Breakthrough: Radical New AI Hardware Design That Nvidia Can’t Ignore | COSM 2025

Summary

The speaker introduces a novel, low-power AI computation design that addresses significant energy efficiency challenges in current computing systems. The innovation involves a "compute in memory" system based on a ferroelectric super lattice that can operate on standard CMOS processing lines, potentially reducing the massive energy and capital investments required by current AI infrastructure. The core breakthrough is a proprietary layer that solves historical reliability challenges, enabling more efficient and cost-effective AI computational capabilities with enhanced performance across extreme operating conditions.