In-Memory Computing: Advances and Prospects

TLDR

In-memory computing (IMC) promises to reduce the high energy and latency costs of data movement and access, requiring integrative research across the computing stack. The study aims to analyze the tradeoffs of IMC, assess proposed approaches, and derive insights on its potential benefits, contextual dependencies, and required support across the computing stack. The discussion has spurred extensive research into IMC.

Abstract

IMC has the potential to address a critical and foundational challenge affecting computing platforms today -that is, the high energy and delay costs of moving data and accessing data from memory. In doing so, it takes a disruptive approach to computing systems that requires integrative research across the computing stack. This has appropriately ignited a wide range of research. It is necessary to understand the foundational tradeoffs arising from IMC to better grasp the challenges that must be addressed and evaluate the different approaches being proposed. From here, perspectives may be derived regarding the potential gains IMC can offer, the contexts and conditions on which these depend, and the support they require across the compute stack.