What This Category of Data Looks Like
This category of data typically originates from wafer fab MES systems, equipment FDC modules, defect detection and metrology tools, process recipe libraries, and experimental datasets. Daily documentation includes process flow cards, batch run records, equipment recipes, SPC control charts, exception handling sheets, defect maps, and yield reports. Data updates sync in real time with wafer batches. Some bulk reports update daily. Most fields contain nanoscale process parameters, batch numbers, defect coordinates, and yield percentages. Common units include nanometers, watts, batch numbers, and other specialized manufacturing units.
Constraints on HTTP Interfaces and External Systems
The multi-source heterogeneous data, real-time update cadence, and precise process parameters of this category require HTTP interfaces to support cross-system data aggregation queries. They must adapt to field formats from multiple data sources such as MES and equipment FDC modules. The ultra-long process chain and limited rework windows require interface response latency to stay within a reasonable range. This avoids disrupting process scheduling. Nanoscale parameter accuracy requires interfaces to retain original units and precision during data transmission. No forced format conversions are allowed. Multi-process documents linked to a single wafer batch require interfaces to support batch recall of relevant data by batch number. They must also support transmission formats for unstructured data such as defect maps and wafer maps.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Authorization | Bearer ${FASTGPTAPIKEY} | Complies with the OpenAI-compatible interface authentication standard for FastGPT v4.17.0, ensuring secure interface call permissions |
| variables | Pass {waferBatchId: "", processStep: "", defectType: ""} | Adapts to queries by wafer manufacturing batch, process step, and defect type dimensions, enabling precise targeting of target process data |
| responseData | Return structured content containing waferBatchId, processDoc, defectMap, yieldRate | Matches standard data extraction needs of wafer fab process departments, reducing secondary parsing costs |
| Single Recall Limit | Top 8 entries | The number of process documents associated with a single wafer batch typically does not exceed 8, preventing excessive return data from the interface |
| Timeout Period | 240–300 seconds | Accounts for data aggregation latency across MES and equipment FDC systems, while meeting real-time requirements for wafer fabrication scheduling |
Three Common Misconfigurations
- Symptom: The interface returns a 405 Method Not Allowed status code. Cause: The interface call was not initiated using the POST request method, or the
Authorizationparameter was not correctly configured in the request header. - Symptom: The agent returns empty fields when calling the interface to upload wafer batch files. Cause: The parameter mapping for the corresponding file upload was not configured in variables, and the batch number field rules for fab documents were not matched.
- Symptom: Request rate limiting occurs even after configuring multi-key load balancing. Cause: Request queues were not split according to the call frequency threshold for individual keys, and traffic sharding was not implemented for real-time data queries in wafer manufacturing.
How to Verify a Correct Configuration
- Send a test request with a known wafer batch number, and verify that the returned data includes the batch’s process documents and associated parameters.
- Check the interface call logs to confirm that the
Authorizationparameter format is correct, with no authentication failure logs. - Simulate multiple concurrent requests to verify that the interface response latency meets the preset process scheduling requirements.
- Upload a test wafer defect map file, and confirm that the interface can correctly receive and associate the file with the corresponding batch data.
The values provided are common starting points and should be evaluated against the reader’s own samples.