Etch Defect Budget

# Etch Defect Budget

An etch defect budget defines the defect classes, inspection coverage and acceptance rules used to control defects associated with an etch module. It keeps measured detections, reviewed classifications and process attribution distinct.

A useful budget does not treat every inspection signal as a confirmed etch-created failure. It records what was inspected, what could be detected and which requirements apply to the resulting evidence.

## Match locations before subtracting totals

In this schematic, the before inspection has five detections and the after inspection seven. Four locations match. One is seen only before, and three are seen only after. The net increase is two, while the number of after-only detections is three.

Subtracting total counts therefore does not recover the after-only population. Even the after-only group needs interpretation: a feature may become visible because the film stack or contrast changed, rather than because a defect was physically created during etching.

Likewise, a before-only detection may have been removed, hidden or missed on the later inspection. Coordinate registration, matching tolerance, review and detection capability affect those conclusions. Retain unmatched and ambiguous cases rather than forcing every signal into a causal category.

## Define which defects count

State the classes of interest, size or signal thresholds, relevant layout locations and classification method. Particles, residues and pattern defects can require different inspections and acceptance rules. Electrical failures and buried defects may need additional methods.

Keep reviewed physical defects, unconfirmed signals and nuisance classifications traceable. A nuisance label is specific to the monitoring objective; it should not silently mean that an observed feature can never matter to reliability or another process step.

Specify how sampled review results relate to the full detection population. A few reviewed examples do not establish the identity of every unreviewed signal.

## Normalize only with a defined denominator

For the illustrative target class, sample A has 20 reviewed detections over 100 cm2, giving an observed density of 0.20 detections/cm2. Sample B has 10 over 25 cm2, giving 0.40 detections/cm2. B has fewer counted detections but twice the observed density.

This comparison assumes comparable inspection sensitivity, defect definitions and relevant coverage. Areal normalization alone does not make unlike layouts or inspection recipes equivalent. Depending on the requirement, a denominator based on inspected features or die may be more appropriate; label it explicitly.

Observed density is not automatically the true physical defect density, a whole-wafer estimate or a yield prediction. Zero detections in an inspected sample also does not prove a defect-free wafer.

## Connect acceptance limits to the integration

Derive limits from process and product requirements rather than reusing a universal defects-per-wafer number. Define whether a rule applies to a critical defect class, a spatial cluster, a wafer statistic or a particular feature population.

A statistical control limit and an engineering acceptance limit serve different purposes. A process can be statistically stable while failing a product requirement, and a shift can deserve investigation before an engineering limit is crossed.

If a yield model is used, retain its critical-area, defect-size and spatial assumptions separately. Do not insert all inspection detections into a yield equation as though they had identical failure probability.

## Preserve spatial and process context

Examine wafer maps, repeated layout locations, edge signatures and clustering alongside totals. Those patterns can guide an investigation, but they do not uniquely identify an equipment fault.

Retain the incoming state, etch recipe revision, chamber condition, strip and clean sequence, handling history and inspection settings. If the inspection follows several operations, the measured change belongs to that combined interval until further evidence isolates its source.

## Close the budget with reviewed evidence

Use matched coverage or another justified comparison strategy, representative sampling and appropriate controls. Document inspection limitations, measurement uncertainty, classification confidence and the disposition of ambiguous cases.

When changing sensitivity or classification settings, establish comparability before calling the new count a process improvement. A lower count can reflect lower capture rather than fewer physical defects.

A defect budget is meaningful when its population, denominator and acceptance rule are visible. Reviewed detections support process decisions; counts alone do not establish etch causation or device failure.

## Sources and scope

• [KLA — Defect Inspection and Review](https://www.spts.com/products/chip-manufacturing/defect-inspection-review): inspection, review and classification for process monitoring.
• [KLA-Tencor — Wafer inspection and review portfolio, July 2016](https://ir.kla.com/news-events/press-releases/detail/121/kla-tencor-introduces-comprehensive-wafer-inspection-and): complementary inspection/review information and nuisance separation.

Original diagrams, counting examples and explanatory synthesis. Counts, areas and classes are teaching assumptions, not measured process capability or an acceptance standard.