1. Introduction

Global information communications demand optical transmission systems with high spectral efficiency, high per-channel data rate, and low cost. Advanced optical modulation formats, together with wavelength- and polarization-division multiplexing, are key enablers to increase the spectral efficiencies and the aggregation data rates within the limited spectrum bandwidth of available optical amplifiers [1, 2]. High capacity coherent optical transmitters and receivers require many advanced optical components, such as modulators, detectors, power splitters/combiners, polarization splitters, wavelength multiplexing filters, and so on. These components are ideally implemented on photonic integrated circuits (PICs) with the advantages of accurate optical path control, compact size, low power consumption, and potentially low packaging cost.

Emerging silicon photonic technologies promise powerful PIC platforms for the optical communications [3–6]. Silicon optoelectronic devices employ high-index silicon core within low-index silica cladding structures. These waveguide structures allow a tight bending radius down to ~1 μm, resulting in extremely compact optical components. Large-scale silicon PICs also promise high yields and uniform performance of optical elements by leveraging the mature complementary metal–oxide–semiconductor (CMOS) microelectronic fabrication infrastructures. Foundry service is available to reduce chip manufacture cost. Further integration of silicon PICs with CMOS driver circuits lead to more complex functions with low power consumption and low packaging cost.

Recently, silicon waveguide modulators were extensively investigated to improve the modulation rates and reduce the drive voltages [7–17]. Nevertheless, the performance of silicon modulators are usually still worse than LiNbO₃ modulators, partially due to the weak electro-optic effect in the silicon modulators using the free-carrier induced index change [18]. As results, most of the reported silicon modulators focused on on-off-keying (OOK) modulation format, geared for short-reach applications which are less demanding in extinction ratios and linearity. Theoretical studies of silicon quadrature phase-shift keying (QPSK) modulation have been reported previously [19–21]; experimental demonstration of 20-Gb/s QPSK signal was only reported recently by using silicon microring modulators [22]. The design and fabrication of nested silicon Mach-Zehnder modulators (MZMs) for QPSK was reported in [23] without successful demonstration of QPSK constellations nor bit-error ratios (BERs). In this paper, a 50-Gb/s QPSK modulation is experimentally realized by using silicon MZMs. Although quadrature amplitude modulation (QAM) has been demonstrated by LiNbO₃ modulators, III-V semiconductors modulators [24], and polymer modulators [25], to our best knowledge, this paper is the first to successfully generate advanced modulation formats utilizing silicon MZMs.

2. Silicon QPSK modulator

QAM is a modulation scheme with optical constellations made of multi-level amplitudes and phases to generate optical signals with high spectral efficiency. Among various QAM formats, QPSK has received significant attention in the optical communication industry as a de facto standard for 100-Gb/s polarization-division-multiplexed QPSK (PDM-QPSK) transponders in long-haul optical communication systems. QPSK transmits four phase shifts (spaced by π/2) with constant amplitudes at a symbol rate of half the aggregate bitrate. A QPSK transmitter can be implemented by two nested MZMs, each operating as a binary-phase modulator. A MZM can produce exact π-phase shift if biased around the zero transmission and driven by a push-pull scheme. To produce QPSK, two modulated fields from binary-phase MZMs are combined with a π/2 phase difference.

Push-pull operation of a MZM can be achieved by using either a dual-electrode configuration with two arms driven by differential data, or a single-electrode where both arms are connected together. While the dual-electrode configuration requires only half of the drive voltages for the single-electrode configuration, the latter may be very useful due to reduced number of RF inputs, especially in the applications where multiple modulators need to be integrated. Previously, we reported single-drive push-pull silicon MZMs [15, 17]. A low V_π of 3.1 V was demonstrated at a modulation speed up to 30 Gb/s [17]. The modulation mechanism in these MZMs is based on the carrier depletion of the silicon pn junction, embedded in the middle of silicon waveguides. A travelling-wave transmission-line electrode is loaded with the junction capacitor. In our modulator design, both modulator arms are symmetrically doped and share a highly n-doped (or p-doped) region in the center. Symmetric coplanar strips are connected with outside highly p-doped (or n-doped) regions of two MZM arms. Since the two junction capacitors in two MZM arms are connected in series, the loaded capacitance on the transmission line is approximately half of that for one arm. This leads to easier design of travelling-wave electrodes for high speeds, while allows both arms are driven in a push-pull fashion with a single input drive signal.

This silicon QPSK PIC has a size of 6.5 mm x 1.4 mm, shown in Fig. 1(a) . The optical power splitters/combiners are implemented by 1x2 and 2x2 multimode interference couplers (MMIs). Thermal phase shifters are employed to the MZM arms to achieve zero-transmission bias (before RF modulation). They are also employed before the final power combiner to adjust 90-degree phase shift between two modulators. These phase shifters are implemented by doped Si-waveguide resistors. The segmentally loaded traveling-wave MZMs here are similar to those reported in Ref [15], with a V_π of ~10 V, an on-chip insertion loss of ~4.2 dB, and a modulation rate up to 30 Gb/s. Detailed design and fabrication have been reported in Refs [12, 15, 17].

 

Fig. 1 Silicon-PIC QPSK modulator. (a) Photograph of a fabricated silicon PIC. (b) Photograph of the packaged silicon PIC with a printed circuit board (PCB). RF and DC represent RF transmission lines and DC lines on the PCB.

Download Full Size | PDF

3. 50-Gb/s QPSK generation

The silicon PIC was wire-bonded to a printed circuit board (PCB) with high-speed transmission lines, DC metal lines, RF connectors, and DC connectors (Fig. 1(b)). Off-chip 50-ohm resistors are bonded to the end of the transmission lines of Si MZMs. Lensed fibers were used to couple the light into the chip.

We performed the QPSK generation using the measurement setup shown in Fig. 2 . The in-phase/quadrature (I/Q) signals use the same 25 Gb/s pseudorandom bit sequence (PRBS) of a length 2¹⁵–1 from a pattern generator, delayed by 57 bits from each other (Fig. 2(a)). The signals are then amplified individually by two broadband electrical amplifiers to a peak-to-peak voltage of ~12 V. A continuous-wave laser of 1540.0 nm from an external cavity laser with ~1 MHz linewidth was fed to the device. The optical output power was adjusted by a variable optical attenuator (VOA) and then amplified by two erbium-doped fiber amplifiers (EDFAs) and filtered by a tunable optical filter, to achieve the desired optical signal-to-noise ratios (OSNRs) (Fig. 2(b)). The received optical signal was measured and analyzed by an optical spectrum analyzer (OSA), a high-speed sampling oscilloscope, and a real-time 80-GSamples/s digital sampling oscilloscope.

 

Fig. 2 Experimental setup for characterizing the packaged Si QPSK modulator. (a) Electrical drive signals. (b) Optical setup. MZM: Mach-Zehnder modulator; PC: polarization controller; VOA: variable optical attenuator; EDFA: erbium doped fiber amplifier; OSA: optical spectrum analyzer; LO: local oscillator, DSP: digital signal processing.

Download Full Size | PDF

Proper tuning of three phase shifters is required in order to achieve QPSK. We first tuned the phase shifters within the MZMs such that a π-phase difference (zero-transmission point) between the MZM arms are obtained. Next, the phase shifter after one of the two MZMs are adjusted to achieve a π/2-phase difference between two MZM outputs. We then turned on the RF drive signals. The measured optical eye diagram and power spectrum are shown in Figs. 3(a) and 3(b), respectively. The three transition levels in the eye diagram and the suppression of the carrier in the spectrum exhibit the characteristics of the QPSK signal. The thick rails in the eye diagrams may come from: (1) the drive voltage is not high enough to achieve 2V_π; and (2) the limited bandwidth of the modulators.

 

Fig. 3 Results of QPSK modulation. (a) QPSK eye diagram. (b) QPSK optical spectrum (resolution bandwidth is 0.01 nm).

Download Full Size | PDF

Next, we measured the optical signal with a real-time digital sampling oscilloscope at a sampling rate of 80 GSamples/s together with an optical coherent receiver. A homodyne detection scheme is utilized by using the local oscillator (LO) source from splitting the CW laser source launched into the device (Fig. 2(b)). The sampled waveforms are processed off-line with digital signal processing (DSP) codes including a digital finite impulse response (FIR) filter, optimized with the constant modulus algorithm (CMA), and Viterbi-Viterbi feedforward carrier recovery [26, 27]. Differential decoding is used in the detection and the BER of the received signal is calculated by direct error counting.

The recovered QPSK constellations after DSP are shown in Figs. 4(a) and (b) with OSNRs of 29.7 dB and 17.8 dB, respectively, where QPSK modulation are successfully generated by this device. Figure 5 depicts the measured BER versus OSNRs (noise in 0.1-nm bandwidth). At a BER of 1x10⁻³ (typical requirement in forward error correction (FEC) techniques), the required OSNR for the 50-Gb/s QPSK is about 12.9 dB, which is approximately 2.7-dB higher than the theoretical limit. This OSNR penalty is only about 1-dB higher than a commercial LiNbO₃ QPSK modulator. The nonlinear BER curve in Fig. 5 indicates some noise floors from this device. Such BER behaviors typically happen in high-speed advanced modulation formats due to hardware limitations (from both electrical and optical components). FEC techniques are generally used to achieve error-free transmissions in coherent optical communications [2].

 

Fig. 4 Results of coherent detection of generated QPSK signals. (a) and (b), constellations after DSP taken at an OSNR of 29.7 dB and 17.8 dB, respectively.

Download Full Size | PDF

 

Fig. 5 BER versus OSNR (0.1-nm noise bandwidth).

Download Full Size | PDF

4. Discussions and conclusion

Since the changes of the effective index of silicon waveguides with a reverse-biased pn junction have a nonlinear dependence on the voltage, residual chirps in the push-pull operation may be induced. Nevertheless, this chirp together with the dispersion during transmission can be compensated by proper DSP algorithms [27]. The V_π in the current device is about 10 V, which is significantly higher than that of LiNbO₃ modulators. Based on our previously demonstrated single-drive push-pull silicon modulators of a V_π of 3.1 V at 30 Gb/s [17], it is possible to further improve the current design to achieve low-voltage and high-performance silicon coherent transmitters.

In conclusion, we have demonstrated a 50-Gb/s QPSK modulator based on two nested single-drive push-pull silicon MZMs. This demonstration illustrates that silicon photonics could find applications in coherent optical communications. With further integration of polarization rotators and splitters/combiners in silicon photonics [28, 29], monolithic dual-polarization coherent transmitters could be implemented in silicon PICs. With many benefits such as low cost, compact size, high yield, and low power consumption, silicon photonic integrated circuits is a promising candidate to drive commercial high capacity coherent transmitters.