A SWIR sorting line needs enough photons in the 970nm to 1700nm band to push a signal through a product stream at line rate. Both the 3535 and the 5050 package can do that, but they reach it from different directions: the 3535 from higher radiant power per die with tighter thermal control, the 5050 from more emitting area and a lower cost per watt of optic.
Start from the camera, not the LED. An InGaAs sensor needs a defined irradiance at the product surface with a falloff the throw distance and beam angle can predict. If the line has 1200mm of throw and a 60 degree optic, the package choice follows from the radiant power the optic can gather.
Where Each Package Fits
| Parameter | 3535 SWIR | 5050 SWIR |
|---|---|---|
| Emitting area | 3.5 x 3.5mm | 5.0 x 5.0mm |
| Typical radiant power | 1.5W to 3.5W per emitter | 2W to 5W per emitter |
| Thermal path | Ceramic base, low Rth | PCB base, higher Rth |
| Best use | Long throw, tight bin | Wide bar, short throw |
| Driver current | 350mA to 1000mA | 500mA to 1500mA |
The 3535 wins where throw is long and every kelvin of junction temperature shortens life. Its ceramic substrate carries heat to the pad with a lower thermal resistance, which keeps the wavelength stable as the bar warms. The 5050 covers more area per board position, so a row of them produces an even band over a wide web at lower board count.
Binning Is the Real Cost Driver
SWIR emitters ship in radiant power bins measured in milliwatts at a stated current, usually 100mA or 350mA. A spec sheet peak figure is a bin ceiling, not a deliverable. Two bars built from the same bin show the same output on the day they are commissioned and then drift apart as the bins age.
Specify the bin you will accept, and ask the manufacturer for the bin distribution across a production run. Sorting lines that blend bins to hit a target often see a visible brightness band across the bar, which shows up as a false edge on the camera. Consistency beats peak number for a line that runs ten years.
Heat and Junction Temperature
Each emitter drops a forward voltage and converts the rest to heat. At 1000mA a 3535 running at 3.2V dissipates around 3W, and the junction needs to stay under the rating set for the bin. Below roughly 85C the wavelength drift over a shift is small enough to stay inside a machine vision band; above it, the peak shifts and the classifier thresholds need re-tuning.
Mount the bar on the aluminium spine of the frame, not on a plastic housing. Spread the emitters across two board rows rather than one dense row, and leave the bar access to the product air stream. Sorting lines are dusty and wet; a bar that runs hot behind a sealed cover loses output faster than one that runs cool in open air.
Line Rate and Camera Matching
The usable light depends on the exposure time the camera can set. A 5000mm/s line with a 1ms exposure gives the sensor 5mm of product to integrate over, so the bar needs a sharp edge rather than a broad wash. Do not fit a bar with a 120 degree optic to a tight line: the spread lands on the housing behind the product and the sensor sees little of it.
Before ordering, run the camera on the actual product with the actual bar at the throw distance. Banding, specular highlights on wet surfaces and the reflectance of the product all change the exposure the line needs. A 20 minute bench test with one 3535 bar and one 5050 bar saves a rework on the gantry.
Key Takeaways
- Choose the package from the camera exposure and throw distance, not from the peak radiant power.
- Specify the radiant power bin and ask for the run distribution, because bins drift over ten years.
- Keep the junction low with a metal spine and two-row spacing; heat moves the wavelength.
- Bench test both packages on the real product before committing the gantry build.
- what FWHM SWIR LEDs need for food sorting applications
- more package and spectral bandwidth notes in LED Diode Q&A
