What data for this category includes
Core data for semiconductor wafer fabrication scenarios typically originates from process integration, equipment, yield, and manufacturing departments at wafer fabs. It covers multiple document types: process flow cards, batch run records, equipment recipes, SPC charts, exception handling sheets, defect maps, and yield reports, among others. Data sources include MES systems, equipment FDC modules, defect detection and metrology devices, wafer map files, process recipe libraries, and experimental datasets. Update frequencies range from per-batch wafer output, real-time parameter reporting, to daily summary statistics. Data fields include nanoscale process parameters, batch identifiers, yield percentages, defect spatial distribution coordinates, and more. Units use category-specific identifiers such as microns, watts, and batch quantities.
Constraints for tool calling and plugins
The data for semiconductor wafer fabrication is multi-source, heterogeneous, and has widely varying update frequencies. Tool calling must support interfaces for multiple data source types, while accommodating both real-time and batch data pulling. Category-specific non-structured documents such as wafer maps and SPC charts require plugins with specialized format parsing capabilities to convert data into content usable for bid documents. The extensive process chains generate massive numbers of fields and cross-module coupled parameters, so tool calling must accurately match data sources for corresponding workflow steps to avoid resource waste from invalid data pulls. Additionally, bid documents require integration of compliance documents from multiple departments, so tool calling must support cross-module data aggregation.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Automatically matches multi-type tool calling requirements for process, yield, equipment, and other scenarios in wafer fabrication |
| Workflow/API function-call configuration | enabled | Supports calling structured process parameter and yield data interfaces to meet data filling requirements for bid documents |
| Tool activation timeout | 600 seconds | Addresses delays from multi-source data pulling, preventing calling interruptions caused by large data volumes |
| MCP service connection address | Calibrated via on-site testing for locally deployed ports | Adapts to locally deployed MCP service running and calling scenarios via npx |
| Output chart syntax | Markdown standard pie chart format | Matches rendering rules of common bid document editors to ensure charts display correctly |
Three Common Configuration Errors
- The interface shows a temporary tool activation status and cannot trigger data pulling. This occurs when the local port for the MCP service is incorrectly configured, interrupting communication between FastGPT and the MCP service.
- Generated pie charts fail to render properly in bid documents. This happens when non-standard Markdown syntax is used to define pie charts, which does not align with the parsing rules of common document editors.
- Tool calling returns an insufficient number of results. This is caused by failing to adjust the automatic matching scope of
toolChoice, only invoking tools for a single scenario without covering data needs across the full process chain.
How to Verify a Correct Configuration
- Navigate to the tool management page in FastGPT v4.17.0, check the MCP service connection status, and confirm it displays as activated.
- Initiate a test call, input a query that includes wafer yield parameters, and verify if the returned results contain structured process data and batch information.
- Generate a segment of standard Markdown pie chart code, paste it into a test document editor, and confirm the chart renders correctly.
- Review the tool calling logs, and confirm there are no error messages related to communication timeouts or parameter mismatches.
The values provided here are common starting points and should be evaluated against the reader's own samples.