A Design of Homogeneous Multi-Core SoC with General-Purpose Floating-Point Processors
Qiao Yuan, Wenlin Pang, Yuhe Shi · 2023
In the era of big data and artificial intelligence, SoC design for computationally intensive applications are a huge challenge, especially for general-purpose processors where the underlying architecture is single-core in nature. This paper exploits customized general-purpose floating-point processors to build a homogeneous multi-core SoC. The general-purpose floating-point processor supports the IEEE 754 single-precision floating-point standard and has effective circuit implementation of matrix operations. The SoC integrates a micro-controller compatible with Microblaze instruction set and a floating-point arithmetic module composed of various functional IPs, which makes the presented homogeneous multi-core SoC programmable and flexible. The homogeneous multi-core SoC is built using a three-layer network-on-chip, and the general-purpose floating-point processor is mounted on the network-on-chip node. Finally, matrix multiplication mapping is implemented on this homogeneous multi-core SoC for performance evaluation.