TY - GEN

T1 - Influence of optical axons on the synaptic weights

AU - Hulea, Mircea

AU - Younus, Othman Isam

AU - Ghassemlooy, Zabih

AU - Rajbhandari, Sujan

PY - 2021/9/7

Y1 - 2021/9/7

N2 - Hardware implementation of the spiking neural networks benefits from parallel processing and transmission of the data. When the neural areas are distanced or are in relative motion, the optical wireless connections represent a reliable alternative to the hard-wiring. Recent research shows that optical axons (OAs) can be successfully used in optical neural networks (NN s) because of the increased tolerance to the optical signal fading to the network activity. This work experimentally evaluates the influence of the optical signal fluctuations caused by the variation of the optical link length and misalignment on the transmitted weights. When the activation threshold of the neuron is reached, the OA triggers the electronic synapses that generate spikes of variable duration according to the synaptic weights. The results show that the duration of the transmitted spike is not affected by the changes in the optical signal intensity when the optical path varies from 0.05 to 1.90 m and the optical misalignment is below 60°. Hence, the OAs can be used to connect simple spiking neural networks (SNN s) such as neuromorphic sensors and the main neural control units or to interconnect high complexity neural areas that control humanoid robots.

AB - Hardware implementation of the spiking neural networks benefits from parallel processing and transmission of the data. When the neural areas are distanced or are in relative motion, the optical wireless connections represent a reliable alternative to the hard-wiring. Recent research shows that optical axons (OAs) can be successfully used in optical neural networks (NN s) because of the increased tolerance to the optical signal fading to the network activity. This work experimentally evaluates the influence of the optical signal fluctuations caused by the variation of the optical link length and misalignment on the transmitted weights. When the activation threshold of the neuron is reached, the OA triggers the electronic synapses that generate spikes of variable duration according to the synaptic weights. The results show that the duration of the transmitted spike is not affected by the changes in the optical signal intensity when the optical path varies from 0.05 to 1.90 m and the optical misalignment is below 60°. Hence, the OAs can be used to connect simple spiking neural networks (SNN s) such as neuromorphic sensors and the main neural control units or to interconnect high complexity neural areas that control humanoid robots.

M3 - Conference contribution

BT - Influence of optical axons on the synaptic weights

ER -