Samsung has officially announced that it has begun mass production and shipping the industry's first commercially available HBM4 (High Bandwidth Memory 4) memory. This not only signifies that HBM technology has officially entered its fourth generation, but also demonstrates Samsung's determination to "overtake on the curve" through its process advantages.
This new product uses Samsung's most advanced 1c (sixth generation 10nm class) DRAM process and combines it with a 4nm logic base die, pushing the transmission speed to 11.7Gbps, and even up to 13Gbps.
Combining 4nm logic process technology, breaking down the boundaries between memory and computing.
The biggest technological breakthrough of Samsung HBM4 lies in the innovation of the "Base Die".
In the past, HBM was mainly handled by major memory manufacturers. However, in the HBM4 generation, in order to overcome bandwidth and performance bottlenecks, the underlying base chips began to use advanced logic processes. Samsung, leveraging its advantages of having both memory and wafer foundry capabilities, directly adopted a 4nm logic process to manufacture the base chips for HBM4.
Key specifications summarized:
• speed:Stable transmission speeds reach 11.7Gbps (46% faster than the industry standard 8Gbps), and can be overclocked to a maximum of 13Gbps.
• bandwidth:The single stack bandwidth reaches 3.3 TB/s, which is 2.7 times that of the previous generation HBM3E.
• Capacity:採用12層堆疊 (12Hi)技術，提供24GB至36GB容量；16層堆疊 (16Hi)最高可擴充至48GB。
• Energy efficiency and heat dissipation:Through low-voltage TSV and power distribution network (PDN) optimization, power consumption is reduced by 40% and heat dissipation performance is improved by 30%.
It's not just about speed, it's also about customization.
Samsung Executive Vice President Sang Joon Hwang stated that HBM4 differs from the traditional approach by directly incorporating 1c DRAM and a 4nm logic process, ensuring ample performance headroom.
Samsung also announced an ambitious timeline:
• HBM4:It is now in mass production and being shipped.
• HBM4E:Sample shipments are expected to begin in the second half of 2026.
• Custom HBM:It is expected to be available starting in 2027.
This demonstrates Samsung's active collaboration with global GPU manufacturers (NVIDIA and AMD) and hyperscale data center providers, leveraging the integration of internal packaging technology and wafer foundry services to shorten lead times and offer more flexible specifications. Samsung anticipates its HBM sales will more than triple in 2026 compared to 2025.
Analysis of viewpoints
This HBM war finally entered the "deep waters" with the 4th generation of HBM.
Previously, in the HBM3 and HBM3E generations, SK Hynix stole the show with its MR-MUF packaging technology, becoming NVIDIA's exclusive or preferred supplier and embarrassing Samsung. But with HBM4, the rules of the game changed—"logic process" became the key.
This is precisely Samsung's biggest competitive advantage. Compared to SK Hynix, which needs to partner with TSMC to produce basic chips, Samsung is one of only two IDM manufacturers globally (the other being Intel) that simultaneously possess top-tier memory and advanced logic process manufacturing capabilities. This theoretically gives Samsung greater autonomy in terms of HBM4 integration, yield control, and cost structure.
Samsung's announcement of "first mass production" carries a strong message. It signifies that Samsung's 1c DRAM process yield has stabilized, and 4nm logic integration is ready. For a market urgently needing higher bandwidth to power next-generation AI chips (such as NVIDIA Rubin or AMD MI400 series), having another supplier is definitely a good thing; and for Samsung, this is a crucial battle in its reclaiming of the "memory king" throne.
