What Data for This Category Looks Like
This category’s data primarily comes from wafer fab MES systems, equipment FDC monitoring tools, defect detection and metrology equipment, process recipe libraries, and experimental data. Daily documents include process flow cards, batch run records, equipment recipes, SPC charts, exception handling sheets, defect maps, and yield reports. Data update rhythms cover each wafer batch generation, real-time parameter reporting, and daily aggregation. Fields include wafer batch identifiers, process parameter values, defect spatial coordinates, yield metrics, and more. Units correspond to process physical quantities and metrology standards.
What Constraints Do These Characteristics Impose on the Workflow Orchestration Link
The ultra-long process chain and nanoscale parameter coupling characteristics require that workflow nodes be strictly arranged in process sequence to avoid cross-node parameter transfer errors. The real-time and batch update rhythms of multiple data sources require workflows to support mixed trigger logic: triggered per wafer batch, real-time parameter pulling, and daily aggregation verification. The limited rework window of the business requires that workflow failure handling nodes quickly jump to the exception handling link to avoid batch scrapping. The cross-department data collaboration requirement requires that the workflow's variables configuration uniformly map the field standards of each department to reduce data ambiguity.
How to Configure Settings
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
variables | Map fields from data sources such as MES and equipment FDC to standardized variables | Unify the format of cross-node data transfer and avoid parameter ambiguity |
nodeId | Configure according to the actual node ID of wafer manufacturing processes | Ensure that node trigger order matches process logic |
| Node trigger timing | Trigger when each wafer batch completes a process node, combined with real-time parameter threshold triggers | Match the update frequency and business trigger logic of this category |
| Timeout threshold | 600 seconds | Adapt to the average waiting window of a single batch manufacturing process |
| Number of failure retries | 2 times | Avoid pollution spread caused by repeated triggering of equipment instructions |
| Data cache upper limit | 100 MB | Adapt to the storage requirements of a single wafer batch |
This document is written based on FastGPT v4.17.0.
Three Common Misconfigurations
- Phenomenon: After the workflow runs,
variablescannot obtain real-time parameters from equipment FDC. Cause: The fields of the real-time data source are not correctly mapped to the workflow variable configuration, causing the node to fail to read the corresponding data. - Phenomenon: After importing a general workflow template, the node order is disordered and cannot match the wafer manufacturing process logic. Cause: The node arrangement is not adjusted according to the ultra-long process chain characteristics of this category, and a non-targeted template is directly reused.
- Phenomenon: After the workflow is executed, the returned results do not meet expectations, and the wafer map data of the corresponding batch cannot be correctly associated. Cause: The correct data source association rules are not configured, causing variables to not be correctly bound to the exclusive data of the corresponding batch.
How to Confirm the Configuration Is Correct
- Verify that the workflow node order completely matches the process sequence listed on the wafer manufacturing process card.
- Verify that the
variablesconfiguration includes all data source fields that require cross-node transfer, with no omissions or incorrect mappings. - Run a test workflow for a single wafer batch, confirm that the input and output of each node match the expected data format and range.
- Validate that the failure branch jump logic triggers normally when an exception handling sheet is triggered, and the jump path aligns with business requirements.
The values provided are common starting points and should be evaluated against the reader’s own samples.