Nagare Labs, a startup specializing in next-generation memory technology, has officially launched following the completion of a $5 million funding round led by Playground Global. Founded in 2026 by Georgia Institute of Technology professors Suman Datta and Shimeng Yu, the company aims to address memory bottlenecks in computing systems with its innovative Z2RAM technology.
Z2RAM offers SRAM-class speed at DRAM-class density and is designed to be integrated directly onto AI accelerators. This configuration is intended to significantly enhance memory bandwidth speeds in these systems. According to Nagare Labs, while on-die SRAM provides high bandwidth, its scalability is limited and often requires excessive logic die area. Conversely, off-die DRAM and High Bandwidth Memory (HBM) solutions, despite offering greater capacity, suffer from increased latency and energy inefficiencies due to the data transfer processes involved.
To combat these issues, Nagare Labs has developed a unique, transistor-based capacitor-less architecture for Z2RAM. This approach combines the speed of SRAM with the density of DRAM into one memory tier. The company noted that traditional HBM setups involve multiple DRAM dies bonded together beside the AI accelerator, necessitating extensive interconnect pathways. Z2RAM, however, simplifies this by utilizing a single memory die with six vertically deposited layers, which can be directly integrated on top of the accelerator.
The architecture reduces heat generation and offers an efficient path for managing thermal output. Z2RAM’s vertical scalability uses standard 300mm processes, allowing for increased capacity without needing to shrink node sizes or employ extreme ultraviolet lithography (EUV).
Data in Z2RAM is stored exclusively in transistors, supporting fast, non-destructive reads while eliminating the complications associated with capacitor manufacturing. The innovative oxide-semiconductor channel of Z2RAM significantly lessens leakage at elevated junction temperatures, extending the time data can be held between refresh cycles.
Datta explained that memory performance has struggled to keep up with advancements in logic performance, contributing to the “memory wall” challenge faced by modern AI workloads. He emphasized that the reimagined memory hardware stack provided by Z2RAM is essential for ensuring that AI systems maintain both the access bandwidth and bit capacity needed for efficient operation while minimizing energy consumption and physical space requirements.
Nagare Labs’ approach promises to reshape memory technologies and enhance the performance capabilities of AI systems in the future.
