A Novel Nonlinear Function Evaluation Approach for Efficient FPGA Mapping of Neuron and Synaptic Plasticity Models.

Area of Science:

• Neuroscience
• Computer Engineering
• Artificial Intelligence

Background:

• Efficient hardware realization of spiking neural networks (SNNs) is crucial for large-scale neural system modeling.
• Field-Programmable Gate Arrays (FPGAs) offer key features for implementing complex neural models.
• Existing methods for approximating nonlinear functions in SNN hardware can be resource-intensive.

Purpose of the Study:

• To present a novel nonlinear function evaluation approach for efficient, low-cost digital implementation of SNNs.
• To leverage uniform piecewise linear segmentation for accurate approximation of nonlinear neuron and synaptic plasticity models.
• To demonstrate the hardware efficiency and performance of the proposed approach on FPGAs.

Main Methods:

• Developed a uniform piecewise linear segmentation method for nonlinear function approximation.
• Designed a high-speed, simple segment address encoder unit for efficient hardware mapping.
• Applied the approach to Izhikevich, FitzHugh-Nagumo, Hindmarsh-Rose neuron models, and a calcium-based synaptic plasticity model.
• Utilized FPGA synthesis and simulations to evaluate hardware performance.

Main Results:

• The proposed approach enables accurate approximation of nonlinear functions with minimal hardware cost on FPGAs.
• Case studies demonstrated hardware implementations producing responses similar to original models.
• Achieved significant improvements in resource utilization and maximum clock frequency compared to prior work.
• Successfully implemented complex 2D and 3D neuron models and a synaptic plasticity model.

Conclusions:

• The novel nonlinear function evaluation approach is highly effective for low-cost neuromorphic circuit design.
• FPGA implementation offers superior resource efficiency and speed for SNN hardware.
• This method provides a scalable and efficient solution for approximating complex biological neuron and plasticity models in hardware.