What the data for this category looks like
Data sources include wafer fab MES systems, real-time equipment monitoring FDC modules, defect measurement equipment, process recipe libraries, and experimental datasets. Update frequency follows each wafer lot transfer, with real-time refresh of equipment parameters, and daily summary of yield and abnormal data. Document types cover standardized process cards, single-lot run logs, equipment parameter recipes, spatially distributed defect maps, and abnormal disposal work orders. Fields include nanoscale process parameters, wafer lot identifiers, process step records, and defect coordinate information. Units include nanometers, milliseconds, lot numbers, and others.
What constraints these characteristics impose on workflow orchestration
The ultra-long process chain of wafer manufacturing and cross-department document linkage requirements require workflow orchestration to support serial and parallel combinations of multiple nodes, covering documents from multiple departments including process integration, equipment, and yield. Real-time updated equipment parameters and lot data require workflow nodes to support dynamic variable calls, to avoid expired fixed context. The precise matching requirement for nanoscale process parameters and spatially distributed defect maps requires locking field mapping rules during configuration, to prevent parameter drift and misjudgment. Limited rework windows require setting fast abnormal trigger branches in the workflow, to shorten review cycles.
How to set the configuration
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
| Variable binding | Bind the wafer lot number and process node ID in variables | Match dynamic data for each wafer lot to avoid confusion between cross-lot documents |
| Node timeout setting | 60–120 seconds | Adapt to the refresh rhythm of real-time equipment data, preventing process interruptions caused by data delays |
| Parallel node count | First 3 core department documents | Cover the three core review dimensions of process, equipment, and yield, avoiding excessive nodes causing lag |
| Field mapping rules | Lock nanoscale parameter and defect coordinate fields | Match the exclusive field characteristics of wafer manufacturing documents, improving cross-review accuracy |
| Trigger condition | Match process nodes by nodeId | Accurately trigger the document review process for the corresponding process step, adapting to the node order of the ultra-long process chain |
| Context cache | 1000–1500 characters | Adapt to the length of single-lot process documents, avoiding context overflow affecting response speed |
Three common misconfigurations
- Phenomenon: Workflow canvas drag lag, batch node operation response delay of several seconds. Cause: Parallel node count is not split by process dimension, exceeding the adaptation threshold of the current version.
- Phenomenon: Workflow run preview is normal, but no results are returned in the chat interface. Cause: The wafer lot parameter in
variablesis not bound correctly, causing the context to fail to match document data for the corresponding lot. - Phenomenon: HTTP request node cannot send form-data parameters of file type. Cause: Form fields and wafer lot log files are not mapped correctly, and the corresponding document path is not associated.
How to confirm the configuration is correct
- Manually trigger the workflow for the corresponding process step, check whether the process card and equipment recipe documents for that step are automatically retrieved.
- Simulate abnormal parameter input to verify whether the workflow triggers the review branch of the corresponding abnormal disposal order.
- Adjust the parallel node count, observe whether the workflow canvas drag and run response speed meet expectations.
- Confirm that FastGPT v4.17.0 is currently in use, check whether the binding configuration of
variablesandnodeIdmatches the current wafer lot fields.
The values given are common starting points and should be measured against the reader's own samples.