Hardware - Software Co-Design Approach In Customizable Programmable Logic Based Neuromorphic System Design

Neuromorphic Computing offers energy efficient AI inference for Edge Computing as well as morphologically adaptive and complex machine learning systems inspired by the nature. Hardware-software codesign approach presents the optimum solution for performing neuromorphic computations. Hardware Software Codesign is an embedded systems approach that target sharing computational costs between the software and hardware with using system-on-chip (SoC) that contains general purpose microprocessors, FPGAs, ASIC cores, memory block periperals and interconnection buses in one chip.In this study, we determine the basic requirements of neuromorphic hardware systems and the problems in the hardware design. We use the hardware-software codesign approach to solve those problems. By comparing the HDL models of spiking neurons with a programmable spiking neuron designed with hardware-software codesign approach in terms of speed, timing and hardware cost, we demonstrate the superiority of the hardware-software co-design approach in neuromorphic computing over pure hardware or pure software solutions.