Former Intel CEO Pat Gelsinger bets on Nagare Labs to vault the memory wall

If you use AI accelerators, they may become five times more powerful and cheaper within a decade.

Nagare Labs, a startup founded by Georgia Tech professors, is developing a new memory technology called Z2RAM, which aims to bridge the gap between SRAM, DRAM, and NAND flash. Pat Gelsinger, former Intel CEO and partner at Playground Global, is investing in the company, believing Z2RAM can overcome the "memory wall" that limits computing performance. Z2RAM promises SRAM-like bandwidth and DRAM-like density using oxide semiconductors and transistor-based storage, potentially reducing chip count for AI workloads by up to five times. The technology is designed for integration via 3D hybrid bonding. While lab prototypes exist, Nagare Labs has secured $5 million in seed funding and is seeking an R&D fab partner. They anticipate sampling working memory tiles within four years and production shipments within a decade. Challenges include the lack of industry standards for integration and ensuring reliability with oxide semiconductors.

The development of Z2RAM signifies a potential paradigm shift in memory technology, addressing the critical bottleneck of memory bandwidth and capacity that hinders advancements in AI and other high-performance computing fields. By combining the strengths of existing memory types into a single, more scalable architecture, Nagare Labs could significantly lower the number of chips required for complex tasks, thereby reducing system costs and power consumption. This innovation is particularly timely given the exponential growth in AI demand, which has outstripped current memory capabilities.

The success of Z2RAM hinges on overcoming substantial technical and market adoption hurdles, including the establishment of industry standards for its integration and ensuring its long-term reliability. While the backing of prominent investors like Playground Global and the involvement of figures like Pat Gelsinger provide significant momentum, the journey from lab prototypes to mass-market deployment is long and historically fraught with failure for memory startups. The path forward will likely involve strategic partnerships with chipmakers and foundries to navigate the complexities of manufacturing and integration at scale.