What the Data for This Category Looks Like
Data for this category comes primarily from wafer test maps, MES systems, bonding and molding equipment, ATE testers, X-ray/SAM inspection equipment, and customer specification documents. Daily processed documents include packaging process cards, bonding and molding parameter sheets, incoming wafer maps, test program documents, yield reports, and failure analysis reports. Data updates occur per production batch, per individual chip test, and daily. Documents include fields such as process parameters, test results, batch identifiers, and equipment numbers. Some drawing-type documents include coordinates and dimension units. Parameter-type documents include numeric process indicators.
Constraints on Document Parsing and Chunking
Documents with mixed drawing and text content, such as incoming wafer maps and failure analysis reports, require the parsing module to handle both structured parameters and unstructured text and images. Multi-batch, high-frequency updated data requires chunking to support small-batch, high-frequency uploads. It also avoids oversized chunks that cause indexing delays. Large structural differences exist across document types. Process cards use linear process steps. Yield reports are statistical tables. Chunking rules must adapt to diverse structures. Some documents contain critical identifiers such as equipment numbers and batch numbers. Chunking must retain contextual associations, to avoid losing field relationships after splitting.
How to Set Configurations
The following configurations apply to FastGPT v4.17.0:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the natural paragraph length of documents such as packaging process cards and yield reports, avoids splitting process steps or statistical rows |
chunkSplitter | By paragraph + semantic boundary | Adapts to the structure of mixed documents, retains contextual associations of process parameters and test results |
customPdfParse | Enable mixed image-text parsing mode | Adapts to drawing and text content in incoming wafer maps and failure analysis reports, extracts image captions and parameter annotations |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Covers parsing time for large yield reports and failure analysis reports, avoids timeout interruptions |
chunkSettingMode | Auto-adapt by document type | Automatically adjusts chunking rules for different documents such as parameter sheets and drawing reports, adapts to multi-document-type scenarios |
Three Common Mistakes
- Uploading a Java interface document with a modified file extension results in no parsing output. The document format is not correctly recognized, and the file with a modified extension is parsed according to its original type. This causes the parsing engine to fail to match the corresponding parsing rules.
- Calling the pushdata API in chunk mode to upload data results in long-term display of "indexing in progress". Chunk parameters are set too large, causing single-chunk data volume to exceed the processing limit of the indexing queue, leading to queue backlog and incomplete indexing.
- Uploading HTML interface documents generated by javadoc results in no parsing output. Custom HTML parsing rules are not enabled, and the default parsing module cannot recognize code blocks and interface parameter content in HTML.
How to Verify Configuration is Correct
- Upload a single packaging process card document. Check that the number of parsed chunks matches the number of original document paragraphs, with no process steps split across multiple chunks.
- Upload a PDF document containing wafer drawings. Confirm that the parsed result includes drawing coordinates, image captions and other graphic and text information from the original document.
- Upload a yield report document under 100 MB. Confirm that parsing time falls within the preset
PARSE_FILE_TIMEOUT_SECONDSrange, with no timeout errors. - Test uploading a Java interface document with a modified file extension. Confirm that core content such as interface parameters and paths can be extracted after custom parsing rules are enabled.
The values provided are common starting points and should be measured against the reader's own samples.