Samsung Electronics has completed the logic chip design for its next-generation high-bandwidth memory (HBM4) and launched 4nm experimental production for its next-generation high-bandwidth memory (HBM4) and launched trial production in 4nm, marking a big step towards mass production in the second half of 2025. This updated schedule is approximately six months ahead of the original 2026 plan, which could allow Samsung to land major orders for Nvidia’s future GPU platforms.
According to available information, Samsung’s storage division has already completed the design of logic chips, while its foundry division is hard at work on 4nm testing. Production of HBM4 takes place in a specialized D1c facility, which uses an advanced 10nm DRAM process. Samsung even skipped the usual D1b development stage to speed things up.
Technical details released at ISSCC 2024 show that the HBM4 will deliver a significant increase in performance, with data transfer rates of up to 2TB/s, approximately 66% higher than the HBM3E. It will also support a 2048-bit interface running at 6.4 GT/s and increase the capacity up to 48 GB, which is 33% more than the current generation.
Meanwhile, SK Hynix, Samsung’s biggest rival in the HBM field, is also said to be accelerating the development of HBM4 to achieve the same 2025 target. Hanwha Investment & Securities analysts expect SK Hynix maintains its market share and may be the first to deliver HBM4 samples to customers.
Besides Nvidia, Samsung customizes HBM4 products for Microsoft and Meta. This memory technology plays a critical role in Nvidia’s upcoming “Rubin” GPU platform, due in 2026, which will feature eight HBM4 chips and build on TSMC’s 3nm process.
Samsung’s current HBM product line is already booming. Sales during the third quarter of 2024 increased more than 70% month over month. HBM3E products, including 8- and 12-layer versions, have entered mass production and are expected to account for about half of Samsung’s total HBM sales by the end of 2024.
Tech
