Infrared is that portion of the spectrum adjacent to the red end of the visible spectrum and extending to about 30 µm.
Infrared imaging comes in two types – active and passive. Active infrared imaging is much like visible imaging in which an object is externally illuminated and the results of this illumination are imaged. In passive imaging, the object itself provides the illumination, in the form of emission of electromagnetic radiation. Of course, passive imaging of very hot objects is possible in the visible band but passive imaging of lower-temperature objects is much more common and more widely useful. Since detectors in each of the infrared bands is always sensitive to emitted energy, careful design of cameras to minimize the effect of stray emissions, even from the camera itself, is essential.
The optical absorption characteristics of the atmosphere have led to the separation of the infrared spectrum into bands. Near infrared is composed of wavelengths adjacent to the visible while the rest of the bands fall farther and farther away. Each band has its own set of useful detection and optical materials that are roughly as exotic and expensive as their relative distance from the visible band. One practical consideration in the implementation of infrared imaging is that essentially all sensors and cameras operating beyond the near infrared are export-controlled.
In defiance of the first atmospheric transmission limit around 1200 nm, the extent of the near-infrared band is essentially defined by the long-wavelength photon absorption limit of silicon – about 1100 nm. Because the transparency of silicon increases rapidly beyond 800 nm, thick image sensors are needed to capture a significant fraction of photons beyond 800 nm. In most CCD and CMOS sensors, even though the silicon may appear to be several hundred micrometers thick, the actual photon collection depth is defined largely by the depleted region, which may only include a few micrometers at the input surface. Even those sensors constructed to collect photons through the full thickness of the silicon still only have a few percent efficiency at 1000 nm.
The advantage of using the NIR band is that the same sensor, and sometimes the same optics, can also be used for visible imaging. This provides additional flexibility in selecting illumination or filters to produce contrast where different wavelengths the visible alone might not be sufficient. For example, bruises can be identified in fruit by NIR imaging. This photograph, taken using a CMOS sensor in sunlight through a 1064 nm filter, shows several potentially useful NIR effects.
Most significantly, the opacity of water at 1064 nm is apparent. Even a few centimeters of water appears nearly black. In addition, the high NIR reflectivity of foliage is demonstrated as it the transparency of the PETE plastic used to make the water bottle. A common use for NIR detection in security systems is imaging through dyes. Most dyes used for color separation in silicon imagers are transparent beyond about 800 nm. Under NIR illumination all of the dyes disappear so that the underlying material can be seen. This is the principle used in film digitizing to remove scratches from the scanned image. Three color channels detect the RGB components while a fourth channel with only NIR response sees through the color dye layers and views only the film substrate material. When this channel is subtracted from each of the color channels, the scratches disappear. Dye transparency in the NIR also lies behind the capacity for image sensors with color filter arrays to see at night using NIR illuminators and imaging of lower paint layers in works of art.
NIR imaging allows imaging hot items by their own emissions, but temperatures of at least 800-900 K are required to produce enough energy in the NIR band. This is typical of soldering irons, some industrial processes and some invisible flames. Heated objects can be imaged by sensors extending to 1000 nm at temperature below those where the eye will see a dull red glow. When the eye can see the glow, then any visible band sensor will do.
Short-wave infrared (SWIR) overlaps with the boundary of the visible, starting around 900 nm, and extends to the atmospheric (and glass) transmission cutoff around 2.5 µm. While SWIR imaging has been in use for temperature profiling of boilers and other equipment and for some aerial surveillance for over 30 years, only the ascendance of infrared laser diodes in telecom applications provided sufficient funding of the technology, especially sensors, to advance the use of this band significantly. For many years, the detector of choice for SWIR was a lead sulfide (PbS) photoconductor in a vidicon-style image tube, but now the pre-eminent detector material for SWIR is the solid-state indium gallium arsenide (InGaAs – pronounced in-gas) array even though most InGaAs detectors respond only to 1.7 µm. The complicating factor in the use of InGaAs arrays for imaging is that InGaAs itself cannot serve as the substrate for the readout circuitry so that the detector array must be bump-bonded to a matched readout array made of silicon. This leads to relatively large pixels compared to those in silicon imagers.
The typical range for InGaAs sensors is 25 to 50 µm but pixels as small as 12 µm have been put into production. The large pixels force larger sensor sizes and raise costs for both the sensors and their optics but they also allow the inclusion of complex silicon circuitry for each pixel to support pixel-rate correction of sensitivity, dark current and fixed-pattern noise and for complex control circuits similar to those found in sophisticated silicon CMOS imagers. The fill factor of InGaAs sensors can be quite high because the circuitry is not in the same plane as the detector. In addition to area arrays, line scan sensors are also readily available . These were originally designed to be used as channel power monitors in telecom systems but also serve well in line-scan cameras. Low-light-level SWIR sensors have also been built, using an InGaAS layer as a photocathode in an electron-bombarded intensifier structure.
InGaAs is actually responsive from the visible through the SWIR range but the visible is blocked from detection because the InGaAs layer is grown on an indium phosphide (InP) wafer substrate. When the InGaAs is bonded to the silicon readout circuit, the InGaAs layer faces the silicon so incoming photons must pass through the InP to reach the InGaAs. Unfortunately, InP blocks everything below about 900 nm so the visible response is lost. Devices have been made in which the InP is etched off after bonding of the InGaAs to the silicon so that the response can reach below 500 nm. These devices are difficult to make and expensive but can provide reasonable response from visible through SWIR, minus the wavelengths blocked by the atmosphere, of course. InGaAs is generally run at room temperature for industrial and surveillance applications but cooling can improve image quality for scientific use. Cooled cameras often include a cold reference plane that can be placed in front of the sensor on command to provide calibration.
Germanium photodiodes are responsive over the full visible-SWIR range but have not found wide use in imaging because the lack of an insulating germanium oxide prevents fabrication of MOS transistors on the germanium wafer. Germanium could be bonded to silicon for readout like InGaAs but the assembly would have to be cooled due to the high dark current of germanium diodes at room temperature. Another possible way to use germanium is to deposit it on silicon so it can function only to generate charge while the silicon provides the diode and readout functions. A device using a silicon array with germanium incorporated into pits in each pixel was manufactured for a time but is now out of production.
One difficulty in using any sensor in the visible and SWIR simultaneously is the extreme burden put on the optics. Ideally, lenses for this application should focus over the entire 400 to 1700 nm band simultaneously. Unfortunately, such a design cannot be realized with any combination of known glasses so techniques such as diffractive elements would have to be employed. At this writing, no such lenses exist. Even if only the 900 to 1700 nm band must be imaged, the design options are limited. The large pixels in InGaAs imagers relieve the design requirements somewhat but large sensors mean long focal lengths, leading to large lenses. In addition, the curvatures of the surfaces must be constrained so that the required wide-band coatings function properly. All of this leads to relatively high cost for lenses specifically designed for SWIR use, especially in applications requiring operation all the way to the 2.5 µm band edge. Lenses intended for visible use can be applied in relatively undemanding SWIR applications, especially where only a narrow band must be imaged as in telecom testing.
Passive SWIR Imaging
Objects at 400K and above produce enough emission in the SWIR band to be detected in contrast to the background. This makes SWIR imaging useful for detecting heating due to thermal insulation defects, electrical connection problems, and even passage of heated fluids. This temperature range is common in industrial processing applications so SWIR imaging has found wide use in both on-line and off-line inspection and in examining machines and facilities for incipient failures. SWIR imaging in surveillance applications is enabled by nightglow from the sky, which is concentrated in the 1.3 to 1.8 µm band. Nightglow is sufficiently bright, even on overcast nights, to permit passive imaging.
The next atmospheric window, extending over the range of roughly 3 to 5 µm, is home for the mid-wave infrared (MWIR) band. Two detector types often used in MWIR cameras, indium antimonide (InSb – pronounced ins-bee) and mercury cadmium telluride (HgCdTe, are narrow-bandgap semiconductor photon detectors operated while cryogenically cooled and built as bonded assemblies like InGaAs. In laboratory applications, the choice between InSb and MCT often depends on the total range of wavelengths required rather than solely on MWIR performance. InSb can detect photons to below 1 µm with appropriate optics but MCT is limited on the short wavelength end to about 2 µm. On the other hand, MCT can detect out to beyond 12 µm, while InSb extends only to slightly beyond 5 µm. Microbolometer detectors can be used in the MWIR band but generally produce insufficient temperature resolution for most applications. Some MWIR detectors are available using quantum well arrays but these generally only detect over a designed narrow band.
In addition to temperature measurement applications, MWIR cameras are very useful in narrowband imaging intended for detection of specific substances. This has led to their wide use in hyperspectral imaging systems in which area images are taken of many narrow spectral bands simultaneously to allow real-time detection of the presence of several target substances. In spectroscopic terms, the range of detection for the MWIR band is about 2000 to 3300 wavenumbers (cm-1), which covers the detection of CHx, various C double and triple bonds, Nitrogen bonds and several other important identifiers for organic compounds. No other band capable of imaging through the atmosphere covers as many individual bond types. As a result, MWIR can be used for imaging uniformity, purity and distribution of a wide variety of organic compounds.
Lenses for the MWIR band are generally made of germanium and silicon with protective windows are made of sapphire (aluminum oxide). All of the usual lens types can be made – fixed focal length, switchable field of view and zoom lenses. Since the ratio of the high and low wavelengths to be covered is relatively small, correction of aberration can be fairly accurate. Diffraction-limited designs are possible. However, lenses containing germanium must be used with care because germanium can experience thermal runaway. As germanium gets hotter, its transmission drops so if it is used in situations where energy arrives at a rate faster than it can be dissipated, then the lens can overheat and suffer permanent damage. To avoid lens overheating in in industrial situations, it is important to block radiation in the range of 1.5 µm and beyond from falling on a germanium lens especially if the radiation is not used to form an image.
Lenses used with cryogenic sensors are usually designed to be used with a cold stop to prevent radiation from outside the imaging field of view from striking the sensor package. It is important to determine the cold stop setup for each particular camera to assure that lenses selected will work properly.
Beyond the next atmospheric absorption band, from 7 to 14 µm, lies the long wave infrared (LWIR) band. This is the band traditionally associated with thermal imaging because it is the band providing the best temperature discrimination for objects around 300k. The first thermal imagers, built back in the 1960s, used MCT cooled with liquid helium. As MCT quality and readout circuits improved, the operating temperature could be allowed to increase until now liquid nitrogen is sufficient. MCT still produces the best temperature resolution in this band, down to 15 milliKelvins, and it is still the detector of choice in the most critical applications but is relatively complicated to support and expensive. As a result, MCT has been supplanted in all but the most critical applications by microbolometer arrays.
A bolometer was originally pair of metal strips, one blackened and exposed to the radiation to be measured, and the other shielded. Incoming radiation heated one strip and the difference in resistance was measured using a Wheatstone bridge. Microbolometer image sensors are essentially of the same construction except that the resistance measurement is typically single-ended – there is no reference strip. Commonly, the absorbing material, which must also have a substantial temperature coefficient of resistance, is amorphous silicon or vanadium oxide prepared to present a black absorbing surface. The strip is silicon etched by chemical micromachining so that it is suspended over an open space. This construction prevents rapid conduction of heat away from the resistive element and improves sensitivity at the expense of lengthening the time constant of the measurement. The tradeoff choice for most imaging applications is to match the time constant to the imaging frame rate, which results in temperature resolution on the order of 50 milliKelvins. The primary advantage of microbolometers is their ability to operate over a wide range of temperatures up to 60°C. This simplifies cameras design substantially. However, to avoid serious temperature-induced drift in image performance, microbolometers are often stabilized at some selected operating temperature with closed-loop thermoelectric plates.
LWIR imaging is primarily used for viewing temperature differences of a object only a few tens of degrees different from the background. The widely-known night-vision applications involving detection of people, animals, vehicles and shelters by their thermal signatures are still the most common use for LWIR cameras. However, the temperature discrimination of these devices is sufficient to reveal small differences in heat transmission by insulation or variations in industrial process temperatures or heating due to friction in machinery and vehicles. These tasks all involve detection of emitted radiation but active illumination is also involved in some uses. In industrial settings, LWIR cameras are used to monitor CO2 laser radiation, especially for its presence in places where it is not desired because stray invisible laser beams, especially at the typical power levels used in industrial processes, can be extremely dangerous. LWIR radiation penetrates smoke well because its wavelengths are not scattered as much as is visible light. This has led to the use of LWIR imaging in monitoring processes that generate smoke and in firefighting.
Because the detector elements in LWIR microbolometer sensors are essentially mechanical, they are typically larger that the silicon photodiodes in silicon visible sensors. Pixel pitches are commonly 25 to 50 µm although pixels down to 10 µm have been produced. The minimum pixel pitch is also constrained, as it is in InGaAs sensors, by the need to bond the detector array to an underlying readout device, which, for microbolometers, can be just a multiplexer since the signal is always available on the detector elements. LWIR line scan sensors have also been made for thermal monitoring of moving objects and panoramic thermal imaging.
Optics for LWIR incorporate primarily germanium. Because of the well-developed set of applications for this band, the selection of optics is quite broad, including fixed, multiple-field and zoom optics, in addition to reflective optics with long focal lengths and remote controls.
The considerations described in the MWIR Optics section above apply to cooled LWIR cameras as well.
Very Long-Wave Infrared
Beyond 14 µm the goal remains object detection and measurement but the objects are those colder than 300K. Since most radiation in this band is absorbed by carbon dioxide in the atmosphere (the “greenhouse effect”), observation typically occurs in space where the targets are discarded shrouds, fragments of old satellites and other objects either in orbit or on space trajectories. MCT works as a detector over the shorter-wavelength part of this band, which extends to some indeterminate wavelength where the radiation is more often termed “terahertz”. At the longer-wavelength end, more exotic materials like microstructured quantum wells, QWIPs and even quantum dots are being explored for use in improved detectors. Outside space or the laboratory, little is seen of VLWIR imaging systems. Wider application will need to wait until significant industrial operations are undertaken in atmosphere-free locations.