Segmented silicon MZM for PAM-8 transmissions at 114 Gb / s with binary signaling

We experimentally demonstrate PAM-8 generation from binary electrical signals driving a silicon multi-electrode Mach-Zehnder modulator acting as an optical digital-toanalog converter. Measured BER in back-to-back configuration is used to evaluate signal quality. We demonstrate 38 GBd PAM-8 transmission below the forward error correction (FEC) threshold using minimum mean square error (MMSE) equalization. The results show that modulators with segmented phase shifters can be advantageously used to eliminate the need for high bandwidth electronic digital-to-analog converters in the generation of multilevel signals. These modulators, that can be designed and fabricated with standard CMOS compatible tools and processes, are of high interest for short range high-speed data links. ©2016 Optical Society of America OCIS codes: (130.3120) Integrated optics devices; (250.7360) Waveguide modulators. References and links 1. C. Cole, “Beyond 100G client optics,” IEEE Commun. Mag. 50(2), s58–s66 (2012). 2. Y. A. Vlasov, “Silicon CMOS-integrated nano-photonics for computer and data communications beyond 100G,” IEEE Commun. Mag. 50(2), s67–s72 (2012). 3. D. Liang and J. E. Bowers, “Photonic integration: Si or InP substrates,” Electron. Lett. 45(12), 578–581 (2009). 4. R. Dubé-Demers, S. LaRochelle, and W. Shi, “Pulse Amplitude Modulation with a Femtojoule Silicon Microring Modulator at 80-Gb/s,” in 12th IEEE Int. Conf. Group IV Photonics, post-deadline session, Vancouver (2015). 5. K. Bédard, A. D. Simard, B. Filion, Y. Painchaud, L. A. Rusch, and S. LaRochelle, “Dual phase-shift Bragg grating silicon photonic modulator operating up to 60 Gb/s,” Opt. Express 24(3), 2413–2419 (2016). 6. G. T. Reed, G. Z. Mashanovich, F. Y. Gardes, M. Nedeljkovic, Y. Hu, D. J. Thomson, K. Li, P. R. Wilson, S.-W. Chen, and S. S. Hsu, “Recent breakthroughs in carrier depletion based silicon optical modulators,” Nanophotonics 3(4–5), 229–245 (2013). 7. D. Patel, S. Ghosh, M. Chagnon, A. Samani, V. Veerasubramanian, M. Osman, and D. V. Plant, “Design, analysis, and transmission system performance of a 41 GHz silicon photonic modulator,” Opt. Express 23(11), 14263–14287 (2015). 8. C. Cole, I. Lyubomirsky, A. Ghiasi, and V. Telang, “Higher-order modulation for client optics,” IEEE Commun. Mag. 51(3), 50–57 (2013). 9. H. Griesser, A. Dochhan, and J.-P. Elbers, “DSP-enhanced Transmission for 100G+ DWDM Data Center Interconnects,” in Adv. Photonics Commun. 2014 Pap. ST1D4, ST1D.4, Optical Society of America (2014). 10. E. Vanin, “Performance evaluation of intensity modulated optical OFDM system with digital baseband distortion,” Opt. Express 19(5), 4280–4293 (2011). 11. M. Chagnon, M. Osman, M. Poulin, C. Latrasse, J.-F. Gagné, Y. Painchaud, C. Paquet, S. Lessard, and D. Plant, “Experimental study of 112 Gb/s short reach transmission employing PAM formats and SiP intensity modulator at 1.3 μm,” Opt. Express 22(17), 21018–21036 (2014). 12. M. Papuchon, C. Puech, and A. Schnapper, “4-Bits digitally driven integrated amplitude modulator for data processing,” Electron. Lett. 16(4), 142–144 (1980). 13. Y. Ehrlichman, O. Amrani, and S. Ruschin, “Improved Digital-to-Analog Conversion Using Multi-Electrode Mach-Zehnder Interferometer,” J. Lightwave Technol. 26(21), 3567–3575 (2008). 14. S. Bhoja, and others, “Study of PAM modulation for 100GE over a single laser,” IEEE Gen 100G Opt. Ethernet Study Group (2012). 15. X. Wu, B. Dama, P. Gothoskar, P. Metz, K. Shastri, S. Sunder, J. Van der Spiegel, Y. Wang, M. Webster, and W. Wilson, “A 20Gb/s NRZ/PAM-4 1V transmitter in 40nm CMOS driving a Si-photonic modulator in 0.13 um CMOS,” in 2013 IEEE International Solid-State Circuits Conference (ISSCC) Digest of Technical Papers 128– 129 (2013). Vol. 24, No. 17 | 22 Aug 2016 | OPTICS EXPRESS 19467 #267499 http://dx.doi.org/10.1364/OE.24.019467 Journal © 2016 Received 13 Jun 2016; revised 3 Aug 2016; accepted 8 Aug 2016; published 12 Aug 2016 16. A. Shastri, C. Muzio, M. Webster, G. Jeans, P. Metz, S. Sunder, B. Chattin, B. Dama, and K. Shastri, “UltraLow-Power Single-Polarization QAM-16 Generation Without DAC Using a CMOS Photonics Based Segmented Modulator,” J. Lightwave Technol. 33(6), 1255–1260 (2015). 17. D. Patel, A. Samani, V. Veerasubramanian, S. Ghosh, and D. V. Plant, “Silicon Photonic Segmented ModulatorBased Electro-Optic DAC for 100 Gb/s PAM-4 Generation,” IEEE Photonics Technol. Lett. 27(23), 2433–2436 (2015). 18. A. Samani, V. Veerasubramanian, E. El-Fiky, D. Patel, and D. V. Plant, “A Silicon Photonic PAM-4 Modulator Based on Dual-Parallel Mach-Zehnder Interferometers,” IEEE Photonics J. 8(1), 1–10 (2016). 19. A Aimone, I. Garcia Lopez, S. Alreesh, P. Rito, T. Brast, V. Höhns, G. Fiol, M. Gruner, J. K. Fischer, J. Honecker, A. G. Steffan, D. Kissinger, A. C. Ulusoy, and M. Schell, “DAC-free Ultra-Low-Power DualPolarization 64-QAM Transmission with InP IQ Segmented MZM Module,” in OFC’2016, post-deadline paper Th5C.6, Anaheim, USA. 20. A. D. Simard and S. LaRochelle, “Complex apodized Bragg grating filters without circulators in silicon-oninsulator,” Opt. Express 23(13), 16662–16675 (2015). 21. P. Cheben, P. J. Bock, J. H. Schmid, J. Lapointe, S. Janz, D.-X. Xu, A. Densmore, A. Delâge, B. Lamontagne, and T. J. Hall, “Refractive index engineering with subwavelength gratings for efficient microphotonic couplers and planar waveguide multiplexers,” Opt. Lett. 35(15), 2526–2528 (2010). 22. M.-J. Picard, Y. Painchaud, C. Latrasse, C. Larouche, F. Pelletier, and M. Poulin, “Novel Spot-Size Converter for Optical Fiber to sub-um Silicon Waveguide Coupling with Low Loss, Low Wavelength Dependence and High Tolerance to Alignment,” in 2015 41st Eur. Conf. Opt. Commun. ECOC, Mo.4.2.4 (2015).


Introduction
The growing popularity of video streaming and cloud computing is pushing the development of high bandwidth short optical links used in data centers.The commonly used 10G technologies for these links, such as parallel vertical-cavity surface-emitting lasers (VCSELs), are considered impractical for next generation systems.Different technologies are suggested as a replacement, such as silicon modulators, indium phosphide modulators, directly modulated distributed feedback lasers or long wavelength VCSELs [1,2].Integrated silicon photonics (SiP) is a technology with great potential for these applications due to its high level of device integration, low-power operation and cost effectiveness [2,3].The main types of SiP modulator are the Mach-Zehnder modulators (MZMs), the microring modulators (MRMs) and Bragg grating modulators (BGMs).MRMs, that show potential for low power consumption and high density, have been demonstrated at 66 Gb/s with a power consumption of 15 fJ/bit [4].Recently, OOK modulation at bit rates reaching 55 Gb/s has been achieved with BGMs [5] demonstrating that the performance of BGMs is now approaching the transmission rates of MRMs.However, the sensitivity to temperature and fabrication variations limit the power efficiency of these two resonant structures [6].Although they are less compact, MZMs offer a reliable temperature insensitive option that shows high performance, for example, a 70 Gb/s modulator with a power consumption of 350 fJ/bit was demonstrated in [7].
In parallel to modulator technology development, higher-order modulation formats such as pulse-amplitude modulation (PAM) or discrete multi-tone are also being investigated [8].These modulation schemes are preferred due to their high spectral efficiency and compatibility with direct detection, which are two properties of interest for short reach interconnect data links in which dispersion is not an issue [9][10][11].Moreover, multilevel formats can also present an advantage in terms of the modulator power consumption provided that the driving voltage requirement is not increased.This aspect has been investigated by comparing OOK and PAM-4 implemented with the same driving voltage, using either MRM [4] or MZM [7], demonstrating a 1.5-3 reduction in the energy per bit consumed by the modulators when using PAM-4.
To generate PAM signals, different schemes can be implemented to eliminate the need for expensive digital-to-analog converters (DACs) and lower the system complexity.For example, segmented MZMs used directly as optical DACs have been proposed as early as 1980 [12].Following the initial demonstration that used LiNbO 3 technology, Ehrlichman et al. theoretically investigated the linearity of these optical DACs [13].More recently, the study of segmented MZMs has found renewed interest in the context of high-capacity short reach data links using integrated silicon photonic transmitters for multi-level formats [14].Demonstrations of PAM-4 transmissions at 10 GBd (20 Gbit/s) have thus been performed using a silicon-insulator-silicon capacitive (SISCAP) junction [15] and the same technology was applied to QAM-16 generation [16].PAM-4 was also demonstrated up to 100 Gb/s using a depletion based phase shifter [17].Similar performances were achieved using a different optical configuration which prevents the use of expensive digital-analog converters, namely dual-parallel Mach-Zehnder interferometers [18].Finally, very recently, 64-QAM at 32 GBd was demonstrated using an indium phosphide IQ segmented Mach-Zehnder modulator [19].
In this paper, we present a silicon MZM DAC with multiple phase shifters in a series push-pull (SPP) configuration achieving PAM-8 generation for transmissions above 100 Gb/s.In section 2, we detail the modulator design and in section 3, we present characterization results of the optical and electrical frequency responses.We also investigate the analog bandwidth (BW) of the electrodes.Finally, in section 4, we perform system test under PAM-8 modulation.To the best of our knowledge, bit error rates below the forward error correction (FEC) threshold of 3.8x10 −3 up to 114 Gb/s and PAM-8 transmission without precompensation are reported here for the first time with a SiP modulator.

Modulator design
Figure 1 shows the schematic of the MZM designed to operate in the C-band.In order to generate multi-level optical output while driving the modulator directly with binary data streams (removing the need for a DAC), the phase shifter on each arm is segmented.The segmented phase shifter design simplifies considerably the generation of multi-level optical signals.To generate a PAM-8 signal, three data streams are sent to the three pairs of phase shifters operating in SPP configuration with integrated 50 Ω terminations.The SPP configuration has been chosen because it reduces the required voltage and the modulationinduced chirp.Furthermore, the integrated 50 Ω terminations were necessary to test the device without packaging.The two arms of the MZM have an intentional asymmetry that results in a free spectral range of 5.6 nm.This length imbalance allows adjusting the operation point by changing the optical wavelength.In a packaged design, a thermal phase shifter would be added to perform this function.The 3-dB couplers at the input and output are multimode interferometers (MMI) identical to those described in [20].
The three phase shifters lengths are L, 2L, and 4L, where L is 500 μm.Assuming a linear optical phase shift as a function of the phase shifter length, the segment lengths were chosen to generate a multilevel optical signal with similar reverse bias and RF power applied to each segment.However, due to fabrication errors and non-linearity in the electrode phase response, fine tuning in RF power and bias voltage applied to each segment must be done.The shorter and longer phase shifters are used for the least and most significant bit respectively.As described in [7], the transmission lines are co-planar strip (CPS) waveguides with Tshaped extensions to reduce the speed of the microwave signal and match the group velocity of the optical mode.The modulator was modeled with the RLGC circuit model of the transmission line augmented with a reverse biased PN junction modeled by a series RC circuit in shunt with the transmission line.The values of the RLGC circuit elements of the standalone CPS transmission line were extracted from a driven terminal s-parameter simulation using HFSS software.The PN junction was subsequently added in MATLAB, from which the propagation constant and the characteristic impedance were calculated.The SPP configuration halves the PN junction capacitance seen across the transmission line leading to a reduction in the microwave loss.However, this reduction in capacitance also results in a decrease of the microwave index.Relative to regular CPS electrodes, these T-shaped extensions increase both the inductance and capacitance of the transmission line, therefore increasing the index of the microwave.The capacitance is determined by the spacing between the edges of the T-shaped extensions, while the inductance is controlled by the width of the stem of the T-shaped extension.The stem and the pad areas were chosen to be the smallest allowed by the fabrication design rules.The modulator electrodes were designed on the top 2-um thick metal layer, the thickest of the metal layers available, to reduce the skin effect loss.Since the characteristic impedance also changes as the ratio of the inductance to the capacitance, a set of geometrical parameters giving the largest microwave index while maintaining an impedance matching close to 50 Ω were chosen by performing various sweeps.The impedance was also designed to match the 50 Ω of the RF drivers and the on-chip resistance.Each section of the modulator was biased through a thin line to provide some inductance and avoid the high speed RF signal from shorting to ground.
The modulator structure was fabricated at IME using 193 nm deep UV photolithography.The wafer had a 220 nm top-silicon layer and a BOX thickness of 3 μm.The optical waveguides outside the phase shifters are full-etched 500 nm wide strip waveguides.The modulator ridge waveguides have a width of 500 nm and a 90 nm slab thickness.The PN junction was positioned at the center of the waveguides.Finally, optical coupling is made with focusing grating couplers optimized for 1550 nm.

Modulator characterization
In this section, the optical, electrical and electrical/optical (E/O) characteristics of the modulator are investigated.The modulator insertion loss (spectral transmission) is measured using a commercial wavelength-swept interferometric system (OVA, Luna Technology).A typical spectral measurement is shown in Fig. 2(a).The 12 dB insertion loss caused by the grating couplers were removed.This coupling loss could be reduced significantly for example by using edge couplers with on-chip subwavelength gratings inverted tapers [21] or Si 3 N 4 multi-rod [22].The on-chip modulator loss (~5 dB) is detailed as follow: the background waveguide propagation loss is about 1.5 dB, the loss due to doping is evaluated to be around 2.5 dB, while waveguide converters and MMI excess loss is ~1 dB.The electro-optical (E/O) characteristics of the device were measured with a 40 GHz vector network analyzer.The normalized E/O S21 measurements are shown in Fig. 2(b) as a function of reverse bias voltage for the long phase shifter.The E/O bandwidth is greater for the short phase shifter since the microwave loss and the impact of the velocity mismatch between the optical and the electrical wave are reduced.The 3 dB bandwidth is ~38 GHz for the medium and long phase shifters while it is higher than 40 GHz for the short phase shifter.Figure 2(b) shows that the 3 dB electrical bandwidth slightly increases with the reverse bias voltage.However, due to the nonlinear dependence of the depletion width variation, and its overlap with the optical mode field, as the reverse bias voltage changes, a larger eye opening (i.e.separation between the levels in the vertical direction) is observed when the reverse bias voltage approaches 0 V.This is expected because, at low reverse bias, modulation of the carrier density will occur in the center of the waveguide where the mode field is higher resulting in a larger phase shift.Consequently, careful optimization of the reverse bias voltage is required to obtain the best compromise between frequency response and eye opening.There is also a trade-off between the modulator extinction ratio and the optical modulation amplitude that, in the present case, is optimized by wavelength tuning.

System analysis
The experimental setup is shown in Fig. 3. To generate the PAM-8 signal, each segment of the MZM is electrically driven by a two-level PRBS of length 2 15 -1 generated by a bit pattern generator (BPG).Two tunable delays (not shown) are used for decorrelation and also for skew adjustment to ensure a perfect phase match.Each binary stream is fed into a RF amplifier.For all experiments, the modulation peak-to-peak voltage is 4 V (V RF,3 in Fig. 3) for the long phase shifter and 3.75 V for the middle and short segments (V RF,2 and V RF,1 in Fig. 3); our experimental setup imposed a limitation on the middle and short segments voltage amplitudes that could not be adjusted independently.Additionally, the reverse bias voltage for all the PN junctions is set to 2 V for the three phase shifters (V DC,1 = V DC,2 = V DC,3 in Fig. 3) for all the experiments.The signal wavelength is set to 1551.18 nm and its polarization is adjusted with a polarization controller (PC) to maximize coupling into the fundamental TE mode.We evaluate that ~1 dBm is injected into the MZM.After modulation, the signal is fed into an erbium-doped fiber amplifier (EDFA) followed by a tunable optical bandpass filter.A variable optical attenuator (VOA) is used to sweep the received power at the photodetector (PD), which has a 70 GHz bandwidth but no integrated transimpedance amplifier.Finally, the detected signal is captured with a real-time oscilloscope (RTO) (63 GHz bandwidth, 160 GSa/s sampling rate, Keysight DSOZ634A) for offline signal processing and BER calculations.The digital signal processing is as follows.First, a second order Gaussian low-pass filter is applied.Retiming and resampling at the symbol rate are next performed and a MMSE equalizer with 50 taps trained over 2000 symbols is then applied.Finally, the vertical decision thresholds are optimized to minimize the BER.For each measurement, the BER is calculated over 50 000 symbols. Figure 4 shows the eye diagrams captured with an optical sampling oscilloscope with 5 dBm received optical power and Fig. 5 shows measured BER against the received power for 10, 20, 30, and 38 Gbaud.Assuming a FEC threshold of 3.8x10 −3 , errorfree transmission is achieved at a received power of approximately −5.4 dBm, −2.2 dBm, −0.6 dBm, and 2.4 dBm for 10 Gbaud (30 Gbps), 20 Gbaud (60 Gbps), 30 Gbaud (90 Gbps), and 38 Gbaud (114 Gbps), respectively.The power consumption of this modulator with SPP electrodes, dominated by resistive power loss, can be estimated as in [7] according to where R is the termination resistance and V rms,i are the rms driving voltages of the three electrode segments.This estimate implies that the line impedance is matched to the source and termination (50 Ω).
The calculated energy per bit is E b ≈7.7 pJ/bit.It should be noted that the modulator was specifically designed to be driven by the same voltages for all three segments and therefore was not optimized to minimize power consumption.Since the bandwidth is limited by the longest electrode, using same length electrodes for all segment could help to reduce power consumption but at the expense of higher optical insertion loss.This aspect requires further investigation as optimization of power consumption should also consider equalizer requirements and the fact that the segmented MZM removes the need for an electronic DAC.

Conclusion
We achieved for the first time, to the best of our knowledge, 114 Gbps PAM-8 modulation with measured BER below the FEC threshold (3.8x10 −3 ) with a multi-electrode silicon MZM DAC and MMSE equalization.Our results demonstrate that silicon multi-electrode MZM DACs offer a practical solution for the needs of short reach 100G and 400G links.

Fig. 2 .
Fig. 2. (a) Insertion loss measurement of the MZM without any voltage applied to the electrodes (b) Normalized E/O |S21| for the long phase shifter with reverse bias voltages from 0 V to 8 V.

Fig. 5 .
Fig. 5. BER as a function of the received power.