VLSI implementation of a bio-inspired olfactory spiking neural network

Area of Science:

• Neuromorphic Engineering
• Artificial Intelligence
• Sensory Systems

Background:

• Electronic noses require efficient odor classification systems.
• Traditional methods often face challenges with power consumption and chip area.
• Neuromorphic computing offers a promising approach for bio-inspired sensory processing.

Purpose of the Study:

• To present a novel low-power neuromorphic spiking neural network (SNN) chip for odor classification in electronic nose systems.
• To leverage sub-threshold oscillation and onset-latency representation for reduced power and area.
• To implement a spike-timing-dependent plasticity learning rule for synaptic weight modification.

Main Methods:

• Developed a neuromorphic spiking neural network (SNN) chip utilizing sub-threshold oscillation and onset-latency.
• Employed a spike-timing-dependent plasticity learning rule for synaptic plasticity.
• Collected odor data using a commercial electronic nose (Cyranose 320) and normalized it.
• Trained and tested the SNN chip, comparing its performance against Support Vector Machine and K-Nearest Neighbor algorithms.

Main Results:

• The SNN chip achieved an average power consumption of approximately 3.6 μW with a 1-V power supply.
• The chip demonstrated a distinct output response for different odor inputs.
• The proposed SNN chip achieved a mean testing accuracy of 87.59% for odor classification.
• The SNN's binary output (high/low response) simplifies decision-making verification.

Conclusions:

• The developed low-power SNN chip is effective for odor classification in electronic nose applications.
• The chip's design efficiently reduces power consumption and chip area.
• The SNN chip demonstrates competitive classification performance compared to established algorithms.