Scalable Digital Neuromorphic Architecture for Large-Scale Biophysically Meaningful Neural Network With...

Area of Science:

• Neuroscience
• Computer Engineering
• Computational Neuroscience

Background:

• Implementing large-scale, biologically realistic neural networks is crucial for advancing neuromorphic systems.
• Existing methods face challenges in computational efficiency and hardware resource utilization for complex neuronal models.

Purpose of the Study:

• To develop a hardware-efficient, scalable, and real-time computing strategy for large-scale multi-compartment neuron (CMN) simulations.
• To enhance the biological realism and computational power of neuromorphic systems.

Main Methods:

• Utilized a hardware platform with four Altera Stratix III field-programmable gate arrays.
• Developed a cost-efficient multi-CMN model for detailed neuronal dynamics and morphology simulation.
• Proposed a scalable network-on-chip (NoC) architecture with a novel routing algorithm.

Main Results:

• Achieved a 56.59% enhancement in computational speed for single-CMN implementation compared to classical methods.
• Demonstrated efficient implementation of a large-scale spiking neural network with biophysically plausible dynamics.
• The proposed NoC architecture improved throughput and reduced computational latency.

Conclusions:

• The presented strategy offers an efficient model and architecture for large-scale biologically meaningful neural networks.
• Hardware synthesis results show low area utilization and high computational speed, supporting scalability.
• This work advances neuromorphic computing by enabling more realistic and powerful neural simulations.