The material was grown on n-type Si (100) substrate with solid source molecular beam epitaxy. The epitaxial layer stacks consist of a 500 nm-thick Ge buffer, and a 650 nm-thick Ge_0.94Sn_0.06 absorbing layer grown at 180 °C. The Ge buffer was grown by a two-step growth method, composed by 100 nm-thick low temperature Ge (LT-Ge) and 400 nm-thick high temperature Ge (HT-Ge), and underwent a multiple-cycle annealing from 600 °C to 800 °C, which was designed to improve the quality of the Ge and GeSn layers. Owing to significant challenges mentioned above, growing high-quality GeSn material directly on Si substrate is difficult. Currently, we always use the high–low temperature growth method (known as two-step growth method) to grow the Ge buffer. This method can release partial misfit dislocations at the Si and Ge interface. Meanwhile, it produces plentiful threading dislocations, which will spread to the upper GeSn material and decrease the material quality. It has been proved that multiple-cycle annealing can minimize the dislocation in the Ge buffer.^[14] High temperature (as close as possible, but not higher than the melting point of Ge) provides energy to the dislocations which exist in the epitaxial film, makes them migrate and ultimately quench. Furthermore, considering the different thermal expansion coefficients between Ge and Si, the epitaxial Ge buffer will induce tensile strain in cycle annealing, as shown in Table 1. This tensile strain can reduce the compressive strain in following epitaxial GeSn material, thus increasing the quality of the GeSn material.

To form p⁺ Ge_0.94Sn_0.06 contact, the top of the GeSn layer was implanted by boron fluoride ions (BF²⁺) with energy of 20 keV (calculated doping depth is 50 nm) and dose of 4.0 × 10¹⁵ cm⁻². In order to repair the lattice damage caused by the implantation of BF²⁺ ions and thermally activate BF²⁺ ions, the material was subsequently annealed at 450 °C for 30 s in an N₂ atmosphere. There is a temperature tradeoff between effectively lattice repairing and preventing Sn segregation. This sample was made out of circular mesas by standard photolithography and Cl-based inductively coupled plasma (ICP) etching with Cl₂, BCl₃, Ar plasma at flow rates of 10 sccm, 30 sccm, 20 sccm respectively, and 60 W RF power. These conditions realized an etch rate of 350 nm/min and produced well-defined mesas with sharp sidewalls, and smooth, residue-free surfaces.

To reduce the reflection and passivate surface of the material, a 500 nm-thick SiO₂ layer was deposited on the material by plasma-enhanced chemical vapor deposition (PECVD). The electrodes were deposited on both top-GeSn layer and Si substrate using electron beam evaporation and then fabricated by a lift-off process, including a 40 nm-thick Ni adhesion layer and a 500 nm-thick Al. Through processing mentioned above, we fabricated different mesa size devices, varying from 30 μm to 200 μm. The top-view microscopic image of the Ge_0.94Sn_0.06 photodetector with a 200 μm-diameter mesa is shown in Fig. 2(a).

Fig. 2.

	Figure Option ViewDownloadNew Window
	Fig. 2. (a) Top-view microscopic image of the GeSn photodetector. The mesa size is 200 μm in diameter. (b) Cross-sectional schematic of the top-illuminated GeSn photodetector structure along the black dash line shown in (a).

The x-ray diffraction (XRD) ω–2θ scanning of the Gegji_0.94Sn_0.06 sample grown on Ge buffered Si(100) substrate around (004) orientation is presented in Fig. 3(a). As we can see, the three peaks from left to right correspond to the Ge_0.94Sn_0.06 layer, Ge buffer, and Si substrate, respectively. At the same time, the full widths at half maximum (FWHM) of the Ge_0.94Sn_0.06 peak and Ge peak are 0.09° and 0.10°, respectively, which indicates that this GeSn layer has good crystalline quality. Furthermore, the substitutional Sn concentration, the lattice constant, and the strain of layer can be calculated by the same way reported in Ref. [16]. Table 1 summarizes the calculated results. It shows that the Ge buffer contains about 0.19% biaxial tensile strain, which verifies the conclusion mentioned above that the different thermal expansion coefficient between Si and Ge will introduce tensile strain in the Ge buffer.^[29] This tensile strain can expand the lattice constant of Ge, and partially relieves compressive strain in the upper GeSn layer. The relieving of compressive strain leads to the narrowing of the GeSn bandgap, and thus extends the cutoff wavelength of the detector.

Fig. 3.

	Figure Option ViewDownloadNew Window
	Fig. 3. HR-XRD ω–2θ scanning of the Ge0.094Sn0.06 sample grown on Ge buffered Si(100) substrate at (004) orientation. (b) TEM image of the Ge0.94Sn0.06 sample (the white lines in the image represent the interfaces of Ge0.94Sn0.06/Ge buffer and Ge buffer/Si, respectively). (c) HR-TEM image of the Ge0.94Sn0.06 layer. (d) HR-TEM image of the interface between GeSn layer and Ge buffer layer.

Table 1.

Table 1.

Table 1. Summary of lattice constant, in-plane strain, and Sn concentration. . Layer a⊥/Å a||/Å ε||/% Sn concentration Ge buffer 5.6489 5.6675 0.19 – GeSn 5.7352 5.6677 –0.75 5.89%

Table 1. Summary of lattice constant, in-plane strain, and Sn concentration. .

The high resolution transmission electron microscopy (HR-TEM) image of the Ge_0.94Sn_0.06 on Ge buffered Si is shown in Fig. 3(b). It indicates that the Ge_0.94Sn_0.06 epitaxial layer is single-crystalline. In addition, a high quality interface between the Ge buffer and Ge_0.94Sn_0.06 layer is shown in Fig. 1(d). Compared to the TEM images that without cycling annealing (not shown here), cycle annealing truly has a positive effect on improving the material quality.

The dark current–voltage (I–V) characteristics of the fabricated silicon-based Ge_0.94Sn_0.06 p–i–n photodetectors with different mesa sizes in diameter are shown in Fig. 4(a). It is observed that these devices exhibit a rectifying behavior of diode and the lowest dark-current of the devices with 30 μm-diameter mesa is 1.53 μA under 1 V reverse bias, which is comparably low compared to other reported values. A high I_on/I_off ratio about four orders of magnitude is measured under 1 V bias voltage, which is comparable to the reported values.^[30,31]

Fig. 4.

	Figure Option ViewDownloadNew Window
	Fig. 4. (a) The Idark–Vbias characteristics of GeSn photodetectors with various mesa sizes. (b) Dark-current density (Jdark) versus 1/D under 1 V reverse bias.

Further analysis of the dark current density (J_dark) has been conducted, which can be expressed as

According to this formula, the J_dark versus 1/D for the GeSn photodetectors at a reverse bias of 1 V bias, where D varies from 30 μm to 200 μm, is depicted in Fig. 4(b). At the same time, J_bulk and J_surf of the photodetectors can be calculated by linearly fitting the data in the scatter plot, which are 44.29 mA/cm² and 0.13 mA/cm, respectively. Moreover, the 200 μm-diameter photodector obtained the lowest dark-current density 0.075 A/cm² under 1 V reverse bias. The J_dark of the GeSn photodetectors typically increases with the Sn component. Increasing Sn component will lead to the narrowing bandgap and the increasing defect density. Since J_bulk is proportional to the threading-dislocation density, the lower J_bulk can indicate the high quality of the GeSn sample processed by circle annealing.

In addition, the main source of J_surf is interface traps around the mesa sidewall, which can introduce energy levels in the forbidden gap and eventually increase the peripheral surface leakage current. These traps were formed in the ICP etching process of the mesa. Currently, it has been reported that the passivation technique can effectively decrease the dark current, for example, forming Si,^[18] yttrium-doped GeO₂,^[32] or GeO_x/GeSnO_x (formed by ozone oxidation)^[33] passivation layer.

Figure 5 shows the I–V characteristics of the GeSn photodetectors (mesa diameter: 200 μm) in the dark and under the light of 1550 nm and 2000 nm laser with power of 1 mW and 0.12 mW, respectively. High responsitives of 0.45 A/W and 0.12 A/W are achieved at 1550 nm and 2000 nm under a reverse bias of 3 V, respectively, which are fairly high values for the Si-based GeSn p–i–n photodetectors as so far. The responsivity of GeSn photodetectors depends on the Sn content, the crystal quality, and the thickness of the GeSn absorbing layer. In the past few years, numerous Si-based GeSn photodetectors have been reported with relatively high responsivity ranging from 0.1 A/W to 0.376 A/W at 1550 nm.^{[15–28,34,35]} Therefore, the 600 nm-thick high-quality GeSn absorbing layer grown by multiple cycle annealing leads to the above-mentioned high responsivity and low dark current device performance. In 2018, Huang et al. reported a Ge_0.0975Sn_0.025 resonant-cavity-enhanced photodetector (RCEPD) fabricated on silicon-on-insulator (SOI) platform. To enhance the light–matter interaction, a vertical resonant cavity is created. This leads to a clear oscillation characteristic of responsivity and peaks at 1543 nm, which is exactly in the telecommunication C-band.^[34] This result shows that the responsivity of the GeSn photodetectors can be further improved by designing the resonant cavity.

Fig. 5.

	Figure Option ViewDownloadNew Window
	Fig. 5. Dark current and photocurrent illuminated by (a) 1550 nm (inset: responsivity of the detector as a function of bias voltage at 1550 nm), and (b) 2000 nm laser (optical power is 0.12 mW). The measured detector is 200 μm in diameter.

The inset displays the responsivity of the devices as a function of the applied voltage at 1550 nm. We find that photon-generated carriers are efficiently collected under a reverse bias of 0.25 V. With the increase of the reverse bias voltage, the responsivity increases slowly, finally peaks at −4.22 V. As the voltage increases further, there is a modest decline in the responsivity. This phenomenon may be caused by drastically increasing leakage current under higher reverse bias voltage. What is more, it has been reported that the Franz–Keldysh effect, which refers to the absorption coefficient dependence on the voltage, could affect the responsivity.^[36]

The optical spectral response (1100–2500 nm) of the GeSn photodetector is shown in Fig. 6. The scatter plot, which was measured by a 1500–1630 nm tunable laser and a 2 μm laser under 3 V reverse bias voltage, has a good agreement with the optical spectral response measured by the FTIR optical spectrometer at 3 V reverse bias voltage. The response of the GeSn photodetector rapidly decreases as the wavelength increases until the wavelength reaches about 1640 nm, which exactly corresponds to the direct-bandgap of tensile-strain Ge.^[1,2] As the wavelength increases further, the response of the GeSn photodetector begins to increase, which attributes to the absorption of the Ge_0.94Sn_0.06 absorbing layer and the Ge indirect-bandgap. Furthermore, there is a sharp decline of response at 1810 nm and the spectral response finally arrives a cutoff at the wavelength about 2.3 μm. This greatly extended cutoff wavelength is the result of the narrow bandgap of GeSn caused by the incorporation of Sn and almost fully strain-relaxed GeSn layer.

Fig. 6.

	Figure Option ViewDownloadNew Window
	Fig. 6. Optical responsivity and spectrum response of GeSn photodetector as a function of wavelength. Devices were measured by lasers (a tunable laser between 1500 nm to 1630 nm, a laser at 2000 nm) and an FTIR optical spectrometer. The data are shown in green scatter plots and blue curve, respectively (all measurements were conducted under 3 V reverse bias voltage).