Root-Cause Dashboard
Bayesian inverse over a forward physics model · clusters failure modes by category and ranks drivers for the current operating point
Top suspect
PECVD i-layer thinning (epi onset)
0% posterior
Dominant cluster
PECVD
0% of probability mass
Diagnostic certainty
100%
entropy 0.00 / max 3.81 bits
Sample window
30 mods
baseline Pmax/cell 6.37 W
Observed KPI deltas vs baseline · manually adjustable
Inferred from the last 30 modules. Drag to inject a hypothesis or simulate a fault signature; ranking re-solves instantly.
ΔPmax (cell)-0.04 W
inferred -0.04 · range -1.5–1.5
ΔVoc (cell)-0.01 mV
inferred -0.01 · range -30–30
ΔFF-0.00 pp
inferred -0.00 · range -3–3
ΔJsc+0.00 mA/cm²
inferred +0.00 · range -1.5–1.5
ΔCrack rate+0.495 /cell
inferred +0.495 · range 0–0.15
ΔBubble risk+0.157
inferred +0.157 · range 0–0.5
Failure-mode clusters · by station/category
PECVD0.0%
TCO0.0%
Print0.0%
Stringer0.0%
Lamination0.0%
Cleaning0.0%
Probability mass aggregated across single-fault hypotheses sharing a process category. A skewed cluster suggests a station-wide drift; a flat distribution indicates ambiguous signature.
Ranked candidates · top 8
| Candidate | Cat. | Prior | Posterior | Distribution | |
|---|---|---|---|---|---|
| 1 | PECVD i-layer thinning (epi onset) | PECVD | 17% | 0.0% | |
| 2 | PECVD i-layer overgrowth | PECVD | 6% | 0.0% | |
| 3 | PECVD chamber temperature drift +25°C | PECVD | 10% | 0.0% | |
| 4 | PECVD RF power excursion | PECVD | 5% | 0.0% | |
| 5 | H2 dilution out-of-band (low) | PECVD | 4% | 0.0% | |
| 6 | TCO target poisoning | TCO | 12% | 0.0% | |
| 7 | TCO sputter power droop | TCO | 5% | 0.0% | |
| 8 | Screen-print finger spreading | 8% | 0.0% |
Selected candidate · signature & remediation
PECVD i-layer thinning (epi onset)
Posterior 0.0% · prior 17.3% · Bayes factor 0.0×
Predicted vs observed signature
Recommended action
Raise SiH4 flow +2 sccm; verify QCM rate trend
Owner: PECVD eng · est. cost $220/hr · expected ΔPmax recovery -4.0 W/module
Hypothesis × KPI signature matrix
Each cell is the candidate's predicted Δ on a KPI (red = decrease, green = increase). The OBSERVED row at top is what the line is actually doing — match its color signature to a row to read off the most consistent driver.
Top-3 posterior history
Collecting history…
Tracks how the leading hypotheses evolve as new modules close. Crossings indicate a regime change in the line signature.