STDP-Based Unsupervised Spike Pattern Learning in a Photonic Spiking Neural Network With VCSELs and VCSOAs

Xiang, Shuiying; Zhang, Yahui; Gong, Junkai; Guo, Xingxing; Lin, Lin; Hao, Yue

doi:10.1109/jstqe.2019.2911565

Cited by 122 publications

(47 citation statements)

References 68 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Additionally, other works have numerically investigated different interconnectivity architectures between coupled VCSEL-Neurons (at telecom wavelengths) using PS for operation [43][44][45]. Further, theoretical works have also recently described the potentials of photonic neuronal models based on VCSELs with a saturable absorbing region in their structure and VCSELs in combination with vertical cavity semiconductor optical amplifiers, for different spiking processing tasks, including spiking memory, spike encoding, spike timing dependent plasticity and pattern recognition [27,[46][47][48][49].…”

Section: Techniquementioning

confidence: 99%

“…Stanford's Neurogrid [2], VCSELs as artificial neuronal models (referred from now onwards as VCSEL-Neurons) was proposed as early as 2010 [35] using different techniques for operation, such as polarization switching (PS) [36] and optical injection (OI) induced nonlinear dynamics [54,55]. Since then, multiple reports on VCSELs for ultrafast spiking photonic neuronal models have emerged [28][29][30][31][32] [45][46][47][48][49][50][51][52][53][54][55][56][57]. Moreover, VCSEL-Neurons operating at both short (e.g.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Toward Neuromorphic Photonic Networks of Ultrafast Spiking Laser Neurons

Robertson

Wade

Kopp

et al. 2020

IEEE J. Select. Topics Quantum Electron.

View full text Add to dashboard Cite

We report on ultrafast artificial laser neurons and on their potentials for future neuromorphic (brain-like) photonic information processing systems. We introduce our recent and ongoing activities demonstrating controllable excitation of spiking signals in optical neurons based upon Vertical-Cavity Surface Emitting Lasers (VCSEL-Neurons). These spiking regimes are analogous to those exhibited by biological neurons, but at sub-nanosecond speeds (>7 orders of magnitude faster). We also describe diverse approaches, based on optical or electronic excitation techniques, for the activation/inhibition of sub-ns spiking signals in VCSEL-Neurons. We report our work demonstrating the communication of spiking patterns between VCSEL-Neurons towards future implementations of optical neuromorphic networks. Furthermore, new findings show that VCSEL-Neurons can perform multiple neuro-inspired spike processing tasks. We experimentally demonstrate photonic spiking memory modules using single and mutually-coupled VCSEL-Neurons. Additionally, the ultrafast emulation of neuronal circuits in the retina using VCSEL-Neuron systems is demonstrated experimentally for the first time to our knowledge. Our results are obtained with off-the-shelf VCSELs operating at the telecom wavelengths of 1310 and 1550 nm. This makes our approach fully compatible with current optical network and data centre technologies; hence offering great potentials for future ultrafast neuromorphic laser-neuron networks for new paradigms in brain-inspired computing and Artificial Intelligence.

show abstract

Section: Techniquementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Toward Neuromorphic Photonic Networks of Ultrafast Spiking Laser Neurons

Robertson

Wade

Kopp

et al. 2020

IEEE J. Select. Topics Quantum Electron.

View full text Add to dashboard Cite

show abstract

“…Alternative methods are therefore desired for high-performance learning, recognition, and neuromorphic computing systems. By combining the high bandwidth and energy-efficiency of photonic devices, photonic neural networks have the potential to be faster than conventional neural networks while consuming less energy [7][8][9][10][11][12][13][14][15][16][17].…”

Section: Introductionmentioning

confidence: 99%

“…STDP was also achieved in different ways and devices including a single SOA, two SOAs and a single VCSOA [8,23,24]. Based on photonic STDP, many tasks such as learning and recognition are achieved in photonic neural network [8,13,23,25]. For instance, in 2015, Ren et al achieved desired outputs through training the photonic neural network based on STDP [8].…”

Section: Introductionmentioning

confidence: 99%

A modified supervised learning rule for training a photonic spiking neural network to recognize digital patterns

Zhang

Xiang

Guo

et al. 2021

Sci. China Inf. Sci.

Self Cite

View full text Add to dashboard Cite

Deterministic conversion rule for CNNs toefficient spiking convolutional neural networks SCIENCE CHINA Information Sciences 63, 122402 (2020); On-line learning from restricted training sets in multilayer neural networks Europhysics letters 51, 691 (2000); Network traffic classification based on ensemble learning and co-training Science in China Series F-Information Sciences 52, 338 (2009); Regional TEC modelling over Africa using deep structured supervised neural network Geodesy and Geodynamics 11, 367 (2020); SCIENCE CHINA Information Sciences

show abstract

“…Modulators combined with photodetectors [23,24], SOA-MZIs [19,38] and laser-cooled atoms with electro-magnetically induced transparency [25] are used to realize the nonlinear activation function in ANN. Meanwhile, lasers, e.g., micro-pillar lasers [36], two-section lasers with saturable absorber regions [26,27], vertical-cavity surface-emitting lasers (VCSELs) [34,35,37,40] and quantum dot (QD) lasers [39] are typically used to implement the functionality of spiking neurons in SNN. Owing to its unique optical properties in different states, phase change material (PCM) has emerged as an attractive alternative to provide in hardware both the basic integrate-andfire functionality of neurons and the plastic weighting operation of synapses [28][29][30][31].…”

Section: Introductionmentioning

confidence: 99%

All-Optical Spiking Neuron Based on Passive Microresonator

Xiang

Torchy

Guo

et al. 2020

J. Lightwave Technol.

View full text Add to dashboard Cite

Neuromorphic photonics that aims to process and store information simultaneously like human brains has emerged as a promising alternative for the next generation intelligent computing systems. The implementation of hardware emulating the basic functionality of neurons and synapses is the fundamental work in this field. However, previously proposed optical neurons implemented with SOA-MZIs, modulators, lasers or phase change materials are all dependent on active devices and quite difficult for integration. Meanwhile, although the nonlinearity in nanocavities has long been of interest, the previous theories are intended for specific situations, e.g., selfpulsation in microrings, and there is still a lack of systematic studies in the excitability behavior of the nanocavities including the silicon photonic crystal cavities. Here, we report for the first time a universal coupled mode theory model for all side-coupled passive microresonators. Attributed to the nonlinear excitability, the passive microresonator can function as a new type of all-optical spiking neuron. We demonstrate the microresonator-based neuron can exhibit the three most important characteristics of spiking neurons: excitability threshold, refractory period and cascadability behavior, paving the way to realize all-optical spiking neural networks.

show abstract

STDP-Based Unsupervised Spike Pattern Learning in a Photonic Spiking Neural Network With VCSELs and VCSOAs

Cited by 122 publications

References 68 publications

Toward Neuromorphic Photonic Networks of Ultrafast Spiking Laser Neurons

Toward Neuromorphic Photonic Networks of Ultrafast Spiking Laser Neurons

A modified supervised learning rule for training a photonic spiking neural network to recognize digital patterns

All-Optical Spiking Neuron Based on Passive Microresonator

Contact Info

Product

Resources

About