What This Category of Data Looks Like
Semiconductor packaging and testing data covers packaging process cards, bonding and molding parameters, incoming wafer maps, test program versions, yield reports, failure analysis reports, and more. Data sources include wafer test maps, MES systems, bonding and molding equipment, ATE test machines, X-ray/SAM inspection equipment, and customer specification requirements. Data update frequencies cover per batch, per chip test, and daily intervals. Core fields include batch numbers, test bin values, process parameters, failure locations, and corresponding physical units.
Constraints on Multi-turn Dialogue and Prompt Engineering
Data sources for this category are scattered and update frequently. Multi-turn dialogue must be bound to a unique batch identifier to avoid cross-batch data confusion. Business documents contain a large number of professional parameters and charts. Prompt engineering must clearly specify field ranges and formats to prevent the model from outputting irrelevant content. The volume of test data per chip is large. Dialogue context must be limited in length to prevent model response timeouts. Support for uploading attachments such as wafer test maps is required to assist in locating failure locations, which imposes requirements for industry-adapted attachment parsing.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chatId | Bind to the unique batch number of the current production batch | Used to associate all dialogue contexts for the same batch, ensuring accuracy of data traceability |
messages | Limit single-turn dialogue context to the most recent 20 interaction records | Avoid excessive historical data causing model response delays, and align with the update frequency of business data |
variables | Preset fields including batch numbers, ATE test bin tables, packaging process cards, and more | Reduce repeated input, and align with the standard format of business data |
| Maximum context length | 800–1200 characters | Adapt to the total number of fields for single-batch test data, preventing exceptions caused by context overflow |
| Supported attachment upload types | .csv, .png, .pdf | Match the formats of business documents such as wafer test maps, yield reports, and failure analysis reports |
Three Common Misconfigurations
- Phenomenon: The model returns content that does not match preset business rules. Cause: The prompt is not configured in the system message position of
messages, or preset parameters are not bound tovariables - Phenomenon: Dialogue records cannot be exported or retrieved. Cause: Session storage configuration for
chatIdis not enabled, resulting in no persistence of session data - Phenomenon: No response is received after uploading a wafer test map. Cause: Attachment parsing rules adapted to semiconductor industry image formats are not configured, or the attachment size exceeds the limit
How to Verify Proper Configuration
- Initiate a test dialogue, input "Query the bonding parameters of the current batch", and confirm that the response automatically associates the preset batch variables
- Upload a packaging process card PDF, and confirm that the parsed text fragments of the attachment are displayed in the dialogue interface
- View the session list, and confirm that dialogue records for the current batch can be retrieved and exported
- Send historical dialogue that exceeds the context length, and confirm that the model automatically truncates redundant content
The values provided are common starting points and should be measured against the reader's own samples.