What this category of data looks like
This category of data comes from MES systems, equipment FDC modules, defect detection and metrology equipment, wafer map files, process recipe libraries, and experimental datasets. Update frequencies cover per wafer batch, real-time, and daily. Daily documents include process flow cards, batch run records, equipment recipes, SPC charts, exception handling sheets, defect atlas, and yield reports. Fields cover process parameters, batch identifiers, defect spatial coordinates, and yield-related data. Units include nanometer-level precision parameters, milliamps, degrees Celsius, and others.
What constraints these characteristics impose on workflow orchestration
The multi-source heterogeneous data and differentiated update rhythms of this category require that trigger timelines for multi-source data pulling be configured in workflow orchestration, to prevent node execution failures caused by unready data. The ultra-long process chain and limited rework window require that the total time taken for workflow node transitions be controlled within a reasonable range, and that fast jump paths for exception branches be set. The strong coupling characteristics of nanometer-level parameters require that parameter interpretation nodes in the workflow accurately match fields and units, to avoid interpretation deviations caused by unit conversion errors. Differences in document types across departments require that the workflow support parsing and adaptation of different format documents.
How to set the configurations
| Configuration Item | Recommended Practice | Rationale for this practice |
|---|---|---|
| Variable Definition | Bind variables to wafer batch numbers and process parameter threshold ranges | Adapt to differentiated data per wafer batch, match business data characteristics |
| Node Timeout Setting | 600 seconds | Cover average execution time for multi-source data pulling and parameter interpretation |
| Maximum Number of Retrieved Entries | Top 10 entries | Adapt to average content length of wafer fabrication process documents, avoid context overload |
| Context Window | 8000 token | Adapt to total content length for multi-document parsing, meet information requirements for parameter interpretation |
| Trigger Rule Configuration | Trigger based on MES data update completion signal | Match real-time/per-batch update rhythms of data sources, prevent unready data |
| Error Branch Jump | Bind nodeId to point to the exception handling node | Quickly jump to exception links, adapt to business requirements of limited rework windows |
Three common mistakes
- Phenomenon: No console.log log output after a code run node executes, and the console cannot be checked for content. Cause: The workflow debug log switch is not enabled, or log output is not bound to a specified node.
- Phenomenon: The code run node throws a syntax error or execution failure prompt. Cause: The code directly references an unbound
variablesfield, or the parameter unit conversion logic for the wafer fabrication scenario is not adapted. - Phenomenon: The AI model selection dropdown list is empty, and no model can be selected for invocation. Cause: The model interface is not bound and permission configured in the workflow configuration, or available model lists are not imported.
How to confirm the configuration is complete
- Based on FastGPT v4.17.0, trigger a simulated wafer batch workflow run, check whether the execution status of each node matches the configured trigger rules.
- View the output logs of the code run node, confirm whether the fields bound to
variableshave correctly loaded wafer batch-related data. - Verify the exception branch jump logic, manually trigger a simulated exception, confirm whether the workflow jumps to the preset exception handling node.
- Check the output of the document parsing node, confirm that different types of process documents can be correctly parsed and associated to corresponding links.
The values provided are common starting points and should be measured against the reader's own samples.