- Inspection
- Detect Anomaly Locations and Signals
- Review
- Review the Morphology of Representative Anomalies
- Metrology
- Quantitative measurement of physical dimensions, film thicknesses, and overlay alignment
- Yield
- The Passing Proportion Defined by a Specified Test and Denominator
Define Which Kind of Quality You Are Evaluating
One detected point need not cause failure; dimensional changes may not be visible in saved images, and electrical failure also depends on design sensitivity. Defect count, metrology conformance, and electrical yield cannot simply be converted into one another.
- 01DetectWhere Do Anomalous Signals Appear?
Optical or electron-beam inspection generates candidate locations, signals, and related images.
- 02ReviewWhat Do Representative Sites Look Like?
More detailed images help classification; morphology judgments also depend on sampling and imaging conditions.
- 03MeasureDo Dimensions or Materials Deviate?
Film thickness, linewidth, overlay, and other properties each have their own measurement methods and criteria.
- 04CorrelateHow Does the Device Perform Electrically and Functionally?
Test structures and chip tests provide results at different levels.
Inspection and Review: A Trade-Off Between Speed and Detail
Optical inspection can use scattering, reflection, or image comparison to detect anomalies. Review examines selected sites, often with SEM; electron-beam inspection also exists. Speed, resolution, and coverage depend on the equipment and inspection recipe.
Metrology Results Also Need Measurement Context
CD, overlay, thickness, refractive index, and composition measure different properties. Record units, measurement locations, recipes, repeatability, and specifications; a process-node name does not define a universal tolerance.
How to Distinguish Between WAT, CP, and Yield
WAT/PCM evaluates process control monitor structures, with test locations and timing determined by line protocols, not exclusively after FEOL completion. CP (circuit probe / wafer sort) coordinates probers, probe cards, and automated test equipment, contacting dies sequentially or in multi-site arrays to produce functional test results and bin maps.
Advanced: What Can a Simple Yield Model Explain?
An ideal random defect Poisson yield model is expressed as Y = exp(-D*A), where D represents effective killer defect density and A is critical area. It illustrates the sensitivity of chip area to random defects, but raw defect counts cannot be substituted directly into the formula to predict electrical sort yield. Systematic defects, spatial clustering, redundancy, and test limits fundamentally alter this relationship.
Correlating Observations Does Not Equal Proof of Causality
Overlaying defect coordinates onto failing dies identifies spatial correlations. However, attributing electrical die failure to a specific microscopic defect requires coordinate precision, physical failure analysis, and ruling out competing mechanisms. No automatic 1:1 mapping exists between localized micrographs and full-wafer sensor readings.
WAFERPILOT / APPLICATION TO THIS TOPIC
Constructing an Auditable Quality Investigation Report
When comparable data is connected, WaferPilot synthesizes defect, metrology, and electrical test evidence, highlighting shared deviations and suspect process steps. Reports distinguish observations, correlations, and proven mechanisms; root cause and recipe changes require engineering validation.
Deliverables for Engineers:A quality investigation report unifying defect, metrology, and electrical test data, clearly distinguishing spatial correlation, candidate mechanisms, and validated conclusions.
View the Current Demo and Capability Scope →Related Reading & References
Below are verifiable industry references; specific processes, permissions, and acceptance criteria are governed by fab site specifications.