Kepler Computing, a San Jose-based startup founded in 2018, has officially emerged from stealth mode with a proposal to break the current AI memory bottleneck. While industry giants are racing to build multibillion-dollar fabs, Kepler claims it can increase memory density using a proprietary 3D stacking approach and new materials that completely bypass the need for extreme ultraviolet (EUV) lithography.
Bypassing EUV Lithography
Most modern chipmakers rely on EUV to shrink transistors and pack more technology into a smaller area. Kepler's strategy is different: by utilizing ferroelectrics and a novel composite material, the company claims it can produce SRAM with densities comparable to 2-nanometer or 3-nanometer chips without the massive investment in EUV equipment. This approach allows Kepler to utilize existing semiconductor fabrication plants, significantly reducing the time and cost required to scale production.
Parallel Development of SRAM and HBM
Originally focused on SRAM, the startup shifted its roadmap following the explosion of demand for High-Bandwidth Memory (HBM) triggered by ChatGPT. Kepler is now developing both in parallel. For HBM, the company uses a 3D-manufacturing technique that fits more memory chips within a fixed footprint, bringing the compute core closer to the memory. This architecture aims to reduce energy consumption during data transfer to levels comparable to SRAM while maintaining HBM's high capacity.
Funding and Scaling Roadmap
The company has raised $468 million from investors including Intel Capital, AMD Ventures, GlobalFoundries, and Bill Gates' Gates Frontier fund. Additionally, the US Department of Commerce has committed up to $245 million to support the development of this high-performance AI memory technology in the US.
Current testing is centered in Singapore at a GlobalFoundries facility, where Kepler has implemented "mini fabs" using 28-nanometer chips. The company's roadmap is ambitious: first HBM samples are expected later this year, followed by a production ramp-up in Singapore next year and the start of US-based production by 2028.
One significant technical hurdle remains: the composite material contains iron, a known contaminant in semiconductor fabs. To manage this, Kepler's process must run on dedicated equipment or be fully encapsulated to prevent contamination of the production flow.
No comments yet. Be the first!