Cameras for SiC Wafer Photoluminescence Defect Imaging

November 19, 2026

Silicon carbide (SiC) power devices depend on the quality of the epitaxial layer they are built in. Some crystal defects in that layer, such as basal plane dislocations (BPDs) and stacking faults, can hurt device yield and long-term reliability, so wafer makers, epi suppliers and device fabs screen for them before devices are built.

Ultraviolet photoluminescence (UV-PL) imaging is one of the main ways to do that screening. Light above the SiC bandgap excites the wafer, and different defects glow in different wavelength bands. A cooled camera behind a set of emission filters records those bands, and software turns the images into a defect map. The method is non-destructive and non-contact, which is why it is used to qualify epitaxial layers before the costly steps of device fabrication.

This post covers what a PL defect-imaging tool asks of its camera, using Finger Lakes Instrumentation (FLI) cameras as examples. It is written for builders of SiC inspection and metrology tools and for the process and reliability labs that use them. For inspection cameras in general, see our post on cooled cameras for display, solar and optics inspection.

How PL defect imaging works

4H-SiC has a bandgap of about 3.26 eV, so excitation has to be shorter than about 380 nm. Published systems use UV lasers or filtered lamps in roughly the 270 to 365 nm range.

The useful information is in where the light comes out:

  • Stacking faults emit in the blue, roughly 420 to 500 nm, and different fault types peak at different wavelengths in that band. One study imaged them through narrow band-pass filters (under 10 nm wide) to tell the types apart (AIP Advances, 2015). Bar-shaped faults have been reported with peaks near 420 and 450 nm (Journal of Applied Physics, 2023).
  • Basal plane dislocations give a broad near-infrared emission, most easily picked out between about 738 and 870 nm. Threading dislocations emit strongly above 900 nm (MRS Proceedings, vol. 911). Studies often image dislocations through a long-pass filter above 700 nm.

So one tool may need to image the same field in a blue band, several narrow visible bands and a near-infrared band. Each band asks something different of the detector.

Spectral response from blue to near infrared

Sensors with high blue response make short work of the stacking-fault bands. Back-illuminated sensors collect more of the light that reaches them, and the Sony back-illuminated CMOS sensors in FLI Aurora and Revelation cameras pair that with very low read noise.

The dislocation bands are harder. Silicon sensors lose response as wavelength rises into the near infrared, and by 900 nm much of the light passes through a thin sensor without being absorbed. Compare published quantum efficiency curves across 700 to 950 nm, not the peak figure. Deep-depletion CCDs use thicker silicon, which improves near-infrared response; FLI's MicroLine ML4710DD is one example. Deep-depletion devices carry higher dark current, so they need deeper cooling or shorter exposures. Our post on cooled CCD cameras for Raman and NIR spectroscopy covers that trade-off.

The camera window matters too. Window materials differ in how far they transmit into the near UV and the near infrared, and the window is chosen before the camera enters our clean-room build, so settle it early.

Rejecting the excitation light

The excitation is far brighter than the defect emission. Any UV that leaks through the emission filter, or any glow from filters and optics under UV, adds a background that can swamp a faint stacking fault. Plan for strong blocking at the excitation wavelength in every emission filter, and check each band with the excitation on and no wafer present.

Signal, exposure and noise

Narrow filters throw away most of the emission, and some defects are faint. That pushes exposures longer in the narrow bands, which makes cooling and dark current part of the noise budget. Our post on long-exposure dark current shows how to work it out.

On the Sony CMOS sensors in Aurora and Revelation cameras, binning is done digitally rather than on the sensor, so it does not lower read noise the way on-chip CCD binning does. If a band is short of photons, a lower magnification or a wider filter will usually help more than binning.

Field of view, pixel size and wafer coverage

Pixel size at the wafer is the pixel size divided by the magnification. Aurora and Revelation cameras use 3.76 µm pixels, so at 2x each pixel covers 1.88 µm of the wafer.

Format sets how many fields cover a wafer. As an illustration, a 150 mm wafer is about 17,700 mm². At 2x, a 9568 × 6380 sensor (AR455, 36 × 24 mm) sees an 18 × 12 mm field, so a full map takes at least about 82 tiles before overlap. A 6244 × 4168 sensor (RV571, 23.5 × 15.7 mm) sees about 11.8 × 7.9 mm and needs about 192. Each tile is imaged in every band, so tile count multiplies quickly. Our post on pixel size and sampling covers the optics side.

Band sequencing, darks and flats

Most tools image each field through several filters, so filter changes add up across a wafer. The FLI High Speed Filter Wheel moves to the adjacent filter in 23 ms on the 6-position model and 30 ms on the 10-position model. See our post on fast filter switching.

Aurora and Revelation cameras have no mechanical shutter, so take dark frames with the excitation off or through a blocked filter position. Take flat fields per band too, because illumination and filter transmission vary across the field, and a defect map should not mistake shading for a defect.

For mapping, hold the stage still during each exposure. These sensors support global reset with rolling readout, so each row's exposure ends at a slightly different time; with the stage settled and the light steady, that does not affect the image.

Keeping tools consistent

Defect maps get compared across tools, shifts and sites, so every camera should behave the same way. Keep gain, temperature and exposure settings fixed per recipe, and measure each incoming camera's gain, noise and linearity. Data-sheet values are typical, and FLI provides a production test summary for each camera by serial number. Our post on camera acceptance testing walks through the checks.

Fitting the camera into the tool

Revelation puts the same Sony sensors as Aurora into a thin, fin-less body that is shorter front to back. It can be liquid-cooled or mounted to your housing as a heatsink, which helps in enclosed tools where heat near the optics matters. See our post on air vs liquid cooling.

External trigger in and out are standard, so the camera can follow the stage controller. Every Revelation camera includes our software development kit and support for integration into your OEM design. Drivers and manuals are on our support page.

Questions to settle before you choose

  1. Which defects matter, and which emission bands will you image?
  2. How much near-infrared response do your dislocation bands need?
  3. How well do your filters block the excitation wavelength?
  4. What pixel size at the wafer do your smallest defects need?
  5. How many tiles and bands per wafer, and what wafer throughput is the target?
  6. How much space, airflow or coolant is available inside the tool?

Talk to our sales team

FLI cameras are Made in USA, and our sensor chambers are assembled in our 1,000+ sq ft ISO Class 7 clean room. See how on our in-house manufacturing page.

Send our sales team your excitation source, emission bands, optics and throughput target, and we'll help you choose a camera and filter wheel for your tool. Request a quote or email sales@flicamera.com.