Yahui Zhang, Shuiying Xiang, Xingxing Guo, Yanan Han, Yuechun Shi, Xiangfei Chen, Genquan Han, Yue Hao, "Spiking information processing in a single photonic spiking neuron chip with double integrated electronic dendrites," Photonics Res. 11, 2033 (2023)
- Photonics Research
- Vol. 11, Issue 12, 2033 (2023)
Fig. 1. Operation principle, chip, and experimental setup of FP-SA neuron. (a) Illustration of a biological neuron model receiving excitatory and inhibitory stimulations at functional dendrites. (b) Microscope image of the FP-SA chip and the operation principle of the FP-SA neuron. (c) Experimental setup of the encoding system based on a single FP-SA neuron with double electronic dendrites. Source, voltage source and current source; FPGA, field programmable gate array; CH1, CH2, two channels of FPGA; Bias Tee 1, Bias Tee 2, bias tees; FP-SA, integrated Fabry–Perot laser with an intracavity saturable absorber; OI, optical isolator; VOA, variable optical attenuator; OC, optical coupler; PD, photodetector; OSC, oscilloscope; OSA, optical spectrum analyzer. Red line, electrical signal; blue line, optical signal.
Fig. 2. Optical spectra and spiking dynamics of the FP-SA neuron. (a) Optical spectra of the FP-SA neuron. (a1) Optical spectrum with I G = 50 mA V SA = 0 V I G = 91 mA V SA = − 3.8 V I G = 91 mA V SA = − 3.6 V I G = 100 mA V SA = − 3.6 V I G = 100 mA V SA = − 3.8 V I G V SA
Fig. 3. Frequency encoding in FP-SA neuron. (a1)–(a6) Excitatory stimulus and corresponding response in FP-SA with I G = 90 mA V SA = − 3.9 V I G = 90 mA V SA = − 3.9 V I G = 81 mA V SA = − 3.9 V
Fig. 4. Spatiotemporal encoding in FP-SA neuron. (a1)–(a4) Excitatory stimulus and corresponding response in FP-SA with I G = 65 mA V SA = − 4.6 V I G = 64 mA V SA = − 4.19 V I G = 61.6 mA V SA = − 3.97 V
Fig. 5. Schematic diagram of photonic SNN for the recognition of Iris data set. (a1) Photonic SNN based on frequency encoding. (a2) Photonic SNN based on spatiotemporal encoding. (b) Detailed connection of presynaptic neuron and postsynaptic neuron.
Fig. 6. (a1), (a2) Recognition accuracy of the Iris data set and (b1)–(c2) weight distribution of corresponding photonic SNNs. (a1)–(c1) Corresponding to the architecture of Fig. 5 (a1) with the frequency encoding; (a2)–(c2) corresponding to the architecture of Fig. 5 (a2) with the spatiotemporal encoding. (b1), (b2) Weight range is [− 1
Fig. 7. Recognition accuracy of the Iris data set in noisy photonic SNNs. (a) Corresponding to the architecture of Fig. 5 (a1) with the frequency encoding; (b) corresponding to the architecture of Fig. 5 (a2) with the spatiotemporal encoding.
|
Table 1. Two-Section Laser Parameters
Set citation alerts for the article
Please enter your email address
© Copyright 2018-2021 | Chinese Laser Press. All Rights Reserved 沪ICP备15018463号-20