Multi-Wavelength, Multi-Level Inputs for an All-Optical SOA-Based Neuron

Bin Shi,Bitao Pan,Nicola Calabretta,Ripalta Stabile

DOI: https://doi.org/10.1364/cleo_si.2021.sm1b.4

2021-01-01

Abstract:We demonstrate the first optical processing of up to 9-bit/symbol multi-level modulated channels on a complete all-optical SOA-based neuron, with an error of 0.08. A higher number of modulation levels and inputs can improve accuracy.

What problem does this paper attempt to address?

Orbital Angular Momentum Mode Demodulation with Neural Network-Assisted Coherent Nanophotonic Circuits

Jiafu Chen,Qingji Zeng,Canming Li,Zebin Huang,Peipei Wang,Wenjie Xiong,Yanliang He,Huapeng Ye,Ying Li,Dianyuan Fan,Shuqing Chen

DOI: https://doi.org/10.1016/j.optcom.2023.129433

IF: 2.4

2023-01-01

Optics Communications

Abstract:Vortex beam carrying orbital angular momentum (OAM) shows promising potential in enlarging communication capacity and improving modulation ability for its mode orthogonality. Although various OAM multiplexing schemes have been proposed to boost channel capacity density, few studies have focused on OAM shift keying (OAM-SK) communication because of the lack of rapid and low-latency mode demodulation technologies. In this study, we propose a neural-network-assisted coherent nanophotonic circuit (CNC) strategy for demodulating OAM modes and investigate its application in OAM-SK communication. The CNC comprises 380 Mach–Zehnder interferometers (MZIs) and achieves a unitary matrix transformation by simultaneously modulating the amplitude and phase of optical pulses via multi-light interference, where a neural network-assisted design algorithm is used to obtain the phase parameters in the CNC that satisfy the mapping relationship between feature information and OAM modes. Since the satisfied mapping relationship is performed in the optical domain, OAM modes can be demodulated at the speed of light. We show that this CNC can effectively identify the OAM modes ranging from -10 to +10, and the identification accuracy exceeds 90% under moderate turbulence intensity. As a proof of concept, we simulated a hexadecimal OAM-SK communication link, and the OAM signals under turbulent perturbations were successfully demodulated with an accuracy of 99.33%.
First demonstration of a two-layer all-optical neural network by using photonic integrated chips and SOAs

Bin Shi,Nicola Calabretta,Ripalta Stabile

DOI: https://doi.org/10.1049/cp.2019.1132

2019-01-01

Abstract:We demonstrate the first all-optical feed-forward neural networks constituted by two full photonic layers of neurons based on SOAs with Broadcast and Weighting protocol. The impact of optical nonlinear function is studied and signal processing with 10 Gbit/s binary sequences is presented which results in 0.21 error.
SOA-based All-optical Photonic Integrated Deep Neural Network with Stable Output Noise

Bin Shi,Nicola Calabretta,Ripalta Stabile

2022-01-01

Abstract:we experimentally emulate the OSNR evolution of the SOA-based integrated all-optical neuron with 7-channel WDM input to single output conversion, resulting in a stable output error <0.1 and providing noise compression. (C) 2022 The Author(s)
Emulation and Modelling of Semiconductor Optical Amplifier-Based All-Optical Photonic Integrated Deep Neural Network with Arbitrary Depth.

Bin Shi,Nicola Calabretta,Ripalta Stabile

DOI: https://doi.org/10.1088/2634-4386/ac8827

2022-01-01

Neuromorphic Computing and Engineering

Abstract:We experimentally demonstrate the emulation of scaling of the semiconductor optical amplifier (SOA) based integrated all-optical neural network in terms of number of input channels and layer cascade, with chromatic input at the neuron and monochromatic output conversion, obtained by exploiting cross-gain-modulation effect. We propose a noise model for investigating the signal degradation on the signal processing after cascades of SOAs, and we validate it via experimental results. Both experiments and simulations claim that the all-optical neuron (AON), with wavelength conversion as non-linear function, is able to compress noise for noisy optical inputs. This suggests that the use of SOA-based AON with wavelength conversion may allow for building neural networks with arbitrary depth. In fact, an arbitrarily deep neural network, built out of seven-channel input AONs, is shown to guarantee an error minor than 0.1 when operating at input power levels of −20 dBm/channel and with a 6 dB input dynamic range. Then the simulations results, extended to an arbitrary number of input channels and layers, suggest that by cascading and interconnecting multiple of these monolithically integrated AONs, it is possible to build a neural network with 12-inputs/neuron 12 neurons/layer and arbitrary depth scaling, or an 18-inputs/neuron 18-neurons/layer for single layer implementation, to maintain an output error <0.1. Further improvement in height scalability can be obtained by optimizing the input power.
InP Photonic Integrated Multi-Layer Neural Networks: Architecture and Performance Analysis

Bin Shi,Nicola Calabretta,Ripalta Stabile

DOI: https://doi.org/10.1063/5.0066350

IF: 5.6

2022-01-01

APL Photonics

Abstract:We demonstrate the use of a wavelength converter, based on cross-gain modulation in a semiconductor optical amplifier (SOA), as a nonlinear function co-integrated within an all-optical neuron realized with SOA and wavelength-division multiplexing technology. We investigate the impact of fully monolithically integrated linear and nonlinear functions on the all-optical neuron output with respect to the number of synapses/neuron and data rate. Results suggest that the number of inputs can scale up to 64 while guaranteeing a large input power dynamic range of 36 dB with neglectable error introduction. We also investigate the performance of its nonlinear transfer function by tuning the total input power and data rate: The monolithically integrated neuron performs about 10% better in accuracy than the corresponding hybrid device for the same data rate. These all-optical neurons are then used to simulate a 64:64:10 two-layer photonic deep neural network for handwritten digit classification, which shows an 89.5% best-case accuracy at 10 GS/s. Moreover, we analyze the energy consumption for synaptic operation, considering the full end-to-end system, which includes the transceivers, the optical neural network, and the electrical control part. This investigation shows that when the number of synapses/neuron is >18, the energy per operation is <20 pJ (6 times higher than when considering only the optical engine). The computation speed of this two-layer all-optical neural network system is 47 TMAC/s, 2.5 times faster than state-of-the-art graphics processing units, while the energy efficiency is 12 pJ/MAC, 2 times better. This result underlines the importance of scaling photonic integrated neural networks on chip.
Optical Implementation of Neural Learning Algorithms Based on Cross-gain Modulation in a Semiconductor Optical Amplifier

Qiang Li,Zhi Wang,Yansi Le,Chonghui Sun,Xiaojia Song,Chongqing Wu

DOI: https://doi.org/10.1117/12.2245976

2016-01-01

Abstract:Neuromorphic engineering has a wide range of applications in the fields of machine learning, pattern recognition, adaptive control, etc. Photonics, characterized by its high speed, wide bandwidth, low power consumption and massive parallelism, is an ideal way to realize ultrafast spiking neural networks (SNNs). Synaptic plasticity is believed to be critical for learning, memory and development in neural circuits. Experimental results have shown that changes of synapse are highly dependent on the relative timing of pre- and postsynaptic spikes. Synaptic plasticity in which presynaptic spikes preceding postsynaptic spikes results in strengthening, while the opposite timing results in weakening is called antisymmetric spike-timing-dependent plasticity (STDP) learning rule. And synaptic plasticity has the opposite effect under the same conditions is called antisymmetric anti-STDP learning rule. We proposed and experimentally demonstrated an optical implementation of neural learning algorithms, which can achieve both of antisymmetric STDP and anti-STDP learning rule, based on the cross-gain modulation (XGM) within a single semiconductor optical amplifier (SOA). The weight and height of the potentitation and depression window can be controlled by adjusting the injection current of the SOA, to mimic the biological antisymmetric STDP and anti-STDP learning rule more realistically. As the injection current increases, the width of depression and potentitation window decreases and height increases, due to the decreasing of recovery time and increasing of gain under a stronger injection current. Based on the demonstrated optical STDP circuit, ultrafast learning in optical SNNs can be realized.
All-optical NRZ Wavelength Conversion Based on a Single Hybrid III-V/Si SOA and Optical Filtering.

Yingchen Wu,Qiangsheng Huang,Shahram Keyvaninia,Andrew Katumba,Jing Zhang,Weiqiang Xie,Geert Morthier,Jian-Jun He,Gunther Roelkens

DOI: https://doi.org/10.1364/oe.24.020318

IF: 3.8

2016-01-01

Optics Express

Abstract:We demonstrate all-optical wavelength conversion (AOWC) of non-return-to-zero (NRZ) signal based on cross-gain modulation in a single heterogeneously integrated III-V-on-silicon semiconductor optical amplifier (SOA) with an optical bandpass filter. The SOA is 500 μm long and consumes less than 250 mW electrical power. We experimentally demonstrate 12.5 Gb/s and 40 Gb/s AOWC for both wavelength up and down conversion.
Artificial Neuron Based on Nonlinear Polarization Rotation in a Semiconductor Optical Amplifier

Qiang Li,Zhi Wang,Huiying Wang,Can Cui,Chongqing Wu

DOI: https://doi.org/10.1016/j.optcom.2018.11.074

IF: 2.4

2018-01-01

Optics Communications

Abstract:We experimentally demonstrated an optoelectronic artificial neuron model, which can perform sigmoid transfer function. The proposed optoelectronic circuit can mimic the behavior of continuous sigmoidal neuron based on the nonlinear polarization rotation (NPR) in a single semiconductor optical amplifier (SOA). Because the current-induced NPR is anti-parallel to the optical power-induced NPR, excitatory and inhibitory stimuli to the proposed artificial neuron can be implemented by optical injection and electrical modulation, respectively. The response of the proposed neuron to excitatory inputs and inhibitory inputs are related to the initial state of polarization (SOP), bias current and power of probe beam. By adjusting the initial SOP, the neuron can be more likely to be excited or inhibited.
Lossless Monolithically Integrated Photonic InP Neuron for All-Optical Computation

Bin Shi,Kristif Prifti,Eduardo Magalhaes,Nicola Calabretta,Ripalta Stabile

DOI: https://doi.org/10.1364/ofc.2020.w2a.12

2020-01-01

Abstract:We demonstrate a monolithically integrated SOA-based photonic neuron, including both the weighted addition and a wavelength converter with tunable laser as nonlinear function, allowing for lossless computation of 8 Giga operation/s with an 89% accuracy. (C) 2020 The Authors
Noise analysis of SOA-based All-optical Photonic Deep Neural Network with WDM input

Bin Shi,Bitao Pan,Nicola Calabretta,Ripalta Stabile

DOI: https://doi.org/10.1364/psc.2021.tu3b.4

2021-01-01

Abstract:We experimentally verify the noise modeling of SOA-based all-optical neural networks on the OSNR degradation of the neuron: The photonic neural network can scale up to 16×16 neurons per layer, resulting in NRMSE <0.10.
All-optical logic OR and AND gates for NRZ-OOK signals based on four-wave mixing in a silicon waveguide

peiran hu,yanqiao xie,xiaoyan wang,qiang yan,shiming gao

DOI: https://doi.org/10.1142/S0218863514500301

2014-01-01

Journal of Nonlinear Optical Physics & Materials

Abstract:In this paper, we report a scheme to achieve an all-optical multiple-function logic unit including OR and AND gates for 10Gb/s nonreturn-to-zero on-off-keying (NRZ-OOK) signals based on four-wave mixing in a silicon waveguide. The functions of the logic OR and AND gates are demonstrated simultaneously using 10-bit OOK sequences. The eye diagrams of the OR and AND signals are realized experimentally as the incident signals are both modulated by OOK sequences originating from 2(31)-1 pseudorandom binary sequences.
Simultaneous All‐optical Amplification, Inversion and Format Conversion Based on Self‐induced Nonlinear Effects in a Semiconductor Optical Amplifier

Yanqiao Xie,Ying Gao,Shiming Gao,Chester Shu

DOI: https://doi.org/10.1002/mop.26009

IF: 1.311

2011-01-01

Microwave and Optical Technology Letters

Abstract:We propose and experimentally demonstrate simultaneous all‐optical amplification, inversion and format conversion based on self‐induced nonlinear effects in a single semicondctor optcial amplifier. All these functions are verified for injected amplitude‐shift keying signals with variable extinction ratios from 5 to 12 dB. © 2011 Wiley Periodicals, Inc. Microwave Opt Technol Lett 53:1320–1323, 2011; View this article online at wileyonlinelibrary.com. DOI 10.1002/mop.26009
Noise and Scalability Investigation of SOA-based All-optical Photonic Deep Neural Network

Bin Shi,Bitao Pan,Nicola Calabretta,Ripalta Stabile

DOI: https://doi.org/10.1364/acpc.2021.t1e.4

2021-01-01

Abstract:We emulate and experimentally validate the scaling of SO A-based all-optical deep neural networks by accurately modeling the OSNR degradation in a chain of SOAs: The photonic neural network can scale up to 16 layers when using only 4 neurons/layer without notable accuracy degradation.
Deep Neural Network Through an InP SOA-Based Photonic Integrated Cross-Connect

Bin Shi,Nicola Calabretta,Ripalta Stabile

DOI: https://doi.org/10.1109/jstqe.2019.2945548

IF: 4.9

2019-01-01

IEEE Journal of Selected Topics in Quantum Electronics

Abstract:Photonic neuromorphic computing is raising a growing interest as it promises to provide massive parallelism and low power consumption. In this paper, we demonstrate for the first time a feed-forward neural network via an 8 x 8 Indium Phosphide cross-connect chip, where up to 8 on-chip weighted addition circuits are co-integrated, based on semiconductor optical amplifier technology. We perform the weight calibration per neuron, resulting in a normalized root mean square error smaller than 0.08 and a best case dynamic range of 27 dB. The 4 input to 1 output weighted addition operation is executed on-chip and is part of a neuron, whose non-linear function is implemented via software. A three feedback loop optimization procedure is demonstrated to enable an output neuron accuracy improvement of up to 55%. The exploitation of this technology as neural network is evaluated by implementing a trained 3-layer photonic deep neural network to solve the Iris flower classification problem. Prediction accuracy of 85.8% is achieved, with respect to the 95% accuracy obtained via a computer. A comprehensive analysis of the error evolution in our system reveals that the electrical/optical conversions dominate the error contribution, which suggests that an all optical approach is preferable for future neuromorphic computing hardware design.
WDM Weighted Sum in an 8x8 SOA-Based InP Cross-Connect for Photonic Deep Neural Networks

B. Shi,Nicola Calabretta,D. Bunandar,D. Englund,Ripalta Stabile

DOI: https://doi.org/10.1109/PS.2018.8751338

2018-01-01

Abstract:We demonstrate high fidelity weighted addition of four 10 Gb/s on-off keyed data channels on a SOA-based monolithically integrated cross-connect, with a reading error dispersion below 0.2 with respect to the target. These results open the route to feasible 8x8 one-layer neuron interconnectivity for photonic integrated deep neural networks.
Optical Signal Processing Based on Semiconductor Optical Amplifier and Tunable Delay Interferometer

Xiaofan Zhao,Caiyun Lou,Yanming Feng

DOI: https://doi.org/10.1007/s12200-011-0143-z

2011-01-01

Frontiers of Optoelectronics in China

Abstract:In this paper, several applications in all-optical signal processing based on a semiconductor optical amplifier (SOA) and variable delayed interferometers (DIs) have been experimentally demonstrated. Wavelength converter based on a nonlinear polarization switch (NPS) and a DI is proposed and presented for the wavelength conversion of nonreturn-to-zero (NRZ) signals. An all-optical nonreturn-to-zero to return-to-zero (NRZ-to-RZ) format converter with tunable duty cycles is achieved by the DI with variable delays. The 40 Gb/s reconfigurable optical OR/NOR gate in a single SOA, followed a tunable optical bandpass filter (OBF) and a DI, optical 2R regeneration using an SOA-DI are investigated. It is found that this combinative realization of filters has been endowed with great flexibility and quality for 40 Gb/s optical logic and 2R regeneration.
Neuromorphic Computing Based on Wavelength-Division Multiplexing

Xingyuan Xu,Weiwei Han,Mengxi Tan,Yang Sun,Yang Li,Jiayang Wu,Roberto Morandotti,Arnan Mitchell,Kun Xu,David J. Moss

DOI: https://doi.org/10.1109/jstqe.2022.3203159

IF: 4.9

2023-03-01

IEEE Journal of Selected Topics in Quantum Electronics

Abstract:Optical neural networks (ONNs), or optical neuromorphic hardware accelerators, have the potential to dramatically enhance the computing power and energy efficiency of mainstream electronic processors, due to their ultra-large bandwidths of up to 10s of terahertz together with their analog architecture that avoids the need for reading and writing data back-and-forth. Different multiplexing techniques have been employed to demonstrate ONNs, amongst which wavelength-division multiplexing (WDM) techniques make sufficient use of the unique advantages of optics in terms of broad bandwidths. Here, we review recent advances in WDM-based ONNs, focusing on methods that use integrated microcombs to implement ONNs. We present results for human image processing using an optical convolution accelerator operating at 11 Tera operations per second. The open challenges and limitations of ONNs that need to be addressed for future applications are also discussed.

engineering, electrical & electronic,optics,physics, applied,quantum science & technology
Neuromorphic computing using wavelength-division multiplexing

Xingyuan Xu,Weiwei Han,Mengxi Tan,Yang Sun,Yang Li,Jiayang Wu,Roberto Morandotti,Arnan Mitchell,Kun Xu,David J. Moss

DOI: https://doi.org/10.1109/JSTQE.2022.3203159

2022-09-08

Abstract:Optical neural networks (ONNs), or optical neuromorphic hardware accelerators, have the potential to dramatically enhance the computing power and energy efficiency of mainstream electronic processors, due to their ultralarge bandwidths of up to 10s of terahertz together with their analog architecture that avoids the need for reading and writing data back and forth. Different multiplexing techniques have been employed to demonstrate ONNs, amongst which wavelength division multiplexing (WDM) techniques make sufficient use of the unique advantages of optics in terms of broad bandwidths. Here, we review recent advances in WDM based ONNs, focusing on methods that use integrated microcombs to implement ONNs. We present results for human image processing using an optical convolution accelerator operating at 11 Tera operations per second. The open challenges and limitations of ONNs that need to be addressed for future applications are also discussed.

Optics,Emerging Technologies
Fully UV Modulated Artificial Synapses with Integrated Sensing, Storage and Computation

Shuyi Sun,Tao Zhang,Shuoqi Jin,Xinhua Pan,Jianguo Lu,Zhizhen Ye,Bin Lu

DOI: https://doi.org/10.1002/adfm.202401403

IF: 19

2024-01-01

Advanced Functional Materials

Abstract:Optical synaptic devices are considered a preferred substitute to electrical synapses for low-energy, high-efficiency neuromorphic computing (NC). However, most optical synapses irreversible optical response severely restricts their application scenarios. In this study, a fully photon-modulated synaptic device constructed from Si-doped beta-gallium oxide (beta-Ga2O3)/ZnO heterojunction is demonstrated. The device exhibits reversible conductance changes and mimics a range of synaptic behaviors, which are modulated by two different wavelengths (255 and 370 nm) of UV light. Owing to the device's plasticity, four types of logic functions including "AND", "OR", "NOR" and "NAND" are implemented in the device by optical stimulation. Furthermore, an artificial neural network (ANN) is simulated based on the device's excitatory and inhibitory characteristics, and an optical programming scheme is employed to enhance the network's performance, resulting in a high recognition rate of 95.5% for handwritten digits. The fully photon-modulated synapse development strategy and programming scheme, as presented in this work, advance the utilization of optical artificial synapses in ANN. In this work, a fully UV modulated artificial synapse is achieved through the Si-doped beta-Ga2O3/ZnO heterojunction. The device conductance response to light is non-volatile, which enables a range of synaptic plasticity. Furthermore, the application of logic gates and handwritten digit recognition are realized by the device, indicating its immense potential in neuromorphic computing (NC). image
Binarized Coherent Optical Receiver Based on Opto-Electronic Neural Network

Zhenming Yu,Xu Zhao,Sigang Yang,Hongwei Chen,Minghua Chen

DOI: https://doi.org/10.1109/jstqe.2019.2931251

IF: 4.9

2019-01-01

IEEE Journal of Selected Topics in Quantum Electronics

Abstract:The resolution of analog to digital converter (ADC) and the power consumption of digital signal processing (DSP) are becoming the bottleneck of optical communication system. To acquire and recover signals with low resolution ADCs, we propose a binarized coherent optical receiver based on opto-electronic neural network, which consists of three layers, the broadcast layer, the optical neural layer, and the electronic neural layer. The broadcast layer that composed of Mach-Zehnder interferometers (MZIs) is to duplicate and select the input optical signal into the opto-electronic neural network. The optical neural layer is constituted by specially designed structures to map the binarized weights into optical domain, which can be integrated on silicon together with the broadcast layer. The one-bit vertical resolution ADCs act as the binarized activation after the optical neural layer. This binarized coherent optical receiver is demonstrated in simulation of 50-Gb/s single polarization (SP) QPSK system and 100-Gb/s polarization multiplexing (PDM) QPSK system. The one-bit vertical resolution ADCs can be utilized to recover the complex modulation format signal successfully. The broadcast layer and the optical neural layer can be integrated based on silicon, which could increase the calculation speed and reduce the energy consumption of the total system.

Multi-Wavelength, Multi-Level Inputs for an All-Optical SOA-Based Neuron

Orbital Angular Momentum Mode Demodulation with Neural Network-Assisted Coherent Nanophotonic Circuits

First demonstration of a two-layer all-optical neural network by using photonic integrated chips and SOAs

SOA-based All-optical Photonic Integrated Deep Neural Network with Stable Output Noise

Emulation and Modelling of Semiconductor Optical Amplifier-Based All-Optical Photonic Integrated Deep Neural Network with Arbitrary Depth.

InP Photonic Integrated Multi-Layer Neural Networks: Architecture and Performance Analysis

Optical Implementation of Neural Learning Algorithms Based on Cross-gain Modulation in a Semiconductor Optical Amplifier

All-optical NRZ Wavelength Conversion Based on a Single Hybrid III-V/Si SOA and Optical Filtering.

Artificial Neuron Based on Nonlinear Polarization Rotation in a Semiconductor Optical Amplifier

Lossless Monolithically Integrated Photonic InP Neuron for All-Optical Computation

Noise analysis of SOA-based All-optical Photonic Deep Neural Network with WDM input

All-optical logic OR and AND gates for NRZ-OOK signals based on four-wave mixing in a silicon waveguide

Simultaneous All‐optical Amplification, Inversion and Format Conversion Based on Self‐induced Nonlinear Effects in a Semiconductor Optical Amplifier

Noise and Scalability Investigation of SOA-based All-optical Photonic Deep Neural Network

Deep Neural Network Through an InP SOA-Based Photonic Integrated Cross-Connect

WDM Weighted Sum in an 8x8 SOA-Based InP Cross-Connect for Photonic Deep Neural Networks

Optical Signal Processing Based on Semiconductor Optical Amplifier and Tunable Delay Interferometer

Neuromorphic Computing Based on Wavelength-Division Multiplexing

Neuromorphic computing using wavelength-division multiplexing

Fully UV Modulated Artificial Synapses with Integrated Sensing, Storage and Computation

Binarized Coherent Optical Receiver Based on Opto-Electronic Neural Network