Etch Metrology Sampling Plan

# Etch Metrology Sampling Plan

An etch metrology sampling plan specifies which wafers, locations, features and repeated measurements will support a defined process decision. It connects the question being asked to the coverage and measurement evidence needed to answer it.

A depth uniformity statistic, a chamber comparison and a drift monitor can require different plans. Recording many values is useful only when their sampling units and intended interpretation are clear.

## Start with the decision

Define the measured quantity and its location: for example, depth inside a specified opening, top CD of a material line, or remaining stop thickness. State whether the decision concerns local geometry, spatial uniformity, wafer-to-wafer variation, chamber matching or change over time.

Then specify the population and process conditions to which the conclusion is intended to apply. Measurements from one pattern class or one chamber state do not automatically cover another.

The two wafer maps are illustrative site layouts, not recommended production counts. Adding diagonal sites changes angular coverage compared with sampling only the center and four cardinal directions. It does not guarantee that a local excursion elsewhere will be found.

Record wafer coordinates, orientation and any excluded edge region. An edge exclusion is part of the reported scope, not proof that the excluded area behaves like the measured region. Site count alone also does not establish spatial resolution or detection probability.

## Separate the sampling levels

A lot contains wafers; a wafer contains sites; a site can contain different structures; a structure can be measured repeatedly. These levels can contribute different kinds of variation.

In this arithmetic example, 2 wafers × 5 sites × 3 feature types × 2 repeat readings produces 60 readings. It still samples only 2 wafers. The repeated readings can help assess measurement repeatability at those features, but they do not turn the experiment into 60 independent observations of wafer-to-wafer process variation.

Different feature types may measure different responses and should not be pooled blindly. Depth, linewidth and remaining-film thickness can have different units, limits and uncertainty models. Even nominally identical features at adjacent sites can be correlated.

## Match coverage to the suspected variation

For spatial questions, include locations that can reveal the relevant radial, angular or localized behavior. For pattern-loading questions, choose representative density and geometry classes. For time-dependent questions, spread observations over the intended process history rather than repeating one short interval.

Use an appropriate selection and run-order strategy for the inference being made. Randomization, blocking or paired comparisons may help separate sources of variation, but the choice depends on the process question. Repeatedly choosing only convenient wafers or easy-to-measure structures can leave important behavior outside the plan.

A development study may need broad coverage to characterize variation before a routine monitoring plan can be justified. Reduced or adaptive sampling requires evidence that the decision remains adequately supported under its stated operating conditions.

## Include measurement capability

Keep the measurement method, recipe, calibration or reference checks, feature definitions and analysis rules with the plan. An apparent process difference can come from tool matching, preparation or the way a feature boundary is interpreted.

Replicate measurements can quantify part of measurement variation under the conditions tested. They do not, by themselves, establish accuracy, complete uncertainty or reproducibility over different instruments and days. Re-measurement can also alter some samples, so confirm that repeats are meaningful for the material and method.

## State the acceptance and escalation rules

Define how site results, spatial summaries and wafer summaries are compared with requirements. Distinguish a measured sample pass from a statement about unmeasured sites or future wafers. If statistical confidence or detection performance is claimed, use a justified model and adequate data for that claim.

Specify what triggers additional measurement, investigation or a change in the monitoring plan. Keep the original sampling and decision rules with the results rather than silently changing them after an inconvenient observation.

A useful sampling plan makes the conclusion's coverage visible. Count the readings, but also identify how many wafers, sites, feature classes and process states those readings actually represent.

## Sources and scope

• [KLA — Practical Applications of Measurement System Analysis](https://www.kla.com/documents/5.3_Practical_Applications_of_MSA_Sutherland_KLA.pdf): site, wafer and lot sampling in relation to combined variability.
• [NIST — Process Characterization: Background and Data](https://www.itl.nist.gov/div898/handbook/ppc/section5/ppc511.htm): an example of designing a wafer/site plan around a study goal.
• [2024 IRDS Metrology](https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=959302): application-dependent sampling and measurement capability considerations.

Original site maps and reading-count example. No universal site count, confidence level or production sampling frequency is prescribed.