Qimiao
Chen
*a,
Shaoteng
Wu
*ab,
Lin
Zhang
a,
Hao
Zhou
a,
Weijun
Fan
a and
Chuan Seng
Tan
ac
aSchool of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798. E-mail: chenqm@ntu.edu.sg; wst@semi.ac.cn
bState Key Laboratory of Superlattices and Microstructures, Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, P.R. China
cInstitute of Microelectronics, A*STAR, Singapore 117685
First published on 18th April 2022
Semiconductor nanomembranes (NMs) have emerged as an attractive nanomaterial for advanced electronic and photonic devices with attractive features such as transferability and flexibility, enabling heterogeneous integration of multi-functional components. Here, we demonstrate transferable single-layer GeSn NM resonant-cavity-enhanced photodetectors for 2 μm optical communication and multi-spectral short-wave infrared sensing/imaging applications. The single-layer strain-free GeSn NMs with an Sn concentration of 10% are released from a high-quality GeSn-on-insulator (GSOI) substrate with the defective interface regions removed. By transferring the GeSn NMs onto a predesigned distribution Bragg reflector (DBR)/Si substrate, a vertical microcavity is integrated into the device to enhance the light–matter interaction in the GeSn NM. With the integrated cavity and high-quality single-layer GeSn NM, a record responsivity of 0.51 A W−1 at 2 μm wavelength at room temperature is obtained, which is more than two orders of magnitude higher than the reported values of the multiple-layer GeSn membrane photodetectors without cavities. The potential of the device for multi-spectral photodetection is demonstrated by tuning the responsivity spectrum with different NM thicknesses. Theoretical simulations are utilized to analyze and verify the mechanisms of responsivity enhancement. The approach can be applied to other GeSn-NM-based active devices, such as electro-absorption modulators or light emitters, presenting a new pathway towards heterogeneous group-IV photonic integrated circuits with miniaturized devices.
Ge has been widely investigated for SWIR photodetection due to its complementary metal oxide–semiconductor (CMOS) compatibility.17 However, Ge can hardly cover wavelength beyond 1.7 μm as limited by its direct bandgap of 0.66 eV.8 By alloying Ge with Sn, GeSn could have a narrower bandgap to extend the photodetection range and covers all the SWIR regime (1.1–2.5 μm) and even mid-infrared.18–20 In contrast to compound semiconductors for traditional SWIR photodetectors, its compatibility with CMOS processing will share the advantages of scaling and low cost. Regarding the benefits of GeSn alloy in the NM format, a few works have been carried out to investigate GeSn photodetectors in the NM/membrane format recently and demonstrated the photodetection ability beyond 2 μm wavelength.19,21 Although promising, the performance, especially the responsivity, of the demonstrated photodetectors is relatively poor (0.001–0.002 A W−1) which will hinder the practical applications of GeSn NM-based photodetectors.
In this work, we demonstrate the transferable single-layer GeSn NM resonant-cavity-enhanced (RCE) photodetectors near the 2 μm band with high efficiency and spectral tunability for 2 μm optical communication and multi-spectral short-wave infrared sensing/imaging applications. The single-layer GeSn NMs with Sn concentration of 10% were released from a high-quality GeSn-on-insulator (GSOI) substrate with defective regions near interfaces removed. Compared to epitaxially grown GeSn thin films, the lattice-mismatch-induced compressive strain could be fully relaxed in the transferable NM without forming dislocations/defects. The high Sn concentration extends the detection wavelength beyond 2 μm. By transferring the GeSn NM onto a predesigned distribution Bragg reflector (DBR)/Si substrate, a vertical microcavity is integrated into the device to enhance the light–matter interaction in the GeSn NM. With the integrated cavity and high-quality single-layer GeSn NM, a record responsivity of 0.51 A W−1 at 2 μm wavelength at room temperature was obtained, which is more than two orders of magnitude higher than the reported values of the multiple-layer GeSn membrane photodetectors without cavities. The response of the photodetector is tuned to be sensitive to a specific spectrum, which is promising for constructing filter-free multi-spectral short-wave infrared sensing systems. The mechanism of responsivity enhancement was analyzed and verified by theoretical simulations. Due to the broad and tunable absorption range of GeSn alloy and broad stopband DBR, the method can be applied to other wavelengths, e.g., MIR, as well.
Following the epitaxy, the GSOI donor substrate was formed to serve as the donor/source substrate for the GeSn NM by low-temperature wafer bonding and layer transfer: an ∼1 μm thick thermal SiO2 layer was grown on another Si substrate. Then, an ∼200 nm thick Al2O3 layer was grown on both the as-grown GeSn substrate and the thermal SiO2 substrate by atomic layer deposition. The Al2O3 layers were smoothened by chemical mechanical polishing (CMP). After that, the thermal oxide substrate and as-grown GeSn substrate were exposed to O2 plasma to improve the hydrophilicity. Then, the two substrates were bonded by direct wafer fusion bonding. To enhance the bonding strength, post-bonding annealing was performed in N2 at 250 °C temperature for 3 hours. Then, backside grinding and wet etching were utilized to remove the Si of the carrier GeSn substrate. Finally, the Ge buffer layer was removed and the GeSn layer was smoothened and thinned down to the desired thickness by CMP (GeSn thickness: 630 nm–920 nm) and the thickness was monitored by spectral ellipsometry.
The current–voltage (I–V) curves of the devices were measured using a Keithley 2450 source meter connected to a probe station at room temperature. The laser (Thorlabs-FPL2000) was used for the photocurrent measurement at 2 μm wavelength. The broad-band supercontinuum light source (tunable 400–2400 nm, SC-Pro-7) connected to the filter (1000–2300 nm, photon, etc.) was used for the responsivity spectrum measurements. The output light from the supercontinuum laser is collimated by the collimator and the laser beam divergence is less than 2 mrad (<0.1°). The power calibration of the light source was done using a commercial extended InGaAs photodetector (Thorlabs-FD10D).
(1) |
(2) |
θ and λ are the refractive angle and the wavelength of the incident light, respectively. n and d are the refractive index and the thickness of the layer, respectively. ε0 and μ0 are the vacuum permittivity and vacuum permeability, respectively. Then the characterized matrix for light transmitting through multilayers can be obtained by
(3) |
Then the reflectivity R(λ) can be obtained by
(4) |
The quantum efficiency (number of electron–hole pairs generated per incident photon) of the GeSn NM RCE photodetector was simulated to make the photodetector resonant at the desired wavelength using Lumerical three-dimensional FDTD. A plane wave was considered as the incident light. The vertical dimension of each layer is from experimental measurements and perfectly matched layer (PML) boundary conditions were used in the top and bottom of the structure. For the lateral directions, period boundaries were used. A power monitor above the GeSn NM was used to obtain the reflectivity (R) and a power monitor below the DBR was used to obtain the transmission (T). The quantum efficiency of the photodetector can be calculated by η = 1 − T − R.
The material quality of the GeSn NM is critical for achieving high-performance NM-based devices. To obtain a high-quality GeSn NM as well as introduce a sacrificial layer, the GSOI substrate serves as the donor wafer for the GeSn NM instead of the as-grown GeSn/Ge/Si. Compared with the as-grown GeSn/Ge/Si substrate, the GSOI as the donor substrate has two advantages: (1) the defective region near the GeSn/Ge or Ge/Si interface in the as-grown GeSn/Ge/Si wafer is removed in the GSOI substrate, resulting in a better NM quality; (2) the oxide layers under the GeSn layer can act as the sacrificial layer which can be easily removed without damaging the GeSn layer. Fig. 2(b) shows the cross-sectional TEM of the developed GSOI substrate, revealing clear interfaces between different layers. Only minor defects are observed in the GeSn layer, indicating the high quality of the GeSn layer. The fast Fourier transform (FFT) pattern of the GeSn layer confirms that the GeSn alloys are single crystalline (inset of Fig. 2(b)).
Next, in order to enable strong light–matter interaction in the GeSn NM through resonance, a highly reflective predesigned substrate without absorption loss is the key. Normally, the DBR to achieve high reflectivity in optoelectronic devices, such as vertical-cavity surface-emitting lasers (VCSELs) or RCE photodetectors, is epitaxially grown with alternate semiconductor layers.24,25 For group-IV devices, strain-balanced GeSn(Si) layers with a wider bandgap might be the material for semiconductor DBR. However, the refractive index contrast of the GeSn(Si) system is low,26 resulting in complex multi-layer growth and a narrow stopband of the DBR. Therefore, we chose the a-Si/SiO2 dielectric DBR and NM transfer instead of semiconductor epitaxy to overcome these issues. Fig. 2(c) shows the cross-sectional SEM image of the DBR/Si receiver substrate with 1.5 pairs of a-Si/SiO2 layers. The DBR consists of alternative a-Si and SiO2 layers with thicknesses of 185 nm/310 nm, respectively. The growth temperature of a-Si and SiO2 is 200 °C so the DBR is compatible with flexible substrates and can be used to extend the design to flexible NM-based optoelectronic devices as well.
Once the donor substrate and receiver substrate were ready, the GeSn NM was released from the GSOI donor substrate and transferred onto the highly reflective DBR/Si receiver substrate. Fig. 2(d) shows an optical micrograph of a transferred GeSn NM on the DBR/Si substrate, showing the etchant access holes, smooth and bubble-free NM. The etchant access holes have a dimension of 20 μm and a pitch of 200 μm. Finally, transferred GeSn NMs were fabricated into photodetector arrays by the processes described in the Experimental section. Fig. 2(e) illustrates an optical microscopy image of one representative fabricated GeSn NM RCE photodetector with a well-defined GeSn NM.
To measure the Sn concentration of the GeSn alloy, SIMS was performed on the as-grown GeSn/Ge/Si substrate, and it confirmed that the Sn concentration of the GeSn alloys is ∼10%. With such a high Sn concentration, the absorption coefficient of GeSn is extended beyond 2 μm and is higher than 6700 cm−1 at 2 μm wavelength. The refractive index of GeSn ranges from 4.30 to 4.36 with different wavelengths. Besides, the strain in the transferred GeSn NM was investigated by Raman spectroscopy. Fig. 3(c) illustrates the measured Ge–Ge longitudinal optical photon peak position of the GeSn NM, as-grown GeSn and Ge bulk. The Ge–Ge mode is at 301.0 cm−1 in the Ge bulk, and shifts to 294.3 cm−1 in the as-grown GeSn due to the introduced Sn, and to 291.9 cm−1 in the GeSn NM. The strain in the GeSn NM can be extracted based on the equation below:27
(5) |
Fig. 4(a) shows the measured normal incidence reflectivity of the DBR with different a-Si/SiO2 pairs (N). The reflectivity of the DBR increases with the a-Si/SiO2 pairs N. With N = 3, the reflectivity at 2 μm wavelength is 97% with a stopband as wide as 600 nm, which covers all the wavelengths of interest attributed to the high contrast of refractive index between a-Si and SiO2. The reflectivity spectra were further verified by a simulation using the transfer matrix method, as indicated by the dashed lines. In contrast to metal mirrors, there is no absorption loss in the DBR. To prove that the DBR has no absorption loss, the optical constants of a-Si and SiO2 were measured and are shown in Fig. 4(b) and (c). a-Si and SiO2 are transparent in the wavelength range of interest and the refractive index contrast between them is as high as 1.1. In the design of the GeSn NM RCE photodetectors, a DBR with 1.5 pairs of a-Si/SiO2 is utilized to obtain a broadband optical response. The measured reflectivity of the DBR (N = 1.5) in free space is 50% (black solid line in Fig. 4(a)). The reflectivity of the DBR (N = 1.5) in the device increases to 82% as estimated by the transfer matrix method (blue dashed line in Fig. 4(a)), which is high enough to demonstrate a strong resonance effect. The reason that the reflectivity of DBR in the device is higher than that of DBR in free space is that the first layer of DBR in the device is SiO2 (n = 1.5) and its refractive index in contrast to the incident media GeSn (n > 4.3) is high. For the free-space DBR, the incident media is air (n = 1) which has a low refractive index in contrast to the first layer (SiO2) of the DBR.
Fig. 4 (a) Measured and simulated reflectivity of DBR with different pairs of a-Si/SiO2. Refractive index, n and extinction coefficient, k of (b) a-Si, and (c) SiO2 measured by spectral ellipsometry. |
The RCE effect which enables responsivity enhancement and wavelength tunability is one of the most unique characteristics of the designed photodetectors. To investigate the RCE effect, the responsivity spectra of the RCE GeSn NM photodetector at voltages of 0.5 and 0.7 V were extracted (Fig. 5(c)). Different from the conventional GeSn photodetector whose responsivity decreases gradually with increased wavelength,20,28 the responsivity spectrum of the RCE GeSn NM photodetector demonstrates a distinct peak that is tuned by the structure design rather than the GeSn absorption coefficient. The resonant peak position is at 1980 nm with a wide wavelength range of 350 nm, which covers the whole 2 μm communication band. The responsivity can be further enhanced by increasing DBR pairs, but the wavelength range will be narrow. There is a tradeoff between the responsivity value and the spectral responsivity range, which can be tuned accordingly based on different applications. Moreover, the spectral responsivity can be improved by increasing the bias voltage. The higher bias voltage provides a stronger electric field in the photodetector, resulting in higher carrier collection efficiency. A responsivity of 0.51 A W−1 at 2 μm wavelength is observed under a 2 V bias voltage at room temperature as shown in Fig. 5(d). The responsivity increases linearly in the voltage range of 0 to 1 V and no saturation is observed within 2 V.
The dependence of the responsivity on the excitation laser power at 2 μm wavelength is shown in Fig. 5(e). The responsivity of the photodetector increases as the excitation laser power decreases, indicating that the photodetector is more sensitive to weak light. By fitting the experimental data using a simple power-law dependence, R = APn, a factor n of −0.54 is deduced. The increased responsivity at low excitation power suggests the existence of the localized states either inside GeSn NMs or at the GeSn NM/SiO2 interface. The localized states could trap one type of carrier and prolong the lifetime of the other type of carrier. At a high excitation laser power, the number of the localized states is limited compared to the photogenerated carriers so that the localization effect is not obvious, and the responsivity decreases due to the decreased carrier lifetime.
The noise-equivalent power (NEP) and specific detectivity (D*) of the photodetector are calculated based on the equations below:29
(6) |
(7) |
In addition to enhancing the responsivity value by increasing the bias voltage, the responsivity spectrum of the device can be tuned by the thickness of the GeSn NM due to the dimensional constraint of light, providing a potential application in multi-spectral sensing. To demonstrate the spectral tunability of the device, GeSn NM RCE photodetectors with a thinner NM thickness of 630 nm were fabricated. Fig. 5(f) shows the measured (solid lines) and simulated (dashed lines) normalized responsivity spectra of photodetectors with NM thicknesses of 630 nm and 918 nm. The spectral tunability of RCE GeSn NM photodetectors is observed clearly. As the GeSn NM thickness decreases, the resonant responsivity peak blue shifts by 155 nm. The experimental responsivity spectra match the simulated responsivity spectra well, as shown in the dashed lines in Fig. 5(f), proving that well-controlled spectral tunability of the photodetector and a high-efficiency multi-spectral GeSn NM photodetector array can be achieved on the same substrate by the multiple-step NM transfer, which has the potential for multi-spectral SWIR sensing applications (ESI Fig. S1†).
To figure out and verify the origin of the peak in responsivity peaks in Fig. 5(c), the FDTD simulation was performed. Fig. 6(a) illustrates the simulated quantum efficiency (i.e., number of electron–hole pairs generated by each incident photon) of the GeSn NM RCE photodetector as well as a reference GeSn NM photodetector without cavities. The dimensional structure of the fabricated device is used for the simulation of the RCE GeSn NM photodetector. The non-cavity structure of GeSn/Ge/Si with the same GeSn thickness was simulated for reference. The simulated quantum efficiency spectrum of the RCE GeSn NM photodetector shows a distinct single peak at 1985 nm wavelength which matches the measured peak in the responsivity spectrum, which is from the 4th order Fabry–Perot resonance. The quantum efficiency at resonance is as high as 84% and it is only 29% for the reference GeSn NM without a cavity. Increasing the a-Si/SiO2 pairs of the DBR would make the GeSn NM close to a perfect absorber. Fig. 6(b) shows the optical field profile (squared magnitude of the electric field) of the photodetector under resonance conditions. The incident light is confined in the Fabry–Perot cavity and a standing wave is formed in the cavity, resulting in a strong spatial overlap between the GeSn NM absorber and the optical field.
Fig. 6 (a) Simulated quantum efficiency spectra of the designed GeSn NM photodetector with and without a cavity. (b) Simulated optical field distribution in the device under resonance conditions. |
Finally, to evaluate the potential of the transferable GeSn NM RCE photodetectors, the responsivity of our device is compared to that of the reported GeSn photodetectors. Remarkably, as shown in the responsivity benchmark shown in Fig. 7, the responsivity of this work (0.51 A W−1) is enhanced significantly by more than two orders of magnitude compared to previously reported results of transferable GeSn nanostructured photodetectors (0.001–0.002 A W−1). The enhanced responsivity is attributed to the resonant cavity effect as confirmed by the resonant peaks in the responsivity spectrum and optical simulations. Besides, the enhancement might also be related to the improved GeSn NM quality since it is released from a high-quality GSOI donor substrate where the defective region near interfaces is removed, resulting in a lower trap density and a higher carrier collection efficiency. Even compared with other reported GeSn photodetectors fixed on bulk substrates, the responsivity of our device is among the highest values while the Sn concentration and thickness of our GeSn NM are at a moderate level. This means that transferable GeSn photodetectors without sacrificing the performance or even with better performance can be achieved by GeSn NM. The results propose that integrating the GeSn NM with a cavity is an effective approach for enhancing the light–matter interaction of GeSn-NM-based active optoelectronic devices. Besides, further device optimization, such as increasing the quality factor of the cavity (increasing pairs of the DBR), utilizing the PIN structure or the ferroelectric polymer gate, could improve the performances of the photodetectors further. Having demonstrated the advantages of the GeSn nanomembrane approach over current GeSn technology, it will be more attractive to heterogeneously integrate GeSn NMs with other NMs, thanks to the transferability of the NMs. By integrating GeSn NMs with different NMs (for example Si, Ge, InP or InGaAs), the photodetectors (arrays) can cover a broadband wavelength from visible to mid-infrared and enable more multifunctional photonic circuits, which can hardly be achieved by conventional direct growth approaches on Si. Considering the resonant-cavity-enhanced optical responsivity, spectral tunability, transferability for heterogeneous integration and device miniaturization, the developed GeSn NM RCE photodetectors are proposed as a promising candidate for efficient photodetection in the SWIR range for various applications, such as next-generation optical communications and multi-spectral SWIR sensing.
Fig. 7 The comparison of responsivity at 2 μm wavelength for the transferable GeSn NM RCE photodetectors and previously reported GeSn-based photodetectors based on different active structures (transferable GeSn19,21,31 and GeSn thin film fixed on bulk substrates18,20,32–38). |
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d1nr07293e |
This journal is © The Royal Society of Chemistry 2022 |