Open Access
Add to BibSonomy
Abstract
Nonlinear optics-based optical signal processing (OSP) could potentially increase network flexibility because of its transparency, tunability, and large bandwidth. A low-loss, high nonlinearity, and compact integrated material platform is always the pursuit of OSP. In this Letter, a high-efficiency, one-to-six wavelength multicasting of 10 Gbaud pulse-amplitude modulation (PAM3/PAM4) signals using a 6 cm long ${{\rm Al}_{0.2}}{{\rm Ga}_{0.8}}{\rm As}$-on-insulator nanowaveguide is experimentally demonstrated for the first time, to the best of our knowledge. The low-loss, combined with the high nonlinear coefficient of the AlGaAsOI platform, enables us to achieve ${-}{11.2}\;{\rm dB}$ average conversion efficiency clear eye diagrams and ${\lt} 2.1\;{\rm dB}$ power penalty at KP4-forward error correction threshold (${2.4} \times {{10}^{- 4}}$) for all the output PAM3/PAM4 multicasting channels. This result points to a new generation of nonlinear OSP photonic integrated circuits.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
The capacity of optical networks is rapidly growing, driven by data-intensive applications, such as high-definition video, cloud computing, artificial intelligence, and 5G. Due to the relatively large bandwidth, optical signal processing (OSP) techniques are considered as potential solutions to help electronics deliver and process high data capacities. OSP brings together various fields of optics and signal processing–namely, nonlinear devices and processes, analog and digital signals, and advanced data modulation formats–to achieve signal processing functions that can potentially operate at the line rate of fiber optic communications [1]. Following the trend of device miniaturization, nonlinear optics-based OSP has been extended from bulky devices, such as highly nonlinear fiber to integrated platforms such as silicon [2], Hydex [3], SiN [3–5], AlN [6], and diamond [7]. Compared to dielectric material platforms, III-V semiconductor materials hold much higher nonlinear coefficients and convenience in active integration. As a promising candidate, aluminum gallium arsenide (${{\rm Al}_x}{{\rm Ga}_{1 - x}}{\rm As}$) offers a high nonlinear index (${{\rm n}_2} \approx {2.6} \times {{10}^{- 17}}\;{{\rm m}^2}\,{{\rm w}^{- 1}}$) [8,9]. It has a large refractive index (${\rm n} \approx {3.3}$) and can be integrated with laser sources and photodetectors. Its bandgap can be tailored by varying the Al mole fraction $x$ to mitigate two-photon absorption (TPA) at two telecom bands around 1310 and 1550 nm. The recently demonstrated AlGaAs waveguide fabricated by heterogeneous integration of III–V materials on a silica buffer layer has attracted a lot of attention in nonlinear optical applications, as the waveguides have a high refractive index contrast and high nonlinear coefficients [10–12]. On such a platform, frequency comb generation [10,11] and efficient second-harmonic generation have been proposed and demonstrated [12].
Among the OSP technologies, wavelength multicasting, which can efficiently deliver a stream of information carried by one input wavelength to several different output wavelengths, is of great importance for enhancing the reconfigurability and non-blocking capacity of future optical networks [13]. Multiple approaches for wavelength multicasting have been demonstrated, such as the research based on bulky HNLF [14] and integrated silicon platforms [15–17] with differential phase shift keying and 16-ary quadrature amplitude modulation as input signals. On the other hand, advanced modulation formats that are compatible with intensity modulation/direct detection (IM/DD) such as pulse-amplitude modulation (PAM) have attracted a lot of attention in short-reach scenarios because of their low complexity and power consumption. PAM4 and PAM3 have been chosen as the standard formats by the IEEE P802.3bs Task Force for 400-G Ethernet and 100/1000BASE-T1 for automotive networking, respectively. Using wavelength multicasting techniques for PAM signals could further enhance the flexibility of PAM-based short-reach networks; yet this has not been reported.
In this Letter, we present an AlGaAsOI strip nanowaveguide with engineered cross-sectional dimension (${400}\;{\rm nm}\;{\rm height} \times {800}\;{\rm nm}\;{\rm width}$) and low propagation loss (0.4 dB/cm). The achieved effective nonlinearity, as defined in [10], is ${\sim}{428}\;{{\rm W}^{- 1}}\;{{\rm m}^{- 1}}$, which facilitates highly efficient $\chi^{(3)}$ nonlinear optical four-wave mixing (FWM) processes. Using this device, we implement one-to-six 10 Gbaud PAM3/PAM4 wavelength multicasting for the first time, to the best of our knowledge. Due to the high nonlinearity of the waveguide, a highly efficient FWM with average conversion efficiency (CE, defined as the ratio of output idler power to the input signal power) up to ${-}{11.2}\;{\rm dB}$ for the multicasting channels is achieved. All of the output multicasting PAM3/PAM4 channels show clear eye diagrams and ${\lt} 2.1\;{\rm dB}$ power penalty at KP4-forward error correction (FEC) (${2.4} \times {{10}^{- 4}}$) [18]. Different from the previous wavelength conversion work based on AlGaAs as reported in [19,20], by adding a second pump laser, multiple degenerate and non-degenerate FWM processes are triggered and used as multicasting output.
The schematic drawing of the waveguide cross section is shown in Fig. 1(a). In this layout, a thin ${{\rm Al}_x}{{\rm Ga}_{1 - x}}{\rm As}$ layer on top of a low-index silica layer resides on a silicon substrate. Here $x$ is chosen to be 0.2 for operating the wavelength multicasting at a C band [21]. For sub-wavelength-sized waveguides, the group velocity dispersion (GVD) can be engineered by tailoring the waveguide geometry to anomalous dispersion that is required to achieve phase matching and high CE in FWM process [20,22]. In this Letter, the ${{\rm Al}_{0.2}}{{\rm Ga}_{0.8}}{\rm As}$ layer is set to be 400 nm in thickness. Figure 1(b) presents the calculated GVD for the waveguides with different widths.
Fig. 1. (a) Schematic drawing of the waveguide cross section, (b) calculated GVD versus wavelength for 400 nm thick ${{\rm Al}_{0.2}}{{\rm Ga}_{0.8}}{\rm AsOI}$ waveguides with different widths, (c) SEM of the sidewall of the as-etched AlGaAs waveguide, and (d) SEM image of the cross section of a waveguide with a ${\sim}{100}\;{\rm nm}$ ${{\rm SiO}_2}$ cladding for protection.
The fabrication of the AlGaAsOI nanowaveguide is based on a heterogeneous wafer-bonding technique. During the fabrication process, it is critical to reduce scattering loss. In this Letter, the scattering loss is reduced by employing a reflow process of the patterned photoresist after the lithography together with an optimized dry etch process. The AlGaAs was epitaxially grown by molecular beam epitaxy, which enables excellent control of film thickness down to the monolayer. For waveguide patterning, we used deep ultra-violet lithography, which can guarantee the dimension control for width with an accuracy of around ${\pm}{10}\;{\rm nm}$. The fabrication details are given in [21]. Figures 1(c) and 1(d) show a scanning electron microscope (SEM) picture of the fabricated AlGaAsOI nanowaveguide with a smooth surface and vertical sidewall. This enables low propagation loss (0.4 dB/cm [21]) and good geometry control, which are important for high-efficiency nonlinear processes.
Figure 2 is a schematic illustration of the PAM3/PAM4 multicasting scheme based on the AlGaAsOI nanowaveguide. With three input lights (pump1 at ${\lambda _1}/\!{f_1}$ indicated by red, PAM3/PAM4 signal light at ${\lambda _2}/\!{f_2}$ indicated by green, and pump2 at ${\lambda _3}/\!{f_3}$ indicated by blue) being injected into the AlGaAsOI nanowaveguide, additional idlers by FWM inside the waveguide are generated. Considering the components generated by degenerate FWM and non-degenerate FWM processes, nine idlers in total will be generated, which possess frequencies of ${f_{\textit{xyz}}} = {f_x} + {f_y} - \;{f_z}$ (${x}$, ${y}\; \ne \;{z}$. ${x}$, and ${y}$ and ${z}\; \in \;{1}$, 2, and 3). $\Delta \!{f_1}$ and $\Delta \!{f_2}$ is the frequency interval of $\langle\! {{f_1},{f_2}}\rangle$ and $\langle\! {{f_2},{f_3}} \rangle$, respectively. To avoid frequency overlap for the new generated idlers, $\Delta \!{f_2}$ is set to be larger than ${2}\Delta\! {f_1}.$
Fig. 2. Schematic illustration of PAM3/PAM4 wavelength multicasting based on a AlGaAsOI nanowaveguide chip.
The PAM sequence of data is modulated on the amplitude of the signal optical carrier, so the power of the new generated idlers is what we care about. The power relationships among the pumps, signal, and idlers are presented in Table 1. ${P_i}$ is the power of each wave, $\gamma$ is the nonlinear coefficient, and $K$ is a constant proportional to the FWM CE. The pump power is constant, provided by a continuous-wave source. From Table 1, it can be seen that the powers of the idlers at ${\lambda _{a\:}} - \;{\lambda _e}$ are proportional to the signal power (${P_2}$) and maintain the original PAM signal information. Besides the output signal at ${\lambda _2}$, six output multicasting channels can be achieved. However, the powers of the idlers at ${f_{223}}$, and ${f_{221}}$ have quadratic dependence on the signal power, and the powers of ${f_{113}}$ and ${f_{331}}$ do not involve the information of the signal, which means the signal information at ${f_{223}}$, ${f_{221}}$, ${f_{113}}$, and ${f_{331}}$ cannot be preserved. As discussed in [23], when the input signal is PAM4 with nonuniform power intervals among the four power levels such as (0, ${({1/3})^{1/2}}$, ${({2/3})^{1/2}}$, 1), the idlers at ${f_{223}}$ and ${f_{221}}$ can be used for signal regeneration.
Table 1. Power Relationship of Signal, Pumps, and Idlers
The experimental setup is shown in Fig. 3. An arbitrary waveform generator (AWG, Tektronix 70002, 25GSa/s) and an intensity modulator are used to generate 10 Gbaud PAM3 (15 Gbps)/PAM4 (20 Gbps) signals. In the coding of PAM3, two PAM3 symbols represent three binary bits. In the coding of PAM4, (00, 01, 11, 10) in binary maps to (${-}{3}$, ${-}{1}$, 1, 3). Three lights (pump1 at 1550.12 nm, signal at 1550.92 nm, and pump2 at 1553.33 nm, with $\Delta\! {f_1} = {100}\;{\rm GHz}$ and $\Delta \!{f_2} = {300}\;{\rm GHz}$), after amplification by erbium-doped fiber amplifier (EDFA), are coupled into the AlGaAsOI waveguide by an edge coupler with a coupling loss of 5 dB/facet. The length of the employed waveguide is 6 cm. The on-chip power of pump1, pump2, and signal are 16.77, 16.92, and 17.04 dBm, respectively. Such pump power is within the power level that an integrated semiconductor laser can provide [24]. At the receiver side, an adjustable optical bandpass filter (OBPF) is used to filter out the desired idler and then amplified by an EDFA. Another OBPF is employed to filter out the added amplified spontaneous emission noise. After being detected by a photodetector (Finisar XPDV2150R), the signals are captured by a real-time oscilloscope (DSA-X 96204Q, 80GSa/s) and processed offline in MATLAB. Square timing clock recovery and least mean square (LMS)-based linear equalizer are employed to enhance the signal quality in the offline processing.
Fig. 3. Experimental setup. PC, polarization controller; OC, optical coupler; EDFA, erbium-doped fiber amplifier; OBPF, optical bandpass filter; ATT, attenuator; PD, photodetector.
Fig. 4. (a) Spectra at the output of the chip, (b) average CE of the multicasting channels across the whole C band which matches well with the simulation results.
Fig. 5. Demodulated signal amplitude, waveform, and eye diagram of BTB, and the multicasting channels at ${\lambda _a}$ (with the highest CE) and ${\lambda _b}$ (with the lowest CE) for PAM3 (a)–(f) and PAM4 (g)–(l). (Insets: corresponding waveforms for PAM3 and PAM4).
Figure 4(a) shows the measured optical spectra at the output of the AlGaAsOI waveguide at point a. The results indicate that all the multicasting channels at ${\lambda _a} - {\lambda _e}$ are successfully generated with high CE (indicated by green circles). The average CE for all the output multicasting channels is ${-}{11.2}\;{\rm dB}$ (${\lambda _a}$ at 1547.72 nm: ${-}{7.5}\;{\rm dB}$, ${\lambda _b}$ at 1549.32 nm: ${-}{14.5}\;{\rm dB}$, ${\lambda _c}$ at 1552.52 nm: ${-}{11.5}\;{\rm dB}$, ${\lambda _d}$ at 1554.13 nm: ${-}{9.5}\;{\rm dB}$, ${\lambda _e}$ at 1555.75 nm: ${-}{13.0}\;{\rm dB}$). The corresponding optical signal-to-noise ratio (OSNR) for ${\lambda _a} - {\lambda _e}$ is 29.6, 22.4, 26.5. 28.3, and 24.3 dB, respectively. The achieved highly efficient CE is the result of the large nonlinear coefficient of AlGaAs material, low loss of the waveguide (${\sim}{0.4}\;{\rm dB/cm}$), and the compact mode area (${\sim}{0.32}\;\unicode{x00B5}{\rm m}^2$). The idlers at ${f_{113}}$, ${f_{223}}$, ${f_{221}}$, and ${f_{331}}$ are also generated [indicated by purple circles in Fig. 4(a)] but, as discussed before, they cannot preserve the original PAM information. Due to the high nonlinearity of the waveguide, some idlers generated by cascaded FWM processes can also be obtained (indicated by blue circles). The CE is also high, as shown in Fig. 4(a). Figure 4(b) shows the measured average CE of the multicasting channels as a function of the wavelength across the C band. In the measurement of Fig. 4(b), the frequency intervals of $\Delta \!{f_1}$ and $\Delta \!{f_2}$ are all fixed, while ${\lambda _1}$ is changed from short wavelength to long wavelength in the C band. From Fig. 4(b), we can see that, when ${\lambda _1}$ is either at the edge (1531.12 and 1564.68 nm) or middle (1537.4, 1542.94, 1550.12, and 1557.36 nm) of the C band, the average CEs of the output PAM3/PAM4 multicasting channels are almost the same (${\sim}{-} {11.2}\;{\rm dB}$), which indicates a good CE performance across the whole C band.
Figure 5 shows the amplitude and eye diagram of back-to-back (BTB) and multicasting channels at ${\lambda _a}$ and ${\lambda _b}$ when the received power is ${-}{9}$ and ${-}{7}\;{\rm dBm}$ for PAM3 (a–f) and PAM4 (g–l), respectively. The performance of all the new generated multicasting channels lie in between ${\lambda _a}$ and ${\lambda _b}$, which represents the highest and lowest CE, respectively. The clearly opened eye proves the good signal quality of the multicasting channels. The insets of Figs. 5(a)–5(c) and 5(g)–5(i) show the corresponding waveforms of PAM3 and PAM4. From the insets, we can see that the different PAM3/PAM4 amplitudes can be clearly distinguished from the waveforms, as labeled in the insets.
To further characterize the performance of the multicasting channels, the bit error rate (BER) performance is measured for the back-to-back (BTB) and the output multicasting channels as a function of the received optical power. The results are shown in Fig. 6. At KP4-FEC threshold ${2.4} \times {{10}^{- 4}}$, all the BER results for the multicasting channels show power penalties lower than 1.0 and 2.1 dB for PAM3 and PAM4, respectively. The power penalty is mainly caused by the difference of OSNR which is determined by the CE of each multicasting channel. Additionally, comes from the residual lights that are not fully suppressed by the OBPF used to select the given multicasting signal.
In our experiment, waveguides with the widths of 600 and 700 nm are also tested using the same experiment settings. The achieved average CE of the multicasting channels for 600 and 700 nm waveguides are ${-}{16.4}$ and ${-}{14.1}\;{\rm dB}$, respectively, which are smaller than that in the case of 800 nm width. Since the zero-dispersion wavelengths of waveguides with different widths are distinct, as shown in Fig. 1(b), when employing the same FWM inputs, the different phase mismatch causes the difference in CE.
In the experiment, by employing the waveguide with a width of 800 nm across the whole C band, an average CE of ${-}{11.2}\;{\rm dB}$ performance can be maintained. High average CE can also be achieved in other wavelength bands by employing an appropriate width waveguide and input pumps because of the dispersion difference, as shown in Fig. 1. As discussed in [25], by dispersion engineering and taking advantage of high-order phase matching, without CE being affected, an ultra-large synergetic FWM bandwidth can be achieved.
In conclusion, a high-efficiency one-to-six wavelength multicasting scheme for a 10 Gbaud PAM3/PAM4 signal using AlGaAsOI nanowaveguides was demonstrated for the first time, to the best of our knowledge. The low loss, combined with the high nonlinear coefficient and small mode area of the AlGaAsOI platform, enable us to achieve an average CE of ${-}{11.2}\;{\rm dB}$ across the whole C band under pump power compatible with an integrated laser.
Disclosures
The authors declare no conflicts of interest.
REFERENCES
1. A. E. Willner, A. Fallahpour, F. Alishahi, Y. Cao, A. Almaiman, P. Liao, K. Zou, A. N. Willner, and M. Tur, J. Lightwave Technol. 37, 21 (2019). [CrossRef]
2. M. Li, L. Zhang, and D. X. Dai, Photonics Res. 6, B13 (2018). [CrossRef]
3. D. J. Moss, R. Morandotti, and M. Lipson, Nat. Photonics 7, 597 (2013). [CrossRef]
4. D. T. H. Tan, K. J. A. Ooi, and D. K. T. Ng, Photonics Res. 6, B50 (2018). [CrossRef]
5. A. Frigg, A. Boes, G. Ren, I. Abdo, D. Choi, S. Gees, and A. Mitchell, Opt. Express 27, 37795 (2019). [CrossRef]
6. Z. Gong, A. Bruch, M. Shen, X. Guo, H. Jung, L. Fan, X. Liu, L. Zhang, J. Wang, J. Li, J. Yan, and H. Tong, Opt. Lett. 43, 4366 (2018). [CrossRef]
7. S. Amirhassan, L. Pawel, O. Yoshitomo, H. Gary, P. Nathalie, G. Alexander, and L. Marko, Opt. Lett. 44, 4056 (2019). [CrossRef]
8. K. Dolgaleva, W. C. Ng, and J. S. Aitchison, Opt. Express 19, 12440 (2011). [CrossRef]
9. C. Lacava, V. Pusino, P. Minzioni, and I. Cristiani, Opt. Express 22, 5291 (2014). [CrossRef]
10. M. Pu, L. Ottaviano, E. Semenova, and K. Yvind, Optica 3, 823 (2016). [CrossRef]
11. H. Hu, F. Ros, M. Pu, F. Ye, K. Ingerslev, E. Silva, Y. Amma, Y. Sasaki, T. Mizuno, T. Morioka, and L. Oxenlowe, Nat. Photonics 12, 469 (2018). [CrossRef]
12. L. Chang, A. Boes, X. Guo, D. T. Spencer, M. Kennedy, and J. E. Bowers, Laser Photonics Rev. 12, 1800149 (2018). [CrossRef]
13. A. Biberman, B. G. Lee, A. C. Foster, M. A. Foster, M. Lipson, A. L. Gaeta, and K. Bergman, Opt. Express 18, 18047 (2010). [CrossRef]
14. H. F. Ting, K. Wang, J. Stroud, K. Petrillo, H. Sun, A. Foster, and M. Foster, Opt. Express 25, 9276 (2017). [CrossRef]
15. M. Pu, H. Hu, H. Ji, M. Galili, and K. Yvind, Opt. Express 19, 24448 (2011). [CrossRef]
16. M. Ghagnon, M. Spasojevic, R. Adams, J. Li, D. Plant, and L. R. Chen, IEEE Photonics Technol. Lett. 27, 860 (2015). [CrossRef]
17. X. Wang, L. Huang, and S. Gao, Opt. Lett. 39, 6907 (2014). [CrossRef]
18. J. Qin, H. Shu, L. Chang, W. Xie, Y. Tao, M. Jin, X. Wang, and J. E. Bowers, in Conference on Lasers and Electro-Optics (Optical Society of America, 2020), paper SM4L.6.
19. J. J. Wathen, P. A. Apiratikul, C. Richardson, G. A. Porkolab, G. M. Carter, and T. E. Murphy, Opt. Lett. 39, 3161 (2014). [CrossRef]
20. M. Pu, H. Hu, L. Ottaviano, E. Semenova, D. Vukovic, L. Oxenlowe, and K. Yvind, Laser Photonics Rev. 12, 1800111 (2018). [CrossRef]
21. L. Chang, W. Xie, H. Shu, Q. Yang, B. Shen, A. Boes, J. Peters, W. Jin, C. Xiang, S. Liu, G. Moille, S. Yu, X. Wang, K. Srinivasan, S. Papp, K. Vahala, and J. B. Bowers, Nat. Commun. 11, 1331 (2020). [CrossRef]
22. X. Liu, R. M. Osgood, and W. M. J. Green, Nat. Photonics 4, 557 (2010). [CrossRef]
23. Y. Long, A. Wang, L. Zhou, and J. Wang, Opt. Express 24, 7158 (2016). [CrossRef]
24. A. Liu, S. Srinivasan, A. Gossard, and J. Bowers, Photonics Res. 3, B1 (2015). [CrossRef]
25. D. Kong, M. Pu, Y. Liu, Y. Zheng, E. Semenova, K. Yvind, L. K. Oxenlowe, M. Galili, and H. Hao, Opt. Lett. 45, 889 (2020). [CrossRef]
