What This Category of Data Looks Like
Semiconductor wafer fabrication category data primarily comes from MES systems, equipment FDC modules, defect detection and metrology equipment, and process recipe libraries. Daily documents include process flow cards, batch run records, equipment recipes, SPC charts, exception handling sheets, defect maps, and yield reports. Data update cadences cover per wafer batch output, real-time equipment parameter reporting, and daily summary statistics. Fields include wafer batch identifiers, process step numbers, nanoscale process parameters, equipment runtime duration, and defect spatial coordinates.
Constraints Imposed on Workflow Orchestration
The data characteristics of this category impose multiple constraints on workflow orchestration. Multi-source heterogeneous data sources require workflows to support cross-system data calls. Cross-node parameter transfer must be implemented via variables. The high-frequency update cadence, covering per wafer batch, real-time equipment reporting, and daily summaries, requires workflow trigger logic to adapt to dynamic data. Bind nodeId to different process step nodes. The ultra-long process chains coupled with nanoscale parameters require workflows to include multi-step verification links to prevent unrecognized parameter drift. The limited rework window requires workflows to set timeout handling nodes to avoid delays in exception handling.
Configuration Settings
This configuration explanation is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Variable Configuration | Bind variables to wafer batch numbers and nanoscale process parameters | Adapt to cross-node parameter transfer requirements for multi-source data |
| Node Binding | Bind nodeId to each process step node | Match the process splitting logic of ultra-long process chains |
| Trigger Condition | Trigger based on data update frequency: per wafer batch / per hour | Adapt to the high-frequency update data cadence |
| Timeout Threshold | 600–1200 seconds | Match the limited rework window to avoid delays in exception handling |
| Knowledge Base Association | Associate corresponding process and defect knowledge bases | Support binding of defect maps and batch data |
| Concurrency Control | Set based on actual testing, no more than 40 concurrent instances | Adapt to single-node hardware configuration to avoid resource exhaustion |
Three Common Mistakes
- Workflows cannot filter target knowledge bases by variables when calling knowledge bases. The cause is that variable association rules are not configured correctly, and variable values are not bound to knowledge base names.
- Excessive single-node concurrency causes service unresponsiveness. The cause is that a reasonable concurrency upper limit is not set based on hardware configuration, exceeding the node's resource carrying capacity.
- Workflows cannot automatically trigger the image recognition link. The cause is that the output parameters of the image upload node are not bound as trigger conditions, and automatic execution logic is not configured.
How to Confirm Successful Configuration
- Trigger a simulated wafer batch data upload, and check whether the workflow executes in sequence according to preset nodes.
- View workflow logs, and confirm that the batch identification parameters passed via
variablesmatch the data source fields. - Adjust concurrency test parameters, and confirm that nodes do not experience resource exhaustion within the preset range.
- Trigger abnormal data upload, and check whether the timeout node triggers alarms or rollbacks according to preset logic.
The values provided are common starting points and should be measured against the reader's own samples.