What this category of data looks like
Data for this category comes from wafer test maps, MES systems, bonding and molding equipment, ATE test machines, X-ray/SAM inspection equipment, and customer specification documents. 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 is updated by production batch, single-chip test result, and daily frequency. Documents contain clear fields such as batch numbers, parameter values, test bin codes, and failure types. Some drawing documents include attached coordinates and dimension units.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The strong association between incoming wafer maps and physical batches requires retrieval to bind the batch number field, to avoid cross-batch matching errors. High-frequency updated single-chip test data and daily yield reports require the retrieval link to support real-time incremental synchronization. Multi-source data includes structured parameter tables and unstructured drawings, so retrieval logic compatible with both document formats is required. Precise matching of bonding and molding parameters requires retrieval recall results to be sorted by parameter precision, to prevent production abnormalities caused by incorrect use of process parameters.
How to set the configurations
The following configurations are based on the knowledge base retrieval function of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for packaging parameters and test programs, and avoids recall of incorrect parameters with low matching scores |
limit | Top 10 entries | Controls the number of single recall results, adapting to the retrieval volume of single-batch process parameters and test data |
searchMode | mixedRecall | Meets the retrieval needs of both structured parameter tables and unstructured drawing documents |
usingReRank | Enabled | Performs secondary sorting on recall results by batch number and parameter precision to optimize matching accuracy |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms related to batch numbers, improving the matching efficiency between incoming wafer maps and physical batches |
collectionIds | Knowledge base collections grouped by production batch | Limits the retrieval scope to avoid invalid recall of cross-batch documents |
Three common mistakes
- Retrieval results work normally during the debug preview phase, but no retrieval results or empty lists are returned after embedding into a self-owned page. The cause is that the page does not correctly carry the context parameters for knowledge base authorization, or the cross-domain configuration does not cover the domain name of the FastGPT retrieval interface.
- Embedded wafer test maps and X-ray inspection maps in retrieval results fail to load normally, with a picture loading failure displayed on the interface. The cause is that the knowledge base has not been configured with external link access permissions for image resources, or the image storage path is not bound to the knowledge base resource whitelist.
- Process parameter documents from non-current production batches appear in retrieval results, failing to accurately match the production needs of the current batch. The cause is that the
collectionIdsconfiguration for limiting retrieval knowledge base collections has not been set, leading to recall of cross-batch documents.
How to confirm configurations are correctly set
- Upload a single batch’s packaging process card and corresponding incoming wafer map, initiate a retrieval, and check whether the returned results are limited to the document collection of the current batch.
- Upload a newly added yield report document, initiate a retrieval, and check whether the result includes the newly added document.
- Enter the knowledge base retrieval configuration panel, confirm that all preset switch states and parameter values match the preset configurations.
- After embedding the retrieval component into a self-owned page, perform a retrieval and check the network request logs to confirm that the interface call parameters match the configuration items.
The values given are common starting points and should be measured against the reader's own samples.