What this category of data looks like
Data related to equipment inspection and maintenance Q&A for semiconductor wafer fabrication typically comes from MES systems, equipment FDC modules, defect detection and metrology equipment, process recipe libraries, experimental databases, and similar sources. Data update cycles cover per wafer batch, real-time, and daily frequencies. Included documents are process flow cards, batch run records, equipment recipes, SPC charts, exception handling sheets, defect maps, yield reports, and more. Data fields include wafer batch numbers, process temperature and pressure parameters, defect spatial coordinates, yield values, and some parameters carry nanometer-level precision and cleanroom-related units.
What constraints these characteristics impose on the tool calling and plugins workflow
Wafer manufacturing data is updated per batch and includes multi-dimensional coupled parameters. This requires tool calling to support binding input parameters by batch to avoid cross-batch data confusion. Long process chains and nanometer-level parameters require tool calling to link multiple knowledge bases or workflows, to cover cross-module exception attribution needs. Real-time updated equipment FDC data requires tool calling response times to align with cleanroom operation rework window limits, and cannot be excessively long. Additionally, equipment inspection and maintenance Q&A processes large volumes of structured and unstructured data. The context length of tool calling must adapt to the scale of single-batch data to avoid interference from redundant information during analysis.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Adapts to scenarios of parallel multi-tool calling in equipment inspection Q&A, automatically matching tool types such as knowledge bases and workflows |
| Workflow/API function-call configuration | enabled | Ensures the parameter format of tool calling complies with the structured field requirements of wafer manufacturing equipment data, avoiding format parsing failures |
| Tool call timeout | 120–180 seconds | Wafer manufacturing equipment parameter analysis requires loading multi-batch historical data, so the timeout must cover the conventional time required for data retrieval and analysis |
| Maximum number of return entries per tool call | Top 8 | Equipment inspection Q&A focuses on core abnormal parameters; too many entries increase context redundancy |
| Workflow trigger parameter binding | Map by wafer batch number | Matches the per-batch update feature of wafer manufacturing data, ensuring parameters passed to workflows match the currently processed batch |
Three common errors
- When a workflow is used as a called tool, the uploaded wafer batch run record file cannot pass its file link as a parameter to the workflow. The interface displays empty workflow input fields or parameter format errors. The cause is that no parameter mapping rule was configured to bind the public access link of the uploaded file to the workflow's input fields.
- When multiple equipment inspection questions are triggered simultaneously, knowledge base call results are mixed, with abnormal data from different questions cross-contaminated. The cause is that no rule was set to call the knowledge base grouped by question, leading to cross-context interference.
- The defect map data returned by tool calling does not match the current wafer batch, showing the defect distribution of historical batches. The cause is that no wafer batch number was bound as a filter condition in the tool calling configuration, resulting in retrieval of cross-batch historical data.
How to confirm the configuration is correct
- Initiate a single equipment inspection question test, check whether tool calling automatically triggers the corresponding knowledge base or workflow, and verify that the trigger mode matches the configured
toolChoicesetting. - Upload a wafer batch run record file, check whether the file link is correctly passed to the bound workflow input fields, and confirm that the parameter mapping configuration takes effect.
- Initiate a multi-question concurrent test, check whether knowledge base calls are processed grouped by question, with no cross-context issues.
- View the FastGPT v4.17.0 tool call logs, confirm that the returned equipment parameter data includes the current batch's fields and units, matching the configured filter rules.
The values provided here are common starting points and should be measured against the reader's own samples.