cd-sem metrology semiconductor
CD-SEM measures the critical dimension of patterned features — line width, space width, contact diameter — by scanning a focused electron beam across the wafer surface and imaging the secondary and backscattered electron signal that emerges from resist, hard-mask, or etched structures. Unlike optical metrology, which infers dimension from a model fit to reflected or scattered light, CD-SEM produces a direct image of the feature edge, and the measured dimension comes from applying an edge-detection algorithm to the intensity profile that the electron beam generates as it crosses from one material to another. This directness is CD-SEM's core advantage — it does not require an assumed optical model of the film stack — but it comes with its own set of systematic uncertainties rooted in how electrons interact with matter near a surface, which is why CD-SEM measurements must be calibrated and interpreted with as much care as any model-based optical technique.
**The secondary electron signal rises sharply near a feature edge because the local escape geometry changes, and this "edge effect" — not a step change in material — is what the CD-SEM edge-detection algorithm actually locates.** As the beam scans across a vertical or near-vertical sidewall, more of the generated secondary electrons find an unobstructed path to the detector than they would from a flat top or bottom surface, producing a bright band at the edge that is a geometric artifact of electron escape probability rather than a direct measurement of where the sidewall physically sits. Different edge-detection algorithms — threshold-based (a fixed percentage of the peak signal), maximum-slope, or linear-approximation methods — place the "edge" at different points within this bright band, so the same physical feature can yield different reported CD values depending on which algorithm and which threshold setting the tool uses, which is why CD-SEM tools must be cross-calibrated against a reference method rather than assumed to report an absolute physical dimension.
**Electron beam interaction volume and landing energy set a floor on CD-SEM resolution and introduce a systematic bias that depends on the material being imaged.** The Kanaya-Okayama relation gives an approximate electron penetration depth for a given landing energy and target material,
$$
R \approx \frac{0.0276 \, A \, E^{1.67}}{Z^{0.89} \, \rho},
$$
where $E$ is the landing energy in kiloelectronvolts, $A$ is the atomic weight, $Z$ is the atomic number, $\rho$ is the density, and $R$ is the penetration depth in micrometers; the practical consequence is that lower landing energies, typically 300 to 800 electron volts for CD-SEM as opposed to several kilovolts for general-purpose SEM, keep the interaction volume confined near the surface, improving edge sensitivity and reducing charging in insulating resist films, but very low landing energies can also reduce signal-to-noise ratio and increase sensitivity to surface contamination. Because secondary electron yield depends on atomic number and local surface chemistry, a CD-SEM recipe tuned and calibrated for one film stack (for example, a photoresist line on a bare silicon substrate) can report a biased CD when applied unmodified to a different stack (for example, a metal hard-mask line), so recipe requalification across process layers is standard practice rather than an occasional check.
**Charging of insulating resist and dielectric features under electron bombardment is a persistent CD-SEM artifact because accumulated charge distorts the local electric field near the feature, which in turn distorts the trajectories of emitted secondary electrons and can blur or shift the apparent edge position.** Photoresist, being a poor conductor, is particularly susceptible, and charging effects accumulate with dose — a feature imaged multiple times or scanned too slowly can show measurable CD drift within a single measurement session purely from charge buildup rather than any real change in the feature. Charge-suppression strategies include reducing beam current and dwell time, using landing energies near the crossover point where secondary and incoming electron flux balance, and applying a light conductive coating for particularly sensitive measurements, though the last option is generally reserved for destructive cross-section analysis rather than routine inline monitoring.
| CD-SEM parameter | Typical range | Primary effect | Trade-off |
|---|---|---|---|
| Landing energy | 300-800 eV | Interaction volume, charging | Lower energy reduces charging but can reduce signal-to-noise |
| Beam current | Picoamp to low nanoamp range | Signal strength, dose accumulation | Higher current speeds throughput but increases charging and possible resist shrinkage |
| Edge-detection algorithm | Threshold, max-slope, linear-approximation | Where within the edge signal the CD is reported | Different algorithms give different absolute CD; consistency matters more than any single "correct" choice |
| Frame averaging | Multiple scans per measurement | Noise reduction | More frames improve precision but increase dose and charging risk |
**Photoresist shrinkage under electron beam exposure is a measurement-induced artifact specific to organic resist materials, in which the imaging beam itself measurably reduces the feature's dimension during the act of measuring it.** Electron-beam-induced outgassing and cross-linking or scission of the resist polymer can shrink line width by a few nanometers or more over the course of repeated scans, an effect that scales with beam dose (current times dwell time times number of frames) and depends on resist chemistry. Because this shrinkage happens during measurement, a CD-SEM recipe must balance the number of frames needed for adequate measurement precision against the cumulative dose that drives shrinkage, and production recipes are typically qualified to a fixed frame count and dose budget specifically to keep this artifact reproducible and correctable rather than eliminating it entirely.
```flowchart
Load wafer and navigate to the target measurement site using pattern recognition or stage coordinates → Select the qualified imaging recipe: landing energy, beam current, frame count, for this film stack and layer → Acquire the SEM image at the target field of view and magnification → Apply the qualified edge-detection algorithm to extract line, space, or contact dimensions → Repeat at multiple sites across the wafer per the sampling plan → Compare measured CD distribution against the process specification and control limits → Flag results for CD-SEM to reference-method correlation checks (cross-section SEM, AFM, or CD-AFM) periodically → Monitor for measurement-induced shrinkage by comparing first-frame and last-frame CD on resist layers → Feed CD trend data back into the lithography or etch process control loop → Requalify the recipe when resist chemistry, hard-mask material, or target CD range changes materially
```
**CD-SEM's role in production has shifted from a standalone final-dimension check toward one input in a metrology suite that also includes optical CD (scatterometry) and, less frequently, CD-AFM, because each technique has complementary strengths and blind spots.** Scatterometry-based optical CD measures faster and can extract additional three-dimensional profile information through model fitting, but like all model-based methods it depends on the accuracy of its optical stack model, whereas CD-SEM provides a direct top-down image at the cost of measurement-induced shrinkage risk and lower throughput. CD-AFM offers direct sidewall-angle and profile measurement without electron-beam artifacts but at substantially lower throughput still, making it a reference and calibration technique rather than a high-volume inline monitor. Production metrology strategies typically use CD-SEM for routine inline dimensional control, scatterometry where three-dimensional profile information or higher throughput is needed, and CD-AFM or cross-section SEM as periodic reference checks to catch drift in either fast method's calibration.
Read CD-SEM through an edge-artifact lens: the bright band the algorithm locates is a geometric electron-escape effect, not a direct picture of the physical sidewall, so every reported CD value is only as trustworthy as the calibration linking that algorithm's edge placement to a reference measurement on the same film stack and feature type.