1. Introduction

With the upcoming 5G era in the next few years, billions of smart devices will seamlessly connect with each other on the cloud, demanding an extremely large data throughput in cloud data centers where these smart devices are hosted [1–3]. As a technology fully compatible with the conventional CMOS fabrication process [4–6], the Si photonics platform provides solutions for high-speed, low-energy, and low-cost optical interconnections, giving rise to the increased adoption of Si photonic transceivers inside data centers [7–9]. An optical modulator, which encodes electrical signals into optical signals, is an indispensable component in an optical transceiver. Energy efficiency and modulation bandwidth are two of the most important performance metrics for optical modulators. Conventional Si optical modulators are all based on the free-carrier dispersion effect, in which the Si refractive index is manipulated by changing the free carrier density via a p-i-n or p-n junction. However, these two types of device are either limited in modulation bandwidth (p-i-n type) or modulation efficiency (p-n type) [10–16].

In 2004, a Si optical modulator based on a Si metal-oxide-semiconductor (MOS) structure was demonstrated [17]. The high-density free carriers that accumulated on the MOS interfaces induce a large refractive index change. Additionally, the improved overlap between the optical mode and the accumulated free carriers resulted in a high modulation efficiency. Moreover, the modulation bandwidth is only limited by the resistance–capacitance (RC) constant, which means that the modulation bandwidth can be easily scaled by changing the thickness of the oxide layer [18,19]. Despite these improvements, there is an unavoidable disadvantage of MOS-type optical modulators, which is the trade-off relationship between energy consumption and modulation bandwidth. A low energy consumption ($\; C\Delta {\textrm{V}^2}/2$) requires a high modulation efficiency, which means a large capacitance. This is because a larger capacitance can induce a large refractive index change through the accumulation of more free carriers on the MOS junction interfaces, reducing the driving voltage swing ($\Delta V$). Assuming an inversely proportional relationship between $\Delta V$ and C [20], if C is increased by n times, the resulting energy consumption will be accordingly reduced by n ($C\Delta {\textrm{V}^2}/nC{({\Delta \textrm{V}/n} )^2}$) times. On the other hand, the 3-dB modulation bandwidth is determined by $1/({2\pi RC} )$, which is also inversely proportional to the capacitance C. Thus, the relationship between the energy per bit and the modulation bandwidth of a Mach–Zehnder interferometer (MZI) optical modulator based on a Si MOS optical phase shifter is as shown by the black line in Fig. 1.

 

Fig. 1. Relationship between energy per bit and bandwidth. The black and blue lines indicate MZI optical modulators based on a Si-oxide-Si MOS optical phase shifter and a III–V/Si hybrid MOS optical phase shifter, respectively. The red line indicates a microring resonator with a III–V/Si hybrid MOS optical phase shifter.

Download Full Size | PDF

In our previous study, we have demonstrated a III–V/Si hybrid MOS optical phase shifter by bonding a thin n-type III–V membrane on a p-type Si waveguide, enabling an efficient optical phase modulation with low optical absorption loss [21,22]. A V_πL as small as 0.047 Vcm has been achieved owing to the efficient electron-induced refractive index change in InGaAsP. Compared with that of a Si MOS optical phase shifter, the modulation efficiency of the III–V/Si hybrid MOS optical phase shifter was improved by fivefold, leading to a 25-fold reduction in power consumption, as shown by the black and blue lines in Fig. 1. To further improve this relationship, it is promising to integrate the III–V/Si hybrid MOS optical phase shifter with a microring resonator. Recently, the thin-film lithium niobate microring modulator [23] or the heterogeneous integration of hybrid material on the microring/racetrack resonator has been reported, such as Barium Titanate/SiN [24], thin film-lithium niobate/SiN platforms [25]. The strong optical resonance of a microring resonator at the resonating wavelengths, represented as a high quality factor ($Q$), enables a low-energy optical intensity modulation with a small $\Delta V$. On the other hand, the improvement of energy efficiency does not degrade the RC-limited modulation bandwidth since both R and C of the phase shifter remain constant. As long as the Q is not high enough to impose limitation on the modulation speed [26–28], the 3-dB modulation bandwidth is still RC-limited. In this case, compared with a MZI optical modulator with the same phase shifter, the microring resonator is capable of enhanced energy efficiency with the same modulation bandwidth. Thus, the relationship between energy consumption and modulation bandwidth breaks, as shown by the red line in Fig. 1. In this work, we present the design, fabrication and characterization of a Si racetrack resonator with a III–V/Si hybrid MOS optical phase shifter. Compared with that of a MZI optical modulator with the same hybrid phase shifter, the energy consumption of the Si racetrack resonator is effectively reduced by 6.25-fold.

2. Device design

A schematic of a Si racetrack resonator with a III-V/Si hybrid MOS optical phase shifter is shown in Fig. 2(a). The passive Si components, including the Si input/output coupler, Si bus waveguide, and Si racetrack, were fabricated on a SOI wafer with a 225-nm-thick Si layer. The width and etching depth of the Si rib waveguide were designed to be 550 and 115 nm, respectively. The radius of the Si bends was assumed to be 5 µm. In order to integrate the III–V/Si hybrid MOS optical phase shifter on the Si rib waveguide, it is necessary to place 45-µm-long III–V tapers at both the input and output ports of the hybrid phase shifter to smoothen the optical mode transition with small insertion loss. There is a 5-µm-long passive Si rib waveguide between the 180-degree Si bend and the III-V taper. It is inserted to minimize the optical loss induced by the optical mode mismatch between the Si bend and straight waveguides. Thus, the total length of the Si racetrack is 200 µm. Figure 2(b) shows a cross-sectional schematic of a III–V/Si hybrid MOS optical phase shifter. A 200-nm-thick n-type InGaAsP layer is bonded to a 225-nm-thick p-type Si waveguide with a 6-nm-thick Al₂O₃ layer as the gate oxide layer. Under the metal contact to n-type InGaAsP, an n⁺⁺-InGaAs/n-InP double layer was inserted to reduce the contact resistance. For the same purpose, the Si layer was also partially heavily doped. Figure 2(c) shows the simulated optical mode profile of the III-V/Si hybrid MOS optical phase shifter. The optical mode centers on the MOS interfaces between InGaAsP/Al₂O₃/Si, where the free carriers accumulate. The overlap of optical mode and carrier accumulation enhances the optical phase modulation efficiency. Figure 2(d) shows the numerically calculated transmissions of a MZI modulator and a racetrack resonator. The transmission function of a MZI modulator is ${T_{MZI}} = \frac{{{P_{out}}}}{{{P_{in}}}} = \frac{{1 + \textrm{cos}({\Delta \varphi + \pi } )}}{2}\; $ [29], where Δφ+π is the optical phase difference between the two MZI arms. To compare the capability of optical intensity modulation with a racetrack resonator, we set π phase shift as the starting phase in the MZI transmission function so that the MZI interference peak and racetrack resonance peak are at the same position in the plot. The transmission function of a racetrack resonator is ${T_{RT}} = \frac{{{P_{out}}}}{{{P_{in}}}} = \frac{{{a^2} - 2ra\cos ({\Delta \varphi } )+ {r^2}}}{{1 - 2racos({\Delta \varphi } )+ {{({ra} )}^2}}}$ [30], where r is the self-coupling coefficient, and a is the single-pass amplitude transmission, which will be discussed in detail later. In the calculation, a = r=0.62 leads to a Q of 3264. From Fig. 2(d), the optical intensity of the racetrack resonator changes more rapidly than that of a MZI modulator near the resonance wavelength, resulting in power efficient optical intensity modulation.

 

Fig. 2. (a) Schematic of Si racetrack resonator with III–V/Si hybrid MOS optical phase shifter. (b) Cross-sectional schematic of III–V/Si hybrid MOS optical phase shifter. (c) Simulated optical mode profile of III-V/Si hybrid MOS optical phase shifter. (d) Numerically calculated transmission of a MZI modulator and a racetrack resonator.

Download Full Size | PDF

In [31], we presented a unique Si racetrack resonator design with a coupling length of 21 µm to integrate the III-V/Si hybrid MOS optical phase shifter. However, in this work, we used a conventional Si racetrack structure by simply engineering the coupling gap. Given a coupling length of 200 µm, the coupling gap of the directional coupler should be well designed to obtain a strong optical resonance and a high Q [30]. We have designed and fabricated Si racetrack resonators with various coupling gaps ranging from 300 to 900 nm. Figure 3(a) shows the transmission spectra of Si racetrack resonators with III-V/Si hybrid MOS optical phase shifters and different coupling gaps. The Si racetrack resonators with 300 and 700 nm coupling gaps show a strong resonance peak with a Q of 2498 and 3368, respectively. In contrast, Si racetrack resonators with 500 and 900 nm coupling gaps show no resonance peak for wavelength range from 1540 to 1560 nm. This can be explained by the variation of cross-coupling coefficient (κ) with the coupling length. The optical coupling was achieved by the directional coupler structure between the Si racetrack and the bus waveguide. The κ is defined as ${\kappa ^2} = si{n^2}\left( {\frac{{\pi L\Delta n}}{{{\lambda_0}}}} \right)$ [30], where L is the coupling length, λ₀ is the resonance wavelength, and Δn is the effective refractive index difference of the first two modes supported by the directional coupler, which can be obtained from simulation [32]. The optical power attenuation in the racetrack is defined as $1 - {a^2} = 1 - {e^{ - \alpha L}}$. Under critical coupling condition, $\; 1 - {a^2} = {\mathrm{\kappa }^2}$. As shown by Fig. 3(b), the Si racetrack resonators with 300 and 700 nm coupling gaps have a large κ² around 0.7∼0.8. In contrast, the Si racetrack resonators with 500 and 900 nm coupling gaps have a small κ² around 0.1. The presence of resonance peaks of Si racetrack resonators with 300 and 700 nm coupling gaps and the absence of resonance peak of Si racetrack resonators with 500 and 900 nm coupling gaps indicate that the power loss inside the racetrack is high (${a^2} \approx 0.7\sim 0.8 > 0.1$). This can be attributed to the adoption of a half-etching Si rib waveguide, a small bend radius, and an imperfect fabrication process. By refining the design and fabrication process in the future, we expect the optical loss of the Si racetrack to be reduced. In the case of a low-loss Si racetrack, the coupling gap should be increased to balance the optical loss and coupling strength for a good resonance condition. In the following measurement, we focus on the Si racetrack resonator with the coupling gap of 700 nm for its high Q value.

 

Fig. 3. (a) Transmission spectra of Si racetrack resonators with III-V/Si hybrid MOS optical phase shifters and different coupling gaps. (b) κ² as a function of coupling length.

Download Full Size | PDF

3. Fabrication process

Figure 4 shows the fabrication process flow of the Si racetrack resonator with a III–V/Si hybrid MOS optical phase shifter. The procedure was conducted on a SOI wafer with a 225-nm-thick Si layer and a 3-µm-thick SiO₂ buried oxide (BOX) layer, as shown in Fig. 4(a). The Si passive components, including Si input/output couplers, Si waveguide, and Si racetrack resonator, were defined by KrF lithography and ICP etching. The width and etching depth of the Si rib were 550 and 115 nm, respectively, as shown in Fig. 4(b). Then, the Si layer was doped as p-type by boron ion implantation. The doping concentrations were 1 × 10¹⁹ cm⁻³ for the heavily doped Si region and 1 × 10¹⁸ cm⁻³ for the remaining Si region. Next, a 3-nm-thick Al₂O₃ layer was deposited by atomic layer deposition (ALD) on both the Si waveguide and the InGaAsP layer which is lattice-matched to an InP wafer. Prebonding annealing was carried out on the Si and InP wafers at 700 and 400 °C for 40 min, respectively. After that, the Si and InP wafers were cleaned by ultrasonic water and bonded together with a pressure of 500 N and at a temperature of 300 °C for 1 h. Then, the InP wafer was selectively removed with HCl solution. The etch-stop layers (InGaAs/InP/InGaAs) on the InP wafer were chemically removed by wet etching. The InGaAs layer was etched using H₃PO₄:H₂O₂:H₂O=1:1:7, whereas the InP layer was etched using HCl:H₃PO₄ = 1:4. As shown in Fig. 4(d), the III–V mesa was defined by EB lithography and reactive ion etching (RIE). Then, a 250-nm-thick SiO₂ layer was deposited as a cladding layer by plasma-enhanced chemical vapor deposition (PECVD), as shown in Fig. 4(e). After the contact via formation, a Ni/Au stack was evaporated on the device using EB evaporator. Finally, as shown in Fig. 4(f), the contact pads were formed by the lift-off process. Figure 4(g) shows a microscopy image of the fabricated Si racetrack resonator.

 

Fig. 4. (a)–(f) Fabrication process flow of Si racetrack resonator with III–V/Si hybrid MOS optical phase shifter. (g) Plan-view images of fabricated Si racetrack resonator.

Download Full Size | PDF

4. Characterization and analysis

The fabricated Si racetrack resonator with a III–V/Si hybrid MOS optical phase shifter was evaluated at the C-band. A Santec TSL-510 tunable laser was used as a light source. The input light was firstly coupled into the Si waveguide through a lensed fiber. The polarization of the input light was adjusted to the transverse electric (TE) mode. The output modulated optical signals were collected using a single-mode fiber via an objective lens. The output power was evaluated using an Agilent 81618A InGaAs power meter. During the measurement, the Si racetrack resonator was mounted on a stage with a Peltier cooler. The temperature was maintained at 23 °C to eliminate the effect of temperature variation. Figure 5(a) shows the transmission spectra from the Si racetrack resonator with different V_g applied. The resonance peak was shifted to the short-wavelength region by increasing V_g. With increasing V_g, the extinction ratio increased and then decreased, which indicated that the racetrack resonator went through over-coupled, critical-coupled and under coupled conditions [33]. At V_g = 0.8 V, the extinction ratio reaches maximum, which is desired for optical intensity modulation. Thus, in the following discussion, we focus on the V_g ranging from 0 to 0.8 V. In this V_g range, Q varies from 3368 to 4077 with the highest Q of 4077 achieved at V_g = 0.8 V. The Q varies with V_g from 3368 to 4077 due to the free-carrier-induced optical loss [30]. The FSR of the racetrack was 1.42 nm. From the FSR and the wavelength shift induced by different V_g, the relationship between phase shift and V_g was obtained, as shown in Fig. 5(b). The modulation efficiency (V_πL) extracted from the gradient of the plot was 0.059 Vcm. From simulation, the oxide thickness is expected to be around 6 nm. Considering the capacitor area, the intrinsic capacitance is 316 fF. However, the accumulation capacitance measured at 1 MHz frequency is 777 fF. This is due to the presence of parasitic capacitances, which can be easily removed by adopting a Si waveguide embedded in SiO₂ layer [34]. In the following discussion, we used the calculated intrinsic capacitance without considering any parasitic effect.

 

Fig. 5. (a). Transmission spectra of fabricated Si racetrack resonator at different V_g and (b) optical phase shift as a function of V_g.

Download Full Size | PDF

The optical intensity modulation characteristics at a wavelength of 1549.38 nm are shown in Fig. 6. For optical intensity modulation from −6 to −16 dB, the ΔV required by the Si racetrack is 0.553 V. The characteristics of optical intensity modulation by a Si MZI optical modulator with the same III–V/Si hybrid MOS optical phase shifter were calculated and plotted in Fig. 6. The transfer function of the MZI optical modulator is given by ${I_o} = \frac{{{I_i}}}{2}[{1 + cos({\Delta \phi } )} ]$, where $\Delta \phi $ is the phase difference between the two arms of the MZI optical modulator induced by the phase shifter [12]. In the calculation of the performance of MZI optical modulator, a V_πL of 0.059 Vcm was assumed, which is the same as that of the Si racetrack resonator. According to the calculation, a ΔV of 1.37 V is required for a Si MZI optical modulator to achieve 10 dB (from −6 to −16 dB) optical intensity modulation. Compared with a Si MZI optical modulator with the same phase shifter, the Si racetrack resonator reduced ΔV by 60% for the same level of optical intensity modulation. Since the static power consumption of the III–V/Si hybrid MOS optical phase shifter is negligibly small [35], the total power consumption is dominated by the dynamic power consumption (CΔV²/2). The energy consumption of the Si racetrack resonator was reduced to 0.16-fold that of a Si MZI optical modulator with the same phase shifter for the same 10-dB optical intensity modulation.

 

Fig. 6. Characteristics of optical intensity modulation by fabricated Si racetrack resonator and Si MZI optical modulator with the same phase shifter.

Download Full Size | PDF

Figure 7 shows the calculated energy consumption ratio (red) of a Si racetrack resonator to a MZI optical modulator with the same phase shifter. The energy consumption ratio is defined as the ratio of dynamic power consumption of a racetrack resonator to that of a MZI modulator, which is ${E_{ratio}} = \frac{{C{{(\Delta {V_{RT}})}^2}/2}}{{C{{(\Delta {V_{MZI}})}^2}/2}} = \frac{{{{(\Delta {V_{RT}})}^2}}}{{{{(\Delta {V_{MZI}})}^2}}}$, where ΔV_RT and ΔV_MZI are the driving voltage of the Si racetrack resonator and the MZI modulator, respectively. With increasing Q, the energy consumption of the Si racetrack resonator decreases owing to the rapid change in optical intensity near the resonating peak. The energy consumption ratio of 0.16, as demonstrated in the measurement, indicated a Q of 3233. On the other hand, the photon-lifetime-limited bandwidth also decreases with Q, as shown by the blue line in Fig. 7. The photon-lifetime-limited bandwidth is expressed as ${f_{photon}} = \frac{c}{{\lambda Q}}$, where c is the light speed in vacuum. Thus, an excessively high Q will limit the achievable modulation bandwidth owing to an excessively long photon lifetime. The photon-lifetime-limited bandwidth is 60 GHz with a Q of 3233. In pseudo-random data transmission, there is 50% possibility of symbol change between “0” and “1”. Thus, the energy per bit is calculated as ${E_{bit}} = \frac{{C{{({\Delta V} )}^2}}}{4} = 24.2\; \textrm{fJ}/\textrm{bit}$ [12]. The energy per bit here does not include the energy consumed for temperature stabilization.

 

Fig. 7. Energy consumption ratio (red) of racetrack resonator to MZI optical modulator and photon-lifetime-limited bandwidth (blue) as a function of Q.

Download Full Size | PDF

Figure 8 shows the benchmarking of 3-dB modulation bandwidth and V_πL. The 3-dB modulation bandwidth is determined by both the RC constant and the photon lifetime, as expressed by $\frac{1}{{f_{3dB}^2}} = \frac{1}{{f_{RC}^2}} + \frac{1}{{f_{photon}^2}} = {({2\pi RC} )^2} + {\left( {\frac{{\lambda Q}}{c}} \right)^2}$[39]. The RF characterization was not performed on the current device due to the presence of parasitic capacitances which can be easily removed by using a Si waveguide embedded in SiO₂ [34]. In the following discussion, we used the theoretically calculated 3-dB modulation bandwidth. The calculated f_3dB is 38 GHz, which is limited by the RC constant. By increasing the oxide thickness of the III-V/Si hybrid MOS optical phase shifter, we can scale up the f_3dB along the blue line as shown in Fig. 8. This trend also indicates another way of improvement, which is to increase the photon-lifetime-limited bandwidth by adopting a smaller Q. However, a small Q will degrade the advantage of low-energy optical modulation of the Si racetrack resonator compared with that of a Si MZI optical modulator, as discussed in the last paragraph. The choice of Q will depend on the balance between energy consumption and bandwidth requirement. Figure 8 also shows the results (red dots) of Si microring [26–29] and racetrack [30] optical modulators based upon the carrier depletion effect. Compared with pure Si optical modulators, our racetrack optical modulator significantly reduced V_πL owing to the efficient III-V/Si hybrid MOS optical phase shifter. For instance, a 50-GHz modulation bandwidth is predicted with a V_πL of 0.127 Vcm by scaling the sandwiched oxide thickness to around 13 nm. In conclusion, a low-energy optical modulation is achievable while maintaining a large modulation bandwidth simultaneously.

 

Fig. 8. Benchmarking with Si microring/racetrack resonator based on free-carrier plasma dispersion effect [36–40].

Download Full Size | PDF

5. Conclusion

In order to improve the energy–and–bandwidth trade-off relationship of Si MOS-type optical modulators, it is very effective to integrate a III-V/Si hybrid MOS optical phase shifter into a Si racetrack resonator. In this study, we have successfully demonstrated a Si racetrack resonator with a III-V/Si hybrid MOS optical phase shifter. The gap between the Si racetrack and the Si bus waveguide was carefully designed for a strong optical resonance condition. The fabricated Si racetrack resonator exhibited a Q of 3233 and a V_πL of 0.059 Vcm. Compared with a Si MZI optical modulator with the same optical phase shifter, the Si racetrack resonator has reduced the power consumption by 84% for 10-dB optical intensity modulation. Hence, it is promising for high-speed and low-power optical modulation applications.