Background limited mid-infrared photodetection with photovoltaic HgTe colloidal quantum dots

Thermal infrared imaging in ranges of atmospheric transparency (the mid-wave infrared (MWIR), 3–5 μm, and long wave infrared (LWIR), 8–12 μm, is presently achieved with semiconductor quantum detectors and microbolometers.^1,2 The former rely mostly on bulk InSb or HgCdTe (MCT) and allow fast imaging. When cooled, semiconductor detectors are much more sensitive than microbolometers and, at cryogenic temperatures, their detectivity reaches Background Limited Infrared Photodetection (BLIP) where the photoresponse to the fluctuation of the background ambient radiation exceeds the intrinsic detector noise. However, cooled semiconductor detectors are an expensive and bulky technology, and strategies to reduce cost are to raise the operation temperature and develop cheaper materials. With that goal, epitaxial quantum dots (QDs) have been pursued for nearly two decades³ and they have some fundamental advantages but have not yet exceeded bulk semiconductors.⁴ Potential advantages of colloidal quantum dots (CQDs) over epitaxial QDs are the flexible choice of the material and substrate, low cost fabrication, and high packing density. In the past few years, HgTe CQDs have been investigated for mid-infrared detection but only as photoconductive (PC) devices and their detectivity was at least one order of magnitude below BLIP^5,6 limited by 1/f noise. The use of photovoltaic (PV) detection allows in principle to avoid the 1/f electrical noise when no bias is applied.¹ There are numerous publications of photovoltaic response with CQDs in the near-infrared with HgTe CQDs^7–10 as well as PbS or PbSe.^11,12 Here, we report background limited CQDs PV devices operating in the mid-infrared.

A schematic of these devices is shown in Fig. 1(a). A CaF₂ window is used as the substrate. A 60 Å NiCr semitransparent electrode¹³ is e-beam evaporated on half of the substrate. This metallic coating has a resistance of ∼1 kΩ square and a transmission of 30% at 4500 cm⁻¹ and 32% at 1500 cm⁻¹ is linear in between. The substrate is treated with mercaptopropyltrimethoxysilane (MPTS), with the aim to produce an adhering layer. The HgTe CQDs¹⁴ solutions are cleaned with two precipitations using a solution of isopropyl alcohol with an added surfactant, didodecylammonium bromide at a concentration of 0.01 M, and dissolution in chlorobenzene. The CQD film preparation follows Ref. 6. Films are made by dip coating, where a drop of the quantum dot solution is coated on the sample and wicked off at an edge. After drying, the sample is immersed in a solution of ethanedithiol, HCl, and ethanol (1:1:20) for 10 s before rinsing with ethanol and drying. The process is repeated 5 to 10 times in order to obtain films of thicknesses around 300 nm. Attempts to make thicker films typically lead to flaking and shorts. The optical density (OD) of the films is measured directly on the samples and is typically 0.1 OD above the band edge shoulder. All the solution and film processing is done in air since the preparation in the glovebox was not observed to lead to significant improvements. After the film is made, the sample is transferred to a glove box in order to deposit manually electrodes with silver paint. Typical electrode areas are ∼mm², measured by an optical picture for each samples. An example is shown in Fig. 1(b). The silver paint is dried for ∼1 h before the sample is taken out of the glove box and mounted on a closed-cycle cryostat. For measurements of the photovoltaic response, the samples are illuminated, through the CaF₂ substrate and the NiCr coating, by a calibrated blackbody (BB-4A) operated at 600 °C and placed a measured distance away, typically 15 cm. The sample photovoltage is measured with a SRS 560 voltage amplifier. The photocurrent at zero bias is measured with a DLCPA-200 current amplifier and the rms noise is measured with a spectrum analyzer (SR760). The blackbody radiation is chopped at 500 Hz.

A typical temperature response is shown in Fig. 1(c). At room temperature, open circuit photovoltages, V_oc, are ∼10–100 μV and short circuit currents, I_sc, are ∼20–200 nA, with the higher values for smaller dots. Both I_sc and V_oc increase upon cooling but I_sc always exhibits a peak at some temperature. The photoresponse is quadratic as a function of distance with the source as shown in Fig. 1(d), and therefore linear with power. The photoresponse is fast as shown in Fig. 1(e) where a time constant of 0.7 μs is observed at 120 K in response to a 808 nm diode laser pulse.

The detector edge is tunable by the size of the dots, as shown by the spectral photoresponses in Fig. 1(f). The spectra are measured with a Nicolet 550 Fourier Transform Infrared (FTIR) spectrometer. The amplified I_sc is directly fed in the outer detector connection of the spectrometer and the scanning speed is set at 0.94 cm/s. The Fourier transforms of the interferogram is measured with 4 cm⁻¹ resolution, averaged 30 times and normalized to a pyroelectric DTGS detector. The log scale illustrates the sharpness of the band edge. For the sharpest edge on device 2, an Urbach tail fit gives an energy of 12 meV, indicating a small size distribution. This is similar to values reported for bulk epitaxial Hg_1−xCd_xTe where the Urbach tail is attributed to alloy and structural disorder.¹⁵ 

Table I shows data obtained for four devices for the responsivity, R, the specific detectivity, D*, and the External Quantum Efficiency (EQE). For the calculation of the responsivity, the band edge of the sample is defined as the ½ point of the signal on the detection edge and the total photon count above the band edge is calculated given the sample area A, the source area (21 mm diameter), its temperature and the sample distance to the source. Errors estimated from the uncertainty on the detector area (10%) and the incident flux (20%), arising from the distance uncertainty to the blackbody source, are about ±30%. For the specific detectivity D*, defined as A^1/2R/I_n, the current noise I_n in a 1 Hz bandwidth is measured at 500 Hz using a spectrum analyzer. It was typically slightly larger than the expected Johnson noise $4kBT/R$ for the measured resistance. Importantly, the noise showed no 1/f component unlike the ubiquitous and large 1/f noise observed in nanocrystal conductors.¹⁶ 

The responsivity and detectivity of the devices are much improved compared to those previously obtained in the photoconductive mode with similar film preparation and wavelength.^5,6 This is attributed to the shorter distance that the carriers have to travel compared to 5–20 μm in previous PC devices, as well as the absence of 1/f noise. The responsivities are as high as 80 mA/W with a maximum EQE of 2.5%. If one includes the transmission of light through the NiCr (30%) and the absorption of light by the film (typically 20%), the internal quantum efficiency (IQE) is estimated to reach η = 40% at the optimum temperature. The D* in Table I are all in excess of 10¹⁰ Jones. An increase in I_sc is easily observed by placing a hand in front of the dewar window; therefore, the detectors respond to a slightly warmer background than room temperature over ∼0.2 sr.

In these devices, the dots are considered intrinsic, but the electrodes can form Schottky barriers. The devices are typically p with a top silver contact (measured from that contact), and always n with a top In/Ga eutectic contact. The rectification is therefore likely determined by the top barrier while the NiCr is indifferent. Silver is a known p-dopant in MCT,¹⁷ and it is possible that the silver paint forms the p contact by diffusion of Ag ions in the HgTe dots. In some instances, the polarity starts as n at room temperature and reverses to p at lower temperatures, a behavior also reported previously for silver doped bulk MCT.¹⁸ 

The best performance was observed with an Ag₂Te nanocrystal¹⁹ layer between the HgTe film and the Ag paint, with device 1 as an example. Such devices are always p-type (silver paint side) and rectifying IV curves at 90 K are shown in Fig. 2. At 90 K, the detectivity of device 1 is 4.2 × 10¹⁰ Jones and covers the full MWIR with a cut-off at 5.25 μm. The noise measured with a closed cold shield is 0.35 pA Hz^−1/2 at 90 K. With the shield open to background 295 K radiation, I_sc rises to 2.7 μA as seen in Fig. 2. Integrating the photon flux below 5.25 μm, the current gives an EQE of 2.6%, consistent with the value in Table I of 2% obtained with a calibrated blackbody source. With the shield open, the noise increases to 1.4 pA Hz^−1/2, slightly larger than the shot noise and larger than the detector intrinsic noise. Therefore, the noise of the detector is limited by the 2π background flux at 295 K and in the BLIP regime.¹ The BLIP detectivity at 5.25 μm is $DBLIP*=1/hν2ϕ$ where ϕ is the 300 K blackbody integrated flux from 5.25 μm to shorter wavelengths over a solid angle of 2π and gives $DBLIP*$ = 0.96 × 10¹¹ Jones. Correcting for the NiCr film transmission Trans ∼ 0.3 and the HgTe film absorption Abs ∼ 0.2, $Ddevice, BLIP*=Trans × AbsDBLIP*$ = 2.3 × 10¹⁰ Jones. This is indeed lower than the detectivity measured in Table I with shield closed. The noise with shield open exceeded the noise with shield closed below ∼140 K.

This can be compared to cryogenically cooled MWIR bulk MCT, InSb, and PbSe detectors.^1,20–22 For optimized bulk MCT PV detectors, there is an empirical rule, Rule-07, which states that the diode dark current, when measured several kT/e in reverse bias where the diode dark current is independent of bias, follows an empirical exponential dependence on the cut-off wavelength λ and temperature T given by the equation J = 8367 × exp(−1.16 hc/λkT) A/cm².²³ Notwithstanding the low shunt resistance likely due to the imperfect film coating, the fitted diode current in Fig. 2 is 0.04 μA corresponding to a current density of 1.3 μA/cm² at 90 K. For the same cut-off wavelength of 5.25 μm, Rule 07 indicates that an optimized crystalline MCT photodiode would have a similar dark current density at 140 K. The resistance area product R_oA = 5 kΩ cm². High values of R_oA correspond to lower dark current and are beneficial. MCT diodes at 5 μm have R_oA ∼ 1 MΩ cm² at a similar temperature.¹ The present performance is therefore worse than the state of the art MCT diodes, but comparable to epitaxial quantum dot detectors while providing many avenues for further improvements.^3,24

To discuss the performance, we focus on the exciton lifetime (radiative and non radiative), $τo$⁠, and the hopping time, $τh$⁠.²⁵ There is no data on $τo$ for this size of HgTe dots, but photoluminescence quantum yield measurements in a solution suggest $τo$ ∼ 1 ns with a weak temperature dependence.²⁶ The efficiency of photogenerated charges after creation of an electron-hole pair is estimated as $η=2τo/(τh+2τo)$ where the factor of 2 is with electron and hole of equal mobility. There is no measurement of the hopping time for this sample but ∼1 ns is consistent with the measured efficiency here, and previous mobility values.²⁷ In nanocrystal solids, $τh$ increases at low temperatures due to energy barriers and follows an Arrhenius behavior, $τh=τhope−Eb/kT$ before entering the regime of variable range hopping.²⁸ Therefore, $η$ is expected to decrease with decreasing temperature and this explains the decreased I_sc at low temperature.

The decreased current at higher temperature is mainly attributed to the decreased carrier lifetime. The probability that an electron or hole encounters an opposite carrier and recombine is $(1−η)n$⁠, where n is the average number of carriers (n ≪ 1) per dot, such that the carrier lifetime in the film is $τ=τhop/(1−η)n.$ The carrier diffusion length is $LD=Dτ=d16(1−η)n$⁠. The intrinsic carrier concentration per dot is $n ∼ NcNv exp(−Eg2kBT)$⁠, where N_c and N_v are the electron and hole states density, considered to be 2 each for simplicity, but a higher value is possible for holes. Using $τ0=τhop$ = 1 ns, E_b = 12 meV, d = 12 nm, N_c = N_v = 2, gives $LD$ ∼ 44 nm at 300 K. This is shorter than the device thickness of l ∼ 300 nm, which explains the low current of the devices at room temperature. As the temperature drops, the diffusion length increases. Assuming that the internal quantum efficiency is $ηe−l/LD$⁠, the calculated responsivity in Fig. 3(a) qualitatively captures the peak responsivity at cryogenic temperature.

To model the detectivity, a precise diode structure is needed and the discussion above indicates that it is too preliminary. However, as an initial attempt, we consider the diode current as $I=I0(exp(eVkT)−1)∼IoeVkT$ and for I_o we take the thermal generation rate in the device of area A, $Io=Aenbl/τ$⁠, where n_b is the bulk density of carriers calculated from n, the volume of the nanocrystals and a filling factor of 0.6. In addition of the thermal carrier density, a doping of n_o = 0.6 × 10⁻⁴ carrier per dot is chosen so that the diode resistance matches the experimental value of R_oA = 5 kΩ cm² at 90 K. Using the Johnson noise, the calculated detectivity is shown in Fig. 3(b) and qualitatively captures the device performance.

Looking forward, the expected fundamental advantage of CQDs compared to bulk MCT is that Auger is not relevant as long as $τo<τA/n$ where $τA$ is the trion lifetime. There are no measurements of $τA$ for any HgTe CQD at the moment. However, the biexciton lifetime for smaller dots of 4–6 nm diameter is ∼50 ps.^26,29 Biexciton lifetimes appear to scale as the volume of CQDs,³⁰ so that extrapolating to a diameter of 12 nm, the biexciton lifetime should be 8 times longer, and the trion at least 16 times longer or 0.8 ns. At 300 K, the carrier density per dot is estimated at 0.015 so that $τA/n$ ∼ 50 ns. This is faster than the radiative lifetime (estimated at ∼0.5 μs), but much slower than the present exciton lifetime. To solidly assess the future potential of the materials, trion lifetimes will therefore need to be measured.

The model is simplistic but it helps guide the search for improved performance. With the same crude device, much higher operation temperature will be obtained with a longer exciton lifetime, a goal that should be possible by controlling the surface chemistry and environment of the HgTe CQDs. In parallel, the operation temperature will also increase with optimized diode and optical design using more transparent electrodes and structured metal plasmonic electrodes³¹ maximizing the optical absorption in the dot layer.

In summary, this report of mid-infrared BLIP detection with colloidal quantum dots motivates interest towards developing colloidal quantum dots for fast, low cost, and sensitive thermal infrared detection.

The work was initially supported by the U.S. National Science Foundation (NSF; Grant DMR-1104755) and received continued support by ARO W911NF-15-1-0110. The authors made use of shared facilities supported by the NSF MRSEC Program under DMR-0820054.