Te chnical Brief

Introduction

Computer systems have used direct attached memory modules in the form of DIMMs for decades. These modules attach

directly to the host motherboard and to the CPU using a parallel bus. Expanding the number of modules a system can

utilize involves adding memory controllers and pins to the CPU which quickly becomes prohibitive.

As modern applications such as Artificial Intelligence (AI) and Machine Learning (ML), image processing, in-memory

databases and real time analytics consume more memory than ever before, a new solution is needed to keep up with the

demand for low latency, high bandwidth memory.

Compute Express Link™ (CXL™) is a high-bandwidth, low latency, CPU-to-Device interconnect standard. It builds on

existing PCI Express

(PCIe

) infrastructure, leveraging the PCIe physical and electrical interface while adding additional

transfer protocols. Memory expansion, memory pooling, including sharing, and system composability are usage models

enabled by CXL. CXL enables support for three types of devices:

• Type 1: Smart NICs & Accelerators

• Type 2: Accelerators with Cache

• Type 3: Memory

Due to the growing demand for high density and cost effective memory solutions, industry efforts are initially focusing on

memory expansion with Type 3 Memory modules. These modules can be built in several different emerging form factors

as described in this article. They include EDSFF (Enterprise and Datacenter Standard Form Factor) and AIC (Add-In-Card)

form factors.