What Data Looks Like for This Category
This category of data includes packaging process cards, bonding and molding parameters, incoming wafer maps, test program versions, yield reports, failure analysis reports, and more. Data sources include wafer test maps, MES systems, bonding and molding equipment, ATE test machines, X-ray/SAM inspection equipment, and customer specification documents. Updates occur per production batch, individual test results, and daily generation. Fields include batch numbers, parameter thresholds, test bin values, yield data, and more. These are dynamically adjusted across production processes, with no unified fixed format.
Constraints for the Citation Sources and Traceability Link
Multi-source heterogeneous production data requires associating batch identifiers from different systems during traceability, to avoid cross-data source matching errors. Frequently updated individual test data requires limiting the time range when citing, to prevent calling expired test programs or yield data. Field differences across documents require unified field mapping during traceability, to ensure cited packaging parameters and bin distribution data align with actual production batches. At the same time, errors such as mixed batches or incorrect program usage require traceability results to clearly link to specific production links and equipment, supporting rapid location of failure analysis.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 8000–12000 token | This category of documents includes long parameter tables and multiple sets of test data. Sufficient traceability content must be accommodated to avoid truncation of critical batch identifiers and parameter values. |
responseData | Return only batch metadata | Redundant raw test data must be filtered out, retaining only core traceable fields such as batch numbers, equipment numbers, and test times, to reduce context length. |
detail | Enabled | Specific entries or log locations of cited documents must be displayed to support detailed traceability for process quality audits. |
| Single search limit | Top 6–10 entries | Production data for this category is aggregated by batch. Too many results will cause context overload, while this range covers multiple types of associated documents for the same batch. |
| Citation matching rule | Exact match by batch number | Unique identifiers of incoming wafer maps and physical batches must be aligned, to avoid invalid citations across batches. |
Three Common Mistakes
- Phenomenon: The generated response does not display the citation document’s source fields, or the fields are empty. Cause: The
detailconfiguration item is not enabled, or the prompt does not explicitly require displaying traceability information. - Phenomenon: The token count of a single response far exceeds the set
quoteMaxToken, triggering a context overflow error. Cause: The number of search results is not limited, or non-associated production documents are not filtered, resulting in excessive redundant data being included in the context. - Phenomenon: Search results include cross-batch production data that cannot align with the physical batch currently being audited. Cause: The exact matching rule by batch number is not configured, or the unique identifier field of the incoming wafer map is not associated.
How to Confirm the Configuration Is Correct
- Initiate a quality audit query for a specific production batch, and check whether the response includes specific entry information of cited documents.
- View the workflow’s context logs to confirm that only production documents associated with the current batch are called, with no cross-batch redundant data.
- Adjust the number of search results, verify that the response length matches the
quoteMaxTokensetting, and there is no context truncation. - Switch to the published knowledge base page, confirm that the citation source fields are displayed normally.
The values given are common starting points and should be measured against the reader's own samples.