Within-Wafer Uniformity (Wiwnu)

Within-wafer non-uniformity (WIWNU) describes how much a measured property varies across one wafer. In CMP, that property might be remaining film thickness, removed thickness, or removal rate. These answer different questions: a uniform removal rate does not guarantee a uniform final film when the incoming film is uneven. Always name the measured quantity before quoting the percentage.

The wafer map comes before the number. A thickness mapper samples selected coordinates rather than every point on the surface. Record the site pattern, valid-site count, edge exclusion, measurement method, and film model. Moving the outermost sites inward can hide an edge problem without changing the process. A dense central cluster can also dominate an equally weighted statistic even though it represents little wafer area. Comparing runs requires a consistent sampling and weighting policy.

“One percent” needs a formula. A coefficient-of-variation convention uses WIWNU = 100 × s / mean, where s is the sample standard deviation of positive thickness measurements. Here s uses the n − 1 denominator; software using n gives a different result, particularly for small maps. Some reports quote three times the standard deviation instead. A half-range convention, 100 × (maximum − minimum) / (2 × mean), is another possible summary, not an interchangeable name for the same calculation. Read the report definition, including its denominator, before comparing values.

For an invented five-site example, thicknesses of 98, 99, 100, 101, and 102 nm have a mean of 100 nm and sample s of about 1.58 nm. The resulting 1s/mean value is 1.58%, the 3s/mean value is 4.74%, and the half-range/mean value is 2.00%. Nothing about the wafer changed between those three numbers. Near zero remaining thickness, relative metrics become unstable; absolute variation and the film-clear endpoint are more useful companions.

Spatial information survives only in the map. Rearranging the same thickness values leaves their mean and standard deviation unchanged. Yet a center-thick profile, an edge ring, and a one-sided gradient suggest different causes. Retain the two-dimensional map alongside radial summaries: averaging around a radius can conceal an asymmetric problem. An isolated bad reading deserves a metrology check, not automatic deletion merely because it worsens the score. Document rejected sites and the reason for rejection.

CMP corrects a profile, not just an average. At a matched coordinate, final thickness equals incoming thickness minus removed thickness, for a simple single-film removal example. A center-thick incoming wafer requires more removal at the center to finish flat. Conversely, excess center removal from a flat starting film makes the final center thin. “Center-fast” refers to removal behavior; it cannot be inferred from a final thickness map alone without knowing the starting profile.

Pressure control has a physical footprint. Zoned CMP heads adjust backside loading to influence the radial removal profile. The retaining ring and pad deformation affect the edge region; slurry delivery, pad condition, and relative motion also matter. Zones are coupled through the mechanical contact, so increasing one pressure is not an isolated correction to one measurement pixel. Establish the tool's response experimentally within its qualified process window. Endpoint control determines when to stop, while profile control determines where removal occurs; both contribute to the result.

Separate wafer-scale uniformity from local planarization. A blanket-film map can look excellent while patterned regions exhibit dishing or erosion. Likewise, a low WIWNU does not prove that the mean thickness meets target: a uniformly over-polished wafer can have a small spread. Report mean, absolute spread, extrema, and map together, then use patterned-feature measurements where local topography is the concern. Within-wafer variation is also distinct from wafer-to-wafer variation across a lot.

Control the measurement before tuning the process. Repeat selected sites to assess measurement repeatability, check the film stack model and coordinate registration, and compare matched pre- and post-polish maps. Track changes against pad and retaining-ring history and incoming-film trends before assigning a cause. Acceptance limits, edge exclusions, and site counts belong to the qualified layer and product, not a universal WIWNU rule. The practical question is whether the remaining film meets its spatial and electrical requirements, not whether one percentage looks small.

Sources: NIST Dataplot, Coefficient of Variation; Applied Materials, CMP; EBARA, Strategic Table of Technological Capabilities; MIT MTL, Wafer Level CMP Modeling; KLA Filmetrics, F50 Automated Film Thickness Mapper. Numerical examples and diagrams are explanatory, not production specifications.