Abstract: Compute is approaching fundamental limits: transistor scaling has been exhausted, and GPU performance has plateaued, stifling progress in AI. This talk introduces a new approach that achieves sub-1 nm compute density on a 22 nm node and reduces data movement by 94%, while preserving full software compatibility. The discussion will focus on the underlying principles and the implications for future large-scale AI systems.
Bio: Murat Onen is the founder and CEO of Eva, a company pioneering a new class of high-performance AI training processors enabled by a fundamental semiconductor breakthrough. Dr. Onen is currently leading the development of next-generation compute clusters that deliver 72x higher performance to unlock advanced AI applications far beyond current reach.