C+L-Band InP-Based Coherent Driver Modulator Enabled Net-1.8 Tbps/λ Transmission

We developed the world's first InP-based C+L-band coherent driver modulator (CDM) that has a 3-dB electro-optic (EO) bandwidth of over 90 GHz. The CDM consisted of an InP-based n-i-p-n heterostructure twin IQ modulator chip with a differential capacitively loaded traveling-wave electrode, a SiGe BiCMOS driver in an open-collector configuration, and an RF package with a flexible printed circuit as the RF interface, with 3-dB bandwidths exceeding 105, 97, and 77 GHz, respectively. In terms of optical characteristics, by using a waveguide with a middle-ridge structure and optimizing the epitaxial structure of the InP modulator chip, the wavelength dependence of the absorption loss, which has been a problem with conventional InP modulators, is suppressed, and the insertion loss per polarization at maximum transmission, including a bias loss of V_π = 2.0 V, is less than 11 dB over the C+L band. In addition, by using the newly developed CDM, we achieved a single-carrier net bit rate of 1.8 Tbps after 80-km of standard single-mode fiber transmission in the C+L band.

SECTION I. Introduction

The demand for high-capacity optical transmission systems continues to grow unabated to handle the ever-increasing IP traffic. To increase the transmission capacity per wavelength, various kinds of research and development on key technologies are underway to achieve optical coherent transmission systems with an over terabit per second class channel rate [1], [2], [3], [4]. It is useful to use high-baud rates and high-order multi-level modulation formats to reduce the cost per bit. Increasing the baud rate is very important from the perspective of suppressing the degradation in transmission distance as capacity increases. However, high baud rates increase the bandwidth of the modulated optical signal, reducing the number of wavelengths that can be accommodated in the C-band. For example, the 130-Gbaud class system, which is already in commercial use in coherent optical communication systems, has a channel spacing of 150 GHz, and the 260 Gbaud class, which is expected to be the target baud rate for the next generation, will extend the spacing to 300 GHz, resulting in approximately half the number of channels that can be accommodated in the same wavelength range. Therefore, to increase the number of channels that can be accommodated, the wavelength range must be extended from the conventional C-band to the super C-band [5] and C+L-band [6].

A high-bandwidth coherent driver modulator (HB-CDM) [7], in which a driver die and a modulator chip are co-packaged to reduce RF losses and the footprint, is known as a high-end transmitter capable of high-speed and high-order modulation formats for metro and long-haul coherent networks. The modulator chips used in high-performance CDMs generally use InP [8], [9], [10] or thin-film LiNbO₃ (TFLN) modulators [11], [12], [13] because both a high bandwidth and low half-wave voltage (V_π) are required. The modulator and driver IC should be in a differential drive configuration to achieve a high signal integrity and efficient connection [14], [15]. InP modulators can be designed with a differential drive configuration [16], while TFLN modulators have a single-ended drive and it is difficult to design a differential drive configuration [13], which may be one of the reasons why InP modulators are popular in the market for CDMs, while TFLN modulators are not. However, unlike LN modulators, InP modulators, due to their operating principle, have a large wavelength dependence of insertion loss when set to a low V_π, such as below 2V, regardless of wavelength. More specifically, since LN modulators use the Pockels effect for refractive index change, they have small wavelength dependence of the insertion loss and can operate over a wide wavelength range, while InP modulators utilize electroabsorption and electrorefraction effects, such as the quantum-confined Stark effect (QCSE) and the Franz-Keldysh effect in addition to the Pockels effect, and generally have a large wavelength dependence [6], [17], limiting their operating wavelength range to the C- or L-band. Therefore, CDMs with InP modulators currently available on the market are designed for the C-band and L-band separately by changing the epitaxial and waveguide design of the InP modulator chip depending on the wavelength range.

We previously reported on the world's first C+L-band CDM including an InP-based n-i-p-n heterostructure twin IQ modulator chip with a differential capacitively loaded traveling-wave electrode (CL-TWE) [18], [19]. The CDM has a 3-dB electro-optic (EO) bandwidth of over 90 GHz in the C+L band. The insertion loss (IL) per polarization at maximum transmission, including a bias loss of V_π = 2.0 V, is less than 11 dB over the C+L band, and the wavelength dependence is smaller compared with previously reported InP modulators [6], [17]. We demonstrated an 80-km transmission with a net bit rate of up to 1.8 Tbps/λ in the C+L band by using the newly developed CDM. In this extended paper, we present the details of the waveguide structure design with suppressed wavelength dependence of the insertion loss compared with the conventional waveguide design in Section II, and the differences in optical characteristics are explained in Section III. Additional transmission experiment results are also described in Section III.

SECTION II. Design of C+L-Band Coherent Driver Modulator

Fig. 1 shows an external view of the fabricated CDM and a schematic diagram of its internal configuration. The package body size is 11.9 × 29.8 × 4.35 mm³, which is more than 1 mm lower in height than our previous CDMs [8] and the maximum package height for CDMs standardized in the Optical Interworking Forum (OIF) [7]. By reducing the thickness of the chip ceramic carrier and using low-loop wires, the package was successfully made to have a lower profile. There are three key components that make up the CDM: an InP-based n-i-p-n heterostructure twin IQ modulator chip with a differential CL-TWE, a SiGe BiCMOS differential linear driver die, and an RF package with a flexible printed circuit (FPC) RF interface.

Fig. 1.

Photograph and schematic diagram of fabricated CDM.

The modulator chip layout is equivalent to that reported previously [20], containing four Mach-Zehnder (MZ) modulators in parallel and using crossed waveguides to achieve symmetric IQ layouts and footprint reduction to 2.5 × 5.0 mm². Since the input/output light to the modulator chip is lens-coupled, a spot size converter (SSC) [21] is integrated to reduce coupling loss and stabilize optical lens coupling. A polarization beam combiner (PBC) and monitor photodiode (MPD) are assembled off-chip in the CDM using spatial optics. Moreover, the modulator chip is mounted on a thermoelectric cooler (TEC) to suppress optical characteristic variations due to environmental temperature changes and to achieve stable operation. Although it has been reported that some InP modulator chips integrate a semiconductor optical amplifier (SOA) [22], [23], [24], an SOA is not integrated in this modulator chip to avoid limiting the operating wavelength range due to the SOA's wavelength dependence since the SOA is generally operated in the C-band or L-band. We changed the waveguide structure from the conventional ridge waveguide to the middle-ridge waveguide (etched about half the thickness of the i-layer) shown in Fig. 2 to increase the optical confinement and the electric field strength applied to the multiple quantum well (MQW) and improve the modulation efficiency. Fig. 2 also shows calculated optical waveguide modes, where the amount of light confined in the MQW below the upper n-InP layer area is improved by about 40%. In addition, the epitaxial structure was revised to minimize the unwanted bulk semiconductor absorption caused by the electroabsorption effect, resulting in an InP modulator chip capable of operating over the C+L band with low wavelength dependence of the absorption loss. Optimizing the structure of the multi-mode interference (MMI) couplers and suppressing the wavelength dependence of transmission loss are also very important to extend the wavelength range [25]. To compensate for the increased transmission loss on the short wavelength side due to fundamental absorption in the semiconductor layer, the center wavelength of the MMI couplers was set to about 1555 nm, which is slightly more on the C-band side than the center wavelength of the C+L band, and the dimensions were designed to cover as wide a wavelength range as possible.

Fig. 2.

(a) Conventional (ridge) and (b) new waveguide (middle ridge) structure with calculated optical waveguide modes.

Figs. 3(a)–​(c) respectively show the measured small-signal RF characteristics of the package with an 8.5-mm-long FPC, the driver die with wire inductance to an input and output driver pad, and the modulator chip. The FPC RF package has a 3-dB electrical (EE) bandwidth of more than 77 GHz, and the roll-off frequency exceeds 125 GHz due to the use of small-diameter vias to suppress electromagnetic leakage around the FPC and the ceramic package connection area. Although the roll-off frequency has not been confirmed within the measured frequency range, it is expected to be above 130 GHz based on the simulation results [8]. Fig. 3(b) shows the measured differential gain Sdd21, re-normalized to an input and output port impedance of differential 100 Ω and 60 Ω respectively, when 100 pH is added to the input and output of the measured S-parameter, accounting for the wire connection between the driver and the package or the modulator. The driver has a 3-dB EE bandwidth of over 97 GHz and has more than 10-dB peaking from 70 to 80 GHz. The differential output amplitude is 2.5 V_ppd for a differential 60-Ω load. To suppress driver oscillation caused by the cavity resonant frequency within the operating frequency range, the spatial vertical distance between the lid and the driver surface in the CDM is sufficiently close [26]. Fig. 3(c) shows the small-signal EO response (Ssd21) of the modulator chip and the EO measurement data were normalized at 1 GHz. The modulator has a 3-dB EO bandwidth of over 105 GHz with V_π = 2.0 V, which is higher than that of our previously reported modulator [8], and is due to lowering the RF loss of the CW-TWE by reducing the metal resistance and lowering the capacitance by changing the waveguide structure.

Fig. 3.

RF characteristics. (a) Sdd21 of FPC package, (b) sdd21 of driver die, and (c) Small-signal EO response (Ssd21) of modulator chip.

SECTION III. Experiment Result

A. CDM's Characteristics in C+L Band

Figs. 4(a)–​(c) show the measured insertion loss (IL) per polarization at maximum transmission from the wavelength of 1527 nm to 1610 nm, including a bias loss of V_π = 2.0 V, for our previously reported C-band CDM with a conventional waveguide structure [8] and our newly developed C+L-band CDM with the new waveguide structure, and the bias loss and voltage for V_π of 2.0 V and 1.5 V, respectively. The IL per polarization of the C+L-band CDM with the new structure was less than 11 dB over the C+L band. It was more than 5 dB less than the IL specified in the OIF [8], despite covering the C+L band. As shown in Fig. 4(a), the conventional structure had a good IL in the C-band, but in the L-band, the IL increased by up to 5 dB or more compared to the C-band, making it impossible to achieve the required transmission performance in the L-band. Since the same structure was used for MMI waveguides, the increase in the IL on the L-band side in the conventional structure was due to the wavelength dependence of the absorption loss. As shown in Fig. 4(b), the wavelength dependence of the absorption loss was small with the new waveguide structure, and very low absorption losses of around 0.5 dB and 1.0 dB were achieved in the C+L band for V_π of 2.0 V and 1.5 V, respectively. The modulator bias voltage was adjusted for each wavelength to obtain the V_π of 2.0 V or 1.5 V, as shown in Fig. 4(c). To reduce V_π, the bias voltage should be lower, which increases the absorption loss. In addition, the measured extinction ratio (ER) of the child MZ and parent MZ were >28 dB and >30 dB, respectively, over the C+L band.

Fig. 4.

Optical characteristics of CDM with the new waveguide structure. (a) Insertion loss per polarization for V_π of 2.0 V compared with the conventional waveguide structure, (b) bias loss and (c) voltage for V_π of 2.0 V and 1.5 V.

Fig. 5 compares the IL after X and Y combining of the C+L band CDM when the structure of the cross waveguide integrated in the modulator chip was the MMI type and the tapered type. The graph shows that the roughly 2.0-dB increase in IL on the L-band side, which was also observed in Fig. 4(a), was due to the wavelength dependence of the MMI-type cross waveguide and that the tapered-type cross waveguide reduced the wavelength dependence by about 1 dB, indicating that the tapered cross waveguide contributed to reducing of the transmission loss in the L-band and extending the wavelength range, and is optimal for the C+L-band CDM. On the other hand, considering only the narrow wavelength bands of the C-band or L-band, the MMI type could lead to a CDM with a lower IL than the tapered type.

Fig. 5.

Optical insertion loss of X+Y polarization for Vp of 2.0 V in comparison with MMI-type and taper-type cross waveguides.

Fig. 6 shows the CDM's measured small-signal EO response normalized at 1 GHz without evaluation board loss, which was de-embedded by using a test coupon's measurement result [8], and the 3-dB EO bandwidth exceeded 90 GHz over the C+L band. Since the modulator can generally operate at baud rates up to at least twice the 3-dB EO bandwidth, the CDM is capable of 180 Gbaud class operation.

Fig. 6.

Small-signal EO response.

B. IQ Modulation

Fig. 7 shows the setup for 180-Gbaud digital coherent transmission. The CDM was soldered to a printed circuit board (PCB) that had a 100-GHz compatible subminiature push-on (SMPS) connector as the RF interface and MIL connectors as the DC interface for device control and evaluation [26]. To emulate a digital signal processor (DSP) and digital-to-analog converter (DAC), an offline PC and 256-GSa/s arbitrary waveform generator (AWG) with a 3-dB analog bandwidth of >80 GHz were used. An external-cavity laser (ECL) with a linewidth of <100 kHz was used as a continuous wave (CW) input light source. For evaluating the transmission characteristics in the C+L band, the light source was set to a total of five wavelengths (1530, 1550, 1565, 1590, and 1605 nm) with an output power of +16 dBm. For each of these wavelengths, the modulator bias was set to obtain the V_π of 2.0 V. The modulated optical signal from the CDM was amplified by an erbium-doped fiber amplifier (EDFA), passed through a VOA, adjusted to the optimum optical power, and transmitted over 80-km standard single-mode fiber (SSMF), an EDFA, another VOA, and an optical band-pass filter (OBPF). The optical signal was received by a coherent receiver frontend with a 3-dB analog bandwidth of >100 GHz, which was followed by a 256-GS/s 110-GHz digital storage oscilloscope (DSO).

Fig. 7.

Experimental setup for IQ modulation.

We used probabilistically constellation-shaped (PCS) quadrature amplitude modulation (QAM) signals. The RF signals were pulsed-shaped by using a root-raised-cosine (RRC) filter with a roll-off factor of 0.05. The length of the symbol sequence was approximately 3 × 10⁵, and the pilot overhead (OH) was set to 0.79%. The RF signals were input by connecting the AWG and the RF connectors on the evaluation board, on which the CDM was soldered, via an RF cable with a total length of about 25 cm. Assuming the use of forward error-correction (FEC) codes with a total code rate of 0.826 [27], the net bit rate was calculated as {H−(1−0.826) × 16}/1.0079 × B Tbps and {H−(1−0.826) × 12/1.0079 × B Tbps for PCS-144QAM and PCS-64QAM, respectively, for a signal with an entropy of H bits/4D-symbol and a baud rate of B Tbaud. For example, the net data rates of dual polarization (DP) 180-Gbaud PCS-144QAM RF signals with an entropy of 12.86 bits/symbol were calculated as {12.86−(1−0.826) × 16)/1.0079 × 0.18} = 1.80 Tbps. On the transmitter side, we used a fixed linear digital equalizer to partially compensate for the frequency channel response without optical pre-equalization. The optical spectrum at 1550 nm with the 180-Gbaud PCS-144QAM signal is shown in Fig. 7. On the receiver side, we used a DSP that was largely similar to that used in [28], which included a frequency-domain 8 × 2 adaptive equalizer with an FFT block size of 4096. We calculated the normalized generalized mutual information (NGMI) as the performance metric.

The NGMIs measured over the C+L band for net data rates of 1.8 Tbps (PCS-144QAM, H = 12.86), 1.7 Tbps (PCS-144QAM, H = 12.30), and 1.6 Tbps (PCS-64QAM, H = 11.05) for 180-Gbaud transmissions and 1.8 Tbps (PCS-144QAM, H = 13.46), 1.7 Tbps (PCS-144QAM, H = 12.86), and 1.6 Tbps (PCS-64QAM, H = 11.57) for 170-Gbaud transmissions are shown in Figs. 8(a) and (b), respectively. The dotted and solid lines respectively show the NGMIs for back-to-back and after 80-km SSMF transmission results. Constellations under each modulation format after 80-km SSMF transmission at 1605 nm are also shown in the figure. The NGMIs measured for all the transmission formats, both back-to-back and after 80-km SSMF transmission, exceeded the NGMI threshold of 0.857 [27], indicating successful 80-km transmissions at data rates up to 1.8 Tbps for both 180 Gbaud and 170 Gbaud over the C+L band. For the transmission formats measured in this study, the transmission results for 170 Gbaud were slightly better than those for 180 Gbaud. For example, comparing the NGMI values, after 80-km transmission at 1.8 Tbps, the NGMI was >0.8668 for 180 Gbaud and >0.8703 for 170 Gbaud in the C+L band. This may be due to the insufficient total EO bandwidth for 180-Gbaud transmission, including the evaluation board and RF cable, and the insufficient analog bandwidth of the AWG, which degraded the generated RF signal quality. At 1.6-Tbps transmission, which is considered to be the transmission capacity for the next generation, the 80-km transmission results suggest that there is a sufficient margin against the NGMI threshold and that longer distance transmission is possible. The significant performance improvements over those of our previous CDMs [8], [29] are due to the optimization of the driver's peaking and the increase in the CDM bandwidth. To the best of our knowledge, this is the world's first CDM to successfully transmit in the C+L band with the highest transmission capacity per wavelength ever reported for a CDM.

Fig. 8.

NGMI of back-to-back and 80-km SSMF transmission of 1.8, 1.7, and 1.6 Tbps for (a) 180-Gbaud and (b) 170-Gbaud operation.

SECTION IV. Conclusion

By introducing a middle-ridge waveguide structure and optimizing the epitaxial structure for an InP modulator chip, the wavelength dependence of the insertion loss due to absorption loss, which is a problem in conventional InP modulators, is suppressed, enabling CDM C+L-band operation. The CDM's insertion loss per polarization is less than 11 dB, and the 3-dB EO bandwidth exceeds 90 GHz over the C+L band. We successfully transmitted up to a net bit rate of 1.8 Tbps using a DP 180-Gbaud PCS-144QAM signal with an entropy of 12.86 bits/symbol and a DP 170-Gbaud PCS-144QAM signal with an entropy of 13.46 bits/symbol over 80-km SSMF in the C+L band.