High-speed uni-traveling carrier photodiode for 2 μm wavelength application

1. INTRODUCTION

Due to the exponentially increasing volume of internet traffic, today’s optical communication systems are rapidly approaching their capacity limit [1,2]. This “capacity crunch” provides an impetus for developing next-generation optical communication systems. A new spectral window at the 2 μm wavelength is a promising solution to increase the system capacity due to the availability of low-loss (0.2 dB/km) hollow-core photonic bandgap fiber (HC-PBGF) and the high-bandwidth (from 1810 to 2050 nm) thulium-doped fiber amplifier (TDFA) at 2 μm wavelength [3,4]. Studies on 2 μm optical components and system architecture have shown the promising potential of this new spectral window [5–7]. An eight-channel wavelength-division multiplexing transmission system with 100 Gbit/s data capacity has been demonstrated at the 2 μm band [6].

A high-speed photodetector is one of the key components of the optical communication system, and among various figures of merit, bandwidth is most commonly used for a high-speed-device benchmark. At the 2 μm wavelength band, high-speed photodiodes have been demonstrated with In-rich InGaAs on InP [5,7–11], InGaAsSb on GaSb [12], GeSn/Ge on Si [13,14], and defect-mediated Si [15]. ${\mathrm{In}}_{0.53}{\mathrm{Ga}}_{0.47}\mathrm{As}$ on InP, which cuts off at 1.7 μm, is the most widely used absorption material for high-speed application due to its high carrier mobility. For 2 μm operation, In-rich InGaAs has to be used to extend thecutoff wavelength. Ye et al. demonstrated a 10 GHz bandwidth photodiode with ${\mathrm{In}}_{0.7}{\mathrm{Ga}}_{0.3}\mathrm{As}$ as an absorption layer [11]. Joshi et al. achieved a 16 GHz bandwidth using a ${\mathrm{In}}_{0.72}{\mathrm{Ga}}_{0.28}\mathrm{As}$ absorber [9]. A partially depleted photodiode with a ${\mathrm{Ga}}_{0.8}{\mathrm{In}}_{0.2}{\mathrm{As}}_{0.16}{\mathrm{Sb}}_{0.84}$ absorber on GaSb has achieved a 3 dB bandwidth of 6 GHz [12]. On the other hand, a GeSn/Ge quantum well grown on Si shows a benefit for integration with other silicon devices, and a ${\mathrm{Ge}}_{0.92}{\mathrm{Sn}}_{0.08}/\mathrm{Ge}$ photodiode with a 10 GHz bandwidth has been reported [14]. By inserting lattice defects, a silicon-on-insulator photodiode can also operate at the 2 μm wavelength, and a 15 GHz bandwidth can be achieved [15]. However, the performance of these devices is sensitive to the crystal quality, making them inconvenient for practical application.

Short-wavelength infrared (SWIR) photodetectors using InGaAs/GaAsSb type-II multiple quantum wells (MQWs) on the InP substrate as an absorber have been demonstrated and well-studied recently [16–19]. The devices enjoy the advantage of lattice-matched property on InP, and show low dark current and high detectivity at room temperature. The crystal quality of the MQW structure can be ensured since it is lattice-matched to InP and the growth technique of InP material system is mature. A PIN-based high-speed photodiode with InGaAs/GaAsSb MQWs for 2 μm operation has achieved a 3 dB bandwidth in the range of 3.5 GHz to 10 GHz, which is mainly limited by the slow transport of an optical-generated hole in the absorption region [20–22].

To further improve the bandwidth, a uni-traveling carrier photodiode (UTC-PD) with a type-II MQW absorber was proposed and theoretically investigated in our previous work [23]. A UTC-PD based on a InGaAs/InP material system has been demonstrated with hundreds of gigahertz bandwidth at a 1.55 μm wavelength band for high-speed application [24–27]. In the UTC-PD structure, the optical-generated carriers are excited in the undepleted p-type absorption layer, and only electrons are injected into the InP drift layers; thus, the effective carrier transit time is shorter than that of the PIN structure with the proper design.

In this work, we demonstrate the normal-incident InGaAs/GaAsSb MQW UTC-PDs at the 2 μm wavelength band with a 3 dB bandwidth of 25 GHz and bit rate of 30 Gbit/s. To the best of our knowledge, this is the highest bandwidth and bit rate demonstrated at the 2 μm wavelength band. By analyzing the RF performance of the devices with different diameters, it is found that the 3 dB bandwidth performance can be further improved by optimizing the fabrication process.

2. DEVICE STRUCTURE AND FABRICATION

The epilayer structure of the photodiode is shown in Fig. 1(a). All layers are lattice-matched to the InP substrate. The structure was grown on semi-insulating double-side-polished InP substrate by a molecular beam epitaxy (MBE) system. The epitaxial growth began with a 200 nm $n+$ InP layer, 20 nm $n+$ InGaAs layer, and 900 nm $n+$ InP layer. The 100 nm n-doped InP layer with a doping concentration of $1\times {10}^{18}{\mathrm{cm}}^{-3}$ was then grown to reduce Si diffusion into the 400 nm intrinsic InP drift layer. The absorption layer consists of 180 nm graded doped 3 nm/3 nm ${\mathrm{In}}_{0.53}{\mathrm{Ga}}_{0.47}\mathrm{As}/{\mathrm{GaAs}}_{0.50}{\mathrm{Sb}}_{0.50}$ MQWs, which is expected to have fast response speed while maintaining a cutoff wavelength larger than 2 μm, according to our simulation [23]. Following the MQW layers is the 30 nm p-doped large bandgap AlGaAsSb electron blocking layer to prevent the diffusing of electrons in the absorption layer towards the p-doped InGaAs contact layer on the top. Room-temperature photoluminescence (PL) of the epitaxial layers was conducted by using the 532 nm wavelength laser as the excitation source and a Fourier transform infrared (FTIR) spectrometer with an In-rich InGaAs detector. Two peaks can be identified in Fig. 1(b), indicating the good crystal quality of the material. The peak at 2.1 μm corresponds to the transition between the ground states in InGaAs and GaAsSb quantum well layers, respectively, and the second peak at 1.75 μm may correspond to the emission from the electron ground state to light hole state, as shown in Fig. 1(d).

 

Fig. 1. (a) Epitaxial structure of the type-II MQW UTC photodiode. (b) Photoluminescence measurement result of the epitaxial structure. (c) Schematic diagram of the fabricated device. (d) Band diagram of the MQW absorber. The wave functions and potential transitions are shown. The wave functions are calculated by a kp method, and the two transitions correspond to the peaks at 2.1 μm and 1.75 μm in the photoluminescence spectrum.

Download Full Size | PPT Slide | PDF

After the material growth, the epitaxial structure was processed into double-mesa structures, as shown in Fig. 1(c). A set of mesa devices with different diameters was formed by standard photolithography and a wet-etching process. A 150 μm pitch ground–signal–ground (GSG) coplanar waveguide (CPW) pad with an air-bridge structure was electroplated for high-frequency measurement. The substrate is backside-polished to support backside illumination with no antireflection coating.

3. MEASUREMENT RESULT

A. Electrical Characteristics

The dark current voltage (I-V) curves of the photodiodes at room temperature are shown in Fig. 2. The dark current value at $-3\mathrm{V}$ is 3.02 nA, 3.88 nA, and 5.82 nA for the photodiodes with diameters of 10 μm, 20 μm, and 40 μm, respectively, which are lower than the In-rich InGaAs photodiodes [5,7–11]. Figure 3 shows the capacitance-voltage (C-V) curves. The device is fully depleted at $-1\mathrm{V}$, and the capacitance at $-3\mathrm{V}$ is 60.6 fF, 154.0 fF, and 432.6 fF for the 10 μm, 20 μm, and 40 μm diameter photodiodes, respectively. Parasitic capacitance of 49.6 fF was found based on the linear fitting of capacitance versus device area, as shown in Fig. 4, which indicates that optimization is needed for future fabrication processes. The origin of the parasitic capacitance is explained in detail in Supplement 1.

 

Fig. 2. Dark current characteristics for three devices with different diameters at room temperature.

Download Full Size | PPT Slide | PDF

 

Fig. 3. Measured capacitance versus reverse bias for three devices with different diameters at room temperature.

Download Full Size | PPT Slide | PDF

 

Fig. 4. Capacitance of devices at $-3\mathrm{V}$ bias. The fitting result indicates a parasitic capacitance of 49.6 fF.

Download Full Size | PPT Slide | PDF

B. Responsivity

The top-illuminated responsivity spectrum was measured using an FTIR spectrometer. A standard blackbody radiation source at 700°C was used to calibrate the responsivity. As shown in Fig. 5, the 100% cutoff wavelength is around 2.25 μm, and the responsivity is 0.07 A/W at 2 μm wavelength and reverse bias (larger than $-1\mathrm{V}$). Considering the 180 nm thickness of the absorption layer, the equivalent absorption coefficient of the absorption layer is about $3600{\mathrm{cm}}^{-1}$, which is higher than the simulation value of $2000{\mathrm{cm}}^{-1}$ in [22].

 

Fig. 5. Responsivity spectrum at various bias voltages. The inset shows the responsivity at 2 μm wavelength.

Download Full Size | PPT Slide | PDF

C. Bandwidth

The 3 dB bandwidth was measured by a heterodyne setup with two beams of laser light at a wavelength of 2004 nm coupled together, which generates an intensity-modulated light with a frequency from hundreds of megahertz to more than 30 GHz. The modulated light was amplified, and then it illuminated the devices from backside. The details of the measurement setup are shown in Fig S2 of Supplement 1.

Figure 6 shows the frequency response of a 10 μm diameter photodiode at different bias voltages. The 3 dB bandwidth increases slightly while reverse bias increases. The highest 3 dB bandwidth of 25 GHz can be achieved at $-4\mathrm{V}$ bias. Theoretically, the 3 dB bandwidth of the photodiode is often limited by transit time and RC time, as expressed by the equation

 

Fig. 6. Frequency response of a 10 μm photodiode at various bias voltages. The photocurrent is set to 1 mA. The inset shows the 3 dB bandwidth at different bias voltages.

Download Full Size | PPT Slide | PDF

Figure 7 shows the frequency response of the 10 μm photodiode at different photocurrents. The 3 dB bandwidth is almost the same (around 25 GHz at $-3\mathrm{V}$) when the photocurrent increases from 1 mA to 10 mA, and no obvious saturation can be found.

 

Fig. 7. Frequency response of a 10 μm photodiode at different photocurrents. The bias voltage is $-3\mathrm{V}$.

Download Full Size | PPT Slide | PDF

Figure 8 shows the frequency response of three photodiodes with different diameters. The 3 dB bandwidths of the 40 μm, 20 μm, and 10 μm devices are 5 GHz, 15 GHz, and 25 GHz, respectively. The smaller photodiode has smaller junction capacitance, leading to a larger RC limit bandwidth, while the transit limit bandwidth remains the same. Thus, the strong bandwidth dependency on diameter indicates that the 3 dB bandwidth of the devices is dominated by the RC limit. To verify the RC limit, we measured the S-parameters and fitted the parameters with an equivalent circuit model, as shown in Fig. 9(a). The fitting curves are shown in Figs. 9(b), 9(c), and 9(d). The extracted model parameters are listed in Table 1. Then, the theoretical RC limit frequency response can be calculated using the equivalent model, as shown in Fig. 9(e). The RC limit bandwidths of the 40 μm, 20 μm, and 10 μm devices are 7.7 GHz, 18.9 GHz, and 41.4 GHz, respectively. Applying Eq. (1), a rough estimation of the transit time limit bandwidth for the 10 μm diameter photodiode is about 31 GHz. As shown in Fig. 4, a parasitic capacitance of 49.6 fF exists in our devices, and so the 3 dB bandwidth of the devices can be further improved by reducing the parasitic capacitance. Figure 10 reviews the 3 dB bandwidth of high-speed photodetectors operating at the 2 μm wavelength in recent years. With proper design, the InGaAs/GaAsSb MQWs UTC-PD can achieve better bandwidth performance than other devices.

 

Fig. 8. Frequency response of photodiodes with different diameters. The bias voltage is $-3\mathrm{V}$, and the photocurrent is 1 mA.

Download Full Size | PPT Slide | PDF

 

Fig. 9. (a) Equivalent circuit model used in parameter fitting. $C_j$ is the junction capacitance, $R_s$ is the series resistance (resistance of the ohm contacts and the CPW pads), $L_s$ is the inductance of the CPW pads, and $R_j$ is the junction resistance, which is hundreds of megaohms and can be regarded as open circuit. The measured and fitting curves of S11 of (b) 10 μm, (c) 20 μm, and (d) 40 μm photodiodes at $-3\mathrm{V}$ bias (the blue curve is the measured data while the red curve is the fitting curve). (e) Calculated RC limit frequency response using the fitting results.

Download Full Size | PPT Slide | PDF

Table 1. Fitting Parameters

View Table | View all tables in this article

 

Fig. 10. Review of the 3 dB bandwidth of high-speed photodiodes operating at 2 μm wavelength reported in recent years.

Download Full Size | PPT Slide | PDF

Figure 11 shows the eye diagram of a 10 μm photodiode biased at $-3\mathrm{V}$ and operating at a 20 Gbit/s, 25 Gbit/s, and 30 Gbit/s data rate. The 128 bits 2^15–1 pseudo-random binary sequence (PRBS) sequences were generated as the data source to drive a 2 μm wavelength Mach–Zehnder modulator, which modulates the optical signal coming from a 2004 nm wavelength single-frequency fiber laser with a modulation depth of 0.25. The optical power illuminating the photodiode is about 7 mW. The output of the photodiode was amplified by a $+23\mathrm{dB}$ microwave amplifier and then displayed on the real-time sampling oscilloscope. A clear eye pattern is demonstrated at a 30 Gbit/s data rate, which indicates the devices can be used at a 2 μm optical communications system.

 

Fig. 11. Eye pattern of a 10 μm photodiode at 20 Gbit/s, 25 Gbit/s, and 30 Gbit/s.

Download Full Size | PPT Slide | PDF

D. Saturation Power

For some applications, such as microwave photonics, the high-speed photodetector works in high-power mode, and the saturation power is another important figure of merit. When the photocurrent is high, a large number of electrons exists in the depletion region, and the electrical field in the depletion region might collapse due to the space charge effect. This limits the carrier sweeping out process, causing a saturation in output power. Figures 12 and 13 show the relationship between the output power and the photocurrent. In this measurement, lasers of 1.55 μm wavelength were used as the optical source due to the power limitation of the 2 μm optical amplifier. Notice that the saturation characteristics should be the same for 1.55 μm and 2 μm input, since the carrier transports in the depletion region are identical. The ideal output power of a photodetector can be calculated by

 

Fig. 12. Output RF power and compression versus photocurrent for a 10 μm photodiode at 25 GHz and at different bias voltages. The gray dashed line shows the ideal output power.

Download Full Size | PPT Slide | PDF

 

Fig. 13. Output RF power and compression versus photocurrent for a 20 μm photodiode at 15 GHz and at different bias voltages. The gray dashed line shows the ideal output power.

Download Full Size | PPT Slide | PDF

Table 2. 1 dB Compression Points

View Table | View all tables in this article

4. CONCLUSION

In this work, we have reported a high-speed photodiode working at the 2 μm wavelength based on InGaAs/GaAsSb type-II MQWs with uni-traveling carrier design. The InGaAs/GaAsSb type-II MQW absorber cuts off at 2.25 μm, and the responsivity is 0.07 A/W at 2 μm. The 3 dB bandwidths at 2 μm are 25 GHz, 15 GHz, and 5 GHz for photodiodes with 10 μm, 20 μm, and 40 μm diameter, respectively. A clear eye pattern can be observed at 30 Gb/s for the 10 μm photodiode. Analysis shows that the bandwidth of the current device can still be improved with further reduction of the parasitic capacitance by optimizing the fabrication process. The excellent high-speed performance of this device paves the way towards more promising feasibility of a new spectral window at the 2 μm wavelength.