Frequency-switched photonic spiking neurons

Area of Science:

• Photonics
• Neuroscience
• Computer Science

Background:

• Biological neurons communicate via electrical spikes.
• Mimicking biological spiking with physical devices is crucial for neuromorphic computing.
• Vertical-cavity surface-emitting lasers (VCSELs) offer potential for high-speed optical signal generation.

Purpose of the Study:

• To propose and demonstrate a VCSEL-based approach for generating neuron-like spikes.
• To achieve ultra-fast spiking speeds and controllable temporal coding.
• To investigate methods for manipulating spiking generation delay for enhanced functionality.

Main Methods:

• Utilizing multi-frequency switching in VCSELs to generate temporal spiking sequences.
• Conducting numerical simulations and experimental validation of the spiking mechanism.
• Designing and implementing a spiking coding scheme with multi-frequency switching.
• Analyzing the influence of injection strength and frequency detuning on spiking delay.

Main Results:

• Achieved stable, sub-nanosecond pulse width neuron-like spikes, significantly faster than biological neurons.
• Experimentally generated a 20-symbol sequence at speeds up to 1 Gbps using a controllable spiking coding scheme.
• Demonstrated control over spiking generation delay up to 60 ns by manipulating frequency detuning, exceeding intensity-controlled delays.

Conclusions:

• The proposed VCSEL-neuron is a single physical device capable of generating spiking signals with variable time delays.
• Multi-frequency switching offers an additional dimension for controlling spiking generation, beneficial for spatial-temporal features in spiking neural networks.
• This technology paves the way for developing future photonic spiking neural networks.