Marvell introduces industry´s first 2 nm custom SRAM for next-gen data centers

Marvell Technology, Inc. has unveiled the industry´s first 2 nanometer custom Static Random Access Memory (SRAM), aiming to advance the performance profile of data-intensive devices such as custom XPUs for cloud data centers and Artificial Intelligence clusters. The new SRAM leverages Marvell´s own custom circuitry and software alongside leading-edge 2 nm process technology, resulting in memory chips that achieve up to 6 gigabits of high-speed data storage. A critical breakthrough comes in the form of drastically reduced memory power consumption and die area at comparable memory densities, delivering tangible efficiency gains for large-scale compute environments.

This innovation marks yet another step in Marvell´s strategy to redefine memory hierarchies in accelerated computing infrastructure. The custom SRAM is part of a broader suite of memory solutions by the company. Marvell recently launched its Compute Express Link (CXL) technology, which is designed for seamless integration into custom silicon, enabling cloud servers to access terabytes of additional memory and supplementary computational capacity. In parallel, the company showcased a custom high-bandwidth memory (HBM) solution that boosts memory capacity by up to 33 percent, optimizing both spatial footprint and energy efficiency for packed silicon environments.

By bringing 2 nm custom SRAM to market, Marvell further entrenches itself as a leader in advanced semiconductor technologies targeting the most demanding enterprise workloads. The company´s commitment to continued innovation in both memory density and power savings reflects growing industry demand for more efficient, high-performance solutions for cloud-scale and Artificial Intelligence-driven applications. These memory advancements position Marvell to address next-generation requirements as data center infrastructure and accelerated compute clusters continue to evolve.