SiC Wafer Defect Inspection: When a Zoom Lens Is Enough and When You Need an APO Objective With Laser Autofocus

Quick answer: A missing 1 µm defect on a SiC wafer is usually a resolution problem, not a pixel problem. At its 7.5× end, the POMEAS PMS-LZL-12104-D1 12.5X motorized zoom reaches a sampling rate of 0.46 µm/pixel, yet its datasheet optical resolution at that same magnification is 3.05 µm (N.A. 0.11). Sampling is not resolution. To actually resolve 1 µm you need an APO microscope objective — the PMS-MPO10 is specified at 1 µm — and you must then deal with depth of field collapsing from 50 µm to 3.5 µm, which is what laser autofocus (PMS-FocusX, ≤ 0.4 s cycle) exists to solve.

What has to be inspected on a SiC wafer, and which optical modality sees it

Silicon carbide device yield is largely a function of substrate and epi-layer defect control. The industry separates SiC defects into two families:

• Crystal defects — basal plane dislocations (BPD), threading screw dislocations (TSD), threading edge dislocations (TED), stacking faults (SF), micropipes, grain boundaries
• Surface / epi defects — scratches, particles, edge chipping, pits and bumps, carrot defects, triangle defects, downfalls, step bunching

No single optical modality captures both families. That is why production SiC final-inspection tools run brightfield, darkfield and photoluminescence (PL) channels in parallel.

Brightfield — particles, scratches, contamination

Brightfield uses coaxial (episcopic) epi-illumination: light goes down through the objective and reflects straight back. A polished, flat area returns strong specular reflection and images bright; a scratch, particle or stain breaks that specular return and appears dark. For polished SiC, coaxial illumination is not a nicety — it is the standard way to suppress specular glare off a mirror-like surface.

Darkfield — sub-micron particles and shallow scratches

Darkfield illuminates at a very low grazing angle. Flat surface reflects light away from the objective and the background goes black; only a defect with topography scatters light back into the lens and appears as a bright point on black. Sensitivity to sub-micron particle scattering is materially higher than brightfield, at the cost of a dedicated grazing illuminator, weak signal, and high sensitivity to vibration and cleanliness.

PL (photoluminescence) — crystal defects

PL excites the SiC with near-UV light. Defective regions emit at different wavelengths and intensities than sound material, which makes dislocations, stacking faults and micropipes visible — defects that neither the eye nor brightfield can see. The method is standardised in China as GB/T 43493.3-2023 (Non-destructive detection and identification criteria for defects in silicon carbide homoepitaxial wafers for power devices — Part 3: Photoluminescence detection method).

Capability boundary, stated plainly. POMEAS supplies optical imaging components. The UV excitation source, dichroic mirror and emission filters that a PL channel requires are not POMEAS products, and objective transmission at a 275 nm excitation band must be confirmed separately. POMEAS does not supply a complete SiC wafer inspection tool, nor does it supply defect classification algorithms.

Where 0.5 µm/pixel actually comes from

There is exactly one formula:

object-space sampling (µm/pixel) = camera pixel pitch (µm) ÷ optical magnification

Using a 2/3″ 5 MP camera (2448 × 2048, 3.45 µm pixels) on the POMEAS PMS-LZL-12104-D1 12.5X motorized zoom (0.58×–7.5×):

Inverting it: reaching 0.5 µm/pixel requires 3.45 ÷ 0.5 ≈ 6.9×. That sits inside the 0.58×–7.5× range, so the claim "a zoom lens can reach 0.5 µm/pixel" is true at the sampling level.

The problem is in the next section.

Can 1 µm defects actually be detected? Sampling is not resolution

This is the load-bearing part of the article.

The PMS-LZL-12104-D1 datasheet lists optical performance per magnification. Placed side by side, the contradiction is obvious:

At the 7.5× stop:

• sampling is 0.46 µm/pixel — a 1 µm feature covers roughly 2.2 pixels, which looks adequate;
• but optical resolution is 3.05 µm (N.A. 0.11) — three times coarser than the defect.

A 1 µm defect is already smeared into a ~3 µm blur spot by diffraction before it ever reaches the sensor, so its contrast collapses and it cannot be detected reliably. Adding pixels does not recover information the optics discarded.

Reliable detection of a 1 µm defect requires two conditions at once:

1. Optical resolution ≤ defect size — which means an objective with sufficient N.A.
2. Enough pixels across the feature — in practice ≥ 3 pixels

Zoom lens vs objective + laser autofocus: how to choose

The division of labour in one line: the zoom lens finds and reviews; the objective plus laser autofocus decides whether the 1 µm call is pass or fail. They are stages in one chain, not substitutes.

Depth of field collapse: why laser autofocus becomes mandatory

Moving from a zoom to a high-magnification objective shows up almost entirely as depth of field:

12.5X zoom @ 7.5×   DOF 50 µm
PMS-MPO10           DOF 3.5 µm
PMS-MPO20           DOF 1.6 µm
PMS-MPO50           DOF 0.9 µm

Sub-micron depth of field means wafer bow, stage travel error and ordinary floor vibration are each enough to throw the image out of focus. At that point focusing capability, not resolution, decides whether the inspection runs at all.

POMEAS PMS-FocusX laser autofocus microscopy system:

A second, practical benefit: real-time surface tracking during scanning. This matters for SiC samples that have been defect-selective etched and therefore present real topography — an application where laser focus is generally preferred over image-contrast autofocus.

What POMEAS supplies — and what it does not

Supplied (catalogue items, published specifications):

Not supplied (stated explicitly):

• ❌ A complete SiC wafer inspection tool — POMEAS supplies optical components, not a finished system
• ❌ PL inspection modules (UV excitation source, dichroic, emission filters)
• ❌ Defect classification algorithms and defect-size calibration standards
• ⚠️ A wafer-specific darkfield illumination module — no catalogue item; engineering confirmation required

Real project evidence: semiconductor grinding equipment

Stated honestly up front: POMEAS has no published SiC wafer inspection case study. The available semiconductor-sector evidence is this project.

For a semiconductor grinding equipment manufacturer, POMEAS configured a system that had to satisfy three simultaneous demands — large-field locating, high-magnification local review, and observation inside a narrow bore:

Key mechanical datums: lens WD 77.8 ± 2 mm, YWD 57 ± 2 mm, XWD 20 ± 2 mm, ring-light standoff approximately 10 ± 5 mm. The inner-bore station carried a ±1 mm height variation, handled by a three-tier strategy: cover it with depth of field where possible, store a separate focus position per station, and add motorized focus / Z-axis / autofocus where neither suffices.

What this case proves: multi-magnification observation, folded optical paths for confined spaces, and height-variation compensation on semiconductor equipment — POMEAS has delivered these. What it does not prove: 1 µm defect detection on SiC wafers. That is a different claim, and without test data it is not made.

Six questions that can only be answered by sample testing

None of these can be settled from a datasheet:

1. Whether 1 µm defects are reliably detectable on your specific wafer — depends on defect type, surface roughness and reflectivity, not on a single resolution number
2. SiC substrate transmission — semi-insulating material can be semi-transparent; the coaxial / ring / transmitted illumination recipe has to be tested
3. Whether darkfield is necessary at all — if sub-micron particles are the target, brightfield may simply never see them; run a modality comparison sample first
4. The PL channel — outside POMEAS scope; validate jointly with your PL module supplier and confirm objective UV transmission
5. Depth of field vs wafer bow — if bow exceeds the objective's DOF, laser autofocus is mandatory, not optional
6. Cycle time — zoom + turret switch + focus + exposure must be timed on the real stage

Seven things to have ready before you request a quote

1. Defect family you must catch (surface particles/scratches vs crystal dislocations/stacking faults/micropipes)
2. Smallest defect size that must be detected (µm)
3. Wafer size (4/6/8 inch) and the area to be inspected
4. Available working distance and mounting envelope
5. Existing camera model, sensor format and pixel pitch
6. Takt time available per wafer or per station
7. Whether brightfield, darkfield or PL channels are required

Frequently asked questions

Ready to specify a configuration?

Send us the defect family, the smallest defect size you must detect, wafer size, working distance and takt time, and the POMEAS engineering team will respond against the seven-point list above. Wherever 1 µm detection or a PL channel is involved, we will say plainly that sample testing is required rather than promise a number.

Further reading:

• Motorized Zoom + APO for Semiconductor Wafer Inspection — how to plan the magnification combination
• Autofocus vs Motorized Zoom Lens: What Each Solves
• Machine Vision Zoom Lenses for Electronics & Semiconductor Inspection

All product figures are taken from POMEAS datasheets and pomeas-vision.com product pages; derived values are shown with their formula; SiC defect taxonomy follows published industry method and GB/T 43493.3-2023. Last updated 2026-11-03.