Electroluminescence (EL) imaging is one of the most useful checks in photovoltaic manufacturing and testing. Drive current through a solar cell or module in the dark, and the cell gives off a faint near-infrared glow. Cracks, broken fingers, inactive regions and shunts show up as dark or uneven areas in that glow, often where a visual inspection sees nothing at all.
The catch is that the glow is weak and sits at the edge of what silicon sensors can see. This post is for teams building EL testers, inspection stations and reliability labs. It covers where the signal is, what that means for the camera, and the questions worth settling before you choose one. We use Finger Lakes Instrumentation (FLI) cooled cameras as examples. For the wider picture, see our post on display, solar cell and optics inspection.
Why EL imaging is a low-light problem
In EL testing, a power supply forward-biases the module, often at currents up to its short-circuit current, and a camera images the light the cells emit. The method was brought to the PV community's attention by Fuyuki and co-workers in 2005, and it's now used to find micro-cracks, shunts, poor contacts and other defects in cells and modules.
The emission is dim compared with ordinary scene lighting, so EL is done in a dark room or enclosure, or outdoors after sunset. Stray light has to be kept out with walls, curtains or baffles. Every photon matters, which is why EL cameras are usually cooled: cooling cuts the dark current that otherwise builds up over a long exposure and buries the faint signal.
Where the signal is, and where silicon sees
For crystalline silicon, EL emission peaks near 1150 nm. Silicon sensors are sensitive out to roughly 1100 nm, and their quantum efficiency falls steeply toward that limit. So a silicon camera sees only the short-wavelength tail of the emission.
That isn't a deal-breaker. Silicon CCD and CMOS cameras are widely used for EL because they offer high pixel counts at reasonable cost, and their low response is made up with longer exposures, cooling and good optics. InGaAs cameras cover the whole silicon emission band and are much more sensitive there, so they allow faster capture, but they tend to have far fewer pixels. Many labs and lines choose silicon for resolution and use InGaAs where speed matters most.
For a silicon camera, three things decide how much of that tail you collect:
- Long-wavelength quantum efficiency. Compare published QE curves at 950 to 1100 nm, not the peak QE figure. Back-illuminated sensors on deep-depletion silicon extend the near-infrared response. FLI's MicroLine ML4710DD, for example, uses a back-illuminated sensor for enhanced near-infrared sensitivity. Deep-depletion devices carry higher dark current, so deeper cooling matters more with them.
- The window. Of FLI's window options, the F116 transmits farther into the near-infrared. Ask for the transmission chart, and choose the window before the camera enters the clean-room build.
- The lens and filter. The lens must not carry coatings or filters that block the near-infrared near silicon's band gap. A long-pass filter in the 850 to 950 nm range is commonly used to reject stray visible light.
Resolution: matching pixels to cells and modules
IEC TS 60904-13, the technical specification for EL imaging of PV modules, defines three resolution classes for comparing EL images: 0.2 mm or finer, 2.0 mm or finer, and 5.0 mm or finer. Micro-cracks and broken fingers need the fine end of that range. Whole-module screening can work at the coarse end.
A quick sum shows where a camera lands. As an example, imaging a 2,000 mm long module across the 9,568-pixel width of an Aurora AR455 puts about 0.21 mm of module on each pixel, before allowing for any margin around the module. Imaging a single 180 mm cell across the sensor's 6,380-pixel height puts about 0.03 mm on each pixel. Run the same numbers for your module size, working distance and the class you need. Our post on pixel size and sampling covers field of view and image circle in more detail.
More pixels mean larger files and longer processing, and each pixel collects less light. The specification places the camera's optical axis perpendicular to the module and as close as practical, which keeps the image sharp and the signal high.
Calibrate before you compare
Labs often go beyond spotting cracks: they compare intensity across cells, track a module before and after stress testing, and estimate local voltage differences from intensity ratios. That depends on a well-behaved camera.
- Linearity. The specification warns that a non-linear camera response gives wrong results when intensities are compared. Scientific cameras are generally linear or corrected. Check it on your own bench, as we describe in our post on acceptance testing gain, noise and linearity.
- Dark and background frames. Take a frame with the supply off and subtract it. That removes the camera's offset and dark signal along with any light leaking into the enclosure. FLI Aurora cameras aren't offered with a mechanical shutter, so dark frames come from a filter wheel's blocked position, such as the CFW8-8RND's blank, or from the dark enclosure itself.
- Flat field. Lens vignetting falls off toward the corners and can look like a weak cell. Correct for it before you compare cells across the image.
- Two bias points. One image at one current can't always tell you the cause of a dark area. Images at two bias currents, combined with photoluminescence or thermography, help separate cracks from resistive losses.
Photoluminescence (PL) imaging uses the same camera physics, but excites the cell with light instead of current, so it works on wafers and cells before they're wired. It needs strong filtering to keep excitation light off the sensor, and the same near-infrared sensitivity questions apply.
Fitting the camera into a tester
On a production line, the camera is one part of a timed sequence: set the current, wait for it to settle, expose, read out and move the next module in. Computer control of both the supply and the camera, so captures follow pre-programmed currents, saves time and improves consistency.
FLI Aurora cameras have external trigger in and out as standard, so a tester's controller can start an exposure when the current is stable. For instruments with tight space, Revelation cameras use the same Sony sensors as Aurora in a thin profile, and every camera includes our software development kit (SDK). If the tester holds the camera in a closed cabinet, our post on air versus liquid cooling covers how to remove the heat.
Some EL programs also look beyond silicon's range. FLI SWIR cameras are coming soon.
Questions to settle before you choose
- What's the smallest defect you need to see, and which resolution class does that put you in?
- Will you image single cells, full modules, or both on one station?
- What exposure time can your cycle afford, and does that point to silicon or InGaAs?
- What is the camera's QE at 1000 to 1100 nm, and which window suits it?
- Will you compare intensities quantitatively, and how will you check linearity and take dark frames?
- How will the controller trigger the camera and the power supply together?
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. If you're specifying a camera for an EL tester, a PL station or a reliability lab, tell our sales team your cell or module size, resolution class and cycle time. We'll suggest a model and window, and walk through how it fits your setup. Request a quote to get started.
