Start with what should repeat
Each cycle in Growth A has the same recipe sequence and nominal exposure duration. The process log records cycle boundaries. The thickness measurement is sampled often enough to estimate an increment across each cycle. Open Analysis, select New Notebook, and enter:Check consistency across the 12 repeated deposition cycles in Growth A. Use the recorded cycle boundaries and the thickness trace to estimate thickness gained per cycle. Compare equivalent points in the recipe, report missing measurements separately, and show whether a change in timing or measurement quality could explain any apparent drift.Click Create Notebook. Confirm the notebook uses the intended trace and recipe log. See Creating a Notebook for the full instructions.
Use recorded boundaries instead of equal time bins
Start by plotting thickness together with the logged cycle boundaries. Each increment should compare thickness at equivalent points in adjacent cycles, such as the settled measurement after the same purge step. Dividing the full run into 12 equal intervals can mix exposure, purge, and idle periods when their durations vary. A requested cycle count alone does not establish where those cycles occurred.Show which timestamps define each increment and the measurement coverage around them. If a boundary falls in a gap or an unsettled measurement, mark that cycle unavailable instead of filling it with zero.Keep the individual-cycle values visible before summarizing groups of cycles.
Locate the change in cycle behavior
Fictional thickness increments measured at equivalent recipe points. Gains remain near 0.10 nm per cycle early in the run, then decline.
The last four cycles add about 14% less thickness per cycle than the first four. Check that measurement uncertainty is small enough to resolve that difference before treating it as meaningful. The recorded exposure duration stays the same. A single average across all 12 cycles would hide when the decline develops.
The graph shows a gradual change rather than one isolated low cycle. Check whether the same pattern appears with another reasonable choice of measurement points or grouping.
Distinguish deposition changes from measurement effects
Ask the assistant to compare explanations:The thickness gain declines late in the run while recorded exposure duration stays constant. Check the corresponding pressure and temperature traces, the thickness-fit quality, and whether readings have settled at the chosen recipe point. Show what each check supports and which data is unavailable.
Unchanged timing alone does not establish that the delivered dose is unchanged. Likewise, a lower apparent thickness increment does not establish a process problem if the measurement model has changed.
Leave a specific next step
A handoff from this example could say:Growth A’s measured thickness gain declines from 0.100 nm per cycle in cycles 1–4 to 0.086 nm in cycles 9–12. Logged exposure durations remain 2.0 s. Review measured exposure conditions and thickness-fit quality, then compare with an independent final-thickness measurement before attributing the decline to deposition.Save the notebook with the boundary-selection method and excluded cycles. If this becomes a routine check, create a reusable template and review its assumptions before applying it to another recipe.
Adapt the investigation
For a multilayer stack, different layers may be designed to have different thicknesses. Compare each layer with its own target or equivalent layers elsewhere in the stack, rather than treating every difference as inconsistency.
Keep spatial uniformity separate
This example measures consistency over time at the observed location. It does not establish thickness uniformity across the wafer. Spatial uniformity requires measurements that distinguish sample positions. If you have those measurements, ask a separate question:Assess thickness uniformity across the sample using the position-resolved measurements. Show measurement locations, coverage, and the definition of non-uniformity used in the summary.See Characterization for the available measurement types, or Diagnose Why a Growth Differs to compare with another run.