Thin Al1−xGaxAs0.56Sb0.44 diodes with extremely weak temperature dependence of avalanche breakdown

When using avalanche photodiodes (APDs) in applications, temperature dependence of avalanche breakdown voltage is one of the performance parameters to be considered. Hence, novel materials developed for APDs require dedicated experimental studies. We have carried out such a study on thin Al1–xGaxAs0.56Sb0.44 p–i–n diode wafers (Ga composition from 0 to 0.15), plus measurements of avalanche gain and dark current. Based on data obtained from 77 to 297 K, the alloys Al1−xGaxAs0.56Sb0.44 exhibited weak temperature dependence of avalanche gain and breakdown voltage, with temperature coefficient approximately 0.86–1.08 mV K−1, among the lowest values reported for a number of semiconductor materials. Considering no significant tunnelling current was observed at room temperature at typical operating conditions, the alloys Al1−xGaxAs0.56Sb0.44 (Ga from 0 to 0.15) are suitable for InP substrates-based APDs that require excellent temperature stability without high tunnelling current.


Introduction
Avalanche photodiodes (APDs) are widely used in optical communication, imaging and sensing applications that require detection of high speed and/or weak optical signals. When the noise is dominated by the amplifier, the APD improves the system signal to noise ratio, by amplifying the photo-generated current through the impact ionization process. A cascade of impact ionization events can transform a single electron (or hole) into an avalanche of new carriers, leading to a large external current. In most semiconductor materials, the ionization process is stochastic and hence a mean avalanche gain, M, is measured in practice.
Avalanche gain versus reverse bias characteristic, M(V), of an APD can be highly temperature-sensitive; therefore, a control circuit is essential for accurately adjusting the reverse bias (or the operating temperature) to maintain M. When the APD is operated in the Geiger mode, the temperature dependence of avalanche gain can lead to significant changes in the breakdown probability, which in turn governs the photon detection probability and the dark counts. Hence, APDs with temperature-insensitive M(V) characteristics that require minimal temperature stabilization control are highly desirable.
Temperature sensitivity of an APD is characterized by the temperature coefficient of its avalanche breakdown voltage, C bd = V bd / T, where V bd and T are changes in breakdown voltage and temperature, respectively. For a given avalanche material, C bd increases with avalanche region width, w [1,2]. This is illustrated by C bd values of 6 and 11 mV K −1 reported for InP diodes with w = 130 and 250 nm, respectively [1]. Although reducing w leads to reduced C bd , there exists a lower limit for w, imposed by the band-to-band tunnelling current from the avalanche region, which is subjected to high electric field. For example, for APDs used in 10 Gbit s −1 optical communication receivers, the approximate lower limits for avalanche materials InP and InAlAs are w = 180 and 150 nm, respectively [3].
The material AlAs 0.56 Sb 0.44 (AlAsSb), which is lattice matched to InP substrates, has been investigated as avalanche material for APDs grown on InP substrates [4]. In [4], an AlAsSb diode with w = 80 nm exhibited the smallest C bd value (0.95 mV K −1 ) reported in the literature for comparable w (where the breakdown mechanism is dominated by avalanche breakdown in all data compared). Besides the low C bd value, AlAsSb also exhibit very low excess noise factor (F = 2.15 at M = 10) [5], comparable to that of silicon APD. While exhibiting low excess noise and C bd value, the high Al composition is vulnerable to oxidization, which gives higher leakage currents [6]. Incorporating Gallium into AlAsSb was reported to significantly reduce the oxidation rate [7]. More recently, alloys of Al 1−x Ga x As 0.56 Sb 0.44 (x = 0, 0.05, 0.1, 0.15), also lattice-matched to InP substrates, were found to have substantially lower room temperature surface leakage current for alloys with x = 0.1 and 0.15 [6]. From AlAs 0.56 Sb 0.44 to Al 0.85 Ga 0.15 As 0.56 Sb 0.44 , the bandgap reduces very slightly from 1.64 to 1.56 eV [6]. Crucially, the reduced bandgap did not lead to significant band-to-band tunnelling current, suggesting that Al 1−x Ga x As 0.56 Sb 0.44 could be an attractive material for low tunnelling, low surface leakage avalanche region. Thus, a thin Al 0.85 Ga 0.15 As 0.56 Sb 0.44 avalanche layer was used to achieve an APD with very high gain-bandwidth product of 424 GHz [8]. For a more comprehensive assessment of Al 1−x Ga x As 0.56 Sb 0.44 alloys, lattice matched to InP, it is important to consider their temperature sensitivity of dark current and avalanche breakdown, which has not been reported. Note that these alloys have larger bandgap than, and distinct band structures from, the Al 1−x Ga x As 1−y Sb y alloys (x = 0.40-0.65 and y = 0.035-0.054) grown lattice-matched to GaSb substrates [9].
In this work, we report the temperature dependence of M(V), from which the C bd values for Al 1−x Ga x As 0.56 Sb 0.44 p + in + diodes, with x = 0, 0.05, 0.1 and 0.15, were obtained. The diodes have nominal w = 100 nm, a useful value for designing high-speed APDs. Levels of dark currents from Al 1−x Ga x As 0.56 Sb 0.44 were also compared with those from InP and InAlAs, current avalanche materials in APDs latticed-matched to InP substrates.

Experimental details
The Al 1−x Ga x As 0.56 Sb 0.44 (Al 1−x Ga x AsSb) p + in + diode wafers with x = 0, 0.05, 0.1, 0.15 used in this work are identical to those described in [6]. The wafers were grown by molecular beam epitaxy on InP substrate using Be and Si as p-type and n-type dopants, respectively. Structure details of the wafers are shown schematically in figure 1a. The wafers were fabricated into circular mesa diodes with nominal diameters, D of 400, 200, 100 and 50 µm using standard photolithography and wet chemical etch. The p-and n-metal contacts were formed by Ti-Au.
Capacitance-voltage (C-V) measurements on each wafer were performed at room temperature. Fittings to the C-V characteristics then gave estimated values of w as well as the approximate doping concentrations in the claddings and the i-region [6]. Secondary ion mass spectroscopy data (p-and n-type dopant atoms profiles) from the x = 0.15 wafer, as shown in figure 1b, support the information obtained from C-V characteristics. The deduced w values range from 110 to 116 nm, slightly thicker than the nominal 100 nm, as summarized in table 1.
Avalanche gain measurements were carried out at temperatures of 77, 150, 200, 250 and 297 K, using a Janis ST-500 low-temperature probe station. The measurements relied on phase-sensitive detection of photocurrent versus reverse bias, I ph (V), so that the photocurrent data were unaffected by background radiation, background noise and diode's dark current. A 542 nm wavelength He-Ne laser    To assess uniformity of V bd value across a given wafer and to support the V bd value derived from M(V), characteristics of reverse dark current versus reverse bias voltage, I-V, was measured from seven devices (three of D = 400 µm, two of D = 200 µm and two of D = 100 µm) from each of the wafers at the same temperature points. These measurements were performed using the low-temperature probe station too, and they confirmed that the breakdown voltage is consistent across a given wafer.

Results
The data of M(V) for all four wafers from 77 to 297 K are shown in figure 2. Results shown were obtained from diodes with D = 400 µm. For each alloy, at a given reverse bias, decreasing the temperature increases M, which is the typical temperature dependence for avalanche multiplication in most wide bandgap semiconductor materials. Data of 1/M versus V and linear fittings to extract V bd values from one of the devices on all four alloys are shown in figure 3.
The deduced V bd values are plotted against temperature for all the alloys in figure 4. The data can be described with linear fittings (with gradients of C bd ) over the range of temperature studied. The values of C bd ranging from 0.86 to 1.08 mV K −1 are listed in table 1. The C bd values for all four alloys are similar, within accuracy of our experimental set-up.
The breakdown voltage values deduced using extrapolation of 1/M from two devices for each alloy are shown in figure 4. These values are supported by the breakdown voltage obtained from the reverse I-V data. For a given wafer, the seven sets of reverse I-V data exhibited abrupt breakdown at highly similar voltages, indicating highly uniform diodes. The typical reverse I-V data of the D = 400 µm diodes at different temperatures are plotted in figure 5 for the four alloys. For each alloy, an abrupt breakdown in dark current can be observed at all temperatures, for reverse bias voltage near the V bd deduced from avalanche gain measurements (indicated in figure 5).

Discussion
In figure 6a, the temperature coefficients of breakdown voltage of this work are compared with relevant reports from the literature. Considering only the AlAs 0.56 Sb 0.44 data from this work and reference [4], C bd increases with w, as observed on other semiconductors. The data from this work are also compared with semiconductors used in the avalanche region of InP-based APDs (InP and InAlAs) in figure 6a. For a given w, the Al 1−x Ga x AsSb diodes have very small C bd , in line with the C bd of AlAs 0.56 Sb 0.44 diodes from Xie & Tan [4] and much lower than those of InP and InAlAs diodes.
Alloy disorder potential analyses in [11] indicated that the very small C bd values for AlAs 0.56 Sb 0.44 diodes could have originated from significant alloy scattering, because of very different covalent radii of the Sb and As atoms. The ratio between the As and the Sb is the same for the AlAs 0.56 Sb 0.44 and the Al 1−x Ga x As 0.56 Sb 0.44 diodes, hence similarly small C bd are perhaps not surprising for the Al 1−x Ga x As 0.56 Sb 0.44 diodes in this work.
The experimental I−V characteristics from figure 5 facilitate a comparison of the reverse current density from Al 1−x Ga x AsSb with those from the current avalanche materials of choices, InP and InAlAs. In figure 6b, simulated band-to-band tunnelling current density versus electric field from InP [1] and InAlAs [10] p + in + diodes with w = 110 nm are plotted, with conditions of 0.95 V bd indicated by symbols. Also plotted are the gain-normalized dark current densities at electric fields corresponding to 0.95 V bd from our diodes with x = 0.10 and 0.15. At conditions of 0.95 V bd , the dark current densities in our diodes are approximately 5 × 10 −6 A cm −2 , at least 5 and 3 orders of magnitude lower than those of InP and InAlAs, respectively. The simulated tunnelling current for thin InAlAs diode is consistent with the level from a waveguide InGaAs/InAlAs APD using an 100 nm thick InAlAs avalanche layer (0.95 V bd  of 0.096 A cm −2 ) [12]. Our results confirm the potential of using thin AlGaAsSb as a low dark current avalanche region.

Conclusion
Four Al 1−x Ga x As 0.56 Sb 0.44 p-i-n diode wafers with 110-116 nm avalanche region width and Ga composition of x = 0, 0.05, 0.10 and 0.15 were characterized, in the experimental study on temperature dependence of breakdown voltage, avalanche gain and dark current. All four wafers showed weak temperature dependence of breakdown voltage and avalanche gain, with C bd ranging from 0.86 to 1.08 mV K −1 , among the lowest values ever reported for a wide range of semiconductor materials. Combined with much lower dark current density (compared with tunnelling current densities in InP and InAlAs), the Al 1−x Ga x As 0.56 Sb 0.44 materials are promising as avalanche layers in InP substrates-based APDs that require excellent temperature insensitivity.