Design of arithmetic units for vector processors based on carbon-based technology

Abstract

Vector processors are widely used in high-performance computing (HPC) due to their flexibility and energy efficiency. However, their performance scaling is increasingly constrained as silicon technology approaches its physical limits. This paper investigates carbon-based arithmetic units as a promising alternative for vector processors. Through simulations, we demonstrate that CNT arithmetic unit achieves 33% lower latency than silicon at the 0.5 μm node. And at sub-10 nm technology nodes, CNT-based designs exhibit a 7× speedup compared to conventional silicon FinFET implementations. These results highlight CNT’s superior scaling characteristics and its strong potential as a next-generation computing technology.