We Are the Frontier

Georgia Spinout's Z2RAM Could Stack Memory Over AI Chips

The research-stage design targets the memory bottleneck, with working samples targeted around 2030.

Updated on Nov. 6, 2026 in Tech

Bare silicon wafer and a short stack of thin gray chip layers on a dark workbench under soft overhead lab light.
Georgia Tech spinout Nagare Labs unveiled Z2RAM, a research-stage memory design, alongside a $5 million seed round led by Playground Global on October 26, 2026. AI Illustration. Upload story photo >

Nagare Labs, a Georgia Tech spinout, unveiled Z2RAM, a memory design built on oxide semiconductors, on October 26, 2026, alongside a $5 million seed round led by Playground Global. The technology is research-stage: demonstrations have confirmed data retention beyond 300 seconds at temperatures up to 85 degrees Celsius, while performance figures remain projections.

Why it matters

Z2RAM targets the memory wall, the widening gap between how fast processors compute and how fast memory can feed them data. Stacking memory in three dimensions directly above an AI accelerator's logic die would shorten the distance data must travel, a direction that matters as AI workloads grow more memory-bound.

The demonstrated result is data retention beyond 300 seconds at temperatures up to 85 degrees Celsius, a laboratory measurement. The headline projections of more than 5x bit density over 2 nm SRAM, 14x faster access than DRAM and one-third the read energy are design targets that have not yet been measured in silicon.

The players

Nagare Labs

A Georgia Tech spinout developing oxide-semiconductor memory stacked over AI logic, with the Z2RAM design as its core technology.

Playground Global

A venture firm that led Nagare Labs' $5 million seed round.

Pat Gelsinger

A former Intel chief executive who serves as a board director of Nagare Labs.

The details

Z2RAM uses a two-transistor memory cell built on oxide semiconductors, a class of materials that can be deposited at low temperatures and still switch reliably. The design eliminates the capacitor, the tiny charge-storing component used in conventional DRAM, which allows fabrication with standard back-end processes. Because the cell is compact and free of capacitors, it can be stacked in layers directly above a processor's logic die, shortening the path data travels.

Timeline

  1. 2026: Nagare Labs was founded by Georgia Tech professors Suman Datta and Shimeng Yu.

  2. October 26, 2026: Nagare Labs unveiled Z2RAM and announced its $5 million seed round.

  3. Around 2030: Nagare Labs targets working memory-tile samples.

  4. Within a decade: Nagare Labs targets mass production.

The Tech Race

Nagare Labs positions Z2RAM as a capacitor-free alternative to DRAM aimed at the memory bottleneck in AI accelerators, a contest also shaped by high-bandwidth memory stacks already used in advanced chips. Its claims are research-stage and must be validated against those incumbent memory technologies before they mean anything in the market.

Z2RAM is a research-stage design with no shipping product, so no user will see it in devices soon. If the memory-tile samples arrive around 2030, the practical benefit would be faster data access for AI accelerators running memory-heavy workloads.

The takeaway

The central trajectory is that AI performance increasingly depends on memory placement, not just compute. Watch for Nagare Labs' working memory-tile samples around 2030, the first point at which Z2RAM's projected density and speed would face measurement in fabricated silicon.

What happens next

Nagare Labs targets working memory-tile samples around 2030, the next concrete checkpoint for whether the projected density and speed figures hold in fabricated silicon.

Further reading

For background on the chip-design work behind this research, see Semiconductors coverage across our tech section.

Source note: This article includes information reported by Crypto Briefing.