South Korea’s memory industry is accelerating its development of Compute Express Link (CXL), a next-generation memory interconnect technology that enables central processing units (CPUs) to scale memory freely without relying on costly graphics processing units (GPUs) or high-bandwidth memory (HBM).

Experts believe CXL could transform how massive data is handled in AI servers.

“CXL allows memory from various devices to be shared, making it a game-changing technology for server environments currently dominated by DDR DRAM,” said Kwon Seok-joon, a professor of chemical engineering at Sungkyunkwan University. “It not only improves speed but also enhances power efficiency and data stability.

The professor noted that CXL is evolving beyond a simple connectivity tool to become a comprehensive solution for next-generation memory infrastructure.

Nicknamed the “ultimate memory expander,” CXL addresses a fundamental constraint in server architecture.

Traditional DRAM requires DIMM slots, while GPUs and solid-state drives (SSDs) use PCIe slots.

Servers typically have only 8 to 16 DIMM slots, limiting scalability and causing bottlenecks when data loads spike.

CXL overcomes this by enabling DRAM-based memory modules to be installed in PCIe slots, allowing terabyte-level expansions and even flexible scaling to hundreds of terabytes.

Major server makers like Hewlett Packard Enterprise (HPE), Dell Technologies, and Lenovo are turning to CXL for precisely this reason.

It allows CPU-centric systems to expand memory resources significantly without the need for expensive GPU or HBM components.

Industry players describe CXL as the “new infrastructure of the AI era.”

Korean memory giants Samsung Electronics and SK hynix, along with U.S.-based Micron, have all entered the CXL memory module market.

The CXL Consortium was launched in 2019 to formalize collaboration in this space.

CPU makers Intel and AMD are also integrating CXL at the architectural level in their server processors.

Intel began supporting CXL versions 1.1 and 2.0 with its 4th-generation Xeon Scalable Processors, while AMD added support starting with its EPYC 9004 Genoa series.

With technological development advancing rapidly, market growth is expected to surge.

According to research firm Yole Intelligence, the global CXL market is projected to grow from $1.7 million in 2013 to $2.1 billion in 2026 and $15.8 billion in 2028.

[ⓒ 매일경제 & mk.co.kr, 무단 전재, 재배포 및 AI학습 이용 금지]