What the data for this category looks like
Production data for semiconductor packaging and testing comes from wafer test maps, MES systems, bonding and molding equipment logs, ATE tester outputs, X-ray/SAM test results, and customer specification documents. Data updates are triggered by production batches, single-chip test results, and daily summaries. Daily production documents include packaging process cards, bonding and molding parameter sheets, wafer incoming material maps, test program version files, yield reports, and failure analysis reports. Fields include batch ID, wafer number, process parameters, test bin codes, failure types, and units such as degrees Celsius, Pascals, and volts.
What constraints these characteristics impose on workflow orchestration
Data sources for this category are scattered, and update frequencies vary widely. Workflow orchestration must accommodate scheduled pulling and real-time synchronization for multi-source data. Log formats from different devices are inconsistent. Configure format adaptation nodes in the workflow to handle heterogeneous data. Single-chip test data volume grows rapidly as batch size increases. Set reasonable batch processing nodes in the workflow to avoid memory overflow. Field requirements for customer specification documents change with each order. The workflow must support dynamic adjustment of ledger summary field mapping rules to ensure generated reports meet current order compliance requirements. Daily summarized production data must be verified and reported within a fixed time window. Configure scheduled trigger mechanisms in the workflow to ensure processes run on schedule.
How to configure the workflow
The following configurations are based on the workflow orchestration feature of FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
| Global variable pool | Bind batch ID and test bin code fields in variables, create independent variable groups per production batch | Isolate storage for multi-batch data, avoid cross-batch data confusion |
| Text content extraction component | Configure field extraction rules for packaging process cards and failure analysis reports | Match the fixed field structure of documents in this category, improve extraction accuracy |
| Scheduled trigger node | Set to trigger full ledger summary at 2:00 AM daily | Align with the daily update rhythm of production data, avoid occupying resources during production hours |
| Loop call configuration | Divide loop batches based on single-batch data volume, process 100–200 entries each time | Adapt to the volume of single-batch test data, prevent workflow timeouts |
nodeId binding | Configure a fixed nodeId for the ATE tester data pulling node | Facilitate precise reference to the output of this component in subsequent verification nodes |
| Failure retry configuration | Set 3 retries for equipment data pulling nodes, with 60-second intervals between each retry | Address temporary device disconnections or temporarily unavailable logs, improve workflow stability |
Three common configuration mistakes
- Phenomenon: The workflow global variable history does not display the output content of a specified component. Cause: The option to mount output to global variables was not enabled in the component configuration, so the variable pool did not include the return data of this node.
- Phenomenon: The text content extraction component returns empty values. Deleting and re-adding the component with the same configuration allows normal extraction. Cause: The data source node of the target document was not correctly associated during initial configuration, and re-adding automatically completed the missing data source link.
- Phenomenon: Some large language models cannot extract text content normally, while other large language models can complete the extraction. Cause: Process documents in this category contain a large number of professional parameters and non-standard formats. Different large language models have varying adaptation capabilities for parsing industrial structured documents.
How to confirm the configuration is complete
- Manually trigger a single workflow run, check if the global variable includes preset fields such as batch ID and test bin code, confirm that data mounting works correctly.
- Run the text extraction component on a packaging process card document, verify that the extraction result includes preset process parameter fields, confirm the extraction rule adaptiveness.
- Simulate a device data pulling failure scenario, check if the workflow triggers the preset retry mechanism, confirm that the failure handling configuration is active.
- View the logs of the scheduled trigger node, confirm that the full summary process starts at the preset time, verify the accuracy of the scheduled configuration.
The configuration values listed above are common starting points. Adjust them based on your own specific samples and operational conditions.