Samsung has begun shipping initial samples of its HBM4E high-bandwidth memory to select global customers, marking a significant step toward mass production for next-generation AI accelerators and hyperscale data centers. The company is offering the first variant as a 12-layer stack with 48GB of capacity, while planning future iterations that include an 8-layer 32GB version and a larger 16-layer 64GB option.
Samsung ships initial 12-layer samples to global customers for AI accelerators.
The HBM4E product line sits within Samsung's broader memory portfolio, designed specifically to handle the intense data throughput required by modern artificial intelligence workloads. Samsung uses its latest manufacturing process to boost speed and capacity while keeping the existing design framework.
Spec comparison
- Layers: 12-layer (initial), with 8-layer and 16-layer variants planned
- Capacity: 48GB (initial), with 32GB and 64GB variants planned
- Pin Speed: Up to 16Gbps (scaling from 14Gbps)
- Bandwidth: Up to 3.6 TB/s per stack
- DRAM Process: Samsung's 1c DRAM process (6th-generation 10nm-class)
Key technical specifications for the initial 12-layer sample include pin speeds scaling up to 16Gbps, which represents a notable increase from the 14Gbps baseline of previous generations. This speed enhancement allows for bandwidth figures reaching up to 3.6 TB/s per stack. The memory utilizes Samsung's proprietary 1c DRAM process, identified as its sixth-generation 10nm-class technology, paired with a 4nm logic base die manufactured at Samsung Foundry.
Secondary details highlight significant improvements in energy efficiency and thermal management compared to the prior generation. Samsung reports that HBM4E achieves a 16% improvement in power consumption while reducing thermal resistance by more than 14%. The 16% improvement in energy efficiency and over 14% reduction in thermal resistance help manage heat in high-density AI accelerators and data centers.
Pricing information has not been disclosed at this stage, as the product remains in the sampling phase. Samsung indicated that mass production schedules will be determined based on customer requirements and the completion of optimization work following initial sample shipments. Samsung will begin mass production of HBM4E stacks according to specific customer schedules after sample shipments and optimization are complete.
Samsung is expanding its HBM4E lineup with 12-layer 48GB initial stacks, alongside planned 8-layer 32GB and 16-layer 64GB variants to meet varying capacity needs. HBM4E builds upon the previous generation by scaling pin speeds up to 16Gbps, increasing bandwidth to 3.6 TB/s per stack, and utilizing a 4nm logic base die from Samsung Foundry.



Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.