This paper presents a summary of current HgCdTe photodiode techology. It discusses the advantages and disadvantages of this material, the operating principles of photodiodes with an emphasis on HgCdTe diodes in particular, the factors that affect their performance, and certain special considerations in the design and manufacture of HgCdTe photodiodes.
Mercury cadmium telluride (HgCdTe) photodiodes have constituted one of the most important types of infrared photodetectors in the past few decades. They are particularly important in electronic focal plane arrays (FPAs), which find extensive use for military applications in IR sensing equipment. For this reason research efforts into improving HgCdTe technology have been particularly intense. In this paper I survey the features and characteristics of HgCdTe photodiodes and the research efforts into them.
This alloy is notoriously difficult to deal with, as we shall see later in discussing passivation layers and substrates. Rogalski characterizes it as having the "most difficult problems of any semiconductor material in mass production." [1, pg. 1395] Aside from the obvious health hazards involved, it has the problems of compositional control associated with any ternary alloy and suffers from poor repeatability in the growth of bulk crystals and layers . That it has become so popular in spite of this limitation is testimony to its superior properties in IR sensing applications.
One of the superior properties of HgCdTe, a direct gap semiconductor, is the wide spectral range of detectors made from it. They can work as long-wavelength infrared (8 - 12 um, LWIR) and medium-wavelength infrared (3 - 5 um, MWIR) devices because its direct bandgap changes with composition. For the usual LWIR and MWIR operation, Hg1-xCdxTe has x between .2 and .3. The bandgap has been found to be given by
Eg/q = -.276 + 1.65x + 5.85x 10^-4(1-2T) + .269 x^4 eV, (1)
roughly linear so that the change in the bandgap for a given stoichiometric change, dEg, is about 1.65dx. This fact allows for exceptionally fine control over the spectral response, on the order of .1 um for dx=.001 in the MWIR-LWIR range .
HgCdTe detectors also operate effectively at higher temperatures than competing technologies such as GaAs/AlGaAs quantum well infrared photodetectors (QWIPs), although QWIPs in particular may eventually supplant them in some applications due to the comparative maturity of GaAs/AlGaAs technology as their properties improve . HgCdTe MWIR devices can give background- limited performance, as discussed later, at temperatures as high as 180 K. This temperature is within the reach of thermoelectric cooling, an attractive property for use in mobile equipment. LWIR devices can be background-limited, as explained later, at 77 K, the temperature of liquid nitrogen. Extrinsic silicon and Schottky barrier detectors must be kept at less than half those temperatures for the same performance. QWIPs are somewhat better (Fig. 1) .
Photodiodes operate by the principle of photogeneration; incident photons generate electron-hole pairs (EHPs) in the depletion region of the forward biased diode. These generated carriers are separated by the electric field and thus contribute to the total current through the diode. The associated electronics of course can then relate this current to the incident photon flux . HgCdTe absorbs the photons across the fundamental energy gap, giving it favorable properties as compared to other types of photodiodes . Such detectors are known as intrinsic detectors.
Several quantities characterize photodiodes, including quantum efficiency, response speed, and noise. Quantum efficiency is the number of EHPs generated per incident photon and is given by h=(Ip/q)(Popt/hn) where Ip is the generated current from the absorption of incident optical power Popt at photon energy of hn. A measure related to h is responsivity, simply Ip/Popt .
The absorptivity a of the material determines quantum efficiency to a large extent and its strong dependence on wavelength determines the spectral response of the diode. At the long wavelength cutoff, a dies off and consequently no photons are absorbed to generate EHPs. At the short wavelength cutoff, the photons are absorbed so readily that they cause generation near the surface, where they recombine before they can traverse the junction .
Three main factors determine the response time. The diffusion of carriers from outside the depletion region produces a delay, for which reason the junction should be formed near the surface to reduce the distance over which such carriers must diffuse. The other two limitations, drift velocity in the depletion region and capacitance of the depletion region, offer competing constraints on device design; a thinner depletion region reduces the drift delay but a wider one reduces the capacitance and the RC time constant. Sze suggests a depletion layer width which results in a transit time on the order of half the modulation period .
The last major photodiode characteristic, noise, seems to garner the most attention in current research. Noise in the diode comes from two main sources, background radiation and dark current . Background radiation is conceptually the simplest. It is primarily a problem in IR photodiodes such as HgCdTe devices. Since the environment, including the instrument itself, constantly emits infrared radiation with temperature-dependent wavelength, this must largely be controlled in instrument design. LWIR and MWIR detectors are often operated at low temperatures to prevent the IR background from swamping the desired signal. This complicates the study of these diodes considerably, as their characteristics must be understood over a wide temperature range, from sub-40K up to room temperature.
Dark current is the physically more complicated form of noise. Its sources vary not only with bias but with the temperature, which varies with the required operational wavelength range. As its name implies, this is the current through the diode when it is receiving neither an optical signal nor background radiation. In journal articles researchers frequently characterize dark current with the RoA product, though its definition is conspicuously absent from many texts on the subject.
At the low voltages typically used in operation, RoA is a approximately proportional to the dark current. The figure is calculated experimentally as the ratio of differential change in current density J (hence the area term) to change in voltage V at zero bias. This may also be calculated separately for different processes from the appropriate theoretical models and the reciprocals of the results for different current components added to obtain the reciprocal of the total RoA product ,
(1/RoA)TOT = (1/RoA)A + (1/RoA)B + (1/RoA)C + ... (2)
Unfortunately, RoA does not always correspond as closely to the dark current level as might be desired, and the validity of its use in characterizing devices has been questioned on this account .
The study and reduction of the dark current is one of the most common topics in current literature on HgCdTe photodiodes, and it is all the more involved in these devices because the stoichiometry varies in addition the operating temperature and bias as mentioned above. Much of the time it is the mechanism that limits the performance of the diode, most other cases, especially at higher temperatures, being limited by the background radiation (background-limited infrared performance or BLIP .)
In general, most sources consider the mechanisms dominating HgCdTe diode behavior below 60K, including dark currents, to be poorly understood. According to Rogalski , trap-assisted tunneling usually dominates in n+p HgCdTe photodiodes below 77K. R.E. DeWames et al.  made a detailed study of Hg1-xCdxTe ion- implanted p-n junctions with x=.224 in 1988. They enumerated three chief mechanisms in high-quality diodes that contribute to the dark current, a diffusion component, a type I tunneling component (internal field emission), and a type II tunneling component (possibly trap-assisted tunneling.) Properties of the thermal dependence (or lack thereof) in the current suggested that generation-recombination (G-R) currents played no significant part in the dark currents of these diodes, as also determined by Lanir et al. . The RoA products (zero bias) from 30K to 500K were believed to be dominated by type II tunneling, though they could not explain the temperature dependence for this component. For small forward bias, they found somewhat exotic mechanisms below 20K, type II tunneling from 30K to 50K, and diffusion for T > 50K. At large forward bias, diffusion currents dominated even at low temperatures. For large reverse bias and small reverse bias below 30K, they observed evidence favoring type I tunneling. An interesting result was that they found high non-uniformity among diode RoA products when tunneling mechanisms dominated. This is of great concern since such photodiodes are often used in large FPAs and uniformity is of some importance in these devices.
Much of the tunneling which contributes to the dark currents seems to occur due to band-bending at the surface (especially drastic due to the small band gap). The surface is indeed the source of a number of effects which often determine the performance of HgCdTe photodiodes [8,9].
In addition to performance reduction from the noise due to the dark currents and background, recombination acts to reduce the quantum yield. If the generated carriers recombine somewhere in the diode, they will obviously not contribute to the photocurrent. The problem can be especially troublesome in this alloy system because of the difficulties in processing the material. Surface effects and defects are hard to control, leading to a number of recombination mechanisms.
The chief difficulty with the surface is that HgCdTe unfortunately does not form a convenient well-behaved oxide at its surface like silicon. Therefore special techniques have been developed to passivate the surface. Passivation of HgCdTe is difficult, in part due to its complex chemical nature. Additionally, the volatility of the mercury component makes the bare HgCdTe surface extremely temperature sensitive, such that passivation and deposition processes are limited to relatively low temperatures . W.B. Tang et al. recently (1992) studied three types of passivation layers grown under 90 C, SiO2, ZnS, and a combination of deposited ZnS and a native oxide film, having produced what they characterize as "excellent" results with the SiO2 . An earlier (1988) survey of passivation techniques by Nemirovsky and Bahir  analyzes these types of layers as well as SiNx passivating layers. They list a number of desirable characteristics for both native layers and dielectric films, including excellent adhesion, a high energy gap, and thermal stability. Rogalski's 1994 survey of IR detector trends  however indicates that CdTe or CdZnTe passivation layers have become the most popular laboratory techniques now.
Passivation is obviously an involved topic, but the topic of the substrate is no less so. An essential consideration in the choice of substrate is whether the optical signal falls directly on the active region or whether it must pass through the substrate. In the latter case, the substrate must clearly not absorb at the wavelength of interest. Two long-popular substrates are CdTe and CdZnTe . These substrates are good when the optical signal passes through the substrate and are metallurgically compatible with HgCdTe, but they have several disadvantages, namely fragility, difficulty in single-crystal growth, and a poor thermal conductivity at cryogenic temperatures . In addition they have a large thermal mismatch with silicon as epitaxial layers, which makes it difficult to integrate them with processing circuitry . Sapphire, Al2O3, is consequently preferred for applications with an MWIR cutoff wavelength since it transmits up to around 4.6 um. The sapphire provides better mechanical characteristics, enhancing large array reliability , and has a 40% smaller thermal mismatch with silicon . One of the most common sapphire-substrate processes, developed by Rockwell, is the "producible alternative to CdTe for epitaxy, 1" or PACE-I in which CdTe provides a buffer layer between the HgCdTe and the sapphire [ 1, 5, 10]. A group at the Rockwell International Science Center is currently exploring processes for growth on composite silicon (CdTe/GaAs/Si -- PACE-III) and lattice-matched CdZnTe substrates .
In the preceding paper, I have discussed the position of the HgCdTe photodiode in the current technological milieu, its operating principles, its characterization, its performance, and aspects of its production. It is a flexible technology, albeit correspondingly complex, with a broad range of uses in infrared sensing applications. These uses will undoubtedly assure its importance if not its preeminence in many areas for some years to come.