A report from ZDNet South Korea describes an early stage 2 nm project underway at Samsung Semiconductor that targets custom logic dies for high bandwidth memory. Industry sources cited in the article say Samsung’s HBM development team is exploring ways to tailor these logic dies to individual product requirements rather than relying solely on standardized designs. The company is rumored to be evaluating foundry process nodes described as ‘as advanced as 2 nm’ for its next wave of HBM products, although the report notes that it is not yet clear whether engineers will use Samsung’s SF2 or SF2P variants.
The article states that Samsung’s sixth generation HBM line, referred to as HBM4, is expected to use a 4 nm process node that is likely drawn from the SF4 family. An anonymous company insider is quoted as saying that ‘Samsung Electronics is designing a custom logic die for HBM under the leadership of the custom SoC (system-on-chip) team newly established within the System LSI Business Division last year…We are building a portfolio ranging from 4 nm to 2 nm to respond to the needs of various customers.’ This indicates a strategic shift toward a broader process portfolio for memory logic that can be tuned for performance, power, and integration needs across different clients.
According to the ZDNet report, Samsung expects that next generation ultra high performance Artificial Intelligence accelerators will depend on the most advanced HBM modules available. The article says that ‘strong’ Artificial Intelligence enterprise industry demand is anticipated once 2 nm logic dies become viable, which is described as possibly post-2027 after the introduction of HBM4E, the seventh generation of Samsung’s HBM products. Together, these details suggest Samsung is positioning its custom HBM logic and process roadmap to capture future demand from cloud, data center, and accelerator vendors that are seeking tighter coupling between memory and compute.
