Correlation Testing is the statistical analysis methodology that quantifies the strength and direction of the relationship between an upstream leading indicator (inline metrology measurement) and a downstream lagging indicator (electrical test parameter or yield) — enabling predictive disposition of wafers at early process stages, virtual metrology systems that replace expensive physical measurements, and real-time process monitoring that provides early warning of yield excursions hundreds of process steps before they would otherwise be detected.

What Is Correlation Testing?

• Definition: Correlation testing uses statistical methods (Pearson correlation, regression analysis, principal component analysis) to determine whether variation in an upstream measurement reliably predicts variation in a downstream outcome. The key metric is R² (coefficient of determination), which quantifies the fraction of downstream variation explained by the upstream predictor.
• Leading vs. Lagging Indicators: In a 500-step semiconductor process, a film thickness measurement at Step 40 is a leading indicator — it is available immediately. The device speed measured at electrical test (Step 500) is a lagging indicator — it is not available for 6–8 weeks after Step 40. Correlation testing links these two measurements across the temporal gap.
• Threshold for Actionability: An R² > 0.7 indicates a strong correlation suitable for process monitoring and early warning. An R² > 0.85 enables predictive disposition — scrapping or reworking wafers at the early step based on the inline measurement, saving all downstream processing cost on wafers that would ultimately fail.

Why Correlation Testing Matters

• Early Scrap (Cost Avoidance): If gate oxide thickness at Step 100 correlates strongly (R² > 0.9) with transistor leakage at electrical test, wafers with oxide thickness outside the predictive window can be scrapped at Step 100 instead of continuing through 400 more process steps — saving $3,000–$10,000 per wafer in wasted processing cost.
• Virtual Metrology: When a physical measurement is destructive, slow, or expensive (e.g., TEM cross-section, reliability testing), correlation with a fast inline measurement enables virtual prediction of the expensive metric for every wafer, not just the sampled ones. This provides 100% coverage without 100% measurement cost.
• SPC Enhancement: Standard SPC monitors individual parameters in isolation. Correlation analysis enables multivariate SPC that monitors the relationship between parameters — detecting shifts in the correlation structure that indicate process degradation even when individual parameters remain within their individual specification limits.
• Root Cause Acceleration: When a yield excursion occurs at electrical test, correlation maps immediately identify which upstream process steps have the strongest statistical linkage to the failing parameter, directing engineering investigation to the most likely root cause rather than searching through 500 process steps.

Correlation Analysis Methods

Correlation Testing is connecting the dots — the statistical bridge between what you can measure early and what you care about later, enabling predictive manufacturing where quality problems are detected and contained at the earliest possible process stage.