NVIDIA Qualifies Samsung’s HBM3E Memory

Samsung Electronics has finally secured a high-bandwidth memory supply deal with NVIDIA. This follows the successful qualification of its new 12-layer HBM3E chips. The agreement is a significant victory for the Korean tech giant.

This deal ends months of speculation about Samsung’s ability to meet NVIDIA’s stringent performance standards. It officially places Samsung alongside SK hynix and Micron in NVIDIA’s elite supplier group.

Overcoming Past Hurdles to Secure the Deal

Samsung initially faced challenges entering NVIDIA’s supply chain. Earlier versions of its HBM3 product reportedly encountered thermal and performance issues. These problems were linked to its previous-generation DRAM technology.

The company then focused on an 8-layer HBM3E design. That product also failed to gain approval from NVIDIA. According to reports from KED Global, Samsung invested heavily to revise its DRAM designs specifically for this goal.

For Samsung, this supply agreement is about more than revenue. Industry analysts note it is a major point of technological pride. Recognition from NVIDIA confirms its HBM technology is back on track and competitive.

Future Prospects and the Road to HBM4

Initial HBM3E orders from NVIDIA are expected to be limited. This is because SK hynix and Micron already handle the bulk of the current supply. However, this qualification is a crucial foothold for future business.

The breakthrough positions Samsung strongly for the next generation of memory. The company is already developing HBM4 technology. Reports suggest it aims to be the first to achieve 11 Gbps speeds with these new stacks.

This will involve using advanced 1c DRAM technology and a 4nm logic die from Samsung Foundry. A successful HBM4 launch could expand its partnerships to include other tech leaders like AMD, Broadcom, and Google.

This deal marks a dramatic turnaround for Samsung’s memory division. The NVIDIA HBM3E qualification solidifies its position in the critical AI semiconductor market. It sets the stage for intense competition in the upcoming HBM4 era.

Info at your fingertips

What is HBM3E memory?

HBM3E is a type of high-performance memory used in AI accelerators and data center GPUs. It offers extreme bandwidth and speed, which is essential for training large AI models. NVIDIA’s latest GPUs rely heavily on this technology.

Why was Samsung struggling with NVIDIA before?

Samsung’s earlier HBM3 and 8-layer HBM3E products had thermal and performance issues. These problems were attributed to its older DRAM manufacturing technology. The new 12-layer stack uses a revised design that meets NVIDIA’s standards.

Who are NVIDIA’s other HBM suppliers?

SK hynix and Micron are NVIDIA’s other primary suppliers for HBM3E memory. Both companies have been key partners in supplying memory for NVIDIA’s AI GPUs. Samsung’s entry creates a three-supplier ecosystem.

What does this mean for the AI chip market?

Increased competition between Samsung, SK hynix, and Micron could lead to better products and pricing. It also diversifies NVIDIA’s supply chain, reducing potential bottlenecks. This is crucial for meeting the massive global demand for AI hardware.

Trusted Sources: KED Global, Reuters, Bloomberg