Skip to content
Semiconductor

Yield excursions, traced to the tool. Lots, tools and metrology together.

We read the wafer map and the tool history together, so an engineer has a disposition within the shift.

Surface defects on a wafer, each with the reason for its class. Illustrative data.
What is hard here

Two kinds of evidence.

A yield excursion shows up as a pattern on the wafer and as a window in tool history. The two live in different systems and are read by different people.

What we have learned

An edge ring already narrows the cause to a handful of steps. We treat the map as data the system can reason about, and link each finding to its lots and tools.

Agent Hive · Run 4471
What caused the edge-ring yield loss on lots from bay 4?
  1. Classifiedwafer maps · edge-ring signature1.2s
  2. Walkedgenealogy · 312 lots, 41 steps2.4s
  3. Delegatedtool-state check · 3 chambers3.1s
  4. Comparedincident 2291 · same signature0.7s
  5. Proposedhold 38 lots · for approval0.3s
Answer4 sources
ETCH-07chamber B
Shared step on 94% of affected lots

The loss tracks chamber B on ETCH-07 after the 14 Oct preventive maintenance. Edge-ring rate is flat on chambers A and C.

wafer_binslot_historypm_log #8812incident 2291
Ask a follow-up…
Illustrative data
Figures

From the wafer to the tool.

Fig. 1 · Wafer map and signatureIllustrative data

Wafer map with defect signature. Wafer map of W07 in lot L-2417: 352 dies, 90 fail, yield 74.4%. 78 failing dies form a edge ring signature; the classifier reads edge ring at 0.86 and links the wafer to investigation INV-20418 and a hold under SOP-QA-001.

Signature
Die bins after sort, with an edge-ring signature.
Fig. 2 · An example fabIllustrative data

Fab 1 overview with agent fleet and twin. Overview of Fab 1: 6 bays and 13 tools (9 running, 1 setup, 1 maintenance, 2 idle), lot carriers on the central aisle, a fleet of 5 agents and a digital twin of the same floor. Metrology tool MET-03 measured lot L-2417; SPC tripped on etch depth from RIE-02. Plant Sentinel flagged it, the RCA Investigator opened INV-20418, the lot is on hold, and the digital twin shows RIE-02's etch rate drifting 3.1% above its model. Agent statuses: Plant Sentinel: watching · 1 open trip; RCA Investigator: INV-20418 · 4 hypotheses proposed, 9 evidence links; Process Engineer: drafting a PM-interval change · pending approval; Wiki Librarian: linked INV-20418 on the RIE-02 page; Supply Planner: focus rings for ICP etch: 2 on hand, 21-day lead.

Bays, tools and lots in flight.
Fig. 3 · Die floorplanIllustrative data

Die floorplan, isometric. Isometric floorplan of a logic test die: an IO ring of pads around Logic cluster (standard-cell rows), SRAM A (4 Mb · bit-cell array), SRAM B (where Bin 7 fails land · 62%), Analog / SerDes (mixed-signal), PLL (clock generation), Interconnect (global routing band). SRAM B is highlighted: the wafer's signature fails land there. Geometry is illustrative.

Chip
Signature fails read against the blocks they land on.

Bring us one hard problem. We will say whether it fits our research.