1. Introduction

Near-infrared photodetectors monolithically integrated on Si are of current interest for low-cost applications in optical communications as well as in optical interconnects. Due to its relatively high absorption coefficient at 1.3-1.55 μm, Ge is regarded as the best candidate for this spectral range detection. Despite the large lattice mismatch (4.2%), high-quality Ge thin films have been successfully grown directly on Si, and high-performance Ge-on-Si photodetectors have also been achieved over the last 10 years [1–3]. Unfortunately, the efficiency of Ge photodetector is drastically reduced at above 1550 nm for its direct band gap of 0.80 eV. As a result, the detector cannot cover the L-band (1565-1625 nm) and U-band (1625-1675 nm) telecommunication windows. Although introducing in-plane tensile strain into Ge through high-temperature annealing can extend its absorption edge into L-band [4], it still fails to cover the entire L-band. GeSn alloy, another group IV semiconductor, is an intriguing alternative since its band gap, which is narrower than that of Ge, will decrease with the increase of Sn concentration. D’Costa et al. demonstrated that even with a very small Sn concentration (x~0.02), Ge_1-xSn_x alloy was sufficient to cover all telecommunication windows and compared with Ge, the optical absorption increased at leat 10-fold in C- and L-bands [5].

However, epitaxial growth of GeSn alloy has been proved to be difficult. Firstly, Sn tends to segregate to the surface and undergo a transition from α phase (diamond cubic) to β phase (body-centered tetragonal) at above 13.2 °C. Secondly, the lattice mismatch between α-Sn and Ge is very large (14.7%). Thirdly, the equilibrium solid solubility between Sn and Ge is rather low (x<0.01). What’s more, GeSn alloy has poor thermal stability [6,7], which prevents its applications for optoelectronic devices as well. Despite these difficulties, some efforts have been made on the epitaxial growth of GeSn alloys [6–8] and fabrication of GeSn devices [9]. A novel gaseous tin source, deuterated stannane SnD₄, has been synthesized and used to grow GeSn alloys by ultrahigh vacuum chemical vapor deposition (UHV/CVD) [8,10]. GeSn photodetector has also been fabricated [9] based on this technique. Nevertheless, SnD₄ is barely stable at near room temperature and difficult to synthesize [10]. Using high-purity solid source, molecular beam epitaxy (MBE) is a much easier method to grow GeSn alloys. However, they were believed to be unsuitable for device applications due to its thermal unstability [6,7]. Fortunately, recently we find that GeSn alloys grown by MBE can have rather good thermal stability when Sn concentrations are as low as about 2.5% [11]. In this article, we report the first GeSn photodetectors grown by solid source MBE that we have known. The devices were fabricated completely compatible with conventional Si CMOS technology and exhibited high responsivities in the whole telecommunication bands.

2. Material growth and device fabrication

The Ge_1-xSn_x (x = 0.03) alloy film was grown at 180 °C in a unique, custom-designed ultra-high vacuum growth system, which can be run at MBE or UHV/CVD model. The system was equipped with pyrolytic BN effusion cells for the deposition of Ge_1-xSn_x alloys. The base pressure of the system was 3.0 × 10⁻⁸ Pa. The substrate was 4-inch Sb-doped Si(001) with a resistivity of 0.01-0.02 Ω cm. High-purity (99.9999%) Ge and Sn solid sources were used for deposition. Before the growth of the alloy, a high-quality 240 nm-thick Ge thin film was deposited as buffer layer [11]. The thickness of the alloy was 820 nm. The root mean square (RMS) surface roughness of the film was 1.95 nm determined by atomic force microscopy (AFM).

In order to form a p-type contact layer, the top of the Ge_0.97Sn_0.03 layer was implanted with BF₂ ⁺ at a dose of 4 × 10¹⁵ cm⁻² and an energy of 40 keV, and then the sample was thermally activated at 400 °C for 30 minutes in N₂ atmosphere. The photodetectors were fabricated by standard lithography, dry etching, and lift-off metallization processing steps. The schematic cross section and top view of the GeSn p-i-n photodetector are shown in Fig. 1(a) and Fig. 1(b), respectively. Circular mesas with diameters ranging from 20 to 200 μm were formed by etching the patterned films down to the Si substrate using an inductively coupled plasma etcher. A 715 nm-thick SiO₂ was deposited on the top of the mesas for passivation. Metal contacts on Si contained a 25 nm-thick Cr adhesion layer, a 300 nm-thick Au film, and a 920 nm-thick Al film, while metal contacts on GeSn were formed only with a 920 nm-thick Al film.

 

Fig. 1 (a) Schematic cross section and (b) top view of a GeSn p-i-n photodetector.

Download Full Size | PDF

3. Experimental results

Figure 2(a) shows the measured 2θ-ω scans around (004) diffraction order of the Ge_0.97Sn_0.03 film by high-resolution X-ray diffraction (HRXRD). For the as-grown sample, a sharp and symmetric GeSn diffraction peak is found at around 65.358° with a full width at half maximum (FWHM) of only 0.117°, which indicates that the alloy is of high crystalline quality. After thermal activation at 400 °C, the alloy also exhibits a strong and sharp GeSn peak and the peak position does not shift (see Fig. 2(a)), implying its very good thermal stability. In Fig. 2(b), we show the Raman spectra of the Ge_0.97Sn_0.03 alloys. An amorphous-like Ge-Ge Raman peak is observed from the as-implanted sample, which is an indication of ion implant damage. The Raman spectra demonstrate that the lattice damage only has been partly repaired after thermally activated at 400 °C. As a result, some defects are expected in the surface layer. The damage can be better repaired by annealing at a higher temperature. Nevertheless, the thermal activation was performed at 400 °C for consideration of avoiding Sn segregation.

 

Fig. 2 (a) X-ray diffraction curves measured from the as-grown Ge_0.97Sn_0.03 film and the thermally activated film after ion implantation; (b) Raman spectra measured with a 488 nm Ar⁺ laser from the as-grown, the as-implanted, and the post-implant annealed Ge_0.97Sn_0.03 alloy films. The annealed samples were annealed in N₂ atmosphere for 30 minutes.

Download Full Size | PDF

In Fig. 3 , we show the dark current-voltage (I-V) characteristics of the GeSn photodetectors. The devices show rectifying diodelike behavior. The I-V curves are very similar to those reported by Mathews et al. [9], but we obtain a much lower dark current density. For a 50 μm diameter device, the diode currents are 35.3 μA, 45.2 nA, and 534 μA at −1 V, 0 V, and 1 V, respectively. The measured dark current density, ~1.8 A/cm² at −1 V, is about one order lower than that reported in Ref. 9. It presented a dark current of 0.38 mA for a 60 μm diameter device at −1 V, which corresponded to a dark current density of 13.4 A/cm². The relatively low dark current further indicates that our material is of high quality. We note that GeSn alloy has been grown on a high-quality Ge buffer layer [11], instead of directly on Si substrate [8,9]. In Ref. 11, we show that GeSn alloy grown with a high-quality Ge buffer layer has a better crystalline quality than that grown directly on Si substrate. Since the large epitaxial mismatch has been greatly reduced through the buffer layer, a better quality GeSn layer with fewer defects is expected.

 

Fig. 3 Current-voltage (I-V) characteristics of the devices with different mesa diameters (d).

Download Full Size | PDF

Assume that the total dark current (I_total) includes two main contributions, bulk leakage current (I_bulk) and surface leakage current (I_surface). Since bulk leakage current is proportional to device area (S) while surface leakage current is proportional to device diameter (d), if surface leakage is negligible, the dark current density, J_total = I_total/S, is independent of d. Otherwise, if surface leakage is prominent, J_total will increase linearly with 1/d. In the case of our device, the dark current density decreases gradually with diameter, from 2.36 A/cm² for d = 20 μm to 1.30 A/cm² for d = 200 μm. Further analysis demonstrates that J_total increases linearly with 1/d, which indicates that the surface leakage current contributes a large part to the total dark current. It’s expected, since there are many unrecovered defects in the surface. The calculated I_surface/I_total ratio, depending on d, ranges from 0.10 to 0.53.

The typical reported dark current densities of Ge photodetectors are in the range of 0.01-0.1 A/cm² [12], which are one or two orders lower than that of our GeSn photodetector. The value of our previously reported Ge-on-Si photodetector [13] is also obviously lower than that of the GeSn photodiode. The reported Ge has a better crystalline quality and fewer defects than that of our GeSn layer. The relatively larger dark current of our GeSn photodetector is probably due to more defects in the material, the prominent surface leakage, and the smaller band gap of the alloy.

The responsivity was measured using a 1310 nm laser and a 1500-1640 nm tunable laser. The light was delivered to the device by an optical fiber and an optical fiber probe. The light spot area was smaller than the active area of the photodiode. The photocurrent increased linearly with the increase of the input optical power and the responsivity was determined by computing the ratio of the measured photocurrent to the measured input optical power. In Fig. 4 , we show the dependence of responsivity on the wavelength at −1 V along with a photocurrent spectrum measured at 0 V. The GeSn photodetector exhibits a relatively high responsivity in all the measured wavelengths. Unlike the drastic decrease of the photoresponse of Ge detector beyond 1550 nm, the responsivity decreases much slowly. Even at a wavelength as long as 1640 nm, the responsivity is still relatively high. At 1310 nm, 1540 nm, and 1640 nm, the responsivities are 0.52 A/W, 0.23 A/W, and 0.12 A/W, respectively, corresponding to external quantum efficiencies (EQEs) of 49.2%, 18.5%, and 9.1%. The responsivities (or EQEs) are two orders higher than those reported in Ref. 9 and are comparable with those of Ge photodetectors [1–4,13].

 

Fig. 4 The responsivity versus wavelength at −1 V along with a photocurrent spectrum of a 100 μm diameter GeSn detector. The responsivities were measured using lasers while the photocurrent spectrum was measured at 0 V using a Fourier transform infrared (FTIR) spectrometer.

Download Full Size | PDF

The photocurrent spectrum (Fig. 4) was measured at 0 V using a Fourier transform infrared (FTIR) spectrometer (Bruker IFS 120HR) and a low-noise current amplifier. The FTIR spectrometer provided a normal incident infrared light source for the photocurrent measurement. Figure 4 shows that the detector covers the whole telecommunication bands with a broad photoresponse range from 800 nm to 1800 nm. Using the direct band gap E₀ of Ge (0.80 eV) and α-Sn (−0.41 eV), and a bowling parameter b = 1.94 eV [14], we predict the direct band gap of Ge_0.97Sn_0.03 to be 0.71 eV, which is much lower than that of Ge. The extended performance of the GeSn detector at above 1550 nm is due to the lower direct band gap of Ge_0.97Sn_0.03. Because of alloy broadening, from Fig. 4 we also find that the absorption edge is not as steep as that of Ge. This phenomenon has been reported in Ref. 9 as well.

Figure 5 shows a typical dependence of responsivity on the photodiode voltage and the inset in the figure shows the photocurrent versus input optical power at different reverse biases. Both figures demonstrate that the responsivity increases gradually with the increase of the reverse bias. For our devices, an external bias of about −3 V is still needed for effectively collecting the photoinduced carriers. The responsivities at 0 V and −1 V are about 25% and 60% of that at −3 V, respectively, independent of the wavelength.

 

Fig. 5 Responsivity (at 1640 nm) versus diode bias of a GeSn photodetector and the inset showing the photocurrent as a function of the input optical power at different biases. In the inset, the points are experimental data, and the solid lines are linear fits.

Download Full Size | PDF

4. Conclusion

A high-quality Ge_0.97Sn_0.03 film has been grown by MBE on Ge-buffered Si(001) substrate and shown to have rather good thermal stability. Using this epitaxial thin film, GeSn p-i-n photodetectors have been successfully fabricated by standard processes fully compatible with conventional Si CMOS technology. The detectors exhibited relatively high responsivities in all the measured wavelengths (1310-1640 nm) and were shown to have a photoresponse up to 1800 nm, covering the whole telecommunication windows. Even at a wavelength as long as 1640 nm, a quite good responsivity of 0.12 A/W has been obtained at −1 V. Therefore, GeSn alloy is a promising candidate for all telecommunication bands detection.