What the data for this category looks like
Production ledger data for semiconductor wafer fabrication comes from MES systems, equipment FDC monitoring data, defect detection and metrology tools, wafer map files, process recipe libraries, and experimental data. Data updates align with wafer lot transfer nodes, real-time collected equipment operating parameters, daily aggregated lot yield rates, and anomaly records. Daily documentation includes process flow cards, single-lot operation records, equipment operating recipes, SPC statistical charts, anomaly handling sheets, defect maps, and yield summary tables. Fields include wafer lot identifiers, process step parameters, equipment operating durations, defect location coordinates, density metrics, and more. Most parameter units use nanoscale and industrial production units such as nanometers, watts, and hours.
What constraints these characteristics impose on the tool calling and plugins workflow
This category’s ultra-long process chain and multi-source data requirements mean tool calls must accurately link full-chain ledgers for individual wafer lots, and avoid mixing data across lots. Real-time equipment parameters and daily aggregated yield data are both used, so plugin call response delays must match production cycles. Excessive wait times are not acceptable. Field formats vary across data sources, so plugins must support custom data mapping rules to align MES lot number fields with coordinate fields from defect detection tools. The nanoscale coupling of process parameters requires tool calls to only extract parameters relevant to the current process step, to prevent redundant data from interfering with analysis. Limited rework windows mean plugins must quickly trace back historical anomaly handling ledgers, to shorten root cause analysis time.
How to set up the configuration
The following configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Adapt to multiple ledger extraction and plugin calling scenarios, automatically select the appropriate tool to complete tasks |
| Workflow/API function-call configuration | enabled | Standardize plugin call parameter formats, ensure ledger data across multiple data sources can be correctly parsed |
| Plugin call timeout threshold | 600 seconds | Match the real-time data collection requirements of wafer fabrication equipment, avoid long waits that interrupt production processes |
| Maximum number of data entries per plugin call | Top 20 entries | Limit the volume of ledger data returned in a single lot call, prevent context overload |
| Plugin data source mapping rules | Calibrated based on actual measurements | Adapt to field differences across multiple data sources such as MES and equipment FDC, ensure accurate data extraction |
| Abnormal data filtering threshold | Calibrated based on actual measurements | Filter parameter data outside normal process ranges, improve the effectiveness of report generation |
Three common configuration errors
- Phenomenon: An OpenApi interface call returns a parameter parsing failure or a
500status code. Cause: ThefunctionCallparameter is not configured correctly, and is mistakenly set to disabled. This causes the plugin call parameter format to not comply with FastGPT interface specifications. - Phenomenon: The content extraction node cannot connect to a self-built model interface, and ledger data extraction cannot be completed. Cause: The model adaptation configuration for plugin calls is not enabled. Only basic conversation parameters are configured, and plugin call permissions for self-built models are not enabled.
- Phenomenon: Internet-connected plugin calls fail, and external process reference data cannot be obtained. Cause: No plugin source or proxy adapted to domestic networks is configured. This causes plugins that rely on overseas services to fail to access normally.
How to confirm the configuration is complete
- Submit a test request for a single wafer lot ledger query, verify that the returned results only include process, equipment, and anomaly data for the specified lot.
- Review plugin call logs, confirm that the actual values of the
toolChoiceandfunctionCallparameters match the configured settings. - Test the process of connecting a self-built model to plugin calls, confirm that ledger content from the corresponding data source can be extracted normally.
- Simulate parameter data outside the normal process range, verify that the plugin processes the relevant content according to the configured filtering rules.
The values provided here are common starting points, and should be evaluated against the reader's own samples.