The Landscape of Compute-near-memory and Compute-in-memory: A Research and Commercial Overview

TL;DR

This survey addresses the data movement bottleneck by examining compute-near-memory (CNM) and compute-in-memory (CIM) paradigms, their underlying memory technologies, and a spectrum of architectures from academic prototypes to commercial solutions. It systematically categorizes CIM/CNM designs, explains device primitives (crossbars, CAMs, logic units), and reviews memory technologies (DRAM, SRAM, PCM, ReRAM, MRAM, FeFET) and their suitability for in-memory or near-memory computation. The paper also maps the evolving commercial landscape, listing startups and industrial efforts (UPMEM, Samsung, IBM, AxDIMM, PIM-HBM, etc.), and highlights open challenges such as programmability, software ecosystems, and validation tools. Collectively, it underscores the potential for dramatic gains in energy and performance for data-intensive tasks, particularly AI workloads, while acknowledging substantial hurdles before widespread adoption. The conclusions point to continued diversification of memory technologies, cross-layer toolchains, and hybrid architectures as paths to realizing practical, scalable CIM/CNM systems.

Abstract

In today's data-centric world, where data fuels numerous application domains, with machine learning at the forefront, handling the enormous volume of data efficiently in terms of time and energy presents a formidable challenge. Conventional computing systems and accelerators are continually being pushed to their limits to stay competitive. In this context, computing near-memory (CNM) and computing-in-memory (CIM) have emerged as potentially game-changing paradigms. This survey introduces the basics of CNM and CIM architectures, including their underlying technologies and working principles. We focus particularly on CIM and CNM architectures that have either been prototyped or commercialized. While surveying the evolving CIM and CNM landscape in academia and industry, we discuss the potential benefits in terms of performance, energy, and cost, along with the challenges associated with these cutting-edge computing paradigms.

Figures (27)

• Figure 1: (a) Conventional computing system where an operation $f$ is performed on data $D$ in the CPU (b) Memory-centric design where $f$ is computed in the proximity of $D$ and CPU is mainly working as a control unit cim_survey.
• Figure 2: High-level overview of systems where computation is performed a) COM (mainstream computing): outside of memory system, b) cnm: using a logic connected to the memory via the memory high-bandwidth channel, c) cim-p: in the memory peripheral circuitry, and d) cim-a: using memory cells within the memory array.
• Figure 3: cim and cnm classification.
• Figure 4: Typical dram system organization.
• Figure 5: Cell structures of various memory technologies