The Defect Your Darkfield Tool Will Never See

Most people think wafer inspection is one tool. It's actually two - and the difference is pure optics.

Brightfield​ collects reflected​ light. Bright background, dark defects. It sees what's wrong with the pattern: opens, shorts, bridge defects, CD variation, missing features, resist residue, dishing, overlay errors.

Darkfield​ collects scattered​ light. Dark background, bright defects. It sees what's wrong with the surface: particles, micro-scratches, pits, crystal-originated defects, edge roughness.

One word explains everything: reflection vs. scattering. That's why they're complements, not competitors.

How fabs actually sequence them:

Bare / unpatterned wafers​ → darkfield (KLA Surfscan class, down to ~7nm). No pattern to compare against, so scatter wins on speed and sensitivity.

Post-litho / post-etch patterned layers​ → brightfield. Catches what scatter simply can't see.

Post-CMP​ → darkfield for residue and scratches, brightfield to confirm dishing and erosion.

Advanced nodes​ → darkfield scans fast, brightfield re-inspects the suspects, e-beam confirms what the defect actually is.

The physics worth remembering: darkfield's oblique angle can't reach the bottom of a high-aspect-ratio contact hole - geometrically shadowed. That's brightfield territory, and its coaxial path even brings back thin-film interference phase data. Flip it around: an atomically thin residue has no steep edge to scatter from, so darkfield misses it entirely - but it shifts the interference color, and brightfield catches it as a gray-level drift.

The honest limit: as features shrink, optics approach the diffraction wall. Brightfield hits resolution limits; darkfield drowns in pattern-generated noise. So the industry stopped asking optics to do everything. Optics find the suspects. E-beam names them.

Brightfield reads the pattern. Darkfield reads the surface. Know which question you're asking.