Etch Rate Drift Budget

# Etch Rate Drift Budget

An etch rate drift budget defines how much a qualified monitor response may change before investigation or other prescribed action is needed. It keeps the baseline, measurement definition, process history and acceptance rule explicit. A changed monitor result is evidence to evaluate, not an automatic diagnosis of a chamber fault.

The chamber-match budget compares chambers. This entry compares a consistent response over time within a defined process context.

## Establish a usable reference

Specify the monitor film, its incoming condition, the measurement sites, recipe version and relevant chamber state. Record the actual duration used in the rate calculation, including what counts as etch time. A change in that definition can create an apparent rate change without the same physical process comparison.

The graph contains invented observations and a hypothetical upper engineering threshold of 102 nm/min around a 100 nm/min baseline. The final observation of 103 nm/min exceeds that example threshold. This does not establish a universal alarm value, a measured trend or a statistical conclusion.

Engineering action thresholds and statistical control limits answer different questions. A control limit derived from a characterized stable process is not the same as a product specification or an allowance chosen from integration needs. The six points shown are insufficient to qualify a complete monitoring plan.

## Calculate apparent removal rate consistently

For a suitable blanket-film thickness measurement, apparent average removal rate is the measured thickness reduction divided by the defined process duration. Use compatible pre/post measurements at specified sites and check that the film and measurement model remain valid.

In this arithmetic example, both monitors start at 200 nm and use a 1 min etch duration. The baseline ends at 100 nm, giving 100 nm/min. The new result ends at 97 nm, giving 103 nm/min. Relative to the baseline, the apparent rate increase is (103 - 100)/100 = +3%.

These values assume a meaningful thickness difference for the same material and measurement definition. Film modification, a changed optical model, nonuniform removal or measurement-system variation can complicate that interpretation. The calculation is an average over the measured interval, not proof of a constant instantaneous rate.

## Connect drift allowance to process requirements

If a constant-rate model were valid for a timed 1 min removal, a +3 nm/min change would produce 3 nm more removal over that minute. Whether that matters depends on the target thickness, protected materials, profile and other acceptance requirements.

An endpoint-controlled sequence may change duration instead, and the mask, landing exposure or profile can respond differently. A blanket monitor does not automatically predict patterned-feature clearing, CD or sidewall behavior. Establish those relationships with the intended process data before assigning a monitor allowance.

Do not automatically compensate time by the inverse rate ratio. Such a correction might restore one average removal while changing other responses or concealing an unresolved change in process state. Follow a qualified response and verify the complete module after any adjustment.

## Separate possible sources of change

Compare incoming film and surface condition, recipe execution, metrology performance and chamber history. Maintenance, conditioning, consumable state and process exposure can be relevant, depending on the equipment and integration. A monitor offset alone does not identify which factor changed.

Use repeated or reference measurements where needed to distinguish measurement variation from wafer response. Preserve site maps when spatial behavior matters; the same mean can coexist with a different center-to-edge pattern. Also check whether the baseline itself remains appropriate after a documented process change rather than silently redefining it to absorb an excursion.

## Define the action and its limits

A monitoring plan should state what is measured, when the measurement occurs, which criteria trigger action and what evidence is needed before resuming normal operation. Monitor frequency limits how much change can go undetected between checks; an acceptable last observation is not proof of the condition of every later wafer.

For a statistical monitoring claim, characterize the underlying measurement and process variation, account for relevant time dependence, and use a method appropriate to the monitored quantity. Neither a small plot nor a single rate percentage establishes capability or product yield.

A useful drift budget turns a well-defined monitor change into a qualified decision. Keep the baseline, physical interpretation and response plan visible together.

## Sources and scope

• [Lam Research — Our Processes](https://www.lamresearch.com/products/our-processes/): metrology and chamber-drift monitoring context.
• [Lam Research — Virtual Sensor Innovation for Higher Productivity](https://newsroom.lamresearch.com/virtual-sensor-innovation-for-higher-productivity?blog=true): process monitoring and chamber comparison.
• [NIST — Measurement Process Characterization](https://www.itl.nist.gov/div898/handbook/toolaids/pff/mpc.pdf): baseline characterization and measurement-process monitoring.

Original plots and assumed monitor values. Rates and thresholds are illustrative rather than a validated control chart or a production response instruction.