Data management method and system for integrated circuit package test

Abstract

The invention relates to the technical field of integrated circuit package testing, and discloses a data management method and system for integrated circuit package testing. The method comprises the following steps: collecting real-time load data and a chip complexity index, and constructing a channel load distribution matrix; carrying out classification and dynamic shunting on the test data by adopting a load balancing algorithm based on the matrix; recording a data distribution path and embedding a timestamp and a source target identifier to form a complete data flow link record; judging a load balancing state according to the record and adjusting parameters; extracting an abnormal signal from the load state, analyzing a test result deviation, and positioning an error correction target channel; carrying out redistribution processing on the problem data subset to obtain corrected data; and summarizing the tracking logs, verifying the integrity, calculating a data reliability index, fusing the data reliability index to a test report after the data reliability index reaches the standard, and generating a final traceable test result set. According to the invention, dynamic distribution, whole-course tracking and rapid error correction of the test data are realized, and the test efficiency and the data reliability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of integrated circuit package testing, and in particular to a data management method and system for integrated circuit package testing. BACKGROUND

[0002] At present, integrated circuit package testing is a key link to ensure chip performance and reliability, and the massive test data generated thereby is an important part of industrial data management. With the increase in chip complexity and the popularity of multi-chip parallel testing mode, the amount of data generated during testing has increased dramatically, posing high requirements for data management.

[0003] In one prior art, test data management often adopts a static allocation strategy, which is difficult to dynamically adjust according to the real-time load of each test channel, and is prone to cause uneven resource allocation problems such as overloading of some channels and idling of other channels. In addition, when test data flows between multiple channels, there is a lack of effective path tracking mechanism, and when test results are abnormal, it is difficult to quickly trace the complete path of the data, locate the problem source, and consume time and effort, which seriously affects the test efficiency and analysis accuracy.

[0004] The prior art lacks an intelligent dynamic load balancing mechanism, cannot adaptively distribute according to the real-time collected channel load and chip complexity indicators, the data flow path is not transparent, and there is a lack of link tracking means that can record the information of each channel switching; the error correction process after an exception occurs is low in automation, relies on manual backtracking, and cannot quickly locate the problem channel and realize automatic redistribution and correction of the data subset.

[0005] Therefore, there is an urgent need for an integrated circuit package testing data management scheme that can realize dynamic distribution, full-process traceability, and automatic error correction. SUMMARY

[0006] The present application provides a data management method and system for integrated circuit package testing to solve the technical problems of uneven load of test channels, untraceable data flow, and difficult exception location in the prior art.

[0007] In a first aspect, to solve the above technical problems, the present application provides a data management method for integrated circuit package testing, comprising: Collecting real-time load data and chip complexity indicators in testing, extracting channel occupancy rate and data volume to be processed, and obtaining a channel load distribution matrix; According to the channel load distribution matrix, a load balancing algorithm is used to preliminarily classify the test data in the real-time load data, and if the channel occupancy rate exceeds a preset occupancy rate threshold, high complexity chip data is transferred to a low load channel to determine a dynamic distribution scheme; Obtain the data allocation path in the dynamic traffic splitting scheme, and embed a timestamp and source / target identifier when switching each channel using a path recording algorithm to obtain a complete data flow link record; Based on the complete data flow link record, determine whether there is uneven distribution. If the complete data flow link record shows that the channel is continuously overloaded, adjust the splitting parameters to obtain the updated load balancing status. Anomaly signal indicators are extracted from the updated load balancing status. Through statistical analysis, the test results are analyzed to obtain test result deviation data. If the test result deviation data exceeds the preset deviation threshold, the data flow link record is traced back to locate the source of the problem and determine the target channel for error correction. Based on the error correction target channel, obtain a subset of associated test data, and inject it into the backup channel for processing through a reallocation algorithm to obtain the corrected test output data. The tracking logs are aggregated from the corrected test output data. The integrity of the tracking logs is verified, the data reliability index is calculated and it is determined whether it meets the standard. If it meets the standard, the corrected test output data is integrated into the overall test report to obtain the final traceable test result set.

[0008] In one optional implementation, the process of constructing the channel load distribution matrix includes: Real-time load data and chip complexity indicators are collected through a multi-chip parallel testing system. Based on the real-time load data, the occupancy rate of each channel is extracted, and the channel occupancy rate distribution is obtained by statistical analysis. Based on the channel occupancy distribution and chip complexity index, calculate the amount of data to be processed for each channel and generate a set of data to be processed. If the amount of data to be processed exceeds the preset threshold, the high-load channel is split into multiple data streams to obtain an adjusted data distribution. A channel load distribution matrix is ​​constructed based on the adjusted data volume distribution and channel occupancy rate.

[0009] In one optional implementation, the process of determining the dynamic traffic splitting scheme includes: Obtain the channel load distribution matrix, and determine the current load status by analyzing the occupancy data of each channel; If the channel occupancy rate exceeds the preset occupancy rate threshold, the K-means clustering algorithm is used to perform preliminary classification of the test data in the real-time load data to obtain groups of high-complexity chip data and low-complexity chip data. Based on the high-complexity chip data grouping, obtain a list of low-load channels and determine the set of transferable target channels; The high-complexity chip data is allocated to the low-load channel using a consistent hashing algorithm to generate a preliminary data offloading scheme. Analyze the load balancing status of each channel for the preliminary traffic splitting scheme to determine if there are any new high-load channels. If the new high-load channel exists, the simulated annealing algorithm is used to optimize the traffic splitting scheme, adjust the data allocation, and obtain the final traffic splitting scheme. Update the channel load distribution matrix according to the final traffic splitting scheme to generate dynamic traffic splitting results.

[0010] In one optional implementation, the process of generating a complete data stream link record includes: Obtain the allocation path in the data splitting scheme, analyze the data flow direction through a preset path recording algorithm, and determine the initial allocation status of each path; Based on the initial allocation state, a timestamp generation mechanism is used to embed the current timestamp during channel switching to generate a path record containing time information; By using the source and target identifier allocation rules, the identifiers of the source and target nodes are extracted from the path records to obtain the labeled path dataset; If duplicate identifiers are found in the labeled path dataset, the identifiers are deduplicated using a hash algorithm to generate a set of paths with unique identifiers. Based on the uniquely identified set of paths, a link tracing algorithm is used to analyze the data flow links and determine the complete data flow transmission trajectory. By obtaining the switching frequency statistics of each channel through the complete data stream transmission trajectory, a dynamic allocation record of the data stream link is obtained. Based on the dynamically allocated records, a consistency check algorithm is used to verify the integrity of the link records, and the final complete data flow link records are generated.

[0011] In one optional implementation, the anomaly detection and problem localization process includes: Abnormal signal indicators are obtained from the updated load balancing status, and the distribution characteristics of the indicators are calculated using statistical analysis methods to obtain an abnormal signal set. If the index values ​​in the abnormal signal set exceed the preset index threshold, the isolated forest algorithm is used for anomaly detection to determine the abnormal point set. Based on the set of anomalies, test result deviation data is obtained. If the test result deviation data exceeds a preset deviation threshold, it is determined to be a deviation anomaly set. Data flow records are extracted from the set of deviations and anomalies, and time series analysis is used to backtrack the records to locate the source set of problems. Based on the set of problem sources, analyze the channel correlation in the data flow link to determine the target channel set for error correction.

[0012] In one optional implementation, the test data correction process specifically includes: Obtain the identification information of the error correction target channel, extract the associated test data subset through database query, and determine the integrity of the subset; If the integrity of the subset meets the preset integrity threshold, the test data subset is grouped by a clustering analysis algorithm to obtain the grouped data set; Based on the grouped data set, the K-means algorithm is used to calculate the matching degree between the data set and the backup channel to determine the channel allocation strategy; The grouped data set is injected into the backup channel for processing using a redistribution algorithm to generate preliminary corrected data. If the deviation between the preliminary corrected data and the preset standard test dataset is less than the preset allowable error threshold, then a verification algorithm is used to verify the consistency of the data to obtain the verified data set. Based on the verified dataset, perform data formatting to generate corrected test output data.

[0013] In one optional implementation, the process of generating the final traceable test result set specifically includes: The corrected data is obtained from the corrected test output data, and the tracking logs are extracted using data parsing technology to obtain a structured log dataset; Based on the structured log dataset, consistency verification technology is used to determine the log integrity and obtain the integrity assessment result; If the integrity assessment result meets the preset integrity threshold, then the support vector machine algorithm is used to calculate the data reliability index, and a reliability index set is obtained based on the data reliability index. Based on the aforementioned set of reliability metrics, the tracking logs are correlated with the overall test data using data mapping technology to obtain a correlated dataset. Key fields are extracted from the associated dataset, and data fusion technology is used to generate a fusion test dataset; Based on the fusion test dataset, a decision tree algorithm is used to generate the final traceable test result set.

[0014] Secondly, the present invention provides a data management system for integrated circuit packaging and testing, comprising: The data acquisition module is used to collect real-time load data and chip complexity indicators during testing, extract channel occupancy and the amount of data to be processed, and obtain the channel load distribution matrix. The traffic splitting module is used to perform preliminary classification of the test data in the real-time load data according to the channel load distribution matrix and the load balancing algorithm. If the channel occupancy rate exceeds the preset occupancy rate threshold, the high-complexity chip data is transferred to the low-load channel to determine the dynamic traffic splitting scheme. The path recording module is used to obtain the data allocation path in the dynamic traffic splitting scheme. By embedding timestamps and source-target identifiers when switching each channel through the path recording algorithm, a complete data flow link record is obtained. The load optimization module is used to determine whether there is uneven distribution based on the complete data flow link record. If the complete data flow link record shows that the channel is continuously overloaded, the flow splitting parameters are adjusted to obtain the updated load balancing status. The anomaly detection module is used to extract anomaly signal indicators from the updated load balancing status, analyze the test results through statistical analysis methods to obtain test result deviation data, and if the test result deviation data exceeds the preset deviation threshold, trace back the data flow link record to locate the source of the problem and determine the target channel for error correction. The data correction module is used to obtain a subset of associated test data according to the error correction target channel, and inject it into the backup channel for processing through a redistribution algorithm to obtain the corrected test output data. The report generation module is used to summarize the tracking logs from the corrected test output data, perform integrity verification on the tracking logs, calculate the data reliability index and determine whether it meets the standard. If it meets the standard, the corrected test output data is integrated into the overall test report to obtain the final traceable test result set.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention achieves intelligent optimization and allocation of test resources by constructing a channel load distribution matrix and dynamically splitting traffic based on preset thresholds, effectively avoiding channel overload and resource idleness, thereby significantly improving test throughput. At the same time, by using a path recording algorithm that embeds timestamps and source-target identifiers, a complete and traceable data flow link record is generated, laying the foundation for quickly locating the data flow direction and the root cause of the problem. Furthermore, by combining anomaly detection and link backtracking mechanisms, it can automatically and accurately locate the error correction target channel and complete the redistribution and correction of the problematic data subset, greatly reducing the cost of manual intervention and shortening the problem investigation time. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the data management method for integrated circuit packaging and testing provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the integrated circuit packaging and testing data management system provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Reference Figure 1 This invention provides a data management method for integrated circuit packaging and testing, specifically including the following steps: S101 collects real-time load data and chip complexity indicators during testing, extracts channel occupancy rate and the amount of data to be processed, and obtains the channel load distribution matrix.

[0019] Specifically, the system, through a performance monitoring agent integrated within the test bench and a standard network management protocol, synchronously collects key performance indicators (KPIs) for all test channels at a high frequency of 100 times per second. These KPIs constitute the core of the real-time load data, primarily including: channel CPU utilization, memory usage, I / O throughput, and the length of tasks waiting to be processed in the queue. This high-frequency collection strategy ensures that the system can capture instantaneous load fluctuations, providing time accuracy for subsequent dynamic decision-making. In the actual deployment, the performance monitoring agent runs as a lightweight daemon process, acquiring resource data through the operating system kernel interface. The preferred network management protocol is SNMP, collecting data such as port traffic from network devices. All collected raw data is encapsulated in JSON format and transmitted in real-time to the central data processing server via message queues such as Kafka to ensure low latency and high throughput of the data stream.

[0020] The complexity of a chip directly impacts the amount of data processing and computing resources required for its testing. For accurate evaluation, the system obtains raw data from two dimensions: first, structural information, primarily the total number of transistors, is extracted from the chip's physical design documents; second, test-level information, including critical path delays in signal transmission and the total number of test modes covering different fault models, is extracted from the logs of the automated test pattern generation tool. These raw data have different dimensions and orders of magnitude, and direct use will introduce bias. Therefore, they must be normalized to map them to a uniform, comparable numerical range. This invention uses a weighted summation method for normalization to generate a comprehensive chip complexity index I, calculated as follows: in, , , The weight coefficients representing the indicators of each dimension reflect the degree to which different factors contribute to the overall complexity and satisfy the following conditions: The constraints. Weights can be set based on domain knowledge or historical data analysis. For example, if the number of transistors is considered the primary indicator of complexity, weights can be set... Critical path delay places high demands on test timing, assigning... The number of test modes reflects the comprehensiveness of the test, giving it a certain level of detail. Now, let's take a specific chip as an example for calculation: Suppose a chip integrates 1.2 billion transistors, its critical path latency is 0.8 nanoseconds (ns), and there are a total of 5000 test modes. Substituting these values ​​into the formula, we get the overall complexity index of this chip as 1.84. This value can be used to intuitively compare the testing difficulty of different chips; a higher value means that more channel resources and computation time may be consumed during testing.

[0021] Channel occupancy is a key metric for measuring its current level of activity. It is obtained by comparing the channel's current actual load with its maximum theoretical processing capacity. For any given channel... its occupancy rate The calculation formula is defined as follows: Among them, molecules Representative Channel The total amount of backlogged, unprocessed test data at the current moment can be measured by the number of data packets, tasks, or total data processing volume. Denominator This represents the maximum data capacity that the channel can handle in its design or configuration, i.e., its upper limit of processing capacity. For example, if a channel has 1800 data packets in its pending task queue, and the channel's maximum processing capacity is set to 2000 data packets, then according to the formula, its instantaneous occupancy rate... This percentage clearly indicates that the channel is operating at high capacity. (Amount of data pending processing) The calculation incorporates a data volume weighting factor. The system maintains a mapping table between task types and data weights, assigning higher weights to "heavy" tasks such as high-resolution image processing, thereby more accurately reflecting the actual data processing load.

[0022] After obtaining the occupancy rate of all channels The amount of data to be processed And the complexity metrics of the chips currently being processed. Then, the system integrates this information into a structured, global view—the channel load distribution matrix. This matrix is The mathematical structure in which This represents the total number of test channels in the system. Each row of the matrix corresponds to one channel, and the three columns correspond to the three key state variables mentioned above: in, This is a comprehensive index of chip complexity for the data processed by the i-th channel. This matrix provides a quantitative data foundation for subsequent load analysis; for example, in a system with 8 test channels, the first row of the matrix might be... This indicates that channel 1 is currently 80% occupied, with 1200 units of data to be processed, and is currently handling a highly complex chip. This matrix acts like a real-time "system health status dashboard," providing unique, accurate, and quantitative data for the next intelligent decision-making step.

[0023] S102, based on the channel load distribution matrix, a load balancing algorithm is used to preliminarily classify the test data in the real-time load data. If the channel occupancy rate exceeds the preset occupancy rate threshold, the high-complexity chip data is transferred to the low-load channel to determine the dynamic load splitting scheme.

[0024] The system first parses the channel load distribution matrix generated in the previous stage. It iterates through the first column of the matrix, which shows the occupancy rate of all channels. and with a predefined occupancy threshold The comparison is performed. This threshold is the system's benchmark for determining whether a channel is overloaded, and it is usually set based on long-term operating experience or system stability goals, for example... All satisfied Conditional channels They will be flagged and grouped into a high-load channel set. This step quickly pinpointed the performance bottleneck of the current system.

[0025] Knowing only which channels are overloaded is insufficient; it's also necessary to understand the types of data causing the overload and how to differentiate between different data types. This invention employs an unsupervised machine learning algorithm—K-means clustering—to intelligently group all currently awaiting test data. The clustering is based on two key attributes of each test task: data volume. and the complexity index of its chip The algorithm aims to divide these data points into two clusters, typically high-complexity data sets. and low-complexity datasets The mathematical objective of the clustering process is to minimize the Within-Cluster Sum of Squares (WCSS), which is expressed by the formula: Where k is the number of clusters, It is a cluster The center of mass, For the first Clusters, For high complexity groups , low complexity group The algorithm iteratively adjusts the affiliation and centroid position of data points until the centroid movement becomes extremely small, at which point the model converges. Through clustering, the system can distinguish between "heavy" tasks (large data volume, high complexity) and "light" tasks, laying the foundation for differentiated task allocation strategies. For the clustered data... Groups, i.e., those "heavy" tasks that may cause or exacerbate channel overload, require the system to find new, "spacious" homes. The system queries all currently functioning channels with normal loads, i.e., those that meet the requirements. passage This constitutes a low-load target channel set. To distribute highly complex data evenly and stably across these target channels and avoid introducing new hotspots, this invention employs a consistent hashing algorithm for mapping. The core of this algorithm is a hash function that maps the unique identifier (DataID) of each piece of data to a channel within the target channel set. This ensures that data with the same ID is always mapped to the same channel, and that when the channel set changes, only a small amount of data needs to be remapped, maintaining good stability. This generates a preliminary data distribution scheme.

[0026] The initial solution may not be optimal; for example, it might unintentionally cause a low-load channel to receive too many high-complexity tasks, thus becoming a new high-load channel. Therefore, this invention introduces a simulated annealing algorithm to globally optimize the initial solution. Simulated annealing is a probabilistic global optimization algorithm inspired by the metallurgical annealing process. It defines an objective function to measure the quality of the solution; in this invention, the objective function is the variance of the occupancy rates of all channels in the system. : in It is the average occupancy rate. This is the total number of test channels in the system. This represents the real-time occupancy rate of the i-th test channel.

[0027] The smaller the variance, the more balanced the load across channels, and the better the system performance. The optimization process begins with an initial solution and a relatively high "temperature." The algorithm randomly generates a new neighboring solution and calculates its objective function value. If the new solution is better, it is accepted; if it is worse, it is accepted with a probability related to the current temperature, which gradually decreases with iterations. This strategy allows the algorithm to escape local optima early on and search towards the global optimum. After multiple iterations, the temperature is reduced to a minimum, the algorithm converges, and finally outputs a stable dynamic load distribution scheme that significantly reduces the system load variance.

[0028] S103, obtain the data allocation path in the dynamic splitting scheme, and embed the timestamp and source target identifier when switching each channel using the path recording algorithm to obtain a complete data flow link record.

[0029] Specifically, the system executes a dynamic traffic splitting scheme, monitoring every data packet switching event between channels. Each time a switching occurs, the path recording algorithm records a triplet of information. ,in For high-precision timestamps, A unique identifier for the source channel used for data switching. A unique identifier for the target channel of the data switching. For a complete data stream... Its link record It can be represented as a series of ordered triples: in, This represents the total number of channel switches for this data stream during the test. This is the initial channel ID for the data stream. This is the final channel ID for the data stream. In this way, a complete, timestamped transmission path record is generated for each test data sample.

[0030] S104. Based on the complete data stream link record, determine whether there is uneven distribution. If the complete data stream link record shows that the channel is continuously overloaded, adjust the splitting parameters to obtain the updated load balancing status.

[0031] Specifically, the system does not make judgments based solely on instantaneous states, but rather periodically checks the most recent... All complete data stream records generated within the time period are aggregated and analyzed. For each test channel... The system needs to calculate its average load level within this time window to smooth out instantaneous fluctuations and reflect its continuous workload. This average load... Calculated using the following formula: Where T is the time window Duration For the i-th channel at time... After calculating the instantaneous occupancy rate and the average load of each channel, the system performs a load balancing status check. The check condition is: does a certain channel exist? Its average load Not only did it exceed the preset occupancy threshold Furthermore, the number of cycles in which this over-threshold state persisted exceeded the set continuous overload threshold. If this condition is met, then the channel is determined. A persistent overload indicates that the current load balancing parameters are not working effectively on this channel, resulting in uneven data distribution within the system. Once a persistent overload channel is confirmed, the system triggers dynamic adjustments to the load balancing parameters. These adjustments target the parameters in the load balancing algorithm that determine the weights for data allocation. New weights It is not randomly assigned, but rather calculated using a reverse weighting method based on the historical average load of each channel, as shown in the following formula: This represents the total number of test channels in the system. This represents the historical average load of the i-th and j-th channels. This represents the maximum historical average load across all channels. These are protection parameters. In the next round of load balancing decisions, the load balancing algorithm will allocate data based on these updated weights, thereby automatically reducing the amount of data flowing to continuously overloaded channels, forming a self-optimizing closed-loop control. In this way, the system can gradually migrate the load away from the overloaded channels until a new, more balanced, and stable state is reached.

[0032] S105, extract abnormal signal indicators from the updated load balancing status, analyze the test results using statistical analysis methods to obtain test result deviation data, and if the test result deviation data exceeds the preset deviation threshold, trace back the data flow link record to locate the source of the problem and determine the error correction target channel.

[0033] Specifically, the system extracts key performance indicators from the test result logs, such as the absolute deviation between the actual measured values ​​and the expected nominal values ​​of the test parameters, forming the original deviation sequence. Since different test items or chips have different deviation benchmarks, directly setting a fixed threshold is unreasonable. Therefore, the system uses the statistical Z-score standardization method to process the raw deviation. This method converts the raw deviation values ​​into Z-scores that follow a standard normal distribution, and its calculation formula is as follows: in, This represents the original deviation between the actual measured value and the expected nominal value of a certain performance parameter in the current test results. This represents the long-term average of the raw deviations of similar tests under normal conditions, calculated based on historical data. Z represents the long-term standard deviation of the raw deviation of similar tests under normal conditions, calculated based on historical data, and Z is the calculated standardized deviation score. Its absolute value represents the current original deviation. The degree of deviation from the historical normal fluctuation range.

[0034] Standardized deviation was calculated Then, the system compares its absolute value with a preset deviation threshold. The comparison is then performed. This threshold is typically set based on the stringency requirements for the test results, for example... .like This means that a low-probability event has occurred in the current test result, which is statistically significant. In this case, the system generates an abnormal behavior flag for the result and marks it as suspicious.

[0035] Upon identifying the abnormal test result, the system immediately initiates a tracing mechanism. Based on the unique identifier of the abnormal result, the system queries the time-series database constructed in step S103 to retrieve the complete data flow link record of the test data stream that generated the result throughout its entire processing lifecycle. This record It precisely indicates which channels the data flowed through during the test, as well as the order and timing of the flow.

[0036] Get Link Records Next, it is necessary to identify the most likely source of the anomaly from all involved channels. This invention employs a probabilistic inference method based on Bayes' theorem. For records... Each channel mentioned in Calculate the posterior probability that it is the source of the fault. Bayes' theorem is as follows: in, Represents data flow link record The identifier for a specific channel involved in the process, Anomaly, represents a currently observed anomaly in the test results that has been determined by the standardized deviation z. It is a passage The prior failure probability represents the probability that channel i itself will fail without observing the current anomaly event. This data comes from the channel historical health records maintained by the system, for example, the proportion of times channel i failed in the past month out of the total uptime. It is the likelihood probability, representing the probability in a known channel. The probability of observing this specific anomaly under fault conditions. This probability is obtained by analyzing a historical fault case database, which records the abnormal signal patterns that accompanied each channel fault in history. It is the posterior probability, representing the probability after the current anomalous event has been observed. Under these conditions, it can be inferred that the anomaly originated from the channel. The system determines the most likely source of the fault by comparing the posterior probabilities of all relevant channels.

[0037] The system calculates and compares The system calculates the posterior probability of all channels and selects the channel with the highest posterior probability as the target channel for error correction. For example, if the calculated posterior probability of channel 3 is 0.72, channel 5 is 0.15, and the remaining channels are all below 0.1, then the system determines channel 3 as the most likely source of the fault. This probability-based localization method significantly improves the accuracy and efficiency of localization compared to simple guessing or polling.

[0038] S106. Based on the error correction target channel, obtain a subset of associated test data, and inject the subset of associated test data into the backup channel for processing through a redistribution algorithm to obtain the corrected test output data.

[0039] Specifically, the system searches the test task manager based on the identified error correction target channel identifier. The search objective is to find all test data that has been processed on that channel and whose processing time overlaps with the time when the anomaly occurred. This data is uniformly defined as a subset of associated test data. Since channel failures may be intermittent, this subset may contain data from multiple test tasks.

[0040] To improve the efficiency and management granularity of reprocessing, the system performs reprocessing on the acquired data subset. Preprocessing is performed. First, its integrity is verified. Then, based on data characteristics or acquisition time, simple clustering or grouping algorithms are used to divide the subsets. Divide into several smaller, logically consistent data groups.

[0041] The system queries the set of all currently healthy and load-level-allowed backup channels. To intelligently allocate data packets to the most suitable backup channels, the system calculates the matching degree between each data packet and each backup channel. The matching degree function can consider various factors, such as the backup channel. Current remaining processing capacity Average complexity of data grouping And the channel's compatibility with tasks of specific complexity. A simplified matching degree. It can be represented as: in, This represents the matching score between data group i and backup channel j. The higher the score, the more suitable the channel is for processing this data group. This represents the current remaining processing capacity of backup channel j. It is a comprehensive indicator that can be calculated by weighting the current CPU idle rate, memory free amount, and I / O bandwidth margin of the channel, reflecting the computing resources immediately available to the channel. This represents the average chip complexity index of the data group i to be assigned. This index is inherited from the aforementioned chip complexity calculation process. The higher the value, the more data processing and computing resources are required for the test. This formula tends to allocate high-complexity tasks to channels with stronger remaining processing capabilities. The matching degree is directly proportional to the remaining processing capability of the target channel and inversely proportional to the complexity of the data group to be assigned. Therefore, the system will automatically prioritize matching high-complexity data groups with channels with stronger remaining processing capabilities during allocation, thereby achieving load balancing during error correction and avoiding the creation of new performance bottlenecks on backup channels.

[0042] Based on the matching degree calculation results, the system injects each data packet into the corresponding backup channel using a redistribution algorithm. These backup channels will use the same testing procedures to retest the data packets. This process is equivalent to reproducing the previous test operations in a supposedly healthy environment.

[0043] After the retest is completed, the system compares the newly generated test results item by item with a pre-prepared, validated standard dataset. The comparison metric can be the numerical difference of key parameters. The system sets a strict allowable error threshold. If the deviation between the retest results and the standard data is less than this threshold, the data is considered to have been effectively corrected. All validated data is collected and uniformly defined as the corrected test output data.

[0044] In step S107, the system extracts the corresponding old and new data flow records from all corrected test output data and summarizes them. Then, using predefined templates or parsing rules, these raw log records are transformed into a structured log dataset that is easier to analyze and store. The system performs integrity verification on the structured trace log dataset. The primary verification metric is log integrity rate. The calculation formula is as follows: in, This is the total number of test samples in this batch that require traceability. This refers to the number of samples with a complete, uninterrupted record of the entire testing process. This metric ensures the foundation for the traceability of the testing process.

[0045] To quantify the overall reliability of the final output data in this batch, the system calculates a comprehensive data reliability score. This score is a weighted combination of multiple sub-metrics. In addition to the log integrity rate mentioned above... The pass rate of data consistency verification is usually also included. The weighted calculation formula is as follows: in, and It is a weighting coefficient used to measure the relative importance of two indicators, and satisfies the following conditions: For example, it can be set , .

[0046] The system will calculate With a preset reliability threshold Compare. If If the overall reliability of the test data in this batch is deemed to meet the standard, then the test data in this batch is considered to be reliable.

[0047] For data that meets the standards, the system performs a final fusion operation: it correlates and encapsulates the corrected test output data, its corresponding complete structured trace logs, and the calculated data reliability indicators, integrating them into the overall test report for that test. This final traceable test result set not only provides accurate test results but also includes detailed "birth certificates" and "health reports," providing a solid guarantee for data credibility and greatly facilitating subsequent problem auditing, quality analysis, and process optimization.

[0048] Reference Figure 2This invention provides a data management system for integrated circuit packaging and testing. This system, through the sequential connection and feedback control of seven dedicated modules, forms a complete "monitoring-analysis-decision-execution-verification" closed loop, achieving fully automated management from data acquisition to final report generation. The specific functions of each module and their collaborative relationships are as follows: The data acquisition module 201, serving as the system's input, is used to collect real-time load data and chip complexity indicators during testing, extract channel occupancy and the amount of data to be processed, and obtain the channel load distribution matrix. This module is the foundation of the entire system, and its output matrix provides data support for all subsequent analysis and decision-making. The traffic splitting module 202, connected to the data acquisition module 201, is used to perform preliminary classification of test data in the real-time load data according to the channel load distribution matrix and a load balancing algorithm. If the channel occupancy rate exceeds a preset occupancy rate threshold, high-complexity chip data is transferred to a low-load channel to determine a dynamic traffic splitting scheme. This module is the core of realizing dynamic load allocation, and its output traffic splitting scheme directly determines the efficient flow path of test data. The path recording module 203, connected to the diversion processing module 202, is used to obtain the data allocation path in the dynamic diversion scheme. It embeds timestamps and source / target identifiers during each channel switch using a path recording algorithm to obtain a complete data flow link record. This module provides the system with end-to-end traceability, and the link record it generates is the sole basis for subsequent problem localization and analysis. The load optimization module 204, connected to the path recording module 203, is used to determine whether there is uneven distribution based on the complete data flow link record. If the complete data flow link record shows that the channel is continuously overloaded, the distribution parameters are adjusted to obtain an updated load balancing state. This module forms a negative feedback closed loop, which can continuously optimize the system's resource allocation strategy based on historical operating data. The anomaly detection module 205, connected to the load optimization module 204, is used to extract anomaly signal indicators from the updated load balancing status and analyze the test results using statistical analysis methods to obtain test result deviation data. If the test result deviation data exceeds a preset deviation threshold, the data flow link is traced back to locate the source of the problem and determine the target channel for error correction. This module is the system's "diagnostic center," correlating and analyzing load status with test results to accurately locate fault points. The data correction module 206, connected to the anomaly detection module 205, is used to obtain a subset of associated test data based on the target channel for error correction, and inject it into a backup channel for processing through a reallocation algorithm to obtain corrected test output data. This module is the system's "repair execution unit," which receives diagnostic results and executes specific error correction actions to ensure that abnormal data is reprocessed. The report generation module 207, connected to the data correction module 206, is used to summarize the tracking logs from the corrected test output data, perform integrity verification on the tracking logs, calculate data reliability indicators, and determine whether they meet the standards. If they meet the standards, the corrected test output data is integrated into the overall test report to obtain the final traceable test result set. This module is the output end of the system; it integrates all pre-processed data and logs to generate a final test report with complete traceability. Each module in this system uses the output of the previous module as the input of the next module. Through the close connection of data flow and control flow, a complete "monitoring-analysis-decision-execution-verification" closed loop is formed, jointly realizing all the functions of the aforementioned method embodiment. Its working principle and beneficial effects correspond one-to-one with the method embodiment, and will not be repeated here.

[0049] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A data management method for integrated circuit packaging and testing, characterized in that, include: By collecting real-time load data and chip complexity indicators in multi-chip parallel testing, the current channel occupancy rate and the amount of data to be processed are extracted from the test data to obtain the channel load distribution matrix. Load balancing is determined based on the channel load distribution matrix. If the channel occupancy rate exceeds the preset occupancy rate threshold, the high-complexity chip data is transferred to the low-load channel to generate a dynamic traffic splitting scheme. Obtain the data allocation path in the dynamic splitting scheme, and embed a timestamp and source / target identifier at each channel transfer using a path recording algorithm to obtain a complete data flow link record; Based on the complete data flow link record analysis of the channel load history, if it is determined that there is a continuous channel overload, the splitting parameters in the dynamic splitting scheme are adjusted to obtain the updated load balancing status. Anomaly signal indicators are extracted from the updated load balancing status, and test result deviation data is calculated. If the test result deviation data exceeds the preset deviation threshold, the data flow link record is traced back to locate the source of the problem and determine the target channel for error correction. A subset of associated test data is obtained from the target channel for error correction, and the subset of associated test data is injected into the backup channel for processing through a reallocation algorithm to obtain the corrected test output data. The integrity of the tracking logs summarizing the test output data is verified. If the data reliability meets the standard, data fusion technology is used to generate the final traceable test result set.

2. The data management method for integrated circuit packaging and testing according to claim 1, characterized in that, The process involves collecting real-time load data and chip complexity metrics during multi-chip parallel testing, extracting initial channel occupancy and the amount of data to be processed from the test data, and obtaining a channel load distribution matrix, including: Real-time load data and chip complexity metrics are collected using a multi-chip parallel testing system. The initial occupancy rate of each channel is extracted based on the real-time load data to obtain the initial channel occupancy rate distribution; Based on the initial channel occupancy distribution and chip complexity index, calculate the amount of data to be processed for each channel; If the amount of data to be processed exceeds the preset data volume threshold, the high-load channel is subjected to data diversion processing to obtain an adjusted data volume distribution. A preliminary channel load distribution matrix is ​​constructed using the adjusted data volume distribution and the initial channel occupancy distribution. The preliminary channel load distribution matrix is ​​optimized using matrix analysis to obtain the channel load distribution matrix.

3. The data management method for integrated circuit packaging and testing according to claim 1, characterized in that, The process of determining load balancing based on the channel load distribution matrix, and if the channel occupancy rate exceeds a preset occupancy rate threshold, involves transferring high-complexity chip data to low-load channels to generate a dynamic traffic splitting scheme, including: The current load status is determined by obtaining the current channel occupancy rate in the channel load distribution matrix; If the current channel occupancy rate exceeds the preset occupancy rate threshold, the K-means clustering algorithm is used to perform preliminary classification of the test data to obtain high-complexity chip data and low-complexity chip data. Based on the channel load distribution matrix, channels with load levels below a preset load threshold are identified, and a list of low-load channels is generated. Based on the list of low-load channels and the grouping of high-complexity chip data, a set of transferable target channels is determined, and a preliminary offloading scheme is generated. The load balancing status of each channel is determined based on the preliminary load distribution scheme. If a new high-load channel exists, the simulated annealing algorithm is used to optimize the preliminary load distribution scheme and adjust the data allocation of the high-complexity chip to obtain a dynamic load distribution scheme.

4. The data management method for integrated circuit packaging and testing according to claim 1, characterized in that, The process of obtaining the data allocation path in the dynamic traffic splitting scheme involves embedding a timestamp and source / destination identifier at each channel transfer using a path recording algorithm to obtain a complete data flow link record, including: Obtain the data allocation path in the dynamic traffic splitting scheme, analyze the data flow direction through a preset path recording algorithm, and determine the initial allocation state of each path; Based on the initial allocation state, a timestamp generation mechanism is used to embed the current timestamp during channel transfer to generate a path record containing time information; According to the preset source and target identifier allocation rules, the identifiers of the source nodes and target nodes are extracted from the path records to obtain the labeled path dataset; If the labeled path dataset contains duplicate identifiers, the identifiers are deduplicated using a hash algorithm to generate a set of paths with unique identifiers. Based on the uniquely identified set of paths, a link tracing algorithm is used to determine the complete data stream transmission trajectory; By obtaining the switching frequency statistics of each channel through the complete data stream transmission trajectory, a dynamic allocation record of the data stream link can be obtained; Based on the dynamically allocated records, a consistency check algorithm is used to verify the integrity of the dynamically allocated records, and the final complete data flow link record is generated.

5. A data management method for integrated circuit packaging and testing according to claim 1, characterized in that, The process involves extracting abnormal signal indicators from the updated load balancing status, calculating test result deviation data, and if the test result deviation data exceeds a preset deviation threshold, then tracing back the data flow link record to locate the source of the problem and determine the target channel for error correction, including: Abnormal signal indicators are obtained from the updated load balancing status, and the distribution characteristics of the indicators are calculated using statistical analysis methods to obtain an abnormal signal set. If the index values ​​in the abnormal signal set exceed the preset index threshold, the isolated forest algorithm is used for anomaly detection to determine the abnormal point set. Based on the set of anomalies, test result deviation data is obtained. If the test result deviation data exceeds a preset deviation threshold, it is determined to be a deviation anomaly set. Data flow records are extracted from the set of deviations and anomalies, and time series analysis is used to backtrack the records to locate the source set of problems. Based on the set of problem sources, analyze the channel correlation in the data flow link to determine the target channel for error correction.

6. A data management method for integrated circuit packaging and testing according to claim 1, characterized in that, The step of obtaining a subset of associated test data from the target channel for error correction, and then injecting the subset of associated test data into a backup channel for processing using a reallocation algorithm to obtain corrected test output data includes: Based on the identification information of the error correction target channel, test data within the time period in which the anomaly occurred is extracted to form a subset of associated test data; If the integrity of the associated test data subset is verified to meet the preset integrity threshold, then the associated test data subset is grouped using a clustering analysis algorithm to obtain a grouped data set. The K-means algorithm is used to calculate the matching degree between the grouped data set and the backup channel. Based on the evaluation result of the matching degree, an allocation strategy is formulated to assign each data group to the backup channel with the highest matching degree. Based on the allocation strategy, the grouped data set is injected into the backup channel for processing through a reallocation algorithm to generate preliminary corrected data. If the deviation between the preliminary corrected data and the preset standard test dataset is less than the preset allowable error threshold, then a verification algorithm is used to verify the consistency of the data to obtain the verified data set. Based on the verified dataset, perform data formatting to generate corrected test output data.

7. A data management method for integrated circuit packaging and testing according to claim 1, characterized in that, The process involves aggregating the tracking logs from the corrected test output data, performing integrity verification on the tracking logs, confirming that the data reliability indicators meet the standards, and then fusing them to generate a final traceable test result set, including: The corrected data is obtained from the corrected test output data, and the tracking logs are extracted using data parsing technology to obtain a structured log dataset; Based on the structured log dataset, consistency verification technology is used to determine the log integrity, and the integrity assessment result is obtained. If the integrity assessment result meets the preset integrity threshold, then the support vector machine algorithm is used to calculate the data reliability index, and a reliability index set is obtained based on the data reliability index. Based on the aforementioned set of reliability metrics, the tracking logs are correlated with the overall test data using data mapping technology to obtain a correlated dataset. Key fields are extracted from the associated dataset, and data fusion technology is used to generate a fusion test dataset; Based on the fusion test dataset, a decision tree algorithm is used to generate the final traceable test result set.

8. A data management system for integrated circuit packaging and testing, characterized in that, include: The data acquisition module is used to extract the current channel occupancy rate and the amount of data to be processed from the test data by collecting real-time load data and chip complexity indicators in multi-chip parallel testing, and to obtain the channel load distribution matrix. The traffic splitting module is used to perform load balancing judgment based on the channel load distribution matrix. If the channel occupancy rate exceeds the preset occupancy rate threshold, the high-complexity chip data is transferred to the low-load channel to generate a dynamic traffic splitting scheme. The path recording module is used to obtain the data allocation path in the dynamic diversion scheme. By embedding a timestamp and source / target identifier during each channel transfer using the path recording algorithm, a complete data flow link record is obtained. The load optimization module is used to analyze the channel load history based on the complete data flow link record. If it is determined that there is a continuous overload of the channel, the load splitting parameters in the dynamic load splitting scheme are adjusted to obtain the updated load balancing status. The anomaly detection module is used to extract anomaly signal indicators from the updated load balancing status, calculate test result deviation data, and if the test result deviation data exceeds the preset deviation threshold, trace back the data flow link record to locate the source of the problem and determine the target channel for error correction. The data correction module is used to obtain a subset of associated test data from the error correction target channel, and inject the subset of associated test data into the backup channel for processing through a redistribution algorithm to obtain the corrected test output data. The report generation module is used to perform integrity verification on the tracking logs summarizing the test output data. If the data reliability meets the standards, data fusion technology is used to generate the final traceable test result set.

Images