Fig. 1. (Color online) (a) Temporal output of the spike encoding based on the modeling-based photonic neuron. (b) Threshold-like response and (c) spike latency property of the modeling-based photonic neuron. © [2020] IEEE. Reprinted with permission from Ref. [39].

Fig. 2. (a) Schematic diagram of inhibitory neuron based on VSCEL-SA. Reprinted from Ref. [31]. (b) The results of inhibition in a photonic neuron. Reprinted with permission from Ref. [33]. © The Optical Society.

Fig. 3. (Color online) (a) Schematic diagram of all-optical exclusive OR (XOR) operator based on a single VCSEL-SA. (b) XOR output for different sets of inputs. (c) The inputs and outputs of XOR for two RZ sequences. Reprinted with permission from Ref. [34]. © The Optical Society.

Fig. 4. (Color online) (a) The experimental setup for spiking firing and inhibition of VCSEL-based neuron. © [2018] IEEE. Reprinted with permission from Ref. [30]. (b) Time series of fired spiking responses. Reprinted with permission from Ref. [26]. © The Optical Society. © [2017] IEEE. Reprinted with permission from Ref. [28]. (c) Time series of suppressed spiking response. © [2018] IEEE. Reprinted with permission from Ref. [30].

Fig. 5. (Color online) (a) The experimental setup for graded-potential-signaling-based neuromorphic processing applications with the optical neuron based on DFB, including (b, c) pattern recognition, (d, e) single-wavelength implementation of STDP, and (f, g) sound azimuth measurement^[37]. Reproduced with permission. © 2020 Springer Nature.

Fig. 6. (Color online) (a) The DFB-based spatiotemporal pattern recognition network with STDP learning module. The network output for patterns with (b) 3 and (c) 4 input branches, respectively. Reprinted with permission from Ref. [38]. © The Optical Society.

Fig. 7. (Color online) (a) Optical implementation of STDP. (b) The measured learning window of Optical STDP with different SOA driving current. Reprinted with permission from Ref. [41]. © The Optical Society.

Fig. 8. (Color online) (a) Schematic diagram of photonic STDP based on VCSOA. (b) The experimental measured output pulse train corresponding to the input pulse pairs with different time interval. (c) Simulated input pulse. (d) Simulated output pulse. (e) The calculated STDP curve. Pulse 1 (Pulse 2): the optical pulse injection beam; VODL: variable optical delay line, OC: optical coupler; Circulator: optical circulator; VCSOA: vertical-cavity semiconductor optical amplifier. Bias and TEC: The bias current and temperature controller for VCSOA; in the box means a bandpass filter. © [2018] IEEE. Reprinted with permission from Ref. [44].

Fig. 9. (Color online) (a) Schematic diagram of photonic SNN based on VCSELs and VCSOAs. (b) PST as a function of the learning cycle. (c) Synaptic weights evolution during the learing process. n photonic presynaptic neurons and one postsynaptic neuron are conneted with optical STDP synapses. VCSEL₁–VCSEL_n: photonic presynaptic neurons; VCSEL_n+1: photonic postsynaptic neuron; T: variable delay line; W_i (i = 1, 2, …, n): variable synaptic weight device connecting VCSEL_i and VCSEL_n+1; STDP array: optical STDP synapses realized by VCSOAs; C: optical coupler. The red dashed box represents the ex-situ approach for updating the synaptic weight. © [2019] IEEE. Reprinted with permission from Ref. [35].

Fig. 10. (Color online) (a) Schematic diagram of a photonic SNN based on VCSEL-SAs for the supervised spike sequence learning. (b-e) Illustration of the spike sequence learning of a typical run. © [2020] IEEE. Reprinted with permission from Ref. [48].

Fig. 11. (Color online) (a) Architecture of the proposed all-optical SNN. (b) An example of a pattern classification task. The network is trained with (b1) a clean character image, and then, the inference was tested with a set of (b2) noisy patterns. (c) Comparison of convergence performance for supervised learning with different I of VCSOA. (d) Accuracy rate of the trained network as a function of the noise strength of the optical digital character. © [2020] IEEE. Reprinted with permission from Ref. [39].

Fig. 12. (a) Schematic structure of a 2 × 2 photonic SNN architecture to detect the sound azimuth, and two PREs correspond to the right ear and left ear, respectively. (b) Responses of POST1 and POST2 when . (c) The calculated as a function of the for different weights ( ). Reprinted with permission from Ref. [49]. © The Optical Society.

Fig. 13. (Color online) (a) Schematic diagram of WTA based on VCSELs-SA. (b) The output of VCSEL_M,A,B-SA for WTA mechanism. (c) Schematic diagram of pattern recognition based on the WTA machine. (d) The inputs and results of pattern recognition. (e) Schematic diagram of max-pooling operation. (f) The results of max-pooling operation. © [2020] IEEE. Reprinted with permission from Ref. [50].

Fig. 14. (Color online) (a) Schematic diagram of associative learning and forgetting processes based on VCSELs and STDP. (b) The emulation of associative learning and forgetting processes. (c) Schematic diagram of pattern recall. (d) Complete and incomplete patterns of number 8 and 5 respectively, visualization initial and final outputs of number 8 and 5 respectively. (e) The change processes of synaptic weight for number 8 and number 5. © [2020] IEEE. Reprinted with permission from Ref. [51].

Fig. 15. (Color online) (a) The architecture of the optical convolution unit (OCU) by modulator arrays. (b) The transmission rate versus the modulation voltage of the single modulator. (c) An illustration of the serialization method. (d) The convolution results of MNIST-handwritten numbers and Fashion-MNIST data sets. Reprinted with permission from Ref. [52]. © The Optical Society.

Fig. 16. (Color online) (a) The conceptual layout of the optical patching scheme with optical delay lines and wavelength-division-demultiplexing (WDM). (b) The experimental setup of the proposed scheme. Delayed copies of the input waveforms corresponding to (c, d) digit 2 and (e, f) 4, respectively. Reprinted with permission from Ref. [53]. © The Optical Society.

Fig. 17. (Color online) Optical matrix computation and the application for polarization processing. (a) Special-purpose processors for optical matrix computing and polarization processing respectively. (b) Self-configuring example for the smart processors. © [2020] IEEE. Reprinted with permission from Ref. [55].

Fig. 18. (Color online) Experimental results for photonic polarization processor chip. (a) Polarization MIMO descrambler. (b) Polarization controller. (c) Polarization analyzer. Reprinted with permission from Ref. [58]. © The Optical Society.

Fig. 19. (Color online) Experimental results for self-configuring optical signal processor. (a) Multichannel optical switching. (b) Optical MIMO descramble. (c) Tunable optical filter. Reprinted with permission from Ref. [59]. © 2017 American Chemical Society.

Fig. 20. (Color online) (a) System design of the Four-channels RC based on MDC-VCSELs. (b) The virtual node states matrix for each channel. (c) The NMSE values of Four-channels RC system based on MDC-VCSELs as a function of bias current for Four-channels RC and One-channel RC, respectively. Reprinted with permission from Ref. [67]. © The Optical Society.

Fig. 21. (Color online) (a) The conceptual scheme of RC based on a semiconductor nanolaser (SNL) with delayed feedback. (b) The NMSE values of SNL-based RC system as a function of the k_d for different F and . © [2020] IEEE. Reprinted with permission from Ref. [69].

Fig. 22. (Color online) (a) Experimental setup of a dual-channels chaotic system with a phase-modulated Sagnac loop. (b) Architecture for reinforcement learning based on dual-channels laser chaos. (c) The CDR as a function of cycle for the CSL-MC system and for the PMSL-MC system. (d) The convergence cycle (CC), at which the CDR reaches 0.9, as a function of coupling strength for the CSL-MC system and the PMSL-MC system. (e) The CDR as a function of the number of cycles for dual-channels and one-channel in the PMSL-MC system. (f) The CC as a function of coupling strength for dual-channels and one-channel in the PMSL-MC system. Reprinted with permission from Ref. [73]. © The Optical Society.

Fig. 23. (Color online) (a) The experimental setup of three globally coupled DFB lasers. (b) A parallel architecture for photonic decision making of 8-armed bandit problem. (c, d) CC and delay concealment as a function of attenuation. (e) The adaptability of the strategy to dynamically changing environment. (f) The scalability to 16-armed problem. Reprinted from Ref. [74]. Copyright (2020) with permission from Chinese Laser Press.