SK Hynix continues to advance its memory technologies and has announced that it has begun sampling HBM4E memory chips. The company says it is on schedule and is shipping the first samples of its memory to customers. These are 12-layer memory modules, offering a total memory capacity of 48 GB. In short, this is the type of memory we can expect to find in the next generation of cutting-edge AI accelerators!
HBM4E: The First Samples Sent to SK Hynix Customers!
48 GB across twelve layers!
HBM4E represents the new standard for HBM-type memory. If all goes well, we should see it in action in future generations of cutting-edge AI accelerators. We’re talking about NVIDIA Rubin GPUs or AMD Instinct Mi500s, for example. In any case, this growth in the AI sector is putting pressure on memory manufacturers.
With these new modules, the South Korean company is offering 48 GB chips consisting of 12 layers, representing a 50% increase in density compared to HBM3E. In terms of speed, this memory can reach 16 Gbps per pin. Efficiency has also been improved—reportedly by 20% compared to the previous generation—while thermal resistance has increased by 17%. SK Hynix’s memory is said to use advanced MR-MUF technology to ensure structural stability. As a reminder, this involves multiple layers of memory stacked on top of one another.
In practice, this results in particularly dense and fast memory, which the AI sector values for the large volumes of data it handles. Furthermore, this announcement comes two weeks after Computex, where the company provided a preview of its memory. Samsung wasn’t far behind either, showcasing its own memory as well as its technology for HBM5.
