DAC-less PAM-4 generation in the O-band using a silicon Mach-Zehnder modulator

We demonstrate 20-Gb/s 4-level pulse amplitude modulation (PAM-4) signal generation using a silicon Mach-Zehnder modulator (MZM) in the O-band. The modulator is driven by two independent binary streams, and the PAM-4 signal is thus generated directly on the chip, avoiding the use of power-hungry digital-to-analog converters (DACs). With optimized amplitude levels of the binary signals applied to the two arms of the MZM, a preforward error correction (FEC) bit-error rate (BER) as low as 7.6 × 10 is obtained. In comparison with a commercially available LiNbO3 modulator, the penalty is only 2 dB at the KP4 FEC threshold of 2.2 × 10. © 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
To cope with the demand of increasing bit rates for a limited bandwidth, 4-level pulse amplitude modulation (PAM-4) is considered as a promising economical solution to replace on-off keying (OOK) with a strong potential for large volume short-reach optical communications such as inside data centers, in high performance computing (HPC) or in centralized radio access networks (CRANs) [1][2][3]. For this purpose, silicon photonics is a well-suited technology, benefiting from both the complementary metal-oxyde-semiconductor (CMOS) mature fabrication process and from its possibility of integration with electronics [4][5][6][7][8][9].
PAM-4 signals are usually generated in the electrical domain by means of power-hungry digital-to-analog converters (DACs), before being converted to the optical domain by the modulator. To reduce the overall power consumption, new DAC-less configurations have been proposed, enabling the generation of the four intensity levels in the optical domain from two independent binary electrical sources [5][6][7][8][9][14][15][16][17][18][19][20][21][25][26][27][28]. A common approach to generate PAM-4 levels with two driving sources is to use segmented electrodes, as it has the advantage of lowering the power consumption and allowing high symbol rates [14]. However, segmented electrode structures are designed only for PAM applications, while dual electrode structures are more flexible and can also be used for phase-shift modulation formats [30].
In this context, we present a carrier-depletion based Mach-Zehnder modulator operating in the O-band and generating 10 Gbaud PAM-4 signals without using any DAC. The modulator is driven by two independent binary streams, with well-chosen amplitudes, to obtain 4 equally spaced optical power levels. After introducing its design, we describe the method used for driving the modulator, and then demonstrate its performance experimentally in terms of eyediagrams and bit-error-rate (BER). The power penalty compared to the same generation scheme but employing an LiNbO 3 MZM is furthermore assessed experimentally.  The device was fabricated at STMicroelectronics using a 300-mm SOI technological platform [31] and has been previously used to demonstrate 25 Gb/s OOK modulation [32]. The crosssection of the phase shifter can be seen in Fig. 1. The waveguide width is 400 nm, its height is 300 nm, and the thickness of the slab region is 50 nm. Targeted doping concentrations are 5 × 10 17 cm −3 and 4.5 × 10 17 cm −3 for the P and N region, respectively. These concentrations were extracted from a numerical model, as a compromise to obtain an efficient phase-shifter while benefiting from sufficiently high bandwidth and low losses [33,34]. Near the contacts, P ++ and N ++ regions with doping levels of 10 19 cm −3 were placed to reduce the access resistance of the device. It can be noticed that the PN junction is slightly shifted (25 nm away from the vertical axis of symmetry of the rib waveguide towards the N-doped region), leading to a higher overlap between the P region and the optical mode, to benefit from the higher efficiency of hole concentration variations on the induced phase shift. The phase shifter length is 1 mm (Fig. 2). Thermal heaters (resistors) were added at the end of each arm. They are used for tuning the operating point of the MZM with a direct current (DC) supply. The data stream is applied to the phase shifters through coplanar travelling wave electrodes (CTWs), designed for GSGSG probes. Light is coupled in and out of the chip by means of grating couplers. The splitter and combiner are 3 dB 1 × 2 multi-mode interference (MMI) designs.

Device design and characterization
An 18 GHz electro-optical bandwidth was measured and the static characterizations reported in [32] showed a 2.5 dB insertion loss (IL). The measured efficiency of the PN junction is in the range of 1.2 V × cm to 1.45 V × cm for 0 V to −7 V bias, respectively.

PAM-4 generation
The principle used for PAM-4 generation was described in [16]. By applying two independent binary non-return-to-zero (NRZ) electrical signals V 1 and V 2 with different peak-to-peak voltages on the arms 1 and 2 of an MZM operating at the quadrature point, four distinct levels are reached in the output power. In case of modulation by carrier depletion in Si PN diode, both the losses and the efficiency of the phase-shifters present a non-linear behavior. The electric field traveling through each arm will therefore go through a non-linear phase-shift and voltage-dependent losses at the same time. Modeling can thus be used to find out the voltages required to reach 4 equally spaced power levels, while exploiting the full dynamic range of the MZM transmission.
The electric field at the output of the arm labelled n of a MZM can be expressed as For a dual-drive silicon modulator, V 1 and V 2 are defined as Where DC V is an offset voltage used to bias the PN junction in reverse way, and n V Δ is the alternative voltage applied on the phase shifter n to control φ Δ .
In Figs. 3(a) and 3(b), the black curve corresponds to the push-pull transmission 1 2  Fig. 3(a), it can be seen that when the separation between the two center power levels is larger than between the extreme levels, which is not optimal in terms of signal-to-noise ratio and BER performance. It is then possible to tune the ratio 1 2 / V V Δ Δ to achieve equally spaced power levels, as illustrated in Fig. 3(b). In this case, V (voltage ratio of 1.74), simultaneously leading to equally spaced power levels and maximum extinction ratio. As it can be seen from this modeling, to reach four equally spaced power levels while benefiting from the full dynamic range of the transmission, the ratio It can be noted that the level differences between the two upper and between the two lower levels of the PAM-4 signal is set by 2 V Δ . On the other hand, 1 V Δ is directly related to the difference between the second and third levels.

Experimental setup
The setup used for the PAM-4 BER experiments is shown in Fig. 4. Light is produced by an O-band tunable external cavity laser (operating at 1310 nm) and amplified by a semiconductor optical amplifier (SOA) before being coupled in the device though grating couplers. Coupling losses are around 5dB per coupler. At the output, the beam is amplified by a praseodymium-doped fiber amplifier (PDFA) and a 3 dB directional coupler splits the light into two paths. One goes directly to an oscilloscope (25 GHz optical module), allowing the visualization of the eye diagram in real time, while the other is input to a 40-GHz PIN photodiode followed by a 35 GHz traveling wave amplifier and feeding the error counter of a BER tester (BERT). For the electrical part, a 10 Gb/s pattern generator generates two complementary pseudorandom binary patterns (PRBSs) with 2 15 -1 lengths. One of them is delayed by 200 ps (corresponding to the duration of 2 symbols) by a radio-frequency (RF) tunable delay line (TDL), so that both outputs can be considered as uncorrelated. The two data signals are then amplified. The gain of each amplifier can be tuned specifically in order to reach 4 equally spaced levels as detailed in the previous section. A −4V DC voltage is added to both signals through bias tees to bias the PN junction, and GSGSG probes transmit the resulting signals to the device. At the output of the MZM RF electrodes, 50 Ω terminations absorb the RF power after DC blocks. A DC voltage is also applied on the thermal heaters to tune the MZM operating point to quadrature.

Experimental results
First, we show in Fig. 5 the eye diagrams that are obtained for different peak-to-peak voltages applied on each arm of a quadrature-biased MZM. On the left panel are reported the equivalent modeling, showing a good correlation of the simulated levels with the experimental results for each of the voltage couples used in the experimental setup. It can be seen from Figs. 5(c) and 5(d) that equally spaced levels can be obtained for a voltage ratio of 2.47, which is different from the optimum ratio of 1.74 mentioned in Section 3. This comes from the fact that the ratio of 1.74 fulfills both the even-spacing and the maximum extinction ratio conditions, whereas in the eye diagrams represented in Figs. 5(c) and 5(d) the extinction ratio is limited.
BER measurements were performed at 20 Gb/s using an error analyzer (EA) designed for conventional binary modulation using the method described in [7,35]. The applied peak-topeak voltages were set to 7 V and 4 V, to benefit from an almost full dynamic range of the transmission while obtaining 4 equally spaced levels. These voltages are slightly lower than the ones stated in Section 3, due to limitations of the RF amplifiers maximum output power, but the voltage ratio of 1.75 is maintained. The received power was swept from −8.5 dBm to 0.5 dBm.
The results were compared to the case of a commercial 32 GHz-bandwidth push-pull LiNbO 3 MZM. The modulator was driven by two NRZ electrical streams with peak-to-peak voltages of 3.25 V and 1.90 V. The static V π of this LiNbO 3 modulator was measured to be 4 V and the extinction ratio was 9 dB in the O-band. The SOA was removed from the set-up as coupling losses in the LiNbO 3 MZM are lower than for the Si modulator. The eye diagrams at 0 dBm received power corresponding to the two different modulators are plotted in Figs. 6(b) and 6(c).

Conclusion
DAC-less modulators for PAM-4 generation have been a topic of increasing interest over the past years within the silicon photonics community. We experimentally demonstrate an Oband 20 Gb/s PAM-4 generation using a simple silicon MZM driven by two independent binary streams. Interestingly, this generation method can be applied to higher bitrates, up to 100 Gbit/s, by phase shifter optimization. Modeling has been used to evaluate the influence of the voltages applied on each arm to the output optical transmission levels. By doing so it is possible to benefit from the full dynamic range of the transmission, while obtaining 4 equally spaced levels. The BER obtained with the Si PAM-4 modulator is below the KP4 FEC threshold of 2.2 × 10 −4 for an average received power of −3.4 dBm and corresponds to a penalty of 2 dB compared to a commercial push-pull LiNbO 3 MZM.