Development of multispectral IR imagers has gained traction over the years due to the increasing need to extract spatially coincident spectral information from distinct wavebands to vastly improve object identification and recognition. A promising technology for such an endeavor is the III-V semiconductor-based type-II InAs/GaSb superlattice (T2SL) [1] and its variants that may prove to be cost effective and possess unique band-engineering capabilities that culminate in Auger suppression [2] and various detector energy profiles [3,4]. Dual-band cameras covering the mid-wave (MW) IR (3–5 μm), long-wave (LW) IR (8–12 μm), and beyond [5–7] have been successfully implemented as a consequence of the band-structure’s design flexibility and the material’s robustness toward fabrication.

Measures to expand the detection range to the short-wave (SW) IR (1.7–3 μm) for the T2SL platform would provide imagery from reflected light comparable to the visible spectrum produced also by active illumination, leading to greater recognition accuracy and familiarity. Additional benefits when compared to visible imagery would include the penetration of SW radiation through smoke and fog because of its longer wavelength and the ability to detect at night under the illumination of ambient night sky radiance from the atmosphere.

Tunable SWIR superlattices were initially designed as tunneling barriers or M-barriers ($\sim 2.26\mathrm{\mu m}$ at 150 K) [8] that were placed at the junction of MWIR photodiodes to lower the dark current of conventional $\mathrm{p}\text{-}\mathrm{n}$ homojunctions in order to achieve high operating temperatures. Based on the same M-barrier architecture, homodiodes were later realized with a modified cutoff of $\sim 2\mathrm{\mu m}$ at 150 K [9]. This Letter aims to demonstrate a T2SL focal plane array (FPA) in the SWIR in conjunction with a spatially coincident MWIR channel using the designs mentioned in the references above. Under low light conditions or situations where use of a light emitter is prohibitive, the MWIR portion of the dual-band detector may prove to be more useful without requiring active illumination.

Further design considerations were needed in terms of a suitable substrate removal and etch-stop layer scheme for back-side-illuminated SWIR detectors. Previously, a lattice-matched ${\mathrm{InAs}}_{0.91}{\mathrm{Sb}}_{0.09}$ etch-stop layer [10] has been used to fully and selectively remove the GaSb substrate to reduce thermal stress and free carrier absorption. While the residual InAsSb layer by itself has proven to be less problematic for MWIR and LWIR back-side-illuminated detectors, the $\sim 4\mathrm{\mu m}$ absorption band edge would significantly reduce the irradiance impinging the SWIR channel, decimating its external quantum efficiency (QE). Therefore, we propose the insertion of a GaSb etch-stop layer following the growth of InAsSb that allows the selective removal of InAsSb using a citric-acid-based solution. Comparative external QE measurements for different etch-stop configurations will be discussed.

The material design is mentioned in brief as additional details will be published elsewhere. The bilayer InAsSb and GaSb etch stop were first grown by molecular beam epitaxy in that sequence on an n-type GaSb substrate with thicknesses of 1 μm each. Thereafter, a $\mathrm{P}\text{-}\mathrm{i}\text{-}\mathrm{N}$ SWIR homodiode with a 1 μm thick $i$-region was grown followed by an $\mathrm{N}\text{-}\mathrm{M}\text{-}\mathrm{\pi }\text{-}\mathrm{p}$ MWIR single heterodiode with a 2 μm thick $\pi$-region. The back-to-back diodes’ active regions aimed to have 50% cutoffs of 2 and 4.2 μm at 150 K using the superlattice periods previously published [8,9].

Several process evaluation chips (PECs) and an FPA were fabricated simultaneously similar to that reported before [6]. The mesas were found to be reticulated down to the highly doped $P$-contact of the SWIR superlattice. After the removal of the GaSb substrate, one PEC was left with the residual InAsSb etch stop, while another had the InAsSb layer selectively etched away. Saturated external QE spectra of these samples under back-side illumination are shown in Fig. 1 for both channels at 150 K, where QE enhancement was observed below $\sim 4\mathrm{\mu m}$ for the sample without InAsSb. Comparison of the MWIR channel found that the average external QE between 2 and 4 μm increased from $\sim 25\%$ to $\sim 45\%$, pointing to the benefit of InAsSb removal for any FPA sensitive to the MWIR spectrum. The energy positions of the Fabry–Perot oscillations seen in both MWIR spectra prove that the cavity sizes are indeed different as observed in the slopes shown within the figure’s top inset, which are inversely proportional to cavity length.

 

Fig. 1. Comparison between external QEs on back-side-illuminated and substrate-removed PECs.

Download Full Size | PDF

 

Fig. 2. Temperature-dependent current-voltage characteristics of the SWIR and MWIR diodes with NEDT performance for the MWIR channel in inset.

Download Full Size | PDF

The SWIR detectors’ external QEs were seen to change more dramatically from a peak QE of $\sim 5\%$ to $\sim 30\%$ with the InAsSb layer removed. The dip in QE seen below $\sim 1.7\mathrm{\mu m}$ was due to the absorption of the GaSb etch-stop layer. This was verified by fabricating a third PEC where the remaining GaSb etch stop was fully removed by means of dry etching, and as a result, the QE was found to be recovered below $\sim 1.7\mathrm{\mu m}$ in the near-IR range (Fig. 1). Applying a negative bias to extract the MWIR detector’s signal, however, found a significant SWIR response indicative of cross talk (not shown). The exact reason for this phenomenon in the absence of the GaSb etch stop is unknown and requires further investigation. Therefore, subsequent characterizations of the PEC and FPA have the GaSb etch stop intact but the InAsSb layer removed.

In order to assess the dual-band FPA’s prospects for high-temperature operation, a cold shield was placed in front of the PEC for temperature-variant current-voltage (I-V) measurements of pixel-sized 27 μm diodes. Shown in Fig. 2 are the dark current trend lines where the biases used for the SWIR and MWIR channels were $+50\mathrm{mV}$ and $-500\mathrm{mV}$, respectively, which were selected at signal saturation. The high bias applied to extract the MWIR signal had the majority of the voltage drop across the SWIR diode with the higher differential resistance since the current flowing through the detectors is equal in magnitude but opposite in polarity. It is also seen from the plot that the decreasing dark current behavior with temperature on both channels could be modeled by diffusion (solid line) and generation-recombination (dashed line). Both diffusion curves reduced proportionally with temperature by $\exp (-E/kT)$, where $E$ is the bandgap of the active region, $k$ is Boltzmann’s constant, and $T$ is temperature. In the case of generation recombination (G-R), the SWIR homodiode followed the trend proportional to $\exp (-E/2kT)$ for midgap traps, while the MWIR heterodiode had an activation energy that is $\sim 78\%$ of the MWIR bandgap (black dashed line) or what is likely to be close to the midgap of the M-barrier. Attributing G-R suppression to the M-barrier located near the depletion region is conceivable and can also be evidenced when comparing the I-V characteristics of MWIR homodiodes and heterodiodes at elevated biases [8].

The performance of the dual-band $320\times 256$ FPA was subsequently assessed and compared to the PEC. The highest temperature operation of the dual-band FPA was determined by the MWIR channel and provided a reasonable performance at 140 K based on the noise equivalent difference in temperature (NEDT), which had a median value of $\sim 49\mathrm{mK}$ using $f/2.3$ optics and a 10 ms integration time (Fig. 2 inset). At this temperature, the MWIR NEDT was limited by the dark current corresponding to a density of $1\mathrm{E}-5\mathrm{A}/{\mathrm{cm}}^2$ at $-500\mathrm{mV}$ on the PEC. This value fell to the measurement system limit of $\sim 2\mathrm{E}-13\mathrm{A}$ at 110 K and exhibited a median NEDT of $\sim 13\mathrm{mK}$ below this temperature, with an operability of 97% ($2\times$ median) using a 30 ms integration time.

The dark current of the SWIR detector on the PEC was seen to reach the system limit at 150 K. However, the same NEDT measurement could not be performed on the SWIR channel of the FPA based on the blackbody temperatures used previously (20°C to 30°C) nor on a noise equivalent input (NEI) measurement because of the limited temperature range up to 100°C of our blackbody system. From a noise perspective, it was observed that the dark noise measured with a cold shield and across 100 frames was found to be invariant below 150 K. The measured dark noise is assumed to be from the readout IC (ROIC) given a negligible SWIR channel dark current at those temperatures. The direct-injection ISC0903 from Indigo Systems ROIC [11] exemplified a noise distribution from $\sim 0.17\mathrm{mV}$ to $\sim 0.29\mathrm{mV}$ shown in the inset of Fig. 3, which is estimated to be over 1000 noise electrons.

 

Fig. 3. Measured temporal noise over 100 frames was in good agreement with the modeled ROIC and photon noise sources.

Download Full Size | PDF

The median temporal noise of the MWIR channel across 100 frames is shown in Fig. 3 as a function of photon flux in the $1\mathrm{E}13\mathrm{ph}·{\mathrm{cm}}^{-2}·{\mathrm{s}}^{-1}$ range at 81 K. The median ROIC noise of 0.23 mV was used along with the photon noise to estimate the total theoretical noise. The photon noise was measured with a narrowband filter centered at 3.69 μm where a QE of $\sim 30\%$ achieved a good fit with the measured noise at higher irradiances. It should be noted that, at 81 K, the cutoff of the MWIR detector decreased to $\sim 4\mathrm{\mu m}$, approaching the center wavelength of the filter. Although the ROIC was seen to be the dominant noise source, the increasing trend in noise with irradiance corresponded well with the photon noise modeled, where the measured noise is equivalent to an NEI range from $\sim 3\mathrm{E}10$ to $\sim 5\mathrm{E}10\mathrm{ph}·{\mathrm{cm}}^{-2}·{\mathrm{s}}^{-1}$.

Imaging from the dual-band SWIR and MWIR FPA using a lens with transmission from $\sim 1$ to $\sim 5\mathrm{\mu m}$ is shown in Fig. 4, demonstrating the advantages of multispectral detection. In Fig. 4(a), the SWIR imagery reveals features similar to that seen in the visible spectrum and details of an apparently “cold” soldering iron on low-heat setting. This is in contrast to the thermal image seen in the MWIR channel shown in Fig. 4(b), where the “cold” nose and the “hot” iron are observed. Pixel coincidence is demonstrated in a composite image shown in Fig. 4(c) at 81 K. A soldering iron on high-heat setting is seen to “glow” in both the SWIR [Fig. 4(d)] and MWIR [Fig. 4(e)] channels at 140 K where the 3.69 μm filter was in front.

 

Fig. 4. Imagery of the (a) SWIR, (b) MWIR, and (c) linear combination of the dual-band images are shown at 81 K. A 3.69 μm narrowband filter held in front of a soldering iron is shown at 140 K in the (d) SWIR and (e) MWIR.

Download Full Size | PDF

In summary, dual-band SWIR and MWIR detectors as well as an FPA are demonstrated on the T2SL material system, where changes in the substrate removal technique enhanced the back-side external QE of the SWIR and MWIR channels. Electro-optical characterizations found the imager operating with NEDT performances of $\sim 49\mathrm{mK}$ at 140 K, and $\sim 13\mathrm{mK}$ at 110 K in the MWIR, while the SWIR channel was limited by ROIC noise below 150 K. This Letter demonstrates the versatility of the T2SL platform in expanding its spectrum range coverage and its potential for high-temperature operation.