Electrical and optical characterisation of low temperature grown InGaAs for photodiode applications

Dilute bismide and nitride alloys are promising semiconductors for bandgap engineering, opening additional design freedom for devices such as infrared photodiodes. Low growth temperatures are required to incorporate bismuth or nitrogen into III–V semiconductors. However, the effects of low growth temperature on dark current and responsivity are not well understood. In this work, a set of InGaAs p-i-n wafers were grown at a constant temperature of 250, 300, 400 and 500 °C for all p, i and n layers. A second set of wafers was grown where the p and n layers were grown at 500 °C while the i-layers were grown at 250, 300 and 400 °C. Photodiodes were fabricated from all seven wafers. When constant growth temperature was employed (for all p, i and n layers), we observed that photodiodes grown at 500 °C show dark current density at −1 V that is six orders of magnitude lower while the responsivity at an illumination wavelength of 1520 nm is 4.5 times higher than those from photodiodes grown at 250 °C. Results from the second set of wafers suggest that performance degradation can be recovered by growing the p and n layers at high temperature. For instance, comparing photodiodes with i-layers grown at 250 °C, photodiodes showed dark current density at −1 V that is five orders of magnitude lower when the p and n layers were grown at 500 °C. Postgrowth annealing, at 595 °C for 15 min, on the two wafers grown at 250 and 300 °C showed recovery of diode responsivity but no significant improvement in the dark current. Our work suggests that growth of the cap layer at high temperature is necessary to maintain the responsivity and minimise the dark current degradation, offering a pathway to developing novel photodiode materials that necessitate low growth temperatures. The data reported in this paper is available from the ORDA digital repository (DOI: 10.15131/shef.data.12562346.v1).


Introduction
Short-wave infrared (SWIR) imaging in the wavelength range of 1 to 3 µm is becoming a topic of great interest in applications such as remote sensing, night imaging, thermal imaging and free space optical communications [1]. Well-established III-V semiconductor detectors capable of SWIR detection include extended-wavelength InGaAs [2], InP-based type-II superlattice (T2SL) [3], GaSb-based T2SL [4] and InAs [5]. The most mature commercial option is the extended InGaAs. However, large lattice mismatch between extendedwavelength InGaAs (of high Indium composition) and InP substrate gives rise to dislocations in the InGaAs crystal and non-ideal optoelectronic properties [6]. The superlattices offer bandgap tuning capability but are more difficult to grow due to complicated interfaces. InAs is a simple binary compound that can circumvent growth related issues in extended InGaAs and T2SL but has an inherently higher leakage current due to its narrower bandgap. In addition to III-V semiconductors, II-VI semiconductors such as PbS and HgCdTe (with appropriate Hg composition [7]) can also be operated in the SWIR wavelength range. The latter incurs high production costs and is incompatible with a global initiative to phase out mercury containing compounds from consumer products [8]. Therefore, having a SWIR detector material compatible with the InP platform, for example utilising dilute bismide [9,10], or nitride [11] incorporated into InGaAs, is attractive. This is due to their relatively low cost and mature processing but also in terms of compatibility to integrated InP photonics systems [12].
Nitrogen (N) [13] or Bismuth (Bi) [14] can be added to III-V compound semiconductors to induce a reduction in bandgap and consequently a longer cutoff wavelength. A significant bandgap reduction per percent of N is attainable [15], offering potentials as 1 eV bandgap material for GaAs-based multijunction solar cells [16] and photodiodes operating in the 1.3 µm telecommunications band [17]. Similarly, dilute Bi incorporation has been demonstrated in III-V semiconductors such as GaAs [18,19], In 0.53 Ga 0.47 As [20] (hereinafter referred to as InGaAs) and InAs [21].
The incorporation of N and Bi atoms requires epitaxial growth at temperatures lower than used for the conventional growth of generic GaAs or InGaAs alloys. Typically, growth temperatures between 375 to 500 • C [22][23][24][25][26][27] and 255 to 400 • C [28][29][30][31] has been used for dilute nitride and bismide alloys respectively. In comparison, conventional InGaAs growth via molecular beam epitaxy (MBE) is usually performed at temperatures above 500 • C [32,33]. It is, however, known that III-As semiconductors grown at low temperature suffer from loss of stoichiometry due to the incorporation of excess As [34,35]. As a result, increased carrier concentration and decreased mobility are commonly reported in low temperature grown InGaAs [36][37][38].
While there has been significant effort in growth and material characterization of In y Ga 1-y As 1-x Bi x , the effects on dark current and responsivity brought upon by the low growth temperature is not well understood [39]. Similarly, a systematic investigation on the low temperature grown InGaAs devices using MBE has not yet been undertaken. This paper aims to provide insights into the influence of growth temperature on fundamental photodiode characteristics in InGaAs photodiodes with no Bi incorporation in preparation for future growth optimization for developing semiconductor materials grown at low temperature, such as Bi containing alloys.

Experimental details
A total of seven InGaAs wafers were grown on lattice matched n-type (001) InP substrates using MBE with growth temperatures ranging from 250 to 500 • C. For each wafer, a 500 nm n-type InGaAs layer was grown on the substrate followed by a 1000 nm undoped InGaAs layer and a 500 nm p-type InGaAs as the top cap layer. The growth rate used is 1.2 µm/hour and p and n layer doping concentrations are nominally 5 × 10 18 cm −3 . Two sets of wafers (A and B) were grown. For a given wafer in set A, all layers were grown at the same temperature. In set B, the p and n layers were grown at 500 • C while the i-layer was grown at a lower temperature. Therefore the i-layers in set B wafers can be considered to have in-situ annealing when the p layer was grown at 500 • C. Table 1 summarises growth details for each wafer and the growth temperatures are calibrated with a kSA BandiT bandedge thermometry system with an uncertainty of ± 10 • C.
To study the effects of postgrowth annealing on photodiode performance, ex-situ annealing on two samples from set A, A300-A and A250-A (annealing denoted by a '-A' suffix) were performed. A thin encapsulating layer of SiO 2 was deposited using plasma-enhanced chemical vapour deposition prior to annealing at 595 • C for 15 min in a nitrogen-rich environment using a rapid thermal annealer.
Standard photolithography process was used to fabricate mesa diodes of varying diameters (70 to 420 µm) from each wafer. Devices were etched with a 1:8:80 mixture of sulfuric acid (95%), hydrogen peroxide (30%) and de-ionised water. Ti/Au were deposited as the p and n-type contacts by thermal evaporation.
On wafer dark current-voltage (I-V) and capacitancevoltage (C-V) measurements were performed at room temperature using a picoammeter (HP Model 4140B) and an LCR meter (HP Model 4275A), respectively. Temperature dependent I-V measurements were performed using a Janis ST-500 cryogenic probe station. For I-V measurements, devices with diameters of 70, 120, 220 and 420 µm were characterised. Responsivity measurements at room temperature were performed by illuminating the top of mesa diodes of 420 µm diameter with 1520 nm wavelength light from a He-Ne laser. The resultant photocurrent was measured with a lock-in amplifier (Stanford Research Systems Model SR830) by a phase sensitive detection method. For electroluminescence (EL) measurements, mesa diodes of 420 µm diameter were wire bonded on 3 mm TO headers. A Keithley 2400 source-meter unit was used for continuous wave characterization and the EL spectra were collected using a 400 µm diameter core multimode optical fibre and a NIRQUEST-512-1.9 Ocean Optics spectrometer operating in 1100-1900 nmss spectral range.

Results
I-V measurements from 3-5 devices of each diameter from all wafers showed excellent uniformity. The excellent agreement in reverse dark current densities (normalising current to device area) of devices, indicates dominance of bulk current mechanisms as seen in examples from wafers A500 and A300 in figure 1(a). In figure 1(b), reverse dark current densities of all wafers plotted against voltage are presented for comparison. Results from set A wafers show significantly increased dark current density as the growth temperature decreases. The dark current density at −1 V changes by more than six orders of magnitude, increasing from 7 × 10 −6 Acm −2 for A500 (grown at 500 • C) to 1.6 Acm −2 for A250 (grown at 250 • C). On the other hand, diodes from set B show only a slight degradation of dark currents, within an order magnitude, as the i-layer growth temperature decreases from 400 to 250 • C. Figure 2(a) plots the dark current densities at a reverse bias of −1 V from all devices to accentuate the impact of the growth temperatures. As can be seen from figures 1 and 2(a), the dark current rises rapidly with reduced growth temperature in set A. Comparing diodes grown at 250 • C, the dark current density sees a significant decrease from 1.6 Acm −2 to 4.4 × 10 −5 Acm −2 by changing the growth temperature used for the p and n layers from 250 • C (A250) to 500 • C (B250). Following a postgrowth anneal, diodes from A300-A and A250-A show lower dark current density, as shown in figure 1(b) and 2(a). The improvement in dark current is more pronounced at higher reverse bias as shown in figure 1 The zero bias responsivity at 1520 nm wavelength of all wafers are shown in figure 2(b). Diodes from set A have responsivity values decreasing from 0.27 to 0.06 A W −1 as the growth temperature is reduced from 500 to 250 • C. On the other hand, diodes from set B display less variance in responsivity, having values of 0.25, 0.30 and 0.23 A W −1 for i-layers grown at 400, 300 and 250 • C respectively. Annealing of wafers A300 and A250 gives rise to a pronounced increase in responsivity to 0.28 and 0.24 A W −1 in wafers A300-A and A250-A, respectively.
The capacitance data as a function of reverse voltage for all wafers are shown in figure 3(a). Excessive dark currents from diodes of A250 and A250-A prevented accurate C-V measurements and, therefore, are not presented in figure  3(a). Diodes from wafers A300 and A300-A show higher capacitance values that decrease with reverse bias whereas diodes from other wafers have relatively constant capacitance after a reverse bias of −3 V. The relatively constant capacitance indicates the i-layer is fully depleted in A500, A400, B400, B300 and B250. By modelling the C-V results, i-layer doping concentration between 1 to 2 × 10 15 cm −3 were obtained for the fully depleted diodes. Doping concentrations that are 20 × higher were found in sample A300 and A300-A resulting in the larger capacitance measured. The extracted doping concentration values are shown in table 1.
The room temperature EL spectra of wafers A500, A300 and A250 are plotted in figure 3(b). The EL intensity is significantly reduced in wafers A300 and A250 in contrast to wafer A500, requiring a scaling factor of 50 for comparison. The peak intensity wavelength from the EL spectra shifts from 1667 nm to 1710 nm and the full-width at half maximum increases from 45.7 to 61.1 meV as the growth temperature reduces from 500 to 250 • C.

Discussion
I-V results suggests that in-situ annealing (during the growth of the p-layer at high temperature in set B wafers) minimises the dark current degradation while ex-situ annealing can produce a small reduction in the dark current. The bulk reverse dark current mechanisms at low electric fields include minority carrier diffusion and generation-recombination (G-R), given by where q, n i and W are the elementary charge, carrier concentration and depletion width respectively while N D , N A , D p , D n , L p , L n and τ are the donor concentration, acceptor concentration, hole diffusion coefficient, electron diffusion coefficient, hole diffusion length, electron diffusion length, and carrier G-R lifetime, respectively. At higher electric fields band-to-band tunnelling (BBT) and trap-assisted tunnelling (TAT) currents can be significant in narrow bandgap materials. They are given by J TAT = qN T W c p p 1 wN c c p p 1 + wN c (4) where the tunnelling rate, wN c , is given by where m * , ξ, V, h, a, and E g are the electron mass, electric field, applied bias voltage, Planck's constant, tunnelling parameter, and bandgap, respectively. N T , c p , p 1 , M and E T are the trap concentration, hole capture coefficient, hole thermal emission rate, matrix element of trap potential energy, and trap energy level, respectively. Poor crystal quality due to large defect densities (in the form of As point defects) commonly associated with low temperature grown InGaAs [40,41]. This can give rise to reduced minority carrier diffusion length and reduced carrier lifetime. At low bias close to −1 V, the increase in dark current in B400, B300 and B250 can be attributed to an increase in J G−R caused by reduced carrier lifetime as the growth temperature of the i-layer is reduced. When the growth temperature for the p and n layers is also reduced in set A diodes, there is a further increase in J diff arising from the reduced minority carrier   diffusion lengths as observed from the higher dark current in A400 compared to B400.
The trends in dark current for A300 and A250 cannot be fully explained using J diff and J G−R which do not increase strongly with voltage. The rapid increase in dark current can be explained qualitatively by fitting the sum of J diff , J G−R , J BBT , and J TAT to the dark current of A250. We extracted an activation energy of 0.06 eV using dark current data at −1 V from 180 to 295 K as shown in figure 4(a). We interpret this as E T to calculate J TAT . In the absence of an accurate value of M, we   assume the value used in HgCdTe [42], where and use N T as the fitting parameter to fit the measured dark current as shown in figure 4(b). We found that J TAT produces a good fit at bias voltages below −0.4 V, suggesting its dominance at low bias in A250. At higher bias voltages, the more rapid increase in dark current can be explained by the onset of J BBT . A good fit to the measured dark current was obtained by assuming the i-layer of A250 has a doping concentration of 5 × 10 17 cm −3 (C-V data cannot be extracted accurately due to high dark current). The parameters used for dark current modelling are listed in table 2. We noted that growing the p and n layers at high temperature will preserve the low background doping concentration in the ilayers of B250 and B300, such that tunnelling current is not significant.
Our hypothesis of increase in defects that leads to the significant increase in dark current in A300 and A250 is supported by EL data. The EL spectra of wafers grown at low temperatures, A300 and A250, show drastically reduced peak intensity. These results suggest the dominance of non-radiative recombination causing the decrease of EL efficiency, which is most likely due to defect-related recombination which increases as the growth temperature decreases [43]. Similarly, these defects reduce the minority carrier diffusion length that leads to lower responsivity in diodes from set A corroborating the trend observed in the dark current. As light absorption follows an exponential decay function, we expect an illumination wavelength of 1520 nm to be absorbed in the p, i and n layers. Poor crystal quality in all three low temperature grown layers could explain the lower responsivity in set A diodes (due to reduced minority carrier diffusion lengths) as compared to set B diodes. The total photocurrent produced in a p-i-n photodiode is the sum of the drift current due to carriers generated in the depletion region, J drift , and the diffusion current from carriers generated in the p and n layers, J p and J n respectively, and are given as J p = qD n ∂n p ∂x where Φ 0 and α are the photon flux and absorption coefficient, respectively. It is clear that reduced minority carrier diffusion lengths will reduce the photocurrent. Hence A250 and A300 have lower responsivity values than B250 and B300. Devices from the ex-situ annealed wafers, A300-A and A250-A, exhibit improved responsivities over devices from unannealed wafers. From the C-V plots, similar capacitances of A300 and A300-A suggests similar depletion widths in these diodes. Therefore, the increased responsivity in annealed wafers is not caused by changes in J drift but can be attributed to the increased minority carrier diffusion lengths after high temperature annealing. The increase in minority carrier diffusion lengths after annealing also produces a very small reduction in the diffusion dark current as expected from equation (1) and observed in figure 2(a). However we found that the diffusion current is several orders of magnitude lower than our measured dark current in A250 as shown in figure 4(b). Therefore the small increase of minority carrier diffusion length does not produce notable change in the dark current.
We believe that the increase in the minority carrier diffusion length can be explained as follows. As point defects nucleate to form precipitates following high temperature anneals above 500 • C as observed in low temperature grown InGaAs [40,41]. The electron capture cross sections are reduced in precipitates as compared to antisite defects [44]. Consequently, an increase in minority carrier diffusion length, can be obtained [44] leading to higher responsivities in ex-situ annealed wafers A250-A and A300-A.
For wafers with i-layer grown at 250 and 300 • C, both insitu and ex-situ annealing increase responsivity to levels comparable to diodes grown at conventional temperature (A500 diodes). However, postgrowth ex-situ annealing does not offer significant improvement to the dark current. This is consistent with reports for GaInNAs(Sb) laser diodes, where in-situ annealing was found to result in superior recovery in performance over ex-situ annealing [45]. Therefore we believe a high temperature cap layer growth is needed to improve device performance in materials that necessitate low growth temperatures.

Conclusion
Experimental data of InGaAs p-i-n photodiodes grown at different temperatures has been presented. Decreasing the growth temperature from 500 to 250 • C in all p, i and n layers leads to a six orders of magnitude increase in dark current at −1 V and 4.5 times lower responsivity at 0 V. The poor diode properties are attributed to defects originating from excess As incorporation in low temperature growth wafers as reported in literature and supported by results from C-V and EL measurements. For wafers grown at 250 and 300 • C, postgrowth ex-situ annealing at 595 • C improves the diode responsivity but fails to produce a significant improvement in dark current. We found that raising the growth temperature of the p layer to 500 • C during growth (in-situ annealing) after the i-layer was grown at a lower temperature provides much better responsivity and dark current values. A five orders magnitude drop in dark current and 3.8 times increase in responsivity is observed when a high growth temperature of 500 • C is employed for the p and n layers in diodes with i-layer grown at 250 • C. Growing the p and n layers at high temperature was also found to maintain the background doping concentration in the i-layer, hence minimising the tunnelling current.
These results suggest that poor device performance due to reduced minority carrier diffusion lengths and increase in the background doping concentration caused by low growth temperature can be avoided by growing the p and n layers at a high temperature.