1. Introduction

Multi-level modulation formats, such as 4 pulse amplitude modulation (PAM-4) and 8 pulse amplitude modulation (PAM-8), have the promise to meet the demands of high-capacity data communication in datacenters [1,2]. High-power and high-linearity photodiodes are of importance in these multi-level communication systems. In addition, considering the increased interest in the applications of optical analog links, microwave photonics, and terahertz photonics, high-power and high-speed performance have been the main requirements for photodiodes [3–5]. On the other hand, as a crucial component of the silicon photonics circuits, the silicon-germanium (Si-Ge) photodiode has been developing rapidly thanks to the high-performance Ge thin film epitaxial growth on silicon [6,7].

However, the power handling capability of Si-Ge photodiode is still limited by the space-charge screening (SCS) effect. Under strong light illumination, the photo-absorption region of the photodiode accumulates a high density of photo-generated carriers, inducing a strong space-charge field that screens the externally applied bias field. The weakened field will reduce the drift velocity of the photo-generated carriers and further degrade the bandwidth and output power performance of the photodiode. Previous studies have been done from mainly three different approaches to overcome the SCS effect. The first solution is to distribute the photocurrent into several sub-parts, such as traveling-wave photodiode [8,9] and arrayed photodiode [10,11]. Another solution is to uniformly distribute the photocurrent within a single photodiode based on the evanescently optical coupling method. Several works have been reported under such a scheme, including mode-evolution-based coupler photodiode [12], edge-coupled photodiode [13,14], and dual-fed photodiode [15]. The third way is to regulate the doping condition, which can fundamentally overcome the limitation of the SCS effect. Some investigations are reported through doping concentration control in the absorption layer, such as the uni-traveling-carrier photodiode (UTC-PD) [16,17]. This method has also been extensively used in photodiodes based on III-V material [18–21]. However, it is difficult to realize the flexible doping concentration control in the absorption layer with the commercial Si-Ge Foundry processes. Fortunately, besides the doping concentration control in the absorption layer, there is also an effective method to regulate the doping condition in the collection layer [22].

In this paper, we investigate the doping condition control in the collection layer of the waveguide Si-Ge photodiode. Based on the conventional fabrication process, we design and demonstrate a high-power Si-Ge photodiode assisted by doping regulation. The optimized photodiode generates a 36.4-mA direct-current (DC) photocurrent under an input optical power of 40 mW, which is 85.7% improvement compared to 19.6 mA of the conventional structure. Aside from that, it achieves the maximum saturated photocurrent density of 1.82 mA/μm³. Furthermore, the radio-frequency (RF) saturation performance of the proposed photodiode exhibits a 57% improvement in -1 dB compression photocurrent measurement. The eye-diagram of the large-signal also shows a significant improvement under high photocurrent.

2. Device design

The 3D illustration of the waveguide Si-Ge photodiode is illustrated in Fig. 1(a). The incident light transmits from the channel silicon waveguide to the Ge absorption region through a 20-μm length tapered coupler. The length and width of the absorption region are 8 and 5 μm, respectively. The schematic cross-section diagram of the photodiode with a conventional design is shown in Fig. 1(b). The detailed structure and doping parameters are listed in the plot. After P⁺ and heavily P⁺⁺ doping in the silicon layer for carriers collection and ohmic contact, a 0.5-μm thickness Ge thin film is selectively grown on the silicon layer, with a 50-nm top Ge layer being heavily implanted by phosphorus for ohmic contact. Typically, the average doping concentrations are controlled at the level of ∼1 × 10¹⁹ cm^-3 for P⁺-Si, ∼1 × 10²⁰ cm^-3 for P⁺⁺-Si, and ∼1 × 10²¹ cm^-3 for N⁺⁺-Ge [23,24].

 

Fig. 1. (a) The 3D illustration of the waveguide Si-Ge photodiode. The length and width of the Ge absorption region are 8 and 5 μm, respectively. The schematic cross-section diagrams of (b) the conventional photodiode and (c) the optimized photodiode.

Download Full Size | PDF

We have preliminarily validated that the high-power handling capability of the photodiode strongly depends on the doping concentration of the collection layer near the absorption layer [25]. The collection layer just like the dike. Its doping concentration is the threshold of the photo-generated carriers concentration in the absorption layer. Increasing the doping concentration of the collection layer over the photo-generated carriers concentration will enhance the built-in electric field and eliminate the SCS effect significantly. Typically, the doping concentration regulation is not flexible and only several kinds of P-type doping concentration can be chosen [23]. To overcome these challenges, we regulate the width of the P⁺ doping region and introduce the heavily P⁺⁺-type doping to the region under the Ge absorption region. As shown in Fig. 1(c), the width of the P⁺ region reduces from 7 to 2 μm in the optimized scheme. On the one hand, the quality of the Ge thin film epitaxially grown on a fully p⁺⁺-type doped silicon layer may not be great [24]. On the other hand, the Ge mesa has a ∼30° sidewall and the width of the top is only ∼3.26 μm [24]. Therefore, the majority of the photo-generated carriers concentrate in the middle 3-μm range. With a comprehensive consideration, we design the width of the P⁺ region to be 2 μm.

Firstly, the optical simulation is implemented through the finite-difference time-domain method. The simulated wavelength is set at 1.55 μm. The simulated optical mode profile in the XZ plane and the XY plane are shown in Figs. 2(a) and 2(b). Due to the butt-coupling approach, strong optical power absorption is concentrate in the first few micrometers of the photodiode. Leveraging the optical power per unit volume, we calculate the number of photo-generated carriers per unit volume. As shown in Figs. 2(c) and 2(d), they are the simulated average photo-generated carriers distribution of the cross-section at the front and the end 1-μm Ge absorption region along the direction of light propagation, respectively. Considering a 10-mW input optical power and 1-ns carrier lifetime, the peak photo-generated carriers concentrations are ∼1.8 × 10¹⁹ cm^-3s^-1 and ∼8 × 10¹⁸ cm^-3s^-1, respectively. It implies that the photo-generated carriers concentration and doping concentration of the collection layer are comparable at 10-mW input optical power. As shown in Fig. 2(e), the simulated absorption ratio of the 8-μm length photodiode is 0.85, which corresponds to a responsivity of 1.06 A/W.

 

Fig. 2. (a)-(b) The simulated optical mode profiles in (a) XZ plane at Y=0 μm and (b) XY plane at Z=0.47 μm. (c)-(d) The simulated average photo-generated carriers concentration distribution of the cross-section at (c) the front and (d) the end 1-μm Ge absorption region along the direction of light propagation. (e) The simulated absorption ratio versus the length of the Ge region of the photodiode.

Download Full Size | PDF

Next, the electrical simulation is performed with 300 K simulated temperature by utilizing the photo-generated carriers concentration distribution. As shown in Figs. 3(a) to 3(c), they present the hole density of the photodiodes with 7 and 2-μm P⁺ width, under different input optical power of 1 and 30 mW but the same bias voltage of 0 V. Moreover, Figs. 3(d) to 3(f) show the simulated electric field profiles corresponding to Figs. 3(a) to 3(c) at a reverse bias voltage of 3 V. For the 7-μm P⁺ width photodiode, as shown in Fig. 3(a), the hole density in the Ge absorption region is much lower than that in the P⁺ doping collection region when inputting 1-mW optical power. Fig. 3(d) shows that its electric field in the Ge absorption region is controlled at the level of ∼5 × 10⁶ V/m, which is strong enough to ensure the carriers transporting at the saturation velocity. While increasing the input optical power to 30 mW in Fig. 3(b), the hole density in the Ge absorption region is comparable to that in the P⁺ doping collection region. Meanwhile, the electric field sharply declines to ∼1 × 10⁴ V/m due to the strong SCS effect, as shown in Fig. 3(e). The neutralize electric field slows down the carrier transporting velocity and increases the carrier transit time and the probability of carrier recombination. Subsequently, the more carrier recombination within the photodiode will decrease the output photocurrent, which eventually leads to saturation. In contrast, as shown in Fig. 3(c), the hole density in the Ge absorption region of the 2-μm P⁺ width photodiode is lower than that of the 7-μm P⁺ width photodiode at the same condition of 30-mW input optical power. Meanwhile, due to the enhanced built-in electric field, Fig. 3(f) indicates that the electric field still maintains a high level of ∼1 × 10⁶ V/m, ensuring the effective carrier transporting velocity. The simulation results verify theoretically the feasibility of doping regulation to enhance the high-power handling capability of the photodiode.

 

Fig. 3. (a)-(c) The simulated hole density distribution of the 7 and 2-μm P⁺ width photodiodes cross-section at input optical power of 1 and 30 mW. (d)-(f) The simulated electric field profile of the corresponding situation for (a)-(c) at a reverse bias voltage of 3 V.

Download Full Size | PDF

3. Device fabrication and characterization

3.1 DC electrical performance and bandwidth

All devices are fabricated on a silicon-on-insulator (SOI) wafer with a 2-µm buried oxide layer and a 220-nm top silicon layer. It leverages the conventional foundry fabrication process at the silicon photonics platform of Advanced Micro Foundry Pte. Ltd. (AMF), Singapore. As shown in the inset of Fig. 4(a), it is the top-view scanning electron microscope (SEM) image of a fabricated waveguide Si-Ge photodiode. For comparison, the photodiodes with P⁺ width of 3 and 5 µm are fabricated simultaneously.

 

Fig. 4. (a) Experiment setup for characterizing the performances of the Si-Ge photodiodes. The red lines denote optical fiber connections and the blue lines denote electrical connections. Channel 1, 2, and 3 are used for bandwidth, RF-power, and large-signal transmission measurements, respectively. LD, laser diode; PC, polarization controller; MZM, Mach Zehnder modulator; OT, optical transmitter; EDFA, erbium-doped fiber amplifier; VOA, variable optical attenuator; DUT, device under test; VNA, vector network analyzer; AWG, arbitrary waveform generator; SA, signal analyzer; BPG, bit pattern generator; Scope, sampling scope; Inset. SEM image of a fabricated device. (b) Measured DC photocurrents of photodiodes with different P⁺ widths as a function of input optical power. Inset. Measured dark currents of the photodiodes, which are independent of the P⁺ width. (c) Measured frequency responses of the photodiodes with the P⁺ width of 2 and 7 μm under different photocurrents.

Download Full Size | PDF

The experimental setup is shown in Fig. 4(a). Channel 1, 2, and 3 are used for bandwidth, RF-power, and large-signal transmission measurements, respectively. As for the DC photocurrent measurement, the source meter undertakes the work independently. The continuous-wave light (1550 nm) emitted from a laser diode (LD) launches into a Mach Zehnder modulator (MZM) or an optical transmitter (OT, SHF 46120A), which is driven by the electric signal from the vector network analyzer (VNA-MS4647B), the arbitrary waveform generator (AWG) or the bit pattern generator (BPG). Then, the power of the modulated signal light is controlled by an erbium-doped fiber amplifier (EDFA) and a variable optical attenuator (VOA). The polarization controllers (PCs) are used to optimize the maximal coupling efficiency. The signal light is coupled into the chip via the tapered fiber and the grating coupler. A bias-tee is used to apply bias voltage on the device under test (DUT) and collect the electrical signal simultaneously. The electrical signal transfers to the VNA, the signal analyzer (Keysight Technologies N9030A), or the sampling scope for the measurements of bandwidth, RF-power, or eye-diagram, respectively.

To characterize the improved DC performance, we measure the photocurrents as a function of input optical power. Deducting the 5.6-dB optical loss of the same grating couplers, Fig. 4(b) shows that all photodiodes have the same responsivity of 0.95 A/W, under low input optical power. With a reverse bias voltage of 3 V, the photocurrent of the photodiode with 7-μm P⁺ width increases linearly with the input optical power until reaching a saturated photocurrent of 19.6 mA. The photodiode with 5-μm P⁺ width has a similar situation, which is due to the heavily doping that has not yet been introduced to the region under the Ge absorption layer. After adopting doping regulation, the photodiode with 3-μm P⁺ width has a larger saturated photocurrent of 25.2 mA. The photodiode with 2-μm P⁺ width shows a photocurrent of 36.4 mA even at a much higher input optical power of 40 mW, demonstrating an 85.7% photocurrent improvement. Unfortunately, the device is damaged due to thermal failure at higher optical power input. Nevertheless, the simulated DC performance shows an increasing trend. As shown in the inset of Fig. 4(b), the dark currents at 3 V bias voltage are 11.6, 18.3, and 14 nA, respectively, for the photodiodes with 2, 3, and 7-μm P⁺ width. The negligible difference may be caused by measurement errors. The result implies that heavily doping implementation will have a negligible effect on the dark current in this case.

The footprint of the photodiode has a significant influence on the DC saturation photocurrent. For a fair comparison, we adopt the normalized photocurrent density [15], which is defined as the ratio of DC saturation photocurrent and the volume of the absorption region. Here, we favorably compare the photocurrent density of our device with the reported works of literature, as shown in Table 1. The proposed device achieves the maximum photocurrent density of 1.82 mA/μm³.

Table 1. Normalized photocurrent density comparison of high-power Si-Ge photodiodes

View Table

Furthermore, small-signal s-parameter measurement is implemented for bandwidth characterization, with the help of a 70-GHz VNA and a commercial MZM. Before the measurement, we calibrate the VNA, the MZM, the RF probe, and the cables to ensure an accurate characterization. The RF responses of the photodiodes with 2 and 7-μm P⁺ width are measured under different photocurrents. As shown in Fig. 4(c), the 2 and 7-μm P⁺ width photodiodes have the same 3-dB bandwidth of 30 and 7.5 GHz at the small photocurrent of 0.1 and 5 mA, respectively. The bandwidth enhancement is not significant here, although the heavily doping regulation could decrease the series resistance and improve the bandwidth characteristics as mentioned in Ref. [23]. It is since the fact that the photodiodes have a small size of 8-μm length and 5-μm width, and their bandwidths are mainly limited by the carrier transmit time. Under a high average photocurrent of 13 mA, the 2-μm P⁺ width photodiode has a bandwidth of 2.1 GHz, which is 17% larger than that of the 7-μm P⁺ width photodiode. It attributes to the alleviated SCS effect assisted by doping regulation.

3.2 RF-power saturation and large-signal data transmission

In the application of optical analog links, the saturated RF-power is an important indicator for the high-power photodiode. Therefore, we measure the RF-power saturation characteristics of the photodiodes. As shown in Figs. 5(a) to 5(c), they are the saturated RF-power and -1 dB compression photocurrents of the photodiodes, at 5, 10, and 20 GHz, respectively. For all devices, the RF-power increases linearly with the photocurrent until it reaches the saturation region. The compression effect is defined as the difference between the measured RF-power and the ideal linear power in dB [27]. Here, the calculated compression is normalized to be 0 dB at the small photocurrent, and the -1 dB compression point is defined as the photocurrent at which the compression drops to -1 dB. For the case of 5 GHz, the -1 dB compression photocurrent of the 2-μm P⁺ width photodiode is 9.6 mA while that of the 7-μm P⁺ width photodiode is 6.1 mA, showing a 57% enhancement benefitting from the doping regulation. Similarly, for the cases of 10 and 20 GHz, the -1 dB compression photocurrents of the 2-μm P⁺ width photodiode are 8.7 and 5.3 mA, respectively. Compared with that of the 7-μm P⁺ width photodiode, corresponding to 4.6 and 4 mA, these are significant improvements. Furthermore, the impact of operation bias voltages at 10 GHz for saturated RF-power and -1 dB compression photocurrent has been investigated. As for the 2-μm P⁺ width photodiode in Fig. 5(d), the -1 dB compression photocurrent increases from 6.2 to 8.7 mA, when increasing the operation bias voltage from 1 to 2 V. When increasing further to a bias voltage of 3 V, the improvement is negligible. The results imply that the improvement of the high-power performance of the photodiode by increasing the bias voltage only works in an appropriate region, and the applied bias voltage of 3 V is enough.

 

Fig. 5. Measured saturated RF-power and -1 dB compression photocurrents of photodiodes with different P⁺ widths at (a) 5, (b) 10, and (c) 20 GHz. (d) The impact of the bias voltage of the photodiode with 2-μm P⁺ width at 10 GHz for saturated RF-power and -1 dB compression photocurrent.

Download Full Size | PDF

To evaluate the feasibility of the high-power photodiode in realistic optical communication systems, we carry out large-signal data transmission measurements under different input optical power. Replacing the AWG with the BPG as the signal source, data streams with pseudo-random bit sequence of 2⁷-1 are generated at 10 and 20 Gbit/s. After being detected by the photodiodes, the RF signals are extracted again and launched to the sampling scope. A 6-dB RF attenuator has been utilized before the scope to avoid the RF signal intensity exceeding the scope range. As shown in Figs. 6(a) and 6(b), all the devices exhibit clear and similar open electric eye-diagrams at the small photocurrent. Meanwhile, the signal-noise ratio (SNR) of the 2, 5, and 7-μm P⁺ width photodiodes exhibit negligible differences under the small photocurrent. However, under the high power condition of 10 mA photocurrent, the eye-diagrams of the photodiodes with 5 and 7-μm P⁺ width have undergone considerable deterioration while that of the 2-μm P⁺ width one still maintains clearness. Moreover, the SNR of the 2-μm P⁺ width device exhibits a significant enhancement than that of the 5 and 7-μm P⁺ width device. Here, the bandwidths of 2, 5, and 7-μm P⁺ width devices are 3.1 GHz, 2.8 GHz, and 2.6 GHz, respectively. These measurements further demonstrate the enhancement of the high-power handling capability provided by the doping regulation method.

 

Fig. 6. The measured eye-diagrams of the photodiodes with 2, 5, and 7-μm P⁺ width under various photocurrents of 1, 5, and 10 mA at the bit rates of (a) 10 and (b) 20 Gbit/s.

Download Full Size | PDF

4. Conclusion

In conclusion, we have demonstrated a high-power Si-Ge waveguide photodiode assisted by doping regulation, utilizing the conventional fabrication process. The proposed photodiode shows an improvement of the saturation photocurrent from 19.6 to 36.4 mA, at the high input optical power of 40 mW, i.e., an 85.7% enhancement. Compared with other reported Si-Ge photodiodes, it achieves the maximum saturated photocurrent density of 1.82 mA/μm³. Furthermore, the proposed photodiode exhibits 57% more -1 dB compression photocurrent (9.6 versus 6.1 mA) at 5 GHz measurement. The eye-diagram characteristic for the application of optical communication systems also shows a significant improvement under high photocurrent. All the results indicate the doping regulation scheme is extremely effective in realizing a high-power Si-Ge photodiode. In addition to that, the combination of dual-feed, edge-couple, and traveling-wave array adds up to the excellent performance of high-power Si-Ge photodiode, which opens more opportunities for future applications.