TEC sensor cooling in SWIR cameras: when dark noise halves

An InGaAs SWIR sensor sees 900-1700 nm light because its bandgap is small, and the same small bandgap lets heat push electrons into the pixel wells with no light at all. Every pixel collects a dark charge that grows with sensor temperature and exposure time, and it lands on your image as noise. ThermoElectric Cooling (TEC) holds the sensor at one fixed, low temperature and halves that dark current for every 6-8 °C of cooling. This explainer gives the scaling law, one worked dark-charge calculation you can repeat for your own pixel and exposure, and a decision rule for when the cooled camera is the right order.

The decision, up front

Order the TEC-cooled SWIR camera when any one of these is true: your exposure runs past about 10 ms with the sensor near 45 °C, which is a 40 °C cabinet plus normal self-heating; the feature you must measure sits within a few dozen electrons of the noise floor; or a gain or radiometric calibration has to hold for months without a re-take.

Stay fanless when exposures sit in single-digit milliseconds under bright SWIR illumination, the enclosure stays near 25 °C, and a fan is the part you least want in a dusty hall. How much an uncooled sensor drifts over a shift, and what that costs a calibration, is the subject of uncooled SWIR camera stability for inspection systems; whether SWIR is the right band for your material at all is settled in SWIR vs NIR vs visible imaging.

What a TEC does under the sensor

A thermoelectric cooler is a solid-state Peltier module in the sensor's heat path. Drive current through it and it pumps heat from the sensor to a heatsink on the camera body, where fins and usually a small fan release it to the air. Reverse the current and the same module heats instead, which is the point: the control loop holds one setpoint from a cold startup to a hot afternoon rather than cooling as hard as it can and floating with the room.

The better camera designs keep the FPGA and memory on a separate heat path, so electronics heat never works its way back across the cold side the TEC is defending. The drive side, closed-loop PWM on the module current from a temperature sensor, is the same technique walked through in PWM temperature control for TEC modules; inside a camera it is integrated, and the only decision you make is the setpoint.

The scaling law: half the dark current every 6-8 °C

Dark current in InGaAs rises exponentially with temperature, and near room temperature the curve is close to a clean rule: 6-8 °C of sensor cooling halves it. Published silicon CMOS sensors behave the same way but more slowly, nearer 8-10 °C per doubling, which is why cooling is a SWIR decision first. One published 1280 × 1024 InGaAs array measured 2.25 nA/cm² at 25 °C, and one camera maker's bench test counted more than 80 times as many pixels above a fixed noise threshold with the cooler off than with it on.

The one caveat worth carrying: the 6-8 °C constant drifts over a wide span, so treat a 40 °C extrapolation as good to about a factor of two.

The next section runs these formulas for a 15 μm × 15 μm pixel, the size this SWIR camera prints:

Worked example: 15 μm pixel, 50 ms exposure, 45 °C versus 15 °C

Inputs, each with its source:

• Pixel: 15 μm × 15 μm, the size printed for the LS-IRC-0029.
• Dark current density: 2.25 nA/cm² at 25 °C, from a published InGaAs array measurement.
• Halving interval: 7 °C, the middle of the 6-8 °C range.
• Exposure: 50 ms, a typical need when the feature is dim and the illumination modest.
• Sensor temperatures: 45 °C for the uncooled camera in a 40 °C cabinet, 15 °C held by a TEC.
• Design ceiling: 2,500 electrons of dark charge per pixel, which is 50 e⁻ rms of dark shot noise.

The arithmetic:

1. Pixel area: 15 × 15 = 225 μm². One cm² is 10⁸ μm², so the pixel is 225 / 10⁸ = 2.25 × 10⁻⁶ cm².
2. Dark current per pixel at 25 °C: 2.25 × 10⁻⁹ A/cm² × 2.25 × 10⁻⁶ cm² = 5.06 × 10⁻¹⁵ A.
3. In electrons: 5.06 × 10⁻¹⁵ A ÷ 1.602 × 10⁻¹⁹ C = 31,600 electrons/s at 25 °C.
4. Warm sensor at 45 °C: 20 °C above the reference is 20 / 7 = 2.86 doublings, a factor of 2^2.86 = 7.2. Dark rate: 31,600 × 7.2 = 229,000 electrons/s. At 50 ms that is 11,400 electrons, with shot noise sqrt(11,400) = 107 e⁻.
5. TEC-stabilized at 15 °C: 10 °C below the reference is 10 / 7 = 1.43 halvings, a factor of 2.7 down. Dark rate: 31,600 / 2.7 = 11,700 electrons/s. At 50 ms that is 585 electrons, with shot noise sqrt(585) = 24 e⁻.
6. Exposure budget at the 2,500-electron ceiling: 2,500 / 229,000 = 11 ms warm; 2,500 / 11,700 = 214 ms cooled.

Decision: at 50 ms the warm sensor's dark noise of 107 e⁻ is more than double the ceiling, so that camera cannot make this inspection no matter what the software does downstream. The stabilized camera sits at 24 e⁻ and could run four times longer before hitting the same ceiling. In this warm cabinet the TEC is not an upgrade; it is the only camera that meets the floor.

Choosing the camera and the lens

Start from the exposure, then the room. Short exposures in a cool, stable room let you stay fanless, which removes the one moving part and suits dusty lines. For that case the pick is a fanless SWIR camera such as this one:

The MV-CI013-GS-NN is a 1.3 MP SWIR camera on a 1/2" IMX990 sensor, C-mount, GigE, and fanless by design. When your numbers look like the worked example instead, with a long exposure, a warm cabinet and a tight floor, specify a camera that states a stabilized sensor temperature at or below 15 °C, and ask for the time from power-up to that spec before the first calibrated frame counts.

The lens is part of the noise budget too. A lens with poor transmission at 900-1700 nm forces the exposure up, which is exactly the regime where cooling pays; the coating side of that trade is covered in lens coating and spectral transmission, and the SWIR and thermal lenses category is where the 900-1700 nm coated options live. The basics of what changes outside visible light are in SWIR lens and UV lens basics.

Checks before ordering

• Write down the enclosure's temperature at the hottest hour of the hottest shift, add about 5 °C of camera self-heating, and use that as the uncooled sensor temperature.
• Divide your longest usable exposure into the dark-charge budget from the table above; if the budget runs out before the exposure does, you need the stabilized sensor.
• Ask for two numbers with the quote: the stabilized sensor temperature, and the stabilization time from power-on.
• Add the cooler's power draw to the camera's supply budget, and its rejected heat to the enclosure's heat load.
• Set the calibration cadence now: a fixed setpoint keeps offsets stable for months, a floating one needs fresh dark frames whenever the room moves.

When cooling is the wrong answer

• Bright light, short exposure. At 1 ms under strong SWIR illumination, dark charge is small next to photon shot noise and read noise, and the fanless camera wins on simplicity.
• Visible or NIR silicon cameras. With a doubling interval nearer 8-10 °C and millisecond exposures, the gain from cooling rarely justifies the power and the fan. It starts to pay when exposures stretch toward seconds: fluorescence, dark-field scatter, astronomy-style low light.
• When dark frames already meet the floor. Subtraction removes the mean dark offset; if the residual shot term sqrt(N) already sits under your ceiling, the TEC buys nothing you will see.

One trap to plan for: a TEC moves heat, it does not delete it. The module dumps sensor heat plus its own drive power into the camera body, and in a small sealed enclosure that heat comes straight back. If the cabinet is already warm, read camera cooling in high temperature environments before you specify the cooled camera that has to live in it.

Next step

Run your pixel size, dark-current figure and exposure through the arithmetic above, then take the result to the infrared cameras category or the camera finder. The system configurator walks the same chain of camera, lens, light and controller in one pass, and the contact page reaches the applications team for cases the table does not cover.