Supplier and Process Quality Audit for Semiconductor Wafer Fabrication: Multi-turn Dialogue and Prompt Engineering

The data for this category primarily comes from MES systems, equipment FDC collection terminals, defect detection and metrology equipment, wafer map

What the data for this category looks like

The data for this category primarily comes from MES systems, equipment FDC collection terminals, defect detection and metrology equipment, wafer map files, process recipe libraries, and experimental datasets. Data update cycles cover each wafer batch production, real-time parameter fluctuations, and daily summary statistics. Daily documents include process flow cards, batch operation records, equipment recipe parameters, SPC control charts, exception handling sheets, defect atlases, and yield reports. Fields include wafer batch numbers, process nodes, nanoscale parameter values, defect spatial coordinates, and yield proportions. Common units are angstroms, pascals, defects per square centimeter, and similar units.

Constraints imposed on multi-turn dialogue and prompt engineering

The data characteristics of this category impose multiple constraints on the multi-turn dialogue and prompt engineering link. First, data is updated across batch, real-time, and daily cycles, so chatId must be used to bind the current wafer batch to avoid mixing data across batches. Second, documents include nanoscale parameters and spatial distribution data, so prompts must clearly define parameter units and associated logic to prevent misinterpretation. Third, the process chain is long and rework windows are limited, so multi-turn dialogue must compress context length and prioritize calling the latest records for the current process step. Fourth, in cross-department collaboration scenarios, variables must be used to pass department and process permissions to ensure the dialogue only returns audit data within the corresponding permissions.

Configuration Settings

This page’s configuration is based on FastGPT v4.17.0.

Configuration ItemRecommended ApproachRationale
chatIdBind to the current wafer batch numberEnsure each wafer batch’s dialogue context is independent, avoid interference from cross-batch data
messagesRetain the most recent 20 turns of dialogueMatch the logic of multi-process audit for wafer manufacturing, avoid context overload
variablesPass batch number, process node, department permissionsAdapt to permission control and data filtering requirements for cross-department audits
Maximum context length800–1200 charactersMatch the field length of core audit documents for a single batch, prevent call timeouts
Dialogue timeout600 secondsAdapt to time requirements for complex parameter interpretation and cross-equipment data retrieval
Question optimizationCalibrate based on actual testingAdapt to correction requirements for nanoscale professional terminology

Three common configuration mistakes

  • Clicking the dialogue opening quick shortcut returns non-preset audit content. The cause is that preset prompts for supplier qualification audit and process parameter interpretation are not bound to the shortcut button configuration.
  • Wafer maps in dialogue history fail to load without a login status. The cause is that a publicly readable image storage path is not configured, so anonymous requests cannot access static resources.
  • The dialog box continues to display when calling a form input node in a nested workflow. The cause is that the option to enable the parent workflow to take over the input box display is not turned on in the nested application configuration, causing the child application’s pop-up rules to take effect independently.

How to verify successful configuration

  • A dialogue bound to a specified wafer batch number may be initiated, and chatId can be checked to confirm correct association with the batch’s MES data.
  • Multi-turn dialogue may be tested, and the context can be confirmed to only retain the most recent specified number of audit records, with no redundant old data.
  • An unauthenticated access scenario may be simulated, and static images in dialogue history can be verified to load normally.
  • A form input node in a nested workflow may be called, and the dialog box can be verified to hide according to the parent workflow’s configuration.
  • Other configured applications may be called, and dialogue logs can be verified to be synchronously recorded to the main workflow.

The values provided are common starting points and should be measured against the reader's own samples.