Systematic study of GeSn heterostructure-based light-emitting diodes towards mid-infrared applications

Semiconductor optoelectronic devices operating in the mid-infrared (IR) range are highly desirable for chemical and biomedical sensing applications because many gas molecules such as water vapor, nitrogen-containing molecules, and hydrocarbon molecules have their characteristic absorption fingerprints in this wavelength range.^1–3 Among the many devices, mid-IR light-emitting diodes (LEDs) as light sources are the key components for developing low cost and light weight sensors. Although III–V based LEDs using InGaAs, GaSb, InAs, etc., feature mid-IR operation and relatively high efficiency,^4,5 their high cost and incompatibility with current complementary metal–oxide–semiconductor (CMOS) processes exclude them from being widely used for on-chip monolithic integration with low cost Si-photonics circuits that play an important role in the emerging field of low cost on-chip biological and chemical sensing. On the other hand, the commonly used group-IV semiconductors Si, Ge, and their alloys suffer from low light-emission efficiency due to their indirectness of bandgap; therefore, they are not ideal candidates to be used as the light source for sensing applications.

Recently, studies on group-IV semiconductor GeSn alloys open a new window for the development of Si-based optoelectronics active devices.^6–15 Since a direct bandgap GeSn has been experimentally identified¹⁶ and an optically pumped GeSn-based laser has been demonstrated,¹⁷ the GeSn-based light emitter as the low-cost Si-based light source is expected to be efficient, with which all necessary components for mid-IR sensing on a Si platform will soon be available. In addition, by varying the Sn composition in the GeSn alloy, the operating wavelength of the GeSn based LED could cover a broad mid-IR range. Preliminary investigation of electroluminescence (EL) from GeSn active layers has been conducted by several research groups.^{10–14,18–22} For the purpose of improving the device performance, some advanced device structures such as double heterostructures (DHS) and multiple quantum wells (MQW) were studied,^9,18,21,22 which offer better carrier confinement so that the light emission efficiency can be significantly improved. Particularly, the band structure of the Ge/GeSn/Ge DHS in this study features a higher density of states compared to the MQW whose energy states are quantized in the growth direction. As a result, the high carrier concentration in an active GeSn layer can be achieved at a high injection level. Therefore, the DHS-based LED holds the possibility of high power emission.

In order to fully develop the heterostructure-based GeSn LED technology, a detailed device study including temperature-dependence, injection-dependence, and Sn compositional-dependence is necessary, which currently is still missing. In this work, we report the comprehensive study of GeSn LEDs based on the Ge/GeSn/Ge p-i-n DHS with the Sn compositions up to 9.2%. Temperature-dependent photoluminescence (PL) and EL spectra have been characterized from 300 to 77 K for each device. The room temperature integrated EL intensity versus current injection was further investigated, which shows the monotonic-increase feature. Based on the experimental data and theoretical calculation, the band structure of the device is discussed, which reveals that the type-I band alignment was achieved.

The DHS GeSn LED samples were grown using an ASM Epsilon^® 2000 Plus reduced pressure chemical vapor deposition (RPCVD) system. The complete layer sequence includes: (1) a 750-nm-thick p-type doped Ge layer (also serving as the buffer layer); (2) a 200-nm-thick unintentionally doped Ge_1−xSn_x active layer; and (3) a 50-nm-thick n-type doped Ge cap layer. Since the concentration of background doping of the Ge_1−xSn_x layer was measured to be slightly p-type doped, in order to achieve a good PN junction, the doping concentrations for p- and n-type Ge layers were selected as 5 × 10¹⁸ and 1 × 10¹⁹ cm⁻³, respectively. The p-type Ge buffer layer was grown on a Si substrate using a two-step growth process (a low-temperature 100 nm undoped seed layer and a higher temperature 650 nm p-type doped layer) followed by an in-situ annealing to reduce the defects. The reactor was then cooled down to 350 °C in H₂ and after which SnCl₄ and GeH₄ were introduced into the chamber to initiate GeSn growth. Finally, a 50-nm-thick n-type Ge cap layer was deposited at the same temperature to avoid Sn-precipitation. The pseudomorphically grown Ge cap layer also served as the passivation layer to reduce the surface recombination. Three samples were grown with the Sn compositions up to 9.2% in this study. A detailed growth mechanism was discussed in our previous report.²³ 

After growth, the material quality, the layer thickness, the Sn composition and the strain of the as-grown wafer were carefully analyzed by using transmission electron microscopy (TEM) and high-resolution X-ray diffraction (XRD) techniques. The 2θ-ω scan is shown in Fig. 1(a). The Si, Ge, and GeSn peaks are clearly resolved. As the Sn composition increases, the GeSn peak shifts towards a lower angle. The reciprocal space map (RSM) is plotted in Fig. 1(b), which reveals that all GeSn layers were compressively strained (in-plane). The asymmetric contour plot of Ge was observed. This is due to the different strain states of the p-type Ge buffer layer (relaxed) and n-type Ge cap layers (in-plane tensile strain). The TEM results showed that the majority of defect was localized at the Ge/Si interface because of the optimized growth of the Ge buffer layer, leading to the low-defect GeSn layers. The density of threading dislocations was extracted from RSM.²⁴ The measured thickness of each layer agreed well with the initial device design. The material characterization results are summarized in Table I. The negative value of strain indicates the compressive strain.

The temperature-dependent photoluminescence (PL) characterization was conducted to further examine the material optical quality. The PL measurements were performed using a standard off-axis configuration with a lock-in technique (optically chopped at 377 Hz). A continuous wave (CW) laser with 532 nm wavelength was used as an excitation source. The laser beam was focused down to a 100 μm spot and the power was measured to be 560 mW. The PL emission was collected by a spectrometer and then sent to a thermoelectric-cooled lead sulfide (PbS) detector with a cutoff at 3.0 μm.

Figure 2(a) shows the PL spectra of samples A, B, and C at the temperatures from 300 to 77 K. Based on our bandgap energy calculation, all GeSn layers in this study are indirect bandgap materials. However, due to the small energy separation between the direct and indirect bandgap (cf. Table III), a portion of photo-generated electrons transfer from L- to Γ-valley via thermal activation, resulting in significantly enhanced direct bandgap emission. Therefore, the main peak of each spectrum in Fig. 2(a) is attributed to the direct bandgap transition.²⁵ The PL peak of indirect bandgap transition cannot be identified because its peak position is close to the direct peak and its intensity is weaker compared to that of direct peak. The PL peak blue-shift at low temperature was observed as expected. At 300 K, the PL peaks at 0.636, 0.605, and 0.548 eV were obtained for samples A, B, and C, respectively. In addition, the reduced PL intensity at low temperature was observed, which further confirms the indirectness of GeSn bandgap. The emission decreases with decreasing temperature can be explained as following:^16,25 due to the small energy difference between direct and indirect bandgap, more thermal activated electrons populate the Γ-valley at 300 K, resulting in direct bandgap transition dominates the PL spectrum. As temperature decreases, the PL intensity decreases with more rapid drop of the direct peak than the indirect peak. This is due to the reduced thermal excitation of electrons and less phonon-assisted indirect radiative recombination. At the temperature below 100 K, the indirect bandgap transition dominates the PL spectrum. However, owing to the dramatically reduced indirect radiative recombination, the overall PL intensity significantly decreases.

The optical properties of the material were further studied by pumping power-dependent PL measurement. The typical PL spectra for sample B are plotted in Fig. 2(b). At low pumping power of 150 mW, both indirect and direct bandgap transitions contribute to the PL emission, resulting in the broad peak line-width. As pumping power increases, more carriers intend to populate the direct Γ-valley, therefore the direct bandgap transition dominates the PL, which reduces the peak line-width at higher pumping power, as shown in Fig. 2(b).

The samples were fabricated into circular mesa structures with diameters of 100, 250, and 500 μm using photo lithography and etching processes. Both dry etch (Reactive-ion etching, RIE) and wet chemical etch methods were investigated for mesa etching. The etching rate of RIE (reaction gases of 30 sccm CF₄ and 30 sccm Ar) was measured as 30 nm/min and it was found to be Sn compositional dependent: higher Sn composition resulting in the lower etching rate. The wet chemical etch (HCl: H₂O₂: H₂O = 1:1:20 at room temperature) showed a stable etching rate of 100 nm/min regardless of Sn composition, which was then utilized in this work. The detailed study of etching mechanism will be reported elsewhere. The p- and n-type metal contacts consist of 10 nm Cr and 200 nm Au defined by metal deposition and liftoff processes. Figures 3(a) and 3(b) show the schematic cross-sectional view and scanning electron microscope (SEM) top view of the device, respectively.

The diode characterization was performed using an independent measurement system (no component shared with the PL measurement system). The current-voltage (I–V) characteristic of the DHS LEDs was measured at temperatures from 300 to 77 K with a Keithley 236 direct current (DC) source unit. A pulsed current source with a pulse width of 250 μs and duty cycle of 10% was applied for temperature-dependent EL measurement. The EL emissions from the device were sent to a spectrometer through a carefully aligned optical system, and the collected light was then detected by a PbS detector.

Figures 4(a)–4(c) show the typical current-voltage (I–V) characteristics of the LED devices measured at temperatures from 300 to 77 K (mesa diameter of 500 μm). The rectifying characteristic was observed at each temperature. As the temperature increases, for each device, the current increases due to more thermally excited carriers. Moreover, the dark current increases as the Sn composition increases under the same bias voltage because of the further reduced bandgap. The dynamic resistances at room temperature were extracted as 52.9, 28.0, and 19.0 Ω (operating at around 0.5 V) for samples A, B, and C, respectively. The relatively high dynamic resistance may arise from the thin n-type contact layer (50 nm Ge cap).

The room temperature dark current under reverse bias was further investigated. The dark current I_dark mainly consists of two components: bulk leakage current I_leak which is usually proportional to the device area and peripheral surface leakage current I_surf which relates to the periphery of the device. The dark current density can be described as

where D is the diameter of the mesa, J_dark, J_leak, and J_surf are dark, bulk leakage, and peripheral surface leakage current density, respectively. By linear fitting of the measured data and by using Equation (1) (see Fig. 4(d)), the J_surf at −1 V were extracted as 10.2, 10.4, and 122.5 mA/cm for samples A, B, and C, respectively.

As the Sn composition increases, the peripheral surface leakage current increases. The relatively high dark currents are mainly due to the following reasons: (1) No passivation for the surface and the sidewall of the mesa, resulting in the large peripheral surface leakage current. (2) The narrowed bandgap of Ge_1−xSn_x alloy leads to more thermally excited carriers. (3) As Sn composition increases, the lattice mismatch between GeSn and Ge increases, resulting in the increased defect density at the Ge/Ge_1−xSn_x interfaces, which contributes to the bulk leakage current. The peripheral surface leakage current can be reduced by depositing a passivation layer on the mesa surface and sidewall (reduces the surface recombination velocity) according to a report that showed the significantly suppressed dark current of a Ge_0.95Sn_0.05 photodiode with a Si passivation technique.²⁶ 

Temperature-dependent EL spectra under an injection current density of 255 A/cm² are shown in Fig. 5. The LEDs with mesa diameter of 500 μm were selected for EL study since they feature lower dark current density. For each device, as the temperature decreases from 300 to 77 K, the EL intensity decrease was observed, which agrees with the emission behavior of indirect bandgap material. The main emission peak at each temperature is determined to be the direct band-to-band transition. The position of the main peak blue-shifts as the temperature decreases from 300 to 77 K, which can be explained by the Varshni relation.¹⁴ For samples A, B, and C, the main peak shifts from 0.638 to 0.650 eV, 0.606 to 0.612 eV, and 0.537 to 0.546 eV, respectively. Note that since a lock-in amplifier and a chopper at frequency of 377 Hz (selected as a trade-off of detector response frequency and the measurement time) were employed for the EL measurement, the signal is averaged over a complete pulse on-and-off cycle (0.2 ms). Considering the low duty cycle, only the first harmonic is extracted by the lock-in amplifier. With the relatively short integration time, the signal-to-noise ratio (SNR) is low, resulting in an increased noise level of the EL spectra.

The peak positions of EL spectra were extracted via Gaussian fitting, as indicated by solid symbols shown in Fig. 6(a). The solid curves were fitted based on the Varshni relation: E(T) = E(0)−αT²/(T+β), where T is the temperature, E(0), α, and β are material-dependent parameters.²⁷ For samples A, B, and C, the values of E(0), α, and β are summarized in Table II.

Figure 6(b) shows the integrated EL intensities of the LED devices under various injection current densities from 102 to 357 A/cm². The monotonic increase of integrated EL intensities was observed for each sample, indicating that the maximum emission has not been reached yet. Theoretically, the device with the higher Sn composition would have the stronger emission intensity at the same current injection and temperature due to more injected carriers populating the Γ-valley in the conduction band (CB). However, sample B shows the highest EL intensity. This may be due to the better material quality compared to sample C. The lower material quality of sample C leads to the enhanced non-radiative recombination such as Shockley-Read-Hall recombination (recombination through the defect energy levels), which reduces the emission efficiency, resulting in the weaker EL intensity. In order to deeply understand the DHS LED characteristics, the electronic band structures of the samples in this study are discussed in Sec. V.

The electronic band structures of Ge/Ge_1−xSn_x/Ge DHS samples were investigated. The bandgap energies were calculated using a quadratic equation with bowing parameters, and the strain induced bandgap changes were obtained using the deformation potential model.^9,22 Figure 7(a) shows the band structure diagram of GeSn alloys in this study. The E_g^Γ and E_g^L are direct and indirect bandgap energies, respectively. Their difference is represented as ΔE_g. It has been reported that the band structure of GeSn alloy is determined by two main factors:^8,10 (1) Sn composition. The incorporation of Sn into Ge pushes down both Γ- and L-valleys in the conduction band (CB), while lifting the valence band (VB); (2) Strain. Since the GeSn alloys were grown on Ge buffer layer in this study, the alloys are under compressive strain, which brings Γ-valley up, altering the band structure of alloy towards indirectness. In addition, the compressive strain splits the heavy hole (HH) and light hole (LH) degeneracy in the VB by lifting the HH band. Our calculation showed that all three samples in this study remain indirect bandgap material, which agrees well with the band structure analysis in Refs. 8, 10, and 22. The direct and indirect bandgap energies are summarized in Table III.

For the estimation of band alignment, the method explained in Ref. 28 was followed, which applied the average band offset for VB based on Jaros' theory,⁷ and took the spin-orbit splitting into account but ignored the HH and LH bands separation. Figure 7(b) shows the band alignment diagram of DHS samples (not to scale). The ΔE_c^Γ and ΔE_c^L are barrier heights at Γ- and L-valleys in the CB, while the ΔE_v1 is the barrier height in the VB between the p-type Ge and GeSn layers, respectively. Due to the in-plane tensile strain, the n-type Ge cap layer exhibits reduced bandgap energy, resulting in the reduced value of barrier height ΔE_v2 between GeSn and n-type Ge layers. The E_g^Γ and E_g^L are energies of GeSn direct and indirect bandgap, respectively. The ΔE₁ represents the energy difference between the CB minimum at the L point of the p-type Ge barrier and the CB minimum at the Γ point of the GeSn layer, while the ΔE₂ is the energy difference between the Γ-valley minimum of GeSn and the L-valley minimum of the n-type Ge cap layer. The reduced bandgap energy of the Ge cap layer leads to the smaller value of ΔE₂ compared to ΔE₁. The band alignment calculation results of all three samples are summarized in Table III.

For sample A, the DHS provides electron confinement at both Γ- and L-valleys in the CB with the barrier heights of ΔE_c^Γ= 120 meV and ΔE_c^L= 28 meV, respectively. However, the CB minimum at the Γ point of GeSn is 20 (28) meV above the CB minimum at the L point of the p- (n-) type Ge layer (cf. Table III, represented as negative values of ΔE₁ and ΔE₂), indicating the lack of electron confinement for Γ-valley. The barrier heights in VB were found to be 43 and 40 meV at p- and n-type Ge layers, respectively.

For samples B and C, the CB minima at the Γ points of GeSn are below the CB minimum at the L points of the p- and n-type Ge barriers, with ΔE₁ = 22, 47 meV, and ΔE₂ = 12, 11 meV for sample B and C, respectively, indicating the type-I band alignment of DHS, which provides a favorable carrier confinement for the device. It is worth noting that although type-I band alignment was achieved, since the barrier heights are smaller than the room temperature thermal energy (1 k_BT ≈ 25 meV), the electron confinement for the Γ-valley is insufficient, resulting in the low efficiency of LED emission.

To optimize the DHS-based LED for improvement of emission efficiency, the following instructions are proposed: (1) using a direct bandgap GeSn well, which can be achieved either by incorporating more Sn or by growing a thicker GeSn layer to relax the material.¹⁶ Based on our theoretical calculation, fully relaxed GeSn with 9% Sn or fully strained GeSn (compressively, pseudomorphic to Ge) with 15% Sn could achieve direct bandgap material. Therefore, the Sn composition can be selected from 9% to 15% with appropriate strain; (2) using ternary material SiGeSn as the barrier instead of Ge, which offers separate tuning of bandgap and lattice constant by varying the Si and Sn compositions.²² Moreover, optimizing the doping profile (including contact layers and background doping of the GeSn active layer) would also improve the device performance in terms of reducing dark current and series resistance.

In summary, the Ge/GeSn/Ge DHS LEDs were investigated at temperatures from 300 to 77 K with Sn compositions up to 9.2%. Systematic study of I–V and EL characteristics has been conducted. The surface leakage current was extracted from I–V characteristics, whose relative high value is mainly due to the non-passivated surface of the device. The temperature-dependent EL spectra show blue-shift of the main peak that is attributed to the direct bandgap transition. The integrated EL intensity exhibits a monotonic increase as the current injection increases, and has not yet reached the maximum output. The electronic band structures of DHS were further investigated. Type-I band alignment has been achieved for samples B and C. Since the carrier confinement for the samples in this study is insufficient, which results in the low efficiency light output, viable solutions for the improvement of emission efficiency are proposed. With their CMOS compatibility and capability of monolithic integration on Si, the GeSn-based LED devices are considered as the robust alternatives to the III–V materials for mid-IR light sources integrated on Si.

The work was supported by the NSF (DMR-1149605), Arktonics, LLC (Air Force SBIR, FA9550-14-C-0044) and AFOSR (FA9550-14-1-0205). Dr. R. A. Soref and Dr. G. Sun acknowledge the support from AFOSR (FA9550-14-1-0196) and from AOARD (FA2386-14-1-4073).