What data for this category looks like
Data for this category primarily comes from wafer fab MES systems, equipment FDC modules, defect detection and metrology devices, wafer map files, process recipe libraries, and experimental datasets. Daily generated documents include process work cards, batch run records, equipment recipes, SPC statistical charts, exception handling sheets, defect maps, and yield reports. Data updates occur in real time as wafer batches are created or equipment runs, or on a daily aggregated cycle. Document fields include wafer batch identifiers, process parameter values, defect location coordinates, recipe parameter ranges, and more. Some parameters use units such as nanometers, milliamps, and millibars.
What constraints these characteristics impose on workflow orchestration
The extremely long process chain and multi-branch rework paths for this category require workflows to support flexible orchestration with multiple connected nodes and conditional branch jumps. The need to access multi-source heterogeneous data requires workflow nodes to support reading and parsing files in different formats, adapting to documents with varying structures such as wafer maps, SPC tables, and recipe parameters. Real-time and batch update cadences require workflows to have configurable trigger timing, supporting process startup based on equipment run events or batch completion nodes. The requirement for precise interpretation of nanoscale process parameters means parameter validation nodes in workflows must have high-precision numerical comparison capabilities, to avoid misjudgments caused by errors.
Configuration Settings
Configuration settings are based on the FastGPT v4.17.0 workflow orchestration interface.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| File size limit | No more than 1000 MB | Adapt to the reading requirements for large files such as wafer maps |
| Number of single retrieval results | Top 10–20 entries | Adapt to fast retrieval of massive process documents |
| File reading node | Support multi-format parsing for images, tables, and text | Adapt to diverse documents such as process work cards, wafer maps, and SPC charts |
variables | Bind wafer batch numbers, process parameter thresholds, and defect judgment rules | Enable cross-node parameter transfer and unified validation |
nodeId | Bind node identifiers for corresponding equipment or processes | Enable precise workflow node calls and status synchronization |
| API call timeout | 300–600 seconds | Cover the total time required for multi-source data reading and parameter interpretation |
Three common mistakes
- Symptom: Calling a published workflow API returns empty fields or a 500 status code. Cause: Required business parameters such as wafer batch numbers are not bound in the
variablesconfiguration, causing nodes to fail to obtain core data and interrupting workflow execution. - Symptom: The workflow cannot retrieve target process recipe documents. Cause: Search keyword parameter rules bound to
variablesare not clearly defined, making it impossible to correctly reference specified documents from the wafer recipe library. - Symptom: Parameter interpretation deviations occur during workflow execution. Cause: The parameter comparison logic of conditional branch nodes does not adapt to nanoscale precision requirements, leading to abnormal judgments that do not match actual process states.
How to confirm configurations are correct
- Manually trigger the workflow, pass a known wafer batch number, and verify that parameters output by each node match batch records in the MES system.
- Call the published workflow API, check that returned results include all preset fields bound to
variables, with no missing or null values. - Simulate parameter drift scenarios, verify that conditional branch nodes trigger corresponding rework or approval paths as expected.
- View workflow run logs, confirm that API call durations do not exceed the preset timeout threshold, and that no workflows terminate abnormally.
The values provided are common starting points and should be measured against the reader's own samples.