Two-colour In0.5Ga0.5As quantum dot infrared photodetectors on silicon

An InGaAs quantum dot (QD) photodetector is directly grown on a silicon substrate. GaAs-on-Si virtual substrates with a defect density in the order of 106 cm−2 are fabricated by using strained-layer superlattice as dislocation filters. As a result of the high quality virtual substrate, fabrication of QD layer with good structural properties has been achieved, as evidenced by transmission electron microscopy and x-ray diffraction measurements. The InGaAs QD infrared photodetector is then fabricated on the GaAs-on-Si wafer substrate. Dual-band photoresponse is observed at 80 K with two response peaks around 6 and 15 μm.


Introduction
Infrared photodetectors have a number of applications, from civilian to military uses, including night version, medical inspection, optical communications and missile tracking [1,2]. HgCdTe (MCT) infrared photodetectors, owning to its tuneable bandgap and high quantum efficiency, have dominated the market for years [3]. However, the low production yield and high cost become the bottleneck for further upgrading the HgCdTe-based infrared imaging systems [4]. Over the past two decades, the demand for the third-generation infrared imaging systems brings interests in quantum structures [5][6][7][8][9]. Particularly, III-V quantum structures have been considered as a candidate that could compete with MCT in terms of performance and cost. For instance, quantum dots (QDs) confine charge carriers in all three dimensions, which promises to reduce dark current, increase operating temperature, and improve detectivity [10,11]. Moreover, the recent demonstration of the direct growth of III-V QDs on Si substrates has attracted a lot of attentions [12][13][14]. The monolithic integration of III-V photodetectors with silicon readout integrated circuits (ROICs) opens new opportunities for enhanced functionality and reduction in manufacturing cost [15,16]. Despite the great promises of monolithic integration of III-V photodetectors and Si ROICs, the epitaxy growth of III-V materials on Si substrates faces great challenges. Due to the large lattice mismatch and difference in thermal expansion coefficient, a high density of threading dislocations (TDs) tends to form at the interface between the Si substrate and III-V epitaxial layers. The TDs propagate into the device active region lead to severe performance degradation. It has been demonstrated that the TD density can be reduced to the extent that the impact on device performance is marginal [12]. Particularly, compared other quantum structures, e.g. quantum wells, QDs are less sensitive to dislocations due to the localization of charge carriers [17]. Although the quantum efficiency of QD infrared photodetectors are yet to be Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. improved, the high tolerance of QDs may lead to monolithic integrated large format and low cost QDIPs on Si substrates.
Recently, we have demonstrated InAs/GaAs QD photodetectors directly grown on silicon substrates with a single band photoresponse (∼5-8 μm) [13]. For many applications, multicolour detection is desired. In this paper, a dual-band InGaAs/GaAs QDIP was grown directly on Si by molecular beam epitaxy (MBE). By using InGaAs QDs instead of InAs QDs, multiple intersubband transitions in shallowly confined QDs are possible. The photoresponse covers both atmosphere transmission windows with detection peaks in the mid-wave infrared (∼6 μm) and long-wavelength infrared spectra (∼15 μm).

Experimental details
An InGaAs/GaAs QDIP has been directly grown on Si by a solid-source Veeco Gen-930 MBE. Epitaxial growth was taken place on an n-doped Si (100) substrate with 4°offcut towards the [01-1] plane. The device structure is illustrated in figure 1(a). The Si wafer was pre-treated in ultrahigh vacuum over 900°C for 30 min to remove the surface oxides. Immediately after de-oxidation, the wafer was cooled down for growth. The initiation of growth started with the deposition of a 5 nm AlAs nucleation layer on the Si substrate by migration-enhanced epitaxy at 370°C. Subsequently, a 1000 nm thick GaAs buffer was deposited followed with four sets of In 0.15 Ga 0.85 As/GaAs strained-layer superlattices (SLSs) for defect reduction. In 0.50 Ga 0.50 As/GaAs QDIP epilayers were grown in a sequence as follows: 500 nm n-type GaAs (Si dopant, 2×10 18 cm −3 ) bottom contact layer, 80 nm undoped GaAs spacer layer, 20 periods of In 0.50 Ga 0.50 As/GaAs QDs separated by 50 nm GaAs spacers, and 300 nm n-type GaAs top contact layer (Si dopant, 2×10 18 cm −3 ). The QD region was directly doped with about two electrons per dot. The x-ray diffraction measurements are performed by using a Jordan Valley Bede D1 system. The surface morphology of the sample is measured by a Veeco Dimension-3100 atomic force microscopy (AFM) in tapping-mode. Structural analysis was performed by crosssectional transmission electron microscopy (TEM) at 300 keV using an FEI Titan 80-300 TEM fitted with a CEOS image corrector.
Detectors in circular mesa structure were fabricated by standard photolithography and wet chemical etching techniques. Ni/GeAu/Ni/Au metals were deposited on both bottom and top n + -GaAs contact layers by a thermal evaporator. The metals were annealed to form ohmic contacts by a rapid thermal processing system. Spectral response of the detectors was recorded by a Bruker Equinox-55 Fourier transform infrared spectrometer at 80 K. The device was configured normally to a glow-bar broadband infrared light source. Photocurrent signal was amplified using a SR570 low-noise current preamplifier that also applies a bias voltage to the device and then converted to spectral response by Fourier transform.  micrographs show a high density of TDs formed at the GaAs/ Si interface, as shown figure 1(c). Some of the TDs were selfannihilated within the 1000 nm GaAs buffer, but about 10 9 cm −2 TDs propagate into the following epilayers and reach the InGaAs/GaAs SLSs. In the low misfit In 0.15 Ga 0.85 As/GaAs layers, gliding of existing mobile TDs assists strain relaxation via introducing misfit dislocations, of which the strain fields enhance the motion of other TDs [18]. As a result, these InGaAs/GaAs SLSs can act as TD filters and reduce the density of TDs by a few orders. The density of TDs is reduced to the level of 10 6 cm −2 , estimated by crosssectional TEM measurements, after the fourth set of SLSs. As shown in figures 1(c)-(e), only a low defect density is visible in the device region with the assistance of InGaAs/ GaAs SLSs.

Results and discussion
Symmetric (004) x-ray ω-2θ scans were performed. The results are plotted in figures 2(a) and (b), showing the ω-2θ curves for the Si (004) and GaAs (004) reflections, respectively. Figure 2(a) shows a full width at half maximum (FWHM) as high as 548 arcsec of the GaAs (004) peak in the ω-2θ curve obtained from the Si (004) plane. The FWHM is directly related to the dislocation density, and the high value of the FWHM indicates a high density of dislocations. However, it should be noted that measured GaAs peak is located at around −5350 arcsec, while an unstrained GaAs layer has a (004) symmetric peak separation of −5538 arcsec with respect to the Si (004) peak [19]. This position difference indicates that the GaAs layer is not fully relaxed and under compressed strain. As the device is grown on a thick buffer (>3 μm), the GaAs layers in the active region are expected to be fully relaxed. Therefore, the probed GaAs may be the highly defective GaAs materials at the GaAs/Si interface region. The ω-2θ curves measured nearly the unstrained GaAs (004) reflection is shown in figure 2(b). The FWHM of the GaAs (004) peak is about ∼226 arcsec, which indicate a high quality of the fully relaxed GaAs layers. The dislocation density ρ d is estimated to be ∼4×10 7 cm −2 , based on the where b is the Burgers vector [20]. The estimated dislocation density is much higher than the TEM measurements because the highly defective GaAs buffer layers also contribute to the FWHM of the GaAs (004) peak due to the deep penetration depth of x-rays. In addition, clear satellite peaks of QD layers are observed up to the fifth order, which confirms the high quality of the epilayers in the device region. The rather broad peaks separated by about 1000 arcsec are the satellite peaks of InGaAs/GaAs strained superlattice layers. To further visualize the quality of the grown material, AFM image is shown in figure 2(c). The AFM line profiles shows surface undulation up to 5 nm, which is due to strain relaxation of lattice-mismatched heteroepitaxy [21]. Additionally, the anisotropic strain relaxation of GaAs causes the distinct surface undulation in the orthogonal á ñ 011 directions [21]. Nonetheless, a relatively smooth surface free of antiphase domain has been achieved with a root mean square roughness of 1.4 nm measured from an area of 5 μm×5 μm.
The temperature-dependent dark current-voltage characteristics for an InGaAs QDIP is shown in figure 3(a). The dark current increases rapidly with increase in the bias voltage due to field-assisted tunnelling of charge curriers. Despite similar dislocation density as the previous study, the InGaAs QDIP shows a higher dark current density (3.95×10 −3 A cm −2 at 60 K and 7.55×10 −3 A cm −2 at 80 K, both at 1 V) [13]. To gain further insight into the electronic transport mechanism of the device, the dark currents measured under +0.1 V bias from 50 to 300 K are plotted in figure 3(b). In sharp contrast to our previous study [13], the dark current increases rapidly even at low temperatures <100 K, at which thermionic emission is suppressed. As shown in figure 3(b), two activation energies, E a1 =31 meV and E a2 =57 meV, can be extracted from the dark currents measured at the low-temperature region. The two activation energies correspond to the longitudinal optical phonon energy E LO and 2E LO , respectively. A dominated phonon-assisted tunnelling may be the reason for the high dark current at low temperatures. With further increasing temperature, the increase in dark current is mainly caused by the thermionic emission of charge carriers. By fitting with Arrhenius equation, the thermal activation energy is calculated as E a3 =219 meV. This energy corresponds to the transition energy from the highest state occupied by electrons to the conduction band continuum. The QDs are only doped with 2e/dot, and the electrons are just enough to fully occupy the ground state of the QDs. Therefore, the dark current activation energy E a3 matches the energy difference between the ground state of the QDs and the GaAs barrier conduction band edge. Figure 3(c) shows photoresponse spectra measured from the InGaAs QDIP at 80 K. The photoresponse spectra show a dual broadband response from 4 up to 17 μm, covering both the atmosphere transmission windows in the infrared region. Similar to the dark current measurement, photoresponse spectra acquired at both negative (+1 V) and negative (−1 V) voltages show minor difference due to slight structural asymmetry of QDs. The mid-wave infrared band has a peak around ∼6.1 μm (203 meV). The transition energy matches well with the thermal activation energy measured from the dark current measurements, indicates optical transitions from the bound states in the QDs to continuum states. The long-wavelength band with Δλ/λ p ∼10% may be attributed to bound-to-bound transitions. To gain further insight into the electronic structure of the QDs, steady state photoluminescence (PL) and transient PL were carried out at 10 K. As shown in figure 3(d), there are two dominated peaks around 1.25 and 1.22 eV for both low (black curve) and high excitations (blue curve). As both peaks appear at relatively low excitation, they may come from the ground states of QDs with two different sizes. The bimodal distribution of QDs is also evidenced in the transient PL. As shown in figure 3(d), optical transitions at the energies of both peaks show similar long decay time over 1 ns, indicating optical transitions from ground states. At high excitation, two additional PL peaks are observed at 1.29 and 1.31 eV. As the peaks at 1.29 and 1.31 eV have distinctly different decay times, they are assigned as transitions from the first excited state and the second excited state. Assuming the hole levels are closely packed, the energy separation between the ground states and the second excited state to range from 60 to 90 meV, which correspond to 13.7-20.6 μm. It should be noted that the PL measurements were carried out at 10 K. At 77 K, it would be expected that the ground state of the smaller QDs was depopulated and the intersublevel transition was mainly from the large QDs with a lower ground state. Considering the slight reduced energy bandgap at 77 K, the transition energy of 90 meV matches with the transition energy of the longwavelength photoresponse band, 72.9-88.6 meV. Additionally, assuming 70% of the bandgap difference is in the conduction band, the energy separation from the electron ground state to conduction band continuum calculated from the PL data is about 189-210 meV, which matches very well with the mid-wavelength photoresponse peaked at 6 μm and the dark current activation energy. This confirms that the transition of the mid-wavelength band between the ground state to continuum states.
The voltage-dependent photoresponse measurement is carried out, as shown in figure 4(a). The detectors were configured normal to the unpolarized IR source. The intensity of both bands increases monotonically with bias voltage. The behaviour can be understood by the capture rate of photoexcited carriers. With increasing electric field, the electron velocity increases and at the same time, the external field lowers the build-in potential barriers and decreases the capture probability. The intensity of the long-wavelength infrared band is much lower than that of the mid-infrared one. The lower intensity can be explained by a higher potential barrier and hence lower escape rate for photo-excited carriers. This is in agreement with our assumption of bound-to-bound transitions for the long-wavelength infrared band. The voltagedependent photocurrents of both bands are compared in figure 4(b). The integrated photocurrents of both band show similar dependence with bias voltage at high electric field. However, at low electric field, the photocurrent of the midwave infrared band increases much faster. The photocurrent from QDIPs can be written as [22] where δ is electron capture cross-section coefficient, q is electron charge, n is the average number of electrons per QD, N QD is the areal QD density, Φ is the photon flux density, and p k is the capture probability. The capture probability is related to potential barrier height as follows [22,23] where τ trans and τ life are the transit time and the capture time, V b is the potential barrier height, and k B is the Boltzmann constant. From equations (1) and (2), it can be the photocurrent of the long-wavelength infrared band is due to a higher potential barrier, in agreement of the assumption of bound-to-bound transitions. When the bias voltage is large enough to overcome the potential barrier, the photocurrents of both bands increase exponentially with the same trend, as shown in figure 4(b). To increase the long-wavelength band response, it may require design of low energy transitions from bound-to-quasi-band in the QDs.

Conclusion
A dual-band photoresponse InGaAs/GaAs QDIP is directly grown on the Si substrate by MBE. The low misfit strained InGaAs/GaAs layers significantly reduce the dislocation density by about three orders to the level of ∼10 6 cm −2 . TEM and XRD measurements confirms the high crystal quality of the III-V epitaxial grown on Si. The dark current is measured to be 7.55×10 −3 A cm −2 at 80 K and 1 V, much higher than previous work, suggesting further optimization of QDs is required. Given the relatively low TD density, the presence defective QDs may attribute to the dominated phonon-assisted tunnelling dark current. Photocurrent measurements show two photoresponse bands covering both mid-wave infrared and long-wavelength infrared regions, from 4 to 17 μm. The InGaAs QDIP may be promising for monolithically integrated multi-spectral imagers on silicon.
The authors acknowledge the support of the EPSRC grants (EP/L018330/1 and EP/P006973/1) and National Science Foundation of the US.