1. Introduction

Silicon photonics is an ideal platform for on-chip and off-chip information routing [1,2]. It provides a cost-effective solution for a wide range of applications, such as sensing [3], optical imaging [4], and quantum computing [5]. The manipulation of light in the silicon involves generation, modulation, and detection. Nevertheless, this is still challenging to achieve in an entirely monolithic platform due to the silicon’s fundamental limits [6,7]. For this purpose, heterogeneous integration of narrow bandgap semiconductors on silicon was successfully demonstrated [8].

High-performance photodetectors and modulators based on III-V compounds and germanium were achieved [9–11], however, the fabrication process is still technologically sophisticated due to mismatch in the lattice constant and thermal expansion coefficients [12,13].

Recently, two-dimensional materials entered the spotlight as a promising building block for the next generation of integrated optoelectronic devices [14–16]. These materials possess distinctive and intriguing optoelectronic properties. Additionally, given their multilayered nature, they are covalently bonded in-plane and stack together out-of-plane by weak van der Waals force. This enables their integration into various substrates without the restriction of lattice matching, offering a tremendous simplification in the fabrication process [17,18]. Recently, remarkable progress has been made toward the hybrid integration of 2D materials on silicon photonics (SiPh) platforms [19–25]. For instance, integrated graphene photodetectors demonstrated a broadband optical absorption and ultrafast response up to 110 GHz with a reasonable responsivity [26,27]. However, these photodetectors exhibit a high dark current which is higher than the photogenerated current when the detector is operated under bias [19,24,28,29]. This is a direct result of its zero bandgap which makes it difficult to turn the channel conductance. This leads to a high shot noise that limits the photodetector sensitivity.

Developing integrated photodetectors based on other layered materials such as transition-metal dichalcogenides (TMDCs) is challenging, since most of their bandgaps fall within the absorption band of silicon, making them unsuitable for short-wavelength infrared (SWIR) and mid-infrared (MIR) active photonic components [30–32]. High-performance integrated infrared photodetectors have been demonstrated in black phosphorous (BP) and black arsenic phosphorus (BAsP) [19,22,33]. They exhibited a thickness-dependent bandgap covering the visible to the mid-infrared spectral range. Additionally, they possess excellent electrical properties, including high hole mobility and unique anisotropy in optical absorption. However, BP lacks chemical stability under ambient conditions [34,35]. This complicates the fabrication process and adds reliability issues. Therefore, exploring novel 2D materials with a strong optical response in the NIR-MIR and excellent air stability is crucial for the development of various integrated optoelectronic devices.

More recently, a novel group IV–P 2D family (combination of group IV element (Si, Ge, or C) with phosphorus) results in a material with distinctive optoelectronic features [36–38]. In particular, 2D GeP uniquely combines narrow and wide tunable bandgap energies (0.51–1.68 eV), offering a broadband absorption from the visible to MIR spectral range [39]. Additionally, experimental and theoretical studies showed that this material exhibits excellent chemical and dynamic stability compared to BP [39–42]. Moreover, it has a low symmetry monoclinic structure that results in strong in-plane anisotropy and directional-dependent absorption [43]. Despite its phenomenal photonic and optoelectronic properties, most of the research on GeP is focused on theoretical studies and investigating its nonlinear optical properties (NLO) [40,43]. In addition, there were few demonstrations of surface illuminated detectors in visible and SWIR [39,43,44], whereas experimental explorations of hybrid integration of GeP on silicon photonics platform haven’t been realized.

Herein, we report to the best of our knowledge, the first multilayer GeP photodetector integrated on a silicon photonic platform operating in the SWIR. The hybrid Si/GeP waveguide photodetectors work under bias with a low dark current of few nano-amps and demonstrate excellent stability and reproducibility. Additionally, devices showed remarkable photoresponsivity and external quantum efficiency with reasonable sensitivity.

This report does not only provide an example of efficient SWIR photodetectors based on multilayer GeP but it also opens prospects for extending the 2D family toward novel MIR optoelectronics.

2. Results and discussion

2.1 2D GeP integration into SiPh

A schematic configuration of hybrid integration of multilayer GeP into SiPh chips is shown in Fig. 1. In this study, we adopted two waveguides integrated Si/GeP photodetectors structures, a micro-ring resonator (MRR) (Fig. 1(a)) and a straight waveguide (SW) (Fig. 1(d)). Both designs employ SOI wafers with a 220 nm top silicon layer and a 2 µm buried oxide layer. The MRR cavity has a waveguide width (d) = 460 nm integrated with three different ring radii of 40 µm, 45 µm, and 50 µm. Light is coupled to the micro-resonator from a bus waveguide using a 100 nm gap spacing and collected at the output via the through port. Two metal pads of Ti/Au (10 nm/120 nm) were symmetrically deposited as electrodes to measure the static and dynamic performance of the photodetector. The contact electrodes are designed with a 3–5 µm gap at both sides of the silicon waveguide. Therefore, losses due to metal absorption are negligible. In addition, the 100 nm height difference between the metal and Si waveguide is critical to ease the transfer of multilayer GeP and it ensures conformal coverage, as we will discuss below. The multilayer GeP was transferred on top of the Si waveguide or MRR by using a deterministic dry transfer process. This technique enables a free cross-contamination transfer without damaging the target device structures [20,45]. More details about the transfer and fabrication process are included in Supplement 1.

 

Fig. 1. Integration of multilayer GeP into SiPh, devices structure, and characterization (a)-(c) schematic configuration of heterogeneous multilayer GeP/silicon photodetector based on micro-ring resonator (MRR) (b) SEM image of the transferred GeP on MRR (c) AFM image scan, the dashed red line shows a thickness of ∼ 65 nm GeP (d)-(f) schematic configuration of heterogeneous multilayer GeP/silicon photodetector based on straight waveguide (SW) (e) SEM image of the transferred GeP on SW (f) AFM image scan, the dashed red line shows a thickness of ∼ 90 nm GeP.

Download Full Size | PDF

The scanning electron microscope (SEM) images of the MRR and SW devices are shown in Fig. 1(b) and 1(e), respectively. As can be seen, the GeP flakes are well aligned and conform to the underneath photonic structure below it which results in efficient light/ GeP coupling. In this study, we chose GeP flakes thickness in the range of 32 nm to 95 nm. The thickness measurements were performed using atomic force microscopy (AFM) as shown in Fig. 1(c), and 1f. Apparently, thicker GeP flakes result in higher mechanical stability and stronger light absorption, hence the detector footprint can be minimized. Additionally, with this range of thickness, we were able to perform a conformal coverage on the top and the sidewalls of the strip waveguide, see Supplement 1 (3D image reconstructed from AFM scan). However, for flake thickness > 100 nm, we find it challenging to cover the waveguide sidewalls due to the material’s inflexibility.

Supplement 1 depicts the phonon vibration of GeP measured using polarization-dependent Raman spectroscopy. The GeP flakes possess a low-symmetry monoclinic structure. Therefore, its Raman modes are anisotropic, and they exhibit a polarization dependence [40,43]. Given the nature of our fabrication process, the crystal orientation of GeP flake couldn’t be determined before the transfer process, hence no optimization of the crystal orientation can be performed to maximize the absorption. More details on Raman measurements conducted on photodetectors can be found in Supplement 1.

2.2 Light–multilayer GeP interaction

Understanding the interaction of the guided light with the exfoliated multilayer-GeP is essential to designing optimal and novel integrated optoelectronic devices on the SiPh platform. Hence, numerical and experimental investigation of the change in the effective refractive index (n_eff) and light absorption loss (κ) of the hybrid Si/GeP structures were performed. The calculation was carried out for different GeP thicknesses and coverage lengths. The results are focused on the telecom wavelengths of 1310 nm and 1550 nm. Experimentally, GeP flake thicknesses of 62–65 nm and 90–95 nm were tested on the MRR and SW structures, respectively.

The optical n, k parameters of the multilayer GeP were extracted using a spectroscopic imaging ellipsometer. This data was fed into the mode analysis of Lumerical Mode Solver Software (more details can be found in Supplement 1).

Figure 2(a) shows the simulated electric-field profiles (${|E |^2}$) of the fundamental guided modes of a bare silicon waveguide and the hybrid loaded structures with 65 nm GeP layer at 1310 nm. As can be seen, the silicon waveguide supports quasi-TE (transverse electric) and quasi-TM (transverse magnetic) modes. The waveguide parameters, the effective index of refraction, optical loss, and mode profile are shown in Supplement 1 Tables S1 and S2 for 1310 nm and 1550 nm wavelengths, respectively.

 

Fig. 2. Guided light-multilayer GeP interaction (a) Electric-field profiles (|E|²) of TE and TM modes of bare Si waveguide (top panel) and 65 nm GeP on Si (bottom panel) at 1310 nm (b) Eigen Mode Expansion (EME) Beam PROP simulation of 65 nm thick GeP layer with a length of 5 µm (top panel) and 35 µm (bottom panel) at 1310 nm (c) measured transmission spectra of TM mode in Si/GeP hybrid MRR with a radius of 50 µm and GeP thickness of 62–65 nm at 1520–1530 nm (top panel) and 1280 nm–1290 nm (bottom panel) (d) cut-back method of 90–95 nm GeP thickness for the TE mode of 1310 nm.

Download Full Size | PDF

It’s worth noting that the guided optical mode of the Si waveguide overlaps with the multilayer GeP flake for both polarizations. At 1310 nm, the calculated optical losses for the 65 nm thick flake are 1.1 dB/µm and 1.3 dB/µm for the TE and TM modes, respectively. The calculated light absorption corresponding to thicknesses of 30 nm, 65 nm, and 95 nm thick GeP are included in Supplement 1 in Table S3 and S4, for 1310 nm and 1550 nm, respectively. It is noticed that as the flake thickness increases above 100 nm, the optical mode resides more in GeP layer, hence this leads to a significant enhancement of absorption.

One important figure of merit is to evaluate the absorption in the GeP flake in terms of its waveguide coverage length. To get a better estimate of the needed flake length, we performed an Eigen Mode Expansion (EME) Beam PROP simulation. Figure 2(b) shows the propagation for the guided TE and TM modes of 1310 nm wavelength for a flake thickness of 65 nm. Results are performed for short (5 µm) and long flake (35 µm) lengths. For the short device, an absorption of ∼ 73% (−5.7 dB) and 79% (−6.8 dB) is observed for the TE and TM light polarizations, respectively. This absorption increases to 84% (−7.9 dB) and 85% (−8.2 dB) when the GeP thickness is increased to 95 nm. The corresponding results for the 1550 nm wavelength and 65 nm thick flake are 60% (−4 dB) and 74% (−5.8 dB) for the TE and TM polarizations, respectively.

These results are in good agreement with previous reports on GeP which demonstrate a broadband optical absorption from UV to MIR regime [39,40]. It is observed that for a 65 nm thick GeP active layer and a device length of ∼ 12–15 µm results in a complete absorption. This leads to a reduced footprint compared to the graphene ones, where > 100 µm length is required to obtain > 90% absorption [46]. A detailed Beam Prop simulation for different GeP layer coverages and thicknesses can be found in Supplement 1 Fig. S3 for the 1310 nm and 1550 nm wavelengths.

To experimentally verify the index of refraction modulation of the waveguide due to the GeP flake, we measured the optical transmission spectra of a MRR-based photodetector in the wavelength range of 1280 nm–1360 nm and 1500 nm–1600 nm. Figure 2(c) shows the MRR’s resonance shift for the TM mode before and after integrating a ∼ 62–65 nm thick GeP flake and with a coverage length of ∼30 µm. The corresponding response for the TE mode is included in Supplement 1 Fig. S4. The measured resonance shift per unit waveguide length is 0.02 nm/µm and 0.03 nm/µm at 1280 nm–1360 nm and 1500 nm-1600 nm, respectively. The ring response at different GeP coverages along with their SEM Images are shown in Supplement 1 Fig. S5. Similar observations have been previously reported for BP and MoTe₂, where, the loss in MRRs can be manipulated by controlling the coverage length and thickness [19,20].

Furthermore, to determine the optical losses due to multi-layer GeP, the cut-back technique is used, as shown in Fig. 2(d). The setup for this method is described in Supplement 1. Supplement 1 Fig. S6 shows SEM images of transferred GeP used in this experiment along with their thickness and coverage lengths. The measured optical losses are shown in the same figure. The absorption coefficient is obtained from the linear fit of the data in Fig. 2(d) and is evaluated as 1.54 ± 0.3 dB/µm. This value is in good agreement with our calculated results. Detailed results for the measured absorption coefficient as a function of GeP thickness and coverage lengths are depicted in Supplement 1 Fig. S7 for the 1310 nm and 1550 nm wavelengths. Optical losses in the GeP loaded waveguides can be attributed to combinations of material absorption, scattering due to flake edge irregularities, and scattering at the coupling interface between the waveguide and the flake. Based on the previous report, GeP is highly absorbing in the telecom wavelengths [39], hence losses are primarily due to material absorption. Additionally, we performed a detailed analysis of mode mismatching and reflection loss at the transition between the passive waveguide (air/Si) to the active region of the strip waveguide (GeP/Si). As can be seen in Supplement 1 Fig. S3 the resulting optical losses in this study are primarily due to the GeP absorption.

2.3 Steady-state photodetector response

The steady-state response of the straight waveguide (SW) and microring resonator (MRR) photodetectors (Fig. 1(a) and (d)) was tested for the TE and TM polarizations at both 1310 nm and 1550 nm wavelengths. As discussed in section 2, the optical mode in the Si waveguide overlaps evanescently with the absorbing GeP (the active section of the detector). The multilayer GeP absorbs photons via indirect band-to-band transitions which results in electron-hole pairs generation. Afterward, the photo-excited carriers are efficiently separated and collected by a lateral electric field applied between the in-plane electrodes. Figure 3(a) shows I–V characteristics of the straight waveguide-based photodetector with a 90-nm-thick GeP flake. Results are shown at dark and under 1310 nm TE mode coupling. The detector exhibits a very low dark current of about ∼2–8 nA at ±1 V bias (device area 281 µm²). The photocurrent as a function of applied bias for the device presented in Fig. 3(a) is included in Supplement 1 Fig. S8.

 

Fig. 3. Photo response of GeP-SW and GeP-MRR photodetector (a) I–V characteristics (semi-logarithmic plot) of the SW photodetector under 1310 nm TE mode light (red) and dark (black) conditions, inset shows SEM image of a 90 nm GeP integrated device (b) Photocurrent versus applied voltage for MRR photodetector at 1310 nm with TE and TM polarization, inset shows the electric field profiles (|E_x|) with 65–62 nm GeP on the top surface of the waveguide (c) Photocurrent versus applied voltage for MRR photodetector at 1550 nm with TE and TM light polarization, inset shows the electric field profiles (|E_x|) with 62–65 nm GeP on the top surface of the waveguide (d) the spectral transmission response of the 1310 nm TM mode of MRR-integrated photodetector along with the photocurrent values at the resonance and off resonance conditions (e) the spectral transmission response of the 1550 nm TM mode of MRR-integrated photodetector along with the photocurrent values at the resonance and off resonance conditions (f) the responsivity and EQE as a function of bias voltage for two devices (flake thickness of 90 nm on SW with coverage length 28.1 µm; flake thickness of 65 nm on MRR with coverage length 31.3 µm and ring radius of 50 µm).

Download Full Size | PDF

The I–V measurements of devices integrated into micro ring resonator with a 65-nm-thick GeP are plotted in Fig. 3(b) and Fig. 3(c) for 1310 nm and 1550 nm, respectively. Results are shown for both polarizations. As it can be seen, the magnitude of the photocurrent increases with the applied bias and it is positive relative to the bias voltage. This is unlike the bolometric effect where the photocurrent has an opposite direction to the applied bias [19,23,47]. Additionally, the design of our photodetectors has structural symmetry, hence the photo-thermoelectrical effect (PTE) can be neglected [47,48]. Therefore, our photodetectors exhibit similar behavior to previously reported 2D photodetectors based on photoconductive effects [20,33]. A summary table of 2D materials/Si waveguides operating in the O- and C-optical bands exploiting different photo generation mechanisms is included in Supplement 1 Table S6. Furthermore, the results indicate that the photocurrent has higher values with TE mode. This is consistent with our modal analysis where strong light confinement in the GeP flake is observed for the TE mode compared to the TM polarization. The TM mode leaks in the air and toward the substrate, exhibiting less optical confinement in the active region, hence lower absorption. As a result, it was difficult to observe the ring’s resonance for the TE polarization due to strong absorption (see Supplement 1 Fig. S4).

The optical transmission spectra of the MRR-integrated photodetector for the TM mode along with its photocurrent response are presented in Figs. 3(d) and 3(e) for 1310 nm and 1550 nm wavelengths, respectively. As noticeable, the photocurrent maximum is at the resonance wavelength (1.284 µm and 1.525 µm) which shows a ∼ 40% enhancement in the photocurrent compared to the off-resonance condition.

Figure 3(f) shows the photoresponsivity and the external quantum efficiency as a function of applied bias for the waveguide and MMR detector configurations. For the TE polarization, devices exhibit a high photoresponsivity (external quantum efficiency) corresponding to 0.54 A/W (51.3%) and 0.43 A/W (40.5%), for SW and MRR, respectively. These values are obtained at 1310 nm under V_SD = −5 V bias. The corresponding value measured from the MMR at 1550 nm wavelength, not shown in the figure, is 0.24 A/W at V_SD = −5 V bias. It is worth mentioning that the resonator design enables a strong light-matter interaction due to higher intra-cavity energy density compared to a straight waveguide [46]. Hence, the MRR amplifies the detector response at the resonance wavelengths. Our results show a photoresponsivity for the 62–65 nm thick GeP flakes comparable to the 90–95 nm flakes integrated into a straight waveguide.

The power dependence of the photocurrent has been discussed in detail in our previous work for a surface illuminated GeP detector [39]. In that work, a saturation of the absorbed optical power is observed for higher illumination intensities. In this work, a similar behavior is observed for waveguide integrated GeP photodetectors. The coupled optical power to the device is in the range of 6–12 µW. Supplement 1 Fig. S8 shows the response which depicts a nonlinear behavior. An exponent of 0.34 is extracted from nonlinear curve fitting. This non-unity exponent 0 < α < 1 is frequently observed in photodetector based on 2D materials [47,49–53]. This is primarily attributed to the complexity of the carrier generation process, density of states, trapping, and recombination within 2D semiconductors [50,54]. It is worth mentioning that an incident power density of two orders of magnitude lower implies a responsivity of two orders of magnitude higher for photodetector with α close to zero.

2.4 Dynamic photoresponse

To get more insight into the dynamic performance of the photodetector, its frequency response was tested at 1310 nm light. In this experiment, a tunable laser was used as the continuous light source, which was modulated using a high-speed optical modulator. The details of the experimental setup are depicted in Supplement 1 Fig. S9. Figure 4 shows the integrated GeP photodetector normalized frequency responses evaluated over the frequency range from 70 kHz to 10 MHz and measured using a lock-in amplifier. A 3-dB bandwidth of 0.4 MHz at V_bias =9 V (see Fig. 4(a)) was measured for the waveguide integrated photodetector.

 

Fig. 4. Dynamic photoresponse of integrated GeP photodetector (a) normalized frequency response at 9 V, inset shows a schematic of measurements setup (b) normalized frequency response as a function of applied voltage showing enhanced 3-dB bandwidth value at higher applied bias (c) alternating current Capacitance-voltage (C-V) curve (f = 1 MHz), inset shows the equivalent circuit of the waveguide integrated GeP photodetector (d) Noise equivalent power (NEP) of the photodetector, red and blue are the total noise equivalent power and the shot noise as a function of applied bias, respectively.

Download Full Size | PDF

Furthermore, we tested the effect of the applied voltage on the frequency response. We noticed that the 3-dB bandwidth increases with the applied bias, see Fig. 4(b). This can be attributed to the rapid separation of the photo-generated carriers at large electric fields. Additionally, the observed sensitivity of the bandwidth to applied bias indicates that drift-diffusive transport governs the speed response of the present photodetectors.

Typically, the 3-dB bandwidth is limited by the carrier transit time, and the RC constant [22]. Thus, it is critical to determine the effective capacitance of our device which could impose a limit on the detector’s time response. For this purpose, we measured the C-V characteristics of the detectors as shown in Fig. 4(c). The measured capacitances are in the range of 0.3 pF. Inset of Fig. 4(c) depicts the equivalent circuit of the present photodetector. From this configuration, we believe that contact resistance can be improved by having a buried contact where the contact pads are deposited after GeP transfer. Additionally, introducing graphene as a contact electrode can further reduce its resistance.

Based on the above results we discern that the temporal response of the present photodetectors is mainly governed by transit time. This is similar to the initial results reported on BP photodetectors [33,55]. Several studies have shown that by optimizing the photodetector designs, BP demonstrated a high-speed operation of a few GHz [22]. For instance, transit time is inversely proportional to mobility, hence, sandwiching GeP between other 2D materials such as graphene or hBN can enhance mobility. Additionally, transit response is proportional to the channel length, thus the speed of the photodetector can be further advanced by reducing the gap distance between electrodes (in this design 3–5 µm).

On the other hand, the in-plane electrodes (lateral electric field) can be reconfigured into a vertical design, where the transit path for the generated carriers is primarily controlled by the active layer thickness (few nm). Hence, we believe that the performance of GeP photodetectors can be optimized in several ways to achieve high-speed operation for NIR applications.

To evaluate the minimum detectable power of the photodetector, we determined its noise equivalent power (NEP). This value provides the amount of excitation power that generates a photocurrent equal to the noise current (NEP = i_n/R, where, i_n is noise current and R is the responsivity) [56]. Typically, the noise consists of three sources: 1/f noise, shot noise, and Johnson noise. However, at a high-speed signal > 1 kHz, the total noise current is determined by the shot noise and the Johnson noise [47,57]. Shot noise is dominated by the dark current (${i_n} = \sqrt {2q{I_D}\Delta f} $, q is the electric charge, I_D is the dark current, Δf is the bandwidth). Figure 4(d) shows the shot noise of the fabricated detector as a function of applied voltage. This is relatively close to the range reported previously for mid-infrared BP photodetector [58]. The thermal noise (Johnson noise) mainly depends on the channel resistance according to the following equation: ${i_n} = \sqrt {\frac{{4{k_B}T\Delta f}}{{{R_0}}}} $, k_B is Boltzmann constant, T is the temperature, R₀ is the channel resistance. At room temperature, the Johnson noise of our photodetector is ∼18 pA/Hz^1/2, which is close to the shot noise values. Therefore, we considered the total noise as the sum of the shot noise and Johnson noise. As can be seen in Fig. 4(d), the NEP at V = 5 V is 229 pW/Hz^1/2, this sensitivity is higher than the graphene’s photodetectors attributed to the lower dark current in the GeP photodetector [59].

3. Conclusion

In summary, we demonstrate high-performance waveguide and micro-ring resonator hybrid-integrated Si/GeP photodetectors for the O- and C-optical bands. The as-assembled devices exhibited a remarkable responsivity of 0.54 A/W and 0.24 A/W at 1310 nm and 1550 nm, respectively. Additionally, the detectors show a low dark current of about ∼2–8 nA and a fairly low NEP of 229 pW/Hz^1/2. Furthermore, a 65 nm thick GeP with coverage length of ∼ 12–15 µm results in a complete absorption, which leads to a reduced footprint. For the present devices, the measured frequency response 3-dB bandwidth is 0.4 MHz. The observed dependence of the bandwidth on the applied bias indicates that drift-diffusive transport governs the photodetector’s speed. It is possible to further improve the device bandwidth by sandwiching GeP between other 2D materials such as graphene or hBN and by reducing the gap distance between the contact electrodes (in this design 3–5 µm). The remarkable photoresponse of 2D GeP at SWIR can be extended to a longer wavelength giving its narrow bandgap. This report does not only provide an example of efficient and robust SWIR waveguide integrated photodetectors based on multilayer GeP, but it also opens prospects for extending the 2D family toward novel MIR optoelectronics applications.

4. Method

4.1 Materials supply

GeP flakes were purchased from a 2D semiconductors vendor (https://www.2dsemiconductors.com/).

4.2 Scanning electron microscopy

Photonic chips were mounted on a SEM stub using carbon tape and imaged under high vacuum mode by using a (FEI) Quanta 450 field emission scanning electron microscope with electron energy of 10 KV.

4.3 Atomic force microscopy

Atomic Force Microscopy was performed using WITec Atomic Force Microscope (AFM) module integrated with a research-grade optical microscope in the tapping mode. The cantilever tip (Scanasyst-air) had a radius of 7 nm, a force constant of 0.2 N/m, and a resonance frequency of 14 kHz.

4.4 Mode analysis and FDTD simulation

The electric field profile in the silicon waveguide and the beam propagation was calculated using the MODE Solutions eigenmode solver, a simulator within Lumerical’s Device Multiphysics Simulation Suite.

4.5 Spectroscopic imaging ellipsometer

The optical parameters of multilayer GeP were determined by Accurion’s Imaging Ellipsometry (https://accurion.com/company). This system combines optical microscopy and ellipsometry for spatially resolved layer-thickness and refractive index measurements. It is highly sensitive to single- and multi-layer ultrathin films, ranging from mono-atomic or monomolecular layers (sub-nm regime) up to thicknesses of several microns. Additionally, Imaging Ellipsometers perform layer thickness measurements with a spatial resolution down to 1 µm. The ellipsometric parameters (Psi (ψ) and Delta (Δ)) were fitted using EP4 model software.

4.6 Optical characterization

The optical transmission was tested by edge coupling the light into the device structure through lensed fiber using a tunable laser operating at O- and C- optical bands (Keysight 8164B Lightwave Measurement System). The output response from the devices is collected by an output lensed fiber and detected by a power meter. The light polarization (TE/TM) was calibrated using reference rings fabricated on the same chips with identical geometries. The output optical power intensities were calibrated before the device testing using a standard photodiode power sensor.

4.7 DC measurements

The steady-state performance of the GeP photodetectors was tested by measuring their dark current and responsivity. A transverse electric-polarized (TE) or transfer magnetic-polarized (TM) 1310/1550 nm light was edge coupled via lensed optical fiber to the devices. A curve tracer/power device analyzer / (Agilent B1505A) was used to control the biases and measure the I-V characteristics in the dark and upon light coupling via a pair of standard DC electrical probes.