What data for this category looks like
Data for this category comes from wafer fab MES systems, equipment FDC monitoring devices, defect inspection metrology equipment, process recipe libraries, and experimental databases. Daily maintenance documents include process flow cards, batch run records, equipment recipe parameters, SPC monitoring charts, exception handling sheets, defect maps, and yield reports.
Data update cadences cover each wafer batch output, real-time equipment parameter fluctuations, and daily yield summaries. Each data entry includes wafer ID, process step, nanoscale process parameters, equipment operating voltage, defect coordinates, and yield values. Supported formats include structured tables and unstructured images.
What constraints these characteristics impose during deployment and upgrade
Wafer fabrication data is multi-source, heterogeneous, and updated at high frequency. Deployment requires adaptation to multi-format data source access, so dedicated database and cache services must be configured.
High-frequency updates of real-time equipment parameters require cache services to support low-latency read/write operations. This avoids impacting response speed of inspection and maintenance Q&A.
Unstructured wafer maps and SPC charts require large-capacity storage support. Appropriate access permission controls must also be configured to prevent internal process data leaks.
Concurrent access to the knowledge base by multiple departments requires database connection count configurations optimized for concurrent scenarios. This avoids query timeouts.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
DB_MAX_LINK | 150 connections | Adapts to concurrent access demands from multiple departments in a wafer fab |
REDIS_URL | redis://:custom password@internal host address:6379/0 | Caches real-time equipment FDC data to reduce multi-source data query latency |
MONGODB_URI | mongodb://username:password@internal host address:27017/wafer_knowledge?authSource=admin | Stores unstructured documents such as wafer batch metadata and defect maps |
STORAGE_S3_ENDPOINT | https://s3.north-1.aliyuncs.com or internal storage node address | Stores large-volume process documents including wafer maps and SPC charts |
FE_DOMAIN | https://wafer-maintenance.your-company.com | Configures frontend access domain to comply with internal network access rules of wafer fabs |
FILE_TOKEN_KEY | randomly generated 32-bit string | Encrypts access tokens for process documents to prevent unauthorized access to internal data |
Three common errors
- Phenomenon: After configuring the model interface address in the frontend, no available options appear in the platform's model list. Cause: The environment variables of the backend container were not updated synchronously, or the interface address does not have access permissions enabled for the wafer fab internal network.
- Phenomenon: The system cannot initiate a conversation, and a key verification failure prompt is displayed. Cause: The AI conversation key was not configured correctly, or the key is not bound to the call permissions of the corresponding large language model.
- Phenomenon: A GPU resource call failure error occurs when running the document parsing container. Cause: The
--gpus allparameter was not added to the docker run command, or the host machine's GPU driver version is incompatible with the container image.
How to confirm the configuration is complete
- Log in to the FastGPT v4.17.0 admin backend, enter the system configuration page, verify the connection information of database configuration items such as
REDIS_URLandMONGODB_URI, and perform connectivity testing. - Upload a wafer batch run record document, perform a retrieval test, and confirm that matching equipment inspection procedures and maintenance plans can be returned.
- View the backend container runtime logs, confirm that there are no abnormal messages such as database connection timeouts or storage permission errors.
- Access the configured
FE_DOMAINaddress in the internal network environment, confirm that the frontend page loads normally and conversations can be initiated.
The values provided are common starting points and should be measured against the reader's own samples.