What the Data for This Category Looks Like
This category of data covers records related to the entire packaging process, sourced from wafer mapping diagrams, manufacturing execution systems (MES), bonding and molding equipment, ATE test machines, X-ray/SAM testing equipment, and customer specification requirements. Daily documents include packaging process cards, bonding and molding parameter sheets, incoming wafer maps, test program version records, yield reports, and failure analysis reports. Data is updated synchronously based on production batches, single-chip test results, and daily production progress. Individual data entries include dedicated fields such as equipment parameters, test bin values, and failure locations.
Constraints Imposed by These Characteristics on Vector Models and Indexes
The strong association between incoming wafer maps and production batches requires vector indexes to support multi-dimensional associated matching to avoid mixed batch risks. High-frequency single-chip test data generates massive short text and structured data, requiring an indexing strategy configured for small-batch high-frequency updates. Differences in fields across document types require vector models to support vectorization of both structured parameters and unstructured failure reports. Daily updated yield data requires real-time incremental indexing to avoid traceability delays caused by latency. Bin distribution data from ATE tests requires precise matching with corresponding batch production parameters, so indexes must support multi-field joint retrieval.
Configuration Settings
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Ensures semantic integrity for long texts such as packaging process cards and failure analysis reports, preventing key production parameters from being split |
| Overlap Length | 100–150 characters | Covers association information between equipment parameters and test bin values across chunks, preventing semantic breaks |
| Similarity Threshold | 0.75–0.85 | Differentiates vector matching accuracy for different failure modes in the same batch, adapting to subtle differences in ATE test bin distributions |
| Single Search Limit | Top 20 entries | Covers multi-dimensional records for a single production batch, avoiding missing key failure association data |
maxToken | Set to the selected embedding model/service maximum supported input tokens | maxToken is the input-token maximum supported by the selected embedding model/service; verify it in that provider configuration. See the v4.17.2 model-field reference. It is not retrieval/LLM context or a FastGPT-wide fixed cap. |
| Reranking | Enabled | Optimizes matching sorting for failure analysis reports and wafer mapping diagrams, improving the efficiency of locating root causes of defects |
Three Common Configuration Errors
- Phenomenon: Milvus fails to start because the compose file includes PostgreSQL-related configurations. Cause: Confused the general vector database deployment template with dedicated configurations for the semiconductor packaging and testing scenario, and did not replace it with the lightweight deployment file adapted to FastGPT v4.17.0.
- Phenomenon: Interface errors occur when connecting to a locally deployed Qwen3-Embedding-8B model. Cause: Did not correctly configure the model's request timeout parameters, causing vectorization requests to exceed preset thresholds.
- Phenomenon: Retrieval results fail to match the corresponding batch wafer mapping diagrams. Cause: Did not set up multi-field joint indexes, only relying on text semantic matching without associating vector links between batch numbers and equipment parameters.
How to Verify a Correct Configuration
- View the vectorization task logs in the FastGPT knowledge base to confirm that all document types have completed vectorization without errors.
- Manually input the test bin values and equipment parameters for a single batch to verify whether the retrieval results include the production records of the corresponding batch.
- Check the storage capacity of the vector database to confirm that the data update frequency matches the production rhythm, with no backlogs or delays.
- After enabling the reranking function, verify whether the sorting of retrieval results conforms to the priority logic of failure root causes.
The values provided are common starting points and should be measured against the reader's own samples.