Near infrared wavelength selection for water and silicon inspection

Visible light shows surfaces. The near infrared shows what water and silicon hide: a wet seal goes dark, and a circuit pattern under a wafer appears. Both effects are strong and simple to use — provided the camera is sensitive at the band you pick, which is where most of these projects fail. This guide gives the selection rule per material, the band table that goes with it, a worked example with the arithmetic shown, and a bench test you can run in half an hour with one light and one camera. It is written for packaging, electronics and film engineers choosing among the common near-infrared bands.

The selection rule

Start from the material in front of the lens, then buy light and camera to match:

• Water or moisture anywhere in the optical path: image at 1450 nm. Wet areas absorb and go dark against a dry background.
• Silicon between the lens and the target: image at 1200 nm. Silicon stops light below about 1107 nm and transmits just above it.
• Short-range work on a standard camera: 850 or 940 nm on a monochrome silicon sensor. Thin coloured plastics pass these bands; colour sensors usually carry an IR-cut filter that blocks them.

One line governs all three cases: the camera sensor must be rated beyond the band you choose. A long-wave LED on a camera that stops at 1000 nm produces a black frame, and no amount of exposure fixes it.

Water: the dark band

Liquid water absorbs in bands near 760, 970, 1190, 1450 and 1950 nm (electromagnetic absorption by water). The strongest short-wave band in that list is the working choice for machine vision: strong enough that a damp patch reads as a dark object, yet covered by affordable SWIR LED sources.

Applications follow directly. Droplets and damp zones on film, paper or non-woven material image dark on a light background. Product trapped in a seal darkens the seal area — the NIR lighting for fill level inspection through packaging guide shows that layout through thin films.

Path length picks the band. The 1950 nm band absorbs harder, so it suits very thin layers where contrast at the water band is too gentle. Millimetre-scale water paths saturate the working band completely; to see features through a wet volume, move to the quieter windows near 1300 or 1550 nm, which sit between absorption bands.

Silicon: clear above 1107 nm

Silicon's band gap is about 1.12 eV, and a photon must carry more energy than that to be absorbed. Converting photon energy to wavelength:

Below roughly 1107 nm a silicon wafer is opaque however thin it is. Above it, transmission climbs fast, which is why 1200 nm is the standard band for looking through silicon: transmitted light for pattern observation, specular reflection for surface films.

The arithmetic fixes the margin. This band sits 93 nm above the edge (1200 − 1107 = 93), an 8.4 percent margin (93 / 1107 × 100), which is why pattern inspection through a wafer works in practice here. Stay near it rather than drifting to 1650 nm: resolution scales with wavelength, so a lens at NA 0.1 resolves about 7 µm at this band against about 3.4 µm at 550 nm (0.61 × λ / NA).

The camera cuts off before the light does

The band gap that clears silicon also ends silicon sensor sensitivity at that same edge. A standard silicon sensor responds to roughly 1000-1100 nm and nothing beyond. Bands above that edge need InGaAs sensors, whose 0.73 eV gap puts the cutoff near 1700 nm (1240 / 0.73 ≈ 1699 nm).

Spectral cameras make the limit explicit. The LS-HSC-0004 is a 5 MP, C-mount hyperspectral camera covering 400-1000 nm with 1200 bands over USB3.

The 970 nm water band falls inside that window, so moisture trials are possible on it today. The bands featured in this guide sit above the cutoff and need SWIR-range hardware: browse the hyperspectral cameras category, and for bands past 1000 nm that are not listed, tell us the material and target through the contact page. How hundreds of bands turn into a classification decision is the subject of the hyperspectral imaging guide.

Band table: what each band buys you

LED sources follow the same ladder: common SWIR emitters come in bands from 1050 to 1650 nm, alongside the everyday 850 and 940 nm. The special lights category carries the infrared models we list, filterable on the lights finder; the custom light series, 41 models deep, takes bands and shapes that are not standard.

Worked example: wet PE seal on a food tray

Task: detect product moisture in the seal area under a clear PE top web.

1. Band. The target is water, so the water band. PE transmits across the near infrared, so the band reaches the seal with little loss in the film.
2. Camera first. A station camera rated to 1000 nm falls 450 nm short (1450 − 1000 = 450). An InGaAs camera with a 1699 nm cutoff leaves about 250 nm of headroom (1699 minus the band = 249). Settle the camera decision before the light.
3. Fallback on the existing 1000 nm camera. The 970 nm water band fits inside its response. Expect grey-level contrast rather than black-and-white, and recover it with a dark background behind the seal, a strobed light for peak power, and the longest exposure the line rate allows — the strobe machine vision light guide gives the overdrive and duty-cycle numbers.
4. Acceptance number. Wet and dry seal patches should separate by at least 2:1 in mean grey value before you trust the band, and 4:1 or better for a threshold that holds across product variation. Under 2:1, change the band or the geometry before anything else.

The 30-minute bench test

1. Mount one band LED at the planned working distance over a dry sample.
2. Set a fixed exposure that puts the dry area at about 70 percent of full well.
3. Wet a second sample or patch; record the mean grey value of a 50 × 50 pixel region on each.
4. Divide the dry mean by the wet mean. Under 2:1, change the band or the geometry before buying anything: try 1950 nm on thinner films, or 1300 / 1550 nm on thicker liquid paths.
5. Repeat at the fallback band so you know what the lower-spec camera option really sees.

The habit matters more than any single band. This contrast-first method is the same one behind the reflected UV surface inspection: test contrast before buying guide, and it transfers unchanged.

When the obvious answer fails

• Carbon-black plastic. Black pigment absorbs across the whole near infrared, so no wavelength recovers a target behind it. Move to a different sensing method.
• Moisture sitting on the pattern you want. Surface water absorbs the water band before the light reaches the silicon below. Inspect the pattern at the silicon band and the moisture in a second, sequential frame.
• A 1050 nm trial that returns black. Most silicon sensors end near 1100 nm with the response already collapsed; a black image usually means sensor cutoff, not a dead LED.

Checks before you order

• Sensor response extends at least 100 nm past the working band, not just up to it.
• The light exists in exactly that band, at a shape that fits the station.
• The lens transmits at the band with refocus planned; the SWIR lens and UV lens basics guide lists what changes outside the visible range.
• The controller strobe output is fast enough for the pulse widths these low-power bands need.
• The bench contrast test passed on your actual product, at the actual band.

Next step

Browse the categories or start the configurator at the lighting step and enter band, field of view and working distance:

For the band-overview decision one level above this one, read SWIR vs NIR vs visible imaging: which band to inspect with. The surrounding guides on wavelength choice, sequencing and moisture imaging sit under lighting guides.