Unamplified Coherent Transmission of Net 500 Gbps/Polarization/λ for Intra-Datacenter Interconnects

To keep up with the growing data traffic demand, spectrally efficient coherent systems are envisioned to replace intensity modulation direct detection (IMDD) systems in short-reach intra-datacenter interconnects (DCI). In this work, we characterize and test an unamplified coherent transmission system employing a C-band thin-film lithium niobate (TFLN) in-phase quadrature modulator (IQM) that has 65 GHz 6-dB electro-optic bandwidth and 1.25 V half-wave voltage (<inline-formula> <tex-math notation="LaTeX">$\text{V}_{\pi }$ </tex-math></inline-formula>). We demonstrate the transmission of 128 (124) Gbaud 32QAM over 2 (10) km on a single polarization with 2.5 Vpp drive signals below the 20% overhead (OH) soft-decision forward error correction (SD-FEC) BER threshold of <inline-formula> <tex-math notation="LaTeX">$2.4\,\,\times \,\,10^{-2}$ </tex-math></inline-formula>, corresponding to a net rate of 533 (516) Gbps. We evaluate and discuss the digital signal processing (DSP) requirements. Our results and analysis show the potential of shifting to coherent systems for high-capacity short-reach links below 10 km.

that IMDD systems cannot support 400G/λ for DCI reach. 28 To date, there have been few demonstrations of IMDD sys-29 tems operating at net 300-350 Gbps/λ [3], [4], [5], [6], [7]. 30 However, achieving higher capacities requires employing a 31 higher PAM order than PAM8 or operating at symbol rates 32 beyond 140 Gbaud, which is very challenging with the existing 33 RF components and electro-optic modulators [8]. Therefore, a 34 paradigm shift towards coherent systems for high-speed short-35 reach links is envisioned [9], [10], [11]. Coherent systems 36 achieve higher data rate capacities because of their higher 37 spectral efficiency at the same symbol rates, which comes at 38 Manuscript received 12  the expense of added DSP complexity and hardware overhead. 39 The developments in ASIC design and fabrication ease the 40 inclusion of the extra DSP blocks of coherent systems. Con-41 sidering the 7 and 5 nm technology nodes, the ASIC power 42 consumption of coherent and IMDD systems are comparable 43 for short-reach applications (< 2km) with negligible difference 44 at 5 nm [12]. Besides, a comparison between 400G coherent 45 transceivers and 4λ 400G IMDD transceivers in the intra-46 datacenter reach showed that coherent transceiver modules 47 achieve a higher optical power budget (achievable optical path 48 loss) at similar ASIC power consumption [13]. Therefore, the 49 specifications of next-generation short-reach coherent trans-50 ceivers are being studied to ensure that the power envelope is 51 competitive versus IMDD solutions [11]. Although coherent 52 receivers offer higher detection sensitivity because of the 53 mixing of the signal with local oscillator (LO), operating with-54 out optical amplification mandates reducing the link overall 55 optical loss. A low loss electro-optic modulator with a low 56 V π (for low modulation loss) is necessary to maintain the 57 optical power at detectable levels even in the absence of trans-58 impedance amplifiers (TIAs), as in this work. High bandwidth 59 electro-optic modulators exist on different material systems, 60 including lithium niobate (LiNbO3), gallium arsenide (GaAs), 61 indium phosphide (InP), and silicon (Si). However, thin-film 62 lithium niobate (TFLN) modulators stand as a promising 63 candidate for unamplified coherent systems because of their 64 very low optical propagation loss (< 0.7 dB/cm), enabling 65 long devices with low V π , while maintaining high bandwidths 66 due to the low RF loss [14], [15].

67
This work evaluates the transmission performance of a 68 C-band unamplified coherent system employing a high band-69 width TFLN IQM and TIA-free PIN photodiodes over short 70 distances (2 to 10 km). We analyze the driving voltage and 71 DSP requirements for optimum performance. We experimen-72 tally demonstrate the transmission of 124 Gbaud 32QAM on 73 a single polarization over 10 km of standard single-mode fiber 74 (SSMF) with 2.5 V pp drive signals below the 2.4 × 10 −2 SD-75 FEC BER threshold, corresponding to a net rate of 516 Gbps. 76 Moreover, we transmit 124 Gbaud 16QAM over 10 km below 77 the 3.8 × 10 −3 hard-decision (HD)-FEC BER threshold, 78 which corresponds to 465 Gbps net rate and is aligned with the 79 envisioned 800G LR1 standard [16]. Our results support the 80 promise of practical unamplified coherent systems with data 81 rates beyond 1 Tbps/λ over the intra-DCI reach with standard 82 polarization division multiplexing and currently available elec-83 tronics analog bandwidths. per symbol (sps) for chromatic dispersion (CD) compensa-129 tion and frequency offset (FO) correction. Then, a 51-tap 130 T/2-spaced 2 × 2 multiple-input-multiple-output (MIMO) 131 equalizer with real-valued coefficients and interleaved with a 132 second-order phase-locked loop (PLL) is employed to track the 133 phase noise and compensate for the frequency response of the 134 RF probes, IQM, and BPDs. The real-valued MIMO equalizer 135 filters each quadrature independently and can correct the power 136 imbalance and timing skew between the I and Q quadratures 137 [18]. Finally, the BER is calculated based on the demapped 138 bit sequence of the received symbols.

139
The TFLN IQM is composed of nested MZMs with 23 mm 140 coplanar waveguide electrodes and on-chip termination close 141 to 50 . The bias points are set with thermal phase shifters, 142 to minimum transmission (null) for the children MZMs and 143 to quadrature for the parent MZM. The measured small-144 signal electro-optic (EO) frequency response (S 21 ) of a MZM 145 identical to those used in the IQM is shown in Fig. 2(a). 146 The characteristic slow roll-off response of TFLN MZMs 147 is observed with a 3-dB bandwidth of 24 GHz and 6-dB 148 bandwidth of 65 GHz. Fig. 2(b) shows the measured RF V π 149 at different frequencies. Each data point is extrapolated from 150 DC to 100 GHz using the measured EO S 21 response. The 151 measured low-MHz V π is 1.25 V that increases to ∼3 V 152 at 60 GHz.

154
The summary of the transmission experiment results is 155 presented in Fig. 3(a). On a single polarization, we transmit 156 128 (124) Gbaud 16QAM over 2 (10) km under the 6.7% OH 157 HD-FEC BER threshold of 3.8 × 10 −3 , which represents a 158 net rate of 480 (465) Gbps. Adopting a higher FEC thresh-159 old, we demonstrate the transmission of 128 (124) Gbaud 160 32QAM over 2 (10) km at a BER below the 2.4 × 10 −2 161 threshold of the 20% OH SD-FEC, corresponding to a net 162  [19]. 180 Considering only the IQM non-linear transfer function, the 181 optimum modulation depth is between 0.5 and 0.6, which 182 addresses the tradeoff between modulation loss and non-183 linear compression [19]. However, the employed RF amplifier is observed at the SD-FEC threshold when NLPD is employed. 201 Although our experimental setup does not include TIAs after 202 the BPDs, it detects signals below −10 dBm owing to mixing 203 with the LO and the higher detection sensitivity of coherent 204 receivers. Fig. 3(f) plots the received RF spectra at different 205 symbol rates. The 128 Gbaud signal experiences ∼7 dB drop 206 at 60 GHz, which corresponds to the combined frequency 207 response of the RF probes, TFLN IQM, and BPDs, and 208 requires large number of MIMO taps for proper equalization. 209 Fig. 3(h) shows the constellations of 124 Gbaud 16QAM 210 and 32QAM after 10 km transmission, which respectively 211 correspond to net rates of 465 and 516 Gbps.

212
Employing an edge-coupled IQM will save an extra 6 dB 213 of optical power, while a DP IQM will result in a 3 dB 214 reduction in the optical power per polarization; yielding a 215 net 3 dB improvement in the power budget per polarization. 216 This supports the potential of extending our results to dual 217 polarization and realizing net 1 Tbps/λ intra-DCIs.

218
It is yet debatable whether to employ O-band or C-band 219 lasers in such short-reach unamplified coherent systems. In the 220 O-band, the impact of CD is negligible; however, the fiber 221 loss is slightly higher (0.35 dB/km) compared to the C-band 222 (0.2 dB/km). Moreover, the O-band optical hybrids and BPDs 223 are commercially available, but not as mature as the C-band 224 components. Thus, the primary drawback of operating in the 225 C-band is the need to digitally compensate for the CD, which 226 increases the ASIC power consumption [9], [10]. For 2 km 227 reach, the accumulated CD is adequately low that it does 228 not require a dedicated DSP block for compensation [9], 229 [20]. In this work, we adopted the conventional time-domain 230 finite impulse response (FIR) CD compensation filter [21]. 231 For short-reach applications up to 10 km, compensating the 232 CD in the time domain is advantageous compared to the 233 frequency domain equalization (FED). Because it requires 234 modest number of filter taps (less than 100), which is com-235 putationally less exhausting compared to the computation of 236 the fast Fourier transform (FFT). Fig. 4(a) shows the BER 237 sensitivity to the length of the MIMO filters for 128 Gbaud 238 The authors would like to thank HyperLight for the TFLN 278 modulator.