An automated method and system for handling misprocessing in wafer testing

Abstract

This invention relates to the field of wafer testing technology, specifically to an automated method and system for handling false positives during wafer testing. The method involves a server that presets false positive blocking conditions and corresponding handling methods, and connects to a probe prober and a testing machine. The testing machine uploads real-time test data to the server. When the server determines a false positive based on the preset conditions, it controls the probe prober to stop testing and issues an alarm, while simultaneously automatically generating a device anomaly report. Operators operate the probe prober according to the handling methods issued by the server, and the probe prober generates an operation record and uploads it in real time. The server compares the operation record with the preset methods for consistency and receives the retest results from the testing machine. If the retest results are satisfactory, the server saves the record, restores the device status, and optimizes and updates the handling method by parsing the operation record. This invention achieves a fully automated closed-loop process for false positive detection, handling, verification, and optimization, improving testing efficiency and quality control.

Description

Technical Field

[0001] This invention relates to the field of wafer testing technology, and specifically to an automated method and system for handling errors in wafer testing. Background Technology

[0002] Wafer testing is a critical step in semiconductor manufacturing, and its quality directly affects chip yield and reliability. During wafer testing, false positives frequently occur due to factors such as probe condition, probe pressure, poor contact, and test parameter settings; that is, qualified chips are mistakenly identified as defective chips. Traditional methods for handling false positives rely on manual experience: operators check, adjust, and retest the probe tester based on their judgment, which has significant limitations. 1. Inconsistent and non-standardized processing procedures, with significant differences in operation among different personnel, lead to unstable processing results; 2. Manual processing is inefficient, results in long downtime, and affects testing capacity; 3. The operation process is difficult to record and trace completely, making it impossible to pinpoint the root cause of quality problems; 4. The testing machine, probe tester, and management system are independent of each other, resulting in significant data silos and making automated linkage impossible; 5. The processing methods rely on experience and cannot be continuously optimized based on historical data. While some existing automated testing systems have anomaly alert functions, they generally suffer from the following shortcomings: It is impossible to achieve integrated linkage of automatic judgment of accidental slaughter, automatic shutdown, and automatic issuance of abnormal orders; It is impossible to perform standardized verification and consistency comparison of personnel operations; There is a lack of intelligent learning and method optimization based on test bins and operation records; The equipment lacks strong interoperability and has not formed a closed-loop automated processing flow.

[0003] Therefore, the industry urgently needs an automated, standardized, traceable, and sustainably optimized solution for handling wafer defects to improve wafer testing efficiency and quality control. To overcome the shortcomings of existing technologies, this invention provides an automated method and system for handling wafer defects during testing, achieving closed-loop control throughout the entire process, including defect detection, shutdown alarm, abnormal start-up, standardized operation, consistency verification, retesting and release, and intelligent optimization. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention provides an automated method and system for handling wafer defects during testing, realizing closed-loop control of the entire process, including defect detection, shutdown alarm, abnormal start-up, standardized operation, consistency verification, retest release, and intelligent optimization.

[0005] To achieve the above-mentioned technical objectives, according to a first aspect of the present invention, the present invention provides an automated method for handling errors in wafer testing, comprising: S100: The server presets the conditions for accidental slaughter and the corresponding accidental slaughter handling methods, and the server communicates with the needle testing machine and the testing machine via the network. The S200 tester uploads real-time test data to the server. When the server determines that a slaughter has been mistakenly committed based on the preset conditions for mis-slaughtering, it controls the needle tester to stop testing and issue an alarm, while automatically creating a device malfunction report. S300: The operator performs the corresponding operation on the needle testing machine according to the error handling method issued by the server. The needle testing machine generates operation records in real time and uploads them to the server. S400. The server compares the operation record with the preset error handling method and receives the retest result from the test machine. S500: If the retest result is qualified, the server saves the operation record, restores the equipment status from abnormal to production status, and optimizes and updates the error handling method by parsing the operation record.

[0006] Specifically, in S100, the conditions for accidental killing include at least one of the following: the electrical parameter threshold of the test program, the needle pressure setting of the probe tester, and the probe contact state parameters.

[0007] Specifically, in S200, the alarm includes an audible and visual alarm and a prompt from the host computer software; the equipment anomaly report includes the type of mis-slaughter, the time of occurrence, the equipment number, and the test location information.

[0008] Specifically, in S300, the operation record includes at least one action information among needle tip inspection, needle position adjustment, needle pressure modification, and probe cleaning.

[0009] Specifically, in step S400, if the server compares and determines that the operation record is inconsistent with the preset error handling method, it prompts the operator to re-execute the standard operation.

[0010] Specifically, in S400, the retest results include electrical test results for open circuit, short circuit, and leakage.

[0011] Specifically, in S500, the method for optimizing the error handling by parsing operation records includes: classifying and clustering operation records according to the Bin of wafer testing, extracting high-frequency effective operations, and generating an optimized processing flow.

[0012] Specifically, the optimized method for handling accidental slaughter is automatically updated to the server for use in handling subsequent accidental slaughter incidents.

[0013] Specifically, the method for handling accidental slaughter includes at least one of the following operations: checking for needle dirt, adjusting needle position, verifying open circuit, and verifying short circuit.

[0014] According to a second aspect of the present invention, an automated processing system for errors in wafer testing is provided, comprising: The server is used to store the conditions for accidental slaughter and the methods for handling accidental slaughter, and is connected to the probe tester and the testing machine for communication. The server is also used to: receive real-time test data uploaded by the testing machine and determine accidental slaughter based on the conditions for accidental slaughter; when accidental slaughter is determined, send a stop command to the probe tester and automatically create a device abnormality report; send the methods for handling accidental slaughter to the client; receive the operation records uploaded by the probe tester and compare them with the preset methods for handling accidental slaughter; receive the retest results fed back by the testing machine; and after the retest results are qualified, save the operation records and update the device status.

[0015] The probe tester is connected to the server and is used to perform wafer testing operations, receive and execute stop commands from the server, and generate operation records and upload them to the server.

[0016] The testing machine is connected to the server and is used to test wafers and upload real-time test data to the server, as well as to perform retests according to the server's instructions or settings and provide feedback on the retest results.

[0017] Beneficial effects: The automated processing method and system for wafer testing provided by this invention have significant advantages over existing technologies.

[0018] First, by centralizing judgment and process control through the server, more accurate and unified handling of misjudgments is achieved. The testing machine is only responsible for uploading real-time test data, while the server judges misjudgments based on preset checkpoint conditions and automatically triggers subsequent shutdowns, alarms, order creation, and process guidance. This approach unifies the judgment criteria, avoids misjudgments or omissions caused by differences in settings across different testing machines, and achieves fully automated triggering from anomaly detection to processing guidance, significantly reducing manual intervention and response time, and improving processing efficiency.

[0019] Secondly, the system establishes a standardized, mandatory closed-loop "operation-verification" process to ensure processing quality. The system not only distributes standardized operating procedures but also compels operators to follow the specifications by comparing operation records with preset methods in real time. This mechanism ensures that every case of miscarriage is handled in accordance with best practices, eliminates fluctuations in processing results caused by differences in personnel experience, and significantly improves the controllability of the processing process and the consistency of results.

[0020] Third, this invention introduces a data-driven artificial intelligence optimization mechanism, enabling the self-evolution of the processing method. The system clusters successful operation records by wafer bin, automatically extracts high-frequency and effective operation patterns, and generates optimized processing flows. This closed loop of "learning-optimization-application" allows the system's processing method to continuously absorb best online experience, constantly improve itself, thereby continuously increasing processing success rate and efficiency, and reducing long-term production capacity losses.

[0021] Finally, this invention breaks down data silos between devices by deeply integrating the probe tester and the testing machine into a server, forming a collaborative and intelligent operating platform. The server acts as the brain, unifying scheduling, calculation, decision-making, and optimization, while the probe tester and the testing machine serve as execution units. This architecture not only achieves a high degree of automation in the current process but also provides a scalable core framework for connecting more devices and realizing more complex intelligent management and control of production lines in the future. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the automated processing method for wafer testing errors provided in a specific embodiment of the present invention. Figure 2 This is a schematic diagram of the system composition of the automated processing system for wafer testing errors provided in a specific embodiment of the present invention. Detailed Implementation

[0023] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments in this application, other similar embodiments obtained by those skilled in the art without creative effort should all fall within the scope of protection of this application. Furthermore, directional terms mentioned in the following embodiments, such as "up," "down," "left," and "right," are only for reference to the directions in the accompanying drawings; therefore, the directional terms used are for illustrative purposes and not for limiting the invention.

[0024] The present invention will be further described below with reference to the accompanying drawings and preferred embodiments.

[0025] Please see Figure 1 This invention provides an automated method for handling errors in wafer testing, comprising: S100. The server presets the conditions for accidental slaughter and the corresponding accidental slaughter handling methods, and the server communicates with the probe tester and the testing machine via the network.

[0026] It should be further explained that in the S100, the server presets the conditions for accidental sterilization and the corresponding handling methods, specifically implemented as follows: The server uses industrial-grade server hardware, such as an Intel Xeon processor, 32GB of memory, and a 1TB hard drive, and installs a Linux operating system and custom software modules. The server communicates with the probe tester and the testing machine via TCP / IP protocol, with a network bandwidth of no less than 100Mbps to ensure real-time data transmission. In specific operation, the server receives the conditions and methods set by the user through the configuration interface and establishes a mapping relationship between them. For example, the sterilization condition "open circuit voltage > 0.7V" is associated with the handling method "check for dirty probes" and stored accordingly. The sterilization conditions include the parameter thresholds of the testing program and the equipment parameters of the probe tester. These conditions are input through the server's graphical user interface (GUI) and stored in the MySQL database. Parameter thresholds, such as the open-circuit voltage threshold, are set to 0.5V to 1.0V (preferably 0.7V, as this value, based on typical wafer test data, balances detection sensitivity and false alarm rate, avoiding missed detections due to excessively high thresholds or false alarms due to excessively low thresholds), and the short-circuit current threshold is set to 1mA to 5mA (preferably 2mA, as this value conforms to the electrical characteristics of wafer materials and can effectively identify short-circuit anomalies). Probe tester parameters include probe pressure settings, ranging from 5g to 20g (preferably 10g, as this pressure ensures reliable contact between the probe tip and the wafer pad, avoiding poor contact or wafer damage). Error handling methods are stored in XML format, containing operation sequences, such as first checking for probe dirt, then adjusting the probe position.

[0027] The S200 testing machine uploads real-time test data to the server. When the server determines that a slaughter has been mistakenly executed based on the preset conditions for mis-slaughtering, it controls the needle testing machine to stop testing and issue an alarm, while automatically creating a device malfunction report.

[0028] It should be further explained that in the S200, during the testing process, the testing machine continuously uploads real-time test data (such as voltage, current, and resistance values) to the server via a custom API interface. The server runs a real-time data stream processing module, which continuously compares the received real-time data with the pre-set error-causing blocking conditions in the S100. When the server determines that the real-time data meets any error-causing blocking condition, it triggers the error-causing processing procedure. The server immediately sends a stop command to the probe tester via the Modbus protocol. The probe tester stops moving and triggers an alarm device, which includes a buzzer (sound frequency of 2kHz to 4kHz, preferably 3kHz, because this frequency is sensitive to the human ear and easily identifiable in noisy environments) and a red LED light (flashing frequency of 1Hz to 3Hz, preferably 2Hz, because this frequency provides a prominent visual cue). At the same time, the server automatically creates a device anomaly report in the database. The anomaly report fields include anomaly ID, timestamp (accurate to milliseconds), error-causing type, device ID, and operator ID.

[0029] S300: The operator performs the corresponding operation on the needle testing machine according to the error handling method issued by the server. The needle testing machine generates operation records in real time and uploads them to the server.

[0030] It should be further explained that in the S300, operators can view the error handling methods issued by the server through the server GUI, such as displaying "Check needle tip contamination" or "Adjust needle mark position" on the interface. When personnel operate the needle testing machine, the built-in sensors (such as pressure sensors and position encoders) generate operation records, which include operation time, action type (such as needle tip cleaning, needle pressure adjustment), and parameter values ​​(such as needle pressure change). The operation records are uploaded to the server in real time in JSON format via the WebSocket protocol, with an upload latency of less than 100 milliseconds.

[0031] S400: The server compares the operation record with the preset error handling method for consistency and receives the retest results fed back by the test machine.

[0032] It should be further explained that in S400, the server uses a rule engine to perform a consistency comparison between the operation record and the preset error handling method. The rule engine is implemented based on the Drools framework. The comparison algorithm calculates the similarity between the operation record and the preset method, and the similarity threshold is set to 90% (preferred value, because this threshold can tolerate slight deviations in the operation while ensuring consistency of the core steps). The similarity calculation formula is as follows: Similarity The number of matched actions represents the number of actions in the operation record that match the preset method, while the total number of actions in the preset method represents the total number of standard actions defined in the preset method. Physical meaning: The similarity percentage reflects the degree of conformity of the user's operation. If the similarity is below 90%, the server sends a prompt message to the GUI, requiring the user to repeat the operation. Simultaneously, the server receives the retest results from the test machine, including a pass or fail status.

[0033] S500: If the retest result is qualified, the server saves the operation record, restores the equipment status from abnormal to production status, and optimizes and updates the error handling method by parsing the operation record.

[0034] It should be further explained that in the S500, if the retest result is qualified, the server archives the operation record to the database and updates the status of the abnormal order from "abnormal" to "running". Simultaneously, the parsing module is started. This module uses a clustering algorithm (such as K-means) to analyze the operation record, grouping it according to the wafer's Bin (e.g., Bin1 to Bin8). The model training steps are: data preprocessing (normalizing operation record features, such as operation time and action type), cluster analysis (cluster number K=5, the preferred value, based on historical data distribution, effectively grouping common operation patterns), and generating optimization rules (e.g., simplifying frequent actions). The training data uses operation records from the past 30 days, with a training cycle of once per day. The optimized error handling method is automatically updated to the database, thus completing the optimization update of the processing method.

[0035] Understandably, this method automates and standardizes the entire process of handling miscarried animals, reducing manual intervention and wasted testing time. Through real-time recording and optimization, continuous improvement is achieved, reducing the miscarriage rate and production capacity loss. Server integration ensures data traceability and improves quality control efficiency. Those skilled in the art can clearly implement this method based on the above details, for example, by configuring a standard server and a commercial needle testing machine, and writing software modules to achieve the aforementioned functions.

[0036] Specifically, in S100, the conditions for accidental killing include at least one of the following: the electrical parameter threshold of the test program, the needle pressure setting of the probe tester, and the probe contact state parameters.

[0037] It should be further explained that this invention clarifies the specific conditions for false detection. The electrical parameter thresholds of the test procedure include key electrical parameters in wafer testing. For example, the open-circuit voltage threshold is set to 0.5V to 1.0V, with a preferred value of 0.7V. The rationale is based on extensive wafer testing statistics; 0.7V can effectively distinguish between a true open circuit and noise interference, avoiding misclassifying normal wafers as abnormal. The short-circuit current threshold is set to 1mA to 5mA, with a preferred value of 2mA, because this value corresponds to the current range of typical wafer short-circuit faults, reducing missed detections. The probe pressure setting range for the probe tester is 5g to 20g, with a preferred value of 10g. The rationale is that experimental verification shows 10g probe pressure ensures stable contact between the probe tip and the aluminum pad while avoiding excessive pressure that could damage the pad. Probe contact state parameters include the contact resistance threshold, set to 0.1Ω to 1Ω, with a preferred value of 0.5Ω. The rationale is that this threshold can effectively identify abnormally high resistance states caused by probe contamination, oxidation, or poor contact, and is one of the important bases for judging false detection. These conditions are set through the server's configuration interface, which provides drop-down menus and numerical input boxes. The server stores the conditions as a configuration file in JSON format, for example: {"Electrical parameter threshold"; {"Open circuit voltage"; 0.7, "Short circuit current"; 2}, "Needle pressure setting"; 10, "Contact resistance"; 0.5}.

[0038] Understandably, by specifying the threshold values ​​of electrical parameters, needle pressure settings, and contact state parameters, this method improves the accuracy and multidimensionality of false kill detection and reduces the limitations of relying on a single judgment criterion; the optimal value is based on empirical data, ensuring the adaptability and reliability of the system under different wafer types and testing environments.

[0039] Specifically, in S200, the alarm includes an audible and visual alarm and a prompt from the host computer software; the equipment anomaly report includes the type of mis-slaughter, the time of occurrence, the equipment number, and the test location information.

[0040] It should be further explained that in the S200, during the testing process, the testing machine continuously uploads real-time test data (such as voltage, current, and resistance values) to the server via a custom API interface. The server runs a real-time data stream processing module, which continuously compares the received real-time data with the preset error-causing blocking conditions in the S100. When the server determines that the real-time data meets any error-causing blocking condition, it triggers the error-causing processing procedure. The server immediately sends a stop command to the probe tester via the Modbus protocol, and the probe tester stops moving and triggers the alarm device. The specific implementation of the audible and visual alarm: The probe tester integrates an audible and visual alarm module. The audible alarm uses a piezoelectric buzzer with a frequency range of 2kHz to 4kHz, preferably 3kHz, chosen because this frequency is within the most sensitive range for the human ear and can be clearly distinguished in a noisy factory environment. The visual alarm uses a high-brightness red LED with a flashing frequency of 1Hz to 3Hz, preferably 2Hz, because this frequency can attract visual attention without causing fatigue. The host computer software prompts the issue via the server GUI. For example, a red warning window pops up on the monitoring interface, displaying "Mistake Alarm: Open Circuit Anomaly Detected," and simultaneously sends push notifications or SMS messages to the terminals of relevant operators. At the same time, the server automatically creates a device anomaly report in the database. The report fields include anomaly ID, mistake type (string, such as "open circuit," "short circuit," "poor contact"), occurrence time (timestamp, format YYYY-MM-DDHH; MM; SS.sss, accurate to milliseconds), device number (unique identifier for probe tester or testing machine), and test location information (such as wafer number, die coordinates X, Y).

[0041] Understandably, centralized server-side fault diagnosis achieves unified logic and centralized management, avoiding inconsistencies in the judgment logic of different testing machines. Multi-mode (audible, visual, and software) alarms ensure that anomalies can be detected promptly and comprehensively, minimizing response delays; equipment anomaly reports containing multi-dimensional information provide a complete and structured data foundation for subsequent problem tracing, equipment management, and production analysis.

[0042] Specifically, in S300, the operation record includes at least one action information among needle tip inspection, needle position adjustment, needle pressure modification, and probe cleaning.

[0043] It should be further explained that in the S300, operators view the automatically issued handling instructions (i.e., miscarriage handling methods) on the server GUI and follow the steps to operate the needle testing machine. The various sensors built into the needle testing machine synchronously generate structured operation records during operator operation. Specific design features include: (1) Needle tip inspection procedure: Personnel visually inspect the needle tip using a high-powered microscope (e.g., 100x magnification) of the needle probe machine, or trigger an automatic image acquisition system (camera resolution not less than 1080p) to capture images of the needle tip. The image processing algorithm is based on the OpenCV library, calculates the grayscale difference of the image, sets the threshold to be greater than 50 (preferred value, because this value can effectively identify obvious contrast changes caused by common contaminants such as oxide layers or residues, while ignoring slight shadows caused by light), and records the inspection results (e.g., "clean" or "contaminated").

[0044] (2) Needle position adjustment action: The operator inputs the offset command through the needle probe machine control software, and the needle probe machine performs position adjustment through a high-precision stepper motor. The system records the adjustment action, including the adjustment direction (X, Y, Z), adjustment amount (unit μm, adjustment accuracy is ±1μm, the preferred value, because this accuracy can meet the alignment requirements of most wafer pads), starting coordinates and target coordinates.

[0045] (3) Needle pressure modification action: This is achieved through a digital pressure sensor. The operator sets the target pressure value in the software interface. The system records the modification action, including the pressure value before modification and the pressure value after modification (the change range is 5g to 20g, in 1g increments).

[0046] (4) Probe cleaning action: The operator performs the cleaning procedure provided with the probe testing machine or uses a dedicated cleaning tool. The system records the type of cleaning action (such as "alcohol wipe" or "vacuum adsorption") and the cleaning duration. All operation records are encapsulated in JSON format and uploaded to the server in real time via the Web Socket protocol. Each record includes a timestamp, action type, action parameters, and result status. The upload network latency is required to be less than 100 milliseconds.

[0047] Understandably, detailed and structured operation records ensure that every step of the process is traceable and auditable; provide rich and high-quality data sources for subsequent analysis; and digitize personnel operations, which is the foundation for achieving operation standardization and process comparison.

[0048] Specifically, in step S400, if the server compares and determines that the operation record is inconsistent with the preset error handling method, it prompts the operator to re-execute the standard operation.

[0049] It should be further explained that in S400, after the server receives the actual operation records uploaded by the probe, it immediately initiates a consistency comparison process. The server uses a rule engine (such as one based on the Drools framework) for comparison. Preset error handling methods are stored on the server in a specific structure (such as instruction sequences in XML or JSON format). The comparison algorithm calculates the similarity between the actual operation record sequence and the preset method sequence. The similarity threshold is set to 90% (preferred value, because this threshold can tolerate minor adjustments to the order of non-critical steps in the operation or small errors in recording time, but ensures that core processing logic and critical steps are not missed). The similarity calculation formula is as follows: The number of matched actions represents the number of actions in the operation record that are consistent with the standard actions defined in the preset method in terms of type and key parameters. The total number of actions in the preset method represents the total number of standard actions defined in the preset method. This percentage reflects the degree to which the actual operation performed by the operator conforms to the standard procedure. If the calculated similarity is less than 90%, the server immediately displays a prominent warning window to the operator through the GUI, showing "Operation procedure inconsistent, please strictly follow the standard procedure to reoperate," and records this inconsistency. At the same time, the server will lock the next retest instruction until it receives an operation record that conforms to the standard. The comparison process is carried out in the server's memory to ensure response speed; the overall response time from receiving the operation record to providing the comparison result is less than 1 second. At the same time or after completing the comparison, the server receives the retest results from the testing machine after retesting the processed wafer.

[0050] Understandably, the automated real-time consistency comparison mechanism acts as a "virtual inspector," ensuring that every case of mis-slaughter is handled in accordance with the preset standard operating procedures (SOPs). This completely eliminates inconsistencies in the processing flow caused by human negligence, differences in experience, or arbitrary operations, fundamentally improving the stability and reliability of processing quality.

[0051] Specifically, in S400, the retest results include electrical test results for open circuit, short circuit, and leakage.

[0052] It should be further explained that the retest results received by the server are obtained after the operator has adjusted the probe tester and re-executed the test program on the original erroneous location or related area. These results specifically include the following types of electrical test results: (1) Open circuit test result: The tester measures the resistance between two specific points. If the resistance value is greater than the preset open circuit resistance threshold (e.g., 1MΩ, preferred value, because this value corresponds to the high resistance state of a typical open circuit fault), it is determined to be "open circuit"; otherwise, it is "non-open circuit".

[0053] (2) Short circuit detection results: The tester measures the conduction resistance between two specific points. If the resistance value is less than the preset short circuit resistance threshold (e.g., 10Ω, preferred value, because this value can effectively identify low resistance short circuits caused by metal bridging, dielectric layer breakdown, etc.), it is determined to be a "short circuit"; otherwise, it is a "non-short circuit".

[0054] (3) Leakage detection results: The tester measures the leakage current under a certain voltage. If the current value is greater than the preset leakage current threshold (e.g., 0.1μA, preferred value, because this value is based on the insulation performance specifications of the wafer and device and can effectively capture defects such as pn junction leakage and gate oxide leakage), it is judged as "leakage"; otherwise, it is "non-leakage". The retest results are sent to the server in the form of structured digital signals, for example, through GPIB or Ethernet interface. The data format is: {"Test type"; "Open circuit", "Measured value"; 1.5, "Unit"; "MΩ", "Judgment"; "Pass"}. The server parses and stores these results for the state transition judgment of S500.

[0055] Understandably, clearly defining the specific electrical test types included in the retest results enables the system to accurately determine whether the problem of misoperation has been truly resolved. These results correspond to the checkpoint conditions in S100, forming a complete technical closed loop of "condition setting - anomaly judgment - processing - result verification", which enhances the rigor and effectiveness of the system.

[0056] Specifically, in S500, the method for optimizing the error handling by parsing operation records includes: classifying and clustering operation records according to the Bin of wafer testing, extracting high-frequency effective operations, and generating an optimized processing flow.

[0057] It should be further explained that in the S500, after the retest result is satisfactory (Pass), the server not only saves the operation record and changes the device status, but also starts the analysis and optimization module. The core task of this module is to automatically optimize the existing error handling method based on historical successful processing experience data. The specific optimization process includes the following steps: I. Data Collection and Preprocessing: (1) Data source: Collect historical successful operation records and their associated information from the server database, including: ① Bin information for wafer testing (e.g., Bin1, Bin2, ..., Bin8); ② A complete sequence of operation records (action type, parameters, and time consumption); ③ The corresponding types of mis-slaughter and the results of retesting.

[0058] (2) Data preprocessing: Normalize the operation time (e.g., divide the actual time by a baseline time such as 300 seconds); perform one-hot encoding or numerical encoding on the action type; remove obviously invalid or abnormal records (e.g., operation time exceeding 500 seconds).

[0059] II. Classification and Cluster Analysis: The parsing module employs an unsupervised learning algorithm. First, operation records are grouped by wafer bin. Then, within each bin group, further cluster analysis is performed on the operation records. The K-means clustering algorithm is preferred. Feature vectors selected for clustering include: normalized operation time, action type encoding vector, and key parameter values ​​(such as needle pressure adjustment). The cluster size K is set to 5 (preferred value, based on elbow rule analysis of historical data; when K=5, the decrease in intra-cluster distance tends to plateau, effectively distinguishing the five most important operation modes, such as "rapid cleaning mode" and "fine positioning mode"). The Euclidean distance formula is used to measure the similarity between records. in, Represents two eigenvectors and The Euclidean distance between them and Representing vectors respectively and The i-th feature component, This represents the dimension of the feature vector. Physical meaning: The smaller the distance value, the more similar the patterns of the two operation records.

[0060] III. Extracting high-frequency effective operations and generating optimized processes: The operation records within each cluster are analyzed to extract action sequences that occur frequently (e.g., frequency > 70%) and ultimately achieve success in retesting. These high-frequency, effective action sequences are then compared with the old processing methods corresponding to the current bin type and the type of miscarriage. Optimization rules may include: simplifying steps (removing low-frequency or invalid actions), adjusting the operation order, merging consecutive actions, and recommending better parameter ranges for specific steps (e.g., for a short-circuit problem in a certain bin type, if the clustering shows that most successful needle pressure modification values ​​are concentrated at 12g, the optimization method could adjust the recommended needle pressure from 10g to 12g). Finally, a new, optimized processing flow is generated for the "specific bin type + specific miscarriage type".

[0061] Furthermore, the "optimization of the erroneous slaughter handling method by parsing operation records" in the above S500 includes the following specific process, the core of which lies in clustering after classification by Bin type and extracting high-frequency patterns to optimize the process: I. Data Collection and Classification by Bin: 1. Data collection sources: (1) Collect historical successful operation records and their associated information from the server database, including: ① Bin information for wafer testing (e.g., Bin1, Bin2, ...); ② Operation sequence (such as needle tip inspection, needle position adjustment, and other action sequences); ③ Operating parameters (such as needle pressure change, operation time, etc.); ④ The corresponding types of miscarriage and the final retest results.

[0062] (2) Data time range: Continuously collect data for the most recent 30 days (preferred value, because it can ensure sufficient data volume and reflect the recent production status).

[0063] 2. Data preprocessing: (1) Normalize the operation time: Divide the actual time by the maximum operation time of 300 seconds to convert it into a range of 0-1.

[0064] (2) Encode the action type, for example: ①Pin tip examination = 1; ② Needle position adjustment = 2; ③ Needle pressure change = 3; ④ Perform verification tests (which can be further divided into verifying open circuits, verifying short circuits, etc.) = 4.

[0065] (3) Remove abnormal records (such as records with an operation time exceeding 500 seconds).

[0066] II. Cluster analysis by Bin: 1. Clustering Algorithms and Strategies: First, the preprocessed historical operation records are grouped according to their respective Bin categories.

[0067] Then, for each Bin group of data, K-means clustering algorithm was used independently for in-depth analysis. The number of clusters K was set to 5 (preferred value, because: when evaluating historical data using the elbow rule, K=5 can effectively distinguish the main operating patterns and avoid overfitting or pattern omission).

[0068] 2. Feature Selection: Construct a feature vector for each operation record for cluster analysis, including: ① Normalized operation time; ② Action type encoding; ③ Normalized parameter values ​​(such as needle pressure variation).

[0069] 3. Clustering result example: Through clustering, the operation records under each Bin are divided into several clusters, each cluster representing a typical processing pattern. For example: (1) For Bin1 data, the typical clusters that may be obtained include: ①Cluster 1: Quick inspection mode (characterized by short operation time, mainly using pinpoint inspection); ②Cluster 2: Fine adjustment mode (characterized by long operation time, including multiple needle position adjustments).

[0070] (2) For Bin2 data, the typical clusters that may be obtained include: ①Cluster 1: Short-circuit fast processing mode (characterized by short operation time, mainly for verifying short circuits and related rapid checks); ②Cluster 2: Complex adjustment mode (characterized by long operation time, including multiple steps such as needle pressure modification).

[0071] III. Extracting high-frequency effective operations and generating optimized processes: 1. Setting rule generation conditions: (1) Minimum support: 30% (that is, the proportion of records in a certain cluster to the total number of records in that Bin must not be less than this value to ensure that the pattern is representative).

[0072] (2) Action condition probability threshold: 0.7 (that is, the frequency of an action within a cluster must be no less than this value in order to be considered a "high-frequency" operation in this mode).

[0073] (3) Rule confidence threshold: 0.9 (the estimated success rate of the optimized rules generated based on this model must not be lower than this value).

[0074] 2. High-frequency operation extraction and rule generation: For each cluster that meets the minimum support requirement, high-frequency action sequences with conditional probabilities exceeding a threshold are extracted. By analyzing the correlation between these high-frequency action sequences and the final successful outcome, optimization rules in the form of "IF-THEN" are generated. The rule generation formula is used to calculate the conditional probability of an action occurring within a cluster: in, This represents the conditional probability of action A occurring in cluster C; This indicates the number of records containing action A in cluster C; This represents the total number of records in cluster C.

[0075] 3. Example of optimization rules: For example, the above process may generate the following optimization rules: (1) Example of Bin1 optimization rule: IF error type is open circuit AND operation record belongs to the "Quick Check Mode" cluster of Bin1; THEN generate optimization process: a) Prioritize automatic pinpoint inspection (limited to 5 seconds); b) If the inspection passes, proceed directly to the verification open-circuit test; c) Non-high frequency needle position adjustment steps can be skipped.

[0076] (2) Bin2 optimization rule example: IF error type is short circuit AND operation record belongs to the "short circuit fast processing mode" cluster of Bin2; THEN generate optimization process: a) First, perform a needle tip cleaning check; b) Automatically set the short-circuit detection parameters of the tester (current threshold <10Ω); c) If the cleaning check passes, proceed directly to the short-circuit verification.

[0077] IV. Rule Validation and Update: 1. Validation Process: (1) Use historical data or independent test sets to validate the newly generated optimization rules.

[0078] (2) Validation metric: Success rate (number of successful tests / total number of tests).

[0079] (3) Success threshold: ≥90% (preferred value to ensure the reliability of the optimized process).

[0080] 2. Update mechanism: (1) Update cycle: 7 days (preferred value, balancing real-time optimization and system stability).

[0081] (2) Conflict handling: When a new rule conflicts with an old rule, the rule with the higher success rate is retained.

[0082] (3) Version management: The optimized error handling method is version managed, and the version number, effective time and update content are recorded.

[0083] V. Optimize the management and application of rule sets: 1. Rule storage: The optimized processing method is stored in the server database in a structured format (such as JSON). For example: { "bin_type"; "Bin1", "error_type"; "open path", "rule_id"; "B1_OPEN_R001", "source_cluster_profile"; "Quick check mode", "optimized_steps"; ["tip check", "verify open circuit"], "confidence_score"; 0.95, "last_updated"; "2025-08-01" } 2. Rule Enforcement: In subsequent handling of accidental wafer destruction, the system will detect the accidental destruction in real time and automatically match and issue the corresponding optimized processing flow to the operator based on the current wafer's bin and the type of accidental destruction, thereby achieving continuous evolution of the processing method.

[0084] Understandably, this process fully implements the analytical and optimization steps of "classifying and clustering by Bin type, extracting high-frequency effective operations, and generating optimized processes." Through a data-driven approach, it continuously learns from historical successes, enabling the error handling method to be continuously optimized and improved. This reduces reliance on reactive experience and enhances the targeting and overall efficiency of testing and handling different Bin types of products.

[0085] Specifically, the optimized method for handling accidental slaughter is automatically updated to the server for use in handling subsequent accidental slaughter incidents.

[0086] It should be further noted that after the module generates the optimized error handling method, the system will not immediately replace the online version. To ensure the robustness of the update, the following automated update mechanism is set up: (1) Validation phase: The newly generated optimization method will be backtested in an independent "sandbox" environment or on a historical dataset. The validation criteria are that the estimated improvement in the processing success rate is not less than 5% and the absolute success rate is not less than 95% (preferred values, because this dual standard can encourage optimization while ensuring the absolute reliability of the new method and avoiding the introduction of defective processes). The validation cycle is usually 24 hours.

[0087] (2) Update and Replace: After successful verification, the server automatically performs the update operation during the set low-load period (such as midnight). The update process is essentially executing database update statements, such as: UPDATE method_table SET content='{new method JSON}'WHERE bin='Bin1'ANDerror_type='short circuit'. The update operation has version management functionality, and each update records the new version number, effective time, and the old version ID being replaced.

[0088] (3) Rollback Mechanism: After the update, the system will monitor the actual application performance of the new method over a period of time (e.g., one week). If the actual success rate drops significantly (e.g., below 90%), the system can automatically or, with administrator confirmation, roll back to the previous stable version. After the update is completed, when the same Bin type and type of false alarm event occurs again in subsequent tests, the server will issue the latest optimized processing method to the operators.

[0089] Understandably, the “optimization-verification-automatic update” process forms a complete closed loop, enabling the system’s processing capabilities to keep pace with the times and continuously absorb excellent experiences. The automated process greatly reduces the workload of manually maintaining and updating the knowledge base, and avoids errors or delays that may be caused by manual updates, ensuring the continuous advancement of the system’s processing strategies.

[0090] Specifically, the method for handling accidental slaughter includes at least one of the following operations: checking for needle dirt, adjusting needle position, verifying open circuit, and verifying short circuit.

[0091] It should be further noted that the default error handling method in the server consists of a sequence of one or more standard operations. These operations specifically include: (1) Check for contamination of probe tips: Operators shall use the built-in microscope or external optical equipment of the probe tester to check for contamination of the probe tips according to the instructions. An automatic cleaning unit can be configured to intervene. The recommended inspection cycle is after each batch of wafers is (preferred value, because this frequency can effectively prevent the accumulation of contaminants and the resulting batch mis-dumping).

[0092] (2) Adjusting the needle position: The operator inputs the compensation values ​​of the X, Y, and Z directions in the needle testing machine control software according to the instructions, and drives the needle testing machine to make fine adjustments. The adjustment accuracy is usually ±1μm.

[0093] (3) Verify open circuit: This is a test action that instructs the operator to trigger the test machine to quickly retest the relevant test items suspected of being open circuits in order to confirm whether the open circuit problem has been resolved.

[0094] (4) Short Circuit Verification: This is also a test action, instructing the operator to trigger the testing machine to quickly retest the relevant test items suspected of being short-circuited, in order to confirm whether the short circuit problem has been resolved. A complete handling method is usually a logical combination of these operations. For example, for suspected open circuit misoperation, the standard method might be "first check the needle for dirt -> if there is no contamination, fine-tune the needle position -> finally verify the open circuit". These operation steps, sequence and parameters are all clearly preset in the server.

[0095] Understandably, clearly defining the specific operations involved in the processing method provides operators with clear and actionable guidelines, reducing the operational threshold and reliance on personnel experience; combining processing actions with verification testing ensures that the effectiveness of the processing actions is reflected in real time, forming a small closed loop of "operation-verification".

[0096] Please see Figure 2 The present invention provides another embodiment, which provides an automated processing system for errors in wafer testing. The automated processing system for errors in wafer testing includes: (1) Server 100 is used to store the conditions for accidental slaughter and the methods for handling accidental slaughter, and is connected to the probe tester 200 and the testing machine 300 for communication; Server 100 is also used to: receive real-time test data uploaded by the testing machine 300, and determine accidental slaughter according to the conditions for accidental slaughter; when determining accidental slaughter, send a stop command to the probe tester 200 and automatically open an equipment abnormality order; send the methods for handling accidental slaughter to the client; receive the operation records uploaded by the probe tester 200 and compare them with the preset methods for handling accidental slaughter; receive the retest results fed back by the testing machine 300; and after the retest results are qualified, save the operation records and update the equipment status.

[0097] It should be further explained that server 100, as the "brain" of the system, undertakes all logical judgments, process control, data storage, and intelligent analysis tasks. Its specific design includes: a) Hardware configuration: High-reliability industrial servers (e.g., Intel Xeon series processors, 64GB or more of memory, and RD-configured solid-state drive arrays) can be used to ensure powerful data processing and storage capabilities. It should run CentOS or other stable enterprise-grade operating systems.

[0098] b) Core software module: The core software system for implementing the method of this invention is deployed on server 100, and its modular design includes: ① Communication gateway module: responsible for establishing and maintaining network connections with probe tester 200 and tester 300, and using protocols such as TCP / IP and Modbus / TCP to issue commands and receive data.

[0099] ② Data Stream Processing and Judgment Module: This is the key module for realizing "centralized judgment by server 100". It receives raw test data streams from test machine 300 in real time and calls preset error-causing and blocking conditions for real-time comparison and analysis, thereby making error-causing judgment decisions.

[0100] ③ Workflow Engine Module: Based on the judgment result, automatically serialize and execute subsequent processes: send a stop command to the needle testing machine 200, trigger an alarm, open an abnormality form, and issue corresponding processing methods and instructions to the lower-level machine and operation terminal.

[0101] ④ Operation Consistency Comparison Module: Receives operation records uploaded by the probe testing machine 200, and uses a rule engine and algorithm to perform real-time consistency comparison with preset methods.

[0102] ⑤ Analysis and Optimization Module: Responsible for periodically performing cluster analysis on historical successful records, extracting high-frequency effective operation patterns, and generating and verifying the optimized processing flow.

[0103] ⑥ Database: MySQL or a similar relational database is used to store all preset conditions, processing methods, equipment fault reports, operation records, historical data and optimization rules.

[0104] (2) The probe tester 200 is connected to the server 100 for performing wafer testing operations, receiving and executing the stop command of the server 100, and generating operation records and uploading them to the server 100.

[0105] It should be further explained that the probe probe 200, as the "precision hand" of the system, is responsible for performing specific wafer contact and position adjustment actions. Its specific design includes: a) Hardware and Communication: A commercial probe station with standard communication interfaces (such as Ethernet ports). Its control system can parse and execute motion control commands (such as stop, move to a specified coordinate) issued by server 100.

[0106] b) Sensing and recording functions: Integrating high-precision pressure sensors, optical encoders, etc., to collect operating parameters such as needle pressure and needle position in real time when the operator performs the processing method, and encapsulate them into structured operation records.

[0107] c) Alarm Execution Unit: Integrated with audible and visual alarm devices (buzzer, LED light), it triggers a local alarm after receiving an alarm command from server 100.

[0108] (3) Test machine 300, which is connected to the server 100 for testing wafers and uploading real-time test data to the server 100, and performing retests and feeding back the retest results according to the instructions or settings of the server 100.

[0109] It should be further explained that the test machine 300, as the "sensor" of the system, is responsible for electrical test excitation and measurement. Its specific design includes: a) Core Function: An automated testing device with communication capabilities. Its main responsibility is to execute predetermined test procedures and upload the measured raw voltage, current, resistance, and other real-time test data to server 100 via a high-speed bus (such as GPIB, PXIe) or Ethernet. It does not contain or execute the "false alarm" judgment logic defined in this solution.

[0110] b) Retest execution function: Receives retest instructions from server 100, re-executes the test at the specified wafer location, and feeds back the retest results to server 100.

[0111] Furthermore, during system operation, data flow and control flow strictly follow the "server 100 center" model: a) Data upload: Test machine 300 continuously uploads real-time test data to server 100.

[0112] b) Core judgment: The data stream processing and judgment module of server 100 analyzes the data to determine whether a mistaken slaughter has occurred.

[0113] c) Downlink Control: If it is determined to be a mistaken slaughter, the server 100 synchronously triggers multiple instructions through the workflow engine: controls the needle probe machine 200 to stop, triggers an alarm, opens an abnormal order, and sends the processing method to the operation terminal.

[0114] d) Operation closed loop: The operator operates the probe testing machine 200, the operation record is uploaded, and the server 100 performs a consistency comparison.

[0115] e) Verify closed loop: Server 100 instructs test machine 300 to perform retest and receives the retest results.

[0116] f) Optimize the closed loop: After the process is completed, server 100 starts the parsing and optimization module to optimize the processing method in the background.

[0117] Understandably, this system architecture completely changes the traditional loose collaboration model between devices, centralizing all core intelligent judgments (error detection, process scheduling, consistency verification, and optimization) on server 100. Tester 300 and probe tester 200 are decoupled into purely functional execution units, bringing multiple benefits: First, it achieves absolute uniformity of judgment standards across the entire production line, eliminating the impact of equipment differences; second, it significantly improves the system's flexibility and scalability, as any updates to processing logic only need to be performed on server 100, without modifying the tester 300 program; finally, it lays a solid foundation for data integration across the entire production line and higher-level intelligent manufacturing analysis. Those skilled in the art can implement this system by selecting commercial probe testers 200 and 300 that support standard communication protocols, and developing server 100 software based on the aforementioned modular design.

[0118] In a preferred embodiment, this application also provides an electronic device, the electronic device comprising: The computer device includes a memory and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, implement the automated processing method for errors in wafer testing. The computer device can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the method of the present invention.

[0119] This invention can be implemented as a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the steps of the methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0120] Those skilled in the art will understand that the method steps of this invention can be performed by a computer program instructing related hardware, such as a computer device or processor, to perform the steps of this invention when executed. Depending on the context, any references herein to memory, storage, databases, or other media may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0121] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0122] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. An automated method for handling errors in wafer testing, characterized in that, Includes the following steps: S100: The server presets the conditions for accidental slaughter and the corresponding accidental slaughter handling methods, and the server communicates with the needle testing machine and the testing machine via the network. The S200 tester uploads real-time test data to the server. When the server determines that a slaughter has been mistakenly committed based on the preset conditions for mis-slaughtering, it controls the needle tester to stop testing and issue an alarm, while automatically creating a device malfunction report. S300: The operator performs the corresponding operation on the needle testing machine according to the error handling method issued by the server. The needle testing machine generates operation records in real time and uploads them to the server. S400. The server compares the operation record with the preset error handling method and receives the retest result from the test machine. S500: If the retest result is qualified, the server saves the operation record, restores the equipment status from abnormal to production status, and optimizes and updates the error handling method by parsing the operation record.

2. The method according to claim 1, characterized in that, In S100, the conditions for accidental killing include at least one of the following: the electrical parameter threshold of the test program, the needle pressure setting of the needle tester, and the probe contact state parameters.

3. The method according to claim 1, characterized in that, In S200, the alarm includes an audible and visual alarm and a prompt from the host computer software; the equipment anomaly report includes the type of mis-slaughter, the time of occurrence, the equipment number, and the test location information.

4. The method according to claim 1, characterized in that, In S300, the operation record includes at least one action information among needle tip inspection, needle position adjustment, needle pressure modification, and probe cleaning.

5. The method according to claim 1, characterized in that, In step S400, if the server compares and determines that the operation record is inconsistent with the preset error handling method, it will prompt the operator to re-execute the standard operation.

6. The method according to claim 1, characterized in that, In S400, the retest results include electrical test results for open circuit, short circuit, and leakage.

7. The method according to claim 1, characterized in that, In S500, the method for optimizing the error handling process by parsing operation records includes: classifying and clustering operation records according to the Bin of wafer testing, extracting high-frequency effective operations, and generating an optimized processing flow.

8. The method according to claim 7, characterized in that, The S500 also includes: automatically optimizing and updating the mistaken slaughter handling method to the server for subsequent mistaken slaughter events.

9. The method according to claim 1, characterized in that, The method for handling accidental slaughter includes at least one of the following operations: checking for needle dirt, adjusting needle position, verifying open circuit, and verifying short circuit.

10. An automated processing system for errors in wafer testing, characterized in that, include: The server is used to store the conditions for accidental slaughter and the methods for handling accidental slaughter, and is connected to the probe tester and the testing machine for communication. The server is also used to: receive real-time test data uploaded by the testing machine, and determine accidental slaughter based on the conditions for accidental slaughter; when accidental slaughter is determined, send a stop command to the probe tester and automatically create a device abnormality report; send the methods for handling accidental slaughter to the client; receive the operation records uploaded by the probe tester and compare them with the preset methods for handling accidental slaughter; and receive the retest results fed back by the testing machine. And after the retest results are qualified, save the operation record and update the equipment status. The probe tester is connected to the server and is used to perform wafer testing operations, receive and execute stop commands from the server, and generate operation records and upload them to the server. The testing machine is connected to the server and is used to test wafers and upload real-time test data to the server, as well as to perform retests according to the server's instructions or settings and provide feedback on the retest results.

Images