1. Introduction

For the past decades, Si photonics has benefited from the use of mutual semiconductor facilities and its volume yields, providing high-performance and low-cost photonic integrated circuits (PICs). Mid-infrared (MIR) group IV photonic devices and systems have also witnessed this remarkable development in various applications including environmental/biochemical sensing, industrial process control, toxic substance evaluation, etc [1–3], and their potential use for the next generation of communication bands [4]. To extend available MIR wavelengths [5], a number of Si photonics platforms including Si-on-Si₃N₄ (SON) [6], Si-on-sapphire (SOS) [7], and suspended Si [8] have intensely been developed. However, all of these platforms suffer from the absorption of Si itself that is transparent only up to a wavelength of approximately 8 µm [9], making it undesirable to cover the most of molecular fingerprint regions [10].

As compared with Si, there are many alternate materials with broader transparency windows in the MIR spectrum, such as chalcogenide glasses [11] and materials in groups III–V [12]. Moreover, Ge has also been considered an ideal platform for MIR applications owing to its wide transmission window up to approximately 15 µm and its compatibility with modern complementary metal-oxide-semiconductor (CMOS) foundries [13]. A Ge-on-Si (GOS) platform [14–18] is promising for using Ge as the core material of a waveguide. A Ge-on-insulator (GeOI) platform [19] is another promising platform that enables the fabrication of ultracompact Ge PICs for MIR wavelengths. Using a GeOI wafer fabricated using Smart-cut [20], we have developed a number of Ge passive/active devices including Ge waveguides [21,22], high-Q Ge resonators [23,24], tunable thermo-optic devices [25–27]. In addition, we have successfully fabricated a suspended Ge waveguide that enables us to use the entire MIR spectrum by removing the SiO₂ buried oxide (BOX) layer [20,23,24]. However, our previous results showed a relatively high propagation loss even with a wide Ge waveguide [19,21,25], which restricted practical applications of Ge waveguides to large-scale GeOI MIR PICs. In particular, Ge exhibits the large hole-induced free-carrier absorption due to the intervalence band transition [28]. Moreover, the suppression of the propagation loss is the key to investigating the nonlinearity of Ge such as supercontinuum generation and frequency combs [29–31]. Although the large free-carrier absorption in Ge makes it a more suitable candidate for developing an optical variable attenuator in the MIR spectrum than Si [20,22], n-type Ge should be used to prevent the hole-induced absorption for a low-loss Ge waveguide. In the case of GOS, the carrier conduction type and density can be controlled during epitaxial growth to obtain a low-loss n-type Ge waveguide [32], whereas the Ge film on a GeOI wafer tends to show p-type conduction even when an n-type Ge wafer is used for Smart-cut owing to the generation of hole induced by crystal defects formed by hydrogen implantation. Thus, the fabrication of an n-type GeOI wafer must take into account the effect of the implantation process on the crystal defect generation.

In this work, we investigated the impact of the implantation energy of hydrogen ions on the distribution of crystal defects in Ge. We found that the higher the ion implantation energy, the deeper from the Ge the surface crystal defects are generated. As a result, the density of crystal defects near the Ge surface was markedly decreased. By high-energy implantation, we successfully fabricated an n-type GeOI wafer from an n-type bulk Ge wafer with a moderate doping concentration. We evaluated the propagation loss of Ge waveguides on n-type and p-type GeOI wafers, and we revealed that the n-type GeOI wafer exhibited the lowest propagation loss.

2. Fabrication of the Smart-cut GeOI wafer

The process flows for the GeOI wafer are illustrated in Fig. 1. Firstly, a 2-inch Ge bulk wafer was pre-cleaned with hydrofluoric acid (HF) solution. Subsequently, we deposited a 100-nm-thick SiO₂ hard mask on the Ge wafer for hydrogen ion implantation as follows. Buffered hydrofluoric acid (BHF) solution was used to clean the hard mask after implantation. Next, the Ge surface was passivated by plasma oxidation [33] before wafer bonding. A 725-µm-thick 2-inch-long Si handling wafer with a 2-µm-thick SiO₂ BOX layer formed by thermal oxidation was also prepared in advance. Consequently, a 5-nm-thick Al₂O₃ layer was deposited simultaneously on the Ge and Si handle wafers by atomic layer deposition (ALD). Next, we manually flipped and bonded the Ge wafer onto the BOX layer of the Si wafer, followed by vacuum annealing at 250 °C under a pressure of 1500 N for 30 min. Thermal splitting was carried out by increasing the annealing temperature gradually from 250 °C to 400 °C on a hotplate in steps of 50 °C per 3 h in atmospheric ambient, and splitting was finally triggered after several hours (∼3 h for 80 keV implantation and ∼15 h for 160 keV implantation) at 400 °C depending on the ion implantation energy. Subsequently, the GeOI wafers were ground by chemical mechanical polishing (CMP) using a commercial alkali slurry (COMPOL 20). Finally, the wafers were annealed at 550 °C for 1 h at a high vacuum (<1×10⁻⁵ mbar) [34].

 

Fig. 1. Process of fabricating Smart-cut GeOI wafer.

Download Full Size | PDF

3. Impact of hydrogen ion implantation energy on crystal defect distribution

In our previous work [33], the initial n-type Ge donor wafer was converted to the p-type Ge donor wafer after thermal splitting owing to the crystal damage induced by hydrogen ion implantation with an ion implantation energy of 80 keV. A GeOI wafer fabricated from an n-type Ge donor wafer with an electron concentration of approximately 3×10¹⁶ cm⁻³ exhibited a p-type conduction with a hole concentration of approximately 1×10¹⁶ cm⁻³.

To reduce the defect density in the Ge device layer, we examined a higher hydrogen ion implantation energy. We performed the stopping and range of ions in matter (SRIM) simulation [35] with implantation energy ranging from 40 keV to 200 keV to evaluate the distribution of hydrogen ions and the irradiation damage. The damage is represented by displacement per atom (dpa) in implanted Ge, as more often discussed for Si [36–38]. We used “Full Damage Cascade” as the simulation type with over 8000 ions to get the reliable distribution of hydrogens in Ge. Figures 2(a) and (b) show the simulated profiles of hydrogen ions and dpa with implantation energy from 40 keV to 200 keV, respectively, at an ion implantation dose of 4×10¹⁶ cm⁻². A 100-nm-thick SiO₂ capping layer was set on the Ge surface as a hard mask. SRIM results showed that the peak positions of the H⁺ concentration and dpa become deeper from the Ge surface when the implantation energy increases. When the implantation energy is 80 keV, which was used in our previous studies, peak tails are still observed in the Ge layer near the surface used for the devices. By using implantation energy as high as 160 keV, one can significantly reduce dpa near the Ge surface. Thus, we expect that the density of defects generated by hydrogen ion implantation can be reduced by using higher implantation energy. To examine this concept, we performed the transmission electron microscopy (TEM) observation of the as-split GeOI wafers fabricated using the implantation energies of 80 keV and 160 keV with an implantation dose of 4×10¹⁶ cm⁻², as shown in Figs. 2(c) and (d), respectively. The thicknesses of the Ge layer after splitting were approximately 650 nm and 1270 nm for 80 keV and 160 keV, respectively, which match the H⁺ peak positions in the SRIM simulation since the distribution profile of vacancy-related point defects implied by the dpa is related to the hydrogen platelets formed during splitting. It was observed that the hydrogen defects penetrated much deeper into the Ge film at the higher implantation energy, leaving a Ge region without defects at a thickness of more than 300 nm from the Ge surface. Hence, we expect that higher implantation energy will help fabricate an n-type GeOI wafer by reducing the density of crystal defects, which induce the generation ion of holes in Ge.

 

Fig. 2. (a) Simulated distributions of hydrogen ions and (b) displacement per atom by collision after implantation in bulk Ge wafer covered with 100-nm-thick SiO₂ hard mask with an implantation dose of 4×10¹⁶ cm⁻². The implantation energy increases from 40 keV to 200 keV in steps of 40 keV. TEM images of as-split GeOI wafers with I/I implantation energies of (c) 80 keV and (d) 160 keV.

Download Full Size | PDF

4. Characterization of the carrier profile in Ge films by Hall measurement

To evaluate the impact of the implantation energy of hydrogen ions on the carrier profile in the split Ge layer, we performed the Hall measurement of the GeOI samples fabricated under the three conditions listed in Table 1. Samples A and B were fabricated at the same implantation energy of 80 keV as previously reported using Ge donor wafers with different doping concentrations, whereas Sample C was fabricated at 160 keV. The implantation dose was fixed to 4×10¹⁶ cm⁻². Note that a higher implantation dose will lead to further damage of the Ge film and the generation of more holes [38]. Thus, in this work, we consider the implantation dose of 4×10¹⁶ cm⁻². We used n-type Ge donor wafers with two doping concentrations to evaluate the n-dopant compensation effect. The Ge wafer for samples B and C had an electron density one order of magnitude greater than that for Sample A.

Table 1. Conditions for Smart-cut processes and type of Ge bulk wafer

View Table | View all tables in this article

We fabricated the Hall devices on GeOI wafers as follows. After BHF cleaning to remove the native oxide layer, the 500-nm-thick Ge film was patterned and etched by reactive-ion etching (RIE). Then, the Ge surface was passivated with Al₂O₃ by ALD. Finally, vias were opened by treating with BHF, and Nickle (Ni) electrodes were formed by sputtering and lift-off processes. We prepared the GeOI samples with different Ge film thicknesses to evaluate the thickness dependence of the effective sheet carrier density ${N_{eff}}$. Samples were thinned by CF₄ RIE followed by surface cleaning with diluted HF solution to remove the native oxide layer and RIE-induced defects.

Figure 3(a) shows ${N_{eff}}$ of Sample A as a function of Ge film thickness. Even when using the n-type Ge donor wafer, the Hall measurement indicated that the polarity of the net charge was positive when the Ge film thickness was greater than 150 nm, owing to the generation induced by defects. As the Ge film thickness decreases, ${N_{eff}}$ decreases, suggesting that there are fewer defects near the bonding interface. When the Ge film thickness is less than 100 nm, the polarity of the net charge becomes negative, which means that the Ge layer near the surface is of the n-type under conventional ion implantation conditions. Figure 3(b) shows ${N_{eff}}$ of Sample B. This sample shows similar trends to Sample A but with a wider Ge range of the n-region. We believe that such a difference is due to the local n-dopant compensation with p-type defects caused by 80 keV implantation since the doping concentration of the Ge donor wafer for Sample B is greater than that for Sample A. As shown in Fig. 3(c), Sample C exhibited a homonymous n-type conductivity, owing to the deeper defect penetration that mitigated the generation of holes near the bonding interface. Figure 3(d) shows ${N_{eff}}$ of Sample C with a wider Ge film thickness range. It was found that the polarity of the net charge for Sample C is positive when the Ge film thickness is greater than 800 nm, indicating a high density of defects at Ge film thickness larger than 800 nm. Finally, we successfully obtained an n-type GeOI wafer by using implantation energy of 160 keV when the Ge thickness was less than 700 nm. Although the CF₄ etching is reported to introduce deep-level accepter-like defects into the Ge films [39], we consider this effect might be negligible by removing the damaged layer through the surface cleaning. Note that the high implantation energy and the thinning process by CMP and RIE etching might be still preferred to obtain a further thinner Ge layer.

 

Fig. 3. Effective sheet carrier density as a function of Ge film thickness: (a) Sample A (I/I energy: 80 keV, Ge donor wafer: ∼3×10¹⁵ cm⁻³), (b) Sample B (I/I energy: 80 keV, Ge donor wafer: ∼4×10¹⁶ cm⁻³) and (c) Sample C (I/I energy: 160 keV, Ge donor wafer: ∼4×10¹⁶ cm⁻³). (d) Effective sheet carrier density as a function of Ge film thickness with an extended range of Ge film thickness for sample C. All samples were annealed at 550 °C for 1 h.

Download Full Size | PDF

5. Evaluation of propagation loss of Ge rib waveguides

Using the three GeOI wafers (Samples A, B, and C), we fabricated spiral waveguides based on Ge rib waveguides to evaluate the impact of free carriers on the propagation loss. To eliminate the error from the bend loss, we designed a series of Ge spiral waveguides maintaining counts of 90-degree bends with a radius of 25 µm. The waveguide width was designed to be 1 µm. The minimum and maximum lengths of the spiral waveguides were 5.766 mm and 15.566 mm, respectively. The Ge spiral waveguides were fabricated with the process shown in Fig. 4(a). Firstly, the Ge film was thinned to approximately 320 nm by RIE with CF₄, noting that the Ge thicknesses were the same for all samples. The grating couplers were patterned with a pitch of 680 nm by electron-beam (EB) lithography using a ZEP-520A EB resist. Then, the grating was etched by 90 nm to obtain a coupling center at a wavelength of 1.95 µm. Subsequently, rib waveguides with a 100-nm-thick slab were formed, followed by the formation of a 400-nm-thick SiO₂ layer by plasma-enhanced chemical vapor deposition (PECVD). The optical microscopy image of the Ge spiral waveguides is also shown in Fig. 4(a). The cross-sectional TEM image of the Ge rib waveguide in Fig. 4(b) shows that the rib structure is well-formed, suggesting a negligible sidewall scattering loss when the waveguide width is larger than 1 µm [21]. Afterward, we measured the propagation loss by the cut-back method. A fiber-coupled amplified spontaneous emission light with a center wavelength of 1.95 µm was coupled to the Ge rib waveguide through the grating coupler. The output power from the waveguide was coupled again to a single-mode fiber and measured using a spectrum analyzer (Thorlabs, OSA203C). The simulated transmission spectrum of the designed Ge grating coupler was shown in Fig. 4(c). Figure 4(d) shows the transmission spectra of the shortest and longest waveguides. The oscillations seen in the spectra are due to the mode interference through the higher-mode excitation in the Ge spiral waveguides, which can be eliminated using a single-mode bend waveguide [40].

 

Fig. 4. (a) Process of fabricating Ge spiral waveguides (inset shows optical microscopy image), (b) Cross-sectional TEM image of Ge rib waveguides, (c) simulated transmission spectrum of the designed grating coupler, (d) transmission spectra of baseline and 7.98-mm-long Ge spiral waveguides.

Download Full Size | PDF

Figure 5(a) shows the measured output powers of Samples A, B, and C obtained by the integration of the total transmission spectrum to minimize the influence from the mode interference. The propagation loss of Sample A with an average p-type carrier density of 3.9×10¹⁶ cm⁻³ was found to be 25.2 ± 0.1 dB/cm. As discussed in Fig. 3, Sample B had a more n-type Ge layer near the bonding interface owing to larger n-type doping concentration of the Ge donor wafers as listed in Table 1. Thus, the propagation loss of Sample B with an average p-type carrier density of 3.4×10¹⁶ cm⁻³ decreased to 10.7 ± 0.2 dB/cm. Moreover, the propagation loss of the n-type Ge waveguide (Sample C with an average n-type carrier density of 4.7×10¹⁶ cm⁻³) was 2.3 ± 0.2 dB/cm owing to the small free-carrier absorption of electrons in Ge. Furthermore, we analyzed the relationship between the carrier concentration and the propagation loss using the free-carrier absorption (FCA) model in Ge [28], as shown in Fig. 5(b). We found that Samples B and C matched the FCA model, whereas sample A had a greater loss than expected by the FCA model probably due to more crystal defects in Sample A.

 

Fig. 5. (a) Propagation losses of Samples A, B, and C evaluated by cut-back method; (b) The relationship between the carrier concentration and propagation loss. The black and red lines are calculated using the free-carrier absorption model in Ge.

Download Full Size | PDF

Finally, we benchmarked our work with other Ge-based photonic platforms, as listed in Table 2. To the best of the authors’ knowledge, the presented propagation loss has the lowest value among all of the ion-cut based GeOI photonic platform and shows comparable value as compared with conventional GOS and Ge-suspended photonic platforms.

Table 2. Benchmarking of the propagation loss of reported Ge waveguides at MIR spectra

View Table | View all tables in this article

6. Conclusions

We investigated the impact of the implantation energy of hydrogen ions on the crystal quality of Smart-cut Ge-on-insulator wafers. We found that the higher the implantation energy, the deeper the crystal defects are generated in Ge from the implanted surface. As a result, the density of crystal defects in the Ge device layer decreased after splitting. By increasing the implantation energy from 80 keV to 160 keV, we can effectively suppress the hole generation induced by the crystal defects in the ∼700-nm-thick Ge layer from the implanted surface. Finally, an n-type GeOI wafer was successfully obtained. We found that the propagation loss of the Ge rib waveguide can be markedly reduced by using the n-type GeOI wafer. This finding paves the way towards large-scale MIR Ge PICs on a GeOI platform for on-chip sensing applications.