Needle mark monitoring method and system after wafer trial punching

Abstract

This application relates to the field of semiconductor technology, specifically to a method and system for monitoring pin marks after wafer trial bonding. The method includes: performing trial bonding on a wafer using a probe and acquiring multi-feature data during the trial bonding process; calibrating the multi-feature data and generating calibration data; using a pin mark inspection model to perform compliance judgment on the calibration data and outputting a judgment result; wherein the judgment result is compliant or non-compliant; when the judgment result is non-compliant, the manufacturing execution system immediately triggers a pre-production checkpoint lockout, prohibiting all production operations and implementing tiered real-time early warning and rectification; after rectification, a re-judgment process is triggered, re-acquiring the multi-feature data from the wafer trial bonding and outputting a judgment result; when the judgment result is compliant, subsequent wafer production operations are initiated. This application provides closed-loop monitoring of pin mark inspection anomalies after wafer trial bonding to ensure that anomalies do not flow into production.

Description

Technical Field

[0001] This application relates to the field of semiconductor technology, and in particular to a method and system for monitoring pin marks after wafer testing. Background Technology

[0002] The wafer testing preparation process includes a trial wafer fabrication procedure, after which the wafer needs to undergo pin mark inspection. However, current pin mark inspection methods have numerous technical flaws, which reduce the quality of wafer manufacturing. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose a method and system for monitoring pin marks after wafer trial production, so as to solve the above-mentioned technical problems.

[0004] To achieve the above objectives, this application provides a method for detecting pin marks after wafer trial bonding, comprising:

[0005] A probe is used to test-brick the wafer, and multi-feature data of the wafer during the test-brick process is acquired; The multi-feature data is calibrated, and calibration data is generated; The calibration data is interpreted for compliance using a pinhole inspection model, and the result is output as either compliant or non-compliant. When the judgment result is non-compliant, the Manufacturing Execution System (MES) immediately triggers a pre-production checkpoint lock, prohibits all production operations, and performs tiered real-time early warning and rectification. After the rectification is completed, a re-judgment process is triggered to reacquire the multi-feature data of the wafer test and output the judgment result. If the determination result is compliant, the subsequent wafer production operation will be initiated.

[0006] Optionally, the multi-feature data is calibrated, and calibration data is generated, wherein, The calibration data includes: continuous feature data and categorical feature data; The continuous feature data includes: execution time feature data, displacement distance feature data, and trajectory coverage feature data; The categorized feature data includes: action initiation validity data.

[0007] Optionally, calibrating the multi-feature data and generating calibration data includes: The multi-feature data is calibrated, and continuous feature data is generated; The continuous feature data is subjected to range normalization.

[0008] Optionally, calibrating the multi-feature data and generating calibration data includes: The multi-feature data is calibrated using a three-level real-time calibration technique, and calibration data is generated. The three-level real-time calibration technique includes spatial calibration, time synchronization calibration, and error compensation calibration.

[0009] Optionally, the pin mark inspection model includes a lightweight gradient boosting tree and a logistic regression model; the step of using the pin mark inspection model to perform compliance judgment on the calibration data and output the judgment result includes: A lightweight gradient boosting tree model is used to process the calibration data to output a first probability value; A logistic regression model is used to enhance the first probability value in order to output a second probability value; The first probability value and the second probability value are fused to generate the target probability value; The target probability value is determined and the determination result is output.

[0010] Optionally, determining the target probability value and outputting the determination result includes: A binary classification target determination method is used to determine the target probability value and output the determination result; wherein, when the target probability value is greater than or equal to 0.9, the determination result is compliant; when the target probability value is less than 0.9, the determination result is non-compliant.

[0011] Optionally, after completing the rectification and triggering the re-judgment process to reacquire the multi-feature data from the wafer test and output the judgment result, the process further includes: The needle mark inspection model is iterated and optimized online.

[0012] Optionally, the online iteration and optimization of the needle mark inspection model includes: The judgment result is retrieved and a checkpoint record is generated. The judgment result, the checkpoint record and the judgment result in the re-judgment process are cross-validated. The needle mark inspection model is updated according to the cross-validation result.

[0013] Optionally, before performing a test run on the wafer using a probe and acquiring multi-feature data during the test run, the method further includes: Obtain abnormal customer complaint samples, combine the obtained abnormal customer complaint samples with calibrated historical compliance samples to construct a labeled pin mark inspection model sample set, complete the initial training of the pin mark inspection model, and determine the initial compliance threshold.

[0014] Based on the same inventive concept, this application also provides a wafer test run needle mark monitoring system, comprising: The multi-source data acquisition and calibration module is used to perform test runs on the wafer using a probe, acquire multi-feature data during the test runs, calibrate the multi-feature data, and generate calibration data. The pin mark inspection module is used to perform compliance judgment on calibration data using the pin mark inspection model and output the judgment result; wherein the judgment result is compliant or non-compliant; The pre-production checkpoint locking module is used to retrieve the judgment result and generate a checkpoint record. When the judgment result is non-compliant, the manufacturing execution system immediately triggers the pre-production checkpoint locking, prohibits all production operations, and performs graded real-time early warning and rectification. The re-judgment and verification module is used to trigger the re-judgment process after the rectification is completed, re-acquire the multi-feature data of the wafer trial production and output the judgment result, retrieve the judgment result, and start the subsequent wafer production operation only when the judgment result is compliant.

[0015] As can be seen from the above, this application, by establishing a method for monitoring pin marks after wafer trial bonding, can better achieve anomaly monitoring during pin mark inspection after wafer trial bonding. By setting a pre-production checkpoint locking function, it can quickly respond to and prohibit abnormal operations when an anomaly is detected, so as to avoid affecting the actual production and processing quality of the wafer. By establishing a rectification and re-judgment function, the abnormal problems of pin mark inspection after wafer trial bonding are separated and processed. After compliance, the system is unlocked and wafer production is started, so as to achieve closed-loop monitoring of anomalies in pin mark inspection after wafer trial bonding and ensure that anomalies do not flow into production. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a method for monitoring pin marks after wafer trial bonding according to an embodiment of this application; Figure 2 This is a block diagram of a wafer test run needle mark monitoring system provided in one embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the process. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method.

[0021] In the wafer testing preparation stage, a wafer probing mark inspection is required to confirm that the probe marks at the chip contact points are free from misalignment and contact issues. Currently, wafer probing mark inspection is mostly performed manually. The accuracy and standardization of its execution directly determine the subsequent wafer testing yield, making it a critical node in wafer quality control during the wafer testing process.

[0022] However, the existing wafer probing trace inspection process involves probing before CP (Chip Probing), with engineers inspecting the traces and recording the probing coordinates. A probe station log is generated 3-5 hours later, and inspection personnel retrieve this log for verification. If no personnel are present in the trace inspection log, it is determined that no traces were inspected. Because manual wafer probing trace inspection lacks a hardware management system for further quality control, it suffers from numerous technical deficiencies and lacks a pre-production anomaly interception mechanism. Problems include: reliance on manual sampling and self-recording, leading to false operations such as "marked as completed without actual inspection," becoming a core cause of quality anomalies; rigid judgment based on fixed parameter comparisons, lacking intelligent interpretation models adapted to industrial scenarios, and incompatible with different operator habits and probe station characteristics, resulting in high rates of misjudgment and missed judgments; and the absence of a pre-production checkpoint mechanism, with the existing process relying solely on manual management confirmation. Even if anomalies are detected, they cannot be technically intercepted before production, allowing them to directly enter production and causing irreversible losses such as invalid wafer testing and a sharp drop in yield.

[0023] Existing technologies do not yet have a dedicated automated monitoring system for post-testing needle mark inspection. Relying solely on post-testing traceability and manual management cannot solve core quality control issues, becoming a key technological bottleneck restricting the future intelligent and refined upgrading of wafer testing.

[0024] Therefore, this application provides a method and system for monitoring pin marks after wafer trial production to solve the above problems.

[0025] Reference Figure 1 This application provides a method for detecting pin marks after wafer testing, including the following steps: Step S100: Use a probe to perform a test run on the wafer and acquire multi-feature data during the test run; Step S200: Calibrate the multi-feature data and generate calibration data; Step S300: Use the pin mark inspection model to perform compliance judgment on the calibration data and output the judgment result; wherein, the judgment result is compliant or non-compliant; Step S400: When the judgment result is non-compliant, the manufacturing execution system immediately triggers the pre-production checkpoint lock, prohibits all production operations, and performs graded real-time early warning and rectification. Step S500: After rectification is completed, trigger the re-judgment process, re-acquire the multi-feature data during the wafer test and output the judgment result; Step S600: When the determination result is compliant, start the subsequent wafer production operation.

[0026] This application aims to address the technical problems in existing wafer testing preparation, such as the lack of objective technical monitoring basis for pin mark inspection after wafer trial, the absence of a dedicated acquisition and monitoring system, low accuracy of pin mark inspection judgment, and the lack of pre-production hard control, which makes it easy for abnormalities to flow into production.

[0027] Meanwhile, by establishing a method for monitoring pin marks after wafer trial production, it is possible to better monitor abnormalities in pin mark inspection after wafer trial production. By setting a pre-production checkpoint locking function, it is possible to respond quickly and prohibit abnormal operations when abnormalities are detected, so as to avoid affecting the actual production and processing quality of wafers.

[0028] In addition, by establishing a rectification and re-judgment function, abnormal issues in the pin mark inspection after wafer trial are separated and processed. Once compliance is achieved, the abnormal issues are unlocked and wafer production is started, thus realizing closed-loop monitoring of abnormal issues in the pin mark inspection after wafer trial and ensuring that abnormalities do not flow into production.

[0029] In some embodiments, in step S100, the multi-feature data is calibrated and calibration data is generated, wherein... The calibration data includes: continuous feature data and categorical feature data; Continuous feature data includes: execution time feature data, displacement distance feature data, and trajectory coverage feature data; Categorized feature data includes: action initiation validity data.

[0030] In this embodiment, the calibration data is divided into continuous feature data and categorized feature data; continuous feature data is used to characterize the compliance of wafer test-brick actions, and categorized feature data is used to detect the status of wafer test-brick signals.

[0031] In some embodiments, in step S200, the multi-feature data is calibrated and calibration data is generated, including: Calibrate multi-feature data and generate continuous feature data; Range normalization is performed on continuous feature data.

[0032] In this embodiment, range normalization is used to eliminate the influence of dimensions in multi-feature data. This is used to transform the collected multi-feature data to the same scale, thereby facilitating subsequent data training and data judgment, and ensuring that the final automated detection results of needle mark inspection are more accurate, reasonable and reliable.

[0033] Optionally, range normalization can be used to map continuous feature data to the range [0,1].

[0034] In some embodiments, in step S200, the multi-feature data is calibrated and calibration data is generated, including: A three-level real-time calibration technique is used to calibrate multi-feature data and generate calibration data; the three-level real-time calibration technique includes spatial calibration, time synchronization calibration and error compensation calibration.

[0035] In this embodiment, a three-level real-time calibration technique based on displacement / time data is established to better achieve accurate acquisition of needle mark inspection feature data.

[0036] Optionally, in this embodiment, the completion of the probe station test is used as the trigger signal to acquire multi-feature data during the wafer test, and a three-level real-time calibration technology is used to calibrate the multi-feature data.

[0037] Furthermore, in this embodiment, the calibration accuracy of the three-level real-time calibration technology can reach a displacement calibration error of ±1mm and a time calibration error of ±0.1s.

[0038] In spatial calibration, after the production line loads the wafer, the manufacturing execution system switches to a new production task. At this time, a probe station is used to locate and calibrate the initial standard calibration point that generates pin marks. After the wafer is tested, the amount of pin mark displacement is checked and the pin mark displacement data is collected. The pin mark displacement data is automatically compensated, calibrated and corrected for nonlinear errors in the pin mark displacement data, thereby realizing spatial compensation of the pin mark displacement data.

[0039] In time synchronization calibration, the timestamps of the probe station and the data acquisition terminal of the manufacturing execution system are uniformly calibrated based on NTP (Network Time Protocol). In this embodiment, the time synchronization error is ±0.05s.

[0040] In error compensation calibration, a linear error prediction model is constructed using historical calibration data to compensate for random errors caused by environmental vibration and equipment temperature drift in real time, thereby dynamically correcting the collected multi-source and multi-feature data.

[0041] In some embodiments, in step S300, the pin mark inspection model includes a lightweight gradient boosting tree and a logistic regression model; the pin mark inspection model is used to perform compliance judgment on the calibration data and output the judgment result, including: Step S310: Process the calibration data using a lightweight gradient boosting tree model to output a first probability value; Step S320: Use a logistic regression model to enhance the first probability value to output a second probability value; Step S330: The first probability value and the second probability value are fused to generate the target probability value; Step S340: Determine the target probability value and output the determination result.

[0042] In this embodiment, a lightweight gradient boosting tree model and a logistic regression model are fused to output the target probability value, which is used to ensure the accuracy of the multi-target feature data interpretation results. The target probability value of the multi-feature data is predicted using the lightweight gradient boosting tree model, and the prediction result is calibrated using the logistic regression model, thereby improving the accuracy, stability and reliability of the final output target probability value.

[0043] Here, the lightweight gradient boosting tree model in this embodiment has a learning rate of 0.05, a number of decision trees of 50, and a maximum depth of 6; the L2 regularization coefficient in the logistic regression model is 0.1. In this embodiment, a target probability value is generated, and the result is determined based on the target probability value to ensure the rigor and accuracy of the determination result.

[0044] In some embodiments, step S340 involves determining the target probability value and outputting the determination result, including: A binary classification target determination method is used to determine the target probability value and output the determination result; wherein, When the target probability value is greater than or equal to 0.9, the judgment result is compliant; When the target probability value is less than 0.9, the result is deemed non-compliant.

[0045] In this embodiment, a binary classification target determination method is used to determine compliance, which transforms continuous target prediction probability values ​​into clear classification decision results, thereby meeting the requirements for compliance / non-compliance determination and enabling the output of the pin mark inspection model to be directly applied to the pre-production checkpoint determination process.

[0046] Furthermore, this embodiment uses a binary classification target determination method to set the target probability value as a fixed technical parameter, and uses precision and recall as core evaluation indicators to evaluate the accuracy of the binary classification target determination results, thereby better completing the compliance determination of actions based on pin mark detection in wafer test.

[0047] In step S400, the pre-production checkpoint locking function in this embodiment employs a combination of hard interface linkage and soft logic locking to achieve "abnormalities trigger checkpoint locking, no unlocking without rectification," thereby better preventing abnormalities from entering production. The hard interface linkage in this embodiment involves bidirectional data interaction between the automated monitoring system and the manufacturing execution system via the OPC UA (Open Platform Communications Unified Architecture) industrial communication interface. Here, the communication latency is less than or equal to 1 second, ensuring the real-time performance of the checkpoint control.

[0048] Furthermore, the triggering conditions for the pre-production checkpoint in this embodiment include: completion of probe station test firing, initiation of automated monitoring process for needle mark inspection, and determination result of needle mark inspection model. Only when the above three technical nodes are completed in sequence can the manufacturing execution system trigger the pre-production checkpoint; otherwise, the pre-production checkpoint triggering request will be directly rejected.

[0049] In step S400, when the determination result is non-compliant, the manufacturing execution system obtains a production prohibition instruction and forcibly locks core operations such as production start-up, wafer loading, and probe station movement, and the locked state cannot be manually released.

[0050] In this embodiment, a pre-production checkpoint locking function is adopted. When an abnormality is detected in the key action of needle mark inspection, the manufacturing execution system can be directly locked to prevent core operations such as production start-up, thus preventing the abnormality from flowing into the subsequent production process and further ensuring production quality and reliability.

[0051] In step S400, when the pre-production checkpoint lock is triggered, all checkpoint triggering, locking and unlocking operations generate checkpoint records with timestamps. These checkpoint records are synchronized to the manufacturing execution system's quality management, which can better realize the full-process traceability function of the pin mark inspection after wafer trial.

[0052] In step S500, after rectification is completed, a re-judgment process is triggered, and the wafer is tested again using a probe to obtain multi-feature data during the wafer test and re-output the judgment result. The judgment result is retrieved, and the subsequent wafer production operation is started only when the judgment result is compliant.

[0053] Here, the specific re-judgment logic includes: After the work unit triggers a re-judgment, the process of "multi-feature data acquisition - multi-feature data calibration - compliance judgment in the needle mark inspection model" is re-executed. Only when the new judgment result is compliant, the automated monitoring system in this embodiment sends an unlocking command to the manufacturing execution system, and the manufacturing execution system automatically unlocks all production operations.

[0054] Optionally, this embodiment establishes a full-process traceability function in the monitoring of pin marks after wafer test bonding. Based on the OPCUA industrial communication protocol, the multi-feature data during wafer test bonding, the pin mark inspection model judgment results, the manufacturing execution system checkpoint records, and the rectification results are synchronized to the EAP (Equipment Automation Program) system and the manufacturing execution system. This achieves bidirectional technical interoperability with the wafer testing EAP system and the manufacturing execution system, as well as full-process traceability and queryability of wafer test bonding data, thereby achieving the purpose of cross-system data traceability.

[0055] In some embodiments, after the rectification is completed in step S500 and the re-judgment process is triggered, after re-acquiring the multi-feature data from the wafer test and outputting the judgment result, the following steps are also included: Step S510: Perform online iteration and optimization of the needle mark inspection model.

[0056] In this embodiment, the pin mark inspection model is iterated and optimized online to dynamically adapt to the dynamic changes in pin mark detection data during wafer trial runs, thereby further improving the accuracy and efficiency of intelligent interpretation of the pin mark inspection model.

[0057] In some embodiments, the needle mark inspection model is iterated and optimized online in step S510, including: Retrieve the judgment results and generate checkpoint records. Cross-validate the judgment results, checkpoint records, and judgment results in the re-judgment process. Update the needle mark inspection model based on the cross-validation results.

[0058] In this embodiment, the judgment results, checkpoint records and wafer test results in the re-judgment process are cross-validated, which can better improve the compliance of the pin mark inspection action. By repeatedly verifying, wafer test quality abnormalities caused by pin mark deviation, no touch frame and other issues can be avoided, thus ensuring wafer test accuracy, yield and production line operating efficiency.

[0059] Optionally, after the operator completes the rectification according to the instructions, they manually trigger the re-judgment process. The manufacturing execution system then re-triggers the automated monitoring process and automatically retrieves the judgment result. Only when the judgment result output by the re-judgment process is compliant will the manufacturing execution system obtain an unlock signal to release the bottleneck. After the bottleneck is released, the judgment result, bottleneck record, and wafer test results in the re-judgment process are cross-validated. The pin mark inspection model is updated based on the cross-validation results and used in the automated monitoring of pin mark inspection after subsequent wafer trial runs.

[0060] Furthermore, based on the results of cross-validation, misjudged, missed, and new customer complaint abnormal samples are extracted. Incremental learning technology is used to re-label the misjudged, missed, and new customer complaint abnormal samples after cross-validation, and the re-labeled samples are added as incremental samples to the pin mark inspection model sample set to update the weights and thresholds of the pin mark inspection model, thus obtaining the optimized pin mark inspection model.

[0061] Meanwhile, in this embodiment, the re-labeled samples are added as incremental samples to the pin mark inspection model sample set. Only the weights and thresholds of the pin mark inspection model are updated. There is no need to retrain the full sample data (multi-feature data, calibration data, judgment results, checkpoint records, rectification and re-judgment data) in the pin mark inspection model sample set. The optimized pin mark inspection model is used for the automated inspection and monitoring of pin marks after the new round of wafer trial testing. This completes the online iterative control of the pin mark inspection model and ensures the online iterative efficiency of the pin mark inspection model.

[0062] Preferably, in this embodiment, the iteration time of the needle mark inspection model is less than or equal to 5 seconds.

[0063] In some embodiments, before step S100, in which the wafer is tested using a probe and multi-feature data is acquired during the wafer test, the method further includes: Step S010: Obtain abnormal customer complaint samples, combine the obtained abnormal customer complaint samples with the calibrated historical compliance samples to construct a labeled pin mark inspection model sample set, complete the initial training of the pin mark inspection model, and determine the initial compliance threshold.

[0064] Optionally, customer complaint anomaly samples related to pin mark inspection can be extracted, and the anomaly types and feature parameters in the customer complaint anomaly samples can be labeled to construct a labeled pin mark inspection model sample set.

[0065] The initial compliance thresholds include: displacement distance feature data greater than or equal to 1400 micrometers, execution time feature data greater than or equal to 10 seconds, trajectory coverage feature data greater than or equal to 95%, and action initiation validity data equal to 1.

[0066] Here, by setting an initial compliance threshold, a judgment standard for pin mark inspection is provided for the pin mark inspection model, thereby distinguishing whether there are abnormal issues in pin mark inspection.

[0067] For example, a displacement distance feature data greater than or equal to 1400 micrometers indicates that the operator must move at least 1400 micrometers during pin mark inspection to demonstrate a complete pin mark inspection; an execution time greater than or equal to 10 seconds indicates that the operator must spend at least 10 seconds performing the pin mark inspection action to prove that the inspection trajectory coverage is >95% and serves its purpose. During the initial training phase of the pin mark inspection model, clear criteria for judging the presence of anomalies in pin mark inspection are established to avoid operator misjudgments and missed detections.

[0068] By setting an initial compliance threshold and using it as a pre-judgment condition for pre-production checkpoint, abnormal conditions are prevented from flowing into production. While the mark inspection model is used for compliance judgment of calibration data, when the mark inspection model determines that the data meets the initial compliance threshold, the judgment result is qualified, and subsequent wafer production operations are started; when the mark inspection model determines that the data does not meet the initial compliance threshold, the judgment result is unqualified, and the manufacturing execution system immediately triggers the locking of the pre-production checkpoint, prohibits all production operations, and prevents abnormalities from flowing into the next process, which is used to intercept abnormalities after trial probing is completed and before formal production, eliminate the outflow of defective products from the source, and avoid subsequent rework and customer complaints.

[0069] By setting the initial compliance threshold, a reference anchor is provided for calibration data and data iteration, and the initial compliance threshold is used as the benchmark in the three-level real-time calibration technology and the online iteration and optimization of the mark inspection model. In the three-level real-time calibration technology, the initial compliance threshold is used to correct systematic / random errors to ensure the accuracy of collected data; in the initial compliance threshold, by dynamically adjusting the threshold (for example, optimizing the displacement threshold from 1400 microns to 1350 microns), the mark inspection model is made more accurate.

[0070] Optionally, in step S010, customer complaint abnormal samples are obtained, and the obtained customer complaint abnormal samples are combined with calibrated historical compliant samples to construct a labeled sample set for the mark inspection model, wherein the ratio of customer complaint abnormal samples to calibrated historical compliant samples is 3:7.

[0071] Wherein, in this embodiment, by constructing a labeled sample set for the mark inspection model, the mark inspection model can accurately obtain the judgment relationship between multi-feature data and customer complaint abnormal causes, which improves the training accuracy and prediction precision of the mark inspection model, ensures that the mark inspection model is more reliable in the automatic monitoring of actual mark inspection, and the training and optimization effect is more objective.

[0072] Here, the ratio of customer complaint abnormal samples to calibrated historical compliant samples is 3:7 to construct the labeled sample set for the mark inspection model. The labeled sample set for the mark inspection model is divided into 80% training set and 20% verification set. The training set and the verification set are divided by a method combining data normalization technology and random oversampling technology, wherein the random oversampling technology is used to solve the problem of sample imbalance and ensure the generalization ability of the mark inspection model.

[0073] Reference Figure 2 As shown, based on the same inventive concept, this embodiment provides a mark monitoring system after wafer trial probing, comprising: The multi-source data acquisition and calibration module is used to perform test runs on the wafer using a probe, acquire multi-feature data during the test runs, calibrate the multi-feature data, and generate calibration data. The pin mark inspection module is used to interpret the calibration data for compliance using the pin mark inspection model and output the judgment result, which is either compliant or non-compliant. The pre-production checkpoint locking module is used to retrieve the judgment result and generate checkpoint records. When the judgment result is non-compliant, the manufacturing execution system immediately triggers the pre-production checkpoint locking, prohibits all production operations, and performs hierarchical real-time warnings and rectification. The re-judgment and verification module is used to trigger the re-judgment process after rectification is completed, re-acquire the multi-feature data of the wafer trial production and output the judgment result, retrieve the judgment result, and start the subsequent wafer production operation only when the judgment result is compliant.

[0074] In this embodiment, a pin mark monitoring system is established after wafer trial bonding. A lightweight pin mark inspection model is selected to train the fused feature data and output compliance judgment results. This system is used to optimize the industrial scenario where manual pin mark inspection is low-dimensional, features-intensive, and requires real-time interpretation. While taking into account the interpretation accuracy of automated inspection and monitoring of pin marks after wafer trial bonding, it also improves the real-time performance of automated monitoring of pin mark inspection in the production line.

[0075] Here, in the multi-source data acquisition and calibration module, the scanning post-manufacturing inspection function of the probe station is combined with the NTP time synchronization transmission method. The probe position is located at the work end and multiple feature data are collected through the equipment operation log interface. The four core feature data of the needle mark inspection are collected and calibrated in a targeted manner: execution time feature data, displacement distance feature data, trajectory coverage feature data, and action start effectiveness data.

[0076] After acquiring the calibrated multi-feature data, the probe station test completion operation command is used as the sole technical trigger signal to automatically start the monitoring process. By locating the feature changes of the signal at the work end, the actual start and completion nodes of the needle mark inspection are intelligently identified without manual marking, thus avoiding missed or false detections.

[0077] In the pin mark inspection module, valid samples related to "non-standard pin mark inspection / no inspection leading to pin problems" are extracted from the abnormal customer complaint samples. Four types of abnormal labels and corresponding feature parameters are labeled: not actually executed, insufficient displacement, too short duration, and incomplete trajectory coverage. Combined with calibrated historical compliant samples, a labeled pin mark inspection model sample set is constructed to determine the initial compliance threshold of the action.

[0078] Meanwhile, based on the trained and optimized pin mark inspection model, the calibrated data is intelligently checked, and a unique technical judgment result of compliance / non-compliance of a single pin mark inspection action is output. At the same time, core monitoring indicators such as actual execution rate, compliance rate, and abnormality rate are generated according to shift, equipment, and operator, which are used to automatically monitor and evaluate the standardization of operators' work.

[0079] In the pre-production checkpoint locking module, the graded real-time early warning includes: based on the judgment results of the pin mark inspection model, the degree of operational abnormality of the pin mark inspection after wafer trial is divided into serious abnormality (not actually performed) and general abnormality (inspection not standardized). The abnormality details, precise rectification guidance and checkpoint reasons are pushed to the operation end through the production line terminal, and the abnormality reminder, key information such as the equipment / personnel / time involved are pushed to the management end, realizing minute-level early warning, thereby realizing precise control of operational abnormality monitoring of pin mark inspection after wafer trial.

[0080] Optionally, the pre-production checkpoint locking module includes a checkpoint determination module, and the core technical logic of the pre-production checkpoint locking module includes: Step S401: Probe station test completed, automated monitoring process for needle mark inspection started, and needle mark inspection model judgment result; Step S402: The work unit submits a production start request to the Manufacturing Execution System; Step S403: Trigger the pre-production bottleneck judgment module of the manufacturing execution system; Step S404: The checkpoint determination module retrieves the determination results of the needle mark inspection model; Step S405: Set up uniqueness verification to verify the validity of the data obtained during wafer trial production; Step S406: Output the compliance / non-compliance determination result; Step S4061: If non-compliant, the pre-production checkpoint judgment module of the Manufacturing Execution System sends a "pin mark inspection non-compliant - production interception" signal to the Manufacturing Execution System; Step S4062: The Manufacturing Execution System locks all core production operations; Step S4063: Trigger an abnormality warning, and at the same time, the work terminal receives the rectification instruction "Needle marks not checked, please check"; Step S4064: The work unit completes the pin mark inspection and rectification according to the instructions, manually triggers the pin mark monitoring system after wafer trial to re-collect and calibrate, inputs the calibration data into the pin mark inspection model and performs compliance judgment, and outputs the compliance / non-compliance judgment result; Step S4066: If compliant, send the “Pin mark inspection compliant - production release” signal; Step S407: The manufacturing execution system records the release data and sends an automatic start command to the probe station.

[0081] In the re-judgment and verification module, after the operator completes the rectification of the abnormality of the pin mark inspection according to the warning guidance, the operator manually triggers the system to re-collect data and re-judge the compliance of the pin mark inspection model. Only when the re-judgment result is compliant, the re-judgment and verification module sends an unlock signal to the manufacturing execution system to unlock the production operation and restore normal production; if the re-judgment is still non-compliant, the blockage lock status will continue to be maintained.

[0082] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0083] The embodiments of the various products and devices in this application are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For ease of description, the above devices are described in functional modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0084] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the above embodiments of this application, which are not provided in detail for the sake of brevity.

[0085] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0086] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description.

[0087] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A method for monitoring pin marks after wafer trial bonding, characterized in that, include: A probe is used to test-brick the wafer, and multi-feature data of the wafer during the test-brick process is acquired; The multi-feature data is calibrated, and calibration data is generated; The calibration data is interpreted for compliance using a pin mark inspection model, and the judgment result is output; wherein the judgment result is compliant or non-compliant. When the judgment result is non-compliant, the manufacturing execution system immediately triggers the pre-production checkpoint lock, prohibits all production operations, and performs graded real-time early warning and rectification. After the rectification is completed, a re-judgment process is triggered to reacquire the multi-feature data of the wafer test and output the judgment result. If the determination result is compliant, the subsequent wafer production operation will be initiated.

2. The method for monitoring pin marks after wafer trial bonding according to claim 1, characterized in that, The multi-feature data is calibrated, and calibration data is generated, wherein, The calibration data includes: continuous feature data and categorical feature data; The continuous feature data includes: execution time feature data, displacement distance feature data, and trajectory coverage feature data; The categorized feature data includes: action initiation validity data.

3. The method for monitoring pin marks after wafer trial bonding according to claim 2, characterized in that, The calibration of the multi-feature data and the generation of calibration data include: The multi-feature data is calibrated, and continuous feature data is generated; The continuous feature data is subjected to range normalization.

4. The method for monitoring pin marks after wafer trial bonding according to claim 1, characterized in that, The calibration of the multi-feature data and the generation of calibration data include: The multi-feature data is calibrated using a three-level real-time calibration technique, and calibration data is generated. The three-level real-time calibration technique includes spatial calibration, time synchronization calibration, and error compensation calibration.

5. The method for monitoring pin marks after wafer trial bonding according to claim 1, characterized in that, The needle mark detection model includes a lightweight gradient boosting tree and a logistic regression model; The process of using a needle mark inspection model to interpret calibration data for compliance and outputting the judgment result includes: A lightweight gradient boosting tree model is used to process the calibration data to output a first probability value; A logistic regression model is used to enhance the first probability value in order to output a second probability value; The first probability value and the second probability value are fused to generate the target probability value; The target probability value is determined and the determination result is output.

6. The method for monitoring pin marks after wafer trial bonding according to claim 5, characterized in that, The step of determining the target probability value and outputting the determination result includes: A binary classification target determination method is used to determine the target probability value and output the determination result; wherein, when the target probability value is greater than or equal to 0.9, the determination result is compliant; when the target probability value is less than 0.9, the determination result is non-compliant.

7. The method for monitoring pin marks after wafer trial bonding according to claim 1, characterized in that, After completing the rectification and triggering the re-judgment process, after re-acquiring the multi-feature data from the wafer test and outputting the judgment result, the process further includes: The needle mark inspection model is iterated and optimized online.

8. The method for monitoring pin marks after wafer trial bonding according to claim 7, characterized in that, The online iteration and optimization of the needle mark inspection model includes: The judgment result is retrieved and a checkpoint record is generated. The judgment result, the checkpoint record and the judgment result in the re-judgment process are cross-validated. The needle mark inspection model is updated according to the cross-validation result.

9. The method for monitoring pin marks after wafer trial bonding according to claim 1, characterized in that, Before using a probe to test-brick the wafer and acquire multi-feature data during the test-brick process, the method further includes: Obtain abnormal customer complaint samples, combine the obtained abnormal customer complaint samples with calibrated historical compliance samples to construct a labeled pin mark inspection model sample set, complete the initial training of the pin mark inspection model, and determine the initial compliance threshold.

10. A system for monitoring pin marks after wafer trial bonding, characterized in that, include: The multi-source data acquisition and calibration module is used to perform test runs on the wafer using a probe, acquire multi-feature data during the test runs, calibrate the multi-feature data, and generate calibration data. The pin mark inspection module is used to perform compliance judgment on calibration data using the pin mark inspection model and output the judgment result; wherein the judgment result is compliant or non-compliant; The pre-production checkpoint locking module is used to retrieve the judgment result and generate a checkpoint record. When the judgment result is non-compliant, the manufacturing execution system immediately triggers the pre-production checkpoint locking, prohibits all production operations, and performs graded real-time early warning and rectification. The re-judgment and verification module is used to trigger the re-judgment process after the rectification is completed, re-acquire the multi-feature data of the wafer trial production and output the judgment result, retrieve the judgment result, and start the subsequent wafer production operation only when the judgment result is compliant.

Images