Embed Size (px)
Citation preview
Silicon Photonic Modulators for High-Capacity Coherent Transmissions Wei Shi, Jiachuan Lin, Hassan Sepehrian, Sasan Zhalehpour, Zhuhong Zhang, and Leslie A. Rusch IEEE/OSA Optical Fiber Communications Conference (OFC 2019) © 2019 IEEE/OSA. Personal use of this material is permitted. Permission from IEEE/OSA must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
Tu2H.1.pdf OFC 2019 © OSA 2019
Silicon Photonic Modulators for High-CapacityCoherent Transmissions
Wei Shi,1∗ Jiachuan Lin,1,2 Hassan Sepehrian,1,2 Sasan Zhalehpour,1Zhuhong Zhang,2 Leslie A. Rusch1
1ECE Department and Center for Optics, Photonics and Lasers (COPL), Universite Laval, QC, Canada2 Huawei Technologies Canada Co., Ltd., ON, Canada
(Invited paper)
Abstract: We discuss system-orientated design and optimization of all-silicon modulatorsfor high-baud-rate (up to 84GBaud) coherent transmissions. We achieved single-carriernet-600Gb/s DP-32QAM and DP-16QAM over 1520km; and 800Gb/s super-channel usinga silicon-modulator optical frequency comb.
OCIS codes: (130.4110) Integrated Modulators; (060.1660) Coherent communications; (250.0250) Optoelec-tronics; (060.233) Fiber optics communications.
1. Introduction
Photonic integrated circuits are a key enabler of future ultra-high-capacity optical transmission systems. Usingdensely integrated high-bandwidth optical transceivers, the footprint and power consumption of WDM systemswill be dramatically reduced compared to current solutions using bulky discrete optical components. High-baud-rate, single-carrier transmission has been extensively explored on photonic integration platforms, such as InP [1,2],thin film polymer on silicon [3] and organic hybrid [4]. Compared to other platforms, silicon photonics (SiP)leverages mature CMOS technology and offers a smaller form factor, leading to lower-cost mass production andlarge-scale photonic integration [5–7]. All-silicon modulators are preferred as they are readily integrated with otherphotonic functions, such as wavelength and polarization multiplexers, optical equalization, and Ge photodectors,on the same fabrication process. Over the last decade, we have seen significant progress in silicon modulators forpulsed amplitude modulation (PAM) links [6] and coherent transmissions [7]. Here we review our recent progressin silicon-modulator-based coherent transmitters for high-baud-rate quadrature amplitude (QAM) and on-chipoptical frequency comb generation for flexible, bandwidth-efficient optical transmission systems.
2. Silicon IQ-modulator for QAM
Figure 1(a) shows the schematic of our silicon IQ modulator consisting of a pair of traveling-wave Mach-Zehndermodulators (MZMs) with phase shifters in lateral p-n junctions working in the depletion mode. A series push-pulldriving configuration is adopted to reduce the capacitance for a higher bandwidth that is dominated by RF losses[8]. As shown in Fig. 1(b), the MZM is designed based on a 220-nm-thick silicon-on-insulator (SOI) platformwith two metal and six doping layers, for which details can be found in [9].
The system-level performance of a silicon modulator in a transmission link can be characterized by the modu-lator power penalty (MPP) with contributions from 1) optical attenuation in doped silicon waveguides, 2) modula-tion loss due to a voltage swing smaller than Vπ , and 3) intersymbol interference (ISI) due to limited electro-opticbandwidth. These three penalty sources are intricately intertwined. Therefore, an optimal design for a give processcan only be achieved by minimizing the total penalty. We recently proposed a bandwidth-aware figure of merit(FOM) for silicon modulators in a pulse-amplitude modulation (PAM) link [10] to minimize this penalty. A nu-merical approach, rather than an analytic FOM, was used for modulator design optimization (minimum MPP) forcoherent transmissions. In either case, we first designed a unit-length MZM for a 50-Ω RF transmission line witha microwave traveling velocity matched to the optical group velocity. Then the phase-shifter length L and the biasvoltage Vb were optimized to minimize the MPP. Figure 1(c) shows the simulated MPP as a function of L at variousVb. We can see that the the minimum value (∼ 10.7dB) is achieved at Vb = 0.75V near L = 4.5 mm that was chosenin our design.The device was fabricated through a CMC-IME MPW run. The measured frequency response of theMZM (Fig. 1(d)) shows a bandwidth of 22 GHz and 35 GHz at zero bias and Vb = −3V, respectively. However,the best modulation performance was achieved at Vb = 0.75V, as shown in Fig. 1(e), due to the optimal trade-offbetween modulation efficiency and bandwidth as predicted in Fig. 1(c). The back-to-back (B2B) bit error rates(BERs) for 60 GBaud 16-QAM with 2Vpp without pre-compensation are shown in Fig. 1(f). Again, Vb = 0.75Vshows the best performance thanks to the lowest MPP.
Tu2H.1.pdf OFC 2019 © OSA 2019> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
2
bandwidth, to probe the device capacity. The rest of this paper is organized as follows. In Section II,
we present a general, unit-length MZM design and its optical and electrical characteristics. Section III optimizes the IQ modulator design. We examine power penalties induced by a SiP IQ modulator in a QAM transmission link; using the optical and electrical characteristics extracted in Section II. We identify the optimal combination of length and bias voltage of the phase shifter, that is, the combination that gives the minimum total modulator-induced power penalty. We found, via simulation, a constraint on IQ branch separation to avoid RF crosstalk between the branches. In section IV, our experiment setup and measured results are presented. Finally, Section V concludes this paper.
II. DESIGN AND CHARACTERISTICS OF TW-MZM Figure 1 presents a schematic of the SiP IQ modulator that
consists of a pair of MZMs nested in an MZ interferometer. Each MZM uses a phase shifter employing lateral p-n junction and a traveling-wave (TW) electrode for high-frequency operation. The TW-MZM is driven in a series push-pull configuration. A negative voltage is applied between the two arms of each MZM; thus, the phase shifter works in the depletion mode.
A. TW-MZM Design Figure 2a shows the cross-sectional schematic of the SiP
TW-MZM (identical for I and Q) using a CMOS-compatible process (A∗STAR IME, Singapore) on a 220-nm SOI wafer with 2 µm buried oxide. The phase shifter applies a lateral p-n junction embedded in the centre of a rib waveguide. While highly doped P++ and N++ regions are used for ohmic contacts, intermediate P+ and N+ doping levels are used to reduce the series resistance without significantly increasing optical propagation loss.
The phase shifter can be modeled as a series of lumped capacitors and resistors [7], whose values depend on the applied voltage. We conduct 2D simulation in Lumerical Mode and Device software at wavelength =1550 nm to predict the electro-optic response of the phase shifter and to calculate the circuit components. Figure 2b shows the calculated junction capacitance (Cj) and resistance (Rj). We calculate the overlap between the optical mode and the carrier distribution to predict the changes in refractive index, Dneff, and loss as functions of reverse voltage, as shown in Fig. 2c. More details can be found in [8].
As shown in Fig. 3a, the TW electrode has a coplanar stripline (CPS) configuration loaded by the p-n junction. For velocity matching between optical waves and RF driving signals, the RF effective index of the loaded CPS waveguide (nRF,l) should be as close as possible to the group index of the optical mode (no,g). We achieve this by employing a slow-wave design that intentionally adds metal capacitance to the transmission line to slow down the RF signal. We added a “T” shaped extension to the metal capacitance of the the CPS transmission line to improve the velocity matching while
maintaining a 50 Ω impedance. Detailed design procedures for slow-wave CPS transmission lines is beyond the scope of this paper and was recently reported in [7]. The bandwidth of the MZM is eventually dominated by the RF propagation loss in the p-n junction loaded CPS [8]. The TW electrode is terminated by an on-chip 50 Ω load implemented using N doped silicon.
B. RF Characteristics of Fabricated Device We had the TW-MZM design in Figs. 1 and 2 fabricated at
IME. We measured the small-signal frequency response using a 67 GHz Keysight network analyzer for two bias voltages, 0 V and -0.75 V. After removing the effects of the RF probes, cables and unloaded section of the transmission line, we extracted the RF parameters including propagation loss, effective index and transmission line input impedance.
Figure 3b presents the microwave effective index as a function of frequency. We can see that at 30 GHz it changes from 3.67 to 3.5, as the bias voltage Vb changes from zero to -0.75 V, respectively. Despite this variation, these values are
l
(a)
(b) (c)
Fig. 2. a) Cross section of lateral P-N junction waveguide with (all dimensions in µm) Wp++ = 5.2, Wp+ = 0.83, Wp = 0.37, Wn++ = 5.2, Wn+ = 0.81, Wn = 0.39, W = 0.5, h= 0.22, t1 = 2, t2 = 0.9, b) resistance and capacitance of the lateral pn junction vs. reverse bias voltage, and c) ∆neff vs. reverse bias voltage for the lateral pn junction in (a).
VIA 1
M2VIA2
BOX
SiO2
t1
W
Wp
M2
M1
M2
M1M1
M2
P++ N++P N+P+hWp+
Wnt2
Wn+
Wn++Wp++
DC
N
S G
Si Substrate
0 2 4 6 80
1.5
3
Reverse Voltage (V)10
4
7
10
∆n ef
f
pn (
dB/c
m)
×10-4
α
2 4 6 8 101.5
1.54
1.58
Reverse Voltage (V)
Res
ista
nce
(Ω .c
m)
2
4
Cap
acita
nce
(pF/
cm)
0
CjRjn Rjp
P+N++ NN+ P P++
Fig. 1. Schematic of the IQ modulator using SiP TW-MZMs with a series push-pull driving scheme.
Vb
Matched load
Heater (I)
p
n
n
G
S
Matched load
Heater(Q)
p
n
n
G
S
Heater
L
VRF, I
Vb
I
Q
S=700µm
G
S
S
G
OutAux,1
OutAux,2
In Out
VRF, Q
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
2
bandwidth, to probe the device capacity. The rest of this paper is organized as follows. In Section II,
we present a general, unit-length MZM design and its optical and electrical characteristics. Section III optimizes the IQ modulator design. We examine power penalties induced by a SiP IQ modulator in a QAM transmission link; using the optical and electrical characteristics extracted in Section II. We identify the optimal combination of length and bias voltage of the phase shifter, that is, the combination that gives the minimum total modulator-induced power penalty. We found, via simulation, a constraint on IQ branch separation to avoid RF crosstalk between the branches. In section IV, our experiment setup and measured results are presented. Finally, Section V concludes this paper.
II. DESIGN AND CHARACTERISTICS OF TW-MZM Figure 1 presents a schematic of the SiP IQ modulator that
consists of a pair of MZMs nested in an MZ interferometer. Each MZM uses a phase shifter employing lateral p-n junction and a traveling-wave (TW) electrode for high-frequency operation. The TW-MZM is driven in a series push-pull configuration. A negative voltage is applied between the two arms of each MZM; thus, the phase shifter works in the depletion mode.
A. TW-MZM Design Figure 2a shows the cross-sectional schematic of the SiP
TW-MZM (identical for I and Q) using a CMOS-compatible process (A∗STAR IME, Singapore) on a 220-nm SOI wafer with 2 µm buried oxide. The phase shifter applies a lateral p-n junction embedded in the centre of a rib waveguide. While highly doped P++ and N++ regions are used for ohmic contacts, intermediate P+ and N+ doping levels are used to reduce the series resistance without significantly increasing optical propagation loss.
The phase shifter can be modeled as a series of lumped capacitors and resistors [7], whose values depend on the applied voltage. We conduct 2D simulation in Lumerical Mode and Device software at wavelength =1550 nm to predict the electro-optic response of the phase shifter and to calculate the circuit components. Figure 2b shows the calculated junction capacitance (Cj) and resistance (Rj). We calculate the overlap between the optical mode and the carrier distribution to predict the changes in refractive index, Dneff, and loss as functions of reverse voltage, as shown in Fig. 2c. More details can be found in [8].
As shown in Fig. 3a, the TW electrode has a coplanar stripline (CPS) configuration loaded by the p-n junction. For velocity matching between optical waves and RF driving signals, the RF effective index of the loaded CPS waveguide (nRF,l) should be as close as possible to the group index of the optical mode (no,g). We achieve this by employing a slow-wave design that intentionally adds metal capacitance to the transmission line to slow down the RF signal. We added a “T” shaped extension to the metal capacitance of the the CPS transmission line to improve the velocity matching while
maintaining a 50 Ω impedance. Detailed design procedures for slow-wave CPS transmission lines is beyond the scope of this paper and was recently reported in [7]. The bandwidth of the MZM is eventually dominated by the RF propagation loss in the p-n junction loaded CPS [8]. The TW electrode is terminated by an on-chip 50 Ω load implemented using N doped silicon.
B. RF Characteristics of Fabricated Device We had the TW-MZM design in Figs. 1 and 2 fabricated at
IME. We measured the small-signal frequency response using a 67 GHz Keysight network analyzer for two bias voltages, 0 V and -0.75 V. After removing the effects of the RF probes, cables and unloaded section of the transmission line, we extracted the RF parameters including propagation loss, effective index and transmission line input impedance.
Figure 3b presents the microwave effective index as a function of frequency. We can see that at 30 GHz it changes from 3.67 to 3.5, as the bias voltage Vb changes from zero to -0.75 V, respectively. Despite this variation, these values are
l
(a)
(b) (c)
Fig. 2. a) Cross section of lateral P-N junction waveguide with (all dimensions in µm) Wp++ = 5.2, Wp+ = 0.83, Wp = 0.37, Wn++ = 5.2, Wn+ = 0.81, Wn = 0.39, W = 0.5, h= 0.22, t1 = 2, t2 = 0.9, b) resistance and capacitance of the lateral pn junction vs. reverse bias voltage, and c) ∆neff vs. reverse bias voltage for the lateral pn junction in (a).
VIA 1
M2VIA2
BOX
SiO2
t1
W
Wp
M2
M1
M2
M1M1
M2
P++ N++P N+P+hWp+
Wnt2
Wn+
Wn++Wp++
DC
N
S G
Si Substrate
0 2 4 6 80
1.5
3
Reverse Voltage (V)10
4
7
10
∆n ef
f
pn (
dB/c
m)
×10-4
α 2 4 6 8 10
1.5
1.54
1.58
Reverse Voltage (V)
Res
ista
nce
(Ω .c
m)
2
4
Cap
acita
nce
(pF/
cm)
0
CjRjn Rjp
P+N++ NN+ P P++
Fig. 1. Schematic of the IQ modulator using SiP TW-MZMs with a series push-pull driving scheme.
Vb
Matched load
Heater (I)
p
n
n
G
S
Matched load
Heater(Q)
p
n
n
G
S
Heater
L
VRF, I
Vb
I
Q
S=700µm
G
S
S
G
OutAux,1
OutAux,2
In Out
VRF, Q
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
5
4.75 mm, with little change in the range from 4.4 mm top 5 mm. Based on this result, we choose a 4.5-mm-long phase shifter in our IQ modulator design and for fabrication. In modulation experiments presented later, we bias that modulation at Vb = -0.75 V. 3) RF crosstalk between I & Q branches
Due to the weak electro-optic effect in silicon, all-silicon TW-MZMs typically are several millimetres in length (4.5 mm in the case examined in this paper). As the I and the Q branches each require such long TW electrodes, spacing between the branches is critical. If placed too close to one another, unwanted RF modes may be created that affect the overall modulator performance. Specifically, when two CPS transmission lines are too closely packed, the signal and ground of one CPS and the ground of the other CPS form an asymmetric coplanar waveguide (CPW). This causes excess RF loss and crosstalk.
To avoid this unwanted phenomenon, the two MZMs need to be sufficiently separated. Using the RF characteristics extracted in Section II., we perform a 4-port electromagnetic simulation of two 4.5-mm-long CPS transmission lines in HFSS. We found a distance of 700 µm to be a safe design, leading to crosstalk between two adjacent CPS transmission lines that is less than -30 dB in the entire simulation frequency range up to 45 GHz.
IV. EXPERIMENT AND RESULTS Figure 6c is a photograph of the fabricated SiP IQ modulator
in the test configuration. A “GSSG” RF probe is used to the drive the modulator and a DC probe controls the bias voltage and thermal phase shifters. We use a fiber array for optical IO through surface grating couplers. Each of the TW-MZMs (in I and Q branches) is characterized individually through the auxiliary optical outputs (OutAux1, OutAux2 in Fig. 1).
We measured the small signal responses of the MZMs. The
Q branch shows slightly better smallp signal performance than the I branch due to fabrication errors. We present results for the I branch. Figure 6a presents the S11 at various bias voltage. We can see that even in the worst case (Vb = -4 V) the reflection is still well under -12.5 dB. The 3 dB modulation bandwidth (S21) is measured to be 36 GHz at Vb = -4 V (Fig. 6b) and decreases to 27 GHz for Vb = -0.75 V. The average static Vπ at Vb = -0.75 V is 7.3 V (Vπ-I = 7.8 V, Vπ-Q = 6.9 V). On-chip insertion loss of the IQ modulator is measured to be 6.8 dB. The coupling loss from the fiber array to the SiP chip is 8.5 dB.
A. Data transmission setup Following characterization, we ran data transmission
experiments using the experimental setup in Fig. 7. We use an EDFA to boost a continuous wave carrier at 1530 nm from an external cavity laser (ECL) to 22 dBm before coupling onto the chip. We use on-chip heaters to implement the 90o degree phase shift between the two branches. The TW-MZMs in the I and the Q branches are operated at their null point to maximize the modulation amplitude of the E field. We achieve this by adjusting the phase shift between the two arms of each MZM through on-chip heaters, Heater (I) and Heater (Q) in Fig. 1.
To generate 16-QAM, the driving voltages VRF, I and VRF, Q feed two uncorrelated four-level signals to the I and Q branches as shown in Fig. 1. We amplify two outputs of an 84 GSa/s, 8-bit DAC with two SHF 50 GHz, 18 dBm drivers and apply the signal to the modulator using an RF probe. In the transmitter digital signal processing (DSP) before uploading to the DAC, the QAM data is grey mapped from a pseudo random bit sequence (PRBS) of order 19. We apply a pre-emphasis filter to compensate the DAC, RF amplifier and RF probe frequency
Fig. 7. Experimental setup. PS: phase shifter; DC: direct current; GC: grating coupler: OBPF: Optical band-pass filter.
DCDSP+DAC
Laser
EDFA
EA
OBPFboost GC GC
PS
PS
DC block
CoRx
Noise loading
DSP
a) b) c) Fig. 6 Small signal electro-optic performance in the I branch of the TW-MZM employing 4.5 mm phase shifter normalized at 1.5 GHz: a) S11 parameter vs. frequency, and b) S21 electro-optic (EO) parameter vs. frequency. c) Photograph of the fabricated TW-SiP IQ modulator in the test setup.
-25
-20
-15
-10
Frequency (GHz)10 20 30 40
S11
(dB
)Vb = -4 VVb= -0.75 VVb = 0 VVb = -4 VVb= -0.75 VVb = 0 V
Frequency (GHz)
-5
10 20 30 40
~27
~22
~35
0
-10
S21
(d
B)EO
I
Q
DC
Fiber array
GS
SG
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
4
of the driving voltage (VRF), decreasing the modulator length decreases the optical loss and increases the electro-optic bandwidth, but it also increases the modulation loss (as a result of increased Vπ and lower extinction ratio of the modulator) and limits the output eye opening. On the other hand, using a longer phase shifter reduces BWE-O of the modulator and increases the total optical loss (αL). Hence, the total penalty from the modulator has to be minimized by simultaneously considering all of the penalty sources. 2) Quantifying modulator power penalty for QAM
The cumulative effect of the three impairments leads to a greatly reduced minimal separation between the constellation points. As illustrated in the upper corner of Fig. 4b, the separation “D” moves to “d”. We can use this change in separation to quantify the impact on bit error rate, neglecting impairments in the receiver.
In order to find the optimal length of the phase shifter, we define the modulator power penalty (MPP) as the ratio of the limited clear distance (d) to the ideal value (D), which is proportional to those impairments noted in Fig. 4a
(3)
where PαL, PER, PISI represent power penalty from the optical loss, modulation loss and limited BW of the modulator, respectively as presented in [9]. The output from the IQ modulator is a coherent combination of the two MZMs. As a result, the d is also affected by the impairments that limit the sub-eye opening in each MZM as shown in Fig. 4a.
We sweep the phase shifter length and examine several bias voltage choices. The RF voltage swings across 2V peak-to-peak; this choice is motivated by the maximum achievable swing by instruments available in our lab. We simulate 60 Gbaud 16-QAM and find minimal separation d numerically.
We plot the modulator penalty from (3) in Fig. 5. By examining multiple values of bias voltage, we can determine the relative importance of each penalty source contribution to MPP in (3).
In Fig (5), we observe an optimal length, Lopt, for each bias voltage. For phase shifter lengths shorter than Lopt, modulation loss is the dominant penalty; for longer lengths, the electro-optic bandwidth is limited and ISI dominates. Optimal length gets monotonically larger as the reverse voltage increases. This is because a higher bandwidth is available as reverse bias voltage increases, which compensates for the reduction in bandwidth at longer phase shifter lengths. While Lopt increases monotonically with bias voltage, the achieved MPP does not. While higher reverse voltage enables higher bandwidth, modulation loss also increases. This trade off leads to minimum MPP that decrease at first, and increase after Vb = -0.75 V.
From Fig. 5, we identify the optimal combination of the phase-shifter length and the bias voltage giving the lowest MPP. For Vb = -0.75 V, the overall minimum point is near
( )10log L ER ISIdMPP P P PD a
æ ö= - µ + +ç ÷è ø
(a) (b) Fig. 4. a) Transfer function of the SiP TW-MZM in one branch (I or Q) operating at the null point. b) Constellation diagram of output data for 16-QAM where constellation points are: black with no impairments, green with only optical loss, red with optical and modulation loss but no ISI, and blue dots are a scatter plot of output four level E field when all three sources of impairment are present. VRF = 2|Vb| = 2V, BR = 30 Gbaud, BWE-O ~ 25 GHz, Vπ ~ 6.4V, L = 4.5mm.
a > 0
VRF < Vπ
optical loss (a > 0) 0.5
0
-0.5
modulation loss (VRF < Vπ)
VRF = 4|Vb|
Nor
mal
ized
am
plitu
de o
f out
put E
fiel
d
input voltage
0 +VRF/2-VRF/2-Vπ Vπ
d
-0.5 -0.25 0 0.25 0.5
Q
-0.5
-0.25
0
0.25
0.5
I
D
eye openingBW = ∞
eye openingBW ≠∞
input 4 level signal
output 4 level E field
ISI penalty
BW≠∞
Fig. 5. Modulator power penalty as expressed in (3) for 16QAM at 60 Gbaud; minimum MPP at Vb=-0.75V and phase shifter length between 4.5 and 5 mm.
2 3 4 5 6 7 8
13
12
11
10
Vb = -0.25 V
Vb = -1.5 V
Vb = -3 V
MPP
(dB
)
Length (mm)
Vb = -0.75 V
Deleted: (3)
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
6
response. This allows us to investigate the impairment from the modulator. The pre-emphasized symbols are further pulse shaped by a raised cosine filter with roll-off filter of 0.01. RF phase shifters are used to precisely de-skew the I and Q channels. We control the optical operation point of each MZM using the DC probed heaters.
We amplify the modulated optical signal with a two-stage EDFA. We place an optical band-pass filter (OBPF after the second EDFA to reject out-of-band amplified spontaneous emission (ASE) noise. To sweep OSNR, we load the signal with ASE noise at the receiver end. The output optical signal is mixed with a 14 dBm local oscillator (LO) and fed to a 70 GHz coherent receiver (CoRx) formed by a discreet optical hybrid and 70 GHz balanced photodetectors. We use a Keysight real-time oscilloscope (RTO) with a 60-GHz analog bandwidth to sample the output of the CoRx at 160 GSa/s.
B. Experimental characterization for MMP In order to verify our design methodology and the numerical
model, we experimentally characterize the MMPs of the IQ modulator at various bias voltages and compare them with simulation results. Imperfection from optical loss (PαL) is applied by employing the measured optical loss of the modulator versus applied reverse bias voltage. To estimate the PER, Vπ at different reverse bias voltages are measured; the modulation loss is then calculated for a driving voltage of 2 Vpp. The experimental characterization for PISI requires a more complex procedure that is described as follows.
To quantify the penalty from the bandwidth limitation of the modulator (PISI), a training-aided channel estimation method is employed. We load the DAC with a Nyquist-16QAM training sequence with a roll-off factor of 0.01, where the frequency responses of the DAC, the RTO and other RF components (such as drivers and RF cables) have been digitally pre-compensated, ensuing a flat spectrum in the driving signal of the modulator. The ISI penalty of the modulator is then estimated following the signal-processing flow as shown in Fig. 8a. The data samples are first synchronized and framed according to the training sequence. After the training-aided frequency offset compensation (TA-FOC) and the training-aided carrier phase recovery (TA-CPR), the channel response is estimated by the minimum mean square error (MMSE) calculation and is averaged over 100 frames to minimize the impact from noise and numerical errors. Since no adaptive equalizer is applied and the frequency responses of other RF components have been de-embedded through the pre-compensation of the training sequence, the obtained MMSE channel response represents the response of the modulator alone.
Using the estimated channel response, we run a post-emulation to visualize the ISI by the convolution of the averaged channel response with an ideal 16QAM signal. The post-emulated 30-Gbaud 16QAM constellations with ISI distortion are shown in Fig. 8b for different reverse bias voltages. Note that the modulator stays in the linear regime for the given voltage swing, thus nonlinear modulation distortion is
negligible in this case. Due to the relatively narrow analog bandwidth (< 20 GHz at -3 dB) of the DAC, the accuracy of the ISI emulation degrades drastically at a higher frequency. Therefore, we have only characterized PISI for up to 30-Gbaud for the purpose of verification.
Finally, the total MPPs of the modulator under test at different bias voltages are calculated using Eq. (3) for 20 Gbaud and 30 Gbaud. The results are shown in Fig 8c. For the lower baud rate (i.e., 20 Gbaud), the electro-optic bandwidth of the modulator is large enough even at zero bias (~22 GHz in Fig. 6.c) with little ISI penalty. In this case, the total MPP increases as the reverse bias voltage due to the higher Vπ and thus higher modulation loss. However, for the higher baud rate (i.e., 30 Gbaud), PISI becomes dominant over the modulation loss (PαL) at zero bias. Increasing the reverse bias to 0.75 V achieves the minimum MPP thanks to the best trade-off between the bandwidth (~27 GHz in Fig. 6c) and the modulation loss as discussed in the previous section.
As shown in Fig. 6.c, the simulation results match very well with the measured MPPs. Although the MPP can only be accurately measured at a relatively low baud rate due to the narrow bandwidth of the DAC used in our experiment, the same physics should work at higher baud rates. Hence, we expect that the simulation results shown in in Fig.5. also present the optimum modulator design for 60-Gbaud 16 QAM modulation. In continue we will present data transmission results to evaluate the system-level performance of the designed modulator.
C. BER performance Experiments are run at 60 Gbaud (240 Gb/s) for three
different bias voltages. The BER performance of 16-QAM with an RF voltage swings across 2V peak-to-peak is shown in Fig. 9a. Two forward error correction (FEC) thresholds are indicated: a 7% overhead FEC threshold at BER = 3.8e-3, and a 20% overhead FEC threshold at BER = 2.4e-2.
In Fig. 9a, we observe that increasing the bias voltage from Vb = -0.25V (where the diodes are slightly forward biased but
(a)
(b) (c) Fig. 8. a) Flow chart of training aided channel estimation. b) post emulated constellation diagrams of received data for 30 Gb 16QAM, when reverse bias voltage changes from 0 V (where the diodes are slightly forward biased) to -3 V. c) simulation and measurement results for MPP versus reverse bias voltage for two different baud rates. (label and unit for (c))
Sync
.
TA-F
OC
TA-C
PR
Ch. E
sti.
RTO
PISI
calc
u.
ISI
Post
-em
ulat
ion
0 -0.5 -1 -1.5 -2 -2.5 -3
13
12
11
9
Bias voltage (V)
10
14 30 Gb (Measurement)30 Gb (Simulation)20 Gb (Measurement)20 Gb (Simulation)
Vb = 0 V Vb = - 0.75 V Vb = - 1.5 V Vb = - 3 V
-0.5 0.50I
Q 0
-0.5
0.5
Commented [LJ3]: Fig.6 b?
Commented [LJ4]: Which figure?
Commented [LJ5]: Figure 8?
Formatted: Font color: Red
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
7
there is still no current passing through diodes) to -0.75 V significantly improves the BER performance of the modulator. In this case, the bandwidth achieved from stronger reverse bias of the phase shifter compensates the modulation loss penalty due to higher Vπ at Vb = -0.75 V. The MPP simulation predicted this behaviour, as shown in Fig. 5. Lower MPP (Fig. 5) indicates a larger clear distance between the constellation points, leading to better BER performance.
Pushing the phase shifter to stronger reverse bias does not improve the BER performance due to a lower efficiency. For example, Vb = -3V shows the worst BER of all the three curves in Fig. 9a despite the highest bandwidth.
We further investigate the achievable baud rate of the modulator. Figure 9b shows the BER at 40, 60 and 70 Gbaud when the reverse bias voltage is Vb = -0.75 V and the RF voltage swings across 2V peak-to-peak. A large margin is available at 40 Gbaud. The BER performance is below the 20% FEC threshold up to 70 Gbaud (net bit rate of 224 Gb/s). A 400 Gb/s dual-polarization transmitter can be achieved by integrating on the same chip two IQ modulators, a polarization rotator (from TE to TM) and a polarization combiner.
V. CONCLUSION We have experimentally demonstrated an integrated IQ
modulator based on a SiP foundry process. Our results show that all-silicon modulators have great potential for high-baud-rate coherent optical communications systems. We demonstrated that the performance of the SiP IQ modulator can be effectively optimized by minimizing the modulation-induced power penalty in an optical transmission link. The design and operation of the TW-MZMs were optimized for 16-QAM at 60 Gbaud. The experimental results verified our theoretical prediction. This system-orientated optimization methodology can also be used in other contexts with various modulation formats and rates. Using the SiP IQ modulator, we have achieved single-polarization 16-QAM at 60 Gbaud (BER
at the 7% FEC threshold) and at 70 Gbaud (BER below the 20% FEC threshold), giving a net data rate of 224 Gb/s and 233 Gb/s, respectively. This indicates a promising solution for 400+ Gb/s dual-polarization transmitter. Note that these results were achieved without pre-compensating for the impairments from the modulator. More sophisticated digital or optical pre-emphasise could be done to overcome the limitations of the modulator and further push the bit rate.
ACKNOWLEDGMENT: The authors thank Dr. Zhuhong Zhang and his team with Huawei Canada for many useful discussions. This work was supported by Huawei Canada and NSERC (CRDPJ 486716-15).
APPENDIX:
A. RF Loss in TW-SIP Series Push-Pull MZM By assuming a quasi-TEM propagation, as it is shown in
[10], we use Telegrapher’s RLGC model to investigate the slow wave propagation of microwave along CPS transmission line in SOI process. The RLGC model of p-n junction loaded CPS transmission line is shown in Fig.10. The two p-n junctions can be modeled by a series combination of the junction capacitance and semiconductor resistance, Rjp, Rjn, Cj (see Fig.2. b). Parallel equivalent of Rjp, Rjn, Cj, beside RLGC model of an unloaded CPS transmission line can be used to present the RLGC model of a p-n junction-loaded CPS transmission line in SOI as it is illustrated in Fig. 10, where Rpn=Rjp+Rjn and:
Ctpn=Cpn
1+Rpn2 .cpn
2 .ω2#at ω1/(RpnCpn)
≈ Cpn (4)
Gtpn=Rpn
.cpn2 .ω2
1+Rpn2 .cpn
2 .ω2#at ω1/(RpnCpn)
≈ Rpn .cpn
2 .ω2 (5)
Where Ctpn and Gtpn are the paralle equivalent of capacitance and conductance per unit length of p-n junction in the Telegrapher’s RLGC model. By applying the low loss approximation, the p-n junction loaded characteristic impedance, Zol and propagation constant, γl can be expressed
(a) (b)
Fig. 9. a) BER performance for different reverse bias voltages at 60 Gbaud. b) BER performance for 40, 60, and 70 Gbaud 16-QAM when the 4.5mm length phase-shifter is biased at Vb = -0.75V.
15 20 25 30 35OSNR (dB)
0
BER
-110
-210
-310
-22.4×10
-33.8×10
16-QAM Vb = -0.75
-410
15 20 25 30 35OSNR (dB)
0
BER
-110
-210
-310
-22.4×10
-33.8×10
16-QAM BR=60 Gbaud
BR = 70 Gbaud
BR = 40 GbaudBR = 60 Gbaud
Vb = -3 VVb = -0.75 VVb = -0.25 V
Commented [WS6]: This seems not the orginal literature. What did they cite?
Commented [WS7]: What does tpn means?
Commented [WS8]: Seems not true
Commented [Office9R8]: According to Fig.2a and i.e. for Vb=-1V, 1/(RpnCpn) »1/(1.5 W ´ 2.8 pF) » 240 GHz so ω1/(RpnCpn)is a valid approximation.
Commented [WS10]: This seems not the orginal literature. What did they cite?
Commented [Office11R10]: It’s a basic concept in microwave and RF design. I think it is a good reference as it shows the application of TL design in SOI with p-n junction. Other ways we have to refer to some books in TL design which is not helpful for the people in optic.
Commented [WS12]: Reasonable?
Commented [Office13R12]: In TL design the “low loss” approximation s considering that Rcps and Gcps in Fig. 10 are small values which is reasonable.
(a)
(b)
(c)
(d)
(e)
(f)
Fig. 1: (a) Schematic of the silicon IQ modulator. (b) Cross section of the silicon MZM. (c) Simulated MPP as a function of the phase-shifter length L for 16-QAM at 60 Gbaud. (d) Frequency response at various bias for L = 4.5mm. (e) Post-emulated constellation diagrams of30-GBaud 16-QAM at various bias. (f) BER performance at 60-Gbaud without pre-compensation.
3. High-baud-rate transmission
Using the silicon IQ modulator, we performed single-carrier dual-polarization QAM transmissions for up to720 Gb/s. The experimental setup is the same as that in [9], plus a physical polarization division multiplexingemulator for dual-polarization operation. Pre-emphasis was applied to compensate the limited transmitter band-width due to the digital-to-analog converter (∼17 GHz) and the silicon modulator (∼27 GHz at Vb = 0.75V). Theuse of an optical filter can reduce the burden of the digital pre-emphasis and lower quantization noise. However,independently designed digital and optical filters cannot achieve the best trade-off of DAC quantization noiseand SNR degradation across the signal frequency band. We have developed a joint optimization procedure of op-tical and digital pre-emphasis filters [9], illustrated in Fig. 2(a), that can effectively improve the signal quality.Figure 2(b) shows constellations of 84-Gbaud 16-QAM and 70-Gbaud 32-QAM signals. B2B BER performancevs. received optical signal-to-noise ratio (OSNR) is shown in Fig. 2(c). Performance for propagation in a standardsingle mode fiber is shown in Fig. 2(d). For 400 Gb/s transmission, we achieved 60-Gbaud dual-polarization (DP)-16QAM reaching a distance of 1,520 km. We meet the BER threshold for 20% forward error correction overhead,at 72 Gbaud DP-32QAM (720 Gb/s) transmitted over 160 km and 84 Gbaud DP-16QAM (672 Gb/s) transmittedover 720 km, achieving net rates of 600 Gb/s and 576 Gb/s, respectively. To the best of our knowledge, theseresults are the highest single-carrier bit rate and the highest spectral efficiency (up to 8.3 bits/Hz) achieved to dateusing a CMOS-compatible technology.
DSP DAC RFamp
SiPIQM OSP
1 1
ripple smooth
effe effff effH Hf fH f H f
1 1 1smooth
1fiteff DAC OSA effH f H f H f H f o
1 1
11
11 1
, ,
, ,
DPE OPE
effeff
efffe efff
H fH f
H fH f
H ff H fH
D E
ED
D E
§ ·¨ ¸©
§ ·¨ ¸©¹ ¹
Measure
Fit & smooth
Decompose
Weight
DSP DAC RFamp
SiPIQM OSA
1DACH f DACH f OSAH f
‐40 ‐20 0 20 40
GHz
E'
10dB/div
1effH f
1 ,DPEH f D
1 ,OPEH f E
1effH f
D'
'
Xk+1
Xk Yk
ek
Systemunder test
LearningController
D
Fig. 1. Block diagram of Iterative learning control method.
D, as detailed in section 2.One of the first ILC methods was formulated by Arimoto in control theory [16,17]. He defined
the ILC learning controller block (ILC law) as a recursive function called F (·)Xk+1 = F
Xk, ek
(1)
Several functions can be used for F (·), depending on the application [16]. In [14] the linear,gain-based and Newton ILC laws were investigated for RF power amplifier linearization. Inthis work, we employ a modified version of the gain-based law, G-ILC, to compensate patterndependent distortion introduced by a SiP IQ modulator. Let the input signal at iteration k begiven by the n-dimensional column vector Xk 2 <n1 where
Xk =h
xk (1) · · · xk (n 1) xk (n)iT
(2)
where xk(i) is the predistorted version of the ith symbol in the training sequence D. The errorcolumn vector ek is the dierence of Y k and D. We define k 2 <nn as a learning gain matrix
k = diagh
G (xk (1)) · · · G (xk (n))iT
(3)
where the function G is an appropriate gain vector for the transmit signal, Xk . We choose thesample-by-sample gain function
G (xk (i)) =yk (i)D (i)
2(4)
With these definitions, (1) can be expressed as
Xk+1 = FXk, ek
= Xk + ↵k
1ek (5)
where ↵ is the step size. We adopt a variable step size to enhance the convergence speed, changingonce or twice during adaptation depending on the amount of distortion. For the case of theidentity matrix, i.e., k = In 2 <nn, we have a linear ILC, which is a special case of G-ILC,given by
Xk+1 = Xk + ↵ek (6)We tested and compared the choice of k = In vs. nonlinear k defined in Eqs. (4) and (5).We found convergence speed was greatly enhanced for the nonlinear version of k . In the caseof higher modulation order, i.e., 256QAM, our choice of G-ILC (nonlinear k ) has sucientdegrees of freedom to combat nonlinear eects, while exhibiting stable convergence. Linear k(k = In) could not achieve minimal error in any reasonable time. The predistortion adaptationcan be initialized with a linear equalizer such as a minimum mean square error (MMSE) equalizer.
(d)
Fig. 3. Received constellations at optical power of 2 dBm. (a-d) 20 Gbaud/256QAM: (a)linear predistortion at TX, (b) linear predistortion at TX and linear postcompensation atRX, (c) G-ILC predistortion at TX, and (d) G-ILC predistortion at TX and linear post-compensation at RX. Black-red-yellow is the transition from lowest to highest density ofsamples.
receiver side 500-tap MMSE filter for post-compensation. The MMSE postcompensation filter isupdated at each iteration to help the G-ILC convergence. At 60 Gbaud, we also use an opticalpre-emphasis stage to provide robust convergence.
5.1. Electrical compensation alone
At moderate baud rates, 20 Gbaud and 40 Gbaud, electrical compensation alone was sucientto achieve convergence in the G-ILC. We present constellation diagrams at 2 dBm receivedpower for various compensation strategies. First we consider no G-ILC, but instead use typicallinear compensation techniques. We consider linear MMSE predistortion at the Transmitter side(a), linear postcompensation at the receiver side, in addition to the linear MMSE predistortion(b). Next, we apply G-ILC at the transmitter without any postcompensation (c) and with linearpostcompensation (d). The case of 20 Gbaud 256QAM is presented in Fig. 3, and 40 Gbaud128QAM is presented in Fig. 4. Note that the standard MMSE equalizer for predistortion in (a)is also the initial input signal, X1, for G-ILC (c). Coloring is used in the constellations to showdensity in the constellation, with black to red to yellow showing the transition from lowest tohighest density of samples.
(a) (b)
(c) (d)
(e)
(f)
Fig. 2: (a) Pre-emphasis frequency responses of optical (H−1OPE ), digital (H−1
DPE ), total with (H−1e f f ) and without smoothing(H−1
e f f ); α and β are
the weighting factors for a maximal excursion of H−1e f f . (b) Constellations of 84-Gbaud 16-QAM and 70-Gbaud 32-QAM. (c) BER v.s. OSNR
curves for 60/75/84 Gbaud 16QAM and 60/72 Gbaud 32QAM. (d) BER for 60/75/84 Gbaud 16QAM and 60/72 Gbaud 32QAM for swepttransmission distance. (e) Block diagram of ILC method. (f) 20 Gbaud/256QAM with the ILC predistortion.
Silicon modulators also suffer from nonlinear pattern-dependent behavior that causes signal distortion in thepresence of very high order modulations or large driving voltages. To address this issue, we recently demonstrated[11] a predistortion method based on the iterative learning control (ILC) technique using quasi-real-time adaptation
Tu2H.1.pdf OFC 2019 © OSA 2019
with hardware-in-the-loop; see Fig. 2(e)). Applying this method with the silicon IQ modulator, we achieved 20-Gbaud 256-QAM (Fig. 2(f)) and 40-Gbaud 128-QAM with an power-sensitivity improvement of up to 5 dB.
4. Frequency comb for super-channel transmission
Modulator-based optical frequency combs provide a flexible solution for generation of optical carriers, especiallyfor applications such as flexible-grid WDM or super-channel, where a tunable frequency spacing is preferred.Using the same fabrication process as the IQ modulator, we developed a dual-drive silicon MZM (Fig. 3(a)) foron-chip frequency comb generation [12]. As shown in Fig. 3(b), five comb lines were achieved with 20 GHzspacing. The worst case tone-to-noise ratio is greater than 40 dB after one stage of optical amplification. Usingthis on-chip frequency comb generator, we achieved 5 × 16-GBd DP-32QAM for a 800 Gb/s Nyquist-WDMsuper-channel meeting a 20% FEC overhead. We also achieved 5×20 Gbaud DP-16QAM with a BER well belowthe 7% FEC threshold. This super-channel without guard band gives a seamless spectrum, as shown in Fig. 3(c),indicating the good stability of the comb. It is challenging to achieve a larger number of comb lines due to therelatively high Vπ (∼ 3.5 V). However, for an aggregate Tb/s data rate, a few sub-carriers should be sufficient fora super-channel using high-speed modulators.
GSSG
DC
(a) (b) (c)
LIN et al.: FREQUENCY COMB GENERATION USING A CMOS COMPATIBLE SiP DD-MZM FOR FLEXIBLE NETWORKS 1497
Fig. 3. Experimental setup for SiP on-chip comb generation and Nyquist-WDM back-to-back transmission.
before QAM data modulation. The Waveshaper is programmedwith a notched comb-like window for enhanced OSNR [19]and flatness. The total power of the 5-line comb is kept at14 dBm. We test back-to-back transmission of 16/20 Gbaud16QAM and 16 Gbaud 32QAM.
The transmitter digital signal processing (DSP) flow chart isshown in the insert of Fig. 3. A pseudo random bit sequence(PRBS) of order 19 is Gray coded to QAM symbols. TheIQ symbols are then pulse shaped using a raised cosine finiteimpulse response (FIR) filter with a roll-off factor of 0.01. Thepulse shaped samples are then resampled to match the DACsampling rate, generating 16/20 Gbaud signal. These samplesare clipped, quantized, and loaded to the 8-bit 64 GSa/s digital-to-analog converter (DAC). The data modulation is realizedby driving a SHF optical QAM modulator with the NyquistIQ signals. After data modulation, the Nyquist-WDM signalis amplified by EDFA3, followed by a polarization divisionmultiplexing emulator (PDME), i.e., a pair of polarizationbeam splitter/combiners (PBS/C) and a delay line.
At the receiver end, the channel of interest is selected bya tunable optical filter. The selected channel is amplified byan EDFA and a 1 nm filter is used to reject out-of-band ASEnoise. The signal is coherently detected using an integratedoptical coherent receiver with 20 GHz electrical bandwidth,where the −3 dBm signal is mixed with the local oscillator(LO). The LO is implemented by a 14 dBm ECL laser,whose linewidth is ∼5 kHz. The four outputs of the integratedcoherent receiver (CoRx) are captured by a 4 × 33 GHzreal time sampling oscilloscope (RTO) at a sampling rate of80 GSa/s. The digitized samples are then processed usingoffline DSP.
The DSP stack is shown as an insert at receiver sidein Fig. 3. We apply a 10th order super Gaussian low passfilter (LPF) to all four captured signals and then resam-ple to 2-samples per symbol. We use a real valued 4 × 4multi-input-multi-output (MIMO) equalizer for polarizationde-multiplexing and residual inter-symbol-interference (ISI)compensation. The frequency offset compensation (FOC) usesa FFT based method. The laser phase drifting is tracked usingblind phase search with 32 test angles. After carrier phaserecovery (CPR), a T-spaced 256-tap decision directed-leastmean square (DD-LMS) filter is applied to further improve thesignal quality. Finally, the recovered symbols are demodulatedfor BER calculation.
IV. RESULTS AND DISCUSSION
Figure 4 shows the spectrum of the 5-line comb with20 GHz spacing that we generated, as measured after
Fig. 4. Spectrum of generated 5-line comb with 20GHz spacing.
the OBPF. For comb generation, the diode reverse bias volt-age of each silicon phase shifter is set to 0 V. This biasvalue trades-off modulation efficiency (bias-dependent Vπ )and bandwidth. For a depletion-mode silicon phase shifter,the modulation efficiency varies with bias voltage due to thenonlinear plasma effect. With a smaller bias voltage, the sili-con phase shifter exhibits better modulator efficiency leadingto a lower Vπ , but by sacrificing modulator bandwidth. Forcomb generation, the conversion efficiency is limited more byVπ than by bandwidth. Across the five lines, the worst tone-to-noise ratio (TNR) measured (at 100 MHz resolution) is around40 dB, corresponding to ∼19 dB OSNR with a 0.1 nm noisereference bandwidth. However, the seed tone is much strongerand cannot be further suppressed. This is caused by the finiteextinction ratio of ∼14 dB of the SiP MZM, and should beimproved through design optimization.
In the Nyquist-WDM test, as mentioned previously,the comb is further shaped using a programmable Waveshaper.This enables more consistent BER performance across all sub-channels. The Waveshaper is programmed as a comb filter withan additional 6 dB notch attenuation at the center frequencythat improves the flatness and also boosts the OSNR. Figure 5ashows the reshaped comb lines. The Waveshaper was usedfor experimental convenience, but a comb filter and spectrumbalancing filter could be realized in SiP [14], [20].
We set three benchmarks on the Nyquist-WDM systemto evaluate the quality of the on-chip comb. We first mod-ulate the 20 GHz spaced 5-line comb with 5 × 16 GbaudNyquist-16QAM. The modulated signal spectrum is shownin Fig. 5b, where we reserve a 4 GHz guard band between sub-channels. Then we expand the modulation constellation to fur-ther stress the OSNR of the comb, testing with a 5×16 GbaudNyquist-32QAM signal. In the final trial, we modulate thecomb with a 5 × 20 Gbaud 16QAM signal, creating a super-channel without guard band. This extreme case demonstrates
1498 IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 30, NO. 17, SEPTEMBER 1, 2018
Fig. 5. Spectrums of (a) 20 GHz spaced 5-line comb after reshaping;(b) 5 × 16 Gbaud 16QAM; and (c) 5 × 20 Gbaud 16QAM.
Fig. 6. BER results of Nyquist-WDM signals of 5 × 16Gbaud 16/32QAMand 5 × 20Gbaud 16QAM.
the good stability of the comb spacing, for which a seamlessspectrum is shown in Fig 5c.
The BER test results are summarized in Fig. 6. We usepre-FEC BER thresholds to evaluate the system performance,assuming 3.8e-3 for 7% overhead hard decision FEC and2e-2 for a 20% overhead soft decision FEC. As indicatedby blue circles, the BERs for 16QAM at 5 × 16 Gbaudare well below 1e-3, achieving a net rate of 598 Gb/s fora 7% FEC overhead and a SE of 5.98 bit/s/Hz. For the32QAM case, indicated by yellow diamonds, the BERs areall below the 20% FEC threshold, but the second and fourthchannels show slightly higher BERs than the low overheadFEC threshold. The achieved net data rate is 667 Gb/s withSE of 6.67 bit/s/Hz. The BER of the super-channel (withoutguard band) is indicated by red squares in the same figure; theBERs are all below 3.8e-3. The results show negligible BERpenalties from crosstalk induced by carrier frequency drift,demonstrating the stable frequency spacing between lockedcomb lines. With super-channel 16QAM, assuming a 7%(20%) FEC overhead we achieve a net rate of ∼747 Gb/s(667 Gb/s) with SE of 7.47 bit/s/Hz (6.67 bit/s/Hz).
V. CONCLUSION
Using an all-silicon DD-MZM, we experimentally achieveda 5-line frequency comb with 20 GHz spacing. The evaluationin the context of a Nyquist WDM system shows that the combcan well support 32QAM and no-guard-interval 16QAM.
An 800 Gb/s super-channel (with a net rate of 747 Gb/s and anSE of 7.47 Gb/s/Hz) was demonstrated, indicating a promisingsolution for ultra-high-capacity optical transmitters using aCMOS-compatible process.
ACKNOWLEDGEMENT
The authors thank Dr. Zhuhong Zhang and his team withHuawei Canada for many useful discussions. They also thankNelson Landry for his technician support, and CMC Microsys-tems for the MPW service.
REFERENCES
[1] V. Lal et al., “Extended C-band tunable multi-channel InP-basedcoherent transmitter PICs,” J. Lightw. Technol., vol. 35, no. 7,pp. 1320–1327, Apr. 1, 2017.
[2] J. Pfeifle et al., “Coherent terabit communications with microresonatorKerr frequency combs,” Nature Photon., vol. 8, no. 5, pp. 375–380,2014.
[3] P. Marin-Palomo et al., “Microresonator-based solitons for mas-sively parallel coherent optical communications,” Nature, vol. 546,pp. 274–279, Jun. 2017.
[4] V. Corral, R. Guzmán, C. Gordón, X. J. M. Leijtens, and G. Carpintero,“Optical frequency comb generator based on a monolithically integratedpassive mode-locked ring laser with a Mach–Zehnder interferometer,”Opt. Lett., vol. 41, no. 9, pp. 1937–1940, 2016.
[5] R. Rios-Mülleret et al., “1-terabit/s net data-rate transceiver based onsingle-carrier Nyquist-shaped 124 GBaud PDM-32QAM,” in Proc. OFC,Los Angeles, CA, USA, Mar. 2015, pp. 1–3.
[6] D. S. Millaret, “Design of a 1 Tb/s superchannel coherent receiver,”J. Lightw. Technol., vol. 34, no. 6, pp. 1453–1463, Mar. 15, 2016.
[7] A. J. Metcalf, V. Torres-Company, D. E. Leaird, and A. M. Weiner,“High-power broadly tunable electrooptic frequency comb generator,”IEEE J. Sel. Topics Quantum Electron., vol. 19, no. 6, pp. 231–236,Nov. 2013.
[8] N. Dupuis et al., “InP-based comb generator for optical OFDM,”J. Lightw. Technol., vol. 30, no. 4, pp. 466–472, Feb. 15, 2012.
[9] N. Yokota, K. Abe, S. Mieda, and H. Yasaka, “Harmonic superpositionfor tailored optical frequency comb generation by a Mach–Zehndermodulator,” Opt. Lett., vol. 41, no. 5, pp. 1026–1029, 2016.
[10] C. Weimann et al., “Silicon-organic hybrid (SOH) frequency combsources for terabit/s data transmission,” Opt. Express, vol. 22, no. 3,pp. 3629–3637, 2014.
[11] P. Dong et al., “Silicon in-phase/quadrature modulator with on-chipoptical equalizer,” J. Lightw. Technol., vol. 33, no. 6, pp. 1191–1196,Mar. 15, 2015.
[12] K. P. Nagarjun et al., “Generation of tunable, high repetition rate opticalfrequency combs using on-chip silicon modulators,” Opt. Express,vol. 26, no. 8, pp. 10744–10753, 2018.
[13] X. Xiao, M. Li, L. Wang, D. Chen, Q. Yang, and S. Yu, “High speedsilicon photonic modulators,” in Proc. OFC, Los Angeles, CA, USA,Mar. 2017, pp. 1–3.
[14] X. Wu and H. K. Tsang, “Flat-top frequency comb generation withsilicon microring modulator and filter,” in Proc. CLEO, San Jose, CA,USA, May 2017, pp. 1–2.
[15] I. Demirtzioglou et al., “Frequency comb generation in a silicon ringresonator modulator,” Opt. Express, vol. 26, no. 2, pp. 790–796, 2018.
[16] Y. Xu et al., “Integrated flexible-grid WDM transmitter using an opticalfrequency comb in microring modulators,” Opt. Lett., vol. 43, no. 7,pp. 1554–1557, 2018.
[17] H. Bahrami, H. Sepehrian, C. S. Park, L. A. Rusch, and W. Shi, “Time-domain large-signal modeling of traveling-wave modulators on SOI,”J. Lightw. Technol., vol. 34, no. 11, pp. 2812–2823, Jun. 1, 2016.
[18] J. Lin, H. Sepehrian, L. A. Rusch, and W. Shi, “Flexible on-chipfrequency comb generation using a SOI dual-drive MZM,” in Proc. OI,Santa Fe, NM, USA, Jun. 2017, pp. 27–28.
[19] J. Lin et al., “Demonstration and evaluation of an optimized RFS combfor terabit flexible optical networks,” IEEE/OSA J. Opt. Commun. Netw.,vol. 9, no. 9, pp. 739–746, Sep. 2017.
[20] Z. Zou, L. Zhou, X. Li, and J. Chen, “Channel-spacing tunable siliconcomb filter using two linearly chirped Bragg gratings,” Opt. Express,vol. 22, no. 16, pp. 19513–19522, 2014.
1498 IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 30, NO. 17, SEPTEMBER 1, 2018
Fig. 5. Spectrums of (a) 20 GHz spaced 5-line comb after reshaping;(b) 5 × 16 Gbaud 16QAM; and (c) 5 × 20 Gbaud 16QAM.
Fig. 6. BER results of Nyquist-WDM signals of 5 × 16Gbaud 16/32QAMand 5 × 20Gbaud 16QAM.
the good stability of the comb spacing, for which a seamlessspectrum is shown in Fig 5c.
The BER test results are summarized in Fig. 6. We usepre-FEC BER thresholds to evaluate the system performance,assuming 3.8e-3 for 7% overhead hard decision FEC and2e-2 for a 20% overhead soft decision FEC. As indicatedby blue circles, the BERs for 16QAM at 5 × 16 Gbaudare well below 1e-3, achieving a net rate of 598 Gb/s fora 7% FEC overhead and a SE of 5.98 bit/s/Hz. For the32QAM case, indicated by yellow diamonds, the BERs areall below the 20% FEC threshold, but the second and fourthchannels show slightly higher BERs than the low overheadFEC threshold. The achieved net data rate is 667 Gb/s withSE of 6.67 bit/s/Hz. The BER of the super-channel (withoutguard band) is indicated by red squares in the same figure; theBERs are all below 3.8e-3. The results show negligible BERpenalties from crosstalk induced by carrier frequency drift,demonstrating the stable frequency spacing between lockedcomb lines. With super-channel 16QAM, assuming a 7%(20%) FEC overhead we achieve a net rate of ∼747 Gb/s(667 Gb/s) with SE of 7.47 bit/s/Hz (6.67 bit/s/Hz).
V. CONCLUSION
Using an all-silicon DD-MZM, we experimentally achieveda 5-line frequency comb with 20 GHz spacing. The evaluationin the context of a Nyquist WDM system shows that the combcan well support 32QAM and no-guard-interval 16QAM.
An 800 Gb/s super-channel (with a net rate of 747 Gb/s and anSE of 7.47 Gb/s/Hz) was demonstrated, indicating a promisingsolution for ultra-high-capacity optical transmitters using aCMOS-compatible process.
ACKNOWLEDGEMENT
The authors thank Dr. Zhuhong Zhang and his team withHuawei Canada for many useful discussions. They also thankNelson Landry for his technician support, and CMC Microsys-tems for the MPW service.
REFERENCES
[1] V. Lal et al., “Extended C-band tunable multi-channel InP-basedcoherent transmitter PICs,” J. Lightw. Technol., vol. 35, no. 7,pp. 1320–1327, Apr. 1, 2017.
[2] J. Pfeifle et al., “Coherent terabit communications with microresonatorKerr frequency combs,” Nature Photon., vol. 8, no. 5, pp. 375–380,2014.
[3] P. Marin-Palomo et al., “Microresonator-based solitons for mas-sively parallel coherent optical communications,” Nature, vol. 546,pp. 274–279, Jun. 2017.
[4] V. Corral, R. Guzmán, C. Gordón, X. J. M. Leijtens, and G. Carpintero,“Optical frequency comb generator based on a monolithically integratedpassive mode-locked ring laser with a Mach–Zehnder interferometer,”Opt. Lett., vol. 41, no. 9, pp. 1937–1940, 2016.
[5] R. Rios-Mülleret et al., “1-terabit/s net data-rate transceiver based onsingle-carrier Nyquist-shaped 124 GBaud PDM-32QAM,” in Proc. OFC,Los Angeles, CA, USA, Mar. 2015, pp. 1–3.
[6] D. S. Millaret, “Design of a 1 Tb/s superchannel coherent receiver,”J. Lightw. Technol., vol. 34, no. 6, pp. 1453–1463, Mar. 15, 2016.
[7] A. J. Metcalf, V. Torres-Company, D. E. Leaird, and A. M. Weiner,“High-power broadly tunable electrooptic frequency comb generator,”IEEE J. Sel. Topics Quantum Electron., vol. 19, no. 6, pp. 231–236,Nov. 2013.
[8] N. Dupuis et al., “InP-based comb generator for optical OFDM,”J. Lightw. Technol., vol. 30, no. 4, pp. 466–472, Feb. 15, 2012.
[9] N. Yokota, K. Abe, S. Mieda, and H. Yasaka, “Harmonic superpositionfor tailored optical frequency comb generation by a Mach–Zehndermodulator,” Opt. Lett., vol. 41, no. 5, pp. 1026–1029, 2016.
[10] C. Weimann et al., “Silicon-organic hybrid (SOH) frequency combsources for terabit/s data transmission,” Opt. Express, vol. 22, no. 3,pp. 3629–3637, 2014.
[11] P. Dong et al., “Silicon in-phase/quadrature modulator with on-chipoptical equalizer,” J. Lightw. Technol., vol. 33, no. 6, pp. 1191–1196,Mar. 15, 2015.
[12] K. P. Nagarjun et al., “Generation of tunable, high repetition rate opticalfrequency combs using on-chip silicon modulators,” Opt. Express,vol. 26, no. 8, pp. 10744–10753, 2018.
[13] X. Xiao, M. Li, L. Wang, D. Chen, Q. Yang, and S. Yu, “High speedsilicon photonic modulators,” in Proc. OFC, Los Angeles, CA, USA,Mar. 2017, pp. 1–3.
[14] X. Wu and H. K. Tsang, “Flat-top frequency comb generation withsilicon microring modulator and filter,” in Proc. CLEO, San Jose, CA,USA, May 2017, pp. 1–2.
[15] I. Demirtzioglou et al., “Frequency comb generation in a silicon ringresonator modulator,” Opt. Express, vol. 26, no. 2, pp. 790–796, 2018.
[16] Y. Xu et al., “Integrated flexible-grid WDM transmitter using an opticalfrequency comb in microring modulators,” Opt. Lett., vol. 43, no. 7,pp. 1554–1557, 2018.
[17] H. Bahrami, H. Sepehrian, C. S. Park, L. A. Rusch, and W. Shi, “Time-domain large-signal modeling of traveling-wave modulators on SOI,”J. Lightw. Technol., vol. 34, no. 11, pp. 2812–2823, Jun. 1, 2016.
[18] J. Lin, H. Sepehrian, L. A. Rusch, and W. Shi, “Flexible on-chipfrequency comb generation using a SOI dual-drive MZM,” in Proc. OI,Santa Fe, NM, USA, Jun. 2017, pp. 27–28.
[19] J. Lin et al., “Demonstration and evaluation of an optimized RFS combfor terabit flexible optical networks,” IEEE/OSA J. Opt. Commun. Netw.,vol. 9, no. 9, pp. 739–746, Sep. 2017.
[20] Z. Zou, L. Zhou, X. Li, and J. Chen, “Channel-spacing tunable siliconcomb filter using two linearly chirped Bragg gratings,” Opt. Express,vol. 22, no. 16, pp. 19513–19522, 2014.
PS-1
PS-2
100 GHz
40 dB
Fig. 3: (a) Micro-image of a dual-drive silicon MZM with 4.5-mm-long phase shifters. (b) Spectrum of a 20-GHz optical frequency comb afterone-stage EDFA amplification (100 MHz resolution). (c) A seamless 800-Gb/s super-channel in 5×20 Gbaud 16-QAM Nyquist-WDM.
5. Conclusion
We have experimentally established the suitability of an all-silicon optical modulator to support ultra-high-capacitycoherent optical transmission links beyond 400 Gb/s. Silicon modulators also provide a flexible solution for on-chip, few-carrier optical frequency comb generation. Monolithic integration of the flexible optical frequency comband IQ modulators provides a promising solution for Tb/s super-channel transmitters.
Acknowledgement This project was supported by Huawei Canada and the Natural Sciences and Engineering ResearchCouncil of Canada (CRDPJ 486716-15). The authors thank CMC Microsystems for the fabrication subsidy and MPW service.
References[1] R. Going, M. Lauermann, R. Maher, H. Tsai, M. Lu, N. Kim, S. Corzine, P. Studenkov, J. Summers et al., “Multi-channel InP-based
coherent PICs with hybrid integrated SiGe electronics operating up to 100 GBd 32QAM,” in Proc. Eur. Conf. Opt. Commun, (2017).[2] Y. Ogiso, J. Ozaki, Y. Ueda, N. Kashio, N. Kikuchi, E. Yamada, H. Tanobe, S. Kanazawa, H. Yamazaki et al., “Over 67 GHz bandwidth
and 1.5 V Vπ InP-based optical IQ modulator with n-i-p-n heterostructure,” Journal of Lightwave Technology 35, 1450–1455 (2017).[3] G. Raybon, J. Cho, A. Adamiecki, P. Winzer, L. Carvalho, J. Oliveira, A. Konczykowska, J.-Y. Dupuy, and F. Jorge, “180-Gb/s (90-GBd
QPSK) single carrier transmitter using a thin film polymer on silicon I/Q modulator,” in ECOC 2016; 42nd European Conference onOptical Communication; Proceedings of, (VDE, 2016), pp. 1–3.
[4] S. Wolf, H. Zwickel, C. Kieninger, M. Lauermann, W. Hartmann, Y. Kutuvantavida, W. Freude, S. Randel, and C. Koos, “Coherentmodulation up to 100 GBd 16QAM using silicon-organic hybrid (SOH) devices,” Optics express 26, 220–232 (2018).
[5] M. M. Milosevic, S. Stankovic, S. Reynolds, T. D. Bucio, K. Li, D. J. Thomson, F. Gardes, G. T. Reed et al., “The emergence of siliconphotonics as a flexible technology platform,” Proceedings of the IEEE pp. 1–16 (2018).
[6] W. Shi, Y. Xu, H. Sepehrian, S. LaRochelle, and L. A. Rusch, “Silicon photonic modulators for PAM transmissions,” Journal of Optics20, 083002 (2018).
[7] C. Doerr and L. Chen, “Silicon photonics in optical coherent systems,” Proceedings of the IEEE pp. 1–11 (2018).[8] H. Bahrami, H. Sepehrian, C. S. Park, L. A. Rusch, and W. Shi, “Time-domain large-signal modeling of traveling-wave modulators on
soi,” J. Lightwave Technol. 34, 2812–2823 (2016).[9] J. Lin, H. Sepehrian, L. A. Rusch, and W. Shi, “CMOS-compatible silicon photonic iq modulator for 84 gbaud 16qam and 70 gbaud
32qam,” in 2018 Optical Fiber Communications Conference and Exposition (OFC), (2018), pp. 1–3.[10] H. Sepehrian, A. Yekani, W. Shi, and L. A. Rusch, “Assessing performance of silicon photonic modulators for pulse amplitude modula-
tion,” IEEE Journal of Selected Topics in Quantum Electronics 24, 1–10 (2018).[11] S. Zhalehpour, J. Lin, H. Sepehrian, W. Shi, and L. A. Rusch, “Mitigating pattern dependent nonlinearity in SiP IQ-modulators via
iterative learning control predistortion,” Opt. Express 26, 27639–27649 (2018).[12] J. Lin, H. Sepehrian, Y. Xu, L. A. Rusch, and W. Shi, “Frequency comb generation using a CMOS compatible SiP DD-MZM for flexible
networks,” IEEE Photonics Technology Letters 30, 1495–1498 (2018).
