Etch Metrology Guard Band

# Etch Metrology Guard Band

An etch metrology guard band separates a product tolerance limit from a measurement acceptance limit under a stated decision rule. It makes room for the measurement uncertainty considered by that rule, rather than treating every reported number inside the product tolerance as equally conclusive.

This guide applies the idea to an illustrative etched dimension. It does not define a universal fab release policy or a required guard-band size.

## Keep the two intervals distinct

The product tolerance describes the permitted dimension. An acceptance interval describes which reported measurement results are accepted under the chosen rule. Changing the measurement rule does not physically change the product tolerance.

For a teaching example, assume a product tolerance of 95-105 nm and an expanded measurement uncertainty U of 1 nm for each result. Choose the explicit rule: accept a result only when its reported interval from result minus U to result plus U lies entirely within the product tolerance, including its boundaries.

With constant U = 1 nm, this rule accepts reported values from 96 to 104 nm. The inward separation is 1 nm at each end. This is one assumed rule, not a claim that every application should use guard bands equal to expanded uncertainty.

## Work through the decision

A reported dimension of 103 ± 1 nm gives an interval of 102-104 nm, entirely within the assumed tolerance. It is accepted under this example rule. A reported dimension of 104.5 ± 1 nm gives 103.5-105.5 nm, which extends above the upper tolerance limit.

The second result is not accepted under the stated rule even though its central value is below 105 nm. That outcome does not, by itself, prove that the physical dimension is outside tolerance. The disposition of a non-accepted result—such as additional evaluation—must be defined by the actual process policy.

A reported uncertainty interval is not a guarantee that the true dimension is inside hard bounds. Its interpretation depends on how uncertainty was evaluated and the documented coverage information. No false-acceptance probability is established by the simple arithmetic shown here.

## Use an uncertainty evaluation for the actual measurement

Specify the measured feature and location: top CD, bottom width, depth or remaining thickness can require different measurement methods and models. Include relevant calibration, repeatability, material or profile dependence, tool matching and analysis effects without counting the same contribution twice.

Expanded uncertainty is distinct from a single repeatability standard deviation or an instrument display resolution. Its evaluation and coverage factor need documentation. Repeating a measurement can reduce some random components under suitable conditions, but it does not automatically remove a shared bias or establish accuracy.

If uncertainty varies across the range or between samples, the accepted central-value interval under this example rule can vary too. Do not silently apply a single guard-band number where the assumed uncertainty is no longer valid.

## Separate measurement uncertainty from process variation

Variation between sites and wafers describes the process population or sampling scope. Measurement uncertainty describes the information associated with a specified measurement result. They can interact in an analysis but should not be merged without an explicit model.

Accepting a measured site under a guard-band rule does not establish that every unmeasured feature passes. The sampling plan, spatial coverage and any population-level conclusion require their own justification. A guard band also does not replace monitoring of drift, profile or protected-layer loss.

## Document the rule before making the decision

Record the tolerance limits, measurement definition, uncertainty evaluation, acceptance rule and handling of boundary or non-accepted results. Keep the rule with the reported result. Choose and qualify it according to the intended application and relevant requirements rather than selecting a convenient margin after seeing the data.

A useful guard band makes the measurement decision transparent. Report what the rule accepts and what the evidence supports without confusing non-acceptance with a demonstrated physical defect.

## Sources and scope

• [JCGM 106:2012 — Measurement uncertainty in conformity assessment](https://www.bipm.org/documents/20126/2071204/JCGM_106_2012_E.pdf/fe9537d2-e7d7-e146-5abb-2649c3450b25): tolerance intervals, acceptance intervals and guard-band concepts.
• [NIST — Basic definitions of uncertainty](https://pml.nist.gov/cuu/Uncertainty/basic.html): interpretation and expression of measurement uncertainty.
• [NIST — Accuracy Considerations for Critical Dimension Semiconductor Metrology](https://www.nist.gov/publications/accuracy-considerations-critical-dimension-semiconductor-metrology): reference measurement and dimensional accuracy context.

Original diagrams and an assumed decision-rule example. Dimensions, uncertainty and guard-band size are teaching values, not a normative interpretation or a production acceptance policy.