What the data for this category looks like
Data for this category primarily comes from wafer fab MES systems, equipment FDC monitoring modules, defect detection and metrology equipment, wafer map databases, process recipe libraries, and experimental datasets. Daily documentation includes process flow cards, batch run records, equipment recipe parameters, SPC control charts, exception handling sheets, defect atlases, and yield reports. Data update cadences cover per wafer batch generation, real-time equipment parameter reporting, and daily summary statistics. Fields include wafer batch ID, equipment runtime (hours), defect coordinates (micrometer scale), process temperature (degrees Celsius), pressure values (pascals), and other high-precision unit fields. Single-batch data volume increases as process complexity rises.
How these characteristics create constraints for HTTP interfaces and external systems
Scattered data sources and varying update frequencies require HTTP interfaces to support concurrent pulling of multi-source data and low-latency responses. High-precision fields and batch-level data require interfaces to retain original formats during transmission, to avoid precision loss and data truncation. Long process chains and limited rework windows require interfaces to transmit complete context data. This data covers full-process parameters and exception records for a single batch. The high-risk consequences of equipment exception handling require interfaces to include complete failure retry and exception alert mechanisms. These mechanisms prevent wafer batch scrap risk caused by interrupted interface calls.
How to set configurations
These configurations are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
Authorization | Bearer ${设备FDC_API_TOKEN} | Complies with HTTP interface authentication standards, and fits the authentication rules of internal wafer fab APIs |
variables | {"wafer_batch_id": "{{batch_id}}", "equipment_sn": "{{eq_sn}}"} | Binds wafer batch and equipment serial numbers, to accurately match corresponding inspection data and process documents |
responseData | Extract fields code, message, answer, reference_docs | Standardizes return of Q&A results and associated process and exception handling documents |
| Request timeout | 30–60 seconds | Fits the latency characteristics of semiconductor equipment data transmission, to avoid session interruption from timeouts |
| Maximum retry attempts | 2 | Prevents excessive load on internal wafer fab equipment monitoring systems from repeated requests |
| Context window length | 8000–12000 token | Covers complete process parameters and exception records for a single wafer batch, to ensure complete Q&A logic |
Three Common Misconfigurations
- Symptom: Configured authentication tokens are lost after system restart. Interface calls return 401 Unauthorized errors. Cause: Tokens are only stored in local session cache, and not persisted to the wafer fab configuration center or local configuration files.
- Symptom: No streaming output is received after calling the interface. Only complete response packets are returned. Cause: HTTP chunked transfer encoding is not enabled, and the interface return is not configured to match the streaming response format.
- Symptom: AI conversations cannot associate retrieved wafer defect atlases or process cards. Reference information is not displayed correctly. Cause: Document ID, title, and other fields from knowledge base retrieval results are not passed to the conversation interface via
variables, so associated references cannot be generated.
How to Confirm Configurations Are Properly Set
- Initiate a test call, pass preset wafer batch parameters and equipment serial numbers, and verify that returned response content includes corresponding equipment inspection rules and exception handling plans.
- Check interface logs to confirm that the
Authorizationfield format meets requirements, and no unauthorized errors appear. - Simulate a short network interruption, verify that the interface triggers retries as configured, and no session interruption or data loss occurs.
- Enable streaming output testing, confirm that response content is returned in segments, and no exceptions of overall delayed loading appear.
The values given are common starting points and should be measured against the reader's own samples.