SWIR camera

A SWIR camera records images in the short-wave infrared, the band beyond silicon's cutoff near 1.1 µm and below the thermal bands used by mid- and long-wave imagers. The standard sensor is a focal-plane array of InGaAs photodiodes lattice-matched to InP, responsive from about 0.9 µm to 1.65 µm, where the 0.75 eV bandgap of In₀.₅₃Ga₀.₄₇As sets the edge. Common formats are 320 × 256, 640 × 512 and 1280 × 1024 pixels, with pixel pitches from about 5 to 25 µm depending on generation. Most laboratory SWIR cameras stabilize the sensor with a thermoelectric cooler to reduce and steady the dark current.

How the sensor is built

InGaAs cannot hold transistor circuits of the quality silicon can, so the photodiode array and the readout are made separately. The InGaAs array is grown on InP, patterned into diodes, and flip-chip bonded to a silicon readout integrated circuit (ROIC) with an indium bump on every pixel; light enters through the InP substrate, which is transparent in this band. The ROIC integrates each pixel's photocurrent on a capacitor and multiplexes the values out, much as a CMOS image sensor does (see CCD vs CMOS). Removing the InP substrate extends the response down into the visible, to about 0.4 µm, in so-called visible-SWIR sensors. Extended-wavelength InGaAs, with a higher indium fraction, reaches 2.2 to 2.6 µm with much higher dark current, and HgCdTe arrays serve the same range where lower noise is needed. Germanium-on-silicon and colloidal quantum-dot sensors are lower-cost alternatives with higher dark current than InGaAs.

Signal levels and a worked number

A photon at 1550 nm carries 0.80 eV (photon energy $E = 1.23984/\lambda$ with $\lambda$ in µm). The number of photoelectrons collected by a pixel receiving power $P$ for an integration time $t$ with quantum efficiency $\eta$ is

$$N = \frac{\eta P t \lambda}{h c}.$$

For $P = 1$ nW, $t = 1$ ms and $\lambda = 1550$ nm, the incident photon number is 7.8 × 10⁶, or 6.2 × 10⁶ electrons at $\eta = 0.8$. SWIR pixel full-well capacities range from tens of thousands to a few million electrons depending on the gain setting, so a nanowatt concentrated on one pixel saturates it. Imaging a focused laser beam therefore needs neutral-density attenuation of several orders of magnitude, or very short integration times.

Noise sources are the same as in visible sensors: photon shot noise, read noise (typically tens of electrons rms, higher than in scientific silicon sensors), and dark current, which in InGaAs is much larger than in silicon because of the narrower bandgap. For diffusion-limited dark current, which scales as $\exp(-E_g/k_BT)$, the increase is a factor of two for about every 7 °C of warming near room temperature, so cooling matters for exposures longer than a few milliseconds.

Where SWIR cameras are used

In photonics laboratories the main uses are viewing and aligning telecom-band beams at 1310 and 1550 nm, beam profiling, inspecting the output of fibers, waveguides and grating couplers, and imaging silicon photonic chips, since silicon is transparent beyond about 1.1 µm and a SWIR camera can look through the substrate. Industrial uses include wafer and solar-cell inspection (including electroluminescence imaging of cracks), sorting of materials and produce through the water absorption band near 1.45 µm, and imaging through haze, since longer wavelengths scatter less from small particles. In surveillance, SWIR cameras image by reflected light, including the night-sky airglow, and so produce images that resemble visible photographs.

Pitfalls

InGaAs arrays have more defective pixels than silicon sensors and a larger pixel-to-pixel spread in offset and gain, so they are shipped with non-uniformity correction tables that are valid only at the calibration temperature and integration time; changing either without new dark and flat frames leaves fixed-pattern noise. Glass optics designed for the visible are usually transmissive in the SWIR but not color-corrected there, and visible anti-reflection coatings may reflect several percent at 1550 nm. Hot objects, above a few hundred degrees Celsius, begin to emit measurably in the SWIR, which can add background in furnaces or near high-power laser optics.

Common questions

What is the difference between SWIR and thermal cameras?

A SWIR camera, like a visible camera, mainly records reflected light, and the 0.9–1.7 µm band carries little thermal emission from room-temperature objects. Thermal cameras work at 3–5 µm or 8–14 µm, where objects near 300 K emit strongly, with cooled photon detectors or uncooled microbolometers.

Can a silicon camera see 1550 nm?

No. Photons at 1550 nm are below silicon's bandgap and pass through without being absorbed. Phosphor-coated upconversion cards or an InGaAs SWIR camera are used instead.

Why are SWIR cameras expensive?

The InGaAs array is grown on small InP wafers and each die is hybridized to a separate silicon readout, so the process has lower volume and more steps than a monolithic silicon sensor.

References: A. Rogalski, Infrared and Terahertz Detectors, 3rd ed. (CRC Press, 2019); S. M. Sze, K. K. Ng, Physics of Semiconductor Devices, 3rd ed. (Wiley, 2007).