A High-Accuracy and Energy-Efficient CORDIC based Izhikevich Neuron

Abstract

Efficient hardware design of biological neuron models is an essential issue in neuromorphic computation research. This paper presents a high-accuracy and energy-efficient hardware design of Izhikevich neuron, in which a fast-convergence COordinate Rotation DIgital Computer (CORDIC) operating in linear system is proposed to calculate square function. A CORDIC error model is also proposed to analyze the error propagation and study the accuracy improvement in the Izhikevich neuron design. Utilizing the fast CORDIC instead of conventional CORDIC, redundant iterations and associated computation are removed, which contributes to both smaller errors and higher efficiency of square calculation. Hence, the proposed fast CORDIC based Izhikevich neuron exhibits higher accuracy and energy efficiency than the conventional CORDIC based design. The FPGA implementation results show that the proposed Izhikevich neuron design achieves 24.2% faster in neuron potential update, 40.7% error reduction, and 45.6% energy-efficiency improvement over the state-of-the-art method, respectively.