Intra-datacenter Interconnects Research Articles

Adaptive Partial-Response Neural Network Equalization for Bandwidth-Limited PAM Transmission in Intra-Datacenter Interconnect

Adaptive Partial-Response Neural Network Equalization for Bandwidth-Limited PAM Transmission in Intra-Datacenter Interconnect

Performance analysis of BTO-based MZM for 200 Gbps/λ O-band intra-datacenter interconnects

Performance analysis of BTO-based MZM for 200 Gbps/λ O-band intra-datacenter interconnects

Simplified THP and M-Log-MAP Decoder Based Faster Than Nyquist Signaling for Intra-Datacenter Interconnect

In this paper, a joint application of simplified Tomlinson-Harashima precoding (THP) and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">M</i> -log-maximum a posteriori probability ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">M</i> -log-MAP) decoder at transceivers is proposed for low complexity faster than Nyquist (FTN) 4-ary pulse amplitude modulation (PAM4) signaling. A THP with only two taps is applied at the transmitter for FTN-induced inter-symbol interference elimination and equalization-induced noise amplification issue mitigation, which effectively relieves the burden of the receiver. At the receiver, a low complexity M-log-MAP decoder is equipped for residual impairments cancellation, thus a feedback channel is not required to accurately obtain the channel state information for the transmitter. Taking advantage of this collaborative approach, the proposed scheme performs much better than applying the two schemes separately. We experimentally demonstrate the proposed scheme in 70, 75, and 80GBaud PAM4 systems with a 32-GHz brick-wall filter, corresponding to 0.914, 0.853, and 0.8 FTN compression factor, respectively. The results show that the 80GBaud PAM4 signal is successfully transmitted over 2-km single mode fiber with the bit-error-rate (BER) under the 7% hard-decision forward error correction (HD-FEC) threshold of 3.8×10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−3</sup> . It is also observed that the proposed scheme shows about 0.7dB receiver sensitivity improvement compared to the error propagation-free decision feedback equalizer, namely the performance upper bound of the THP, at BER of 7% HD-FEC threshold in a 75GBaud FTN PAM4 system. According to the experimental results, the proposed scheme is a promising solution for bandwidth-starved low-cost intra-datacenter interconnect.

3D Lumerical simulation of silicon photodiodes with microholes for high-speed short-reach intra-datacenter interconnects.

3D simulations are conducted using Lumerical software to study the performance of surface illuminated silicon positive-intrinsic-negative photodiodes with microholes. Drift-diffusion equations are solved including the effects of carrier lifetime due to Shockley-Read-Hall and Auger recombination mechanisms, as well as high field mobility. Lumerical's FDTD tool is used to determine the light absorption in the device. The generation profile is imported to Lumerical's CHARGE tool to determine the transient-limited impulse response. An equivalent circuit of the photodiode with microholes is developed for the simulation of an end-to-end high-speed system. Simulation results show an open eye diagram at 50 Gbps for 20µm×20µm devices.

Low-latency wavelength-switched clock-synchronized intra-data center interconnects enabled by hollow core fiber.

Fast (nanoseconds) optical wavelength switching is emerging as a viable solution to scaling the size and capacity of intra-data center interconnection. A key enabling technology for such systems is low-jitter optical clock synchronization, which enables sub-nanosecond clock and data recovery for optically switched frames using low-cost methods such as clock phase caching. We propose and demonstrate real-time low-latency wavelength-switched clock-synchronized intra-data center interconnection at 51.2 GBd using a fast tunable laser (with ns scale switching time) and ultra-stable-latency hollow core fiber (HCF) for optically-switched data center networks. For wavelength-switched systems, we achieve a physical layer latency below 46 ns, consisting of 28 ns transceiver latency and a 18 ns inter-packet gap. Finally, we show that by exploiting the low chromatic dispersion and thermally-stable latency features of HCF, active clock phase tracking can be entirely eliminated.

A Multi-Rate Approach for Nonlinear Pre-Distortion Using End-to-End Deep Learning in IM-DD Systems

Modern intra-data center (IDC) interconnects leverage robust and low-cost intensity modulation (IM) and direct detection (DD) optical links, based on multimode fibers (MMFs) and vertical-cavity surface-emitting lasers (VCSELs). Current solutions, based on on-off keying (OOK) modulations, reach up to 25-50 Gbps per lane over nearly 100 meters. The actual target for IDCs is to increase VCSEL-MMF links capacity up to 100 Gbps, using PAM-4 on the same devices. To counteract the consequent linear and nonlinear distortions affecting the transmitted signals, an effective solution is to exploit digital signal processing (DSP). In this manuscript, we propose a novel method to optimize a nonlinear artificial neural network (ANN) digital pre-distorter (DPD), based on End-to-end (E2E) learning, that, trained jointly with a Feed-Forward Equalizer (FFE), fulfills physical amplitude constraints and handles different ratio between the sampling rates incurring along with an optical IM-DD system. We indeed propose an E2E ANN system operating simultaneously at different sampling frequencies. We moreover propose in our training method a substitution to the time-domain injection of the receiver noise in the system with an additive regularization term in the FFE gradient loss. We experimentally show the advantages of our proposed DPD comparing the bit error rate (BER) performance against the same scenario without DPD. We assess the gain in terms of Gross Bit Rate and Optical Path Loss (OPL), at given BER targets, for different fiber lengths.

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 xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <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 xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <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.

Multi-band DFT-S 100 Gb/s 32 QAM-DMT transmission in intra-DCI using 10 G-class EML and low-resolution DAC.

In this paper, 100 Gb/s/λ 32 quadrature amplitude modulation discrete multi-tone (QAM-DMT) transmission using 10 G-class electro-absorption modulated laser (EML) and 4/5-bit digital-to-analog converters (DACs) are experimentally demonstrated to meet the requirement of intra-datacenter interconnection (intra-DCI). Unequal length multi-band (ULM) discrete Fourier transform spread (DFT-S) precoding is investigated to alleviate the distortion induced by the high peak-to-average power ratio (PAPR) of DMT. The results show that the required computational complexity of ULM DFT-S precoding with 2-bands (k1=256, k2=64) decreases sharply compared to the traditional DFT-S technique with only about 0.5 dB receiver sensitivity penalty. In addition, compared to the equal length multi-band (ELM) DFT-S precoding, the ULM DFT-S precoding can bring about 2.5 dB receiver sensitivity improvement with slight added computational complexity. With the assistance of ULM DFT-S precoding and noise shaping (NS) technique, the bit-error ratio (BER) of 100 Gb/s 32 QAM-DMT signal generated by 5-bit DAC over 2-km single-mode fiber (SMF) transmission can reach the hard-decision forward error correction (HD-FEC) threshold with received optical power (ROP) of -6.5 dBm, with only additional 39.9% multiplier and 33.7% adder.

Silica Microsphere WGMR-Based Kerr-OFC Light Source and Its Application for High-Speed IM/DD Short-Reach Optical Interconnects

Kerr optical frequency combs (OFCs) based on silica microsphere whispering gallery mode resonator (WGMR) have various applications where they are used as a light source. For telecommunication purposes, WGMR-based Kerr-OFC comb generators can be physically realized using silica microsphere resonators and can be used to replace multiple laser arrays. In such a realization, these novel light sources have the potential to demonstrate an attractive solution for intra-datacenter interconnects (DCI). In this paper, we show an experimental demonstration of a silica microsphere WGMR-based Kerr OFC light source where newly generated 400 GHz spaced carriers together with powerful linear equalization techniques, such as a linear symbol-spaced adaptive decision-feedback equalizer (DFE) with feed-forward (FF) and feedback (FB) taps, provide an alternative to individual lasers ensuring low-cost and low-complexity IM/DD scheme for the transmission of NRZ-OOK modulated signals at data rates up to 50 Gbps/λ over 2 km SMF link. Finally, we demonstrate a record 50 Gbps per λ transmission of NRZ-OOK modulated signals with a novel silica microsphere WGMR-based Kerr-OFC as a light source operating in the optical C-band, surpassing the previously demonstrated data rate record by five times.

Machine Learning Applications for Short Reach Optical Communication

With the rapid development of optical communication systems, more advanced techniques conventionally used in long-haul transmissions have gradually entered systems covering shorter distances below 100 km, where higher-speed connections are required in various applications, such as the optical access networks, inter- and intra-data center interconnects, mobile fronthaul, and in-building and indoor communications. One of the techniques that has attracted intensive interests in short-reach optical communications is machine learning (ML). Due to its robust problem-solving, decision-making, and pattern recognition capabilities, ML techniques have become an essential solution for many challenging aspects. In particular, taking advantage of their high accuracy, adaptability, and implementation efficiency, ML has been widely studied in short-reach optical communications for optical performance monitoring (OPM), modulation format identification (MFI), signal processing and in-building/indoor optical wireless communications. Compared with long-reach communications, the ML techniques used in short-reach communications have more stringent complexity and cost requirements, and also need to be more sensitive. In this paper, a comprehensive review of various ML methods and their applications in short-reach optical communications are presented and discussed, focusing on existing and potential advantages, limitations and prospective trends.

A 50-GBaud QPSK Optical Receiver With a Phase/Frequency Detector for Energy-Efficient Intra-Data Center Interconnects

This paper describes the energy-efficient realization of a QPSK optical receiver (CoRX) for short-reach intra-datacenter interconnects based on analog coherent detection. The CoRX comprises inphase and quadrature channels for each polarization and a high-speed phase-frequency detector (PFD) that provides feedback to stabilize an optical local oscillator (LO) and maintain coherence with the received optical signal. Each receive (RX) channel consists of a transimpedance amplifier (TIA) based on a Cherry-Hooper emitter follower (CHEF). The electronic RX is implemented in a 130-nm SiGe HBT technology ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$f_{T} = 300$ </tex-math></inline-formula> GHz), consumes 534 mW of DC power for a total electrical RX energy efficiency of 5.34 pJ/bit, and occupies 2.8 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$mm^{2}$ </tex-math></inline-formula> . Electrical characterization of the CoRX on an FR-4 PCB assembly demonstrates operation up to 60 GBaud with a bit error rate (BER) of less than 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−12</sup> . A co-packaged optical/electrical CoRX assembly with a silicon photonic receiver is characterized using a commercial-off-the-shelf quadrature phase-shift keying (QPSK) transmitter for constellations up to 50 GBaud (100 Gbps) at BER below KP4-FEC ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2.2\times 10^{-4}$ </tex-math></inline-formula> ).

Scaling Laws for Unamplified Coherent Transmission in Next-Generation Short-Reach and Access Networks

International standardization bodies (IEEE and ITU-T) working on the evolution of transmission technologies are still considering traditional direct detection solutions for the most relevant short-reach optical link applications, that are Passive Optical Networks (PON) and intra-data center interconnects. Anyway, future jumps towards even higher bit rates per wavelength will require a complete paradigm shift, moving towards coherent technologies. In this paper, we thus study both analytically and experimentally the scaling laws of unamplified coherent transmission in the short-reach communications ecosystems. We believe that, given the extremely tight techno-economic constraints, such a revolutionary transition towards coherent in short-reach first requires a very detailed study of its intrinsic capabilities in largely extending the limitation currently imposed by direct detection systems. To this end, this paper focuses on the ultimate physical layer limitations of unamplified coherent systems in terms of bit rate and power budget. The main parameters of our performance estimation model are extracted through fitting with a set of experimental characterizations and later used as the starting point of a scaling laws study regarding local oscillator power, modulator-induced attenuation, bit rate, and maximum achievable power budget. The analytically predicted performance is then verified through transmission experiments, including a demonstration on a 37-km installed metropolitan dark fiber in the city of Turin. Our findings show that coherent detection without optical pre-amplification and using PM-QPSK can tolerate optical power budget (OPB) well above, for instance, the 29 dB imposed by the current PON standards even at extremely high raw bit rates up to 800 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$Gbps$</tex-math></inline-formula> . PM-16QAM, on the other hand, can provide up to 190 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$Gbps$</tex-math></inline-formula> at 29 dB OPB only if combined with soft FEC algorithms. Even higher bit rates are also shown for the less demanding power budget needed in intra-data centers links.

Reduced Dimensionality Multiphysics Model for Efficient VCSEL Optimization

The ICT scene is dominated by short-range intra-datacenter interconnects and networking, requiring high speed and stable operations at high temperatures. GaAs/AlGaAs vertical-cavity surface-emitting lasers (VCSELs) emitting at 850–980 nm have arisen as the main actors in this framework. Starting from our in-house 3D fully comprehensive VCSEL solver VENUS, in this work we present the possibility of downscaling the dimensionality of the simulation, ending up with a multiphysics 1D solver (D1ANA), which is shown to be capable of reproducing the experimental data very well. D1ANA is then extensively applied to optimize high-temperature operation, by modifying cavity detuning and distributed Bragg’s reflector lengths.

Silicon modulator design using a system-oriented methodology for high-speed data center interconnect PAM-4 applications

A Silicon (Si) modulator and a design methodology using a system-oriented numerical approach are proposed. The main contributions of this work are to provide a Silicon modulator as an alternative solution for intra-data center interconnect (DCI) applications and a single figure-of-merit (FOM) for the modulator design, which takes as input the main device parameters (modulation efficiency (Vπ×L), optical loss, and electro-optical bandwidth (f3dB)) and provides a global optical system optimization. By using a numerical approach, the modulator physical parameters are analyzed according to system performance metrics such as the required receiver sensitivity to reach a target bit error rate (BER), and the required transmitter launch power. A capacitive Silicon device and a 4-level pulse amplitude modulation (PAM-4) system simulator is presented. The results show an efficient modulator (Vπ=3.5V) with high bandwidth (f3dB>38GHz), which is promising for data centers solutions, where power consumption is critical and high-speed is required.

Intra-Datacenter Interconnects With a Serialized Silicon Optical Frequency Comb Modulator

Powerful warehouse-scale datacenters form the fabric of cloud computing. Efficient computing requires intra-datacenter interconnects with large capacity, low energy consumption, and high scalability. These unique goals present brand new scientific and technological challenges for short-reach optical communications. Silicon photonics (SiP) modulators are promising for intra-datacenter interconnects for its low energy consumption, with the superiority of low cost, small footprint, and complementary metal-oxide-semiconductor (CMOS) compatibility. While currently, the main research focus is to increase the modulation speed of a single SiP modulator for a large interconnect capacity, parallel schemes exhibit better scalability. In this article, we propose intra-datacenter interconnects using dense wavelength-division multiplexing (DWDM) techniques with a single SiP optical frequency comb modulator (OFCM) to achieve potential large capacity and high scalability. The SiP-OFCM consists of serially cascaded microring modulators (MRMs). The MRM based SiP-OFCM is intrinsically compatible with DWDM, performing simultaneously data modulation for multiple optical carriers, DWDM demultiplexing and multiplexing functionalities. We demonstrate an interconnect line rate of 400 Gbit/s with a PAM 4 modulation format using the SiP-OFCM. All four DWDM channels exhibit bit-error ratios below the 33% hard-decision forward-error correction threshold after 2-km single-mode fiber transmission.

100G PAM-6 and PAM-8 Signal Transmission Enabled by Pre-Chirping for 10-km Intra-DCI Utilizing MZM in C-band

In this paper, we experimentally demonstrate the transmission of 100G PAM-6 and PAM-8 signals for 10-km intra-data center interconnects (intra-DCI) in C-band, which is enabled by pre-chirping technique utilizing a commercial dual-arm Mach-Zehnder modulator (MZM). At the transmitter side, the use of pre-chirping technology can overcome frequency selective fading (FSF) caused by fiber dispersion (CD) in intensity modulation and direct detection (IM/DD) systems. The pre-chirping is realized by adjusting the bias voltage of dual-arm MZM operating at single driven state. At the receiver, Volterra Non-Linear Equalizer (VNLE) is used to eliminate nonlinear impairments caused by pre-chirping. With the joint pre-chirping and nonlinear elimination in single driven dual-arm MZM based IM/DD system, 45-Gbaud PAM-6 and 40-Gbaud PAM-8 signals are successfully transmitted over 10-km single mode fiber (SMF) with a bit-error-ratio (BER) below 3.8 × 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-3</sup> . While in the chirp-free IM/DD system with dual-arm MZM operating at push-pull state, BERs of 45-Gbaud PAM-6 and 40-Gbaud PAM-8 signals over 10-km SMF transmission deteriorate dramatically to nearly 1 × 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> . The experimental results show that our proposed pre-chirping scheme can effectively enhance the robustness of IM/DD systems against the CD-induced FSF, and this scheme is a promising candidate for 10-km 100G intra-DCI.

Multi-core fiber based coherent transceiver utilizing self-homodyne detection and ICA based channel equalization for optical intra-datacenter interconnects

We have proposed a cost-effective full-duplex datacenter inter-(or intra-) connection transmission scheme based on self-homodyne coherent detection and multi-core fiber. Independent component analysis (ICA) based channel equalization is applied to remove the requirement of training symbols (TSs) for enhancement of spectral efficiency. We validate in a proof-of-concept experiment using polarization division multiplexing (PDM-) 16QAM orthogonal frequency division multiplexing (OFDM) format over 10-km 7-core fiber in full-duplex way. The results show that the performances can below the forward error correction (FEC) threshold of 3.8 × 10−3. In addition, we investigate the effects of several parameters in the proposed scheme, including laser linewidth and distance mismatch. Finally, we experimentally compared the performance of the proposed scheme to that of employing the TSs for channel compensation to quantify the performance penalty resulting from the TSs removal.

Beyond 1 Tb/s Intra-Data Center Interconnect Technology: IM-DD OR Coherent?

We discuss technology options and challenges for scaling intra-datacenter interconnects beyond 1 Tb/s bandwidths, with focus on two possible approaches: pulse amplitude modulation (PAM)-based intensity modulation-direct detection (IM-DD) and baud-rate sampled coherent technology. In our studies, we compare the performance of various orders of PAM modulation (PAM4 to 8). In addition to these fixed PAM signaling options, a flexible PAM (FlexPAM) technique leveraging granularity in spectral efficiency (SE) is proposed to maximize link margin. For baud-rate sampled coherent technology, we propose a simplified digital signal processing (DSP) architecture to bring down power consumption of the coherent approach closer to that of IM-DD PAM. We also propose two new phase noise tolerant 2D coherent modulation formats to relax the laser linewidth requirement. In closing, a comparative study of fixed IM-DD PAM versus coherent polarization multiplexed-quadrature amplitude modulation (PM-QAM) is presented for a 1.6 Tb/s solution (200 Gb/s per dimension), with consideration of link loss/reach budget, power consumption, implementation complexity, as well as fan-out granularity.

High-Speed VCSEL-Based Transceiver for 200 GbE Short-Reach Intra-Datacenter Optical Interconnects

The soaring demand for higher speeds in datacenters to address the relentless growth of the global IP traffic places optical interconnects in the spotlight. In this manuscript, we present a high-speed optical transceiver for intra-datacenter connectivity. The transceiver is based on single-mode, single-polarization high-speed vertical-cavity surface-emitting lasers (VCSELs), a VCSEL driver chip, and a linear receiver. Following a step-by-step approach, we present the architectures, assembly processes, and experimental results from the different modules. More specifically, we demonstrate (1) a data transmission experiment at 80 Gb/s using PAM-4 (four-level Pulse Amplitude Modulation) modulation for a reach of up to 500 m by employing a single-mode VCSEL module, and (2) a full-link experiment proving up to 64 Gb/s per lane capacity using PAM-4 signaling of the VCSEL-based optical transceiver test vehicles in back-to-back configuration and up to 56 Gb/s for 500 m and 2 km transmission distances. The acquired experimental results verify the suitability of the optical transceiver for intra-datacenter interconnects’ applications.

Exploitation of wavelength, hardware, and path redundancies in fault-tolerant all-optical DCNs

Abstract Data center performance is affected by three main factors; bandwidth, latency, and reliability of intra-data center interconnection network. Bandwidth and latency are definitely improved by adopting optical technology for intra-data center communication, but fault tolerance of the corresponding optical networks has been raised less. Recently, we introduced two Torus-based, all-optical, and non-blocking networks, i.e. O-TF and O-FTF, addressing reliability of optical networks, and now, in this paper, to address the scalability problem, we propose a novel Optical Clos-based architecture which reduces minimum number of required wavelength channels, as well as, the switch size in each node. Moreover, we examine three different types of redundancies; including (a) wavelength (adopted in O-TF network), (b) hardware (adopted in O-FTF network), and (c) path (adopted Optical Clos)) redundancies, in terms of throughput drop under failures, implementation cost, optical power loss, electrical power, latency, topology, and wiring complexity. In addition to simulating networks utilizing generated traffic based on Microsoft data, realistic traffic has also been considered. Our simulation results confirm that in addition to improving power consumption and latency, O-TF, O-FTF, and O-Clos architectures lead to throughput increment up to 80%, 90%, and 70%, respectively, comparing to WaveCube at the presence of network failure, as result of adopting our proposed fault-tolerant method, named as RAOP.