Hardware-efficient implementation of FitzHugh–Nagumo neural networks on FPGA: A scalable approach for neuromorphic system design

The primary objectives of neuromorphic engineering involve researching, simulating, and utilizing neural processes in the human brain. Our solution is designed to replicate a brain cell-based system inspired by nature, focusing on duplicating its functions. This study presents a model that accurately replicates a range of dynamic behaviors observed in the original FitzHugh–Nagumo (FHN) framework, a simplified mathematical model used to simulate neural spike activity. By modifying the nonlinear term and using reduced terms (reduced means reducing equation complexity), we aim to achieve high accuracy in matching and minimize computational errors. By using time-domain simulations and dynamic behavior analysis (i.e., how system states evolve over time), we validate the proposed model, demonstrating its exceptional precision and minimal error in replicating all features of the FHN model. As a proof of concept, we construct the hardware on an FPGA using Hardware Description Language (HDL), a programming method used to define electronic circuits. The results of the FPGA deployment demonstrate that the suggested model utilizes merely 0.19% of the resources accessible on a Zynq FPGA platform, a type of programmable chip developed by Xilinx. Furthermore, the results of the static timing analysis, an assessment of the circuit's timing performance without simulating signal transitions, demonstrate that the circuit is capable of functioning at a peak synthesis frequency of 597.118 MHz. Additionally, we propose a practical digital hardware strategy for implementing large-scale neuron models, incorporating techniques such as FSM flattening, parallelization, operation chaining with retiming, resource sharing, dynamic precision scaling, and related optimizations. The synthesis frequency in this case for 300 neurons along with astrocyte (a type of brain cell that supports neuron communication and helps regulate signal flow between connected neurons) and synapse (the points where neurons communicate by transmitting electrical or chemical signals) connections is 406.721 MHz.