The explosive volume growth of deep-learning (DL) applications has triggered an era in computing, with neuromorphic photonic platforms promising to merge ultra-high speed and energy efficiency credentials with the brain-inspired computing primitives. The transfer of deep neural networks (DNNs) onto silicon photonic (SiPho) architectures requires, however, an analog computing engine that can perform tiled matrix multiplication (TMM) at line rate to support DL applications with a large number of trainable parameters, similar to the approach followed by state-of-the-art electronic graphics processing units. Herein, we demonstrate an analog SiPho computing engine that relies on a coherent architecture and can perform optical TMM at the record-high speed of 50 GHz. Its potential to support DL applications, where the number of trainable parameters exceeds the available hardware dimensions, is highlighted through a photonic DNN that can reliably detect distributed denial-of-service attacks within a data center with a Cohen’s kappa score-based accuracy of 0.636.
Modern IoT and 5G applications are driving the growth of Internet traffic and impose stringent requirements to datacenter operators for keeping pace with the increasing bandwidth and low-latency demands. At the same time, datacenters suffer from increasing number of interconnections dictating the deployment of novel architectures and high-radix switches. The ratification of 400 GbE standard is driving the market of optical transceivers nevertheless, a technology upgrade will be soon necessary to meet the tremendous traffic growth. In this paper, we present the development of 800 Gb/s and 1Tb/s optical transceivers migrating to 100 Gbaud per lane and employing wafer-scale bonding of InP membranes and InP-DHBT electronics as well as advanced co-packaging schemes. The InP membrane platform is also exploited for the development of novel ultra-fast optical space switches based on a modular architecture design for scaling to large number of I/O ports.
Existing transceiver technology inside data centers will soon reach its limits due to the enormous traffic growth rates driven by new, bandwidth-hungry applications. Efforts to develop the next generation of 800Gbps and 1.6Tbps transceivers for intra-DC optical interconnects have already kicked-off to address the demands in traffic, the exhaustion of the ports at the digital switches and the power consumption limitations inherent to the use of many lower capacity modules. The new generation of optical modules must also provide Terabit capacities at low cost, necessitating the use of high-volume manufacturing processes. TERIPHIC is an EU funded R and D project that aims at developing transceiver modules with up to 1.6 Tbps capacity over 16 lanes in duplex fiber and cost less than 1 € per Gbps for distances up to 2 km, utilizing PAM-4 modulation for 100Gbps per lane and high-volume production compatible transceiver designs. At the component level, TERIPHIC will rely on arrays of high-speed electronics, InP Externally Modulated Lasers (EMLs) and InP photodetectors, and at the integration level it will rely on a polymer photonic platform as a host motherboard, leveraging its flexibility and powerful toolbox. A summary of the progress on the TERIPHIC transceiver modules concept, both at the component level and integration level is presented in this paper.
P. Bakopoulos, K. Tokas, C. Spatharakis, I. Patronas, G. Landi, K. Christodoulopoulos, M. Capitani, A. Kyriakos, M. Aziz, D. Reisis, E. Varvarigos, E. Zahavi, H. Avramopoulos
The soaring traffic demands in datacenter networks (DCNs) are outpacing progresses in CMOS technology, challenging the bandwidth and energy scalability of currently established technologies. Optical switching is gaining traction as a promising path for sustaining the explosive growth of DCNs; however, its practical deployment necessitates extensive modifications to the network architecture and operation, tailored to the technological particularities of optical switches (i.e. no buffering, limitations in radix size and speed). European project NEPHELE is developing an optical network infrastructure that leverages optical switching within a software-defined networking (SDN) framework to overcome the bandwidth and energy scaling challenges of datacenter networks. An experimental validation of the NEPHELE data plane is reported based on commercial off-the-shelf optical components controlled by FPGA boards. To facilitate dynamic allocation of the network resources and perform collision-free routing in a lossless network environment, slotted operation is employed (i.e. using time-division multiple-access - TDMA). Error-free operation of the NEPHELE data plane is verified for 200 μs slots in various scenarios that involve communication between Ethernet hosts connected to custom-designed top-of-rack (ToR) switches, located in the same or in different datacenter pods. Control of the slotted data plane is obtained through an SDN framework comprising an OpenDaylight controller with appropriate add-ons. Communication between servers in the optical-ToR is demonstrated with various routing scenarios, concerning communication between hosts located in the same rack or in different racks, within the same or different datacenter pods. Error-free operation is confirmed for all evaluated scenarios, underpinning the feasibility of the NEPHELE architecture.
KEYWORDS: Signal detection, Optical interconnects, Pulse shaping, Modulation, Digital filtering, Singular optics, Data conversion, Electro optics, Single mode fibers, Digital signal processing, Nyquist pulse, Receivers, Modulators, Forward error correction
Faced with surging datacenter traffic demand, system designers are turning to multi-level optical modulation with direct
detection as the means of reaching 100 Gb/s in a single optical lane; a further upgrade to 400 Gb/s is envisaged through
wavelength-multiplexing of multiple 100 Gb/s strands. In terms of modulation formats, PAM-4 and PAM-8 are
considered the front-runners, striking a good balance between bandwidth-efficiency and implementation complexity. In
addition, the emergence of energy-efficient, high-speed CMOS digital-to-analog converters (DACs) opens up new
possibilities: Spectral shaping through digital filtering will allow squeezing even more data through low-cost, low-bandwidth
electro-optic components.
In this work we demonstrate an optical interconnect based on an EAM that is driven directly with sub-volt electrical
swing by a 65 GSa/s arbitrary waveform generator (AWG). Low-voltage drive is particularly attractive since it allows
direct interfacing with the switch/server ASIC, eliminating the need for dedicated, power-hungry and expensive
electrical drivers. Single-wavelength throughputs of 180 and 120 Gb/s are experimentally demonstrated with 60 Gbaud
optical PAM-8 and PAM-4 respectively. Successful transmission over 1250 m SMF is achieved with direct-detection,
using linear equalization via offline digital signal processing in order to overcome the strong bandwidth limitation of the
overall link (~20 GHz). The suitability of Nyquist pulse shaping for optical interconnects is also investigated
experimentally with PAM-4 and PAM-8, at a lower symbol rate of 40 Gbaud (limited by the sampling rate of the AWG).
To the best of our knowledge, the rates achieved are the highest ever using optical PAM-M formats.
Datacenter traffic is exploding. Ongoing advancements in network infrastructure that ride on Moore’s law are unable to
keep up, necessitating the introduction of multiplexing and advanced modulation formats for optical interconnects in order
to overcome bandwidth limitations, and scale lane speeds with energy- and cost-efficiency to 100 Gb/s and beyond. While
the jury is still out as to how this will be achieved, schemes relying on intensity modulation with direct detection (IM/DD)
are regarded as particularly attractive, due to their inherent implementation simplicity. Moreover, the scaling-out of
datacenters calls for longer transmission reach exceeding 300 m, requiring single-mode solutions.
In this work we advocate using 16-QAM sub-cycle Nyquist-SCM as a simpler alternative to discrete multitone (DMT),
but which is still more bandwidth-efficient than PAM-4. The proposed optical interconnect is demonstrated at 112 Gb/s,
which, to the best of our knowledge, is the highest rate achieved in a single-polarization implementation of SCM. Off-the-shelf
components are used: A DFB laser, a 24.3 GHz electro-absorption modulator (EAM) and a limiting photoreceiver,
combined with equalization through digital signal processing (DSP) at the receiver. The EAM is driven by a low-swing
(<1 V) arbitrary waveform generator (AWG), which produces a 28 Gbaud 16-QAM electrical signal with carrier frequency
at ~15 GHz. Tight spectral shaping is leveraged as a means of maintaining signal fidelity when using low-bandwidth
electro-optic components; matched root-raised-cosine transmit and receive filters with 0.1 excess bandwidth are thus
employed. Performance is assessed through transmission experiments over 1250 m and 2000 m of SMF.
We demonstrate a polarization-insensitive coupler interfacing multicore-fiber (MCF) to silicon waveguides. It comprises a 3D glass fanout transforming the circular MCF core-arrangement to linear and performing initial tapering, followed by a Spot-Size-Converter on the silicon chip. Glass waveguides are formed of multiple overlapped modification elements and appropriate offsetting thereof yields tapers with symmetric cross-section. The Spot-Size-Converter is an inverselytapered silicon waveguide with a tapered polymer overcladding where light is initially coupled, whereas phase-matching gradually shifts it towards the silicon core. Co-design of the glass fanout and Spot-Size-Converter obtains theoretical loss below 1dB for the overall Si-to-MCF transition in both polarizations.
New broadband applications are causing the datacenters to proliferate, raising the bar for higher interconnection speeds.
So far, optical board-to-board and rack-to-rack interconnects relied primarily on low-cost commodity optical
components assembled in a single package. Although this concept proved successful in the first generations of opticalinterconnect
modules, scalability is a daunting issue as signaling rates extend beyond 25 Gb/s. In this paper we present
our work towards the development of two technology platforms for migration beyond Infiniband enhanced data rate
(EDR), introducing new concepts in board-to-board and rack-to-rack interconnects.
The first platform is developed in the framework of MIRAGE European project and relies on proven VCSEL
technology, exploiting the inherent cost, yield, reliability and power consumption advantages of VCSELs. Wavelength
multiplexing, PAM-4 modulation and multi-core fiber (MCF) multiplexing are introduced by combining VCSELs with
integrated Si and glass photonics as well as BiCMOS electronics. An in-plane MCF-to-SOI interface is demonstrated,
allowing coupling from the MCF cores to 340x400 nm Si waveguides. Development of a low-power VCSEL driver with
integrated feed-forward equalizer is reported, allowing PAM-4 modulation of a bandwidth-limited VCSEL beyond 25
Gbaud.
The second platform, developed within the frames of the European project PHOXTROT, considers the use of
modulation formats of increased complexity in the context of optical interconnects. Powered by the evolution of DSP
technology and towards an integration path between inter and intra datacenter traffic, this platform investigates optical
interconnection system concepts capable to support 16QAM 40GBd data traffic, exploiting the advancements of silicon
and polymer technologies.
We demonstrate a flexible multi-format vector modulator for coherent passive optical network (PON) architectures based on a non-interferometric cascade of a Semiconductor Optical Amplifier (SOA) and an Electro-Absorption Modulator (EAM) that exhibits increased simplicity and high energy-efficiency at low cost. 12- and 16- Quadrature Amplitude Modulation (QAM) constellations are generated, achieving modulation efficiencies of 3.6 and 4 bits/symbol respectively. The presented results demonstrate successful transmission over 25 km of fiber below the Forward Error Correction (FEC) limit, with sufficient compatible loss-budgets, making the proposed modulator concept an attractive candidate for urban network deployments, where high user density demands spectrally efficient formats. In addition, we describe the operation principle of the proposed modulator in detail and finally, the complete set of the digital signal processing (DSP) functionalities and algorithms that follow the standard coherent detection scheme is given, emphasizing on novel methods for QAM signal demodulation with uncommon constellation diagrams.
KEYWORDS: Signal to noise ratio, Quadrature amplitude modulation, Amplitude modulation, Digital signal processing, Statistical analysis, Receivers, Modulation, Interference (communication), Signal processing, Error analysis
We present a novel non-data-aided algorithm that uses only the magnitude of the received signal for accurate estimation of the signal-to-noise ratio (SNR) in M-QAM optical coherent digital receivers. The Koay inversion method that only works with constant-modulus signals, is extended through analytically exact expressions so as to allow application to any multi-level, complex modulation scheme. Performance is evaluated via simulation for formats up to 64-QAM and is shown to be superior than the decision-directed error vector magnitude (EVM) method at low SNR, while outperforming schemes based on the method of moments at high SNR.
We present experimental results obtained in the frame of the QOMA project involving the design and development of a diode-pumped solid state (DPSS) Nd:YAG laser, operating at passively and actively Q-switched pulsed mode. Continuous wave (cw) and passively Q-switched operations are demonstrated obtaining 7.5 W and 2 mJ at 100 Hz, respectively with excellent beam quality using a multi-segmented (0.1%, 0.23%, 0.6% at Nd) Nd:YAG crystal. Comparisons with a uniform Nd:YAG rod reveals improvements of up to 64% in normalized optical conversion efficiency and 67% in output power for retaining the same beam quality. Active Q-switching operation was also obtained with the same multi-segmented crystal, demonstrating pulse energy of 1.20 mJ at 5 kHz repetition rate, preserving an almost Gaussian beam profile.
We present preliminary simulation and experimental results obtained in the frame of QOMA project funded by the European Space Agency (ESA), involving the design and development of a diode-pumped solid state (DPSS) Nd:YAG laser. The simulation results were obtained using the LASCAD software code, while the experimental results were obtained at the Laboratoire Charles Fabry (France) and the National Technical University of Athens (NTUA).
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.