What Data for This Category Looks Like
Core data for semiconductor packaging and testing scenarios comes from wafer test maps, MES systems, bonding and molding equipment, ATE test machines, X-ray/SAM detection equipment, and customer specification requirements. Daily documents include packaging process cards, bonding and molding parameter tables, incoming wafer maps, test program version files, yield reports, and failure analysis reports. Data update frequency adjusts based on production batches, individual chip test results, and daily equipment operating status. Most documents contain structured fields such as equipment parameters, batch numbers, yield bin values, and failure locations. Some drawing-type documents require extraction of pixel-level metadata.
Constraints Imposed on Retrieval and Recall
This category of data characteristics creates multiple constraints for retrieval and recall. First, frequently updated batch-level documents require the retrieval system to support incremental indexing and real-time synchronization, to prevent retrieval results from being out of sync with current production status. Second, multimodal data including wafer maps and equipment parameter tables requires the retrieval system to support hybrid matching of text vectors and image features. Third, structured fields such as test bin distribution and bonding temperature parameters have high matching precision requirements, requiring filtering of low-correlation non-specialized process documents. Finally, single-batch documents have strong specificity, and excessive recall results will interfere with on-site engineers’ rapid decision-making.
Configuration Settings
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Packaging process parameters have high matching precision requirements. This range filters low-match irrelevant documents and avoids incorrect reference of non-batch-specific process parameters. |
limit | Top 8 entries | Process documents in packaging and testing scenarios are mostly batch-specific. Too many results will distract engineers. This value balances comprehensiveness and decision-making efficiency. |
searchMode | mixedRecall | It is necessary to match both text-based process parameters and multimodal data such as wafer maps and equipment logs. Hybrid retrieval covers the recall needs of all document types. |
usingReRank | Enabled | The original ranking from hybrid retrieval may have logical deviations. Reranking places high-match key parameter documents first, improving the availability of retrieval results. |
datasetSearchUsingExtensionQuery | Enabled | Process questions often include context information such as equipment models and batch numbers. This configuration automatically expands retrieval keywords to improve recall accuracy. |
collectionIds | Bound to exclusive packaging and testing process document collections | Avoid mixing documents from wafer manufacturing, complete machine assembly and other categories. Ensure retrieval results only cover process procedures and technical standards for this scenario. |
Three Common Misconfigurations
- Issue: After uploading image-format packaging process cards, retrieval fails to return parameterized content. Cause: The image OCR parsing plugin was not configured in FastGPT v4.17.0. Uploading only raw image files cannot extract structured process parameters.
- Issue: The number of retrieval results exceeds the preset upper limit, but no limit prompt is triggered. Cause: The
limitparameter was not configured correctly, or the interface setting items were not synchronized with background parameters, causing the restriction rule to not take effect. - Issue: Hybrid retrieval response time exceeds 10 seconds, and single retrieval time exceeds the tolerance range of commercial scenarios. Cause: Full-text retrieval and multimodal vector recall were enabled at the same time, and document collections were not optimized with separate indexing based on production batches, leading to excessive retrieval load.
How to Verify Proper Configuration
- Upload an image of a packaging process card, perform a keyword search for "bonding temperature parameters", and confirm that structured results containing specific temperature values are returned.
- Initiate a retrieval, check that the number of returned results does not exceed the preset
limitparameter value, to verify that the number restriction rule is active. - Perform pure text retrieval and hybrid retrieval separately, compare response times, and confirm that hybrid retrieval time meets actual commercial scenario requirements.
- Adjust the
similarityparameter to a low value outside the specified range, verify that low-match irrelevant documents are not included in retrieval results.
The values provided here are common starting points and should be measured against the reader's own samples.