Owing to the unique advantages of low power consumption, large bandwidth, and high data rate, electronic–photonic integrated circuits (EPICs) have received increasing attention and can be used in telecommunications, sensing, on-chip optical interconnection, and quantum computing. Fully compatible with the standard complementary metal oxide semiconductor (CMOS) process [1,2], group IV materials are very suitable for high-volume, scalable manufacturing of monolithic EPICs. The emergence of low-loss hollow-core photonic bandgap fiber and Si-based waveguides makes the 2 μm band become a promising optical communication band. The fundamental components operated at 2 μm band have been developed rapidly. As a group IV material, GeSn alloys have great potential in fabricating 2 μm band optoelectronics. And GeSn-based photonic devices such as photodetectors [3–7] and modulators [8–12] were realized. Despite the laser, broadband light-emitting diodes (LEDs) with emission in the shortwave infrared (SWIR) region support a large variety of applications, such as biomedical imaging, spectroscopy, and environmental monitoring [13–16]. The epitaxial heterostructures of III-V inorganic semiconductors (for example, InGaAs and InGaAlAs) were already reported for making NIR-LEDs [17,18]. However, the III-V device preparation process is not compatible with CMOS technology and is difficult to integrate into the Si EPICs. Other materials, such as organometallic perovskites, perovskite–metal complexes, and organic semiconductor dyes, are limited by the emission wavelength range (700–1600 nm) [15,19].

Although there is much literature about infrared LEDs, an efficient electrically excited group IV light source remains challenging to achieve EPICs due to the indirect bandgap property of Si and Ge. Recently, germanium tin (${\mathrm{Ge}}_{1-x}{\mathrm{Sn}}_x$) alloys, as group IV materials, have adjustable bandgap between 0.66 eV (Ge) and 0 eV with varied Sn content and can turn to direct bandgap semiconductors as Sn content exceeds about 8% [20,21], showing very promising applications in preparation of efficient light emitters. However, the lattice mismatch between germanium tin and silicon and easy Sn segregation make it very difficult for epitaxial high quality germanium tin films on Si substrate. After researchers’ years of efforts, GeSn lasers operated by optical and electrical pumping ways have been demonstrated at low temperature, showing that GeSn material is very promising for making an efficient light source [22–26].

Although most reported GeSn alloys were grown by molecular beam epitaxy (MBE) [20,27] and chemical vapor deposition (CVD) [28–30], the rapid melting growth (RMG) technique offers another method for growing nearly defect-free GeSn alloys by terminating the threading dislocations at the seed region [31]. Another advantage of the RMG technique is that the composition is gradually changed along the growth direction of GeSn nanostrips. This allows for the preparation of light-emitting devices with different bandgaps on one substrate. And by combining these LEDs, a broadband light source can be obtained on chip. Recently, single crystal materials such as Ge, GeSn, and GePb have been prepared on Si substrate by RMG [32–35]. High performance photodetectors grown by RMG technology were fabricated, pioneering the potential applications of RMG method in Si photonics [4,33]. In addition, GeSn grown by RMG has a small thermal tensile strain ($<0.3\%$), induced by the expansion difference between Ge and Si substrate, which can facilitate the direct bandgap transition. The photoluminescence characterization of GeSn prepared by the RMG method has been investigated, and strong light emission from the direct bandgap was achieved [36].

In this work, single crystal GeSn nanostrips were grown on Si substrate by RMG. GeSn lateral p-i-n light-emission diodes on insulator were formed by a CMOS compatible process. The broad-spectrum (BS) light source with light emission ranging from 1600 to 2200 nm was achieved in the GeSn LED array by controlling the Sn content of active layer varying from 2.1% to 5.2%. Compared with the single GeSn LED, the full width at half-maximum (FWHM) of BS-LED is about 1.77 times wider, reaching about 340 nm. These results indicate that the GeSn LED arrays prepared by RMG have promising applications for fabricating the broad-spectrum SWIR light source.

For single crystal GeSn grown by RMG, the Sn content ($x$) distribution follows the Scheil equation, i.e., $x=k{x}_0{(1-d/L)}^{k-1}$, where $x$ is the Sn content, $x_0$ is the initial doped content, $d$ is the distance from the Si seed, $L$ is the length of the GeSn strip, and $k$ is segregation coefficient of Sn at the solid Ge/liquid Ge interface. Although the Sn content is gradient along the strip, for the single GeSn LED, the electrically injected carriers are more likely to move towards the regions with narrower bandgaps, leading to the narrow light emission band, as shown in Fig. 1(a). From the Scheil equation, for the same $d$ value, the Sn content can be controlled by the length of GeSn strip, offering GeSn LEDs with different bandgaps on one substrate. Therefore, the GeSn LED arrays with varied Sn content were proposed, as shown in Fig. 1(b). The light emission wavelength range can be broadened effectively through making LED arrays on parallel GeSn nanostrips.

The material growth was conducted on a (100)-oriented Si substrate. First, a ${\mathrm{Si}}_3{\mathrm{N}}_4$ film with thickness about 350 nm was deposited on a Si(100) substrate by low-pressure chemical vapor deposition (LPCVD). Then, the Si seed windows were defined by etching the ${\mathrm{Si}}_3{\mathrm{N}}_4$ film and the patterned substrates were transferred into a molecular beam epitaxy (MBE) chamber. A 500 nm thick amorphous GeSn film was deposited at substrate temperature 50°C. After the growth of the amorphous GeSn film, the GSOI (GeSn on insulator) strips were defined by photolithography and inductively coupled plasma (ICP) etching. Finally, a 1.5 μm thick ${\mathrm{SiO}}_2$ layer was deposited to form a microcrucible structure. The samples were annealed at 930 °C for 1 s by rapid thermal annealing (RTA) treatment to investigate the crystalline GeSn growth.

Figure 2(a) shows the RMG GeSn LED array fabrication process flow chart. The p-i-n light-emitting diodes were fabricated on the GeSn strips after RMG growth. After etching the upper layer of ${\mathrm{SiO}}_2$ from 1500 to 300 nm, the ion implantation windows were opened through photolithography and ICP. Then, the B and P were implanted into GeSn strips and annealed to form ${\mathrm{p}}^{+}$ GeSn and ${\mathrm{n}}^{+}$ GeSn regions. Subsequently, in order to activate the implanted ions and repair the lattice, the sample was annealed at 450°C for 30 s, and this annealing condition will not cause Sn segregation. Finally, a 1000 nm thick metal electrode layer was formed on contact holes by the metal deposition and lifting off process. The GeSn strips of various lengths were designed to prepare broad-spectrum LED (BS-LED), and the strips are named A, B, C, D, and E in descending order of length. The schematic drawing of GeSn LEDs array is shown in Fig. 2(b). For comparison, the single GeSn LED and the LED array near the seed region named as controlled LED (C-LED) were also fabricated.

Figure 3(a) shows the cross-sectional transmission electron microscopy (XTEM) image and SAED pattern of the GSOI strip grown by the RMG method. The interface of ${\mathrm{SiO}}_2/\mathrm{GeSn}/{\mathrm{Si}}_3{\mathrm{N}}_4$ is sharp and the SAED pattern with a single set of diffraction spots indicates the high quality of GeSn alloys grown by the RMG method. The clear lattice fringes without defects shown in Fig. 3(b) also testify to the excellent quality of the GSOI strip.

To evaluate the performance of the lateral GeSn LED, a single GeSn LED on the GeSn strip is studied first, and the inset of Fig. 4(a) shows the scanning electron microscopy (SEM) image of the LED. The microregion Raman spectra were used to determine the Sn content in the LED active region. Raman spectra were tested at room temperature in backscattering geometry using a Horiba Jobin Yvon micro-Raman system equipped with an Ar⁺ laser excited at 488 nm and a 600 lines/mm grating. The Ge–Ge peak in Raman spectra is decided by the Si content, the Sn content, and the strain. In order to obtain accurate strain values, it is necessary to analyze the Sn and Si components. First, based on our previous GeSn on insulator (GSOI) work [34], the Sn content is about 0.3% (equilibrium solid solubility of Sn in Ge) at the Si seed region. The Si content is calculated based on the ratio of Ge–Ge and Si–Ge peak intensities, and it is about 12.8% near the seed region [37]. The strain of ${\mathrm{Si}}_y{\mathrm{Ge}}_{1-x-y}{\mathrm{Sn}}_x$ strip with a low Si content could be calculated from the following equation [33,34]: $${\epsilon }^{\mathrm{SiGeSn}}=\frac{{\omega }_{\mathrm{Ge}-\mathrm{Ge}}^{\mathrm{bulkGe}}-{\omega }_{\mathrm{Ge}-\mathrm{Ge}}^{\mathrm{SiGeSn}}-82x_{\mathrm{Sn}}^{\mathrm{SiGeSn}}-16y_{\mathrm{Si}}^{\mathrm{SiGeSn}}}{563},$$(1)where ${\omega }_{\mathrm{Ge}-\mathrm{Ge}}^{\mathrm{bulkGe}}$ and ${\omega }_{\mathrm{Ge}-\mathrm{Ge}}^{\mathrm{SiGeSn}}$ are the wave number of the Ge–Ge Raman peak of Ge and SiGeSn film and ${\epsilon }^{\mathrm{SiGeSn}}$ is the strain of SiGeSn. After substituting the content of Si and Sn, the ${\epsilon }^{\mathrm{SiGeSn}}$ is calculated as 0.25%, and the type of the strain is tensile, which is attributed to the expansion difference between Si and Ge. As shown in Fig. 4(a), the calculated Sn content from Raman spectra is about 2.5%. The current-voltage (I–V) characteristics of the single LED are shown in the inset of Fig. 4(b). The perfect rectifying behavior of the single RMG GeSn LEDs is observed, and the on/off ratio is about ${10}^6$ for $\pm 1\mathrm{V}$. Moreover, the turn-on voltage is 0.5 V, and the series resistance is $1.06\times {10}^3\mathrm{\Omega }$. The value of ideality factor obtained by fitting the I–V curve is 1.24. Figure 4(b) shows the room temperature electroluminescence (EL) spectra with different injection currents for the single LED. The peak position is at about 1850 nm and almost constant under different injection current. The EL intensity increases with increasing injection current, along with the broadened spectra. This is likely because when the carrier injection density increases, although the direct transition still dominates, the indirect bandgap transition starts.

Figure 5(a) shows the SEM image of GeSn LED arrays. Five GeSn strips with lengths between 105 and 145 μm were fabricated to control the Sn distribution along the strips, and the spacing between the strips is about 3 μm, which is used to form compact LED arrays. The width, length, and thickness of the single LED are 2 μm, 3 μm, and 700 nm, respectively. Two kinds of LED arrays are fabricated by connecting five LEDs that are denoted as broad-spectrum LED and controlled LED. The strip lengths are 145 μm, 135 μm, 125 μm, 115 μm, and 105 μm for A, B, C, D, and E, respectively. Figure 5(b) shows the current-voltage (I–V) characteristics of devices. The GeSn LEDs show good rectifying behavior about ${10}^6$ for $\pm 1\mathrm{V}$ like the single LED shown above, confirming that the lateral GeSn PIN LED arrays are fabricated. The forward and reverse currents of both C-LED and BS-LED are about five times that of the single LED above for 1 V, indicating that the LEDs are connected successfully. The low dark current indicates the high crystal quality of GeSn grown by RMG.

The micro-Raman spectra of LEDs are shown in Figs. 6(a) and 6(b). Compared with bulk Ge, the shift of the Ge–Ge LO peak is nearly the same in C-LED, about $1.7{\mathrm{cm}}^{-1}$, while the maximum shift of the Ge–Ge LO peak is about $4.2{\mathrm{cm}}^{-1}$ in BS-LED. Since the Sn distribution follows the exponential function, the farther the position away from the seed region, the higher the Sn content. From Raman test, the calculated Sn content of the active GeSn layer is about 2.1% in each strip for C-LED, which is likely because the LED array is fabricated near the Si seed region and the Sn content changes little. However, due to the exponential increase of Sn content along the strips, the situation is much different in BS-LED. The Sn component in strips A and B in BS-LED is similar, about 2.1%, and it increases to 2.5% in strip C. The Sn fraction comes to 5.0% and 5.2% for strips D and E in BS-LED. According to our previous work, although GeSn grown by RMG exhibits tensile strain, it still needs about 7.8% Sn content to realize the direct bandgap material [36].

The bandgap of GeSn strip is studied by the room temperature photoluminescence (PL) measurement, carried out by a monochromator with a 300-groove grating equipped with 785 nm laser and a liquid-nitrogen-cooled InGaAs detector with cutoff wavelength at 2.4 μm. As shown in Figs. 7(a) and 7(b), strong light emission related with GeSn direct bandgap transition is observed, confirming the high quality of the GeSn grown by RMG. The experimental and theoretical bandgaps of C-LED and BS-LED are shown in Figs. 7(c) and 7(d), respectively. As expected, the energy bandgaps of GeSn in C-LED array are nearly the same, which are in agreement with the Raman results that Sn content is almost the same. However, the peak positions change greatly for BS-LED, that is, 0.711 eV (1743 nm), 0.707 eV (1755 nm), 0.684 eV (1812 nm), 0.605 eV (2049 nm), and 0.603 eV (2055 nm), respectively. The direct bandgap of the GeSn layer is calculated by the deformation potential theory [38,39]. The measured direct bandgaps are good matches with the theory prediction. As expected, the energy difference between the direct bandgap and indirect bandgap decreases with the increase of Sn content. This can also explain that the spectrum for D and E in the BS-LED is relatively broad due to the small energy difference between the direct bandgap and indirect bandgap. Thus, by choosing the Sn content for the LED active layer, broad-spectrum LEDs can be achieved on one chip.

Electroluminescence (EL) spectra were measured at room temperature using a standard off-axis configuration with a lock-in technique (optically chopped at 124 Hz). The output light was collected by a Princeton Instruments SP-2300 spectrometer in step scan mode, and the supporting detector is a liquid nitrogen-cooled InGaAs detector with a cut-off wavelength of 2.4 μm. Figures 8(a) and 8(b) show the power dependence of light emissions at room temperature from C-LED and BS-LED. Compared with the single LED, the electroluminescence intensity of LED arrays is higher due to the connection of five LEDs. The emission wavelength of C-LED ranges from 1600 to 2000 nm, and the FWHM is approximately equal to that of a single LED, which is attributed to the nearly same Sn component in the active layer. It is clear that the EL intensity of both devices increases with the increasing current density, and only one peak is observed in both EL spectra. It is worth noting that the light emission wavelength of BS-LED ranges from 1600 to 2200 nm, which results from the increased Sn content of the active layer in D and E. The peak wavelength of EL spectra is summarized in the inset of Figs. 8(a) and 8(b). Due to the Joule heat generated inside the devices, the emission peaks of LEDs gradually shift to longer wavelengths as the injection current increases. The emission wavelength (1789 nm) in C-LED is in good agreement with the calculated direct bandgap (0.724 eV) by the deformation theory. The peak wavelength of the BS-LED shifts to 1875 nm, which is a mixed interaction of the five LEDs with different strip lengths.

The integrated EL intensity and FWHM as a function of injection current are shown in Figs. 9(a) and 9(b). The FWHM increases as the injection current increases in both LEDs. Compared with the C-LED, the FWHM increases obviously for BS-LED. By utilizing different length strips with different components at the same position, the LEDs in BS-LED can emit light with different wavelengths. When their luminescence merges, the light emitting wavelength of the LED array is broadened. The FWHM of C-LED increases little, from 175 nm to 191 nm, at higher injection current. However, this situation is much obviously in BS-LED, and the FWHM increases from 282 nm to 340 nm. At injection current 15 mA, the FWHM of BS-LED is about 1.77 times that of C-LED. These results indicate RMG is a potential method to realize the broadband LED. The integrated EL intensity ($L$) as a function of injection current $I$ is characterized by $L\sim I^m$. The exponent $m$ is used to characterize the emission mechanism. The exponent $m$ is about 1.38 for C-LED and 1.30 for BS-LED, exhibiting superlinear dependence. This superlinear dependence ($m>1$) indicates the radiation recombination from the injected unbalanced carrier is dominant [40–42].

Although the broadband LED is demonstrated in this work, there are still some approaches to improve the LED performance. First, the range of Sn components in the LED active region is increased by introducing more different length strips. Second, the luminescence performance is limited by the size of the devices, and the width of the nanostrips can be further increased to improve the luminous efficiency. Third, the LED array can work in the waveguide mode, which offers more applications for multifunctional EPICs.

In conclusion, room temperature GeSn SWIR LEDs were demonstrated on insulators. The GeSn strips with different length were paralleled to form LEDs devices with different Sn component distributions. Utilizing the characteristics of the Sn component gradient from the rapid melting growth method, a GeSn LED array with Sn content ranging from 2.1% to 5.2% was fabricated. Electroluminescence was observed at room temperature under continuous current injection, and the wavelength range of light emission is from 1600 to 2200 nm. Theoretical analysis showed the observed electroluminescence originates from the direct-gap transition. The superlinear dependence between the integrated EL intensity and the injection current indicates the excellent emission mechanism. The FWHM is about 340 nm for BS-LED, which is 1.77 times that of the single component LED. These results pave the way toward efficient electrically injected on-chip broadband light emitters.