Samsung Ships Industry-First HBM4E Memory For Next-Gen AI Chips
Shares of Samsung are surging after the South Korean chip maker revealed it has begun shipping samples of its 12-layer HBM4E to major customers around the globe, giving it temporary bragging rights over rivals Micron and SK hynix. The cutting edge memory chips will help fuel the next round of AI hardware with Samsung touting stable speeds of up to 14 gigabits per second (Gbps).
There's also headroom to push the pedal to the metal, with Samsung claiming performance is scalable to a blistering 16Gbps. At the top end, Samsung's HBM4E chips represent a better than 20% increase over HBM4, found in solutions like AMD's Instinct MI455X accelerator, with memory bandwidth calculated at up to 3.6 terabytes per second (TB/s) per stack.
"Following the successful mass production of HBM4, Samsung has once again demonstrated its distinct technological edge with HBM4E," said Sang Joon Hwang, Executive Vice President and Head of Memory Development at Samsung Electronics. "Through our advanced manufacturing capabilities and preemptive infrastructure investments, we will continue to drive the growth of the global AI memory market."
The announcement comes amid a feeding frenzy by big tech to capitalize on AI demand. Fast memory with lots of bandwidth gives large language models (LLMs) and next-generation AI systems something to feast on as AI computing and hypserscale infrastructure continue to evolve at a breakneck pace.
Samsung said its HBM4E is available in 48-gigabyte (GB) capacity, delivering more than a 30% uptick compared to the previous generation. It's also planning to make available 32GB (8-layer) and 64GB (16-layer) options.
In addition to more bandwidth for next-gen AI systems, Samsung says its HBM4E chips leverage optimized packaging structures to boost energy efficiency by 16% and thermal resistance by more than 14% versus the previous generation. That means better heat dissipation and, in theory, longer term reliability.
Following initial shipments, Samsung says it will coordinate mass shipment plans based on customer demand and roadmaps. While not specified, that undoubtedly includes NVIDIA for solutions like its Vera Rubin AI superchip.
