WAT drift root cause

The scenario and numbers in this demonstration project are simulated data.

Industry: Semiconductor (foundry)
What was built: Agents, skills, semantic models and write tools, built on demonstration data
Where the knowledge came from: The furnace vendor's service notice, plus demonstration data from the ERP, MES, EQP, mask and YMS systems

What the weekly report misses

Before the weekly yield meeting, the YMS weekly report shows the VTH_SHIFT margin of one product closing in over consecutive lots, still within spec.
The YMS weekly report shows only that the parameter drifts, not what the drifting lots have in common.

This demonstration project has semantic models for ERP, MES, EQP, mask, YMS and the data warehouse. The data warehouse model covers the data of all these systems.
The agent uses the data warehouse model to scan YMS, lists the lots with a margin below 15% and confirms the VTH_SHIFT trend.
The agent joins the lot history and tools of the drifting lots and the normal lots. Every drifting lot went through one furnace, and the normal lots went through other furnaces.
The agent checks the PM records in EQP and finds that the PM of this furnace is overdue.
The furnace vendor's service notice is in the Drive. The agent reads it and confirms that running the furnace past its PM cycle causes temperature field drift, which matches the VTH_SHIFT symptom.
A manager asks in Mimir which tool the drifting lots share. The agent answers with reports on the WAT margin trend and on the tools the drifting lots share.

The agent uses the PM compliance and downtime impact skill to assess an early PM. During the downtime, work in progress moves to the other furnaces, and capacity can absorb it.
Before the lot holds, the early furnace PM and the note to the customer are written back to YMS, EQP and ERP, the agent stops for approval.

Get started

AI feasibility review of a live workflow