Defect Inspection and Review Workflow is the systematic multi-stage process of detecting, locating, imaging, classifying, and dispositioning wafer defects throughout the semiconductor fabrication flow, providing the yield-learning feedback loop that enables rapid identification and elimination of process excursions to maintain die yields above 90% in high-volume manufacturing at advanced technology nodes.

Inspection Stage 1 — Defect Detection:

• Broadband Plasma Optical Inspection: KLA 39xx series tools use broadband deep-UV illumination (200-400 nm) with multiple collection angles to detect particles, pattern defects, and residues at 10-15 nm sensitivity on bare and patterned wafers
• Laser Scattering Inspection: SP7/Surfscan tools detect particles and surface anomalies on unpatterned wafers and films using oblique laser incidence—sensitivity to 18 nm particles (LSE equivalent) on bare Si
• E-beam Inspection: multi-beam SEM tools (ASML/HMI eScan, Applied SEMVision G7) detect voltage-contrast defects (buried opens, shorts, non-visual defects) invisible to optical inspection—throughput of 2-10 wafers/hour limits to sampling
• Scatterometry-Based Inspection: optical CD metrology tools detect systematic patterning defects through spectral signature deviation from baseline—fast whole-wafer coverage at >50 WPH
• Inspection Frequency: critical layers (gate, contact, M1, via) inspected on every lot; non-critical layers on 10-25% sampling basis—inspection cost of $1-3 per wafer per layer

Inspection Stage 2 — Defect Review:

• High-Resolution SEM Review: detected defects are relocated and imaged at 1-3 nm resolution using dedicated review SEMs (e.g., KLA eDR-7380)—captures defect morphology, size, and surrounding pattern context
• Automatic Defect Classification (ADC): machine learning algorithms classify defect SEM images into 20-50 categories (particle, bridge, break, residue, void, scratch, etc.) with >90% classification accuracy
• Review Sampling: typically 50-200 defects per wafer reviewed from total detected population of 1000-50,000—statistical sampling targets root cause identification with 95% confidence

Defect Disposition and Analysis:

• Pareto Analysis: defects ranked by frequency, class, and spatial signature (random, clustered, systematic, edge)—top 3-5 defect types typically account for 60-80% of yield loss
• Spatial Signature Analysis (SSA): mapping defect locations reveals process-specific patterns—radial distributions indicate CVD uniformity issues; arc patterns suggest CMP retaining ring problems
• Killer Defect Ratio: kill ratio varies from 10-30% for particles to >80% for pattern defects on critical layers
• Baseline Management: each layer maintains a defect density baseline (D₀)—excursions >2σ trigger hold-lot investigation

Yield Learning Feedback Loop:

• Defect-to-Yield Correlation: Poisson yield model Y = exp(-D₀ × A_die) relates defect density to die yield—at N3 with 100 mm² die, D₀ must be <0.05/cm² per critical layer for >90% yield
• Inline-to-Electrical Correlation: linking inline defect locations to electrical test failures validates that inspection is capturing yield-relevant defects—correlation coefficient >0.7 indicates effective inspection strategy
• Excursion Response Time: time from defect detection to root cause identification and corrective action—target <24 hours for critical defects to minimize wafer-at-risk (WAR) from 500 to <50 wafers
• Tool Commonality Analysis: when defect excursion occurs, comparing defect rates across parallel process tools identifies the offending chamber—requires normalized defect tracking per tool and chamber

Advanced Defect Challenges at Sub-3 nm:

• Stochastic Defects: EUV-induced random patterning failures (missing contacts, bridging) cannot be distinguished from systematic defects without statistical analysis over large populations—requires die-to-die inspection at high sensitivity
• Buried Defects: defects in lower metal layers obscured by subsequent depositions—voltage-contrast e-beam inspection detects electrical impact without physical access
• Nuisance Defect Filtering: as inspection sensitivity increases to detect 10 nm defects, nuisance rate (non-yield-relevant detections) increases 10-100x—requires advanced AI-based filtering with false-positive rate <5%
• Throughput vs Sensitivity: optical inspection at maximum sensitivity processes 5-15 WPH; reduced sensitivity achieves 50+ WPH—optimizing this tradeoff per layer is key to cost-effective defect management

The defect inspection and review workflow is the yield management backbone of every advanced semiconductor fab, where the speed and accuracy of defect detection, classification, and root cause analysis directly determine how quickly process problems are resolved and whether a new technology node can ramp to profitable high-volume manufacturing within its target timeline.