High-performance end-to-end deep learning IM/DD link using optics-informed neural networks.

In this paper, we introduce optics-informed Neural Networks and demonstrate experimentally how they can improve performance of End-to-End deep learning models for IM/DD optical transmission links. Optics-informed or optics-inspired NNs are defined as the type of DL models that rely on linear and/or nonlinear building blocks whose mathematical description stems directly from the respective response of photonic devices, drawing their mathematical framework from neuromorphic photonic hardware developments and properly adapting their DL training algorithms. We investigate the application of an optics-inspired activation function that can be obtained by a semiconductor-based nonlinear optical module and is a variant of the logistic sigmoid, referred to as the Photonic Sigmoid, in End-to-End Deep Learning configurations for fiber communication links. Compared to state-of-the-art ReLU-based configurations used in End-to-End DL fiber link demonstrations, optics-informed models based on the Photonic Sigmoid show improved noise- and chromatic dispersion compensation properties in fiber-optic IM/DD links. An extensive simulation and experimental analysis revealed significant performance benefits for the Photonic Sigmoid NNs that can reach below BER HD FEC limit for fiber lengths up to 42 km, at an effective bit transmission rate of 48 Gb/s.

Temperature and wavelength drift tolerant WDM transmission and routing in on-chip silicon photonic interconnects.

We demonstrate a temperature and wavelength shift resilient silicon transmission and routing interconnect system suitable for multi-socket interconnects, utilizing a dual-strategy CLIPP feedback circuitry that safeguards the operating point of the constituent photonic building blocks along the entire on-chip transmission-multiplexing-routing chain. The control circuit leverages a novel control power-independent and calibration-free locking strategy that exploits the 2nd derivative of ring resonator modulators (RMs) transfer function to lock them close to the point of minimum transmission penalty. The system performance was evaluated on an integrated Silicon Photonics 2-socket demonstrator, enforcing control over a chain of RM-MUX-AWGR resonant structures and stressed against thermal and wavelength shift perturbations. The thermal and wavelength stress tests ranged from 27°C to 36°C and 1309.90 nm to 1310.85 nm and revealed average eye diagrams Q-factor values of 5.8 and 5.9 respectively, validating the system robustness to unstable environments and fabrication variations.

Unbiased Plasmonic-Assisted Integrated Graphene Photodetectors.

Photonic integrated circuits (PICs) for next-generation optical communication interconnects and all-optical signal processing require efficient (∼A/W) and fast (≥25 Gbs-1) light detection at low (

Programmable photonic neural networks combining WDM with coherent linear optics.

Neuromorphic photonics has relied so far either solely on coherent or Wavelength-Division-Multiplexing (WDM) designs for enabling dot-product or vector-by-matrix multiplication, which has led to an impressive variety of architectures. Here, we go a step further and employ WDM for enriching the layout with parallelization capabilities across fan-in and/or weighting stages instead of serving the computational purpose and present, for the first time, a neuron architecture that combines coherent optics with WDM towards a multifunctional programmable neural network platform. Our reconfigurable platform accommodates four different operational modes over the same photonic hardware, supporting multi-layer, convolutional, fully-connected and power-saving layers. We validate mathematically the successful performance along all four operational modes, taking into account crosstalk, channel spacing and spectral dependence of the critical optical elements, concluding to a reliable operation with MAC relative error [Formula: see text].

Adaptive sigmoid-like and PReLU activation functions for all-optical perceptron.

We present an approach for the generation of an adaptive sigmoid-like and PReLU nonlinear activation function of an all-optical perceptron, exploiting the bistability of an injection-locked Fabry-Perot semiconductor laser. The profile of the activation function can be tailored by adjusting the injection-locked side-mode order, frequency detuning of the input optical signal, Henry factor, or bias current. The universal fitting function for both families of the activation functions is presented.

Optical RAM and integrated optical memories: a survey.

The remarkable achievements in the area of integrated optical memories and optical random access memories (RAMs) together with the rapid adoption of optical interconnects in the Datacom and Computercom industries introduce a new perspective for information storage directly in the optical domain, enabling fast access times, increased bandwidth and transparent cooperation with optical interconnect lines. This article reviews state-of-the-art integrated optical memory technologies and optical RAM cell demonstrations describing the physical mechanisms of several key devices along with their performance metrics in terms of their energy, speed and footprint. Novel applications are outlined, concluding with the scaling challenges to be addressed toward allowing light to serve as both a data-carrying and data-storage medium.

Initializing photonic feed-forward neural networks using auxiliary tasks.

Photonics is among the most promising emerging technologies for providing fast and energy-efficient Deep Learning (DL) implementations. Despite their advantages, these photonic DL accelerators also come with certain important limitations. For example, the majority of existing photonic accelerators do not currently support many of the activation functions that are commonly used in DL, such as the ReLU activation function. Instead, sinusoidal and sigmoidal nonlinearities are usually employed, rendering the training process unstable and difficult to tune, mainly due to vanishing gradient phenomena. Thus, photonic DL models usually require carefully fine-tuning all their training hyper-parameters in order to ensure that the training process will proceed smoothly. Despite the recent advances in initialization schemes, as well as in optimization algorithms, training photonic DL models is still especially challenging. To overcome these limitations, we propose a novel adaptive initialization method that employs auxiliary tasks to estimate the optimal initialization variance for each layer of a network. The effectiveness of the proposed approach is demonstrated using two different datasets, as well as two recently proposed photonic activation functions and three different initialization methods. Apart from significantly increasing the stability of the training process, the proposed method can be directly used with any photonic activation function, without further requiring any other kind of fine-tuning, as also demonstrated through the conducted experiments.

4-channel 200 Gb/s WDM O-band silicon photonic transceiver sub-assembly.

We demonstrate a 200G capable WDM O-band optical transceiver comprising a 4-element array of Silicon Photonics ring modulators (RM) and Ge photodiodes (PD) co-packaged with a SiGe BiCMOS integrated driver and a SiGe transimpedance amplifier (TIA) chip. A 4×50 Gb/s data modulation experiment revealed an average extinction ratio (ER) of 3.17 dB, with the transmitter exhibiting a total energy efficiency of 2 pJ/bit. Data reception has been experimentally validated at 50 Gb/s per lane, achieving an interpolated 10E-12 bit error rate (BER) for an input optical modulation amplitude (OMA) of -9.5 dBm and a power efficiency of 2.2 pJ/bit, yielding a total power efficiency of 4.2 pJ/bit for the transceiver, including heater tuning requirements. This electro-optic subassembly provides the highest aggregate data-rate among O-band RM-based silicon photonic transceiver implementations, highlighting its potential for next generation WDM Ethernet transceivers.

Alignment tolerant, low voltage, 0.23†V.cm, push-pull silicon photonic switches based on a vertical pn junction.

In this paper, we present the design, fabrication and characterization of a carrier depletion silicon-photonic switch based on a highly doped vertical pn junction. The vertical nature of the pn junction enables the device to exhibit a modulation efficiency as high as 0.23 V.cm. Fast switching times of 60ps are achieved in a lumped configuration. Moreover, the process flow is highly tolerant to fabrication deviations allowing a seamless transfer to the 350 nm process node of a commercial complementary-metal-oxide semiconductor (CMOS) foundry. Overall, this work showcases the possibility of fabricating highly efficient carrier depletion-based silicon photonic switches using medium resolution lithography.

10 Gb/s optical random access memory (RAM) cell.

Optical random access memories (RAMs) have been conceived as high-bandwidth alternatives of their electronic counterparts, raising expectations for ultra-fast operation that can resolve the ns-long electronic RAM access bottleneck. However, experimentally demonstrated optical RAMs have been limited to up to 5 GHz only, failing to validate the speed advantages over electronics. In this Letter, we demonstrate the first all-optical RAM cell that performs both Write and Read functionalities at 10 Gb/s, reporting on a 100% speed increase compared to state-of-the-art optical RAM demonstrations. To achieve this, the proposed RAM cell deploys a monolithically integrated InP optical Flip-Flop and a Semiconductor optical amplifier-Mach-Zehnder Interferometer (SOA-MZI) On/Off switch configured to operate as a strongly saturated differentially-biased access gate. Error-free operation is demonstrated at 10 Gb/s for both Write and Read operations with 6.2 dB and 0.4 dB power, respectively, achieving the fastest reported RAM cell functionalities.

Aluminum plasmonic waveguides co-integrated with Si3N4 photonics using CMOS processes.

Co-integrating CMOS plasmonics and photonics became the "sweet spot" to hit in order to combine their benefits and allow for volume manufacturing of plasmo-photonic integrated circuits. Plasmonics can naturally interface photonics with electronics while offering strong mode confinement, enabling in this way on-chip data interconnects when tailored to single-mode waveguides, as well as high-sensitivity biosensors when exposing Surface-Plasmon-Polariton (SPP) modes in aqueous environment. Their synergy with low-loss photonics can tolerate the high plasmonic propagation losses in interconnect applications, offering at the same time a powerful portfolio of passive photonic functions towards avoiding the use of bulk optics for SPP excitation and facilitating compact biosensor setups. The co-integration roadmap has to proceed, however, over the utilization of fully CMOS compatible material platforms and manufacturing processes in order to allow for a practical deployment route. Herein, we demonstrate for the first time Aluminum plasmonic waveguides co-integrated with Si3N4 photonics using CMOS manufacturing processes. We validate the data carrying credentials of CMOS plasmonics with 25 Gb/s data traffic and we confirm successful plasmonic propagation in both air and water-cladded waveguide configurations. This platform can potentially fuel the deployment of co-integrated plasmonic and photonic structures using CMOS processes for biosensing and on-chip interconnect applications.

High-port low-latency optical switch architecture with optical feed-forward buffering for 256-node disaggregated data centers.

Departing from traditional server-centric data center architectures towards disaggregated systems that can offer increased resource utilization at reduced cost and energy envelopes, the use of high-port switching with highly stringent latency and bandwidth requirements becomes a necessity. We present an optical switch architecture exploiting a hybrid broadcast-and-select/wavelength routing scheme with small-scale optical feedforward buffering. The architecture is experimentally demonstrated at 10Gb/s, reporting error-free performance with a power penalty of <2.5dB. Moreover, network simulations for a 256-node system, revealed low-latency values of only 605nsec, at throughput values reaching 80% when employing 2-packet-size optical buffers, while multi-rack network performance was also investigated.

All-optical 10Gb/s ternary-CAM cell for routing look-up table applications.

We experimentally demonstrate the first all-optical Ternary-Content Addressable Memory (T-CAM) cell that operates at 10Gb/s and comprises two monolithically integrated InP Flip-Flops (FF) and a SOA-MZI optical XOR gate. The two FFs are responsible for storing the data bit and the ternary state 'X', respectively, with the XOR gate used for comparing the stored FF-data and the search bit. The experimental results reveal error-free operation at 10Gb/s for both Write and Ternary Content Addressing of the T-CAM cell, indicating that the proposed optical T-CAM cell could in principle lead to all-optical T-CAM-based Address Look-up memory architectures for high-end routing applications.

Silicon photonic 8 × 8 cyclic Arrayed Waveguide Grating Router for O-band on-chip communication.

We report an 8 × 8 silicon photonic integrated Arrayed Waveguide Grating Router (AWGR) targeted for WDM routing applications in O-band. The AWGR was designed for cyclic-frequency operation with a channel spacing of 10 nm. The fabricated AWGR exhibits a compact footprint of 700 × 270 µm2. Static device characterization revealed 3.545 dB maximum channel loss non-uniformity with 2.5 dB best-case channel insertion losses and 11 dB channel crosstalk, in good agreement with the simulated results. Successful data routing operation is demonstrated with 25 Gb/s signals for all 8 × 8 AWGR port combinations with a maximum power penalty of 2.45 dB.

Active plasmonics in WDM traffic switching applications.

With metal stripes being intrinsic components of plasmonic waveguides, plasmonics provides a "naturally" energy-efficient platform for merging broadband optical links with intelligent electronic processing, instigating a great promise for low-power and small-footprint active functional circuitry. The first active Dielectric-Loaded Surface Plasmon Polariton (DLSPP) thermo-optic (TO) switches with successful performance in single-channel 10 Gb/s data traffic environments have led the inroad towards bringing low-power active plasmonics in practical traffic applications. In this article, we introduce active plasmonics into Wavelength Division Multiplexed (WDM) switching applications, using the smallest TO DLSPP-based Mach-Zehnder interferometric switch reported so far and showing its successful performance in 4×10 Gb/s low-power and fast switching operation. The demonstration of the WDM-enabling characteristics of active plasmonic circuits with an ultra-low power × response time product represents a crucial milestone in the development of active plasmonics towards real telecom and datacom applications, where low-energy and fast TO operation with small-size circuitry is targeted.

Two dimensional polymer-embedded quasi-distributed FBG pressure sensor for biomedical applications.

We report on the development of a flexible 2D optical fiber-based pressure sensing surface suitable for biomedical applications. The sensor comprises of highly-sensitive Fiber Bragg Grating elements embedded in a thin polymer sheet to form a 2x2 cm(2) sensing pad with a minimal thickness of 2.5mm, while it is easily expandable in order to be used as a building block for larger surface sensors. The fabricated pad sensor was combined with a low physical dimension commercially available interrogation unit to enhance the portability features of the complete sensing system. Sensor mechanical properties allow for matching human skin behavior, while its operational performance exhibited a maximum fractional pressure sensitivity of 12 MPa(-1) with a spatial resolution of 1x1cm(2) and demonstrated no hysteresis and real time operation. These attractive operational and mechanical properties meet the requirements of various biomedical applications with respect to human skin pressure measurements, including amputee sockets, shoe sensors, wearable sensors, wheelchair seating-system sensors, hospital-bed monitoring sensors.