Understanding Yield
Yield definitions, loss categories, and the economics of yield improvement
Yield Fundamentals
Yield Fundamentals
Yield is the fraction of manufactured dies that work correctly. It's the single most important metric in semiconductor manufacturing:
- Die yield: (Good dies / Total dies per wafer) × 100%. This is the headline number.
- Wafer yield: Fraction of wafers that complete the fab process without being scrapped.
- Parametric yield: Fraction of dies that meet performance specifications (speed, power, leakage).
- Bin yield: Fraction of dies in each speed/power bin (premium vs. budget grades).
The classic Poisson yield model: Y = e^(-D₀ × A), where D₀ is defect density (defects/cm²) and A is die area (cm²). Key insight: larger dies have exponentially lower yield.
Key Concept: The Yield Learning Curve
New process technologies start at low yield (30–50%) and improve over months through yield learning — systematically identifying and eliminating defect sources. Mature processes achieve 90–99% yield. Accelerating yield learning by even a few weeks translates to hundreds of millions in additional revenue.
Categories of Yield Loss
Categories of Yield Loss
Yield loss can be decomposed into several categories:
| Category | Description | Typical Contribution |
|---|---|---|
| Random defects | Particles and random pattern defects | 30–50% of yield loss |
| Systematic defects | Design-process interaction failures | 20–40% |
| Parametric failures | Devices out of spec (speed, leakage) | 10–30% |
| Edge/peripheral | Dies near wafer edge with poor process control | 5–15% |
| Test/packaging | Failures during probe test or packaging | 2–5% |
Understanding which category dominates helps focus improvement efforts. ML models can decompose yield loss by correlating failures with upstream process data, defect inspection results, and design features.
Knowledge Check
Knowledge Check
1 / 2According to the Poisson model, what happens to yield when die area doubles?