Thin-film lithium niobate dual-polarization IQ modulator on a silicon substrate for single-carrier 1.6 Tb/s transmission Open Access

With the explosive growth of Internet data traffic, the demand for optical interconnection capacity is increasing in both data-center and backbone network. According to the Ethernet roadmap,¹ 800 G and 1.6 T will be the next-generation optical interface speeds after 2023. Considering the limited bandwidth of optical fibers and amplifiers, high spectrum efficiency transmission has become the focus of attention. Coherent transmission technology has proven to effectively improve spectrum efficiency by utilizing advanced modulation format and polarization division multiplexing (PDM).^2,3 In-phase/quadrature (IQ) modulator is a core device for a coherent transmission system, which allows encoding of information onto both the amplitude and phase of light. For decades, much effort has been made to achieve IQ modulators on various material platforms, including silicon (Si),^4–6 indium phosphide (InP),^7–9 plasmonic,^10,11 and silicon-organic hybrid (SOH).^12,13 The modulators on each above-mentioned platform have their own limitations, such as high driving voltage and limited bandwidth for Si, complex and expensive monolithic integration fabrication processes for InP, large loss for plasmonic, and the instability of material for SOH. It is always believed that lithium niobate (LN) is an excellent material for high performance electro-optic (EO) modulators due to its strong linear EO effect, wide transparency bandwidth, and long-term good temperature stability.

Thin-film lithium niobate (TFLN) has recently emerged as an appealing material platform, which is expected to overcome the low modulation efficiency and limited EO bandwidth in traditional bulk LN modulators.^14–17 The world’s first IQ modulator has been demonstrated on the TFLN platform with 48 GHz EO bandwidth, supporting modulation data rate up to 320 Gb s⁻¹.¹⁸ The voltage-bandwidth limitation in the TFLN based modulator was then surpassed by introducing capacitively loaded traveling-wave electrode, showing the potential to enable sub-volt modulator with >100 GHz bandwidth.¹⁹ In addition, a 1.58 Tb/s net rate transmission was demonstrated with a TFLN based IQ modulator and an external PDM emulator,²⁰ which proves that TFLN based IQ modulators can be used for ultra-high-speed interconnection. Until now, the dual-polarization (DP) IQ modulator composed of monolithic integrated IQ modulators, spot-size converters (SSCs), and polarization rotator combiners (PRCs) is rarely reported on the TFLN platform, due to the more complex fabrication processes and low yield. An excellent voltage-bandwidth performance DP-IQ modulator was recently achieved on the TFLN platform with quartz substrate.²¹ The quartz substrate helps to reduce microwave attenuation and thus contributes to higher EO bandwidth. However, the production of the TFLN wafer with quartz substrate suffers from higher cost, lower yield, and smaller size than that with Si substrate. In addition, the Si substrate may be a better candidate for heterogeneous bonding with III–V distributed feedback (DFB) lasers to achieve an integrated TFLN optical transmitter.²² As coherent transmission technology sinks to data-center interconnect, low-cost and highly integrated optical components are demanded. Therefore, it is suitable to use conventional Si substrate wafers for practical production. In particular, a recently reported method of partially removing the Si substrate underneath the electrode of LN modulators significantly reduces the microwave loss, which enables the EO bandwidth close to modulators with quartz substrate.^23,24

In this paper, we demonstrate a high performance, TFLN based monolithic integrated DP-IQ modulator on a silicon substrate. The proposed modulator exhibits an insertion loss of 12 dB per polarization, 6 dB electro-optic bandwidth larger than 67 GHz, and polarization extinction ratio of 33 dB. Furthermore, a single-carrier probability-shaping 256 quadrature amplitude modulation (QAM) transmission with a net bitrate of 1.6 Tb/s via the proposed DP-IQ modulator is successfully demonstrated.

The schematic of the proposed TFLN based DP-IQ modulator is shown in Fig. 1(a). The device is designed on the 500 nm-thick X-cut TFLN wafer. The DP-IQ modulator consists of two monolithic integrated IQ modulators, several SSCs, and one PRC. The IQ modulators are driven by 1.2 cm-long ground–signal–ground–signal–ground (GSGSG) traveling-wave electrodes in a push–pull mode. The coupling of light between the fiber and the LN waveguide is through the low loss and polarization-independent SSCs.²⁵ The 1 × 2 multimode interference (MMI) couplers are used as symmetrical optical power splitters/combiners. Six thermo-optical phase shifters (PSs) are employed to control the modulation bias points of the Mach–Zehnder modulators (MZMs), and π/2 phase shift between I and Q components. The thermo-optical PSs are formed by Ti thin-film heaters, which can avoid the DC drift phenomenon in the EO effect.¹⁸ To achieve PDM, a PRC cascading an adiabatic taper and an asymmetric Y-junction is applied to combine lights from two IQ modulators into orthogonal polarization states.²⁶ The microscope images of the thermo-optical PS and PRC are shown in Figs. 1(b) and 1(c), respectively.

The high-speed modulation section is the core part of the modulator, which consists of optical waveguides and coplanar waveguide (CPW) electrodes. The etching depth of the LN ridge waveguide is 260 nm, and the 70°-waveguide sidewall angle is caused by the fabrication processes. The cross-sectional view of the modulation section is shown in Fig. 2(a). The LN ridge waveguide has a width of W = 1.5 µm in the modulation section. The electrode gap is set to G = 5 µm by a Finite Difference Eigenmode (FDE) solver to achieve the maximum modulation efficiency while ensuring that the electrodes will not introduce additional optical absorption loss. The calculated mode field distribution of the TE0 mode in the LN waveguide is shown in Fig. 2(b). Then, the CPW electrode is designed for velocity matching of the microwave and optical signals, as well as impedance matching. We optimize The CPW electrode structure by Finite Element Method (FEM). The width of the signal and ground electrodes are designed as W_S = 40 µm and W_G = 50 µm, respectively. The thickness of the Au electrode is H = 1200 nm. A 2 μm-thick SiO₂ layer covers the electrode as the upper cladding. Figure 2(c) shows the calculated electric field distribution when a 1–V voltage is applied; thus, we can get the theoretical voltage-length product of 2.43 V cm. The RF effective index is 2.3 matched to the optical group index, and the characteristic impedance of the CPW electrode is 38 $Ω$⁠. The electrode impedance is less than 50 $Ω$ as a compromise between the characteristic impedance and the microwave loss. A narrower signal electrode could be used to increase the characteristic impedance. However, this will increase the microwave loss and further degrade the EO bandwidth. The EO response is also simulated as shown in Fig. 2(d). We get the 3 dB bandwidth of 51 GHz when the termination impedance is matched.

The DP-IQ modulator is fabricated on a commercial lithium niobate on insulator (LNOI) wafer (NANOLN), which includes 500 µm silicon substrate, 4.7 µm buried oxide, and 500 nm LN thin film. Electron beam lithography (EBL), inductively coupled plasma reactive-ion-etching (ICP-RIE), electron-beam evaporation (EBE), plasma enhanced chemical vapor deposition (PECVD), and end face polishing processes are used during the fabrication. After the fabrication of the TFLN based DP-IQ modulator chip, the optical coupling between the LN waveguide and the polarization maintaining fiber array is achieved by UV-curable glue, as shown in Fig. 3(a). We use polarization maintaining fibers to ensure TE mode coupling into the LN waveguide. A load resistor (38 $Ω$⁠) is wire bonded to the termination pads of the CPW electrode to achieve an impedance match. The modulator chip is then mounted on a printed circuit board (PCB), and the electrode pads of PSs are connected to the PCB pads through gold wires. To measure the fiber-to-fiber insertion loss and polarization extinction ratio, a polarization controller (PC) is put behind the tunable laser (Santec TLS-510) to control the polarization state of light. The light is coupled into the chip from the output port of the modulator. The measured fiber-to-fiber insertion loss of the simple packaged DP-IQ modulator is 12 dB per polarization, including 6 dB from two SSCs and on-chip loss of 6 dB. On the other hand, we get the polarization extinction ratio over 33 dB at 1550 nm by comparing the transmittance of TE and TM modes. The coupling loss of SSC is evaluated to be 3 dB/facet via a straight waveguide with SSC at both ends on the same chip. Compared to our previous results,²⁵ the coupling loss of SSC increases significantly. The coupling loss is dominated by the mode field mismatch between polarization maintaining fiber (mode field diameter is 6.5 µm) and SSC (mode field diameter is 3.2 µm). Meanwhile, the position shift of the fiber array caused by UV-curable glue during the 80 °C thermal cure also affects the coupling loss. The on-chip loss is mainly caused by the optical propagation loss. We will further optimize the chip fabrication and optical packaging processes to reduce the insertion loss of the modulator.

When measuring the half-wave voltage of the modulator, a 1 kHz triangle electric signal is applied to the MZMs. As shown in Fig. 3(b), the half-wave voltage is 2.18 V, corresponding to the half-wave voltage length product of 2.6 V·cm. The extinction ratios of the MZMs are around 24 dB. The small-signal EO response of the modulator is also characterized. The RF signal is added to the GSGSG electrodes via two 67 GHz RF probes. The measured S21 and electrical reflections (S11) of the XI, XQ, YI, and YQ MZMs are shown in Fig. 3(c), where the dashed line shows the simulated S21 curve. It is worth mentioning that the S11 and S21 curves are obtained without excluding the effect of RF probes. We observe a 3 dB bandwidth roll-off around 30 GHz, and S11 of all MZMs are less than −14 dB. The measured S21 curve drops slightly faster than simulated results due to the RF attenuation caused by the probes. However, it is possible for the modulator to operate at a frequency far beyond the 3 dB bandwidth considering the flatness of the S21 curve. We can observe a 6 dB electro-optic bandwidth greater than 67 GHz. We successfully demonstrate a TFLN based DP-IQ modulator, which is more complex in fabrication than the previously reported single polarization TFLN modulator.

We use our DP-IQ modulator to demonstrate high-speed coherent transmission. The experimental setup is shown in Fig. 4(a). A 136 Gbaud single carrier probability-shaping DP-256QAM signal is generated by an arbitrary-waveform generator (AWG, Keysight 8199A) with a typical bandwidth of 70 GHz and a sampling rate of 224 GSa/s; the entropy is about 7.0 bits/symbol.²⁷ Open forward error correction (FEC) with 15% overhead is assumed, and a 3.5% overhead pilot is used.²⁸ The data signal output from the DAC consists of four real components, which are connected to four single-ended RF amplifiers (SHF S804 B) with 3 dB bandwidth of 60 GHz and 22 dB gain to drive our DP-IQ modulators. The optical spectra are shown in the inset of Fig. 4(a). The modulated signal is transmitted through a 2 km-long single-mode fiber, and then down-converted to the electrical domain using four balanced photodetectors (PDs) (6 dB attenuation at 100 GHz) at the coherent receiver. The down-converted signal is sampled by a 256 GSa/s 110 GHz real-time scope followed by offline digital signal processing (DSP). After mapping and framing, pulse shaping with 0.01 roll-off factor, nonlinear and linear pre-distortions are performed in the transmitter DSP. At the receiver, the DSP steps are (i) resampling, (ii) frame synchronization, (iii) a real-valued 4 × 2 least mean square (LMS) equalizer, (iv) down-sampling, (v) carrier phase recovery aided by training and pilot symbols, (vi) 2 × 2 LMS equalizers, (vii) whitening filter, (viii) 2 × 2 decision feedback equalizer, and (ix) symbol decision.

The measured bit error rate (BER) performances as a function of optical signal-to-noise ratio (OSNR) are shown in Fig. 4(b). The pre-FEC BER threshold is 2.0 × 10⁻². When OSNR is 36 dB, the BERs are below the pre-FEC threshold, which indicates zero error will be achieved after FEC decoding. The corresponding recovered probability-shaping 256 QAM constellations are shown in Fig. 4(c). We successfully demonstrated a single-carrier 1.6 Tb/s net bitrate transmission via our TFLN based DP-IQ modulator.

A monolithic integrated DP-IQ modulator is realized on the TFLN platform with a silicon substrate, which consists of IQ modulators, on-chip SSCs, and a PRC. After simple packaging, the proposed modulator exhibits an insertion loss of 12 dB, a 6 dB electro-optic bandwidth larger than 67 GHz (including the effect of RF probe), and a polarization extinction ratio of 33 dB. Furthermore, a single-carrier probability-shaping DP-256 QAM transmission with a net bitrate of 1.6 Tb/s via the proposed DP-IQ modulator is successfully achieved. We believe this is a promising solution for future high speed optical interconnects.

This research was supported by the National Key Research and Development Program of China (Grant Nos. 2021YFB2800104 and 2019YFB2203501) and the National Natural Science Foundation of China (Grant Nos. 61835008, 62175079, 61905079, and 61905084). The authors acknowledge the facility support from the Center of Micro-Fabrication and Characterization (CMFC) of WNLO, HUST, and the Center for Nanoscale Characterization and Devices (CNCD) of WNLO, HUST.

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding authors upon reasonable request.