Near memory computing for data-centric architectures

PI: Joachim Rodrigues (LU); co-PI: Oscar Gustafsson (LiU)

A promising technique to boost performance and energy efficiency in data-centric computing architectures is the integration of computation logic in the near proximity of on-chip memory. In this project, we propose hardware/software co-optimized near memory computing (NMC) techniques, having the advantage of being scalable for various architectures and applications (e.g., ML/AI or baseband processing). Our techniques will improve computation speed by several orders of magnitude at a negligible hardware cost, increase energy efficiency, be validated by hardware/software co-verification in a virtual platform, and will be confirmed with measurements of a RISC-V processor based silicon realization.

Project number: D11