A5068387703- Optical Network Technologies
- Neural Networks and Reservoir Computing
- Photonic and Optical Devices
- Advanced Memory and Neural Computing
- Advanced Photonic Communication Systems
- Semiconductor Lasers and Optical Devices
- Semiconductor Quantum Structures and Devices
- Advanced Fiber Laser Technologies
- Random lasers and scattering media
- Magneto-Optical Properties and Applications
- PAPR reduction in OFDM
- Advanced Optical Network Technologies
- Blind Source Separation Techniques
- Microwave Engineering and Waveguides
- Optical Coherence Tomography Applications
Aristotle University of Thessaloniki
2019-2023
University of Belgrade
2011-2019
The explosive growth of deep learning applications has triggered a new era in computing hardware, targeting the efficient deployment multiply-and-accumulate operations. In this realm, integrated photonics have come to foreground as promising energy technology platform for enabling ultra-high compute rates. However, despite photonic neural network layouts already penetrated successfully era, their rate and noise-related characteristics are still far beyond promise high-speed engines. Herein,...
Photonic artificial neural networks have garnered enormous attention due to their potential perform multiply-accumulate (MAC) operations at much higher clock rates and consuming significantly lower power chip real-estate compared digital electronic alternatives. Herein, we present a comprehensive consumption analysis of photonic neurons, taking into account global design parameters concluding analytical expressions for the neuron's energy- footprint efficiencies. We identify optimal...
We present an approach for the generation of adaptive sigmoid-like and PReLU nonlinear activation function all-optical perceptron, exploiting bistability injection-locked Fabry–Perot semiconductor laser. The profile can be tailored by adjusting side-mode order, frequency detuning input optical signal, Henry factor, or bias current. universal fitting both families functions is presented.
The relentless growth of Artificial Intelligence (AI) workloads has fueled the drive towards non-Von Neuman architectures and custom computing hardware. Neuromorphic photonic engines aspire to synergize low-power high-bandwidth credentials light-based deployments with novel architectures, surpassing performance their electronic counterparts. In this paper, we review recent progress in integrated neuromorphic analyze architectural hardware-based factors that limit performance. Subsequently,...
Abstract Neuromorphic photonics has relied so far either solely on coherent or Wavelength-Division-Multiplexing (WDM) designs for enabling dot-product vector-by-matrix multiplication, which led to an impressive variety of architectures. Here, we go a step further and employ WDM enriching the layout with parallelization capabilities across fan-in and/or weighting stages instead serving computational purpose present, first time, neuron architecture that combines optics towards multifunctional...
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 the brain-inspired computing primitives. transfer deep neural networks (DNNs) onto silicon (SiPho) architectures requires, however, analog engine that can perform tiled matrix multiplication (TMM) at line rate support DL a large number trainable parameters, similar approach followed by...
Neuromorphic photonics aims to transfer the high-bandwidth and low-energy credentials of optics into neuromorphic computing architectures. In this effort, photonic neurons are trying combine optical interconnect segments with that can realize all critical constituent functions, including linear neuron stage activation function. However, aligning new platform well-established neural network training models in order allow for synergy hardware best-in-class algorithms, following requirements...
Linear optics aim at realizing any real- and/or complex-valued matrix operator via optical elements, addressing a broad field of applications in the areas quantum photonics, microwave photonics and neural networks. The transfer linear operators into photonic experimental layouts typically relies on Singular Value Decomposition (SVD) techniques combining meshes cascaded 2x2 Mach Zehnder Interferometers (MZIs), with main challenges being precision representation targeted matrix, referred to as...
We review different technologies and architectures for neuromorphic photonic accelerators, spanning from bulk optics to photonic-integrated-circuits (PICs), assess compute efficiency in OPs/Watt through the lens of a comparative study where key technology aspects are analyzed. With an emphasis on PIC we shed light onto latest advances plasmonic modulation realization weighting elements training inference applications, present recently introduced scalable coherent crossbar layout. Finally,...
Abstract Non-von-Neumann computing architectures and deep learning training models have sparked a new computational era where neurons are forming the main architectural backbone vector, matrix tensor multiplications comprise basic mathematical toolbox. This paradigm shift has triggered race among hardware technology candidates; within this frame, field of neuromorphic photonics promises to convolve targeted algebraic portfolio along circuitry with unique speed, parallelization, energy...
Reprogrammable optical meshes comprise a subject of heightened interest for the execution linear transformations, having significant impact in numerous applications that extend from implementation switches up to neuromorphic computing. Herein, we review state-of-the-art approaches realization unitary transformations and universal operators photonic domain present our recent work field, allows fidelity restorable low-loss circuitry with single-step programmability. These advantages unlock new...
Neuromorphic photonics came to the fore promising neural networks (NNs) with orders of magnitude higher computational speeds compared electronic counterparts. In this direction, research efforts have been mainly concentrated on development spiking, convolutional and Feed-Forward (FF)-NN architectures, aiming solve complex cognitive problems. However, in order time-series classification prediction tasks, state-of-the-art deep-learning models require most cases employment Recurrent-NNs (RNNs)...
We experimentally demonstrate a coherent SiPho neuron that relies on EAM for both on-chip data generation and weighting. A record-high 32GMAC/s/axon compute rate an accuracy of 95.91% is reported, when the deployed as hidden layer MNIST classifier neural network.
The field of neuromorphic photonics has been projected to comprise the next-generation Neural Network platform, expected lead remarkable advances in compute energy- and area-efficiency metrics. Herein, we review performance state-of-the-art photonic demonstrators, summarizing impact circuit architecture employed weight technology on system credentials terms scalability, footprint-efficiency. We provide an overview a recently demonstrated crossbar multi-port interferometer, holding...
We present a comprehensive numerical model for intrinsic small-signal modulation response of both reflective, and traveling-wave semiconductor optical amplifiers. investigate the photon carrier density spatial distribution, response, -3 dB bandwidth uniform lossless current model. The analysis shows that does not significantly affect as long frequency is within bandwidth. One most important results our discovery maximum in case reflective amplifier operating with low to moderate input powers...
Linear optics aim at realizing any real- and/or complex-valued matrix operator via optical elements, addressing a broad field of applications in the areas quantum photonics, microwave photonics and neural networks. The transfer linear operators into photonic experimental layouts typically relies on Singular Value Decomposition (SVD) techniques combining meshes cascaded 2x2 Mach Zehnder Interferometers (MZIs), with main challenges being precision representation targeted matrix, referred to as...
We summarize recent developments in neuromorphic photonics, including our work and the advances it brings beyond state-of-the-art demonstrators terms of architectures, technologies, training models for a synergistic hardware/software codesign approach.
This paper presents a detailed numerical model of reflective and traveling-wave semiconductor optical amplifiers (SOAs), based on self-consistent iteration method. The method is fully transparent to the input parameters provides stable efficient convergence all relevant SOA variables as long sufficient number previous iterations taken into account. accounts for spectral carrier density dependence radiative recombination rate, material gain, refractive index, confinement factor. analysis...
Neuromorphic photonics has turned into a key research area for enabling neuromorphic computing at much higher data-rates compared to their electronic counterparts, improving significantly the (multiply-and-accumulate) MAC/sec. At same time, time-series classification problems comprise large class of artificial intelligence (AI) applications where speed and latency can have decisive role in hardware deployment roadmap, highlighting need ultra-fast implementations simplified recurrent neural...
We present a photonics platform targeting optical connectivity at the point of compute in high-power ASICs. The uses bias-controlled electro-absorption modulators and is differentiated by broad temperature stability coupled with high bandwidth density.
Reflective semiconductor optical amplifier fiber cavity lasers (RSOA-FCLs) are appealing, colorless, self-seeded, self-tuning and cost-efficient upstream transmitters. They of interest for wavelength division multiplexed passive networks (WDM-PONs) based links. In this paper, we compare RSOA-FCLs with alternative colorless sources, namely the amplified spontaneous emission (ASE) spectrum-sliced externally seeded RSOAs. We differences in output power, signal-to-noise ratio (SNR), relative...
We discuss recent advances in the field of neuromorphic photonics, presenting our work and perspective towards optimizing architecture, enabling technology Deep Learning training models through a hardware/software co-design co-development framework.
Neuromorphic computing has emerged as a highly-promising compute alternative, migrating from von-Neuman architectures towards mimicking the human brain for sustaining computational power increases within reduced consumption envelope. Electronic neuromorphic chips like IBM’s TrueNorth, Intel’s Loihi and Mythic’s AI platform reveal tremendous performance improvement in terms of speed density; at same time, photonic layouts are constantly gaining ground exploiting their large component...
Universal multiport interferometers that can be programmed to perform any unitary or linear transformation turn into an important building block for both classical and quantum photonics. These typically utilize the mathematical framework of U(2) matrix decomposition techniques comprise a mesh 2 × beam splitters phase shifters. All them are, however, inherently fidelity limited as their U(2)‐based deployment approach leads imbalanced path losses without supporting restoration mechanism....
We experimentally demonstrate distributed denial of service (DDOS) attack identification using Deep Learning over a photonic neuromorphic engine that supports both input signal and weight update at 50 GHz, reporting Cohen's κ-score 0.636.