Comparison and Selection of Spike Encoding Algorithms for SNN on FPGA

Area of Science:

• Neuromorphic Engineering
• Computational Neuroscience
• Artificial Intelligence

Background:

• Spiking Neural Networks (SNNs) process information via discrete spikes, necessitating efficient conversion between spike and real-value signals.
• Spike encoding algorithms are crucial for SNN performance, impacting encoding efficiency and overall network capabilities.

Purpose of the Study:

• To evaluate and compare four common spike encoding algorithms for SNNs.
• To provide insights into algorithm selection for neuromorphic implementations based on performance metrics.
• To propose a scoring method for optimizing spike encoding algorithm selection.

Main Methods:

• Evaluation of four spike encoding algorithms based on Field-Programmable Gate Array (FPGA) implementation results.
• Assessment criteria included calculation speed, resource consumption, accuracy, and anti-noiseability.
• Validation using two real-world applications to verify evaluation findings.

Main Results:

• The sliding window algorithm demonstrates lower accuracy, suitable for signal trend observation.
• Pulsewidth modulated-based and step-forward algorithms offer accurate signal reconstruction, except for square waves.
• Ben's Spiker algorithm effectively handles square wave signals, complementing other methods.

Conclusions:

• Algorithm characteristics and application ranges were summarized, guiding selection for different SNN tasks.
• The proposed scoring method aids in selecting appropriate spike encoding algorithms.
• Enhanced algorithm selection can significantly improve the encoding efficiency of neuromorphic SNNs.