Silicon becomes transparent beyond 1micron so CCDs are useless in the Infra-Red part of the spectrum. Alternative semiconductor materials are required such as InSb, InGaAs or Mercury-Cadmium-Tellurium (MCT). The latter dominates and is used by Teledyne-e2v in their range of Hawaii detectors. Since nobody has found a way of building CCDs from these materials, IR detectors generally consist of arrays of IR photodiodes, each individually bonded to a silicon readout circuit known as the Multiplexer. So each sensor consists of two layers: a silicon layer containing the read-out electronics and an MCT layer containing the photodiodes. Aligning and mating these two layers is known as “hybridisation” . This complex process contributes to the extremely high prices of these detectors.
Teledyne-e2v dominates the market in large format IR photodiode arrays with its H2RG and H4RG products. A European project to produce a competing 4Mpix device known as ALFA was abandoned due to technical difficulties.
Hawaii devices have excellent QE and can have their wavelength range modified during manufacture to suite the intended application. The cutoff wavelength can be varied from 1.7um to ~14um by modifying the Cadmium fraction used in the MCT alloy. Hawaii devices have poor cosmetics with up to tens of thousands of inoperable pixels. They also suffer from charge trapping within the pixel structure. This causes the complimentary problems of reciprocity failure and persistence. Reciprocity failure means that up to 1% of the collected photocharge is not measured at the end of the current exposure, with the trapped fraction increasing for longer exposure times. This trapped charge then slowly de-traps into subsequent exposures as a persistence signal where a faint version of the previous exposure is still visible. Bad persistence can take hours to fully disappear. Hawaiis have 3T pixel format meaning that each pixel contains only three transistors. They use DC-coupled video amplifier chains which are wide-open to temperature induced drifts and 1/f noise. They therefore must be operated under very stable conditions (thermally and electrically) to get good noise performance. Even so, their noise is considerably higher than CCDs and much higher than visible-wavelength photodiode arrays that use 4T (or more complex) pixel structures. Both the James Webb Space Telescope and the European Extremely Large Telescope use, almost exclusively, Hawaii detectors. Modern astronomy is highly dependent on this critical technology, and it comes from a single US supplier.
Teledyne also offer the Geosnap detector for high frame rate applications. This is fully digital with much more on-chip circuitry than the Hawaii detector. Crucially, it uses a Capacitive Trans Impedance Amplifier within each pixel. This maintains a constant bias across each of the MCT phtotodiodes thus eliminating persistence and improving linearity. It is also a hybridised detector. The light sensitive layer can be MCT or Silicon thus giving potential cutoff wavelengths ranging from 0.4 to 15um.
Another very promising IR-sensor technology, known as LmAPD (Linear Mode Avalanche Photo Diode) is being pursued by Leonardo UK. They offer 2.5um sensitive detectors that achieve around 1e RMS read noise through the use of an internal avalanche gain mechanism. Their sensors already form part of the highly successful Gravity instrument on the ESO VLT telescopes. Investments from ESA and NASA have since lead to the development of the larger format sensors IBEX (2k x 2k pixels) and Ike Pono (512 x 512 pixels).
One great strength of this technology is that the multiplication mechanism is very clean in HgCdTe (MCT) material and introduces very little extra noise. This is in contrast to Silicon EMCCDs where the internal gain mechanism doubles the variance in higher-signal flat fields. LmAPD pixels have an ingenious design where the bandgap of the MCT material is modulated throughout their depth. The incident photons first encounter an MCT absorber layer whose bandgap corresponds to a cutoff wavelength of 2.5um. The photoelectrons that are generated are then accelerated through a gain region with a much lower bandgap, i.e. one in which electrons can be elevated to the conduction band with much less energy. In traversing this gain region an avalanche multiplication occurs. The electronic gain can be controlled by an externally applied bias voltage. The electron packet that exits the gain region is then sensed using low-noise amplifiers constructed on a silicon layer to which the MCT is bonded. In the past this technology has suffered from high dark current (at higher gain settings) and light emission from the active circuitry within the detector. These problems have been addressed and as of 2022 the read noise of the Ike Pono sensor was <1e RMS with a dark current of 10-3 electrons per pixel per second.
Photodiode arrays produce a signal proportional to the number of IR photons absorbed . An alternative to photodiode array technology is the Microbolometer. These sensors produce a signal proportional to net incident power. They are available in large 2D arrays and are principally intended for military imaging in the thermal infra-red (7-14microns). Each pixel consists of an absorber structure with an integrated temperature dependent resistor. Changes in absorbed power change slightly the temperature of the absorber and the value of its associated resistor. This change is then sensed by a low-noise amplifier within each pixel. They are much less sensitive than photodiode arrays but are cheaper and do not require cooling (although they do need to be temperature stabilised).
Further information on IR detectors:
Presentation by Jim Beletic (Chief Scientific Officer, Teledyne)
Lynred PICO microbolometer array
JATIS, 2020, Predictive Model of Persistence in H2RG detectors
Jupyter Notebook with IR Radiometry routines (useful for IR testbench design)
Jupyter Notebook with Persistence Model of an H2RG detector