General semiconductor test data full life cycle management system

Abstract

The invention provides a general semiconductor test data full life cycle management system, and the system comprises a test source data access module which is used for uploading a source test file in a test machine to the system; the full test data analysis and standard model conversion module is used for converting the uploaded source test file into a test file in a standard data format and transmitting the test file to the multi-dimensional KPI atomic model operation and persistent storage module; the multi-dimensional KPI atomic model operation and persistence storage module is used for calculating a multi-dimensional KPI atomic result according to the data in the test file in the standard data format and storing the multi-dimensional KPI atomic result to a persistence layer database; and the unified service output module is used for receiving a user query request, performing aggregation calculation on the multi-dimensional KPI atomic result in the persistent layer database according to the user query request, and returning a KPI result meeting the user query request. The system can improve the utilization efficiency of semiconductor test data and reduce the operation cost.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor test data management technology, and in particular to a general semiconductor test data lifecycle management system. Background Technology

[0002] As a cornerstone of modern technology, the semiconductor industry's position is increasingly prominent in today's globalized world. With continuous technological advancements and innovation, the semiconductor industry is experiencing robust growth, not only driving the upgrading of electronic products but also providing impetus for global economic growth. The semiconductor industry is the core of the information technology industry, and its development level directly impacts a nation's technological strength and economic competitiveness. With the rise of new technologies such as 5G, the Internet of Things, and artificial intelligence, the semiconductor industry has ushered in unprecedented development opportunities. Globally, the semiconductor market continues to expand, technological innovations are emerging in rapid succession, and new application scenarios are constantly appearing, demonstrating strong vitality and broad market prospects.

[0003] OSAT (Outsourced Semiconductor Assembly and Testing) plants are a crucial link in the semiconductor industry chain. OSAT plants serve diverse clients, including large multinational corporations and small startups. Their semiconductor products have wide-ranging applications, including consumer electronics, automotive, medical, and industrial control. Different sectors have different testing requirements for semiconductors. The miniaturization and high integration of semiconductors are significant trends in modern electronic technology development, posing higher demands and challenges for testing.

[0004] Testing is an indispensable part of the semiconductor manufacturing process. It not only ensures product quality but also enables the timely detection of problems in the production process, thereby improving production efficiency and product reliability. Tiny defects that are difficult to detect may occur during testing, potentially leading to system malfunctions or performance issues in practical applications. To ensure high-quality testing, OSAT (Outsourced Semiconductor Testing and Assembly) facilities need to employ scientific data analysis methods to meet the needs of different products and customers. The collection, analysis, and application of semiconductor test data are crucial for ensuring product quality, improving production efficiency, promoting technological innovation, and meeting customer demands.

[0005] In the semiconductor manufacturing and testing process, the three major stages of CP (Chip Probing), FT (Final Test), and SLT (System-Level Test) output a large number of heterogeneous raw files. Besides the industry-standard STDF (Standard Test Data Format), there are also SUM (Summary Data Format) and proprietary formats such as KDF, GDF, and HDF defined by various customers. Adopting a two-stage implementation model of "raw machine files → factory server → customer-written scripts," the actual implementation process faces the following five prominent problems: (1) The system is strongly coupled with customers and the coverage is insufficient: The existing solution is a customized system jointly developed with a few strategic customers. New customers cannot quickly access it, resulting in insufficient customer coverage. (2) Offline analysis can only be performed on chip-level data within a single Lot (test batch). The processing time for a single Lot is relatively long and cannot be scaled in parallel. This function is only applicable to large customers with IT capabilities. New customers and small and medium-sized customers cannot reuse it, resulting in a "limited access" phenomenon. (3) Aggregation relies on "manual Excel pivot": When customers need aggregate analysis across batches, sites, and test stages (such as multi-lot pivot, chip test result trend), they must manually export the data in the batch number from the MES (Manufacturing Execution System) and then do Excel pivot and splicing with the test file, which takes several hours and is prone to errors. (4) Different departments have different data business needs. Currently, relying solely on MES cannot provide configurable, reusable, and shareable data services, resulting in a double waste of data silos and redundant development.

[0006] In summary, the industry urgently needs an end-to-end test data lifecycle management platform that is "fast in parsing, easy in aggregation, fast in access, and usable by multiple customers and departments" to solve the pain points of full import of test data, efficiency of cross-Lot aggregation, customer scalability, and reuse in multiple business scenarios. Summary of the Invention

[0007] This disclosure provides a general semiconductor test data lifecycle management system to solve the problems of slow parsing, difficult aggregation, difficult expansion, and departmental silos in existing semiconductor test data management.

[0008] Based on the above problems, in the first aspect, a general semiconductor test data lifecycle management system is provided, the system including: a test source data access module, a full test data parsing and standard model conversion module, a multi-dimensional KPI atomic model calculation and persistent storage module, and a unified service output module; The test source data access module is used to upload the source test files in the test machine to the system; The full test data parsing and standard model conversion module is used to convert the uploaded source test file into a test file in a standard data format and transmit it to the multidimensional KPI atomic model calculation and persistent storage module. The multidimensional KPI atomic model calculation and persistent storage module is used to calculate multidimensional KPI atomic results based on the data in the test file of the standard data format and store them in the persistent layer database; the multidimensional KPI atomic results include at least one of yield KPI, quality KPI, efficiency KPI, time KPI, and equipment hardware health. The unified service output module is used to receive user query requests, aggregate and calculate the multidimensional KPI atomic results in the persistent layer database according to the user query requests, and return KPI results that meet the user query requests.

[0009] In conjunction with the first aspect, in one possible implementation, the test source data access module is used for: The source test files are queried from a specified storage path on the test machine by manual triggering or timed scanning; if the source test files have been completed, they are uploaded to the system.

[0010] In conjunction with the first aspect, in one possible implementation, the test source data access module is used for: When the size of the source test file on the test machine exceeds a preset threshold, the source test file is compressed, and then the compressed source test file is uploaded to the system.

[0011] In conjunction with the first aspect, in one possible implementation, the test source data access module is used for: Calculate the hash algorithm checksum of the source test file, and upload the source test file and the corresponding hash algorithm checksum to the system.

[0012] In conjunction with the first aspect, in one possible implementation, the full test data parsing and standard model conversion module is used for: Based on the client, test platform, and test file format corresponding to the source test file, a customized parser extracts basic test data from the source test file, and fills the extracted basic test data into a preset XML standard template to generate an XML format test file.

[0013] In conjunction with the first aspect, in one possible implementation, the test source data access module is used to: configure the storage path for uploading source test files according to the three-dimensional isolation rule of "customer-test platform-test file type", and upload different types of source test files generated by different customers and different test platforms to the corresponding storage paths respectively; The full test data parsing and standard model conversion module is used to: monitor the uploaded source test files, and identify the client, test platform, and file type corresponding to the source test files based on the storage path where the source test files are uploaded.

[0014] In conjunction with the first aspect, in one possible implementation, the multidimensional KPI atomic model calculation and persistent storage module is used to: extract the index information of the test batch according to the test file in the standard data format and the information of the MES system, and store it in the product-level model in the persistent layer database; calculate the multidimensional KPI atomic results of the test batch based on the test file in the standard data format and store them in the multi-level model in the persistent layer database; the multi-level model includes at least one of the following: test result model, test hardware model, test batch model, and die-level model; The unified service output module is used to: receive user query requests, which include filtering conditions, aggregation logic, and target KPIs; match target test batches that meet the filtering conditions using the index information of the product-level model; call multi-level model data adapted to the aggregation logic; aggregate and return the target KPIs to the front end.

[0015] In conjunction with the first aspect, in one possible implementation, the multidimensional KPI atomic model computation and persistent storage module is used for: Based on the test result data in the test file of the standard data format, the number of corresponding chips is counted according to the test result type of the first test and the test result type of the final test, and the statistical results are stored in the test result model. Based on the test file in the standard data format, the multidimensional KPI atomic results of the test hardware are calculated according to the test hardware dimension and stored in the test hardware model. The test hardware dimension includes at least one of the test machine dimension, board dimension, and test location dimension. Based on the test file in the standard data format, calculate the multidimensional KPI atomic results of the test batch according to the test batch dimension, and store them in the test batch model; Extract basic file information and specific test results from the test file in the standard data format, and store the basic file information and specific test results in the Die-level model.

[0016] In conjunction with the first aspect, in one possible implementation, the unified service output module is used for: Using one or more of Trie, Bitmap, and Lot inverted indexes, the target test batch that meets the user's filtering conditions is matched with the index information of the product-level model.

[0017] In conjunction with the first aspect, in one possible implementation, the unified service output module includes a Web layer and a unified service interface layer; The Web layer is used to provide a visual operation interface for users to configure filtering conditions, aggregation logic and target KPIs, and to receive user query requests; The unified service interface layer is used to receive user query requests transmitted from the Web layer, and based on the user query requests, call the product-level model and multi-level model in the persistent layer database to perform data queries and aggregation calculations, and feed back the target KPIs that meet the user filtering conditions and aggregation logic to the Web layer.

[0018] The beneficial effects of the embodiments disclosed herein include: 1. Achieve unified access and parsing of heterogeneous files: The full test data parsing and standard model conversion module converts various heterogeneous source test files into test files of standard data format, breaking down data format barriers and achieving normalization processing of multiple file types.

[0019] 2. Enhance system adaptability and scalability: Based on standardized parsing and flexible access design, new customers and small and medium-sized customers can quickly adapt to the system without customized development. At the same time, it supports parallel processing of multiple Lots, breaking the monopoly pattern of a few customers and improving system reusability and scalability.

[0020] 3. Achieve efficient and accurate cross-dimensional aggregation analysis: Through multi-dimensional KPI atomic model calculation and persistent storage module, multi-dimensional KPI atomic results are calculated and stored. Combined with the automated query function of the unified service output module, it replaces manual Excel pivot operations, realizes rapid aggregation across batches, sites, and stages, and ensures accurate results.

[0021] 4. Break down data silos and reduce development costs: A persistent layer database enables unified storage of test data throughout its entire lifecycle. Combined with a unified service output module for querying and aggregation services, departments can access data resources as needed, achieving data sharing and service reuse. This eliminates the waste of data silos and redundant development, providing efficient data support for enterprise decision-making.

[0022] 5. Improve data upload reliability and efficiency: Relying on the optimized upload mechanism of the test source data access module, combined with large file compression and hash verification design, the transmission volume is reduced, file transmission integrity is ensured, and full retransmission is avoided after network anomalies, which greatly improves upload efficiency and stability. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the structure of a general semiconductor test data lifecycle management system provided in an embodiment of the present disclosure; Figure 2 A schematic diagram of a semiconductor test data visualization dashboard provided in an embodiment of this disclosure; Figure 3 A schematic diagram illustrating data interaction in a unified service output module provided in an embodiment of this disclosure; Figure 4 This is a data processing architecture diagram of a multi-dimensional KPI atomic model operation and persistent storage module provided in an embodiment of the present disclosure; Figure 5 A logic diagram for constructing a model of semiconductor test data provided in this embodiment of the disclosure; Figure 6 A flowchart illustrating the query and aggregation process for semiconductor test data provided in this embodiment of the disclosure; Figure 7 This is an architecture diagram of a unified service output module provided in an embodiment of this disclosure. Detailed Implementation

[0024] This disclosure provides a general semiconductor test data lifecycle management system. Preferred embodiments of this disclosure are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the scope of this disclosure. Furthermore, the embodiments and features described herein can be combined with each other unless otherwise specified.

[0025] This disclosure provides a system for the semiconductor industry that performs stream parsing of test files, generation of general data models, and conversion of KPI atomic models for multi-dimensional positioning. It can automatically and efficiently complete the automatic data access of test files after semiconductor production, parse them into standard data, convert them into multi-dimensional KPI atomic models, and store them persistently. It provides plant-wide test data services through browsers, system integration, and other methods, meeting the requirements of modern semiconductor production, equipment, quality, finance, and industrial engineering departments. A detailed description follows.

[0026] Figure 1 This is a schematic diagram of the structure of a general semiconductor test data lifecycle management system provided in an embodiment of this disclosure, as shown below. Figure 1As shown, the system includes: a test source data access module 100, a full test data parsing and standard model conversion module 200, a multi-dimensional KPI atomic model calculation and persistent storage module, and a unified service output module 400; The test source data access module 100 is used to upload the source test files in the test machine to the system.

[0027] In some embodiments, the functionality of the test source data access module 100 can be achieved through deployment and integration within the test machine.

[0028] The full test data parsing and standard model conversion module 200 is used to convert the uploaded source test files into test files in a standard data format and transmit them to the multidimensional KPI atomic model calculation and persistent storage module 300.

[0029] The multidimensional KPI (Key Performance Indicator) atomic model calculation and persistent storage module 300 is used to calculate multidimensional KPI atomic results based on data in test files with standard data formats and store them in the persistent layer database. The multidimensional KPI atomic results include at least one of yield-related KPIs, quality-related KPIs, efficiency-related KPIs, time-related KPIs, and equipment hardware health.

[0030] The unified service output module 400 is used to receive user query requests, aggregate and calculate the multidimensional KPI atomic results in the persistent layer database according to the user query requests, and return KPI results that match the user query requests.

[0031] This disclosure provides a general semiconductor test data lifecycle management system that effectively integrates front-end and back-end systems, rapidly analyzes large amounts of data, and provides strong decision support. By normalizing source test files (including CP, FT, and SLT files), a unified multi-dimensional KPI atomic model is abstracted, and a simple aggregation service is provided. This allows functional departments such as production, equipment, quality, finance, and industrial engineering to access the full lifecycle indicators of test data across different plant areas, thereby eliminating data silos and redundant modeling costs, enabling enterprises to maximize the use of data assets and reduce operating costs.

[0032] As one possible implementation, the test source data access module 100 is used to: query the source test files under the specified storage path of the test machine by manual triggering or timed scanning; and upload the source test files to the system after confirming that the source test files have been written.

[0033] Specifically, the configuration file predefines the monitoring folder path and file extension rules for the source test files. The system scans and captures files that conform to the extension rules under the corresponding path according to a preset period and executes the upload process. During the file upload process, the system verifies whether the current file is not being called by any process and whether any other process holds a handle to the file, thereby confirming that the file has been written and avoiding operations on files that have not been written.

[0034] Traditional semiconductor test data management solutions lack a mechanism for uploading large files in bulk during the data upload phase, which can lead to excessively long file upload times.

[0035] As one possible implementation, the test source data access module 100 is used for: When the size of the source test file on the test machine exceeds a preset threshold, the source test file is compressed before being uploaded to the system to improve file transfer efficiency.

[0036] Traditional semiconductor test data management solutions only perform raw file transfer during the data upload phase: lacking an upload integrity verification mechanism, once an anomaly occurs such as network interruption or file byte misalignment, the system cannot automatically locate the error point and can only re-upload the entire file, which prolongs the time required to upload the source test file.

[0037] As one possible implementation, the test source data access module 100 is used for: Calculate the hash algorithm checksum of the source test file, and upload the source test file and the corresponding hash algorithm checksum to the system.

[0038] In this embodiment, by verifying the correspondence between the hash algorithm checksum and the uploaded source test file, file transmission anomalies caused by network interruptions, byte misalignment, etc., can be accurately identified, preventing incomplete files from entering the system and ensuring file transmission integrity. When a transmission anomaly occurs, the system can locate missing or erroneous data segments based on the verification result, requiring only the abnormal portion to be retransmitted instead of the entire file, significantly reducing transmission time and network bandwidth usage, and improving data upload efficiency. The hash checksum provides a unique identifier for the authenticity and integrity of the file, preventing the file from being tampered with or replaced, ensuring that the source test data accessed by the system is authentic and reliable, and providing a reliable data foundation for subsequent parsing and calculation processes.

[0039] In some embodiments, after the upload is completed, the source test file is stored in the local backup folder of the test machine, while retaining the association record between the verification code and the file to achieve secure archiving of local data.

[0040] As one possible implementation, the full test data parsing and standard model conversion module 200 is used for: Based on the client, test platform, and test file format corresponding to the source test file, a customized parser extracts basic test data from the source test file, and populates the extracted basic test data into a preset XML (eXtensible Markup Language) standard template to generate an XML format test file.

[0041] In this embodiment of the disclosure, multi-dimensional positioning is performed by combining time dimension, material attributes, test hardware information and test result data. A standardized semiconductor test data model is constructed around the core indicators commonly used in the semiconductor testing field, such as yield, quality, efficiency, time, and equipment hardware health. Based on this model, the raw data in various test files are mapped to a unified XML data model.

[0042] Specifically, the preset XML standard template includes the following two types of data: one type is the conditions for the current batch test, including batch number, packaging method, product type, test version information, test code referenced in the test version, test machine, test board, start time, end time, etc.; the other type is the test information for each chip, including test start time, end time, pure test time, hardware information during test, chip serial number, QR code number, test result, chip category (HardBin), and sub-category defined by the test software (SoftBin).

[0043] Meanwhile, based on the unified parsing platform, customized parsers are configured for source test files (STDF, SUM standard format and KDF, GDF, HDF and other proprietary formats) corresponding to different customers, different formats and different products, to achieve unified parsing and conversion of various files to standard XML format.

[0044] As one possible implementation, the test source data access module 100 is used to: configure the storage path for uploading source test files according to the three-dimensional isolation rule of "customer-test platform-test file type", and upload different types of source test files generated by different customers and different test platforms to the corresponding storage paths respectively; The full test data parsing and standard model conversion module 200 is used to: monitor uploaded source test files, and identify the customer, test platform, and file type corresponding to the source test files based on the storage path of the uploaded source test files.

[0045] In this embodiment of the disclosure, the test source data access module 100 supports both automatic and manual upload modes through flexible configuration. It adopts a three-dimensional isolation upload rule of "client-test machine-file type" to clearly distinguish data from different clients, test machines and file types, avoid cross-contamination and meet the data management needs of parallel testing of multiple clients and multiple platforms.

[0046] Simultaneously, the test file generation directory is monitored periodically to ensure that correct and complete test data is automatically uploaded at the correct time and in the correct format, achieving precise isolation, real-time performance, and standardization of test data uploads.

[0047] As one possible implementation, the multidimensional KPI atomic model calculation and persistent storage module 300 is used to: extract the index information of the test batch according to the test file in the standard data format and the information of the MES system, and store it in the product-level model in the persistent layer database; calculate the multidimensional KPI atomic results of the test batch based on the test file in the standard data format and store them in the multi-level model in the persistent layer database; the multi-level model includes at least one of the following: test result model, test hardware model, test batch model, and die-level model.

[0048] The main data for the product-level model comes from the MES system and mainly includes the customer, product name, semiconductor packaging method, product type, test station, test code, test program version, packaging batch, etc. for the Lot. This data is mainly used to build a global index at the Lot level to support subsequent querying and aggregation analysis of test data.

[0049] The unified service output module 400 is used to: receive user query requests, which include filtering conditions, aggregation logic, and target KPIs; match the target test batch that meets the filtering conditions through the index information of the product-level model; call multi-level model data adapted to the aggregation logic; aggregate and return the target KPIs to the front end.

[0050] In this embodiment of the disclosure, the multidimensional KPI atomic model operation and persistent storage module 300 can perform the following three functions.

[0051] Underlying data storage: Extract specific test information and Lot test information for each product from test files in standard data format, and complete persistent storage based on product-level model and test batch model.

[0052] Multidimensional KPI Atomic Model Calculation Data Storage: Based on the above underlying data, according to commonly used semiconductor KPI categories such as yield, quality, efficiency, time, and equipment hardware health, multidimensional KPI atomic results corresponding to models such as product, test result, and test hardware are calculated and persistently stored in categories.

[0053] Lot-level data dimension expansion: By integrating with the MES system's master data, the Lot-level data dimensions are supplemented and improved.

[0054] The Unified Service Output Module 400 adopts a browser / server (B / S) architecture, providing users with an operation interface based on a single-page rich application, and supporting users to flexibly configure filtering conditions, aggregation logic and target KPIs.

[0055] The filtering criteria include time period, Lot attribute (source test file and MES system), machine number, etc.

[0056] The aggregation logic includes time (year, month, week, day, shift), Lot attributes (semiconductor packaging method, product type, test station, test code, test program version, packaging batch, etc.) and test hardware information (test machine, test board, test location, etc.).

[0057] The output KPI values ​​include semiconductor testing yield-related KPIs (first test yield, final yield, yield difference, etc.), quality-related KPIs (overall yield, retest rate, etc.), efficiency-related KPIs (overall equipment efficiency, production cycle, etc.), and time-related KPIs (average test time, hourly capacity, sorting time, comprehensive test time), etc. For example, Figure 2 This is a schematic diagram of a semiconductor test data visualization dashboard provided in an embodiment of this disclosure.

[0058] Figure 3 This is a schematic diagram illustrating data interaction in a unified service output module provided in an embodiment of this disclosure, such as... Figure 3 As shown, the unified service output module 400 uses the API gateway as a unified entry point, integrating functional units such as Lot-level models, test result aggregation, test hardware class aggregation, and test detail query. These functional units call model data from the persistent MySQL database, and the output results are encapsulated into JSON format data such as product class aggregation, test result aggregation, test hardware class aggregation, and test detail data. Finally, it is transmitted to a browser-based single-page rich application developed based on Vue / Nodejs, providing users with an interactive interface that includes filtering conditions, aggregation logic, and target KPI selection, enabling the query and display of test data.

[0059] Specifically, the unified service output module 400 can implement the following aggregated services: 1. Product-level KPI Aggregation: Based on filtering criteria, product-level models matching the filtering conditions are queried from the persistent layer database, aggregated, and the corresponding KPI values ​​are output (such as yield KPIs, quality KPIs, efficiency KPIs, and time KPIs for semiconductor testing). For example, you can view the average test yield and average test time of Customer A's product F at various test stations last week, and the number of products tested weekly and the corresponding first-test yield of Customer B at the FT test station in the last 12 weeks.

[0060] 2. Test Result KPI Aggregation: Through aggregation logic, the system queries the persistent layer database for test result models that meet the criteria, aggregates them, and outputs the corresponding KPI values ​​(e.g., number of tests, first-test yield, final yield, yield difference, etc.). For example, it can view the types of failures that customer A's product F had the most in the FT test station in the last 3 days; and the first-test yield of each machine in the SLT test station on that day, as well as the main failure issues of each test head.

[0061] 3. Hardware Test KPI Aggregation: Through aggregation logic, the system queries the persistent layer database for test hardware models that meet the criteria, aggregates them, and outputs the corresponding KPI values ​​(such as yield KPIs, quality KPIs, efficiency KPIs, and time KPIs for semiconductor testing). For example, it can view the first-test yield and total number of tests on test machines at the FT test station in the last 6 hours; or view the number of tests conducted per hour on each machine at the SLT test station for customer A's product B in the last 7 days.

[0062] 4. Test Details Inquiry: You can find the corresponding Die-level model by using any of the following conditions: Lot batch number, test time, or individual product number.

[0063] The system disclosed herein can complete the entire process from the end of Lot testing to the visualization of the user dashboard within 30 minutes, and can be deployed and replicated in any OSAT factory in a short time to form a test data lifecycle management platform with equivalent capabilities.

[0064] As one possible implementation, the multidimensional KPI atomic model computation and persistent storage module 300 is used for: Based on the test result data in the test file of the standard data format, the number of corresponding chips is counted according to the test result type of the first test and the test result type of the final test, and the statistical results are stored in the test result model.

[0065] In some embodiments, the test result type can be distinguished and defined according to Bin (Bin classification code), wherein Bin includes HardBin (hardware classification code) and / or SoftBin (software classification code).

[0066] Based on the test file in the standard data format, the multidimensional KPI atomic results of the test hardware are calculated according to the test hardware dimension and stored in the test hardware model. The test hardware dimension includes at least one of the test machine dimension, board dimension, and test location dimension.

[0067] Based on the test file in the standard data format, calculate the multidimensional KPI atomic results of the test batch according to the test batch dimension, and store them in the test batch model.

[0068] Extract basic file information and specific test results from test files in standard data format, and store the basic file information and specific test results in the Die-level model.

[0069] Specifically, the multidimensional KPI atomic results include yield-related KPIs for semiconductor testing (first test yield, final yield, yield difference, etc.), quality-related KPIs (overall yield, retest rate, etc.), efficiency-related KPIs (overall equipment efficiency, production cycle, etc.), and time-related KPIs (average test time, hourly capacity, sorting time, comprehensive test time).

[0070] The hardware health of the equipment in the multidimensional KPI atomic results can be calculated based on basic test data: The test equipment and products used in semiconductor testing have standardized characteristics, and the yield standards required by customers are usually achieved through initial testing and retesting processes. Therefore, the hardware health of the corresponding equipment can be evaluated from the following dimensions: 1. Yield-related indicators: (1) First test yield: The first test yield of the same product under standardized testing scenarios should be consistent. By comparing the differences in the first test yield of different equipment, the performance stability of the equipment can be judged intuitively. (2) Yield difference: The larger the difference between the final yield and the first test yield, the higher the proportion of abnormal results in the initial test corrected through the retest process, which indirectly reflects the possible fluctuations in the initial test accuracy or stability of the equipment. (3) Test results: By using the classification data of HardBin (hardware classification code) or SoftBin (software classification code), abnormal problems of the device hardware can be directly identified, such as temperature failure, voltage test deviation, etc.

[0071] 2. Quality indicators: By statistically analyzing the retest rate of the same product over a certain period of time, the hardware stability of the testing equipment can be inferred. A higher retest rate may indicate that there are potential problems in the equipment hardware that could cause fluctuations in test results.

[0072] 3. Efficiency and time-related indicators: By combining data such as the average test duration of the equipment, machine operation time, and hourly output, the normal operation status of the equipment hardware can be comprehensively evaluated.

[0073] Figure 4 This disclosure provides a data processing architecture diagram for a multi-dimensional KPI atomic model computation and persistent storage module, as shown in the following embodiment. Figure 4As shown, the input layer aligns and merges data with the MES system interface through standard XML format test files; then, in the model layer, a unified event flow (including memory caching) is used to build product-level models (Device / Package, etc.), test result models (Bin), test hardware models (test machines / boards / test slots), test batch models (test station, test version, quantity, etc.), and die-level models (files, single-chip test information), and completes unified Lot aggregation; based on the aggregated data, the calculation layer performs yield calculations (first test, final yield), quality calculations (Bin distribution), efficiency calculations (number of tests per hour, waiting time, etc.), time calculations (test time, year, month, day, week, shift), and health calculations (Tester-related information) to generate KPI atomic results; finally, the persistent layer assembles and stores the data in the database, realizing the link from test data integration, modeling, aggregation, calculation to storage.

[0074] Figure 5 A logic diagram for constructing a model of semiconductor test data provided in this disclosure embodiment, such as Figure 5 As shown, the original data source includes MES Lot data (including customer / product / process) and XML test files (including die / test item / value / time). These two types of data are mapped to five models through different dimensions of information: a product-level model is built from the index and aggregation conditions of the MES Lot data and the test machine and Handler information of the XML test files; a test result model is built from the first and last test Bin information of the XML test files; a test hardware model is built based on the test machine / board / test location and test results of the XML test files; a test batch model is built through the test results and time in the XML test files; and a die-level model is built based on the file information and test results of the XML test files. In this way, the scattered data is structured according to business dimensions, providing support for subsequent KPI calculation and data query.

[0075] In some embodiments, the data output by the multidimensional KPI atomic model operation and persistent storage module 300 is stored in a relational database according to the data structure and logic of the product-level model, test hardware model, test batch model, test result model, and die-level model.

[0076] In some embodiments, caching data structures can be used to assist in fast retrieval.

[0077] As one possible implementation, the unified service output module 400 is used for: Using one or more of Trie (prefix tree), Bitmap (bitmap), and Lot inverted index, the target test batch that meets the user's filtering conditions is matched with the index information of the product-level model.

[0078] Trie: The index information of the product-level model contains a large amount of string data (such as customer name, product model, test station name, test code, etc.). When users set such filtering conditions, the Trie tree can directly locate the test batch corresponding to the relevant string through prefix matching, without traversing all the data. For example, if the user's filtering condition is "product F series of customer A", the Trie tree can first match the string prefix of customer A, and then match the prefix of product F, quickly narrowing down the range of target test batches.

[0079] Bitmap: The index information of the product-level model contains a large amount of enumerated data (such as product type, test station type, packaging method, Bin classification, etc.), which can be mapped to binary bits in the Bitmap.

[0080] Lot Inverted Index: Using product-level model index information (such as Lot batch number, test time, customer, product name, test hardware, etc.) as keywords, a mapping table is constructed between keywords and Lot test batches. When users set multi-dimensional filtering conditions (e.g., product F of customer A is tested in the FT test station in May 2025), the inverted index can be used to obtain the Lot batch set corresponding to each keyword, and then the final target batch is obtained through the intersection operation of the sets.

[0081] Figure 6 A flowchart illustrating the query and aggregation process for semiconductor test data provided in this disclosure is shown below. Figure 6 As shown, the front-end parameters corresponding to the user query request are first passed to the product-level model for indexing. The second-level LotID list is quickly filtered out by one or more of Trie, Bitmap, and Lot inverted index. Then, based on this list, data from the four major models (test batch model, test result model, test hardware model, and Die-level model) are called to finally generate the aggregated KPI results.

[0082] The multidimensional KPI atomic model operation and persistent storage module 300 outputs and stores the product-level model and multi-level model in the persistent layer database, providing data support for the unified service output module 400.

[0083] As one possible implementation, the unified service output module 400 includes a Web layer and a unified service interface layer.

[0084] The web layer provides a visual interface for users to configure filtering conditions, aggregation logic, and target KPIs, and to receive user query requests.

[0085] In some embodiments, the Web layer adopts the SOA (Service-Oriented Architecture) design pattern to provide users with a simple and intuitive operation page, supporting users to select filtering conditions, aggregation logic and target KPIs through simple operations.

[0086] The unified service interface layer is used to receive user query requests transmitted from the Web layer, and based on the user query requests, call the product-level models and multi-level models in the persistent layer database to perform data queries and aggregation calculations, and feed back the target KPIs that meet the user filtering conditions and aggregation logic to the Web layer.

[0087] In some embodiments, the API layer uses the RESTful specification. The API layer interacts with the Web layer to provide data support for the web page and operates on the database to query and store data. The API layer also handles data interaction with third-party systems, supporting functions such as big data integration within the factory and MES system calls.

[0088] For example, Figure 7 An architecture diagram of a unified service output module provided in this disclosure embodiment is shown below. Figure 7 As shown, the unified service output module 400's data storage layer uses a database cluster to carry data, distinguishing between cold and hot data; the core interaction layer includes web pages and API interfaces, supporting users to access web pages and generate email reports through a browser, and also supporting other systems to call via API; the HA (High Availability) layer achieves high availability and high load through a high-availability transparent proxy, supporting the display of test data. In addition, the unified service output module 400 covers security systems (including identity recognition, access control, data encryption, security auditing, backup and recovery, users and permissions) and infrastructure-related dimensions (including operating system, performance, log recording, and anomalies).

[0089] This invention relates to the subfield of test data management in the field of semiconductor manufacturing and testing technology, and in particular to an end-to-end test data lifecycle management system for various types of raw data used in semiconductor testers, such as wafer probe testing, final product testing, and system-level testing, which enables real-time uploading, streaming parsing, and multi-dimensional slice querying of raw data.

[0090] Compared to traditional test file viewing methods, this disclosure provides a system for test data management in OSAT factories. This system can cover test data for all types of test files from all customers and has the following advantages: 1. Unified access across platforms, clients, and file types: By adapting customized parsers to various test files for different clients, combined with a dual-queue processing mechanism for large and small files and the support of multiple machines, efficient and unified access and parsing of all types of test files can be achieved while ensuring file upload consistency.

[0091] 2. General Data Model and Multidimensional KPI Atomic Model: By constructing a general data model, the standardization and unification of test data within the factory are achieved; based on preliminary research and multiple iterations, a multi-dimensional KPI atomic model covering all scenarios is established to achieve comprehensive integration and normalized management of core KPI indicators.

[0092] 3. Unified Service Output: The system provided in this disclosure can complete data archiving within 30 minutes of test completion. By providing convenient aggregation services, it enables functional departments such as production, equipment, quality, finance, and industrial engineering to access the full lifecycle indicators of test data across different factory areas, thereby eliminating data silos and redundant modeling costs, and helping enterprises maximize the use of data assets and reduce operating costs.

[0093] In summary, the system provided in this disclosure can improve production efficiency, reduce costs, decrease error rates, and save on operation, maintenance, and development human resources.

[0094] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes in the drawings are not necessarily essential for implementing this disclosure.

[0095] The sequence numbers of the embodiments disclosed above are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0096] Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from its spirit and scope. Therefore, if such modifications and variations fall within the scope of the claims of this disclosure and their equivalents, this disclosure is also intended to include such modifications and variations.

Claims

1. A general-purpose semiconductor test data lifecycle management system, characterized in that, The system includes: a test source data access module, a full test data parsing and standard model conversion module, a multi-dimensional KPI atomic model calculation and persistent storage module, and a unified service output module; The test source data access module is used to upload the source test files in the test machine to the system; The full test data parsing and standard model conversion module is used to convert the uploaded source test file into a test file in a standard data format and transmit it to the multidimensional KPI atomic model calculation and persistent storage module. The multidimensional KPI atomic model calculation and persistent storage module is used to calculate multidimensional KPI atomic results based on the data in the test file of the standard data format and store them in the persistent layer database; the multidimensional KPI atomic results include at least one of yield KPI, quality KPI, efficiency KPI, time KPI, and equipment hardware health. The unified service output module is used to receive user query requests, aggregate and calculate the multidimensional KPI atomic results in the persistent layer database according to the user query requests, and return KPI results that meet the user query requests.

2. The system according to claim 1, characterized in that, The test source data access module is used for: The source test files are queried from a specified storage path on the test machine by manual triggering or timed scanning; if the source test files have been completed, they are uploaded to the system.

3. The system according to claim 1, characterized in that, The test source data access module is used for: When the size of the source test file on the test machine exceeds a preset threshold, the source test file is compressed, and then the compressed source test file is uploaded to the system.

4. The system according to claim 1, characterized in that, The test source data access module is used for: Calculate the hash algorithm checksum of the source test file, and upload the source test file and the corresponding hash algorithm checksum to the system.

5. The system according to claim 1, characterized in that, The full test data parsing and standard model conversion module is used for: Based on the client, test platform, and test file format corresponding to the source test file, a customized parser extracts basic test data from the source test file, and fills the extracted basic test data into a preset XML standard template to generate an XML format test file.

6. The system according to claim 5, characterized in that, The test source data access module is used to: configure the storage path for uploading source test files according to the three-dimensional isolation rule of "customer-test platform-test file type", and upload different types of source test files generated by different customers and different test platforms to the corresponding storage paths respectively; The full test data parsing and standard model conversion module is used to: monitor the uploaded source test files, and identify the client, test platform, and file type corresponding to the source test files based on the storage path where the source test files are uploaded.

7. The system according to claim 1, characterized in that, The multidimensional KPI atomic model calculation and persistent storage module is used to: extract the index information of the test batch according to the test file in the standard data format and the information of the MES system, and store it in the product-level model in the persistent layer database. Based on the test file in the standard data format, the multidimensional KPI atomic results of the test batch are calculated and stored in the multi-level model in the persistent layer database; the multi-level model includes at least one of the following: test result model, test hardware model, test batch model, and die-level model. The unified service output module is used to: receive user query requests, which include filtering conditions, aggregation logic, and target KPIs; match target test batches that meet the filtering conditions using the index information of the product-level model; call multi-level model data adapted to the aggregation logic; aggregate and return the target KPIs to the front end.

8. The system according to claim 7, characterized in that, The multidimensional KPI atomic model computation and persistent storage module is used for: Based on the test result data in the test file of the standard data format, the number of corresponding chips is counted according to the test result type of the first test and the test result type of the final test, and the statistical results are stored in the test result model. Based on the test file in the standard data format, the multidimensional KPI atomic results of the test hardware are calculated according to the test hardware dimension and stored in the test hardware model. The test hardware dimension includes at least one of the test machine dimension, board dimension, and test location dimension. Based on the test file in the standard data format, calculate the multidimensional KPI atomic results of the test batch according to the test batch dimension, and store them in the test batch model; Extract basic file information and specific test results from the test file in the standard data format, and store the basic file information and specific test results in the Die-level model.

9. The system according to claim 7, characterized in that, The unified service output module is used for: Using one or more of Trie, Bitmap, and Lot inverted indexes, the target test batch that meets the user's filtering conditions is matched with the index information of the product-level model.

10. The system according to claim 7, characterized in that, The unified service output module includes a Web layer and a unified service interface layer; The Web layer is used to provide a visual operation interface for users to configure filtering conditions, aggregation logic and target KPIs, and to receive user query requests; The unified service interface layer is used to receive user query requests transmitted from the Web layer, and based on the user query requests, call the product-level model and multi-level model in the persistent layer database to perform data queries and aggregation calculations, and feed back the target KPIs that meet the user filtering conditions and aggregation logic to the Web layer.

Images