous bias correction described in the foregoing Embodiment 1 are also applicable to the semiconductor component cutting control system supporting autonomous bias correction in this embodiment. Through the foregoing detailed description of the semiconductor component cutting control method supporting autonomous bias correction, those skilled in the art can clearly understand the semiconductor component cutting control system supporting autonomous bias correction in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.
[0079] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
[0080] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.
Claims
1. A semiconductor component cutting control method supporting autonomous bias correction, characterized in that, include: Perform semiconductor component status identification and cutting equipment initialization, and perform cutting stage positioning to obtain the cutting cycle; Establish an anomaly predictor, and deploy a correction plan for the cutting cycle based on the intrinsic state information using the anomaly predictor. Real-time micro-correction control is executed based on the aforementioned correction plan, and the predicted anomalies are compressed back into the allowable tolerance range to obtain the correction processing result. Establishing a mutation predictor includes: Based on historical cropping data, historical anomaly locations and historical deviation vectors are extracted to obtain deviation features; The mutation predictor is established by supervised training based on the deviation characteristics and historical clipping control data in the historical clipping data. The mutation predictor has decision linkage with the clipping control interface, and the decision linkage is visualized on the clipping control interface. Based on the mutation predictor, the error correction plan is deployed for the intrinsic state information under the trimming period, including: Based on the mutation predictor, the intrinsic state information is used to predict mutations, and the predicted mutations are obtained. Based on the predicted mutation location of the predicted mutation, the cutting cycle is used to locate the predicted mutation stage and the corresponding mutation stage status monitoring threshold. Based on the predicted deviation vector, anomaly correction simulation is performed within the state monitoring threshold of the anomaly stage to generate the correction plan. If the number of corrective action plans exceeds the preset number of plans, a conditional triggering mechanism is set to trigger the corrective action plans: Static threshold triggering is performed based on the real-time acquisition status of the semiconductor component; When the predicted anomaly changes to continuous growth, dynamic trend triggering is performed; Based on the auxiliary correction index, and combined with the state monitoring threshold of the anomaly stage, multiple conditions are jointly triggered.
2. The semiconductor component cutting control method supporting autonomous bias correction as described in claim 1, characterized in that, Perform semiconductor component status identification and cutting equipment initialization, and perform cutting stage positioning to obtain the cutting cycle, including: The intrinsic state information is obtained by acquiring the material properties of the semiconductor component, measuring its geometric dimensions, detecting its surface stress, and determining its structural stability. The target cutting index and target cutting index value are imported into the cutting equipment to obtain the cutting task requirements; Based on the intrinsic state information and the cutting task requirements, the cutting stage is located, and the cutting cycle is obtained.
3. The semiconductor component cutting control method supporting autonomous bias correction as described in claim 2, characterized in that, Based on the intrinsic state information and the trimming task requirements, the trimming stage is located, and the trimming cycle is obtained, including: A control cycle is generated using the intrinsic state information as the control start point and the trimming task requirements as the control end point. The control cycle is divided by spatial dimensions to obtain a spatially divided control cycle; The control cycle is divided into equal periods by the time dimension to obtain the time-divided control cycle; The spatial partitioning control cycle and the temporal partitioning control cycle are weighted and calculated to obtain the trimming cycle, wherein the trimming cycle includes a state monitoring threshold obtained based on the intrinsic state information and the trimming task requirements.
4. The semiconductor component cutting control method supporting autonomous bias correction as described in claim 1, characterized in that, Based on the aforementioned correction plan, real-time micro-correction control is executed, and the predicted anomalies are compressed back into the allowable tolerance range to obtain the correction processing results, including: The prediction clipping control data corresponding to the predicted anomaly is adjusted by the correction plan to generate an initial correction processing result. The real-time acquisition status is obtained based on the initial correction processing results; The real-time acquisition status is compared with the pre-correction status of the correction plan to obtain the correction status to be supplemented. The allowable tolerance range of the pre-defined correction plan is preset, and the corresponding allowable tolerance correction execution result is obtained based on the allowable tolerance range; If the allowable tolerance correction state of the allowable tolerance correction execution result does not meet the supplementary correction state, the next stage of the prediction anomaly stage is performed for supplementary correction until the supplementary correction state is met, and the correction processing result is obtained.
5. The semiconductor component cutting control method supporting autonomous bias correction as described in claim 4, characterized in that, The micro-correction control includes automatic correction and human-computer interaction correction. The automatic correction is based on the real-time acquisition status triggering a correction plan. The human-computer interaction correction is based on the selection of fine-tuning parameters through the cutting control interface to obtain the correction processing result.
6. The semiconductor component cutting control method supporting autonomous bias correction as described in claim 1, characterized in that, The mutation predictor is retrained, wherein the retraining process is based on incremental learning using a dynamic sampling window, and the mutation predictor is updated based on a preset stability threshold.
7. A semiconductor component cutting control system supporting autonomous correction, characterized in that, The steps for implementing the semiconductor component cutting control method supporting autonomous bias correction as described in any one of claims 1 to 6 include: The cutting stage positioning module is used to perform semiconductor component status recognition and cutting equipment initialization, and to perform cutting stage positioning to obtain the cutting cycle; The deviation correction plan deployment module is used to establish an anomaly predictor and deploy deviation correction plans for the cutting cycle based on the intrinsic state information of the anomaly predictor. The correction processing result acquisition module is used to perform real-time micro-correction control based on the correction plan and compress the predicted anomalies back to the allowable tolerance range to obtain the correction processing result.