Optical Axons for Electro-Optical Neural Networks.

Area of Science:

• Neuromorphic Engineering
• Bio-inspired Robotics
• Optical Computing

Background:

• Neuromorphic sensors convert analog signals to spiking frequencies for neurorobotics.
• Optical synapses improve spiking neural network performance but are sensitive to light intensity variations.
• Existing optical synapse limitations hinder systems with multiple neuromorphic sensors.

Purpose of the Study:

• To propose and experimentally verify optical axons with optically activated synapses using digital signals.
• To address the limitations of optical synapses in multi-sensor neuromorphic systems.
• To investigate the impact of optical intensity and misalignment on synaptic activation delay.

Main Methods:

• Developed optical axons with synapses activated by digital optical signals.
• Encoded synaptic weights using the energy of optically transmitted stimuli.
• Experimentally tested transmission distance, delay, and the effect of illuminance on axon delay.

Main Results:

• Demonstrated line-of-sight transmission up to 190 cm with an 8 μs delay for the proposed optical axon.
• Showed that optical intensity fluctuations and link misalignment cause synaptic activation delays.
• Developed a fitted model for axon delay versus illuminance with a 0.95 RMS similarity.

Conclusions:

• Optical axons offer a viable solution for connecting multiple neuromorphic sensors without the drawbacks of traditional optical synapses.
• The proposed system demonstrates robust optical data transmission for neurorobotic applications.
• The study provides a model to predict and understand axon delay based on environmental illuminance.