Processing near memory : a paradigm shift

Abstract

This B.Tech project embarked on an educational journey through the dynamic field of computational architectures, initiating with an in-depth study of Processing-in-Memory (PIM) systems and their challenges in the context of modern computing. A significant aspect of this learning process was the exploration of RISC-V architectures, which provided essential insights into the potential and capabilities of vector processors within Single Instruction, Multiple Data (SIMD) frameworks. The transition from PIM to SIMD, particularly to vector architectures, stemmed from the discovery of their superior efficiency, scalability, and broader application range. Throughout the project, a focus was placed on learning about the architectural nuances of SIMD and vector processors. This exploration unveiled their proficiency in managing large datasets and executing parallel operations, effectively addressing the energy and scalability constraints associated with data movement in PIM systems. The study underscored the benefits of SIMD in terms of energy efficiency and performance enhancement, especially in scenarios favoring sequential programming while harnessing parallel data processing. This educational shift towards SIMD architectures, especially in understanding RISC-V and its vector extensions, represents a significant stride in computational technology, highlighting the evolving landscape of efficient and adaptable data processing methods in the current era of data-driven computing.