Pat Gelsinger backs Nagare Labs’ bid to expand AI memory capacity
Former Intel chief executive Pat Gelsinger is backing Nagare Labs, a start-up developing Z2RAM, a memory technology intended to ease the bandwidth and capacity constraints affecting AI computing. The company says it could combine SRAM-like speed with much greater density and simpler manufacturing, though its claims face a difficult path from research to commercial production.
Founded by Georgia Tech professors Sumem Datta and Shimeng Yu, Nagare builds Z2RAM from oxide semiconductors and plans to integrate it with processors using 3D hybrid bonding. Its transistor-based cells store charge in a transistor’s gate, enabling non-destructive reads, and the company says its first version could provide over five times SRAM’s bit density. That could mean 320 MB in the space used by a 64 MB SRAM tile, potentially reducing the number of chips needed for some AI systems; the article cautions that many memory technologies have failed before.
In short
- Pat Gelsinger is backing Nagare Labs’ Z2RAM memory development.
- Z2RAM aims to pair SRAM-like speed with over five times its bit density.
- The technology could reduce chip counts for some AI workloads.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Nagare’s approach could address a growing bottleneck in AI systems if it delivers SRAM-like speed with substantially higher density. A successful design might put more fast memory close to processors, reduce the number of chips required and make large AI workloads more efficient; backing from experienced technology figures may help the company pursue the difficult path to production.
The case against
The reported capacity and speed advantages are company claims, and translating a research-stage memory technology into reliable, economical products is a major challenge. Integrating Z2RAM with processors through 3D hybrid bonding adds manufacturing complexity, while the history of memory technologies that failed to reach commercial scale gives reason to wait for independent evidence.
Explained simply
AI processors need memory to hold data close by so they can access it quickly. SRAM is a fast type of memory used in small amounts on processors, but it takes up space, which can limit how much data is available nearby.
Nagare Labs is a start-up founded by Georgia Tech professors Sumem Datta and Shimeng Yu. It is developing Z2RAM, which uses oxide semiconductors and is designed to combine speed with greater storage density; former Intel chief executive Pat Gelsinger is backing the company.
If it works at scale, denser fast memory could help AI systems handle more data with fewer chips. Moving a memory technology from research into reliable, affordable production is difficult, so Z2RAM’s proposed benefits remain to be demonstrated.
Americas Business Companies Entertainment Music World
Read the full article at the source →
Originally published by The Register as “Former Intel CEO Pat Gelsinger bets on Nagare Labs to vault the memory wall”.