WAT probe station control system, methods, equipment and storage media

Abstract

This invention belongs to the field of semiconductor testing technology and discloses a WAT ​​probe station management system, method, equipment, and storage medium. The method achieves efficient integration and traceability of multi-source heterogeneous data through full-process metadata collection and unified association; it introduces an intelligent reliability judgment mechanism to prioritize the identification of test system anomalies and risks, avoiding misattribution of measurement problems to process defects; error compensation technology based on a gold reference set effectively reduces systematic offset; a predictive maintenance module enables early warning of equipment performance degradation and accurate estimation of remaining lifespan; full lifecycle management of recipes ensures the consistency and traceability of test recipes; and a knowledge base and diagnostic recommendation module accumulates expert experience, improving the efficiency of anomaly handling. This achieves full-link traceability, predictability, and controllability of the WAT probe station testing process, significantly reducing yield fluctuations, unplanned downtime, and misjudgment risks.

Description

Technical Field

[0001] This invention relates to the field of semiconductor testing technology, and in particular to a WAT ​​probe station control system, method, device and storage medium. Background Technology

[0002] The WAT (Wafer Acceptance Test) probe station is a core testing device in the back-end processes of semiconductor manufacturing, used to comprehensively test the electrical parameters of each chip before wafer dicing. Its testing accuracy requirements are extremely high (sub-micron level positioning, millinewton level contact force control), the equipment is highly complex (integrating multiple subsystems such as precision mechanics, motion control, temperature management, machine vision, and signal measurement), and maintenance costs are exorbitant (unplanned downtime can result in losses of tens of thousands of US dollars per hour).

[0003] However, the existing WAT probe station control has the following technical problems:

[0004] The problem is severe: probe station equipment, testing machines, needle card management system, and Recipe server each form independent data silos, lacking the ability to perform correlation analysis. When yield is abnormal, engineers need to manually export data from multiple systems and manually compare timelines, which takes several hours to several days and is highly dependent on personal experience.

[0005] The maintenance strategy is crude: the current maintenance is based on fixed cycles or simple counting, ignoring the actual condition of the equipment, resulting in over-maintenance or under-maintenance.

[0006] Lack of predictive capability: Alarms based on single-parameter thresholds cannot identify slow performance degradation, multivariate joint anomalies, and early fault characteristics.

[0007] Disorganized Recipe Management: Inconsistent test recipe file versions and unmonitored execution deviations severely impact test consistency.

[0008] Fragile knowledge transfer: Experience in equipment fault diagnosis and parameter optimization relies on oral instruction from senior engineers, which can easily lead to knowledge gaps due to staff turnover.

[0009] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0010] The main objective of this invention is to provide a WAT ​​probe station management system, method, device, and storage medium, aiming to solve the technical problems of data silos, extensive maintenance, missing predictions, chaotic recipes, and fragile knowledge transfer in existing WAT probe station management systems.

[0011] To achieve the above objectives, the present invention provides a WAT ​​probe station management system, which includes: a data access layer, a stream processing and alignment layer, a standardization and quality control layer, an associated storage layer, an error compensation layer, a prediction and early warning layer, a process recipe management layer, and a knowledge base and diagnostic recommendation layer.

[0012] The data access layer is used to collect multi-source heterogeneous data from the test machine, probe station controller, sensor, needle card management system and Recipe server in real time;

[0013] The stream processing and alignment layer is used to align the multi-source heterogeneous data with timestamps based on a unified time base and test event anchor points, and generate a unified primary key.

[0014] The standardization and quality control layer is used to standardize the aligned multi-source heterogeneous data, generate a confidence label for each standardized data record based on a multi-factor dynamic calculation model, and classify the standardized data according to the confidence label. Low confidence data is marked and preferentially routed to the anomaly investigation branch of the test system.

[0015] The associated storage layer is used to store wafer-level test data, probe station status timing data, and probe card history data containing the credibility tag into a timing database and a relational database, using the unified primary key as the associated index.

[0016] The error compensation layer is used to filter candidate data for matching gold benchmark group based on the confidence label, construct gold benchmark group based on the candidate data, calculate the systematic parameter offset under the current probe station-probe card-temperature range combination based on the gold benchmark group, use the systematic parameter offset to compensate the original test values ​​of each chip unit on the wafer, and output the compensation residual sequence and the compensated wafer map to the prediction and early warning layer.

[0017] The prediction and early warning layer is used to receive the compensated residual sequence and the compensated wafer image, and combine it with the probe station health feature time series data read from the associated storage layer to perform trend decomposition, significance test and sudden anomaly detection.

[0018] The prediction and early warning layer is also used to estimate the remaining lifespan of the probe card and the remaining lifespan of the probe station drive unit based on the equipment degradation model, generate graded early warnings based on trend significance, anomaly score and remaining lifespan, and send the early warning event and the corresponding wafer ID, probe card ID and fault feature vector to the knowledge base and diagnosis recommendation layer.

[0019] The Recipe management layer is used to version and process-oriented approval management of Recipe files, and to perform consistency verification and deviation monitoring before, during and after test execution.

[0020] The knowledge base and diagnostic recommendation layer are used to structure and store historical anomaly cases and their handling strategies, and to provide diagnostic recommendations for current anomalies through similarity matching.

[0021] In one embodiment, the stream processing and alignment layer is used to synchronize the clocks of each data source through a preset protocol, establish a time window with the start and end events of the test machine as anchor points, map each source data into the same time window, and use interpolation or nearest neighbor algorithms for alignment when there is a sub-second offset.

[0022] In one embodiment, the formula for generating the credibility tag is:

[0023]

[0024] in, This is the credibility score, with a value in the range [0, 1]. Let be the base of the natural logarithm. The weights of each factor are w1+w2+w3+w4=1. To calibrate the status indicators, As an indicator of environmental stability, The degree of pin card degradation, Alarm density.

[0025] In one embodiment, the error compensation layer is used to dynamically match or revert to the optimal gold benchmark set based on the current tested product, process conditions, and environmental parameters, and to calculate the systematic offset of the current batch relative to the gold benchmark set using a robust statistical method. The calculation formula is as follows:

[0026]

[0027] in, The systematic offset of parameter p, This represents the original test value of parameter p for the i-th wafer pair in the current condition group. Let p be the test value of parameter p for the i-th wafer in the gold group. This is a median function.

[0028] Calculate the confidence level of this compensation. If the confidence level is lower than the threshold, trigger an alarm and enter the compensation degradation mode.

[0029] The original test values ​​are compensated using the aforementioned systematic offset, calculated as follows:

[0030]

[0031] in, To compensate for the test values, These are the original test values. The systematic offset of parameter p, Dimensional weighting coefficients;

[0032] Calculate the compensated residual:

[0033]

[0034] in, The compensation residual for the i-th wafer is used for subsequent early warning. This represents the original test value of parameter p on the i-th wafer in the current batch. This represents the median value of the parameter p in the gold reference set. The systematic offset of parameter p;

[0035] If the compensation residual exceeds the threshold, an early warning will be triggered.

[0036] In one embodiment, the prediction and early warning layer is used to construct a multi-dimensional health feature vector that includes compensation residual statistics, sensor time-series features, and the rate of change of needle card health indicators; perform time series decomposition and trend significance testing on the health feature vector to obtain trend detection results; identify sudden anomalies by combining statistical rules and outlier detection algorithms; estimate the remaining service life based on the equipment degradation model; and generate early warnings of different severity levels and suggested actions including maintenance windows based on the trend detection results, anomaly identification results, and remaining service life.

[0037] In one embodiment, the similarity matching employs a comprehensive similarity algorithm, calculated using the following formula:

[0038]

[0039] in, The overall similarity score ranges from 0 to 1. Cosine similarity weights For cosine similarity, For Euclidean distance, This is the current anomaly feature vector and the historical case feature vectors. This is the largest Euclidean distance in the historical database.

[0040] In one embodiment, the Recipe control layer includes: a version management unit, an electronic approval unit, and a consistency verification unit;

[0041] The version management unit is used to calculate a hash value as a version identifier when a Recipe file is uploaded, and to associate and store change descriptions, approval information, and parameter differences.

[0042] The electronic approval unit is used to define and execute the electronic process from uploading, reviewing to approval, and to control the state migration of Recipe versions;

[0043] The consistency verification unit is used to compare the hash values ​​of the work order-specified version and the machine-loaded version before test execution, and to monitor the deviation between the actual readback value and the set value of key parameters during execution. When a mismatch or deviation exceeds the limit, an alarm is triggered or the test is blocked.

[0044] Furthermore, to achieve the above objectives, this invention also proposes a WAT ​​probe station control method, which is applied to the WAT probe station control system described above, and the method includes:

[0045] The data access layer collects multi-source heterogeneous data in real time from the test equipment, probe station controller, sensors, needle card management system and Recipe server;

[0046] The stream processing and alignment layer timestamps the multi-source heterogeneous data based on a unified time base and test event anchor points, and generates a unified primary key;

[0047] The standardization and quality control layer standardizes the aligned multi-source heterogeneous data and generates a confidence label for each standardized data record based on a multi-factor dynamic calculation model. The standardized data is then graded according to the confidence label, with low-confidence data being marked and preferentially routed to the anomaly investigation branch of the test system.

[0048] The associated storage layer uses the unified primary key as the associated index to store wafer-level test data, probe station status timing data, and probe card history data containing the credibility tag into the timing database and relational database.

[0049] The error compensation layer filters candidate data for matching gold benchmark groups based on the confidence label, constructs a gold benchmark group based on the candidate data, calculates the systematic parameter offset under the current probe station-probe card-temperature range combination based on the gold benchmark group, compensates the original test values ​​of each chip unit on the wafer using the systematic parameter offset, and outputs the compensation residual sequence and the compensated wafer map to the prediction and early warning layer.

[0050] The prediction and early warning layer receives the compensated residual sequence and the compensated wafer image, and combines it with the probe station health feature time series data read from the associated storage layer to perform trend decomposition, significance test and sudden anomaly detection.

[0051] The prediction and early warning layer estimates the remaining lifespan of the probe card and the remaining lifespan of the probe station drive unit based on the equipment degradation model. It generates graded early warnings based on trend significance, anomaly score and remaining lifespan, and sends the early warning event and the corresponding wafer ID, probe card ID and fault feature vector to the knowledge base and diagnosis recommendation layer.

[0052] The Recipe management layer performs versioning and process-based approval management of Recipe files, and conducts consistency verification and deviation monitoring before, during and after test execution.

[0053] The knowledge base and diagnostic recommendation layer store historical anomaly cases and their handling strategies in a structured manner, and provide diagnostic recommendations for current anomalies through similarity matching.

[0054] Furthermore, to achieve the above objectives, the present invention also proposes a WAT ​​probe station control device, which includes: a memory, a processor, and a WAT ​​probe station control program stored in the memory and executable on the processor. The WAT probe station control program is configured to implement the steps of the WAT probe station control method described above.

[0055] In addition, to achieve the above objectives, the present invention also proposes a storage medium storing a WAT ​​probe station management program, wherein when the WAT probe station management program is executed by a processor, it implements the steps of the WAT probe station management method as described above.

[0056] This invention achieves efficient integration and traceability of multi-source heterogeneous data through end-to-end metadata collection and unified association; it introduces an intelligent reliability judgment mechanism to prioritize the identification of test system anomalies and risks, avoiding misattribution of measurement problems to process defects; error compensation technology based on the golden benchmark group effectively reduces systematic bias; the predictive maintenance module enables early warning of equipment performance degradation and accurate estimation of remaining lifespan; full lifecycle management of the Recipe ensures the consistency and traceability of test recipes; and the knowledge base and diagnostic recommendation module accumulate expert experience, improving the efficiency of anomaly handling. This invention achieves end-to-end traceability, predictability, and controllability of the WAT probe station testing process, significantly reducing yield fluctuations, unplanned downtime, and misjudgment risks. Attached Figure Description

[0057] Figure 1 This is a structural block diagram of the first embodiment of the WAT probe station control system of the present invention;

[0058] Figure 2 This is a flowchart illustrating the first embodiment of the WAT probe station control method of the present invention.

[0059] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0060] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0061] This invention provides a WAT ​​probe station control system, referring to... Figure 1 , Figure 1 This is a structural block diagram of a first embodiment of a WAT ​​probe station control system according to the present invention.

[0062] It's important to note that the WAT (Wafer Acceptance Test) probe station is a core testing device in the back-end processes of semiconductor manufacturing, used to comprehensively test the electrical parameters of each chip before wafer dicing. Its testing accuracy requirements are extremely high (sub-micron level positioning, millinewyn-level contact force control), the equipment is highly complex (integrating multiple subsystems such as precision mechanics, motion control, temperature management, machine vision, and signal measurement), and maintenance costs are exorbitant (unplanned downtime can result in losses of tens of thousands of US dollars per hour). However, existing WAT probe station management suffers from the following technical problems: Severe data silos: The probe station equipment, testing machine, probe card management system, and Recipe server each form independent data silos, lacking the ability to correlate and analyze data. When yield is abnormal, engineers must manually export data from multiple systems and manually compare timelines, taking hours to days and heavily relying on personal experience. Inefficient maintenance strategies: Current maintenance relies mainly on fixed cycles or simple counting, ignoring the actual condition of the equipment, leading to over-maintenance or under-maintenance. Lack of predictive capabilities: Alarms based solely on single-parameter thresholds cannot identify slow performance degradation, multi-variable combined anomalies, or early fault characteristics. Recipe management is chaotic: inconsistent test recipe file versions and unmonitored execution deviations severely impact test consistency. Knowledge transfer is fragile: experience in equipment fault diagnosis and parameter optimization relies on verbal instruction from senior engineers, making it susceptible to knowledge gaps due to staff turnover.

[0063] To address the aforementioned technical challenges, this embodiment achieves efficient integration and traceability of multi-source heterogeneous data through end-to-end metadata collection and unified association; introduces an intelligent reliability judgment mechanism to prioritize the identification of test system anomalies and risks, avoiding misattribution of measurement problems to process defects; utilizes error compensation technology based on a gold benchmark set to effectively reduce systematic bias; a predictive maintenance module provides early warning of equipment performance degradation and accurate estimation of remaining lifespan; full lifecycle management of the Recipe ensures the consistency and traceability of test recipes; and a knowledge base and diagnostic recommendation module accumulates expert experience, improving the efficiency of anomaly handling. This achieves end-to-end traceability, predictability, and controllability of the WAT probe station testing process, significantly reducing yield fluctuations, unplanned downtime, and misjudgment risks. Specifically, this can be implemented as follows.

[0064] In this embodiment, the overall architecture of the system is first described. The system in this embodiment includes: a data access layer, a stream processing and alignment layer, a standardization and quality control layer, an associated storage layer, an error compensation layer, a prediction and early warning layer, a recipe management and control layer, a knowledge base and diagnosis recommendation layer, a visualization and interface layer, and an operation and maintenance and security audit layer. The data access layer connects to the test equipment, probe station controller / sensor, needle card management system, and Recipe server. The stream processing and alignment layer connects to the data access layer and is used to perform time alignment on multi-source data and generate a unified primary key. The standardization and quality control layer connects to the stream processing and alignment layer and is used for field mapping, unit conversion, missing data handling, and credibility scoring. The association storage layer connects to the standardization and quality control layer and is used to write data associated by a unified primary key into the time series database and relational database. The error compensation layer, prediction and early warning layer, and Recipe management layer read data from the association storage layer and output compensation results, early warning results, and Recipe consistency reports, respectively. The knowledge base and diagnosis recommendation layer receives abnormal events from the prediction and early warning layer and the Recipe management layer and outputs recommended handling strategies. The visualization and interface layer outputs reports and APIs to the engineer terminal and MES / EDA platform. The operation and security audit layer runs through all the above layers and is used for access control, operation auditing, and alarm closure.

[0065] Furthermore, the complete operation process of this system includes the following main steps executed in sequence (supporting both real-time stream processing and batch offline processing modes, with real-time mode prioritized for online monitoring and batch mode for in-depth analysis): Step S1: Real-time acquisition and synchronization alignment of multi-source data When the system starts wafer testing, multi-source acquisition is triggered; after acquisition, timestamp alignment and preliminary association are performed immediately. If alignment fails, an anomaly is marked and an alarm is triggered. Step S2: Data standardization, quality control and dynamic labeling of credibility Aligned data enters the cleaning pipeline for unit / field unification, missing data handling and credibility calculation; low credibility data is prioritized for routing to the test system anomaly investigation branch. Unified data model and storage structure: The system adopts a hybrid storage of "relational database + time series database / object storage": (1) Relational database is used for master data and auditing; (2) Time series database is used for high-frequency sensors and health features; (3) Object storage is used for large files: wafer diagrams, raw logs, Diff details and report files, which are associated with foreign keys in the relational database using Object_Key. All cross-system associations are implemented using a unified primary key (Key) and trace_ID: Key is used for multi-source aggregation within the same wafer / test window, and Trace_ID is used for end-to-end tracing of a single alarm / a single closed-loop response. Step S3: Error source tracing modeling and real-time / post-compensation processing. Based on the current context, the optimal golden benchmark group is matched, multi-dimensional offsets are calculated, and compensation is applied; when the compensation residual exceeds the threshold, an early warning input S4 is triggered. Step S4: Health feature construction and predictive early warning decision-making. Health features are extracted from the compensated data and historical sequences, and trend / anomaly detection and remaining lifetime estimation are performed; the early warning level (low / medium / high) and maintenance suggestions are output. Step S5: Recipe end-to-end consistency verification and execution deviation monitoring are performed in parallel with S1-S3, with version verification, deviation detection, and execution log recording performed before / during / after testing, respectively. Step S6: Intelligent matching of abnormal patterns and diagnostic assistance recommendation. The abnormal signals generated in S2-S5 are integrated, and multi-dimensional similarity matching is performed with the historical knowledge base to output the top-ranked recommended handling strategies and their confidence levels. Step S7: Unify result output, visualization reports, and closed-loop feedback. Integrate the outputs of all modules, generate structured reports, and archive them to the database to support subsequent auditing and model iteration. Step S8: Model self-update and feedback loop (periodic execution). Batch process historical data daily / weekly, automatically retrain the gold benchmark group, compensation model, and prediction model to achieve system adaptive optimization. The training data for the prediction and early warning model comes from a historical event loop: using equipment failure / maintenance work orders as weak labels, construct sample {feature sequence X, result Y}, where Y can be the failure category, whether it failed within T hours, or the RUL regression target.The training process includes: sample cleaning, time window segmentation (e.g., rolling by week), feature standardization, and importance selection. Model validation uses time-series cross-validation. For classification tasks, AUC / recall / false positive rate are output, while for regression tasks, MAE and confidence interval coverage are output. Deployment adopts a "dual-track" system: new models initially issue warnings in shadow mode but do not trigger mandatory actions; once the metrics meet thresholds during the continuous validation period (e.g., 2 weeks), they switch to production mode; if the false positive rate exceeds the threshold, the system automatically rolls back to the previous model version and records the MODEL_ROLLBACK event.

[0066] In the specific implementation, the data access layer 10 is used to collect multi-source heterogeneous data from the test machine, probe station controller, sensor, needle card management system and Recipe server in real time. Interface registration and subscription: When the system starts, it registers multiple protocol interfaces (SEC / GEM protocol for test machine, OPC-UA for probe station sensor, file listening for Recipe server, database trigger for needle card management system). (2) Real-time acquisition trigger: The wafer test start event triggers acquisition, synchronous acquisition: WAT test result data structure: {Wafer_ID, Lot_ID, Die_X, Die_Y, Param_Name, Raw_Value, Limit_High, Limit_Low, Timestamp} Context dimension data structure: {ProbeCard_ID, Usage_Count, Contact_Resistance, Temp_Set / Actual, Humidity, Vibration_RMS, PowerBoard_ID, Alarm_Log[].

[0067] In one embodiment, the stream processing and alignment layer 20 is used to synchronize the clocks of each data source through a preset protocol. A time window is established using the start and end events of the test equipment as anchor points, mapping each source data to the same time window. When sub-second offsets exist, interpolation or nearest neighbor algorithms are used for alignment. For example, the time deviation Δt = |Timestamp_source - Timestamp_reference| (reference based on the test equipment) is calculated. If Δt > 1s, it is marked as "time deviation abnormal" and aligned using interpolation. A unified primary key Key = Hash(Wafer_ID + Lot_ID + Start_Time_Window) is generated. To ensure the alignment of multi-source data, the system adopts a "unified time base + drift correction" strategy: the edge gateway periodically (e.g., every 60 seconds) performs NTP or PTP time synchronization with the factory time server and records the local clock offset and jitter; all collected source timestamps (Timestamp_source) retain their original values ​​and generate a unified timestamp (Timestamp_unified = Timestamp_source - offset). When a data source does not provide a reliable timestamp, the system uses the edge gateway's received time, Rx_Time, as a substitute, and appends a Time_Source_Flag identifier (such as SOURCE, RX, or INFER). Alignment is performed using test machine events (Wafer_Start, Wafer_End, Step_Start) as anchor points, prioritizing event window alignment: mapping each source data to the same Start_Time_Window to End_Time_Window; interpolation or nearest neighbor matching is only used if sub-second offsets still exist within the same window. Furthermore, it is important to emphasize that data is temporarily stored in a real-time buffer; if acquisition times out (>30s), a retry mechanism is triggered (maximum 3 times) and an alarm is issued.

[0068] Furthermore, the standardization and quality control layer 30 is used to standardize the aligned multi-source heterogeneous data, generate a confidence label for each standardized data record based on a multi-factor dynamic calculation model, and perform hierarchical processing on the standardized data according to the confidence label. Among them, low confidence data is marked and preferentially routed to the anomaly investigation branch of the test system.

[0069] Specific implementation details include the following processing steps (input: raw multi-source data packets; output: standardized records + credibility labels):

[0070] (1) Units and fields are consistent: Traverse the field mapping table (e.g., the conversion formula for "°F" → "℃" is: T_℃=(T_°F-32)×5 / 9).

[0071] (2) Handling missing values: If a key field (such as Raw_Value) is missing, mark it as "invalid record" and exclude it from subsequent analysis; if a non-key field is missing, use forward fill or mean fill to complete it, and record "fill-method code".

[0072] (3) Quality Tag and Error Code System: The system generates a Quality_Tag and Error_Code set for each record. The Quality_Tag includes at least VALID, INVALID, IMPUTED, LOW_CONF, MED_CONF, and HIGH_CONF. The Error_Code includes at least TIME_SKEW (time deviation anomaly), SRC_TIMEOUT (collection timeout), FIELD_MISSING (key field missing), UNIT_CONFLICT (unit conflict), OUTLIER_RAW (original outlier), UNSTABLE (baseline unstable), RECIPE_MISMATCH (version mismatch), and RECIPE_DRIFT (execution deviation). The above tags and error codes are written to the storage along with the data and are used as part of the feature input and interpretation output in the early warning and diagnostic recommendation module.

[0073] In one embodiment, the formula for generating credibility tags is:

[0074]

[0075] in, This is the credibility score, with a value in the range [0, 1]. Let be the base of the natural logarithm. The weights of each factor are w1+w2+w3+w4=1. To calibrate the status indicators, As an indicator of environmental stability, The degree of pin card degradation, Alarm density.

[0076] It should be noted that, ; (The value is 1 when all environmental parameters (such as temperature, humidity, and vibration) are within the set threshold range, and 0 otherwise; For indicator functions, (For the allowable deviation threshold of each parameter). For the current environmental parameters, To set a threshold. ; .like Mark "Low credibility - test system risk priority" if Mark as "Medium confidence" otherwise "High confidence".

[0077] Furthermore, the associated storage layer 40 is used to store wafer-level test data, probe station status timing data, and probe card history data containing the credibility tag into the timing database and relational database, using the unified primary key as the associated index.

[0078] Furthermore, the error compensation layer 50 is used to filter candidate data for matching gold benchmark groups based on the confidence label, construct a gold benchmark group based on the candidate data, calculate the systematic parameter offset under the current probe station-probe card-temperature range combination based on the gold benchmark group, compensate the original test values ​​of each chip unit on the wafer using the systematic parameter offset, and output the compensated residual sequence and the compensated wafer map to the prediction and early warning layer.

[0079] Specific implementations include: (1) Dynamic maintenance of the gold benchmark group: The system maintains the gold benchmark group library (classified by product / structure / temperature range), automatically filters batches with C>0.9 and no alarms within the last 30 days as candidates each month, and selects those with CV<5% within the group. (2) Determination of gold benchmark group thresholds and rollback strategy: The CV threshold, MAD threshold, and Confidence threshold can be determined in two ways: one is configuration based on engineering experience (grouped according to product / process node); the other is data-driven adaptive determination: the P90 of the CV distribution calculated from historical stable period data is used as the upper limit of the threshold, the P90 of the MAD distribution is used as the robust threshold, and the P95 of the compensated residual distribution is used as the alarm threshold candidate. When the monthly update of the gold benchmark group fails (insufficient candidate samples or instability), the system automatically rolls back to the available gold benchmark group of the previous period and records the gold benchmark_UNSTABLE event; if it fails N times in a row (e.g., N=3), the matching dimension is reduced (e.g., the temperature range is relaxed) or switched to the "no compensation but enhanced warning" degradation mode. (3) Conditional group stratified matching: The current data is stratified by [temperature range ±2℃, pin card ID, distribution panel ID], with priority given to accurate matching. If no match is found, it will fall back to the previous dimension (such as only temperature range).

[0080] In one embodiment, the error compensation layer 50 is further configured to dynamically match or revert to the optimal gold benchmark set based on the current tested product, process conditions, and environmental parameters, and to calculate the systematic offset of the current batch relative to the gold benchmark set using a robust statistical method. The calculation formula is as follows:

[0081]

[0082] in, The systematic offset of parameter p, This represents the original test value of parameter p for the i-th wafer pair in the current condition group. Let p be the test value of parameter p for the i-th wafer in the gold group. Median function (robust estimation, reducing the impact of outliers)

[0083] Calculate the confidence level of this compensation. If the confidence level is lower than the threshold, trigger an alarm and enter the compensation degradation mode.

[0084] The original test values ​​are compensated using the aforementioned systematic offset, calculated as follows:

[0085]

[0086] in, To compensate for the test values, These are the original test values. The systematic offset of parameter p, Dimensional weighting coefficients;

[0087] Calculate the compensated residual:

[0088]

[0089] in, The compensation residual for the i-th wafer is used for subsequent early warning. This represents the original test value of parameter p on the i-th wafer in the current batch. This represents the median value of the parameter p in the gold reference set. The systematic offset of parameter p; if the compensation residual exceeds the threshold, an early warning is triggered, for example, if the average residual |R|>3σ, an early warning is triggered.

[0090] Furthermore, the prediction and early warning layer 60 is used to receive the compensated residual sequence and the compensated wafer image, and combine it with the probe station health feature time series data read from the associated storage layer to perform trend decomposition, significance testing and sudden anomaly detection; the prediction and early warning layer 60 is also used to estimate the remaining service life of the probe card and the remaining service life of the probe station drive unit based on the equipment degradation model, generate graded early warnings according to trend significance, anomaly score and remaining service life, and send the early warning event and the corresponding wafer ID, probe card ID and fault feature vector to the knowledge base and diagnosis recommendation layer.

[0091] It should be noted that the prediction and early warning layer 60 constructs a multi-dimensional health feature vector that includes compensation residual statistics, sensor time-series characteristics, and the rate of change of needle card health indicators; performs time series decomposition and trend significance test on the health feature vector to obtain trend detection results; identifies sudden anomalies by combining statistical rules and outlier detection algorithms; estimates the remaining service life based on the equipment degradation model; and generates early warnings of different severity levels and suggested actions including maintenance windows based on the trend detection results, anomaly identification results, and remaining service life.

[0092] Specific implementations include vectors. (Dimensions ≥ 20), including: compensation residual statistics (mean / variance), sensor trends (temperature control fluctuation RMS), pin card indicator change rate, alarm frequency, etc.

[0093] (2) Trend and anomaly detection (combining statistics and models):

[0094] STL time series decomposition:

[0095]

[0096] in, The original health characteristic observations at time t (or the t-th sampling point) For the trend component at time t, For the seasonal component of time t, Let be the residual component at time t.

[0097] right Perform the Mann-Kendall trend test and calculate the statistic:

[0098]

[0099]

[0100] in, This is the Mann-Kendall statistic, used to detect the significance of trends. The sign function is 0 (positive = +1, negative = -1, zero = 0). Values ​​in the time series of health characteristics (i) <j)。 , Let S be the expected value and variance of S (standard Mann-Kendall formula, with a modified version to handle the tie).

[0101] For standardized statistics, if |Z|>1.96, the trend is significant (corresponding to p<0.05). If |Z|>1.96 (p<0.05) and the direction deteriorates, a "trend warning" is triggered.

[0102] Residual anomaly detection: 3σ rule + IsolationForest outlier score > 0.7 triggers "burst anomaly".

[0103] (3) Remaining useful life estimation (extended to a degradation model): A Paris crack propagation-like degradation model is adopted, with the health index HI_t = 1 - (cumulative damage / failure threshold), and RUL estimation:

[0104]

[0105] in, For standardized statistics, if |Z| > 1.96, the trend is significant. This is an estimate of the remaining useful life. This is the health index threshold when the system fails. For the current health index, The degradation rate parameter is obtained by least squares fitting of historical failure data, and the RUL confidence interval [μ-σ, μ+σ] is output.

[0106] (4) Maintenance Decision Output: Early Warning Levels (Low: Monitoring, Medium: Parts Preparation, High: Immediate Shutdown), Suggested Window = Current Time + Lower Limit of RUL ± Tolerance. Early warning output uses a structured record: {Alert_ID, Key, Severity, Trigger_Rule, Evidence (TopK features), RUL_Range, Suggested_Action, Action_Constraint, Approver, Approve_Time}. Action_Constraint describes action constraints: for example, high-level early warnings require dual approval before blocking testing; medium-level early warnings only prompt for spare parts and scheduling windows; low-level early warnings only strengthen monitoring and increase sampling frequency. The system saves approval and execution results to form closed-loop training data.

[0107] Furthermore, the Recipe management layer 70 is used to perform versioning and process-based approval management of Recipe files, and to perform consistency verification and deviation monitoring before, during and after test execution.

[0108] In one embodiment, the Recipe control layer 70 includes: a version management unit, an electronic approval unit, and a consistency verification unit; the version management unit is used to calculate a hash value as a version identifier when the Recipe file is uploaded, and associate and store change descriptions, approval information, and parameter differences; the electronic approval unit is used to define and execute an electronic process from upload, review to approval, and control the state migration of the Recipe version; the consistency verification unit is used to compare the hash values ​​of the version specified in the work order with the version loaded on the machine before test execution, and monitor the deviation between the actual readback value and the set value of key parameters during execution, and trigger an alarm or block the test when a mismatch or deviation exceeds the limit is found.

[0109] For example, versioned storage involves calculating a SHA-256 hash as the version ID when uploading a Recipe file, and recording {Version ID, Change Description, Approver, Timestamp, Parameter Difference Diff}.

[0110] Recipe access control and approval workflow define role permissions for the Recipe management module: Uploader, Reviewer, Approver, and Operator. Any new Recipe version must meet at least the "upload-review-approval" workflow. Only after approval is it written to T_RecipeVersion and generated as an executable (Released) version; unapproved versions remain in Draft state and cannot be loaded by machines. Version rollback is supported when anomalies are found in production: the system forces loading based on the Version_ID specified in the work order and records the rollback reason; canary releases are also supported: only effective on specified machines / lots, with the scope expanded after consistency audits are passed.

[0111] Before execution, the verification is performed by querying the version ID specified in the work order versus the current loading hash of the machine before the test starts. If they do not match, the test is blocked and an alarm is issued: "Version mismatch - Risk: System drift".

[0112] Deviation monitoring during execution involves collecting key parameter readback values ​​every minute. The determination of the key parameter set K includes at least two methods:

[0113] Rule-based approach: The process engineer configures a parameter whitelist and sets weights (e.g., assigning higher weights to temperature control, Z-axis contact force, and alignment threshold).

[0114] Data-driven approach: Based on the correlation between historical deviations and yield fluctuations, the Top-M correlation parameter is selected as K and updated periodically.

[0115] Weighted normalization is applied to the parameters within K during deviation calculation to avoid misjudgments caused by differences in dimensions. The deviation vector is calculated as follows:

[0116]

[0117] in, For the overall deviation of key parameters, Set a value vector for the Recipe. This is the vector of actual readback values ​​from the machine. For the set value and actual value of the j-th parameter, This is the overall deviation threshold. If D > θ or the deviation of a single parameter > 5%, the deviation event is recorded and an alarm is triggered in real time.

[0118] Post-execution audit involves comparing the entire process logs and generating a consistency report.

[0119] Furthermore, the knowledge base and diagnostic recommendation layer 80 are used to structure and store historical anomaly cases and their handling strategies, and to provide diagnostic recommendations for current anomalies through similarity matching. The similarity matching employs a comprehensive similarity algorithm, calculated using the following formula:

[0120]

[0121] in, The overall similarity score ranges from 0 to 1. Cosine similarity weights For cosine similarity, For Euclidean distance, This is the current anomaly feature vector and the historical case feature vectors. This is the largest Euclidean distance in the historical database.

[0122] Finally, this embodiment generates JSON / XML format reports, including: a compensated wafer diagram (heatmap display), a multi-system timeline Gantt chart, and an alert / maintenance list, supporting interactive dashboard queries. Alternative solutions include data acquisition using a pure batch file import mode (suitable for offline analysis); the compensation model using a neural network regression model (input context dimension, output prediction offset); and the prediction module using a pure rule engine (threshold + count) as a low-computation version.

[0123] This embodiment achieves efficient integration and traceability of multi-source heterogeneous data through end-to-end metadata collection and unified association; introduces an intelligent reliability judgment mechanism to prioritize the identification of test system anomalies and risks, avoiding misattribution of measurement problems to process defects; error compensation technology based on the golden benchmark group effectively reduces systematic deviations; the predictive maintenance module enables early warning of equipment performance degradation and accurate estimation of remaining lifespan; full lifecycle management of the Recipe ensures the consistency and traceability of test recipes; and the knowledge base and diagnostic recommendation module accumulates expert experience, improving the efficiency of anomaly handling. This achieves end-to-end traceability, predictability, and controllability of the WAT probe station testing process, significantly reducing yield fluctuations, unplanned downtime, and misjudgment risks.

[0124] Furthermore, this embodiment of the invention also proposes a storage medium storing a WAT ​​probe station management program, which, when executed by a processor, implements the steps of the WAT probe station management method described above.

[0125] Reference Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the WAT probe station control method of the present invention.

[0126] like Figure 2 As shown, the WAT probe station control method proposed in this embodiment of the invention includes:

[0127] Step S10: The data access layer collects multi-source heterogeneous data in real time from the test machine, probe station controller, sensor, needle card management system and Recipe server.

[0128] Step S20: The stream processing and alignment layer timestamps the multi-source heterogeneous data based on a unified time base and test event anchor points, and generates a unified primary key.

[0129] Step S30: The standardization and quality control layer standardizes the aligned multi-source heterogeneous data and generates a credibility label for each standardized data record based on a multi-factor dynamic calculation model. The standardized data is then graded according to the credibility label.

[0130] Step S40: The associated storage layer uses the unified primary key as the associated index to store the wafer-level test data, probe station status timing data, and probe card history data containing the trustworthiness label into the timing database and relational database.

[0131] Step S50: The error compensation layer filters candidate data for matching gold benchmark groups based on the confidence label, constructs a gold benchmark group based on the candidate data, calculates the systematic parameter offset under the current probe station-probe card-temperature range combination based on the gold benchmark group, compensates the original test values ​​of each chip unit on the wafer using the systematic parameter offset, and outputs the compensation residual sequence and the compensated wafer map to the prediction and early warning layer.

[0132] Step S60: The prediction and early warning layer receives the compensated residual sequence and the compensated wafer image, and combines it with the probe station health feature time series data read from the associated storage layer to perform trend decomposition, significance test and sudden anomaly detection.

[0133] Step S70: The prediction and early warning layer estimates the remaining lifespan of the probe card and the remaining lifespan of the probe station drive unit based on the equipment degradation model. It generates graded early warnings based on trend significance, anomaly score and remaining lifespan, and sends the early warning event and the corresponding wafer ID, probe card ID and fault feature vector to the knowledge base and diagnosis recommendation layer.

[0134] Step S80: The Recipe management layer performs versioning and process-based approval management of Recipe files, and conducts consistency verification and deviation monitoring before, during and after test execution.

[0135] Step S90: The knowledge base and diagnostic recommendation layer structurally store historical anomaly cases and their handling strategies, and provide diagnostic recommendations for the current anomaly through similarity matching.

[0136] This embodiment achieves efficient integration and traceability of multi-source heterogeneous data through end-to-end metadata collection and unified association; introduces an intelligent reliability judgment mechanism to prioritize the identification of test system anomalies and risks, avoiding misattribution of measurement problems to process defects; error compensation technology based on the golden benchmark group effectively reduces systematic deviations; the predictive maintenance module enables early warning of equipment performance degradation and accurate estimation of remaining lifespan; full lifecycle management of the Recipe ensures the consistency and traceability of test recipes; and the knowledge base and diagnostic recommendation module accumulates expert experience, improving the efficiency of anomaly handling. This achieves end-to-end traceability, predictability, and controllability of the WAT probe station testing process, significantly reducing yield fluctuations, unplanned downtime, and misjudgment risks.

[0137] This application embodiment also provides a WAT ​​probe station control device, including a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other through the communication bus. The memory is used to store the WAT probe station control program. When the processor executes the program stored in the memory, it implements the above-mentioned WAT probe station control method.

[0138] The communication bus mentioned in the WAT probe station control equipment can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc.

[0139] The communication interface is used for communication between the aforementioned WAT probe station control device and other devices.

[0140] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0141] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0142] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0143] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0144] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0145] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0146] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solution of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.

[0147] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0148] In addition, for technical details not described in detail in this embodiment, please refer to the WAT probe station control method provided in any embodiment of the present invention, which will not be repeated here.

[0149] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0150] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0151] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0152] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

[0153] It is understood that the system provided in the embodiments of the present invention corresponds to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant content can be referred to the corresponding parts of the above method.

Claims

1. A WAT probe station control system, characterized in that, The WAT probe station management system includes: a data access layer, a stream processing and alignment layer, a standardization and quality control layer, an associated storage layer, an error compensation layer, a prediction and early warning layer, a process recipe management layer, and a knowledge base and diagnostic recommendation layer. The data access layer is used to collect multi-source heterogeneous data from the test machine, probe station controller, sensor, needle card management system and Recipe server in real time; The stream processing and alignment layer is used to align the multi-source heterogeneous data with timestamps based on a unified time base and test event anchor points, and generate a unified primary key. The standardization and quality control layer is used to standardize the aligned multi-source heterogeneous data, generate a confidence label for each standardized data record based on a multi-factor dynamic calculation model, and classify the standardized data according to the confidence label. Low confidence data is marked and preferentially routed to the anomaly investigation branch of the test system. The associated storage layer is used to store wafer-level test data, probe station status timing data, and probe card history data containing the credibility tag into a timing database and a relational database, using the unified primary key as the associated index. The error compensation layer is used to filter candidate data for matching gold benchmark group based on the confidence label, construct gold benchmark group based on the candidate data, calculate the systematic parameter offset under the current probe station-probe card-temperature range combination based on the gold benchmark group, use the systematic parameter offset to compensate the original test values ​​of each chip unit on the wafer, and output the compensation residual sequence and the compensated wafer map to the prediction and early warning layer. The prediction and early warning layer is used to receive the compensated residual sequence and the compensated wafer image, and combine it with the probe station health feature time series data read from the associated storage layer to perform trend decomposition, significance test and sudden anomaly detection. The prediction and early warning layer is also used to estimate the remaining lifespan of the probe card and the remaining lifespan of the probe station drive unit based on the equipment degradation model, generate graded early warnings based on trend significance, anomaly score and remaining lifespan, and send the early warning event and the corresponding wafer ID, probe card ID and fault feature vector to the knowledge base and diagnosis recommendation layer. The Recipe management layer is used to version and process-oriented approval management of Recipe files, and to perform consistency verification and deviation monitoring before, during and after test execution. The knowledge base and diagnostic recommendation layer are used to structure and store historical anomaly cases and their handling strategies, and to provide diagnostic recommendations for current anomalies through similarity matching.

2. The WAT probe station control system of claim 1, wherein, The stream processing and alignment layer is used to synchronize the clocks of each data source through a preset protocol, establish a time window with the start and end events of the test machine as anchor points, map each source data into the same time window, and use interpolation or nearest neighbor algorithms for alignment when there is a sub-second offset.

3. The WAT probe station control system of claim 1, wherein, The formula for generating the credibility tag is: wherein, is a confidence score, with a value range [0, 1], is the base of the natural logarithm, is the weight of each factor, w1+w2+w3+w4=1, is a calibration status indicator, is an environmental stability indicator, is a needle card deterioration degree, is an alarm density.

4. The WAT probe station control system of claim 1, wherein, The error compensation layer is used to dynamically match or revert to the optimal gold reference set based on the current tested product, process conditions, and environmental parameters. It employs a robust statistical method to calculate the systematic offset of the current batch relative to the gold reference set, using the following formula: wherein, a systematic offset for parameter p, is the original test value for parameter p for the i-th wafer of the current condition group, is the test value for parameter p for the i-th wafer of the gold group, is a median function; Calculate the confidence level of this compensation. If the confidence level is lower than the threshold, trigger an alarm and enter the compensation degradation mode. The original test values ​​are compensated using the aforementioned systematic offset, calculated as follows: in, To compensate for the test values, These are the original test values. The systematic offset of parameter p, Dimensional weighting coefficients; Calculate the compensated residual: in, The compensation residual for the i-th wafer is used for subsequent early warning. This represents the original test value of parameter p on the i-th wafer in the current batch. This represents the median value of the parameter p in the gold reference set. The systematic offset of parameter p; If the compensation residual exceeds the threshold, an early warning will be triggered.

5. The WAT probe station control system as described in claim 1, characterized in that, The prediction and early warning layer is used to construct a multi-dimensional health feature vector that includes compensation residual statistics, sensor time-series features, and the rate of change of needle card health indicators; the health feature vector is subjected to time series decomposition and trend significance test to obtain trend detection results; and sudden anomalies are identified by combining statistical rules and outlier detection algorithms. Estimating remaining useful life based on equipment degradation models; Based on the trend detection results, anomaly identification results, and remaining service life, warnings of different severity levels and suggested actions including maintenance windows are generated.

6. The WAT probe station control system as described in claim 1, characterized in that, The similarity matching uses a comprehensive similarity algorithm, and the calculation formula is as follows: in, The overall similarity score ranges from 0 to 1. Cosine similarity weights For cosine similarity, For Euclidean distance, This is the current anomaly feature vector and the historical case feature vectors. This is the largest Euclidean distance in the historical database.

7. The WAT probe station control system as described in claim 1, characterized in that, The Recipe control layer includes: a version management unit, an electronic approval unit, and a consistency verification unit; The version management unit is used to calculate a hash value as a version identifier when a Recipe file is uploaded, and to associate and store change descriptions, approval information, and parameter differences. The electronic approval unit is used to define and execute the electronic process from uploading, reviewing to approval, and to control the state migration of Recipe versions; The consistency verification unit is used to compare the hash values ​​of the work order-specified version and the machine-loaded version before test execution, and to monitor the deviation between the actual readback value and the set value of key parameters during execution. When a mismatch or deviation exceeds the limit, an alarm is triggered or the test is blocked.

8. A method for controlling a WAT ​​probe station, characterized in that, The WAT probe station control method is applied to the WAT probe station control system as described in any one of claims 1 to 7, the method comprising: The data access layer collects multi-source heterogeneous data in real time from the test equipment, probe station controller, sensors, needle card management system and Recipe server; The stream processing and alignment layer timestamps the multi-source heterogeneous data based on a unified time base and test event anchor points, and generates a unified primary key; The standardization and quality control layer standardizes the aligned multi-source heterogeneous data and generates a confidence label for each standardized data record based on a multi-factor dynamic calculation model. The standardized data is then graded according to the confidence label, with low-confidence data being marked and preferentially routed to the anomaly investigation branch of the test system. The associated storage layer uses the unified primary key as the associated index to store wafer-level test data, probe station status timing data, and probe card history data containing the credibility tag into the timing database and relational database. The error compensation layer filters candidate data for matching gold benchmark groups based on the confidence label, constructs a gold benchmark group based on the candidate data, calculates the systematic parameter offset under the current probe station-probe card-temperature range combination based on the gold benchmark group, compensates the original test values ​​of each chip unit on the wafer using the systematic parameter offset, and outputs the compensation residual sequence and the compensated wafer map to the prediction and early warning layer. The prediction and early warning layer receives the compensated residual sequence and the compensated wafer image, and combines it with the probe station health feature time series data read from the associated storage layer to perform trend decomposition, significance test and sudden anomaly detection. The prediction and early warning layer estimates the remaining lifespan of the probe card and the remaining lifespan of the probe station drive unit based on the equipment degradation model. It generates graded early warnings based on trend significance, anomaly score and remaining lifespan, and sends the early warning event and the corresponding wafer ID, probe card ID and fault feature vector to the knowledge base and diagnosis recommendation layer. The Recipe management layer performs versioning and process-based approval management of Recipe files, and conducts consistency verification and deviation monitoring before, during and after test execution. The knowledge base and diagnostic recommendation layer store historical anomaly cases and their handling strategies in a structured manner, and provide diagnostic recommendations for current anomalies through similarity matching.

9. A WAT probe station control device, characterized in that, The WAT probe station control device includes: a memory, a processor, and a WAT ​​probe station control program stored in the memory and executable on the processor, wherein the WAT probe station control program is configured to implement the steps of the WAT probe station control method as described in claim 8.

10. A storage medium, characterized in that, The storage medium stores a WAT ​​probe station management program, which, when executed by a processor, implements the steps of the WAT probe station management method as described in claim 8.

Images