What this category’s data looks like
Data sources for semiconductor packaging and testing scenarios include wafer test maps, MES systems, operation logs of bonding machines and molding equipment, data collected by ATE test machines, X-ray/SAM test results, and customer specifications. Daily generated documents include packaging process cards, bonding and molding process parameter tables, incoming wafer batch maps, test program version files, yield analysis reports, and failure analysis reports. Update frequency adjusts based on production pace. Process and parameter documents are updated per production batch. Corresponding test data is generated after each chip is tested. Yield and failure data is summarized daily. Most document structures include fields such as process parameters, batch numbers, test bin codes, and equipment numbers. Units cover industrial standard units including voltage, current, temperature, and time.
What constraints these characteristics impose on knowledge base retrieval and recall
Multi-source heterogeneous data sources require unified vector embedding adaptation for documents in different formats. The high-frequency update feature requires the retrieval system to support real-time or near-real-time document synchronization. Process and test documents have highly specialized fields, so retrieval must accurately match parameter values and batch information to avoid mixed batches or incorrect use of process programs. Unlike front-end wafer manufacturing processes, the retrieval focus in this scenario is on parameters in back-end packaging and testing links, and irrelevant data from front-end processes must be excluded. At the same time, retrieval result accuracy directly affects process compliance and the risk of defective products being shipped, so the precision and scope of recall must be strictly controlled.
How to configure the setup
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirement for packaging process parameters, filters low-correlation general documents |
limit | Top 8 entries | Adapts to the number of documents per production batch, prevents workflow context overflow |
usingReRank | Enabled | Optimizes retrieval sorting precision for specialized parameters such as bonding and molding |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands process query terms, supplements implicit parameter requirements from customer specifications |
collectionIds | Bound to the exclusive knowledge base collection for packaging and testing | Isolates irrelevant data from front-end wafer manufacturing, avoids retrieval interference |
Three common mistakes
- Enabling tool calling prevents knowledge base queries: The symptom is that tool calls return "No matching knowledge base content found" with a 404 status code. The cause is that the corresponding packaging and testing knowledge base collection is not linked in the tool calling configuration, or the collection permissions are not granted to the current workflow.
- Retrieval results are not filtered by tags: The symptom is that retrieval returns documents with irrelevant tags, such as mixed-in wafer manufacturing documents. The cause is that exclusive tags are not configured for packaging and testing documents during knowledge base import, or the tag filtering switch is not enabled in the retrieval module.
- Retrieval returns an unexpected number of results: The symptom is that too many or too few results are returned, exceeding the preset range. The cause is that the
limitparameter value does not match the number of documents per production batch, or thesimilaritythreshold is set too high, filtering valid process parameter documents.
How to confirm the configuration is complete
- Log in to the FastGPT backend, enter the exclusive knowledge base for packaging and testing, check if the document import interface supports tag configuration, and confirm that tag fields such as batch numbers and process parameters have been configured for document types such as process cards and test reports.
- Enter the workflow test panel, enable the tool calling function, enter "Query the bonding process parameters for batch 2024", and verify that the packaging process card content in the corresponding knowledge base is returned.
- Call the FastGPT knowledge base API interface, submit a new molding parameter document, wait 1 minute, then perform a retrieval, and confirm that the new document has been included in the recall scope.
- Enter a query term with a specific tag, such as "Retrieve cases of bonding failure in failure analysis reports", and verify that the retrieval results only return documents with the "bonding failure" tag.
The values provided here are common starting points, and should be evaluated against the reader's own samples.