Critical Dimension Control
# Critical Dimension Control
Critical dimension (CD) control is the measurement-and-control discipline used to keep patterned feature dimensions within functional limits from mask and exposure through resist development and final pattern transfer. A CD may be a line width, space, hole diameter, trench opening, gate length, fin or sheet dimension, or another geometry defined at a specific height and measurement convention.
“CD variation” is not one error term. Engineers separate mean bias, within-wafer and wafer-to-wafer nonuniformity, field and layout dependence, local stochastic variation, edge/width roughness, vertical-profile variation, pattern-transfer bias, and measurement uncertainty. Combining all of them into a single standard deviation can hide the mechanism and cause the wrong process correction.
## The CD budget
A useful CD budget keeps deterministic and random components visible. Depending on the application, it may include reticle or template dimension, scanner dose and focus, illumination and aberrations, overlay interactions, resist thickness and chemistry, bake and development, etch bias, loading, chamber state, layout context, stochastic printing, sampling, and metrology noise.
Variance terms may sometimes be combined in quadrature, but only when independence and stationarity are justified. Correlated terms need covariance, and deterministic signatures should be modeled rather than treated as random variance. The budget must also specify whether it describes one wafer, a lot, tool matching, a time horizon, or long-term manufacturing capability.
| Metric | Meaning | Common misuse |
|---|---|---|
| Mean CD / bias | Offset from target for a declared feature and location | Treating global mean as proof of uniformity |
| CDU | Distribution across declared wafer/field/layout population | Mixing spatial signatures and local roughness |
| LCDU | Local feature-to-feature CD distribution under defined sampling | Comparing unlike pitches, counts, or algorithms |
| LER/LWR | Edge or width fluctuation along features | Treating it as wafer-level CDU |
| Profile/CD by height | Sidewall shape, taper, footing, top loss | Assuming top-down CD equals transfer geometry |
| Metrology uncertainty | Repeatability, reproducibility, model and sampling contributions | Subtracting noise without a validated model |
## Patterning and transfer controls
Exposure dose primarily moves resist CD for a given focus and process, but the response is nonlinear and layout-dependent. Focus changes image contrast and profile as well as CD. Resist thickness, PAB, PEB, develop, rinse, and queue state change the chemical and mechanical conversion of the aerial image. Pattern transfer then adds its own bias through mask erosion, sidewall passivation, ion angular distribution, loading, microloading, charging, and chamber condition.
Feed-forward and feedback must use the correct actuator. Reticle or OPC corrections address repeatable layout signatures. Scanner dose/focus control addresses qualified imaging signatures. Track adjustments address demonstrated resist-process shifts. Etch controls address transfer bias and profile. No actuator should be moved solely because it correlates with the output; the causal response, interaction terms, latency, limits, and risk of compensating one module for another must be established.
## Metrology and uncertainty
CD-SEM, optical scatterometry, critical-dimension AFM, cross-sectional SEM/TEM, profilometry, and electrical structures observe different properties and volumes. Top-down CD-SEM provides local edge information but is sensitive to focus, charging, contamination, edge-detection model, pixel size, sampling, and beam interaction. Scatterometry is fast and profile-sensitive through a model but can be non-unique and depends on optical constants and prior assumptions. Cross-sections reveal profile directly at destructive, sparse locations.
Every reported CD needs a measurement definition: feature, layer, material boundary, height, algorithm, sampling plan, calibration/reference, uncertainty, and whether values are raw or corrected. Gauge repeatability and reproducibility, tool matching, long-term stability, reference-material behavior, and model uncertainty must be small enough relative to the control decision. A stable metrology number is not necessarily an accurate physical boundary.
## Qualification and production control
Qualification should span dose/focus, resist and track inputs, layout context, wafer position, etch conditions, chamber state, product stack, and relevant time/queue variables. Use patterned monitors covering dense and isolated lines, spaces, holes, tips, corners, and orientations. Compare ADI to after-etch inspection (AEI) so lithographic and transfer contributions can be separated.
Production control should preserve raw spatial coordinates, tool/chamber/module identity, material lot, recipe version, reticle location, design context, and timestamps. Statistical process control detects shifts; run-to-run control may compensate qualified systematic behavior; advanced process control can combine feed-forward and feedback. All require actuator limits, delay handling, fault detection, and safeguards against chasing noise.
The practical signoff question is: Does the final transferred feature meet mean, spatial, local, roughness, profile, defect, and uncertainty limits across the declared manufacturing population—and is each dominant signature controlled at its causal module?